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
Many species that run or leap across sparsely vegetated habitats , including horses and deer , evolved the severe reduction or complete loss of foot muscles as skeletal elements elongated and digits were lost , and yet the developmental mechanisms remain unknown . Here , we report the natural loss of foot muscles in the bipedal jerboa , Jaculus jaculus . Although adults have no muscles in their feet , newborn animals have muscles that rapidly disappear soon after birth . We were surprised to find no evidence of apoptotic or necrotic cell death during stages of peak myofiber loss , countering well-supported assumptions of developmental tissue remodeling . We instead see hallmarks of muscle atrophy , including an ordered disassembly of the sarcomere associated with upregulation of the E3 ubiquitin ligases , MuRF1 and Atrogin-1 . We propose that the natural loss of muscle , which remodeled foot anatomy during evolution and development , involves cellular mechanisms that are typically associated with disease or injury . Muscles in the feet of birds , reptiles , and mammals were lost multiple times in the course of limb evolution , usually coinciding with the loss of associated digits and elongation of remaining skeletal elements ( Hudson , 1937; Raikow , 1987; Pavaux and Lignereux , 1995; Botelho et al . , 2014; Abdala et al . , 2015; Berman , 1985; Cunningham , 1883; Souza et al . , 2010 ) . Despite its frequent occurrence , the developmental mechanisms that lead to the natural absence of adult limb muscle are not known . We focus here on a representative example of distal limb muscle loss in the bipedal three-toed jerboa ( Jaculus jaculus ) , a small laboratory rodent model for evolutionary developmental biology , to determine if evolutionary muscle loss conforms to expectations based on what was previously known about muscle cell biology . The hindlimb architecture of the adult jerboa is strikingly similar by convergence to the more familiar hooved animals , like horses and deer , including the disproportionately elongated foot that lacks all intrinsic muscle ( Berman , 1985; Cunningham , 1883 ) . The tendons were retained and expanded in each of the anatomical positions where flexor muscles are absent ( Figure 1A , B and Figure 1—figure supplement 1A , B ) and serve to resist hyperextension when the terminal phalanx contacts the ground during locomotion ( Lochner et al . , 1980; Moore et al . , 2017 ) . The evolutionary origin of jerboa intrinsic foot muscle loss lies deep in the phylogenetic tree of Dipodoid rodents . Compared to the ancestral state , the number of intrinsic foot muscles are reduced from sixteen to six in pygmy jerboas ( Stein , 1990 ) which diverged from the three-toed jerboa lineage more than 20 million years ago ( Wu et al . , 2012; Pisano et al . , 2015 ) . The mechanisms of limb muscle development have been extensively studied in traditional model systems , and its degeneration has been studied after injury and during disease . Briefly , limb muscle progenitors are specified from mesodermal cells at the ventrolateral edge of the dermomyotome in somites aligned with the prospective limb . These cells delaminate and migrate into the limb bud as dorsal and ventral muscle masses that proliferate and initiate a myoblast specification program ( Chevallier et al . , 1977; Christ et al . , 1977; Hayashi and Ozawa , 1991; Murphy and Kardon , 2011 ) . The muscle masses are then subdivided into individual muscle groups in response to cues from the developing muscle connective tissue , which is derived from limb field lateral plate mesoderm ( Hayashi and Ozawa , 1991; Kardon , 1998; Kardon et al . , 2003; Wortham , 1948 ) . They then initiate a differentiation program , which includes cell fusion to form aligned multinucleated myofibers ( Abmayr and Pavlath , 2012; Kelly and Zacks , 1969 ) . Each differentiated myofiber produces an assemblage of Z-body proteins , Actin filaments , and non-muscle Myosin that form premyofibrils ( Ono , 2010; Rhee et al . , 1994; Sanger and Sanger , 2008; Sanger et al . , 2002 ) . Desmin , α-Actinin , and the Z-body portion of Titin also begin to organize ( Furst et al . , 1989; Sanger et al . , 2002 ) . Subsequent uncoiling of Titin increases Z-body spacing , and integration of embryonic skeletal muscle Myosin results in formation of nascent myofibrils ( Ono , 2010; Sanger et al . , 2010 ) . Further maturation of the nascent myofibril into a mature myofibril involves incorporation of additional proteins that are important for sarcomere structure and function , and Z-lines are aligned and properly spaced to bring sarcomeres into register ( Ehler and Gautel , 2008; Sanger et al . , 2010 ) . Failure at any point of myoblast specification , migration , myofiber differentiation , or myofibril maturation compromises muscle function and manifests as muscle degenerative disease in humans ( Bönnemann and Laing , 2004; Laing and Nowak , 2005; Morita et al . , 2005 ) . Working backward in time from the adult jerboa phenotype , we found that two of the three flexor muscle groups differentiate as multinucleated myofibers that initiate sarcomere assembly , as in other species . However , almost all jerboa intrinsic foot muscle is lost within a few days shortly after birth . Despite the rapid and near complete loss of myofibers , we found no molecular or ultrastructural evidence of apoptotic or necrotic cell death , no accumulation of autophagic vesicles , and no macrophage infiltration . Instead , we observed evidence of ordered sarcomere disassembly and upregulation of muscle-specific ubiquitin ligases , MuRF1 and Atrogin-1 . Although the ultimate fate of intrinsic foot myofibers after loss of muscle identity remains unknown , these data suggest that the mechanism of myofiber loss is similar to atrophy , which is typically considered a pathological response to injury or disease . The absence of intrinsic foot muscle in the adult jerboa could be due to a failure of early myoblasts to migrate into and/or to differentiate in the distal limb . Alternatively , embryonic muscles may form but not persist through development to the adult . In transverse sections of newborn mouse feet , immunofluorescent detection of skeletal muscle myosin heavy chain reveals each intrinsic muscle group ( Figure 1—figure supplement 1G ) . In newborn jerboas , we observed two of the three groups of flexor muscles . While the m . lumbricales never form , the jerboa has a single m . flexor digitorum brevis and three pinnate m . interossei that are not present in adults ( Figure 1—figure supplement 1H ) . Postnatal growth of vertebrate skeletal muscle typically involves an increase in myofiber number ( hyperplasia ) within the first week , followed by an increase in myofiber size ( hypertrophy ) ( Chiakulas and Pauly , 1965; Gokhin et al . , 2008; White et al . , 2010 ) . In order to understand the dynamics of muscle growth and loss , we quantified the rate of myofiber hyperplasia at 2-day intervals after birth of the mouse and jerboa , focusing on the representative interosseous muscle that is associated with the third metatarsal ( Figure 1E , F ) . As expected in the mouse , we observed a steady increase in the average number of myofibers in cross section from birth to P8 ( Figure 1C ) . In contrast , the number of myofibers in the third interosseous of the jerboa foot rapidly declines beginning at approximately P4 , and few myofibers remain by P8 ( Figure 1D ) . It is possible that the rate of myofiber loss outpaces a typical rate of new cell addition such that muscles with the potential to grow are instead steadily diminished . Alternatively , myofiber loss may be accelerated by a compromised ability to form new myofibers and to add nuclei to growing myofibers . To distinguish these hypotheses , we analyzed cohorts of animals 2 days after intraperitoneal BrdU injection at P0 , P2 , or P4 . Since multinucleated jerboa foot myofibers are postmitotic ( Figure 2—figure supplement 1 ) , we reasoned that BrdU+ nuclei present within Dystroglycan+ myofiber membranes were added by myocyte fusion during the 2-day window after they were labeled as myoblasts or myocytes in S-phase ( Figure 2A ) . When normalized to the total number of myofiber nuclei , we found that myocytes fuse to form multinucleated myofibers in jerboa hand muscle at a consistent rate from P0 to P6 . However , their incorporation into jerboa foot muscle decreased significantly after P2 ( Figure 2B ) . These results suggest that myofiber loss , which begins at P4 , is preceded by reduced myogenesis . The reduced rate of myocyte incorporation could be due to reduced numbers of muscle progenitor cells or to an inability of these cells to mature and fuse . To distinguish these possibilities by quantifying proliferative muscle progenitor cells , we analyzed animals 2 hr after BrdU injection at P0 , P2 , and P4 and counted the number of BrdU+ nuclei located between the Dystroglycan+ myofiber membrane and the Laminin+ basal lamina ( Figure 2C ) . Normalized to the total number of myofibers , we found that the number of proliferative progenitor cells in jerboa foot muscle significantly decreased from P0 to P4 compared to hand muscles that showed no change over time ( Figure 2D ) . These results suggest that a reduced number of muscle progenitor cells might contribute to the reduced prevalence of myocyte fusion events . We next tested whether compromised proliferation and differentiation of jerboa foot muscle progenitors is cell autonomous or non-cell autonomous . We isolated single cells , including myoblasts and myocytes but excluding myofibers , by mechanical trituration and enzymatic digestion of P1 jerboa and mouse lower leg and foot muscles ( Danoviz and Yablonka-Reuveni , 2012 ) . After 6 days and 9 days of culture , we detected Myogenin+ differentiating myocytes and Myosin+ fully differentiated myofibers in primary cell cultures isolated from each muscle ( Figure 2E ) . We did not detect a significant decline in the number of differentiated cells over time ( Figure 2—figure supplement 2 ) . Jerboa foot muscle cell differentiation and survival in vitro days after cell number begins to decline in vivo suggests that loss of jerboa foot myofibers is initiated non-cell autonomously . The rapid and almost complete loss of differentiated myofibers in vivo from P4 to P8 suggested these cells die , since individual cells or groups of cells are commonly eliminated by apoptosis during development ( Brill et al . , 1999; Fernández-Terán et al . , 2006 ) . We therefore tested the hypothesis that neonatal intrinsic foot muscles undergo apoptosis by implementing the TUNEL assay to detect DNA fragmentation and by immunofluorescent detection of cleaved Caspase-3 , a key protein in the apoptotic program ( Elmore , 2007 ) . Each revealed keratinocyte apoptosis in hair follicles , which are known to undergo programmed cell death , as a positive control in the same tissue sections ( Magerl et al . , 2001 ) . However , TUNEL or cleaved Caspase-3-positive jerboa foot myofibers or cells in their vicinity were an extreme rarity ( 0 . 25% of myofibers ) in animals ranging from P0 to P8 and comparable to mouse myofibers suggesting muscle is not eliminated by apoptosis ( Figure 3A , B and Figure 3—figure supplement 1 ) . Alternatively , myofiber loss may occur through a cell death mechanism that is first characterized by plasma membrane permeability , such as necrosis ( Vanden Berghe et al . , 2014 ) . To test this hypothesis , we injected Evans blue dye ( EBD ) , a fluorescent molecule that accumulates in cells with compromised plasma membranes ( Hamer et al . , 2002; Matsuda et al . , 1995 ) , into the peritoneum of P3 and P4 neonatal jerboas 24 hr before euthanasia . Although we detected EBD in mechanically injured myofibers of the gastrocnemius as a control , we saw no EBD fluorescence in jerboa foot myofibers or in surrounding cells ( Figure 3C and Figure 3—figure supplement 2 ) . We also saw no Annexin V immunofluorescence on the surface of jerboa foot myofibers , another hallmark of dying cells that flip Annexin V to the outer plasma membrane ( Figure 3D and Figure 3—figure supplement 2 ) . Since we observed no direct evidence of cell death , we asked whether there was an immune response that might be an indirect proxy for undetected death . Dying muscle cells frequently recruit phagocytic macrophages that engulf cellular debris ( Arnold et al . , 2007; Londhe and Guttridge , 2015; Tidball and Wehling-Henricks , 2007 ) . We predicted that myofibers that die by any mechanism that produces cellular debris might recruit macrophages that are detectable by expression of the F4/80 glycoprotein . However , consistent with the lack of evidence of cell death in the jerboa foot , no F4/80+ macrophages were found among myofibers from birth to P7 ( Figure 3—figure supplement 3 ) . Since immune cells other than mature macrophages might be recruited to a site of cell death , we also assessed expression of CD45 and found no evidence of T-cells , B-cells , dendritic cells , natural killer cells , monocytes , or granulocytes near jerboa foot myofibers from P4 to P8 ( Figure 3E and Figure 3—figure supplement 3 ) . The absence of any clear indication of muscle cell death motivated us to re-evaluate muscle maturation at greater resolution in order to capture the earliest detectable signs of muscle cell loss . We collected transmission electron micrographs of jerboa hand and foot muscle at P0 , P2 , and P4 . We identified criteria for three categories of maturation , as described previously ( Borisov et al . , 2008; Raeker et al . , 2014; Sanger et al . , 2006 ) , and two categories of degeneration . Category A cells have pre-myofibrils with thick and thin filaments and poorly resolved Z-discs , but the M-lines and I-bands are not yet apparent ( Figure 4A ) . In Category B , Z-discs of nascent myofibrils are better resolved , and M-lines and I-bands are apparent , but parallel sarcomeres are not in register ( Figure 4B ) . The mature myofibrils of Category C have Z-lines that are aligned with one another ( Figure 4C ) . In Category D , early degeneration , some sarcomeres appear similar to Category A , but other areas of the cell contain disorganized filaments ( Figure 4D ) . Category E includes those in the worst condition where less than half of the cell has any recognizable sarcomeres , and much of the cytoplasm is filled with pools of disorganized filaments and Z-protein aggregates ( Figure 4E ) . Additionally , Category D and E cells have membrane-enclosed vacuoles and large lipid droplets ( Figure 4—figure supplement 1 ) . However , consistent with a lack of evidence for cell death , none of these cells or their organelles appear swollen , nuclear morphology appears normal , plasma membranes seem to be contiguous , and we do not observe an accumulation of autophagic vesicles that typically characterize cell death associated with unregulated autophagy ( Mizushima , 2007; Kroemer and Levine , 2008; Denton and Kumar , 2019 ) . We then coded and pooled all images of hand and foot myofibers from P0 , P2 , and P4 jerboas and blindly assigned each cell to one of the five categories . Quantification of the percent of myofibers in each category after unblinding revealed the progressive maturation of jerboa hand myofibers and the progressive degeneration of jerboa foot myofibers ( Figure 4G ) . Compared to later stages , there is little difference in the maturation state of hand and foot sarcomeres at birth . Loss of ultrastructural integrity is therefore initiated perinatally , prior to complete myofibril maturation in the jerboa foot . Our analysis of transmission electron micrographs also revealed the presence of filamentous aggregates that we did not include in our quantifications because they are enucleate , lack all other recognizable organelles , and are not bounded by a plasma membrane . Although these aggregates do not appear to be cellular , they are always closely associated with cells of a fibroblast morphology , and most lie between remaining myofibers in a space we presume was also once occupied by a myofiber ( Figure 4F , H ) . To determine if these unusual structures contain muscle protein , we performed immunofluorescence on sections of P4 jerboa foot muscle and found similar aggregates of intensely fluorescent immunoreactivity to skeletal muscle myosin heavy chain . We also found that the surrounding cells , which correlate with the positions of fibroblasts in electron micrographs , express the intracellular pro-peptide of Collagen I ( Figure 4I ) , the major component of tendon and other fibrous connective tissues and of fibrotic tissue after injury ( Mann et al . , 2011 ) . Given the apparent deterioration of nascent sarcomeres , we asked whether individual sarcomere proteins are lost from myofibrils in a temporal order or if proteins disassemble simultaneously . We assessed the organization of sarcomere proteins by multicolor immunofluorescence at P0 , P2 , and P4 . Alpha-Actinin , Desmin , Myomesin , Myosin , Titin , and Tropomyosin are each localized to an ordered series of striations in a subset of myofibers suggesting all are initially incorporated into immature sarcomeres ( Figure 5A and Figure 5—source data 1–5 ) . By assessing all combinations of immunologically compatible primary antibodies , we identified populations of cells where Desmin was no longer present in an ordered array , but each of the other proteins appeared properly localized to the sarcomere ( Figure 5B and Figure 5—source data 2 ) . Although we could not distinguish such clear categories of mislocalization for each protein relative to all others , we inferred a relative timeline whereby Desmin disorganization is followed together by Myosin and Tropomyosin , then Titin , and lastly Myomesin and α-Actinin ( Figure 5B–F and Figure 5—source data 1–5 ) . Desmin forms a filamentous network that connects parallel sarcomeres to one another and coordinates myofibril contraction within cells and between neighboring cells ( Bär et al . , 2004; Capetanaki et al . , 2015; Goldfarb et al . , 2008 ) . Mutations that cause desminopathies illustrate that Desmin is essential to maintain sarcomere integrity ( Clemen et al . , 2013 ) . In mouse models of muscle atrophy triggered by fasting or denervation , phosphorylation of Desmin removes the protein from the sarcomere and targets it for ubiquitination and proteolytic degradation prior to degradation of other sarcomere proteins ( Volodin et al . , 2017 ) . The observation that Desmin is the first of an ordered sarcomere disassembly in the jerboa foot may reflect targeted degradation of muscle proteins that is similar to muscle atrophy . The ubiquitin-proteasome system is the main pathway through which cellular proteins are degraded during muscle atrophy , and MuRF1 and Atrogin-1 are E3 ubiquitin ligases among the ‘atrogenes’ that are highly upregulated ( Bodine and Baehr , 2014; Schiaffino et al . , 2013 ) . To test the hypothesis that muscle loss in the jerboa foot exhibits molecular hallmarks of atrophy , we performed quantitative reverse transcriptase PCR ( qRT-PCR ) of MuRF1 and Atrogin-1 mRNA from intrinsic foot muscles and the flexor digitorum superficialis ( FDS ) of the mouse and jerboa . The FDS , which originates in the autopod during embryogenesis and translocates to the forearm ( Huang et al . , 2013 ) , is the most easily dissected of the analogous forelimb muscles and serves as a control for typical muscle maturation in both species . When normalized to expression in the FDS at birth of each species , Atrogin-1 expression is 3 . 1-fold higher in the jerboa foot at P3 ( Figure 5G ) . MuRF1 mRNA expression is already significantly elevated at birth and remains elevated at P3 ( Figure 5H ) . The NF-κB pathway is an upstream mediator of skeletal muscle atrophy ( Li et al . , 2008 ) and is both necessary and sufficient to induce MuRF1 expression ( Cai et al . , 2004; Wu et al . , 2014 ) . To lend further support to the hypothesis that jerboa foot muscle loss involves an ‘atrophy-like’ mechanism , we performed qRT-PCR of NF-κB2 and its binding partner , Relb . We observed that each gene is expressed greater than three-fold higher in jerboa foot at birth and at P3 ( Figure 5—source data 1 ) . The progressively disordered ultrastructure of the sarcomere that begins with loss of Desmin localization , the increased expression of multiple genes that are typically upregulated during atrophy , and the lack of evidence for cell death or macrophage infiltration are consistent with observations of atrophying muscle in mice and rats ( Volodin et al . , 2017; Bonaldo and Sandri , 2013; Sakuma et al . , 2015; von Haehling et al . , 2010 ) . Despite the similarities to muscle atrophy , myofiber loss in the jerboa foot does not seem to be simply explained by an atrophic response to denervation . First , and in contrast to the rapid rate of jerboa foot myofiber loss , chronic denervation in mice ( 100 days after nerve transection at P14 ) reduced the size but not the number of individual myofibers ( Moschella and Ontell , 1987 ) . Additionally , we found that the post-synaptic Acetylcholine Receptor ( AchR ) exclusively coincides with the presynaptic neuronal protein , Synaptophysin , in neonatal jerboa foot muscles ( Figure 5—figure supplement 1 ) . In the mouse , AchR clusters are present in a broad domain of fetal muscle prior to innervation and are refined to nerve terminals in response to chemical synapse activity before birth ( Yang et al . , 2001 ) . The refinement of AchR clusters in jerboa foot muscles suggests that the muscles are not only innervated at birth but are also responsive to motor inputs . Beginning only with knowledge of adult jerboa foot anatomy , we reached evidence for a cellular mechanism of intrinsic muscle loss during neonatal development that is surprising in the context of what is known about muscle development and pathology in more traditional laboratory species . Although we found multinucleated myofibers in the feet of neonatal jerboas , all muscle protein expression rapidly disappears from the jerboa foot shortly after birth . We were perplexed to find no evidence of apoptotic or necrotic cell death by a variety of assays and throughout the time when muscle cells are lost , nor did we observe immune cells that are commonly recruited to clear the remains of dying cells . Instead , we saw structural and molecular similarities to muscle atrophy , although atrophy in young mice leads to reduced myofiber size rather than number as in the jerboa ( Bruusgaard and Gundersen , 2008; Moschella and Ontell , 1987 ) . In an effort to functionally connect the atrogenes , MuRF1 and Atrogin-1 , to a mechanism of muscle loss , we endeavored to determine whether knocking down both genes in jerboa foot muscle in vivo could be sufficient to rescue myofibers or to delay their loss . Although we developed a system to validate shRNAs targeting each jerboa gene after first inducing MuRF1 and Atrogin-1 expression in cultured primary jerboa myofibers , shRNA delivery into rodent neonatal foot muscle was not feasible ( not shown ) . Lentiviral injection bathes the ensheathing muscle connective tissue , and infection rarely reaches myofibers of these small muscles in either mouse or jerboa . Plasmid injection and electroporation , which is feasible in adult rodent feet ( DiFranco et al . , 2009 ) , is not efficient in neonatal feet of either species . This is likely because the neonatal foot lacks a cavity to contain injected DNA , provided by separation of the overlying skin in mature animals , which is required for efficient plasmid uptake by electroporation ( Krull , 2004 ) . Even if genetic manipulations were technically feasible , intrinsic foot muscle loss first appeared in the jerboa lineage more than 20 million years ago . For phenotypes that diverged over such long evolutionary distances , a large set of genes and mechanical acommodation of integrated tissues are likely at play , which were collectively honed by millions of years of evolution; thus , manipulation of one or two genes may not be sufficient to rescue or delay muscle loss . It is therefore important to consider that a standard of ‘molecular mechanism’ applied to the experimental manipulations of traditional laboratory species may not be appropriate in the context of understanding complex macroevolutionary processes . As for why the phenotype is limited to the distal limb , it is possible that disuse contributes to jerboa foot muscle loss , since jerboas and ungulates each fuse metatarsals into a single cannon bone , which would be expected to physically impair lateral motion of the digit elements ( Cunningham , 1883; Moore et al . , 2015 ) . However , the rapid and complete loss of myofibers in the neonatal jerboa foot does not appear to simply reflect a species-level difference in the animal’s generalized response to disuse atrophy , since hindlimb denervation or immobilization in adult jerboas causes gradual loss of muscle mass , primarily through a significant reduction in the diameter of individual myofibers ( AlWohaib and Alnaqeeb , 1997; Aryan and Alnaqeeb , 2002 ) . These observations are very similar to what has been shown in disuse atrophy models in mice and in rats ( Bonaldo and Sandri , 2013; Moschella and Ontell , 1987 ) and differ from what we see in the neonatal foot . Why would an embryo expend energy to form muscles that are almost immediately lost ? The formation and subsequent loss of intrinsic foot muscles in jerboas and hooved animals may simply reflect a series of chance events in each lineage with no fitness cost , or these similarities in multiple species may reveal true developmental constraints . Muscle is not required for autopod tendon formation or maintenance in mice , but the tendons that develop in a muscle-less or a paralyzed mouse are thinner and less well organized ( Huang et al . , 2015 ) . It is therefore possible that muscle is initially required in the fetus and neonate for tendons to establish sufficient architecture from origin to insertion so that the tendon , after further growth , can withstand high locomotor forces in the adult ( Lochner et al . , 1980; Moore et al . , 2017 ) . Regardless of whether these nascent muscles serve an essential purpose , we are left wondering what is the ultimate fate of jerboa foot myofibers . If these cells do indeed die , perhaps death is too rapid for detection . However , programmed cell death is thought to occur over hours or even days from the initial trigger to the final corpse ( Green , 2005 ) . Alternatively , death may result from a mechanism that does not proceed through DNA fragmentation , plasma membrane permeability , macrophage recruitment , or stereotyped ultrastructural changes , and yet this would seem to eliminate most known forms of regulated cell death ( Galluzzi et al . , 2007 ) . Alternatively , multinucleated myofibers may transform to another cellular identity after degrading all sarcomere proteins . Although a fate transformation would be surprising , it would not be without precedent . The electric organ of fish that can produce an electric field ( e . g . knifefish and elephantfish ) is thought to be derived from skeletal muscle . Electrocytes of Sternopygus macrurus express skeletal muscle Actin , Desmin , and α-Actinin , and electrocytes of Paramormyrops kingsleyae retain sarcomeres that are disarrayed and non-contractile ( Gallant et al . , 2014; Unguez and Zakon , 1998 ) . If myofibers in the jerboa foot indeed transdifferentiate , it is possible that they transform into the pro-Collagen I positive fibroblasts that are entangled with the filamentous aggregates , although these could also be phagocytic fibroblasts recruited to consume the enucleate detritus without stimulating inflammation ( Heredia et al . , 2013; Joe et al . , 2010; Monks et al . , 2005; Schwegler et al . , 2015 ) . Unfortunately , the lineage labeling approaches required to track the ultimate fate of jerboa myofibers are exceptionally challenging in this non-traditional animal model . It is clear , however , that regardless of the ultimate fate of jerboa foot myofibers , their path passes through a phase marked by cell biology that is typical of atrophy , including the ordered disassembly of sarcomeres and expression of the E3 ubiquitin ligases , MuRF1 and Atrogin-1 . However , skeletal muscle atrophy is typically associated with pathology in the context of disuse , nerve injury , starvation , or disease . In this context , we were struck by a statement in the 1883 anatomical description of the fetal and adult suspensory ligament of four species of hooved mammals: ‘It is an instance of pathological change assisting a morphological process’ ( emphasis his ) ( Cunningham , 1883 ) . Indeed , there are remarkable similarities in the histology of jerboa and horse foot muscle compared to human clinical observations of tissue remodeling that follows rotator cuff tear characterized by muscle atrophy , myofiber loss , and fibrosis ( Souza et al . , 2010; Gibbons et al . , 2017 ) . Foot muscle atrophy in the jerboa may be one of many cellular responses associated with injury or disease in humans that is utilized in the normal development and physiology of other species . These data suggest that there is less of a clear divide between natural and pathological states than typically thought . Studies of non-traditional species may not only reveal the mechanisms of evolutionary malleability , but may also advance our understanding of fundamental biological processes that are typically associated with pathological conditions . Jerboas were housed and reared as previously described ( Jordan et al . , 2011 ) . CD1 mice were obtained from Charles River Laboratories ( MA , USA ) , housed in standard conditions , and fed a breeder’s diet . All animal care and use protocols for mice and jerboas were approved by the Institutional Animal Care and Use Committee ( IACUC ) of the University of California , San Diego . The following primary antibodies and dilutions were used for immunofluorescence of tissue sections: Col1A1 ( SP1 . D8 , 1:20 ) , Dystroglycan ( 11H6C4 , 1:10 ) , Myosin heavy chain ( MF20 , 1:20 ) , Myomesin ( B4 , 1:20 ) , Myogenin ( F5D , 1:5 ) , Titin ( 9D10 , 1:10 ) , Tropomyosin ( CH1 , 1:10 ) , Developmental Studies Hybridoma Bank; Desmin ( D33 , 1:300 ) , α-actinin ( EA-53 , 1:1000 ) , Sigma Aldrich; Annexin-V ( ab14196 , 1:100 ) , Desmin ( ab32362 , 1:500 ) , CD45 ( ab10558 , 1:200 ) , F4/80 ( ab6640 , 1:200 ) , Abcam; Cleaved Caspase-3 ( Asp175 ) ( #9661 , 1:100 ) , Cell Signaling Technologies; Alexa 488 conjugated Wheat Germ Agglutinin ( W11261 , 1:200 ) , Invitrogen; BrdU ( MCA2060 , 1:100 ) , Biorad . Secondary antibodies were obtained from Invitrogen and used at 1:250 dilution: Alexa Fluor 594 conjugated goat anti-mouse IgG2b , Alexa Fluor 488 or 647 conjugated goat anti-mouse IgG1 , Alexa Fluor 488 conjugated goat anti-mouse IgM , Alexa Fluor donkey anti-mouse IgG , Alexa Fluor 488 conjugated goat anti-rat IgG , Alexa Fluor 488 or 647 conjugated goat anti-rabbit . Mouse and jerboa limbs were dissected and fixed in 4% PFA in 1x PBS overnight . Tissues were washed in 1X PBS twice for 20 min and placed in 30% sucrose in 1x PBS overnight at 4°C . Tissues were then mounted in a cryomold in OCT freezing media , and blocks were frozen and stored at −80°C until cryosectioned . Blocks were sectioned at 12 µm thickness , and sections were transferred to Super-Frost Plus slides ( Thermo Fisher ) . For immunofluorescence , slides were washed for 5 min in 1x PBS and subject to antigen retrieval by incubation in Proteinase K ( 5 µg/mL ) for 10 min followed by 5 min postfix in 4% PFA in PBS and three washes in 1x PBS . Slides were then blocked in a solution of 5% heat inactivated goat serum , 3% BSA , 0 . 1% TritonX-100 , 0 . 02% SDS in PBS . Slides were incubated in the appropriate primary antibody dilution in block overnight at 4°C . On the second day , slides were washed three times for 10 min in PBST ( 1x PBS + 0 . 1% TritonX-100 ) and incubated at room temperature in secondary antibodies and 1 µg/ml DAPI for 1 hr . Slides were then washed three times for 10 min in PBST and mounted in Fluoro Gel with DABCO ( EMS ) . For TUNEL , slides that had been previously processed for MF20 immunofluorescence were placed immediately into the TUNEL reaction mixture following manufacturer’s instructions ( Roche In Situ Cell Death Detection Kit , TM-Red ) for 60 min at 37°C , rinsed three times in 1x PBS , and mounted in Fluoro Gel with DABCO . Sections were imaged with Olympus compound microscope model BX61 , Leica SP5 confocal , or Olympus FV1000 confocal . Blocks containing embedded mouse or jerboa feet were cryosectioned at 12 µm thickness in transverse orientation onto two serial sets of slides . Slides of the second series were used as back up in case certain sections of the first series contain folded tissue and thus cannot be used . Slides of the first series were stained with MF20 and WGA and analyzed to locate the proximal and distal ends of the third interosseous muscle . Using this information we estimated the middle area of each muscle and selected 10 sections for subsequent analysis . We analyzed the third interosseous muscle of the hindlimb , spanning approximately 240 µm in muscle length . For each selected section , all cross-sectionally oriented myofibers were manually counted and recorded using the plugin cell counter in ImageJ . The average number of myofibers from 10 sections represents an estimate of the myofiber number for the middle transverse section of the third interosseous muscle . For each developmental stage , data from three animals were collected , and one-way ANOVA with Tukey’s multiple comparisons test was performed to determine the statistical significance of mean myofiber number differences between developmental stages in each species . BrdU solution was intraperitoneally injected to achieve 100 μg/g ( BrdU/animal body weight ) in P0 , P2 , and P4 jerboas . Injected animals were sacrificed 2 days later . The feet and hands of each animal were fixed in 4% PFA/PBS overnight , processed through a sucrose series , and embedded in OCT freezing media . Blocks of embedded tissue were cryosectioned in transverse orientation at 12 µm thickness and placed in serial sets on Superfrost Plus slides . Slides were stained with BrdU and Dystroglycan antibodies as indicated above . As in the methods to count myofibers , we chose ten sections near the midpoint of the interosseous muscle associated with the third metatarsal and counted all BrdU+ nuclei within a Dystroglycan+ myofiber as well as all myofiber nuclei in each section . Data is represented as the total number of BrdU+ myofiber nuclei divided by the total number of myofiber nuclei , and this ratio was averaged for all 10 sections in each animal . The data was plotted using Prism8 ( GraphPad ) , and the statistical significance between datapoints at each time interval was calculated with one-way ANOVA with Tukey’s multiple comparisons test in each of forelimb and hindlimb . BrdU solution was intraperitoneally injected to achieve 100 μg/g ( BrdU/animal body weight ) in P0 , P2 , and P4 jerboas . Injected animals were sacrificed two hours after injection . The feet and hands of each animal were fixed in 4% PFA/PBS overnight , processed through a sucrose series , and embedded in OCT freezing media . Blocks of embedded tissue were cryosectioned in transverse orientation at 12 µm thickness and placed in serial sets on Superfrost Plus slides . Slides were stained with BrdU and Myosin or BrdU , Laminin , and Dystroglycan antibodies as indicated above for assessment of proliferation in myonuclei . As in the methods of fusion assay , we chose 10 sections near the midpoint of the interosseous muscle associated with the third metatarsal and counted all BrdU+ nuclei within a Laminin+ basal lamina and outside Dystroglycan+ myofiber membrane as well as number of Dystroglycan+ myofiber in each section . Data is represented as the total number of BrdU+ myofiber nuclei divided by the total number of myofiber , and this ratio was averaged for all 10 sections in each animal . The data was plotted using Prism8 ( GraphPad ) , and the statistical significance between datapoints at each time interval was calculated with one-way ANOVA with Tukey’s multiple comparisons test in each of forelimb and hindlimb . Intrinsic foot muscles ( m . flexor digitorum brevis and m . interossei ) and lower leg muscles ( tibialis anterior and gastrocnemius ) were manually dissected from three animals of P1 jerboas and mice and pooled . After connective tissues were manually removed with forceps , muscle stem/progenitor cells were isolated and cultured as described in Danoviz and Yablonka-Reuveni , 2012 . Briefly , the tissues were enzymatically with 10 mg/ml Pronase ( EMD Millipore ) and mechanically dissociated . The cells were plated onto matrigel-coated 8-well chamber slides ( Nunc Lab-Tek , Thermo fisher ) coated with Matrigel ( Corning ) at 1 × 104 cells/well . The cells were cultured for 9 days with DMEM ( Thermo Fisher ) , 20% fetal bovine serum ( Thermo fisher ) , 10% horse serum ( Thermo Fisher ) and 1% chicken embryonic extract ( Accurate Chemical ) . During the culture period , the medium was changed at days 3 , 6 , and 8 . After 6 and 9 days , cells in replicate cultured wells were fixed with 4% PFA/PBS at 4°C for 15 min and washed with PBS . After permeabilization with 1% Triton-X 100 in PBS at room temperature for 10 min , the cells were blocked with 5% BSA/PBS for 30 min and stained with BrdU , anti-Myogenin and Myosin antibodies and secondary antibodies . At each time point of each experimental group , the total number of nuclei and nuclei within Myosin+ myofibers were counted in 10 images taken from eight wells using the Olympus compound microscope at 4x magnification . The numbers in 10 images were averaged and the difference between day 6 and day 9 were statistically analyzed with paired sample t-test in each experimental group . We injected Evans Blue Dye as 1% solution by animal body weight ( 1 mg EBD/100 µl PBS/10 g ) 24 hr prior to sample collection ( Hamer et al . , 2002 ) . As a positive control for EBD uptake , we create an injured muscle area by inserting a 21-gauge needle 2–3 times into the jerboa gastrocnemius muscle . Samples were fresh frozen in OCT and cryosection at 12 µm thickness . Slides were processed for MF20 fluorescence with primary antibody incubation for 1 hr at RT before secondary antibody incubation . Slides were mounted for analysis: EBD signal is detected using the Cy5 filter and imaged using the Olympus compound microscope or imaged using the Leica SP5 confocal laser 633 nm . ORO stock solution: 2 . 5 g of Oil red O to 400 ml of 99% ( vol/vol ) isopropyl alcohol and mix the solution by magnetic stirring for 2 hr at room temperature ( RT; 20–25°C ) . ORO working solution: 1 . 5 parts of ORO stock solution to one part of deionized ( DI ) . Cryosections were fixed with 4%PFA in 1x PBS for 5 min . Slides were washed with 2x with PBS for 10 min each and stained with ORO working solution for 10 min followed three 30 s washes with DI water . Slides were then washed in running tap water for 15 min followed by three 30 s washes with DI water and mounting in aqueous medium . Animals were perfused with 2% glutaraldehyde and 2% PFA plus 2 mM CaCl2 in 0 . 15M sodium cacodylate buffer , pH 7 . 4 @ 35°C for 2–3 min . The hands and feet were removed , skinned , and fixed on ice for 2 hr . Samples were then rinsed six times for 5 min in cold 0 . 15M cacodylate buffer and then post-fixed in 1% OsO4 in 0 . 15M cacodylate buffer on ice for 1 hr . Samples were then rinsed in cold double distilled water ( DDW ) six times for 5 min and placed into 1% uranyl acetate in DDW on ice overnight . Fixed tissue was then rinsed in ice cold double distilled water three times for 3 min and dehydrated in an ethanol series ( 50% , 70% , 90% in DDW ) on ice for 5 min each . Samples were further dehydrated into 100% ethanol twice for 5 min at room temperature and then transitioned to 1:1 ethanol:acetone for 5 min followed by two times 5 min in 100% acetone . Dehydrated samples were infiltrated with 1:1 acetone:Durcupan ACM resin for 1 hr at room temperature followed by 100% resin twice for 1 hr and then placed in fresh resin overnight . On the next day , samples were transferred to fresh resin , which was polymerized in a 60°C vacuum oven for 48–72 hr . Resin embedded samples were stored at room temperature until ready for sectioning . Seventy nanometer thick sections were stained in lead solution and image using Tecnai Spirit TEM scope ( 120 kV ) . Foot muscles were dissected and stored in RNAlater solution ( Thermo Fisher ) at −80°C until ready for use . RNA extraction was performed using the PicoPure RNA Isolation Kit ( Thermo Fisher ) according to the manufacture instructions . RNA was reverse transcribed to generate cDNA using QuantiTect Reverse Transcription Kit . cDNA was used as template for quantitative PCR with PCR amplification detected with Sybr green ( SYBR Green Real-time PCR master mixes , Invitrogen ) . See the table below for the sequences of primers used to quantify real time amplification . Each quantitative reverse transcriptase PCR experiment was conducted twice with technical triplicates in each experiment . Cq values that are significant outliers were determined using Grubb’s test in GraphPad software and eliminated . Expression of MuRF-1 , Atrogin-1 , NF-κB2 , and Relb was normalized to SDHA , quantitation of gene expression was determined by the equation 2−ΔΔCT , and the fold-change of mRNA expression was calculated relative to the mRNA level of P0 FDS samples in each species , which was set to 1 . One-way ANOVA with Tukey’s multiple comparisons test was performed to determine the statistical significance of fold change differences between samples in each species . mouseMuRF1_FTGCCTGGAGATGTTTACCAAGC ( Dogra et al . , 2007 ) mouseMuRF1_RAAACGACCTCCAGACATGGACA ( Dogra et al . , 2007 ) mouseAtrogin_FTGGGTGTATCGGATGGAGAC ( Files et al . , 2012 ) mouseAtrogin_RTCAGCCTCTGCATGATGTTC ( Files et al . , 2012 ) jerboaMuRF1_FCCGCGTGCAGACTATCATCAjerboaMuRF1_RGCAGCTCGCTCTTTTTCTCGjerboaAtrogin_FGCATCGCCCAAAAGAACTTCAjerboaAtrogin_RACTTGCCGACTCTTTGGACCmouseSDHA_FGGAACACTCCAAAAACAGACCT ( Xu et al . , 2016 ) mouseSDHA_RCCACCACTGGGTATTGAGTAGAA ( Xu et al . , 2016 ) jerboaSDHA_FACTGGAGGTGGCATTTCTACjerboaSDHA_RTTTTCTAGCTCGACCACAGATGmouseNF-κB2_FGCCCAGCACAGAGGTGAAAGmouseNF-κB2_RCATTCAGTGCACCTGAGGCTmouseRelb_FTGTCACTAACGGTCTCCAGGACmouseRelb_RCAGGCGCGGCATCTCACTjerboaNF-κB2_FCTAGCCCACAGACATGGACAjerboaNF-κB2_RTAGGGGCCATCAGCTGTCTCjerboaRelb_FCCTACAATGCTGGCTCTCTGAjerboaRelb_RGTCATAGACAGGCTCGGACA	Intrinsic muscles are a group of muscles deep inside the hands and feet . They help to control the precise movements required , for example , for a pianist to play their instrument or for certain animals to climb with remarkable agility . Some animals , such as horses and deer , have evolved in such a way that they no longer grasp objects with hands and feet . Where intrinsic muscles were once present in the hands and feet of their ancestors , these animals now have strong ligaments that prevent over-extension of the wrist and ankle joints during hard landings . Given their size , it is difficult to study horses and deer in the laboratory and understand how they lost their intrinsic muscles during evolution . Tran et al . therefore focused on a small rodent called the lesser Egyptian jerboa , which also displays long legs with strong ligaments and no intrinsic muscles . Newborn jerboas have foot muscles that look very much like the intrinsic muscles found in mice , but these muscles disappear within 4 days of birth . A mechanism called programmed cell death is often responsible for specific tissues disappearing during development , but the experiments of Tran et al . revealed that this was not the case in jerboas . Instead , their intrinsic muscles were degraded by processes triggered by genes that disassemble underused muscles . In mice and humans , fasting , nerve injuries , or immobility trigger this type of muscle degradation , but in jerboas these processes appear to be a normal part of development . This unexpected discovery shows that development and disease-like processes are linked , and that more studies of nontraditional research animals may help scientists better understand these connections .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "developmental", "biology", "evolutionary", "biology" ]	2019	Evolutionary loss of foot muscle during development with characteristics of atrophy and no evidence of cell death
Mothers are crucial for mammals’ survival before nutritional independence , but many social mammals reside with their mothers long after . In these species the social adversity caused by maternal loss later in life can dramatically reduce fitness . However , in some human populations these negative consequences can be overcome by care from other group members . We investigated the consequences of maternal loss in mountain gorillas and found no discernible fitness costs to maternal loss through survival , age at first birth , or survival of first offspring through infancy . Social network analysis revealed that relationships with other group members , particularly dominant males and those close in age , strengthened following maternal loss . In contrast to most social mammals , where maternal loss causes considerable social adversity , in mountain gorillas , as in certain human populations , this may be buffered by relationships within cohesive social groups , breaking the link between maternal loss , increased social adversity , and decreased fitness . Maternal loss , along with a number of other indicators of early-life adversity , is one of the strongest predictors of lifespan in humans and other social mammals ( Snyder-Mackler et al . , 2020 ) . In mammals , mothers are vital for the survival of young offspring , providing nutrition , thermoregulation , and protection ( Clutton-Brock , 1991 ) . In some species , particularly social species with slow life histories ( Mitani et al . , 2013 ) , mothers continue to provide benefits to their co-residing offspring throughout immaturity and even into adulthood ( Surbeck et al . , 2019; Surbeck et al . , 2011; Andres et al . , 2013; Stanton et al . , 2020 ) . The active support of mothers can increase the rank of their offspring ( Strauss et al . , 2020; East et al . , 2009; Maestripieri and Mateo , 2009; Lea et al . , 2014 ) and improve their integration in the group ( Tung et al . , 2016 ) , both of which are linked with greater survival ( Snyder-Mackler et al . , 2020; Archie et al . , 2014 ) . Maternal presence can also influence nutrition at these later stages of development by buffering against feeding competition ( Samuni et al . , 2020 ) , providing access to valuable ecological knowledge ( Stanton et al . , 2020; Foley et al . , 2008; Brent et al . , 2015 ) or increasing opportunities for the social learning of complex feeding techniques ( Lonsdorf et al . , 2004; Estienne et al . , 2019 ) . In social mammals maternal loss can therefore reduce the fitness of offspring across a broad age range through long-term effects on their social environment , negatively influencing their social integration and social status throughout their lives ( Snyder-Mackler et al . , 2020; Strauss et al . , 2020; Tung et al . , 2016 ) . Multiple studies have now confirmed effects on survival for individuals orphaned well past the period of nutritional dependency , with these negative changes to their social environment , often termed social adversity , posited to be the key mechanism by which this occurs ( Andres et al . , 2013; Stanton et al . , 2020; Tung et al . , 2016; Watts et al . , 2009; Foster et al . , 2012 ) . In social mammals , the social environment can have extreme consequences for health , fitness , and lifespan , mediated through pathways such as chronic stress , immune function , or environmental exposure ( Snyder-Mackler et al . , 2020 ) . The impact of maternal loss varies based on the loss in benefits relative to individuals with mothers present . In species with sex-biased dispersal , the consequences of maternal loss can differ between the sexes due to longer periods for potential investment in non-dispersing offspring ( Clutton-Brock , 1991; Fairbanks , 2009; Altmann and Alberts , 2005; Greenwood , 1980 ) . In female-philopatric red deer ( Cervus elaphus ) , maternal loss increases mortality for males and females , but this effect is only detectable in males under 2 years of age , whilst for females , maternal benefits continue throughout their lives ( Andres et al . , 2013 ) . In male-philopatric chimpanzees ( Pan troglodytes ) , males that suffer maternal loss before reaching 15 years of age have lower survival , whilst females only show reduced survival from maternal loss under the age of 10 ( Nakamura et al . , 2014; Stanton et al . , 2020 ) . In killer whales ( Orcinus orca ) , where neither sex disperses , maternal loss when offspring are over 30 years old reduces survival for both sexes , although considerably more so for males ( Foster et al . , 2012 ) . However , maternal loss between 15 and 30 years appears to reduce male but not female survival . This is thought to be due to higher maternal investment in males which mate outside the group and whose offspring therefore do not increase within-group feeding competition . As a result of the numerous benefits mothers can provide , it is not surprising that maternal loss not only influences offspring survival but can also impact other components of their offspring’s fitness , such as reproduction and the survival of grand-offspring . Male bonobos ( Pan paniscus ) residing in groups with their mothers sire three times the number of offspring ( Surbeck et al . , 2019 ) , whilst maternal loss before weaning negatively affects antler development in male red deer – a trait found to correlate with reproductive success ( Andres et al . , 2013 ) . In chimpanzees , females mature faster , first give birth younger ( Walker et al . , 2018 ) and enter the dominance hierarchy higher ( Foerster et al . , 2016 ) if their mothers are present which is expected to considerably increase their lifetime reproductive success . In savannah baboons ( Papio cynocephalus ) , if mothers had themselves suffered maternal loss in the first 4 years of their life , their offspring had 48% higher mortality throughout the first 4 years of their life , suggesting an intergenerational effect of maternal loss driven by lifelong developmental constraints ( Zipple et al . , 2019 ) . Maternal loss in social mammals with extended maternal care can therefore have long-term fitness consequences mediated through multiple pathways that detrimentally affect survival and reproduction . Due to the extended periods of mother–offspring co-residence and the important social support mothers can provide , highly social species often have the most to lose from maternal loss . However , social groups also provide the potential for support from other group members following maternal loss . Both kin and non-kin group members have been suggested to compensate for the loss of close kin to varying extents ( Hamilton et al . , 1982; Engh et al . , 2006; Goldenberg and Wittemyer , 2017; Reddy and Mitani , 2019 ) and the strengthening of relationships with remaining group members may buffer against changing social environments ( Firth et al . , 2017 ) . In chacma baboons ( Papio ursinus ) , social support from group members is thought to alleviate the stress of losing a close relative ( Engh et al . , 2006 ) . Similar social support has been suggested in African savannah elephants ( Loxodonta africana ) which associate more with group members of a similar age and siblings in response to maternal loss ( Goldenberg and Wittemyer , 2017 ) . However , these orphaned elephants interact less with matriarchs which may decrease their access to key knowledge and high-quality resource patches . In chimpanzees , older siblings can ‘adopt’ younger siblings after maternal loss , increasing their social contact and showing heightened vigilance in dangerous situations ( Reddy and Mitani , 2019; Hobaiter et al . , 2014 ) . But despite these compensatory social behaviours , the negative consequences of maternal loss post-weaning are well documented in all three genera ( Stanton et al . , 2020; Tung et al . , 2016; Samuni et al . , 2020; Nakamura et al . , 2014; Goldenberg and Wittemyer , 2017; Goldenberg and Wittemyer , 2018 ) . Humans ( Homo sapiens ) are a rare example of a group-living mammal in which compensatory social behaviours have been suggested to have the capacity to consistently overcome these negative consequences of maternal loss . In a meta-analysis of historic and contemporary human populations , the death of a mother was associated with increased child mortality in all 28 populations studied ( Sear and Mace , 2008 ) . However , this effect appeared to decline substantially with age , disappearing for children that suffered maternal loss over 2 years of age in 5 of the 11 populations in which it was investigated ( Sear and Mace , 2008 ) . This reduced mortality was thought to be due to the care provided by other kin , particularly after weaning , suggesting that social buffering from other group members can overcome the negative effect of maternal loss on survival in certain circumstances . Whilst the effects of care from specific kin members varied across populations , at least one kin member significantly impacted child survival in all studies ( Sear and Mace , 2008 ) , and there is evidence for the importance of maternal grandmothers ( Lahdenperä et al . , 2004; Sear et al . , 2000 ) and fathers ( Hurtado and Hill , 1992; Hill and Hurtado , 2017 ) in particular . In killer whales , maternal grandmothers , especially those that are post-reproductive , are also known to improve grand-offspring survival ( Nattrass et al . , 2019 ) . Whilst the specific effect of killer whale grandmothers on orphan survival has not been investigated , this finding suggests that killer whales may represent a further species in which care from other kin has the capacity to overcome the effects of maternal loss . In humans , there is also evidence for the benefits of care provided by non-kin such as step-mothers ( Andersson et al . , 1996; Campbell and Lee , 2002 ) and through the modern practices of non-kin adoption ( Bentley and Mace , 2009 ) . Mountain gorillas ( Gorilla beringei beringei ) show extended maternal care with offspring remaining in their natal groups at least until sexual maturity and approximately half remaining beyond sexual maturity ( 48% of females [Robbins et al . , 2009a] and 55% of males [Stoinski et al . , 2009a] ) . Females that disperse from their natal group tend to do so earlier ( mean age of 7 . 9 years [Robbins et al . , 2009a] ) than males ( mean age of 15 . 3 years [Stoinski et al . , 2009a] ) and therefore have a shorter period of potential maternal investment . The complexity of gorilla social structure with numerous types of differentiated social relationship both within and among groups ( Morrison et al . , 2019; Mirville et al . , 2018; Morrison et al . , 2020a; Morrison et al . , 2020b ) suggests that detrimental long-term effects on individual gorillas’ social environments could have particularly negative fitness consequences . However , these stable , cohesive , social groups also have the potential to provide a social buffer to the negative consequences of maternal loss . Mountain gorilla groups either contain a single adult male ( approximately 64% of groups ) or multiple adult males ( approximately 36% of groups ) ( Gray , 2010 ) , at least one adult female , and their offspring ( Robbins , 1995 ) . Single male groups are polygynous whilst multimale groups have high reproductive skew towards the dominant male who sires the majority of offspring ( Nsubuga et al . , 2008; Stoinski , 2009b; Bradley et al . , 2005 ) . Infants ( <4 years of age ) are nutritionally dependent on their mothers until being weaned at a mean of 3 . 3 years ( Eckardt et al . , 2016 ) and are reliant on their mothers for thermoregulation and transport , being carried for prolonged periods ( Breuer et al . , 2009 ) . Juveniles ( 4–6 years old ) are nutritionally independent but remain in close proximity to their mothers the majority of the time ( Breuer et al . , 2009 ) . Dominant males’ primary form of care is through protection from out-group males and potential predators ( Harcourt and Greenberg , 2001 ) . But they also show high levels of affiliative behaviour towards infants , grooming and resting in contact with them , with no evidence that they discriminate between infants based on paternity ( Rosenbaum et al . , 2018; Rosenbaum et al . , 2015 ) . In this study , we use the long-term demographic records of the Dian Fossey Gorilla Fund’s Karisoke Research Center collected over 53 years ( 1967–2019 ) to ( A ) quantify the effects of maternal loss on multiple fitness measures: survival , female age at first birth , female survival of first offspring through infancy , and male dominance; and ( B ) investigate the social responses of group members to maternal loss by immature gorillas . We hypothesize that as demonstrated in chimpanzees ( Stanton et al . , 2020 ) , gorillas may face greater fitness costs if they suffer maternal loss at an earlier age and that males may face greater costs from maternal loss than females due to their longer periods of mother–offspring co-residence . Alternatively , as observed in many human populations , the cohesive , stable social groups of mountain gorillas may enable social buffering from group members to compensate for the social costs of maternal loss with minimal fitness consequences to maternal loss . In particular , dominant males may take on crucial roles in buffering the social adversity faced by maternal orphans ( hereafter , orphans ) , as past research has demonstrated the strong bond between dominant males and young orphans who may regularly share a nest at night ( Robbins et al . , 2005; Gatesire et al . , 2016 ) . To determine the effect of maternal loss on survival , we carried out a Cox-proportional hazards analysis separating individuals into four orphan classes based on their age when their mother died: ( a ) infants ( 2–4 years old ) , ( b ) juveniles ( 4–6 years old ) , ( c ) subadults ( 6–8 years old ) , and ( d ) non-orphans ( >8 years old ) if their mothers died after they had reached maturity . Due to the small sample sizes for the juvenile and subadult classes , analysis was also run with these two classes merged . We found no significant differences in survival between all orphan classes and the non-orphan class for both sexes irrespective of using three or four classes ( Table 1 , Figure 1 , Figure 1—figure supplement 1 , Supplementary file 1 -Table 1 ) . Bayesian survival trajectory analysis ( Colchero and Clark , 2012; Colchero et al . , 2012 ) showed similar results , whereby the model with highest support for both sexes was the null model without orphan classes as covariates ( Table 2 ) . We examined the effect of maternal loss on the likelihood of females dispersing before giving birth to their first offspring using a binomial generalized linear model ( n = 51 ) . Female orphans were not significantly more likely to disperse from their natal group prior to first birth than female non-orphans ( Table 3 ) . However , there was a close to significant increase in the likelihood of dispersal for females that lost their mothers as juveniles or subadults . 37 . 5% of non-orphan females ( n = 32 ) dispersed prior to their first birth ( mean dispersal age ± SD: 7 . 96 ± 1 . 55 years ) compared to 54 . 5% ( n = 11 ) of infant orphans ( dispersal age: 7 . 75 ± 0 . 60 years ) and 75 . 0% ( n = 8 ) of juvenile and subadult orphans ( dispersal age: 8 . 21 ± 2 . 36 years ) . We examined the effect of maternal loss on male dispersal based on whether a male had dispersed from their natal group prior to the age of 16 years ( the median age of male dispersal ) . Only three males that reached the age of 16 had lost their mothers as infants , but all three remained in their natal group . Due to this small sample size this was not examined statistically . However , a binomial generalized linear model demonstrated that juvenile and subadult orphan males were significantly more likely to disperse before reaching 16 years of age ( 84 . 6% , n = 13 ) than non-orphan males ( 37 . 5% , n = 40 , Table 3 ) . Using a generalized linear model , we found that maternal loss had no significant effect on the age at which females first gave birth ( Supplementary file 1 - Table 2 , n = 53 ) . The mean age at first birth ( ± SD ) for non-orphans was 10 . 24 ± 1 . 61 years compared to 9 . 72 ± 0 . 73 years for those orphaned as infants and 9 . 67 ± 1 . 82 years for those orphaned as juveniles or subadults . After accounting for age at first birth and dispersal , there was also no evidence that maternal loss influenced whether a female’s first offspring survived infancy ( Supplementary file 1 - Table 3 , n = 50 , binomial generalized linear model ) . 51 . 5% of non-orphan females’ first-born offspring ( n = 33 ) survived infancy compared to 60% ( n = 10 ) of first-born offspring of those orphaned as infants and 57 . 1% ( n = 7 ) of first-born offspring of those orphaned as juveniles or subadults . The mean age ( ± SD ) at which a male first became the dominant male of a group was 17 . 88 ( ± 2 . 56 ) years . The oldest that a male first reached dominance was 22 . 99 years , with all males that had not become dominant by this age , never reaching dominance . We therefore compared the proportion of males over the age of 23 that had attained dominance in each orphan class . 52% of non-orphan males had become the dominant male of a group for at least 6 months ( n = 21 ) by the age of 23 compared to all three infant-orphaned males and no juvenile- or subadult-orphaned males ( n = 5 ) . This finding suggested maternal loss as a juvenile or subadult male could limit a gorilla’s ability to become dominant . However , this could also be purely an artefact of the small sample sizes . We therefore examined the dominance status of the seven juvenile- or subadult-orphaned males that were over the age of 16 but had not yet reached 23 years by the end of the study period . 71% ( n = 7 ) of these males had already become dominant despite their younger age , indicating the capacity for males orphaned in any age category to become dominant later in life . The social responses of group members to 21 incidents of maternal loss were investigated ( Supplementary file 1 - Table 4 ) . Focal data collected daily in each group were used to construct social networks based on ( a ) 2 m proximity and ( b ) affiliative contact ( resting , playing , or feeding in physical contact and grooming , but excluding physical aggression ) for the 6 months leading up to a maternal loss incident and the 6 months immediately after a maternal loss incident . In the 6 months prior to maternal loss , orphans had spent a mean of 13% ( ± 8 , n = 31 ) of their time while monitored in affiliative contact with their mother and 39% ( ± 16 , n = 31 ) of their time within proximity ( <2 m ) of their mother , who was predominantly their closest social partner ( Table 4 ) . After maternal loss , orphan’s affiliative contact with other group members increased on average by 28% and the proportion of their time spent within 2 m of other group members increased on average by 42% . Overall , this resulted in a net decrease in orphan’s affiliative contact of 31% and a net increase in orphan’s proximity to others of 11% following maternal loss . To investigate how these changes influenced the social network position of orphans we compared orphan’s change in binary degree ( number of connections ) , weighted degree ( strength of connections ) , and eigenvector centrality ( how connected they were to other well-connected individuals ) with that of non-orphans within the same networks ( n = 136 ) . These networks included only individuals that were present both before and after the maternal loss incident ( excluding the mothers of orphans ) to enable direct comparison . Using generalized additive mixed models ( GAMMs ) with node-level permutations of orphan status we found that within proximity-based networks , orphans’ eigenvector centrality ( Est = 0 . 169 ± 0 . 037 , t = 4 . 594 , p < 0 . 001 , Pnull < 0 . 001 ) and weighted degree ( Est = 0 . 075 ± 0 . 032 , t = 2 . 337 , p=0 . 021 , Pnull = 0 . 024 ) increased significantly more after maternal loss than non-orphans within the same pair of networks . This led to orphans and non-orphans having similar weighted degree and centrality values following maternal loss despite orphans losing a key social partner ( Figure 2 , Supplementary file 1 - Table 5 ) . However , within contact-based networks orphans did not show these same gains , with no significant change in eigenvector centrality ( Est = −0 . 034 ± 0 . 057 , t = −0 . 593 , p = 0 . 554 , Pnull = 0 . 598 ) or weighted degree ( Est = 0 . 031 ± 0 . 047 , t = 0 . 656 , p = 0 . 513 , Pnull = 0 . 442 ) relative to non-orphans ( Figure 2—figure supplement 1 , Supplementary file 1 - Table 5 ) . Binary degree did not increase to a greater extent in orphans than non-orphans in either contact or proximity-based networks ( Supplementary file 1 - Table 5 ) . GAMMs were also used to investigate changes in individual pairwise relationships pre- and post-maternal loss . Our first pair of models included all pairwise relationships involving an immature gorilla ( both orphans and non-orphans ) . They demonstrated that affiliative contact with dominant males , subordinate males , and subadult females increased to a greater extent for orphans than other immature gorillas within the same group during the same time period ( n = 3486 , Figure 3; Supplementary file 1-Table 6 ) . Based on proximity , orphan’s relationships with dominant males , subordinate males , adult females , subadult females , and juveniles strengthened more than those between other immature gorillas and the same age-sex classes of group members ( n = 3486 , Figure 3; Supplementary file 1 Table 6 ) . Our second pair of models examined only pairwise relationships involving an orphan ( n = 755 ) to provide more detailed information on how orphans relationships changed . The extent to which an orphan’s affiliative contact with and proximity to other group members increased following maternal loss did not differ depending on the orphan’s sex ( Table 5 ) . The increase in proximity with other group members following maternal loss was smaller for older orphans but this difference was not significant for affiliative contact ( Table 5 ) . This suggests that social support after maternal loss through proximity with other group members is lower for older orphans who may already have been less reliant on their mothers . Age-mates ( those within 2 years age of the orphan ) showed a greater increase in proximity after maternal loss relative to other group members . However , this was not the case for affiliative contact ( Table 5 ) . The change in relationship strength between maternal siblings ( hereafter , siblings ) after maternal loss depended on the age-sex class of the sibling ( Figure 3—figure supplement 1 ) . Subordinate adult males and subadult females had more affiliative contact with younger siblings following maternal loss but siblings in all other age-sex classes did not ( Table 5 ) . Both forms of social support ( affiliative contact and proximity ) showed the greatest increase from dominant males ( Figure 3 , Table 5 ) . For affiliative contact this was significantly greater than all other age-sex classes , but for proximity the increase was only significantly greater than that of subordinate adult males . For 67% of orphans of known paternity ( n = 18 ) , the dominant male at the time of maternal loss was their genetic father . Paternity did not influence the social support provided by adult males after maternal loss ( contact: z = 1 . 130 , p=0 . 262; proximity: z = −0 . 552 , p = 0 . 583 ) but adult male maternal siblings increased both their affiliative contact and proximity more than non-siblings ( contact: z = 3 . 807 , p < 0 . 001; proximity: z = 2 . 237 , p = 0 . 028 ) . However , the increased social support from adult male maternal siblings relative to non-siblings was largely driven by an effect in subordinate males , whilst social support from dominant males did not differ greatly by kin relationship ( Supplementary file 1 - Table 7 , Figure 3—figure supplement 2 ) . We found that immature mountain gorillas do not appear to face increased social adversity or a detectable reduction in fitness following maternal loss . It is not yet possible to demonstrate the direct link between the strengthening of relationships with other group members after maternal loss and the absence of fitness costs to maternal loss . However , our analyses show that at least in the short-term , a key mechanism by which maternal loss is hypothesized to lead to reduced survival and fitness in other social species – social adversity - does not apply in mountain gorillas over the age of 2 years . The social support provided by other group members within mountain gorillas’ cohesive social groups , particularly from dominant males , siblings , and those close in age , appears to buffer against the negative consequences of maternal loss . In mountain gorillas , like humans ( Sear and Mace , 2008; Lahdenperä et al . , 2004; Sear et al . , 2000; Hurtado and Hill , 1992; Andersson et al . , 1996; Campbell and Lee , 2002; Bentley and Mace , 2009 ) social support appears to come from a number of group or family members . This could provide a buffer to the loss of any single relationship , even one as important as the mother–offspring relationship , once an individual can be nutritionally independent . In the absence of nepotistic matriarchal dominance hierarchies and when social buffering is possible due to cohesive strongly bonded social groups , it may matter less who is providing care as long as care is provided . Mountain gorillas in the Volcanoes National Park , Rwanda , have been monitored almost continuously by the Dian Fossey Gorilla Fund’s Karisoke Research Center since 1967 . Habituated mountain gorilla groups are monitored daily by field teams who collect data on demography , behaviour , ranging , and health . From 1967 to 2015 ( inclusive ) , 59 ( 28 males , 31 females ) out of the total 200 immature mountain gorillas ( 102 males , 98 females ) that reached the age of at least 2 years suffered maternal loss between the ages of 2 and 8 through the death or permanent transfer of their mother . Gorillas were classified as infants up to 4 years of age , as juveniles from 4 to 6 years of age and as subadults from 6 to 8 years of age ( Breuer et al . , 2009 ) . From 8 years of age females were classified as adults . Males were classified as blackbacks from 8 to 12 years . From 12 years , males were classified as either subordinate adult males or dominant adult males from their dominance hierarchy . Male dominance hierarchies were based on displacements and avoidances using the Elo-rating method ( Albers and de Vries , 2001; Neumann et al . , 2011 ) . Dominance hierarchies were calculated using the R package EloRating , version 0 . 43 ( Neumann and Lars , 2014 ) as described by Wright et al . , 2019 . Only one adult male was classified as dominant in each group at a given time . Dominant males were those with the highest dominance status unless they were the only adult male in the group in which case they were automatically classified as dominant . The youngest infant to survive maternal loss was a 2 . 45-year-old female . Our data set included only one infant younger than this that suffered maternal loss at 0 . 67 years and died after 1 day . Three infants aged 1 . 91 , 2 . 42 , and 2 . 52 became separated from their mothers after a suspected poacher encounter . During this separation they travelled with a small number of group members not including their mothers . The 1 . 91-year-old died after 6 days of separation . The 2 . 42-year-old died after 9 days of separation . The 2 . 52-year-old survived until they were reunited with their mother and the larger group 18 days after the initial separation . These infants were not considered as orphans in the data set as their mothers did not permanently transfer or die , but in combination with those that did , they suggest that infant mountain gorillas cannot survive independently from their mothers under the age of at least 2 . We therefore investigated the effect of maternal loss after the age of 2 . To determine the effect of maternal loss on survival , we carried out a Cox-proportional hazards analysis separating individuals over the age of 2 , based on four general age classes of maternal loss: ( a ) infants , ( b ) juveniles , ( c ) subadults , and ( d ) non-orphans , where mothers did not die or leave the group before the individual reached 8 years of age . First , we ran a Cox-proportional hazards model for each sex ( males: n = 102 , females: n = 98 ) with a time-varying covariate for the age at maternal loss and using the four classes as covariates . Due to the small sample size for the subadult class , we merged this with the juvenile class into a single juvenile/subadult class and ran a new set of Cox-proportional hazards models on these new classes . In all cases we truncated the analysis to start at age 2 . In order to verify our results and to account for the uncertainty in some of the dates of birth , we ran a Bayesian survival trajectory analysis ( Colchero and Clark , 2012; Colchero et al . , 2012 ) for each sex truncated at the age of maternal loss for all orphans , and at 2 years of age for non-orphans . We used orphan status as a binary covariate ( orphan vs non-orphan ) and , using the Siler , 1979 mortality model for the baseline mortality , we tested three models: ( a ) no covariates ( i . e . null model where all individuals have the same hazard rate ) ; ( b ) proportional hazards ( where mortality differs proportionally between orphan classes ) ; ( c ) covariates modifying all Siler mortality parameters ( where each orphan class has a different age-specific mortality ) . We used deviance information criterion for model fit and selection ( Spiegelhalter et al . , 2002; Celeux et al . , 2006 ) . Model ( b ) is equivalent to the Cox-proportional hazards model . However , these tests facilitate further exploration of the hypotheses on the effect of maternal loss on mortality , namely that there is no effect ( model a ) or that the entire age-specific trajectory of mortality changes for each category . Between 1967 and 2019 , 66 females gave birth to what was known to be their first offspring . For 53 of these females , their age could be accurately estimated within a 90-day period . For these individuals , we extracted their age at first birth , whether they dispersed from their natal group prior to first birth and whether they had suffered maternal loss when immature , from the long-term database . Maternal loss was investigated with the orphan age classes described above with juvenile and subadult classes merged due to small sample sizes . We investigated the effect of maternal loss on the decision of females to disperse from their natal group prior to their first birth using a binomial generalized linear model . We investigated the effect of both maternal loss and dispersal prior to first birth on age at first birth using a generalized linear model with a Gaussian distribution . Due to the positive skew of age at first birth , we used the square root of age at first birth minus 8 ( the earliest recorded age at first birth ) as the response variable . We checked Q–Q plots to verify the normal distribution of residuals and ran Levene tests using the ‘rstatix’ package to check for heteroskedasticity . Finally , we examined survival of each female’s first offspring through infancy ( 1: survived to 4 years , 0: died before reaching 4 years ) according to age at first birth , dispersal prior to first birth ( 1: yes , 0: no ) , and maternal loss , with merged juvenile and subadult classes as above , using a binomial generalized linear model . Multicollinearity of all models with multiple variables was checked using variance inflation factors in the ‘car’ R package . Between 1967 and 2019 , 56 males whose age could be accurately estimated within a 90-day period reached the median age of dispersal ( 16 years ) . We used a binomial generalized linear model to predict whether each of these males dispersed from their natal group prior to this age according to orphan age classes ( as above ) . To investigate the effect of maternal loss on dominance , we gave males that became the dominant male of a stable group for at least six consecutive months a dominance score of 1 . This included adult males of single-male groups and the most dominant male of multi-male groups based on Elo-ratings . Males that never reached dominance or only transiently ( for <6 months ) received a score of 0 . We recorded the age at which a male first became dominant for those for which this could be accurately estimated within a 90-day period . This represented the age at which they first successfully attracted and retained a female to join their group , the age at which they split from their natal group with at least one adult female to form a new group , or the age at which their Elo-rating surpassed that of all other adult males in their group , if those groups remained independent and did not disintegrate within 6 months of that date . The mean age ( ± standard deviation ) at which a male reached dominance was 17 . 88 ± 2 . 56 years . The median was 17 . 29 years . The oldest age at which a male first became dominant was 22 . 99 years . Therefore , to investigate the influence of maternal loss on dominance status we analysed only males that had survived and remained in the study population until at least 23 years ( n = 40 ) . We did not attempt to statistically examine the effects of maternal loss on dominance status due to small sample sizes . Habituated gorilla groups were monitored for up to 4 hrs daily and all gorillas were individually identified by physical characteristics . Behavioural data were collected on each group member via 50 min focal sampling during which the researcher would typically be within 10–20 m of the focal individual . Researchers systematically worked their way through a randomly ordered list of all individuals in the group . If an individual could not be observed ( e . g . obscured by dense vegetation ) , the researcher moved on to the next individual on the list and returned to them the subsequent day . During focal sampling , a focal scan was completed every 10 min which recorded all gorillas within 2 m of the focal individual and all gorillas in physical contact with the focal individual . This data was therefore in the form of frequencies ( the number of focal scans during which individuals were associating ) rather than durations . Affiliative contact included resting , playing , or feeding in contact and grooming , and excluded physical aggression . This focal sampling approach limited the extent of potential sampling bias due to individual-level differences in observation propensity , with all group members systematically observed and an extremely low likelihood of individuals within 2 m or in physical contact of the focal individual being missed at these close proximities . The social response of group members to an incident of maternal loss was investigated in 31 of the 59 total cases – those that suffered maternal loss after 2003 for which adequate social behaviour data was available ( more than 12 focal scans of the individual were recorded in the 6 months prior to maternal loss and the 6 months after maternal loss , Supplementary file 1 - Table 3 ) . For each case of maternal loss , two types of weighted social network were constructed based on ( a ) 2 m proximity and ( b ) affiliative contact . For each type , edge values of the networks were calculated using the Simple Ratio Index ( SRI ) ( Whitehead , 2008 ) with edges between a pair of individuals calculated as the proportion of focal scans of either individual during which the pair was recorded as associating . These values represented an estimate of the proportion of time two individuals were either within 2 m of each other or in physical contact . For example , in the contact network a value of 1 would indicate that the two individuals were in physical contact every time a focal scan of either individual was conducted , whilst 0 would indicate that they were never observed in physical contact during a focal scan . Both network types were constructed for two time periods for each case of maternal loss: pre-maternal loss ( the 6 months leading up to maternal loss ) and post-maternal loss ( the 6 months immediately after maternal loss ) . Social networks were constructed using all focal scans during these time periods . Only gorillas for which more than 12 focal scans were available in both of the 6 month periods ( pre- and post-maternal loss ) were included in the networks ( Farine and Whitehead , 2015 ) . This meant that only the social relationships with group members that were present both pre- and post-maternal loss were analysed , except for the mother–offspring relationships pre-maternal loss which were extracted separately . The mean ( ± SD ) number of focal scans used to estimate edge values pre-maternal loss was 144 . 81 ± 90 . 10 and post-maternal loss was 149 . 16 ± 88 . 56 . Binary degree , weighted degree , and eigenvector centrality in the pre- and post-maternal loss networks ( excluding orphan’s mothers ) were calculated using the ‘igraph’ package for both network types ( contact and proximity ) . To enable comparison across networks of the same type , binary degree and weighted degree metrics were calculated as a proportion of the maximum value for an individual within the network ( as is already the case for eigenvector centrality ) . This meant that for all networks , each metric could have a maximum value of 1 and a minimum value of 0 . Metrics were extracted for all immature gorillas that were between the ages of 2 and 8 years on the date that an immature individual within their group suffered maternal loss . We then calculated the change in these network metrics between time periods for all immature gorillas in each set of group networks and used GAMMs in the ‘gamm4’ R package to determine whether orphan’s network metrics changed differently to those of non-orphans within the same group during the same time period . To ensure significant changes were not driven by unusually high or low initial values , the deviance of the network metric in the initial network from the mean value for immatures in the network was calculated . Orphan status , age , and the deviance of the initial value from the group mean were included in the model as fixed factors . The number of focal scans per individual across both time periods was included as a smoothing factor to account for any potential differences driven by sampling effort . The specific set of group networks ( the pair of networks for each incident of maternal loss ) was included as a random effect to account for differences in network composition ( Supplementary file 1 - Table 5 ) . Due to the non-independence of network metrics , null models generated through node-level permutations were used to assess the significance of orphan status on the change in network metric . Permutations were run by swapping orphan status between immature gorillas ( orphans and non-orphans ) within the same set of paired group networks ( same maternal loss incident ) . Orphans that suffered maternal loss in groups with no other non-orphaned immature gorillas were excluded from the analyses , leaving a sample of 19 incidents of maternal loss , 28 orphans , and 108 non-orphans ( Supplementary file 1 - Table 3 ) . The mean age ± SD of orphans was 5 . 12 ± 1 . 49 years and for non-orphans was 4 . 71 ± 1 . 71 years . 10 , 000 sets of node-permutations were generated by permuting orphan status between immature gorillas in the same network and extracting node labels every 200th permutation . The same GAMMs were then run on all 10 , 000 sets of node permutations to produce a null distribution of t-values for the effect of orphan status . Pnull was calculated using a two-tailed approach . For observed t-values greater than the median of the null distribution , Pnull was calculated as:2×numberofnullt-valuesgreaterthanobservedt-valuetotalnullt-values For observed t-values lower than the median of the null distribution , Pnull was calculated as:2×numberofnullt-valueslowerthanobservedt-valuetotalnullt-values We extracted SRI edge values ( representing the strength of relationship between a pair of gorillas ) between the orphan and all other group members pre- and post-maternal loss . We also extracted SRI edge values between all immature gorillas ( aged 2–8 years ) within the same group that had not suffered maternal loss and all other group members , for the same time periods . We then calculated the change in relationship strength between an immature gorilla ( both orphans and non-orphans ) and other group members following an incident of maternal loss within the group as the change in SRI value between periods ( SRI post-maternal loss – SRI pre-maternal loss ) in both network types ( contact and proximity ) . GAMMs were used to predict this change for both affiliative contact and proximity which enabled the non-independence of relationships involving the same individual to be accounted for through random effect structures and sampling variation to be accounted for using a smoothing term . We ran an initial set of GAMMs on the change in SRI values for relationships involving the orphans or other immature gorillas within the same group that had not suffered maternal loss , excluding mother–offspring relationships . This was to verify that changes in the relationships of orphans with other group members were significantly different to those observed for other immature gorillas present within the same group during that period . These GAMMs included the identity of the immature gorilla as a random factor , nested within the specific maternal loss incident ( group and time period ) , nested within the group . The identity of the other group member was included as a further random factor ( Supplementary file 1 - Table 6 ) . The mean number of focal scans used to estimate the SRI value across both time periods was included as a smoothing term in the model to account for any biases from sampling intensity . The age and sex of the immature gorilla were included as fixed factors , along with the age-sex class of the other group member , whether the immature gorilla suffered maternal loss during this period ( 1: Yes , 0: No ) , and the interaction between maternal loss and the age-sex class of the other group member . Non-orphan relationships where the other group member was an orphan were excluded from the analysis . We then ran GAMMs on only the relationships involving orphans to investigate in more detail how these changed following maternal loss . The identity of the orphan was included as a random factor , nested within the group . The identity of the other group member was included as a further random factor ( Table 5 ) . The mean number of focal scans used to estimate the SRI value across both time periods was again included as a smoothing term . The age and sex of the orphan were included as fixed factors to predict the social response of group members , as well as whether those group members were maternal siblings ( 1: Yes , 0: No ) and whether those group members were age-mates ( 1: <2 years age difference with the orphan , 0: ≥2 years age difference ) . This 2-year cut-off was chosen to be consistent with the width of age categories , such that age-mates would be within the same age class during large proportions of their immature life . The age-sex class of the other group member at the date of maternal loss and the interaction between this and maternal sibling status were also included as fixed factors for predicting the change in relationship . Paternity was known for 18 of the 31 orphans for which social data was available , from a previous study ( Vigilant et al . , 2015 ) . To investigate the influence of paternity , we ran an additional set of GAMMs to predict the change in relationship ( both contact and proximity ) between adult males and orphans following maternal loss ( n = 121 ) . The identity of the orphan and the identity of the adult male were included as random factors ( Supplementary file 1 - Table 7 ) . The mean number of focal scans was included as a smoothing term . Orphan age , orphan sex , male dominance , and the binary kinship variables: maternal sibling and paternity were included as fixed factors . The interaction between dominance and sibling status was also included in the model , but the interaction between dominance and paternity could not be , due to possible issues of multicollinearity ( variance inflation factors > 3 ) .	Most mammals depend entirely upon their mothers when they are born . In these species , losing a mother at a young age has dramatic consequences for survival . In cases where orphaned individuals do reach adulthood , they often suffer negative effects , like reduced reproductive success or lower social status . But this is not the case for humans . If a child loses their mother , relatives , friends and the wider community can take over . This does not tend to happen in nature . Even our closest relatives , chimpanzees , are much less likely to survive if their mothers die before they reach adolescence . Although orphan survival is not the norm for mammals , humans may not be entirely unique . Mountain gorillas also live in stable family groups , usually with a dominant male and one or more females who care for their offspring for between 8 and 15 years . It is possible that gorillas may also be able to provide community support to orphans , which could buffer the costs of losing a mother , just as it does in humans . To answer this question , Morrison et al . examined 53 years of data collected by the Dian Fossey Gorilla Fund to assess the effects of maternal loss in mountain gorillas . The analysis examined survival , reproduction and changes in social relationships . This revealed that , like humans , young gorillas that lose their mothers are not at a greater risk of dying . There is also no clear long-term effect on their ability to reproduce . In fact , gorillas who lost their mothers ended up with stronger social relationships , especially with the dominant male of the group and young gorillas around the same age . It seems that gorilla social groups , like human families , provide support to young group members that lose their mothers . These findings suggest that the human ability to care for others in times of need may not be unique . It is possible that the tendency to care for orphaned young has its origins in our evolutionary past . Understanding this in more depth could provide clues into the social mechanisms that help to overcome early life adversity , and have a positive impact on future health and survival .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "ecology", "evolutionary", "biology" ]	2021	Social groups buffer maternal loss in mountain gorillas
Host factors restricting the transmission of respiratory viruses are poorly characterized . We analyzed the contribution of type I and type III interferon ( IFN ) using a mouse model in which the virus is selectively administered to the upper airways , mimicking a natural respiratory virus infection . Mice lacking functional IFN-λ receptors ( Ifnlr1−/− ) no longer restricted virus dissemination from the upper airways to the lungs . Ifnlr1−/− mice shed significantly more infectious virus particles via the nostrils and transmitted the virus much more efficiently to naïve contacts compared with wild-type mice or mice lacking functional type I IFN receptors . Prophylactic treatment with IFN-α or IFN-λ inhibited initial virus replication in all parts of the respiratory tract , but only IFN-λ conferred long-lasting antiviral protection in the upper airways and blocked virus transmission . Thus , IFN-λ has a decisive and non-redundant function in the upper airways that greatly limits transmission of respiratory viruses to naïve contacts . Influenza and other respiratory viruses are readily transmitted in community settings . The resulting infection chains and concomitant diseases represent an enormous economic and public health burden ( Osterhaus et al . , 2015 ) . At present , it is largely unknown if and how the innate immune response influences virus transmission efficacy . However , a better understanding of this issue is likely to lead to new therapeutic strategies that reduce viral transmission from infected individuals to naïve contacts . Type I and type III interferon ( IFN ) are virus-induced cytokines that potently restrict viral replication during the first days of infection before activation of the adaptive immune system occurs ( Lazear et al . , 2015; Wack et al . , 2015 ) . The type I IFN family consists of several IFN-α subtypes , a single IFN-β and several minor family members that all bind to and act via the IFN-α/β receptor complex ( IFNAR ) , which is expressed on most nucleated cells ( Lazear et al . , 2015; Wack et al . , 2015 ) with the possible exception of intestinal epithelial cells ( Lin et al . , 2016; Mahlakõiv et al . , 2015 ) . The members of the type III IFN family ( IFN-λ ) bind to a different receptor complex ( IFN-λ receptor; IFNLR ) , which is highly expressed on epithelial cells ( Kotenko et al . , 2003; Lazear et al . , 2015; Sheppard et al . , 2003; Sommereyns et al . , 2008; Wack et al . , 2015 ) and neutrophils ( Blazek et al . , 2015; Broggi et al . , 2017; Espinosa et al . , 2017; Galani et al . , 2017 ) . Transmission of influenza and other respiratory viruses is traditionally studied in ferrets and occasionally in guinea pigs ( Bouvier , 2015 ) . In contrast to mice , these animals are not readily accessible to genetic manipulation , and the role of IFN and IFN-regulated host restriction factors is not easily dissected . A recent report provided evidence that the mouse represents a good alternative animal model for studying influenza virus transmission , as long as suitable virus strains are employed ( Edenborough et al . , 2012; Ivinson et al . , 2017 ) . Influenza virus replication is strongly restricted by IFN-regulated Mx genes ( Haller et al . , 2015 ) . Since most standard inbred mouse strains carry defective Mx alleles ( Haller et al . , 2015 ) , the full magnitude of influenza virus restriction by the IFN system only becomes apparent when mouse strains carrying functional Mx1 alleles derived from wild mice are employed ( Haller et al . , 2015 ) . We previously introduced defective alleles of the type I IFN receptor α chain ( Ifnar1−/− ) or the IFN-λ receptor one chain ( Ifnlr1−/− ) into Mx1-competent C57BL/6 mice and thereby generated a set of mouse lines that carry functional Mx1 alleles but differ in their ability to respond to type I IFN and IFN-λ ( Mordstein et al . , 2008 ) . Using these knockout mouse strains we demonstrated that type I IFN plays a prominent role in the defense against respiratory viruses , including influenza viruses , respiratory syncytial virus or human metapneumovirus , whereas protection mediated by IFN-λ was surprisingly small ( Mordstein et al . , 2010b ) . In contrast , the protective role of IFN-λ during viral infections of the small intestine was found to be far more pronounced than that of type I IFN ( Baldridge et al . , 2017; Baldridge et al . , 2015; Mahlakõiv et al . , 2015; Nice et al . , 2015; Pott et al . , 2011 ) . This suggested either that ( i ) the gut and the respiratory tract use fundamentally different antiviral defense strategies or ( ii ) IFN-λ is important in both organ systems but its beneficial role in the respiratory tract cannot be demonstrated easily due to experimental limitations . Recent work with influenza virus-infected mice ( Galani et al . , 2017 ) indicated that IFN-λ is produced more quickly than type I IFN , suggesting that IFN-λ plays a non-redundant role in suppressing early virus growth in the respiratory tract . Nevertheless , protective effects of IFN-λ were observed only if very low infection doses were used , presumably because mice with defective Mx1 alleles were employed for these studies . Furthermore , the important question whether IFN-λ might play a role in restricting transmission of respiratory viruses from infected animals to naïve contacts was not addressed . According to standard protocols for experimental influenza virus infections , the virus is delivered intranasally to anesthetized mice in a 40–50 µl volume , allowing immediate infection of the entire respiratory tract . This experimental setup might override the natural barrier function of the nasal mucosa by allowing the infection to artificially initiate in the lower respiratory tract ( Ivinson et al . , 2017 ) . In this study , we applied an infection protocol that guarantees a selective delivery of the virus inoculum to the upper respiratory tract , thus more closely mimicking the natural course of infection . Under such experimental conditions , virus transmission to naïve contacts as well as virus spread from the upper airways to the lower respiratory tract was inhibited far more efficiently by IFN-λ than by type I IFN . Interestingly , the antiviral effect induced by type I IFN in the upper airways was more transient than that of IFN-λ , while no such difference was observed in the lungs . Furthermore , a small number of epithelial cells in the proximal upper respiratory tract failed to respond sufficiently to type I IFN but did respond to IFN-λ . These observations provide a mechanistic explanation for the superior role of IFN-λ in the upper airways . Most previous studies failed to assign a prominent role to IFN-λ in the resistance of mice against influenza and other respiratory viruses ( Galani et al . , 2017; Mordstein et al . , 2008 ) . Similarly , when we applied 104 PFU of the H7N7 influenza A virus strain SC35M intranasally in a standard 40 µl volume , we noted that Mx1-competent animals carrying a defective IFN-λ receptor ( Ifnlr1−/− ) supported similar virus replication in the lungs on day three post infection , compared with Mx1-competent wild-type ( WT ) animals . In contrast , Mx1-competent mice lacking functional type I IFN receptors ( Ifnar1−/− ) showed significantly elevated viral titers under such experimental conditions ( Figure 1A ) . Respiratory viruses enter the human body via the upper respiratory tract , and the majority of acute respiratory viral infections in humans are confined to the upper airways ( Cotton et al . , 2008 ) . To mimic a natural infection in which virus replication initiates in the upper airways , we established an upper respiratory tract infection model in mice . Using poliovirus that is unable to infect mouse cells as an indicator we found that virus delivery can be targeted specifically to the upper respiratory tract of mice if the inoculum is applied in a volume of 10 µl ( Figure 1—figure supplement 1 ) . In contrast , a large proportion of the inoculum reached the lungs if the inoculum was applied in a volume of 40 µl . When a 10 µl volume was used to deliver influenza virus strain SC35M specifically to the upper respiratory tract of WT mice , the virus grew well in the upper airways , but was absent or present at only very low titers in the tracheae ( 3 of 22 animals ) and the lungs ( 1 of 22 animals ) on day five post-infection ( Figure 1B ) . In Ifnlr1−/− mice , viral titers in the upper airways were significantly higher than in WT mice and , most interestingly , many infected Ifnlr1−/− mice contained high virus levels in the tracheae ( 11/23 ) and lungs ( 13/23 ) ( Figure 1B ) . As expected , Ifnar1−/− mice also had significantly higher virus titers in the upper airways than WT mice , and more animals contained virus in the tracheae ( 5/23 ) and lungs ( 9/23 ) ( Figure 1B ) . In comparison , Ifnar1−/− mice seemed to control the virus spread from the upper airways to the lungs slightly better than Ifnlr1−/− mice ( Figure 1B ) , indicating that IFN-λ is more potent in containing viral infections within the upper respiratory tract than type I IFN . To determine whether this unexpected phenotype of Ifnlr1−/− mice is restricted to the SC35M virus , we performed similar infection experiments with the H3N2 influenza A virus strain Udorn . After selective delivery to the upper respiratory tract , Udorn virus was rarely found in the tracheae ( 1/16 ) or lungs ( 0/16 ) of WT mice , but was frequently present in the tracheae ( 15/16 ) and lungs ( 14/16 ) of Ifnlr1−/− mice ( Figure 1C ) . Ifnar1−/− mice showed an intermediate phenotype under these experimental conditions , and infectious virus was found at lower frequency in the tracheae ( 8/15 ) and lungs ( 9/15 ) of Ifnar1−/− mice compared with Ifnlr1−/− mice . A similar picture emerged when a murine respiratory virus ( Sendai virus; SeV ) was employed in infection experiments . SeV reached the lungs of all ( 15/15 ) Ifnlr1−/− mice when selectively applied to the upper respiratory tract ( Figure 1D ) . In contrast , SeV reached the lungs of WT mice at significantly reduced frequency ( 6/15 ) under these experimental conditions , and virus titers at day five post-infection in the lungs of WT mice were generally lower compared with Ifnlr1−/− mice . Taken together , these data clearly indicated that IFN-λ is an indispensable antiviral cytokine that restricts respiratory viral infections to the upper airways and limits virus spread to the lungs . To better characterize the early events of the antiviral defense in the upper airways , we measured the baseline expression levels of various IFN genes . Expression of the IFN-λ2 and IFN-λ3 genes was significantly reduced in snout tissue of Ifnar1−/− mice compared with WT and Ifnlr1−/− mice , whereas baseline expression of IFN-α and IFN-β genes was comparable ( Figure 2A ) . Consequently , baseline expression of the Mx1 gene was also lower in snout tissue of Ifnar1−/− mice than in WT and Ifnlr1−/− mice ( Figure 2A ) . Upon infection with Udorn virus , expression of the IFN-λ genes in Ifnar1−/− mice returned to comparable levels of WT and Ifnlr1−/− mice within two days ( Figure 2B ) . Interestingly , expression of the IFN-λ genes in snout homogenates of WT and Ifnlr1−/− mice was only slightly induced after virus infection , whereas expression of the IFN-β gene was induced about tenfold in all mouse strains ( Figure 2B ) . These data indicate that functional type I IFN signaling is needed for proper baseline expression of the IFN-λ genes , which contributes to early virus defense in the upper airways . When the infection was initiated by applying the inoculum selectively to the upper airways , none of the influenza viruses used in this study caused morbidity or mortality , although virus titers in lungs of Ifnlr1−/− mice reached levels above 106 PFU in some animals on day five post-infection ( Figure 1B ) . These observations are in good agreement with recent work indicating that upper airway infections with influenza viruses are usually benign in mice presumably because saliva components delay the spread of influenza viruses to the lungs ( Ivinson et al . , 2017 ) . Delayed virus arrival in the lungs may permit timely adaptive immune responses to eliminate the virus before it causes irreversible tissue damage . We concluded from these observations that the upper airway infection model is not suitable for studying influenza virus-induced pathology in mice , but it allows addressing the question which factors might control virus growth at the entry site . The efficacy of respiratory virus transmission depends , among other parameters , on the production of mucosal secretions or aerosols containing a sufficiently high number of infectious particles ( Herfst et al . , 2017 ) . To determine whether nasal excretions of Ifnlr1−/− mice contain more infectious virus than excretions of WT mice , we sampled the nostrils of Udorn-infected mice with wet cotton swabs . This analysis revealed that Ifnlr1−/− mice shed significantly more infectious virus than WT or Ifnar1−/− mice ( Figure 3A ) . Already at 12 hr post infection , Ifnlr1−/− mice secreted remarkable amounts of infectious virus . Interestingly , virus shedding was only transient and peaked between 24 and 36 hr post infection ( Figure 3A ) . When the nostrils of SeV-infected mice were swabbed ( Figure 3B ) , we found at least 10-fold more infectious SeV in mucosal secretions of Ifnlr1−/− mice compared with WT controls between days 2 and 4 post infection . Levels of infectious SeV in mucosal secretions of Ifnar1−/− mice were only slightly increased compared with WT mice . Importantly , Ifnar1−/− mice shed significantly less virus than Ifnlr1−/− mice at all time points between days 2 and 4 post infection ( Figure 3B ) , indicating that IFN-λ is the dominant limiting factor for virus shedding . Next we evaluated whether increased virus shedding by Ifnlr1−/− mice would translate into enhanced spread of respiratory viruses among mice . We set up transmission experiments in which infected mice were cohoused with highly susceptible sentinel mice that lack functional receptors for type I IFN and IFN-λ ( Ifnar1−/− Ifnlr1−/− ) . In a first series of experiments , we infected WT , Ifnar1−/− or Ifnlr1−/− mice with 105 PFU of Udorn . To mimic natural virus transmission in a family setting , groups of three infected mice were cohoused with four sentinels in a single cage for a period of four days . Viral titers were then determined in the upper airways of the sentinels . Under such experimental conditions , 79% of the sentinels that were in contact with infected Ifnlr1−/− mice became infected ( Figure 4A ) . As expected from the virus excretion data ( Figure 3A ) , transmission of Udorn to sentinel mice was observed much less frequently ( 17% and 42% , respectively ) when infected WT or Ifnar1−/− mice were employed as virus spreaders ( Figure 4A ) . In a second experiment we infected WT , Ifnar1−/− or Ifnlr1−/− mice with 105 PFU of a different H3N2 influenza A virus strain ( HK68 ) before cohousing with Ifnar1−/− Ifnlr1−/− double-deficient sentinels . Under these experimental conditions , 10 of 11 ( 91% ) sentinels in contact with infected Ifnlr1−/− mice contracted the virus , whereas only 27% of the sentinels cohoused either with infected WT or Ifnar1−/− mice became infected ( Figure 4B ) . To determine whether transmission of other respiratory viruses might also be restricted by IFN-λ , we studied mouse-to-mouse transmission of SeV using similar experimental conditions , except that we cohoused each infected mouse individually with one Ifnar1−/− Ifnlr1−/− sentinel animal . All sentinels co-housed with SeV-infected Ifnlr1−/− mice contracted the virus , whereas only 38% of the sentinels cohoused with WT mice and 75% of the sentinels cohoused with Ifnar1−/− became infected with SeV ( Figure 4C ) . Together , these data show that IFN-λ effectively limits the spread of respiratory viruses in a contact transmission setting . Transmission of respiratory viruses between humans presumably occurs at low doses . A recent study in mice suggests that endogenous IFN-λ confers better protection against influenza virus infection compared with type I IFN if very low viral doses are used ( Galani et al . , 2017 ) . In our experiments , viral titer differences in the upper airways of WT and Ifnlr1−/− mice were relatively small ( Figure 1 ) . This suggested that the high infection dose used in those experiments had largely masked the protective effect of IFN-λ . Indeed , when the challenge dose of Udorn was reduced to 100 PFU per animal , the differences in the upper airways of WT and Ifnlr1−/− mice became more obvious ( Figure 5—figure supplement 1 ) . In agreement with the virus excretion data ( Figure 3 ) , which had indicated a minor role for type I IFN in this process , replication of Udorn in the upper airways was not increased in Ifnar1−/− animals when compared with WT mice ( Figure 5—figure supplement 1 ) , supporting the view that IFN-λ plays a more prominent role in antiviral protection of the upper respiratory tract than type I IFN . To confirm that IFN-λ is more potent than type I IFN in the upper respiratory tract , we used a reciprocal experimental approach . We compared the antiviral potency of a broadly cross-reactive hybrid IFN-αB/D and mouse IFN-λ2 prophylactically applied into the airways . Pilot experiments with differentiated primary mouse airway epithelial cell cultures , known to express functional receptors for both type I and type III IFN ( Crotta et al . , 2013 ) , demonstrated that hybrid IFN-αB/D and mouse IFN-λ2 can induce the IFN-responsive genes Isg15 , Stat1 and Mx1 to similar levels if used at identical concentrations ( Figure 5—figure supplement 2 ) . To exclude unwanted interference from virus-induced endogenous IFN , we used Ifnlr1−/− and Ifnar1−/− mice and treated them intranasally with IFN-α or IFN-λ , respectively , one day before infection with the Udorn virus . The infection was performed with a 40 µl volume to ensure virus delivery to all parts of the respiratory tract . IFN-λ pretreatment potently inhibited virus replication in the upper airways , whereas IFN-α pretreatment had no such effect ( Figure 5A ) . Of note , the differences between IFN-α and IFN-λ were far less pronounced in the lungs ( Figure 5A ) . Next , we determined whether the replication of SeV in the upper respiratory tract of mice is affected by IFN-λ and type I IFN in a similar manner as observed for influenza virus . WT mice were treated intranasally with identical doses of IFN-α or IFN-λ one day before infection with SeV and co-housed with Ifnar1−/− Ifnlr1−/− double-deficient sentinel mice . On day six post infection with SeV , viral titers in the upper airway of IFN-λ-treated mice were on average about 5-fold lower than in mock- or IFN-α-treated mice ( Figure 5B ) . Only 33% ( 3/9 ) of the IFN-λ-treated mice transmitted SeV to sentinel animals , whereas 67% ( 6/9 ) of the IFN-α-treated mice infected the sentinels upon co-housing ( Figure 5C ) . Mock-treated WT mice transmitted SeV at a frequency of 88% ( 7/8 ) under the same conditions ( Figure 5C ) . Taken together , these experiments demonstrated that IFN-λ is able to inhibit virus replication in the upper respiratory tract to a much greater extent than IFN-α . Furthermore , only IFN-λ strongly decreased the rate of successful virus transmission to cage mates . To investigate the possibility that IFN-α might not efficiently reach the epithelial cells of the upper airways , we applied the various IFN preparations via the subcutaneous route before the mice were infected with SeV in a 40 µl volume to ensure infection of the entire respiratory tract . IFN-λ applied subcutaneously provided strong and long-lasting protection against SeV in the upper airways and lungs of mice ( Figure 6A ) . This result is consistent with experiments in which IFN-λ was applied intranasally ( Figure 5B ) . In contrast , IFN-α-treated animals sacrificed on either day 4 or six post infection had similar levels of SeV in nasal swabs and upper airways as mock-treated mice , but the IFN-α-treated animals contained reduced viral titers in the lungs ( Figure 6A ) . Interestingly , IFN-α-treated animals sacrificed on day two post-infection showed significantly reduced viral titers in swabs and upper airways compared with mock-treated controls ( Figure 6A ) . Thus , although this latter observation demonstrated that IFN-α is anti-virally active in the upper airways irrespective of the application route , our data clearly indicated that the antiviral effect of IFN-α , specifically in the upper airways , is surprisingly short-lived . In a second experiment we strove to extend these findings and asked whether IFN-α might similarly inhibit the replication of influenza virus in the upper airways only transiently . In this experiment , the IFN preparations were applied by the intranasal route . Indeed , IFN-α and IFN-λ both exhibited pronounced inhibitory effects on Udorn virus replication in the upper airways at 24 hr post infection ( Figure 6B , left panel ) . However , the antiviral effect of IFN-α in the upper airways could no longer be detected at 72 hr post infection , while virus inhibition by IFN-λ remained prominent ( Figure 6B , right panel ) . Interestingly , there were no differences between the antiviral effects of IFN-α and IFN-λ in the lungs at 72 hr post infection ( Figure 6B ) , indicating that the short-lived nature of the type I IFN response represents a unique feature of the upper respiratory tract . In concordance with these data , we found that IFN-α applied subcutaneously at doses ranging from 0 . 1 to 1 . 0 µg failed to provide long-lasting antiviral protection in the upper airways , but provided substantial antiviral protection in the lungs ( Figure 6—figure supplement 1 ) . The short-lived nature of the IFN-α response was also evident from experiments in which Mx1 gene expression was analyzed in IFN-treated differentiated primary airway epithelial cells derived from mouse tracheae ( Figure 6C ) . Mx1 mRNA levels in IFN-α-treated cells already peaked at 4 hr post onset of treatment and showed a sharp decrease afterwards , nearly reaching baseline levels after 48 hr . In contrast , IFN-λ-mediated induction of Mx1 gene expression increased until 24 hr post onset of treatment , and even after 72 hr Mx1 mRNA levels were still close to peak values in IFN-α-treated cells ( Figure 6C ) . In summary , these data demonstrated that the antiviral effect of IFN-αB/D especially in the upper airways is surprisingly short-lived , irrespective of the application route . Next , we set out to visualize virus-infected cells in the upper airways of mice . At 24 hr post infection with the Udorn virus , many epithelial cells lining the nasal cavity ( rostral naso- and maxilloturbinates ) of IFN receptor-deficient mice were strongly positive for viral antigen ( Figure 7A and B , top panels; Figure 7—figure supplement 1 ) . Epithelial cells in other parts of the upper airways showed no sign of viral infection . In Ifnar1−/− mice that were treated intranasally with 2 µg IFN-λ before infection , no virus-infected cells could be detected by this method ( Figure 7A ) . In contrast , in Ifnlr1−/− mice pretreated with 2 µg IFN-α , a small number of virus antigen-positive cells were present in most sections of the nasal cavity which included parts of the rostral naso- and maxilloturbinates ( Figure 7B ) . These rare virus antigen-positive cells in IFN-α-treated Ifnlr1−/− mice are most likely epithelial cells , due to their location and expression of the epithelial marker EpCAM . These cells did not contain detectable levels of IFN-inducible nuclear MX1 protein , whereas uninfected neighboring cells did ( Figure 7C ) . We concluded from these observations that the upper airways of mice contain some influenza virus-susceptible epithelial cells which appear to rely entirely on IFN-λ for long-lasting antiviral defense . IFN-λ has previously been recognized as an important component of the innate immune system that limits the replication of viruses infecting epithelial cells of the intestinal tract ( Lazear et al . , 2015; Wack et al . , 2015 ) . Since Ifnlr1−/− mice showed only minimally enhanced susceptibility towards influenza viruses compared with wild-type mice in previous studies ( Crotta et al . , 2013; Galani et al . , 2017; Mordstein et al . , 2008 ) , it was assumed that the contribution of IFN-λ to protection against respiratory viruses is minor and becomes detectable only when the type I IFN system is defective ( Crotta et al . , 2013; Mordstein et al . , 2008; Mordstein et al . , 2010a; Mordstein et al . , 2010b ) or when infections are performed with very low doses of virus ( Galani et al . , 2017 ) . Prior studies neglected the upper respiratory tract , where the decisive battle between invading viruses and the host defense system usually takes place . In most previous studies the inoculum was applied to the airways of the animals in a relatively large volume , which delivers the virus directly to the lungs . In this study we used an experimental setup which ensures that the virus is delivered specifically to the upper airways , thus mimicking the natural infection scenario of respiratory viruses . Our work indicates that previous studies have grossly underestimated the non-redundant antiviral potential of IFN-λ in the respiratory tract , and we demonstrate that IFN-λ is of central importance for antiviral defense of the upper airway mucosa during respiratory virus challenge . Galani et al . similarly concluded from a recent study that IFN-λ mediates non-redundant frontline protection against influenza virus infections ( Galani et al . , 2017 ) . Their work focused exclusively on the IFN-λ-mediated control of viral replication in the lungs . Although our data is compatible with the published findings , it further demonstrates that the non-redundant role of IFN-λ against respiratory viruses is far more prominent in the upper airways than in the lungs . Our results also show that IFN-λ plays a more important role than type I IFN in limiting the transmission of respiratory viruses to naïve contacts and that IFN-λ confers effective and long-lasting protection . Our new data reveals a compartmentalization of the IFN-based antiviral defense system in the upper respiratory tract that is reminiscent to what was previously described for the intestinal tract ( Mahlakõiv et al . , 2015 ) . Thus , a picture emerges which indicates that IFN-λ is essential for barrier integrity in tissues where viruses most frequently attack the host , and that the type I IFN system ramps up at these sites only after the epithelial barrier has failed to contain the virus . A likely explanation for the evolution of such functional compartmentalization is that this strategy allows shielding the body against minor viral attacks without causing a strong activation of immune cells ( Davidson et al . , 2015; Davidson et al . , 2016; Trinchieri , 2010; Wack et al . , 2015 ) . Since IFN-λ predominantly acts on epithelial cells rather than immune cells ( Sommereyns et al . , 2008 ) , it is well suited to limit viral replication at epithelial surfaces without inducing a strong inflammatory response commonly associated with type I IFN ( Davidson et al . , 2014; Davidson et al . , 2016; Galani et al . , 2017; Lee-Kirsch , 2017 ) . Different mechanisms seem to favor IFN-λ over type I IFN in the antiviral defense of the gut and the respiratory tract . Epithelial cells of the intestinal tract of adult mice express no or only very low levels of functional type I IFN receptors ( Lin et al . , 2016; Mahlakõiv et al . , 2015 ) and , consequently , these cells are largely blind to the protective activity of IFN-α and IFN-β . In contrast , replication of influenza and Sendai viruses was clearly inhibited in the upper airways of IFN-α-treated mice during the first days of infection ( Figure 6 ) , excluding the possibility that airway epithelial cells in general are devoid of functional type I IFN receptors . However , the IFN-α-mediated antiviral protection of the upper airways , but not the lungs , was surprisingly short-lived and IFN-α-treated and untreated animals contained comparable viral titers in the upper airways at later times post infection . Furthermore , treatment of primary AECs with IFN-α resulted in a very short-lived upregulation of ISG expression . The situation was strikingly different when IFN-λ was applied . IFN-λ treatment resulted in long-lasting ISG induction in primary AECs and sustained antiviral protection of both the lower and the upper respiratory tract in vivo . Analysis at the single cell level revealed that a small number of epithelial cells in the nasal cavity of mice remained virus-susceptible even if the animals were treated with IFN-α . Interestingly , these cells contained no detectable levels of the IFN-inducible MX1 protein , although the animals were treated with a high dose of IFN-α before virus infection ( Figure 7 ) . Since the antiviral effect of IFN-α in the upper airways was poor irrespective of whether it was applied intranasally or systemically , it is unlikely that IFN-α simply failed to penetrate the mucosa of the upper airways . It remains possible that the influenza virus-susceptible cells in the upper airways of IFN-α-treated mice represent a rare specialized epithelial cell type which possesses no functional type I IFN receptors ( Lin et al . , 2016; Mahlakõiv et al . , 2015 ) . More likely , however , these cells contain high levels of factors that inhibit type I but not type III IFN receptor signaling such as USP18 , SOCS1 and SOCS3 ( François-Newton et al . , 2011; Makowska et al . , 2011; Porritt and Hertzog , 2015 ) . IFNAR1 can be rapidly internalized and degraded upon activation ( Fuchs , 2013 ) , while no such effects were reported for the IFN-λ receptor complex . Thus , it remains possible that negative feed-back regulation of IFNAR1 expression is exceptionally strong in epithelial cells of the upper airways . It is important to note that unhindered virus replication in these rare IFN-α-unresponsive cells might initiate a second wave of virus infection in the upper airways shortly after the antiviral state induced by the IFN-α treatment has waned , which could provide an explanation for the transient antiviral effect of IFN-α in the upper airways . Since IFN-λ exhibits selective , sustained and highly powerful antiviral activity in the upper airways that efficiently reduces morbidity and mortality ( Davidson et al . , 2014 ) as well as transmission of respiratory viruses in mice ( our current work ) , we suggest to consider IFN-λ not only for treating diseased patients ( Davidson et al . , 2014; Davidson et al . , 2015; Davidson et al . , 2016; Wack et al . , 2015 ) but also for the management of respiratory virus outbreaks in community settings . According to our results , such clinical use of IFN-λ may not only improve the health status of patients with respiratory symptoms , but is also expected to strongly inhibit virus transmission from infected individuals to healthy contacts . Clinical studies indicate that the side effects of IFN-λ are minimal ( Muir et al . , 2014 ) . Thus , the use of IFN-λ as prophylactic drug seems justified in health care units during epidemics with respiratory viruses as an effort to protect people who are not fully protected by vaccines . The following influenza A virus strains were used: A/Seal/Massachusetts/1/1980 ( H7N7 ) designated SC35M , A/Hong Kong/8/68 ( H3N2 ) designated HK68 , and A/Udorn/72 ( H3N2 ) designated Udorn . SC35M represents a mouse-adapted variant of an avian-like H7N7 virus that was originally isolated from a diseased seal . HK68 and Udorn are human virus isolates and have no passage history in mice . MDCK cells were used for the preparation of influenza virus stocks and for virus titration by plaque assay . Recombinant Sendai virus ( SeV ) expressing green fluorescent protein ( GFP ) was grown in the allantoic cavity of 9-day-old embryonated chicken eggs for 3 days at 33°C ( Strähle et al . , 2007 ) . SeV titers of mouse organ extracts were determined by counting GFP-positive foci after infection of Vero cells at different dilutions . Polio virus type I Mahoney , a strain derived from the infectious cDNA clone pOM , was used in this study ( Shiroki et al . , 1995 ) . Polio virus titration by plaque assay was performed on Vero cells . B6 . A2G-Mx1 mice ( designated WT ) are C57BL/6 mice carrying functional Mx1 alleles ( Mordstein et al . , 2008 ) . B6 . A2G-Mx1-Ifnar1−/− mice ( designated Ifnar1−/− ) lack functional type I IFN receptors , B6 . A2G-Mx1-Ifnlr1−/− mice ( designated Ifnlr1−/− ) lack functional IFN-λ receptors . Ifnar1−/−Ifnlr1−/− lack both IFN receptor systems ( Mordstein et al . , 2008 ) . All mice used in this study were bred locally in our facility or purchased from Janvier ( Strasbourg ) . Animals were handled in accordance with guidelines of the Federation for Laboratory Animal Science Associations and the national animal welfare body . Animal experiments were performed in compliance with the German animal protection laws and were approved by the university’s animal welfare committee ( Regierungspräsidium Freiburg; permit G-15/59 ) . Mice used for the experiments were 6 to 13 weeks old . To achieve infection of the entire respiratory tract , virus in OptiMEM medium containing 0 . 3% BSA was administered intranasally to ketamine/xylazine-anesthetized animals in a volume of 40 µl . For selective infection of the upper respiratory tract , the virus was administered intranasally under light anesthesia ( 3% isoflurane in oxygen ) in a volume of 10 µl . For transmission experiments , infected mice of different genetic backgrounds were cohoused with naïve Ifnar1−/− Ifnlr1−/− contact mice in a fresh cage for 3–5 days starting at 24 hr post infection . Transmission experiments with Udorn and HK68 were performed by cohousing three index mice with four naïve contact mice , whereas transmission experiments with SeV were performed by cohousing individual index mice with one contact mouse each . To determine viral titers in various organs , mice were sacrificed by cervical dislocation at the indicated time points . The upper airways , tracheae and lungs were harvested and stored at −80°C until further proceedings . For immune staining , complete heads without skin and fur were collected and fixed by incubation in 4% formaldehyde at 4°C for 16 hr . To determine viral titers , organs were homogenized in 800 µl ice-cold PBS using FastPrep tubes , spheres , and homogenizer ( MP Biomedicals , USA ) . Two cycles of homogenization at 6 . 5 m/s for 16 s were performed , with samples resting on ice in between . Homogenates were centrifuged at 2300 x g for 5 min at 4°C , supernatants were collected and 10-fold serial dilutions in Opti-MEM with 0 . 3% BSA were applied to MDCK or Vero cells for plaque assay or determination of TCID50 , respectively . Mice were treated with the indicated doses of either human IFN-αB/D which is active in mice ( Horisberger and de Staritzky , 1987 ) or recombinant mouse IFN-λ2 ( Dellgren et al . , 2009 ) by either the intranasal ( 20 or 40 µl ) or the subcutaneous ( 100 µl ) route . Formaldehyde-fixed heads were decalcified in Osteosoft at room temperature on a roller shaker for 96 hr . Samples were then incubated in 15% sucrose at 4°C for 4 hr , followed by incubation in 30% sucrose overnight . Samples were embedded with Tissue-Tek O . C . T . compound and stored at −80°C until use . 4 µm sections were prepared using a cryotome , dried overnight at 37°C and permeabilized for 5 min with 0 . 5% Triton-X in PBS . Blocking was performed with 10% donkey normal serum in PBS for 45 min . Primary antibodies were goat anti-influenza A ( Purified , AbD Serotec ( OBT1551 ) ) , rabbit anti-MX1 ( anti-AP5 peptide [Meier et al . , 1988] ) , rat anti-EpCAM ( Purified , BD Biosciences ( 552370 ) ) and mouse anti-E-Cadherin A647 ( Purified , BD Biosciences ( 560062 ) ) . Secondary antibodies were donkey anti-goat A488 ( Purified , Jackson Immuno Research ( 705-545-147 ) ) , donkey anti-rat A647 ( Purified , Abcam ( ab150155 ) ) and donkey anti-rabbit A555 ( Purified , Invitrogen ( A-31572 ) ) . Nuclei were stained with DAPI and slides were mounted with FluorSave reagent . Isolation and culturing of primary mouse airway epithelial cells were performed as previously described ( Crotta et al . , 2013 ) . Briefly , cells were isolated from the tracheae of WT mice by enzymatic treatment and seeded onto a 0 . 4 µm pore size clear polyester membrane ( Corning ) coated with a collagen solution . At confluence , the medium was removed from the upper chamber to establish an air-liquid interface ( ALI ) . Fully differentiated , 7–10 day-old post-ALI cultures were routinely used for experiments . For analysis of the biological activity of IFN-αB/D and IFN-λ2 , cells were stimulated for 4 hr by basolateral supplementation of the indicated IFN concentrations . For continuous treatment , cells were stimulated by basolateral supplementation of 1 ng/ml IFN-α B/D or IFN-λ2 . Media containing IFN were refreshed every 24 hr . AEC cultures were lysed directly in the transwells using the RNeasy Plus Mini kit ( Qiagen ) according to the manufacturer’s instructions . Snouts were homogenized in 800 µl peqGOLF TriFastTM FL using FastPrep tubes , spheres , and homogenizer ( MP Biomedicals , USA ) . Four cycles of homogenization at 6 . 5 m/s for 16 s were performed and samples subsequently centrifuged at 12 , 000 x g for 10 min at 4°C . Supernatants were diluted ( 1:2 . 5 ) in peqGOLF TriFast FL , combined with 200 µl BCP ( 1-Bromo-3-chlorpropane , Sigma ) per ml TriFastTM FL and phase separation performed with 12 , 000 x g for 15 min at 4°C . Supernatants were combined with one volume ethanol ( Sigma ) and RNA was isolated using the RNeasy Plus Mini kit according to the manufacturer’s instructions starting with the RNeasy spin column . DNA was removed by using DNAse I , Amplification Grade ( Invitrogen ) . cDNA was generated using the Thermoscript RT-PCR system , following the manufacturer’s instructions ( Invitrogen ) . The cDNA served as a template for the amplification of genes of interest ( IAV-M1: forward: 5’-AAGACCAATCCTGTCACCTCTGA-3’; reverse: 5’-CAAAGCGTCTACGCTGCAGTCC-3’ , Ifnl2/3 ( mm0420156_gH , Applied Biosystems ) , Ifnb1: forward: 5’-CCTGGAGCAGCTGAATGGAA-3’; reverse: 5’-CACTGTCTGCTGGTGGAGTTCATC-3’; probe: 5’-[6FAM]CCTACAGGGCGGACTTCAAG[BHQ1]−3’ , Ifna4 ( QT01774353 , QuantiTect Primer Assay , Qiagen ) , Isg15: forward: 5’-GAGCTAGAGCCTGCAGCAAT-3’; reverse: 5’-TTCTGGGCAATCTGCTTCTT-3’ , Stat1: forward: 5’-TCACAGTGGTTCGAGCTTCAG-3’; reverse: 5’-CGAGACATCATAGGCAGCGTG-3’ , Mx1: forward: 5’-TCTGAGGAGAGCCAGACGAT-3’; reverse: 5’-ACTCTGGTCCCCAATGACAG-3’ and Hprt ( mm00446968_m1 , Applied Biosystems ) ) by real-time PCR , using TaqMan Gene Expression Assays ( Applied Biosystems ) , Universal PCR Master Mix ( Applied Biosystems ) and the ABI-Prism 7900 sequence detection system ( Applied Biosystems ) . The increase in mRNA expression was determined by the 2-ΔCt method relative to the expression of Hprt or by the 2-ΔΔCt method relative to mock . Mice were randomly assigned to the various experimental groups , but the study was not blinded . Group sizes and endpoints were predefined based on results from suitable pilot experiments . Comparison between more than two groups was evaluated using one-way analysis of variance ( ANOVA ) with Tukey’s multiple comparisons . Two-way ANOVA with Tukey’s multiple comparison was used to evaluate more than two groups at different time points . All tests were performed using log10 transformed values for viral titers or relative gene expression . Transmission and viral spread frequencies were analyzed by Fisher's exact test . P-values considered to indicate a significant difference are indicated in the figures as follows: p<0 . 05; p<0 . 01; **p<0 . 001 . GraphPad Prism six software ( GraphPad Software , USA ) was used for statistical analysis .	Influenza ( ‘the flu’ ) and other respiratory viruses make millions of people ill every year , placing a large burden on the healthcare system and the economy . Unfortunately , few options for preventing or treating these infections currently exist . The flu virus spreads from infected individuals , enters a new host through the nose and establishes an infection in the upper airways . If the infection stays restricted to this region of the respiratory tract – which consists of the nasal cavity , sinuses , throat and larynx – it causes a rather mild disease . However , if it spreads to the lungs it can cause potentially life-threatening viral pneumonia . Epithelial cells line the upper respiratory tract , forming a physical border between the outside world and the human body . These cells are therefore the first to face the incoming virus . In response , the epithelial cells release messenger molecules termed interferons that warn nearby cells to increase their antiviral defenses . There are several subtypes of interferons , such as IFN-α , IFN-β and IFN-λ , but it was not known how each subtype helps to combat respiratory viruses . To investigate , Klinkhammer , Schnepf et al . exposed mice to flu viruses in a way that mimicked how an infection would naturally start in the upper airways in humans . Some of the mice were genetically engineered so that they could not respond to either IFN-α/β or IFN-λ . The virus spread most effectively from the nasal cavity to the lungs in mice whose IFN-λ system was defective . Infections in mice that lacked IFN-λ were also more likely to spread to other individuals . Furthermore , treating mice with IFN-λ , but not IFN-α , gave their upper respiratory tract long-lasting protection against flu infections and prevented the spread of the virus . IFN-λ therefore has a specific and significant role in protecting the upper airways against viruses , and could potentially be used as a drug to block the spread of infections between humans . Currently , IFN-λ is in clinical trials as a potential treatment for hepatitis D . To repurpose it for upper respiratory tract infections , its effectiveness against specific respiratory viruses will first have to be evaluated .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "microbiology", "and", "infectious", "disease", "immunology", "and", "inflammation" ]	2018	IFN-λ prevents influenza virus spread from the upper airways to the lungs and limits virus transmission
Some oscine songbird species modify their songs throughout their lives ( ‘adult song plasticity’ or ‘open-ended learning’ ) , while others crystallize their songs around sexual maturity . It remains unknown whether the strength of sexual selection on song characteristics , such as repertoire size , affects adult song plasticity , or whether adult song plasticity affects song evolution . Here , we compiled data about song plasticity , song characteristics , and mating system and then examined evolutionary interactions between these traits . Across 67 species , we found that lineages with adult song plasticity show directional evolution toward increased syllable and song repertoires , while several other song characteristics evolved faster , but in a non-directional manner . Song plasticity appears to drive bi-directional transitions between monogamous and polygynous social mating systems . Notably , our analysis of correlated evolution suggests that extreme syllable and song repertoire sizes drive the evolution of adult song plasticity or stability , providing novel evidence that sexual selection may indirectly influence open- versus closed-ended learning . Song is a learned behavior with a complex evolutionary history in the oscine songbirds . Birds’ songs have multiple functions , including species recognition , territory defense , and mate attraction ( Catchpole and Slater , 2003 ) . All species studied in this expansive clade have certain life stages during which they are more likely to learn and acquire songs , termed sensitive periods ( Brenowitz and Beecher , 2005; Marler , 1990; Marler and Peters , 1987; Rauschecker and Marler , 1987 ) . In some species , learning is restricted to a short sensitive period early in life , also called a ‘critical period’ ( e . g . ~day 25–90 in zebra finches ) , after which no new song elements are acquired ( Böhner , 1990; Immelman , 1969; Nottebohm , 1984 ) . Other species appear to delay song crystallization until some time in adulthood ( Dowsett-Lemaire , 1979; Kipper and Kiefer , 2010; Martens and Kessler , 2000 ) ; for example , chipping sparrows appear to have a second sensitive period immediately after their first migration , following which their song crystallizes ( Liu and Kroodsma , 2006; Liu and Nottebohm , 2007 ) . Still other species can continue to acquire new syllables or songs throughout their lives ( Adret-Hausberger et al . , 2010; Espmark and Lampe , 1993; Gil et al . , 2001; Hausberger et al . , 1991; Mountjoy and Lemon , 1995; Price and Yuan , 2011 ) . Typically , this spectrum of variation in the timing of the sensitive period is simplified into a dichotomy of ‘open-ended learning’ and ‘closed-ended learning . ’ While these temporal differences in the song-learning window have been studied for decades , it is unknown how they interact with the evolution of song itself . Previous hypotheses have suggested that seasonal factors , such as environmental variation and breeding season length , play a role in shaping adult song learning ( Nottebohm et al . , 1986; Smith et al . , 1997; Tramontin et al . , 2001 ) . However , evidence from a small-scale comparative analysis suggests that a longer learning window in a species may be associated with larger average syllable repertoire sizes ( Creanza et al . , 2016 ) . Birdsong is composed of both culturally and genetically inherited features , any of which may be subject to evolutionary pressures . Two key modes of selection on song might act in conjunction: female choice can favor certain song characteristics , such as superior repertoire size , learning quality , or song performance ( Beecher and Brenowitz , 2005; Gil and Gahr , 2002; Searcy and Marler , 1984; Searcy and Andersson , 1986 ) , while the inherent metabolic cost of neuroplasticity should theoretically favor a shorter song-learning window and thus reduce the a for a bird to alter its song in adulthood ( Garamszegi and Eens , 2004; Nottebohm et al . , 1986; Tramontin and Brenowitz , 2000 ) . Therefore , while learning in adulthood or elongated sensitive periods have not been shown to be directly under positive selection ( sexual or natural ) or to play an explicit role in female preferences , sexual selection acting on certain song features could indirectly favor longer or shorter song-learning windows . However , this theorized connection between sexual selection and adult learning hinges on establishing the evolutionary relationship between song as the target of sexual selection and the neurobehavioral phenotype of song learning , which has not yet been done . Furthermore , sexual selection is hypothesized to be amplified in species with polygynous social mating systems or high rates of extra-pair paternity ( EPP ) . A recent large-scale study found that polygyny drives faster , but non-directional , evolution of syllable repertoire size , and that syllable repertoire size is negatively correlated with the rate of EPP ( Snyder and Creanza , 2019 ) . Because of the evolutionary links between these mating strategies and song features , higher rates of EPP and polygyny could potentially have an effect on learning windows . We therefore investigate the evolutionary relationship between the critical learning period and mating strategies . Here , we take a comparative , computational approach to the evolutionary history of open- and closed-ended song learning . We mined the literature for longitudinal field and laboratory observations to classify species as exhibiting ‘adult song stability’ or ‘adult song plasticity’ . This classification is a quantifiable proxy for closed- and open-ended learning , as the true length of the song-learning window is difficult to assess directly in nature; ultimately , we obtained data for the classification of 67 species . For these species , we also compiled a database of seven species-level song characteristics that can represent either song complexity ( syllable repertoire , syllables per song , and song repertoire ) or song performance ( song duration , inter-song interval , song rate , and song continuity ) . We then performed phylogenetically controlled analyses to evaluate the evolution of song and mating strategies alongside the relative plasticity of song over time . We find that adult song plasticity has evolved numerous times in bird species . Further , we find evidence of correlated evolution between adult song stability and plasticity and social mating system , with shifts in social mating system occurring more rapidly in lineages with adult song plasticity . In addition , we find a significant evolutionary pattern: species with plastic songs generally have larger repertoires than species with stable songs . Specifically , the evolution of larger syllable and song repertoires appears to drive an evolutionary transition toward open-ended learning . We were first interested in examining the rate of evolution of adult song stability versus adult song plasticity , as well as when and where evolutionary transitions in these traits occurred on a phylogenetic tree ( Jetz et al . , 2012 ) using ancestral state reconstruction . As with any reconstruction of evolutionary history , these simulations cannot exactly predict the ancestral states but aim to approximate them . Furthermore , we note that only a subset of oscine families were represented in our analysis , and most of the early branching lineages that would be required to assess the ancestral state for all oscine species were not included in our dataset . Ultimately , we could not make a conclusion about whether the last common ancestor for the species included in this study had adult song plasticity , but our results hint that there might have been several early transitions in this trait , leading to clades that predominantly have adult song stability or plasticity , coupled with a number of more recent transitions ( see pie charts in Figure 1A for the predicted likelihood of each state at each node ) . We found that a model allowing the transition rate from song stability to plasticity to be different from the transition rate from plasticity to stability ( all-rates-different model [ARD] ) did not fit the data significantly better than a simpler model allowing for only one rate of transition back and forth between song stability and plasticity ( equal rates model [ER] ) ( LogLikelihoodER = −38 . 22 , LogLikelihoodARD = −38 . 21 , p=0 . 87 ) . At least 14 transitions were required to explain the current binary song-stability states of our subset of bird species . Explaining the distribution of song plasticity in our subset of species most parsimoniously requires at least nine transitions to adult song plasticity if the last common ancestor was song-stable and seven transitions to song stability if the common ancestor was song-plastic ( Figure 1—figure supplement 1 ) . We next tested whether song characteristics were affected by the length of the song-learning window on an evolutionary scale . Intuitively , it makes sense that a species that has a longer time-window to learn might be able to accumulate a larger repertoire . Indeed , this relationship is consistent with the pattern of song stability and repertoire size in several clades , such as the Phylloscopus species ( Figure 2 ) . However , many individual species do not follow this prediction: for example , Acrocephalus palustris appears to learn a large repertoire in a single year ( Dowsett-Lemaire , 1979 ) , and Philesturnus rufusater modifies its song for multiple years but maintains a small repertoire ( Jenkins , 1978 ) . Further , numerous species with adult song plasticity do not significantly increase their repertoire sizes over time ( Eriksen et al . , 2011; Galeotti et al . , 2001; Garamszegi et al . , 2005; Nicholson et al . , 2007 ) . Thus , an evolutionary link between adult song plasticity and larger repertoire sizes cannot be assumed . Using a phylogenetically controlled ANOVA ( Garland et al . , 1993; Revell , 2012 ) , we found that species with adult song plasticity did possess significantly larger syllable repertoires than species with adult song stability ( Figures 1A and 3A , Table 1 ) . This concurs with a previous analysis using a smaller dataset ( Creanza et al . , 2016 ) . Similarly , we found that song-plastic species had significantly larger song repertoires than song-stable species ( Figure 3B , Figure 1—figure supplement 2 , and Table 1 ) . There were no significant differences between song-plastic and song-stable species for the other song characteristics that we tested: syllables per song , inter-song interval , song duration , song rate ( calculated as 60/ ( interval + duration ) ) , or song continuity ( calculated as duration/ ( duration + interval ) ) ( Table 1 , Figure 1—figure supplements 3–7 ) . When we used the classification scheme with three states , we could only test for differences in syllable repertoire , song repertoire , and syllables per song between groups , as there were very few early song-stable species for which we had data on the other song traits . We found no significant differences between early song-stable and delayed song-stable species for any tested traits , but both of these groups had significantly smaller syllable and song repertoires compared to song-plastic species ( Figures 1B and 3C , Tables 2 and 3 ) . When performing a phylogenetic generalized least squares ( PGLS ) analysis using continuous estimates of the duration of song plasticity , we found similar results; both syllable repertoire and song repertoire were correlated with duration of song plasticity , such that repertoire size increased with the song-plasticity duration ( Figure 3D , Table 4 ) . Our result that species with adult song plasticity had significantly larger syllable and song repertoires raised the question of whether song stability versus plasticity also affected the rate of evolution for any of the song characteristics . To examine this possibility , we used the Brownie algorithm ( O'Meara et al . , 2006 ) , which tests whether a model with two rates of evolution for each song characteristic—one rate for ancestral periods of song stability and another rate for song plasticity—fits the data significantly better than a model that allows for only a single rate of evolution of each song characteristic regardless of the ancestral states of song stability . Each calculation of the two-rate model is based on one stochastic projection of the ancestral traits across the phylogenetic tree , so we generated 1300 different stochastic simulation maps to use with Brownie . We plotted the distribution of potential rates ( Figures 4–5 ) and compared the average log likelihood of the two-rate models to the log likelihood of the one-rate model ( Table 5 ) . We found that allowing for two different rates of song trait evolution depending on song stability or plasticity did not lead to a significantly better fit model than using one Brownian-motion rate for either syllable repertoire size or song repertoire size , even though syllable repertoires and song repertoires were both significantly larger in species with adult song plasticity ( Figure 4A and B and Table 5 ) . In contrast , the two-rate model led to a significantly better fit for syllables per song , song rate , inter-song interval , and song duration ( Figures 4C and 5 and Table 5 ) , indicating that evolution of these song characteristics was faster in song-plastic lineages ( Figures 4C and 5 , red traces ) . We repeated this analysis with the three-state categorization for syllable repertoire , song repertoire , and syllables per song; for other song characteristics , we did not have enough species in the early song stability group . We found that the three-rate model was significantly better than the one-rate model for syllables per song and song repertoire , but not for syllable repertoire ( Figure 4D–F and Table 6 ) . However , the three-rate model was only significantly better than the two-rate model for song repertoire ( Table 7 ) . Thus , the two-rate model sufficiently approximated the evolution of syllables per song . We noticed that for both song repertoire and syllable repertoire , the rate of evolution in delayed song-stable lineages ( purple traces in Figure 4D , E ) was very similar to the rate in song-plastic lineages ( corresponding red traces ) . We tested one more set of models where we combined delayed song-stable species with song-plastic species to create a ‘longer learning’ group , while early song-stable species were assigned to a ‘shorter learning’ group . For this comparison of shorter versus longer learning , the two-rate model was significantly better than the one-rate model for song repertoire and trending in that direction for syllable repertoire ( Figure 4G , H and Table 8 ) . The three-rate model was not significantly better than the longer/shorter-learning two-rate model for either syllable or song repertoire ( Table 9 ) . Taken together with our phylANOVA results , this pattern suggests that species with early song stability evolve their song repertoires and potentially their syllable repertoires at a slower rate than delayed song-stable and song-plastic species; however , only song-plastic species directionally evolve towards larger song and syllable repertoires . While the Brownie algorithm tested whether adult song plasticity affected the rate of evolution for song characteristics , it did not address whether the opposite might be true . We used BayesTraits ( Pagel , 1994; Pagel and Meade , 2006 ) to test whether the rate and order of evolutionary transitions in one trait is dependent on the state of another trait . Because the song features were continuous variables , we binarized them by setting a series of threshold values to delineate ‘low’ and ‘high’ categories , using each observed song feature value as a threshold in turn . We then tested whether there was correlated evolution between the binary classifications of adult song plasticity versus stability and each of the seven song characteristics . In the lowest third of syllable repertoire thresholds , adult song plasticity with small syllable repertoires was an evolutionarily unstable state , with rapid transitions primarily toward a song-stable state and secondarily toward larger repertoires ( 82% of runs significant in this range , Figure 6 ) . In the middle third of syllable repertoire thresholds , song stability with smaller syllable repertoires is an evolutionarily stable attractor state , with high rates of transition observed from large to small syllable repertoires in song-stable species and from plasticity to stability with a small syllable repertoire . These rate differences are highly significant ( 100% of runs significant in this range ) . In the highest third of syllable repertoire thresholds , adult song stability with a large syllable repertoire is an evolutionarily unstable state , transitioning primarily toward adult song plasticity ( 86% of runs significant in this range , Figure 6 ) . We found similar trends when using two , four , and five bins for the song characteristic threshold values , with subtle differences . When using four or five bins , we still observe that song stability with larger syllable repertoires is an unstable combination . However , for the highest bin of threshold values , the transition rates are faster when changing to song plasticity , whereas for the second-highest bin , we observe faster transition rates toward repertoire size increases ( Figure 6—source data 1 ) . To rule out the possibility that syllable repertoire size evolution is faster in species with larger repertoire sizes regardless of learning program , we tested the rates of evolution of syllable repertoire size across monophyletic species pairs in our dataset . We found that lineages with larger syllable repertoire sizes do not systematically undergo faster or slower syllable repertoire size evolution ( Figure 1—figure supplement 8 ) . At low song repertoire threshold values , song plasticity with a small repertoire is an evolutionarily unstable state; there is rapid transition away from this combination , predominantly trending toward song stability but also transitioning secondarily to a larger song repertoire , with very high significance ( 100% of runs significant in this range ) . At moderate song repertoire thresholds , the highest rate is observed for species with small repertoires transitioning from song plasticity to song stability , also with very high significance ( 100% of runs significant in this range ) . At high song repertoire thresholds , the primary shift is from song stability to song plasticity in species with large song repertoires ( 89% of runs significant in this range ) ( Figure 7 ) . When analyzing the results using five bins , transitions in the upper range of song repertoire values becomes more nuanced; in the highest bin , song stability with a larger song repertoire is very unstable , but is relatively stable in the second-highest bin . In the lower four bins , the dominant transition is from plastic to stable song with smaller song repertoires ( Figure 6—source data 1 ) . The results for syllables per song show some general trends that are complicated by the strong effect of the mimid species ( see results of jackknife analysis below ) . We did not find evidence for correlated evolution between adult song stability or plasticity and any of the other song characteristics ( Figure 6—figure supplements 1–5 ) . In addition , it has been proposed that polygyny and extra-pair paternity ( EPP ) may increase sexual selection pressures on sexually selected traits , including song ( Emlen and Oring , 1977; Payne , 1984 ) , and increased selection pressure due to polygyny was theorized to accelerate the evolution of song learning in a mathematical model ( Aoki , 1989 ) . We tested for correlated evolution between adult song plasticity versus stability and both social mating system ( polygyny vs . monogamy ) and extra-pair paternity ( low vs . high EPP ) , with the caveat that many species in our dataset lacked mating behavior classifications ( 57 species with social mating system data , 41 with EPP data ) . We did not find evidence for correlated evolution between song stability and EPP . There was , however , evidence for correlated evolution between polygynous/monogamous mating systems and song plasticity ( 100% of runs significant ) , with elevated rates of transition between polygyny and monogamy in the song-plastic state ( Figure 6—figure supplement 6 ) . In many cases , there were multiple studies that gave different estimates for a given song trait in one species , so we used the median values across studies for our main analysis . To test whether our results depended on the particular values we used , we repeated the PhylANOVA and Brownie analysis using the either the maximum or minimum values reported in the literature . This did not significantly alter our PhylANOVA results for any song feature when species were split into song-stable and song-plastic groups ( Supplementary file 1 Table S1 ) , though when the early song-stable , delayed song-stable , and song-plastic dataset was used , using the maximum values for song repertoire led to non-significant results ( Supplementary file 1 Table S2-3 ) . When species were split into those with adult song plasticity and adult song stability , the Brownie analysis suggested that syllables per song was evolving significantly faster in song-plastic lineages when we used the median value , but using the minimum values led to non-significant results ( Supplementary file 1 Table S4 ) . When species were split into three states ( early song stability , delayed song stability , and song plasticity ) , we found that the three-rate model was not better than the one-rate model when the minimum values for song repertoire were used , while the three-rate models for syllable repertoire became significant when either the maximum or the minimum values were used ( Supplementary file 1 Table S5 ) . In our dataset , 24 songbird families were represented by 1 to 11 species each . This meant that families had unequal influence on the outcomes of the analyses . We performed a jackknife analysis to examine whether our results were affected by excluding individual families represented by four or more species in the full dataset . Exclusion of individual families did not significantly alter any of the phylANOVA results ( Supplementary file 1 Tables S6-11 ) . In general , exclusion of individual families did not affect the Brownie results ( Supplementary file 1 Tables S12-17 ) or the BayesTraits results ( Figure 6—source datas 2–3 , Figure 7—source data 1 ) . However , there were several notable exceptions , detailed below . In several cases , removing a single family altered the significance of our findings . Removal of the Muscicapidae ( three species ) from the Brownie analysis of inter-song interval created a two-rate model that did not fit the data significantly better than the one-rate model ( Supplementary file 1 Table S13 ) . For song duration , removal of Fringillidae ( three species ) led to a two-rate model that was not a significantly better fit than the one-rate model ( Supplementary file 1 Table S14 ) . Removal of Mimidae ( four species ) from the Brownie analysis of syllables per song drastically changed the results , such that the two-rate model was no longer a significantly better fit than the one-rate model ( Supplementary file 1 Table S15 ) . All of the included Mimidae species are song-plastic , so these results suggest that mimids may be driving the the fit of the two-rate model for syllables per song . In addition , removal of Mimidae from the BayesTraits analysis of song plasticity and syllables per song altered the observed trends . At low threshold values of syllables per song , song plasticity with low syllables per song is an unstable state , with high rates of transition toward either higher syllables per song or toward song stability ( 77% of runs significant in this range ) ; however , when mimid species are removed , gaining more syllables per song in the song-plastic state becomes far more likely . At moderate values of syllables per song , there are elevated rates of transition both toward more syllables per song and towards fewer syllables per song in song-plastic species ( 100% of runs significant in this range ) ; there is a comparable trend when mimid species are removed . At high threshold values , more syllables per song with song stability is an unstable state ( 63% of runs significant in this range , Figure 6—figure supplement 2 ) ; however , with mimids removed , more syllables per song with song plasticity becomes the most unstable state ( Figure 6—source data 3 ) . Due to the strong effect that the inclusion of Mimidae had on the Brownie analysis of syllables per song , we performed a second jackknife analysis , in which each of the four mimid species was removed in turn . Exclusion of Toxostoma rufum or Dumetella carolinensis had little effect on the results ( Supplementary file 1 Table S18 ) , leading to significant support for the two-rate model . However , exclusion of Mimus polyglottos ( p=0 . 074 ) or Mimus gilvus ( p=0 . 509 ) led to a two-rate model that did not fit the data significantly better than the one-rate model ( Supplementary file 1 Table S18 ) . Therefore , we concluded that these two Mimus species drove the estimated evolutionary rate of syllables per song in song-plastic species to be much greater than in song-stable species , and that faster evolution of syllables per song may not necessarily be a universal trend for song-plastic species . Members of Mimidae are renowned for their vocal mimicry , frequently exhibiting improvisation and invention of syllables beyond simple imitation , and thus they lack the generally stereotyped song structure shown in other oscine families . Furthermore , mimids often have periods of continuous singing with minimal repetition of elements and irregular syllable spacing . Thus , quantification of song duration or number of syllables per song for mimids could be highly susceptible to listener perception ( Wildenthal , 1965 ) . Therefore , although we acknowledge that mimids are an important case study in extended learning durations , our results for the evolutionary rate of syllables per song might be more meaningful across all bird species when mimids are excluded , in which case we find the rate of evolution of syllables per song is be independent of adult song plasticity . Previously , it was unknown whether the song-learning window evolved in concert with song features associated with sexual selection , as predicted by a computational model of song learning ( Creanza et al . , 2016 ) . This is a critical missing piece of the puzzle of song learning evolution , as previous evidence has suggested that sexual selection only acts upon the features of song and not the length of the song-learning window or maintenance of the song-learning pathways in the brain . Here , we performed phylogenetically controlled analyses to assess the interactions between the length of the song-learning window—using adult song stability versus plasticity as a proxy—and the evolution of song characteristics . Interestingly , we noted that several evolutionary events relatively early in passerine evolution accounted for much of the diversity in the song-plasticity period in our sample of species . We show that a bird’s ability to modify its song as an adult affects the characteristics of its species' song: adult song plasticity corresponds to larger syllable and song repertoires . Further , our results suggest that sexual selection for large repertoires could indirectly favor individuals with longer learning windows , driving the evolution of increased song plasticity . We found two key trends in the trait correlation ( phylANOVA ) and evolutionary rate ( Brownie ) analyses . First , song plasticity affected the direction of evolution in traits that can be considered metrics of song complexity ( syllable and song repertoire size , Figures 1 and 3 , Table 1 ) , leading to larger repertoires in species with adult song plasticity . Further , species with early song stability evolved their repertoires at a slower rate than species with longer learning ( delayed song stability and song plasticity , Table 5 ) , but song-plastic species did not evolve their repertoires at a faster rate than species with early or delayed song stability combined ( Tables 6–9 ) . Thus , while repertoires only evolve directionally in song-plastic species ( Tables 2–3 ) , our results suggest that extended learning through the first breeding season allows for faster , but nondirectional , evolution of repertoire size . A possible explanation is that delayed song learning allows individuals to modify their songs after migration and thus adapt their song to their new surroundings once they establish a territory , without necessarily corresponding with sexual selection for larger repertoires . This ability for an individual to adapt to a new local song might be beneficial , particularly when species have local dialect structure; however , this would not lead to directional evolution for any particular song feature , consistent with our findings . Second , song plasticity increased the rate of evolution primarily in traits that can be considered metrics of song performance ( song duration , intersong interval , and song rate Figure 5 , Table 2 ) . While these performance-related song traits evolved faster in song-plastic lineages , this rapid evolution did not lead to significant differences in those song characteristics compared to song-stable lineages . A possible explanation may be that increases in repertoire size necessitate changes to song structure , but multiple structural aspects of song can be altered to accommodate these changes . Thus , there is no overall pattern of directional evolution in these other song characteristics . Alternatively , bird species may be required to adapt to increasing repertoire sizes while maintaining species-specific constraints imposed by innate aspects of song structure or female preferences for different performance characteristics . In the latter case , if information about innate characteristics and female preferences are known , it may be possible to predict how song traits will change in response to increasing repertoire sizes and greater adult song plasticity . With our analyses of correlated evolution , we aimed to detect whether the state of the repertoire size or adult song stability versus plasticity consistently changes first in evolutionary history , facilitating a change in the other trait . Overall , our results suggest that there is not a consistent order of evolutionary transitions ( Figures 6–7 ) . For example , song stability with very large syllable or song repertoires , and song plasticity with very small syllable or song repertoires both formed evolutionarily unstable states , with high evolutionary rates of transition in both repertoire size and song-learning window . However , we do note that the fastest rates of transition in our analyses were those switching between song stability and song plasticity to leave those unstable states . This trend suggests that the magnitude of a species’ repertoire may be more likely to drive the evolution of learning window than vice versa ( Figures 6–7 ) . This is consistent with the idea that selection acting upon song features could indirectly place selective pressures on the learning window . We propose several hypotheses that could explain these evolutionary dynamics: 1 ) it may be disproportionately costly to maintain song plasticity when syllable or song repertoire sizes are very small , perhaps because the benefit of extra time to learn does not outweigh the metabolic cost of maintaining plasticity , or 2 ) species with small syllable or song repertoires may have highly stereotyped songs which are selected for based on accuracy of learning and consistency of song production , favoring males that only learn from their fathers or early-life neighbors . Alternatively , in species where females prefer larger repertoires , 3 ) it may simply require more time to learn a large song or syllable repertoire than is available with a short learning window , or 4 ) learning large repertoires may require too much energy devoted to song practice during the crucial period of development before and during fledging , favoring birds that can learn for longer periods . Further research into the physiological or reproductive costs of song plasticity is needed . Beginning with Darwin ( 1872 ) , numerous researchers have proposed that polygynous mating systems could lead to amplified sexual selection ( Emlen and Oring , 1977; Payne , 1984 ) . The elevated rates of transition that we observed between social mating systems in song-plastic lineages suggest an interesting hypothesis for further investigation: perhaps having a plastic song-learning program facilitates evolutionary transitions in mating systems . Our results provide key evidence that sexual selection upon song characteristics might indirectly act upon the song-learning window . We do not fully understand the mechanisms underlying the maintenance or reopening of the song learning window in adulthood , but genetic , environmental , hormonal , and social factors are likely contributors ( Eales , 1987; Eales , 1985; Kroodsma and Pickert , 1980; Morrison and Nottebohm , 1993; Nottebohm , 1969 ) . For example , when zebra finches were reared in isolation , their sensitive periods were lengthened . These isolated birds maintained both gene expression profiles associated with song learning in the song system and high levels of neuronal addition to the HVC ( a key region in the song system of the songbird brain ) for longer than birds reared with an adult male tutor ( Kelly et al . , 2018; Wilbrecht et al . , 2006 ) , linking the neural underpinnings of song learning to the length of the song-learning window ( Brenowitz and Beecher , 2005; Nordeen and Nordeen , 1990; Nordeen , 1997; Nottebohm , 1992 ) . Furthermore , a positive association between HVC volume and song repertoire size has been demonstrated both intraspecifically ( in song sparrows; Pfaff et al . , 2007 ) and interspecifically ( Devoogd et al . , 1993 ) . In light of our findings that adult song plasticity correlates with an increase in song repertoire size , there is a logical prediction that extended song learning may be associated with increased HVC volume across species . This is an important avenue for future research . Although our dataset includes species from 24 different songbird families , many families are not represented due to a lack of data about song stability . It will be important to expand this dataset in future studies . It would also be interesting to explore the evolutionary interactions between adult song plasticity and mimicry of heterospecific sounds , which has been observed in Mimidae and numerous other clades across the songbird lineage ( Goller and Shizuka , 2018 ) . With our current dataset , we could not adequately explore the effects of mimicry on the evolution of song learning outside of the mimids , but the repeated evolution of mimicry makes it a particularly interesting topic for follow-up studies on the length of the song-learning window . In addition , different song metrics that are tailored to mimicry would be important in studying the evolution of vocal mimics and the dynamics of their unique vocal patterns . Furthermore , there is increasing interest in the importance of female song in species , which is more common than previously thought ( Langmore , 1998; Odom et al . , 2014 ) . Our dataset includes numerous species wherein females are known to sing at least occasionally ( Langmore , 1998 ) , but the length of the song-learning window in females has not been assessed in any of these species . There is , however , some evidence that female birds can modify their song preferences in adulthood ( Nagle and Kreutzer , 1997 ) . Thus , it remains an open question whether song plasticity affects the evolution of female song in the same way it affects male song , and whether species with adult song plasticity in males also have adult song plasticity in females . Our findings shed new light on the broader subject of song evolution , specifically the evolution of the process of song learning . We hypothesized that sexual selection on certain aspects of song could in turn alter the selection pressures on the length of the song-learning window . Here , we performed phylogenetically controlled analyses across 67 songbird species to assess the evolutionary interactions between song characteristics and song plasticity in adulthood . With these analyses , we show the first evidence for this evolutionary relationship . Adult song stability versus plasticity may be evolutionarily dependent upon the properties of the song itself: large syllable and song repertoires appear to drive the evolution of adult song plasticity and thus open-ended song learning . This provides context for the remarkable interspecific variation in song-learning windows across the songbird lineage and suggests an evolutionary mechanism by which sexual selection might have influenced the evolution of songbird brains . It is difficult to precisely measure the length of the song-learning window in both field and lab studies . In field studies , if a bird is recorded singing a new song element in its second year , researchers often cannot rule out the possibility that it learned that element during its first year but did not incorporate it into its song until later ( Marler and Peters , 1981; Vargas-Castro et al . , 2015 ) . Likewise , if a bird learns a new element in its second year but elects not to produce it , a recordist would be unlikely to capture it . In contrast , raising birds in the laboratory with known song exposure enables researchers to assess when a bird learned a particular song element ( Chaiken et al . , 1994; Marler and Peters , 1988; Nelson ( 1998 ) ; Nottebohm , 1969 ) but also raises questions about whether the lab-reared birds are exhibiting their normal behaviors ( Baptista and Petrinovich , 1984; Kroodsma and Pickert , 1984 ) . In addition , these lab-rearing procedures have only been performed in a handful of species . In this paper , we examined studies that include longitudinal measures of adult song stability versus plasticity ( e . g . Nordby et al . , 2002 ) to determine whether song is modified across time as a proxy for the length of the song learning window . We performed a literature search to gather information on the stability or plasticity of male song in oscine species . Studies with information about learning style were found via Google Scholar using the following search terms: [species name] or [common name] in combination with ‘open-ended , ’ ‘close-ended , ’ ‘closed-ended , ’ ‘age-limited , ’ ‘crystal* , ” ‘adult learning , ’ and ‘song changes . ’ We used three strategies to assign the song stability classification for a species . We first defined a species as having adult song stability ( ‘song-stable species’ ) if the literature indicated that males did not modify their songs after the first breeding season . Species in which males modified their syllable repertoires after their first breeding season were classified as having adult song plasticity ( ‘song-plastic species’ ) . This strategy was meant to approximate the dichotomy of open- and closed-ended learners often used in the birdsong field . We made two exceptions to this for the species Cacicus cela and Phoenicurus ochruros , in which males do not gain their mature plumage until their second breeding season and may therefore be delayed in reaching sexual maturity relative to other bird species ( Draganoiu et al . , 2014; Trainer and Parsons , 2002 ) . Because these birds cease modifying their repertoires before reaching their second breeding season with mature plumage , they were considered song-stable ( Source data 1 ) . Additionally , past research defined Melospiza lincolnii as an ‘open-ended improviser . ’ It is unclear whether improvisation throughout the lifespan is equivalent to learning throughout the lifespan , however it does fit our definition for adult song plasticity . For the main analysis , we considered this species to have adult song plasticity , but we repeated the main analysis with this species reclassified as having adult song stability . This reclassification produced a negligible effect on our results ( Supplementary file 1 Tables S19-20 ) . For our second strategy , we separated song-stable species into two sub-groups: early song-stable ( species that cease modifying their songs before the first breeding season ) and delayed song-stable ( species that modify their songs during their first breeding season but not after ) . There was not enough information to make this determination for some species , so our dataset was reduced to 59 species . We used these same 59 species to generate a continuous measure of song stability for our third strategy . Because no information was available about the prevalence of song changes beyond the second breeding season for most of the species in our dataset , this measure only ranges from 0 to 2 years , and all song-plastic species were assigned a value of 2 . Furthermore , it was not clear exactly when most of the delayed song-stable species stopped learning , so they were given a value of 1 . 33 , at which point the first breeding season should have ended . The two species mentioned above that display delayed song and plumage maturation were assigned a value of 2 . To gather data on the song characteristics , we performed a literature search via Google Scholar and Web of Science using the search terms ‘Passeriformes’ or [species name] in combination with ‘syllable repertoire’ or ‘song repertoire . ’ This yielded a mix of primary sources and studies that had previously aggregated repertoire size data . We also gathered data from the curated field guides Birds of North America ( Rodewald , 2015 ) and Handbook of Birds of the World ( del Hoyo et al . , 1992 ) . We did not perform explicit searches for any of the other included song characteristics , but we collected this data whenever we encountered it . Song characteristic nomenclature is variable across studies , so , when possible , we read the methods of the primary sources to ensure that the data collected were consistent with our definitions for song characteristics . We utilized the following definitions for song characteristics: For three species with song-learning window data , syllable repertoire size estimates were not available in the literature , so we estimated these repertoire sizes from published sources or song recordings . For Philesturnus rufusater ( Supplementary file 1 Table S21 ) , ( Jenkins , 1978 ) coded songs into different types and gave the song repertoire types of each male studied in Table 2 of that paper . Information from one male was missing from this table . We deduced the repertoire of the missing male by first looking at Table 1 from that paper ( Jenkins’s Table 1 ) , which showed that 16 males had a repertoire size of one song . Only 15 of the males in Jenkins’s Table 2 had a repertoire size of one song . Thus , the missing male had a repertoire size of one song . We then compared the bands of males present in Jenkins’s Table 2 to the territory map in Figure 7 of that paper , and A_RW was the only male missing from Jenkins’s Table 2 . A_RW was located in the DC region of Jenkins’s Figure 7 , so we assigned that as his repertoire . Jenkins notes that neighboring males share song types , so the only other song A_RW could have known instead of DC was ZZ , which has the same number of unique syllables as DC . For Geospiza fortis ( Supplementary file 1 Table S22 ) and G . scandens ( Supplementary file 1 Table S23 ) , we used recordings from the Macaulay Library at the Cornell Lab of Ornithology ( Cornell Lab of Ornithology , 2019 ) to estimate the syllable repertoire size and syllables per song . For G . fortis , we also used the sonogram examples present in Grant and Grant ( 1996 ) . When the song repertoire for a species equalled one , we assumed that its syllable repertoire was equal to its number of syllables per song . In many cases , there were multiple studies that gave different estimates for a given song variable in one species . To handle these discrepancies , we created three datasets ( Source data 2 ) . For one , the main dataset , we used the median value across studies . For the second dataset , we used the minimum value reported in the literature for each species , and for the third , we used the maximum value reported in the literature . We log normalized all song trait data . The three datasets revealed similar results; analyses using the maximum and minimum literature values are presented in Supplementary file 1 Tables S1-S5 . We also cataloged data on mating behavior at the species level . In particular , we assembled binary classifications of social mating system ( monogamy vs polygyny ) and extra-pair paternity ( low EPP vs high EPP ) . We considered a species to be monogamous or polygynous when a source unambiguously categorized that species’ social mating system; that is , we did not assign a social mating system to species labeled ‘probably , ’ ‘usually , ’ ‘mostly , ’ ‘normally , ’ ‘typically , ’ and ‘generally monogamous/polygynous , ’ etc . unless quantitative measurements were also provided . When quantitative data were available , species were defined as polygynous when at least 5% of males had more than one social mate , as in Snyder and Creanza ( 2019 ) . A review of extra-pair paternity studies estimated an average of ~11% of offspring per nest were attributable to extra-pair mates across species ( Griffith et al . , 2002 ) . In accordance with this estimate and with previous studies that used a binary classification of EPP ( Snyder and Creanza , 2019; Soma and Garamszegi , 2011 ) , we used a 10% threshold for either extra-pair young or nests containing at least one extra-pair chick to estimate the frequency of extra-pair paternity in that species ( <10% = low EPP; ≥10% = high EPP , Source data 2 ) . To predict the rate of transition between adult song stability and adult song plasticity , we used the ace function from the R package Phytools and a phylogeny from Jetz et al . ( 2012 ) . We note that this phylogeny has broad coverage of oscine songbirds but is based on limited genetic data ( often mitochondrial ) , so the topology could change as more avian genomes are sequenced ( Jarvis et al . , 2014; Lamichhaney et al . , 2015 ) . With this tree , we tested whether an all rates different ( ARD ) model fit the data significantly better than the equal rates ( ER ) model using an ANOVA . We then used the better-fit equal rates model to generate 10 , 000 trees with make . simmap ( Phytools ) . This function uses the rate from ace and a phylogenetic tree with annotated tips to create stochastic simulation maps for the potential evolutionary transitions between the song-stable and song-plastic states . We found the predicted ancestral state for each of these 10 , 000 simulations and used countSimmap ( Phytools ) to count the total number of transitions that occurred in each map . The minimum number of predicted evolutionary transitions across these 10 , 000 simulations was considered to be the most parsimonious; we also compared this to a manual count of evolutionary transitions starting from either ancestral state . To test whether there were significant differences between song-stable species and song-plastic species for the song traits , we performed a phylogenetically controlled ANOVA ( phylANOVA , Phytools ) for each song characteristic . We repeated this analysis with the subset of species we classified into early song-stable , delayed song-stable , and song-plastic . Because there were relatively few early song-stable species in this dataset , we only performed this re-analysis for song traits that had data for at least nine early song-stable species ( syllable repertoire size: nine species with early song stability , song repertoire size: nine species , and syllables per song: ten species ) . In this paper , we visualize the predicted ancestral traits on the phylogenetic tree with color and pie graphs , however , the raw values are available in Figure 1—source data 1 . To test for correlations between song characteristics and the continuous values for the duration of song plasticity , we performed a phylogenetic generalized least squares ( PGLS ) analysis . We used the function gls ( R package: nlme ) , with the ‘correlation’ parameter lambda computed using the function corPagel ( R package: ape ) . To test whether adult song stability or plasticity affected the rate of evolution for the song characteristics , we used the function brownie . lite ( R package: phytools ) . This function first calculates a one-rate model of evolution for a song characteristic using a phylogenetic tree and the current states of the tips for that song trait . This one-rate model assumes that change in the value of the song characteristic is random across evolutionary time and can be approximated by Brownian motion . Next , a model is generated wherein two different rates are calculated; this two-rate model assumes that the evolution of the song characteristic has one rate in the song-stable state and a different rate in the song-plastic state . This model requires estimations for the ancestral states of song stability for each branch of the phylogeny . To create these estimates , we used the function ace to calculate the rate of transition between the song-stable and song-plastic states for the full dataset . We then used these transition rates to generate 1300 different stochastic simulation maps ( make . simmap ) for the subset of species that had data for each song trait . For the Brownie analysis , we tested whether the two-rate model fit the data significantly better than the one-rate model by performing a chi-square test on the mean log likelihoods of the two models . We repeated this analysis for the set of species we classified as early song-stable , delayed song-stable , and song-plastic for traits for which we had data on at least nine early song-stable species . We compared the three-rate model to the one-rate model . We also reran the two-rate model in this reduced dataset by combining the early and delayed song-stable groups and testing whether the three-rate model was better than the two-rate model . Because the delayed song-stable trace peaked at a similar position to the song-plastic trace for syllable and song repertoire size ( Figure 4D–E ) , we also compared the three-rate model to another version of the two-rate model , in which one group was early song-stable ( shorter learning ) , and the other was delayed song-stable and song-plastic combined ( longer learning ) . We used BayesTraits to test for correlated evolution between song stability and song characteristics , or , in other words , whether the rate and direction of evolutionary transitions of one trait are dependent on the state of another trait , and whether an order of transition events can be inferred . Specifically , we tested the hypothesis that an evolutionary change in song stability increases the likelihood of an evolutionary change in certain song variables or mating behaviors , or vice versa . BayesTraits compares two models of discrete trait evolution for a pair of binary traits and a given phylogenetic tree: 1 ) an independent model ( i . e . the evolution of one trait does not depend on the other trait ) and 2 ) a dependent model ( i . e . the evolutionary transitions of each trait depend on the state of the other trait , suggesting correlated evolution ) . Using the maximum likelihood method , BayesTraits reports marginal likelihoods for the computed dependent and independent models ( function Discrete in package btw [BayesTraitsWrapper] ) , allowing us to determine whether the dependent model describes the data significantly better than the independent model . We used function LRtest ( package: lmtest ) to perform the likelihood ratio test . Since this model required both input traits to be binary , we classified the continuous song characteristics as binary groups ( ‘low’ or ‘high’ ) based on a delineating threshold . Instead of choosing the threshold arbitrarily , we used each unique value of the song characteristic data as the threshold and repeated the analysis 100 times at each threshold . This method of using a spectrum of thresholds to delineate the ‘low’ and ‘high’ value categories resulted in transition rates that varied dramatically depending on where the threshold was placed . In essence , when the threshold was set as a value in the bottom third of the unique trait values present in the data , the analysis evaluated the rate of transition from low-to-moderate and larger values for a song trait and vice versa . When the threshold was set as a value in the upper range of the unique trait values present in the data , the analysis calculated the rates of transition from higher song trait values to medium-to-low values . To account for this nuance , we binned the threshold data into two to five bins , with three bins as the default: low ( bottom third of unique trait values ) , medium ( middle third ) and high ( top third ) . We then calculated the mean of each state transition rate in each bin . In addition to the song characteristics , we also analyzed song stability versus social mating system ( i . e . social monogamy or polygyny ) and rate of EPP . These analyses were performed for 1000 runs each . Some families of birds were well represented in our sample , while others were only represented by one or two species . To test whether any well-represented family significantly skewed our results , we removed each family that was represented by four or more species in the full dataset in turn , and repeated the phylANOVA , brownie . lite , and BayesTraits analyses . Jackknife analyses were only performed when significant results were obtained in the main analysis . Thus , all song traits except continuity were tested in the phylANOVA and brownie . lite jackknife analysis , while only syllable repertoire , song repertoire , and syllables per song were tested in the BayesTraits jackknife analysis . Each Brownie analysis was run on 1300 unique stochastic character maps , and each BayesTraits analysis was repeated 20 times . We determined the family of each species based on its classification in the 2017 version of the eBird Clements Integrated Checklist ( Clements , 2007 ) . The family Locustella was combined with Acrocephalidae , as Acrocephalidae was paraphyletic when Locustella was considered to be a separate family . The Mimidae family alone had a large effect on the syllables per song metric , so we performed another jackknife analysis with phylANOVA and brownie . lite by removing each mimid species in turn . We used a Holm-Bonferroni correction to control for testing multiple hypotheses with the same data ( Holm , 1979 ) . This correction is appropriate for data wherein the outcome of one test is likely to be related to the outcome of another test , as would be the case for song characteristics .	Every morning , people all over the world are greeted by the sound of songbirds singing to attract mates and defend their homes . Each type of songbird has its own unique song that it learns from other birds in the same species . Maintaining these signature songs is important , because songbirds that sing the wrong tune are unlikely to succeed in breeding . This type of vocal learning is rare in the animal kingdom and is similar to how humans learn to speak . The time it takes for songbirds to learn their unique song varies between species: some species pick up their song within a few months after hatching , while others may continue learning for years or even the rest of their lives . It remains unclear , however , why certain species spend a longer time learning , and how changing this length of time affects the song they sing . One possibility is that differences in the amount of time spent learning evolved from female songbirds preferring mates with more impressive songs . To address this possibility , Robinson et al . used previously published data to compare the song characteristics , learning periods , and mating strategies of 67 songbird species . This confirmed that longer learning windows allowed songbirds to develop a larger repertoire of songs with more unique sounds . Robinson et al . also found songbirds that learn elaborate songs over a short period of time quickly evolve either simpler songs or longer phases of learning . Meanwhile , songbirds that learn simple songs over a longer period tended to evolve more elaborate songs or shorten the amount of time they spent learning . Furthermore , species with longer learning were more likely to switch between having one mating partner and multiple partners over several generations . These findings suggest that by choosing mates with elaborate songs and larger repertoires , female songbirds can end up favoring the evolution of longer learning . Studying the conditions that led to longer or shorter learning in songbirds could help scientists understand why some things , like human language , are easier to learn early in life .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "evolutionary", "biology", "computational", "and", "systems", "biology" ]	2019	Correlated evolution between repertoire size and song plasticity predicts that sexual selection on song promotes open-ended learning
Dynamic cellular systems reprogram gene expression to ensure appropriate cellular fate responses to specific extracellular cues . Here we demonstrate that the dynamics of Nuclear Factor kappa B ( NF-κB ) signalling and the cell cycle are prioritised differently depending on the timing of an inflammatory signal . Using iterative experimental and computational analyses , we show physical and functional interactions between NF-κB and the E2 Factor 1 ( E2F-1 ) and E2 Factor 4 ( E2F-4 ) cell cycle regulators . These interactions modulate the NF-κB response . In S-phase , the NF-κB response was delayed or repressed , while cell cycle progression was unimpeded . By contrast , activation of NF-κB at the G1/S boundary resulted in a longer cell cycle and more synchronous initial NF-κB responses between cells . These data identify new mechanisms by which the cellular response to stress is differentially controlled at different stages of the cell cycle . One of the most important functions in a cell is the accurate interpretation of the information encoded in extracellular signals leading to context-dependent control of cell fate . This is achieved via complex and dynamic signal transduction networks , through which gene expression is re-programmed in response to specific environmental cues ( Barabási et al . , 2011 ) . Many signalling systems are subject to temporal changes , involving dynamic alterations to the states of their constituent genes and proteins , with time scales ranging from seconds ( Calcium signalling [Berridge , 1990; Schmidt et al . , 2001] ) , to hours ( DNA damage response [Lahav et al . , 2004] , inflammatory response [Ashall et al . , 2009] ) , to days ( circadian clock [Welsh et al . , 2004] , cell cycle [Sakaue-Sawano et al . , 2008] ) . Although previous studies have indicated interactions between proteins associated with different dynamical systems ( Wilkins and Kummerfeld , 2008; Bieler et al . , 2014; Feillet et al . , 2014 ) , how and when signalling systems are dynamically integrated to determine important cell fate decisions is not well understood . Nuclear Factor kappa B ( NF-κB ) is an important signalling system , implicated in many diseases including autoimmune diseases and cancer ( Grivennikov et al . , 2010 ) . Inflammatory cues such as Tumour Necrosis Factor alpha ( TNFα ) can trigger the nuclear translocation of the NF-κB RelA subunit and activation of target gene transcription ( Hayden and Ghosh , 2008 ) . Nuclear NF-κB activates feedback regulators , including the inhibitory kappa B alpha ( IκBα ) and epsilon ( IκBε ) inhibitors ( Arenzana-Seisdedos et al . , 1997; Kearns et al . , 2006; Paszek et al . , 2010 ) , which bind and transport NF-κB back into the cytoplasm . In response to TNFα , this system shows nuclear-cytoplasmic ( N:C ) oscillations in the localization of the NF-κB complex associated with out-of-phase cycles of degradation and re-synthesis of IκB proteins ( Nelson et al . , 2004; Ashall et al . , 2009; Lee et al . , 2009; Sung et al . , 2009; Tay et al . , 2010; Turner et al . , 2010; Ruland , 2011; Hughey et al . , 2015 ) . Through systems biology and experimental approaches , the frequency of these oscillations has been proposed to be a key parameter that regulates the pattern of downstream gene expression ( Ashall et al . , 2009; Lee et al . , 2014; Williams et al . , 2014 ) . NF-κB signalling has also been suggested to have a role in controlling cell division through a number of different mechanisms ( Perkins and Gilmore , 2006 ) . Many NF-κB family members have been characterised as oncoproteins ( e . g . c-Rel and Bcl-3 [Hayden and Ghosh , 2008] ) . Also , a number of cell cycle control proteins have been shown to be NF-κB transcriptional targets , including Cyclin D , ( Guttridge et al . , 1999; Sée et al . , 2004 ) and p21 , an inhibitor of Cyclin Dependent Kinase ( CDK ) activity ( Hinata et al . , 2003 ) . Although interactions between NF-κB and the cell cycle have been reported ( Kundu et al . , 1997; Phillips et al . , 1999; Perkins and Gilmore , 2006 ) ; observing the dynamics of such interactions is challenging via traditional biochemical techniques , which often fail to capture the heterogeneity in a cellular population . Analysis of cell-to-cell heterogeneity has revealed novel regulatory mechanisms for diverse cellular processes ( Pelkmans , 2012 ) and it has been suggested that this is a fundamental property of the NF-κB response ( Paszek et al . , 2010 ) . The E2 Factor ( E2F ) proteins are differentially expressed during the cell cycle to control cell proliferation ( Bertoli et al . , 2013 ) . They are a family of transcription factors that play a key role in the G1/S cell cycle checkpoint . Previous studies have provided preliminary evidence for physical interaction between NF-κB and E2F proteins ( Tanaka et al . , 2002; Lim et al . , 2007; Garber et al . , 2012 ) In the current study , a combination of single cell imaging and mathematical modelling was applied to investigate reciprocal co-ordination of the NF-κB response and cell proliferation driven by dynamic interactions between RelA and E2F proteins . We investigated the effect of cell cycle timing on the NF-κB response in HeLa cervical cancer and SK-N-AS neuroblastoma cells . SK-N-AS cells showed repeated oscillations in response to TNFα stimulation that were more damped than those seen in HeLa cells ( see Appendix 1—figure 1 for longer time course data [Nelson et al . , 2004; Ashall et al . , 2009] ) . In previous studies it was observed that when SK-N-AS cells were treated with a saturating dose of TNFα ( 10 ng/ml ) the initial response of NF-κB ( i . e . immediate RelA nuclear translocation ) was relatively synchronous between cells ( Nelson et al . , 2004; Ashall et al . , 2009; Turner et al . , 2010 ) ( Figure 1A; Appendix 1—figure 1 ) . However , these data showed a variation in timing and amplitude when cells were treated with a lower dose of 30 pg/ml TNFα , even though this was functionally close to a saturating dose that gave a strong population-level NF-κB response ( Turner et al . , 2010 ) ( Figure 1B ) . In common with treatment of SK-N-AS cells at 30 pg/ml , HeLa cells showed greater heterogeneity in their initial response at a saturating 10 ng/ml dose of TNFα , with some cells showing little or no response and others showing a variable delay ( Nelson et al . , 2002; 2004 ) ( Figure 1C ) . This is in agreement with data showing heterogeneity of the initial response in other cell types ( Tay et al . , 2010; Zambrano et al . , 2014 ) . HeLa cells showed no significant translocation in response to 30 pg/ml TNFα ( Figure 1D ) , suggesting that these cell types have differential dynamic NF-κB responses at varying TNFα doses . 10 . 7554/eLife . 10473 . 003Figure 1 . NF-κB dynamics following TNFα treatment in HeLa and SK-N-AS cells: Mapping the NF-κB response over the cell cycle in synchronized HeLa cells . ( A , B , C and D ) The dynamics of RelA-dsRedxp following 10 ng/ml TNFα treatment in transiently transfected SK-N-AS ( A ) , or following 30 pg/ml TNFα treatment in SK-N-AS , and 10 ng/ml TNFα treatment in HeLa cells ( C ) , and at 30 pg/ml for HeLa ( D ) cells ( n=30 cells analysed per condition ) . ( E ) The localization of endogenous RelA in different cell cycle phases , observed by immunocytochemistry at 2 hr ( G1/S transition ) , 4 hr ( mid S-phase ) , post-release from double thymidine block and with 15 min TNFα treatment . ( F and G ) The dynamics of RelA-dsRedxp in transiently transfected HeLa cells synchronized by a double thymidine block , following 10 ng/ml TNFα treatment at G1/S ( F ) , or passing through S-phase ( G ) ( n=20 cells analysed per condition ) . ( H ) Western blot of Ser536phopho-RelA ( p-RelA ) , IκBα , and cyclophilin-A ( cyclo-A ) levels in synchronized HeLa cells harvested at 1 hr time intervals over the G1/S transition following 15 min treatment with TNFα . Also shown are asynchronous , non-stimulated ( ASY NST ) and asynchronous , stimulated ( ASY ST ) controls , harvested at t=0 . DOI: http://dx . doi . org/10 . 7554/eLife . 10473 . 003 We hypothesised that this cell-to-cell heterogeneity in response might be a consequence of cell cycle phase . To test this hypothesis , we investigated the role of cell cycle in both the HeLa and SK-N-AS cells , as these show different dynamic responses to TNFα that are typical of the profile of a wide range of cell lines ( Tay et al . , 2010; Turner et al . , 2010; Zambrano et al . , 2014; Hughey et al . , 2015 ) . Initially , HeLa cells were treated with 10 ng/ml TNFα at various stages of the cell cycle ( Figure 1E–H ) , as they could be easily synchronized at late G1 by a double thymidine block ( see Appendix 1—figure 2 ) . When endogenous RelA was examined using immunocytochemistry , HeLa cells treated with 10 ng/ml TNFα in S-phase displayed a reduced nuclear localization , compared to those treated in late G1 ( Figure 1E ) . These results were confirmed using time-lapse imaging of synchronised HeLa cells transiently transfected with RelA-DsRedxp . Cells treated in late G1 showed a strong synchronous translocation of RelA , whereas cells treated in S-phase showed reduced RelA translocation ( Figure 1F , G ) . These cell cycle-dependent differences following TNFα treatment of synchronized cell populations were supported by alterations in the extent of IκBα degradation and RelA Serine536 phosphorylation at different stages of the cell cycle as measured by western blot ( Figure 1H ) . To further investigate the effect of cell cycle on the NF-κB response , unsynchronized populations of HeLa and SK-N-AS cells were followed by time-lapse imaging through successive cell divisions . 30 hr after the start of this time-course , HeLa cells were stimulated with 10 ng/ml TNFα . Cells were assigned to different cell cycle phases based upon their mitosis-to-mitosis and mitosis-to-treatment timings ( Figure 2A and B ) . 10 . 7554/eLife . 10473 . 004Figure 2 . Mapping the NF-κB response over the cell cycle through virtual synchronization . ( A ) Selected images from time-lapse imaging of RelA-dsRedxp transiently expressing Hela cells treated with 10 ng/ml TNFα . ( B ) Virtual synchronization of HeLa cells treated with 10 ng/ml TNFα . Cells were imaged through two successive divisions ( M ) allowing correlation of cell cycle timing of TNFα treatment ( parameter 1 ) to RelA dynamics ( parameters 2 , 3 and 4 ) and cell cycle duration ( parameters 1 plus 5 ) . ( C ) Representative cells of RelA-dsRedxp dynamics following TNFα treatment in asynchronous cells , then virtually synchronized into G1 ( n=115 ) , G1/S ( n=32 ) , S ( n=52 ) and G2 ( n=38 ) phases . DOI: http://dx . doi . org/10 . 7554/eLife . 10473 . 00410 . 7554/eLife . 10473 . 005Figure 2—figure supplement 1 . Analysis of cell cycle duration and G1/S timing in HeLa and SK-N-AS cells . ( A ) Time series for FUCCI expression in single representative HeLa and SK-N-AS cells . White arrows mark cells before and after the fluorescence levels were detectable . ( B ) Analysis of cells in ( A ) , showing the G1/S crossing point in fluorescence levels from reporters of SCF ( SKP2 ) ( Orange ) and APC ( Green ) E3 ubiquitin ligase activity . ( C ) Analysis of the G1/S crossing point and cell cycle duration in populations of HeLa and SK-N-AS cells transfected with FUCCI vectors ( n ≥ 30 cells for all conditions ) . DOI: http://dx . doi . org/10 . 7554/eLife . 10473 . 00510 . 7554/eLife . 10473 . 006Figure 2—figure supplement 2 . Statistical analysis of NF-κB translocation in HeLa cells at inferred cell cycle stages following 10 ng/ml TNFα stimulation . ( A ) Analysis of dynamics of initial RelA-dsRedxp translocation with respect to cell cycle phase , using virtual synchronization in HeLa cells . Data were analyzed using nonparametric Anova analysis with Dunn correction for multiple comparisons . Red lines indicate mean normalised amplitude of NF-κB nuclear translocation for different cell cycle phases , and the population average ( dotted line ) . ( B ) Analysis of nuclear RelA occupancy assessed in non-synchronized cells expressing RelA-dsRedxp following treatment with 10 ng/ml TNFα . Statistical analysis showed significant difference between cell cycle phases with respect to distribution of amplitude of the response ( Anova analysis with Dunn correction for multiple comparisons ) . DOI: http://dx . doi . org/10 . 7554/eLife . 10473 . 00610 . 7554/eLife . 10473 . 007Figure 2—figure supplement 3 . Statistical analysis of NF-κB translocation in SK-N-AS cells at inferred cell cycle stages following 30 pg/ml TNFα stimulation . ( A ) Correlation of estimated cell cycle timing with RelA-dsRedxp N:C peak amplitude following 30 pg/ml TNFα treatment ( n=138 ) . ( B ) Analysis of dynamics of initial RelA-dsRedxp translocation with respect to cell cycle phase . Statistical analysis showed a difference between G1 and S with respect to distribution of amplitude of the response ( Anova analysis with Dunn correction for multiple comparisons ) . DOI: http://dx . doi . org/10 . 7554/eLife . 10473 . 007 To ensure the accuracy of the inferred cell cycle stage in these experiments , the cycle timing of cells at the point of TNFα treatment was calibrated through control experiments using Fluorescent Ubiquitin-based Cell Cycle Indicators ( FUCCI ) in both HeLa and SK-N-AS cells ( Figure 2—figure supplement 1A–B ) . The crossing point of Red and Green FUCCI reporters was determined , and defined as the G1/S checkpoint . The average and distribution of the cell cycle duration in populations of HeLa and SK-N-AS cells was also measured ( Figure 2—figure supplement 1C ) . The resulting data suggested that HeLa cells treated with TNFα in late G1 ( inferred to be G1/S ) showed an increase in the translocation amplitude compared to the unsynchronized population average ( Figure 2C ) . By contrast , cells treated in S-phase appeared to show a damped or delayed response ( Figure 2C ) , with markedly reduced amplitude of nuclear NF-κB translocation . In G2 phase the NF-κB response appeared to be restored . Analysis of the complete data set confirmed that there was statistically significant higher nuclear translocation amplitude in HeLa cells at G1/S and significantly reduced amplitude in S-phase , compared to G1 and G2 ( Figure 2—figure supplement 2 ) . A smaller data set from SK-N-AS cells treated with 30 pg/ml TNFα , showed once again a statistically reduced translocation in S-phase compared to G1-phase . Visually the data are consistent with increased translocation in late G1 and a restored level of translocation in G2- compared to S-phase . However more cells would be required for a statistical analysis of possible differences between these cell cycle phases . ( Figure 2—figure supplement 3 ) . We also measured the effect of TNFα treatment on HeLa cell cycle duration ( Figure 3 ) . It was found that mean cell cycle duration for cells treated with TNFα showed a small , but statistically significant increase of 1 . 9 hr ( ~10% ) compared to untreated cells , with the variability in the total population increasing by ~2-fold ( Figure 3 ) . Within this TNFα-treated population , cells treated in late G1 were more susceptible to cell cycle elongation with a cell cycle duration that was ~1/3 longer than the untreated population average . TNFα treatment in S-phase had no statistically significant effect on the timing of mitosis . These data suggest a potential direct or indirect role for the NF-κB system in controlling cell cycle duration through an unknown mechanism at the G1/S phase of the cell cycle . 10 . 7554/eLife . 10473 . 008Figure 3 . Cell cycle length and variability is modified by TNFα addition at G1/S . Analysis of the timing and variability of mitosis ( parameter 1 plus 5 from Figure 2B ) following 10 ng/ml TNFα treatment of asynchronous untransfected HeLa cells , compared to subsets of those cells stimulated at late G1- or S-phase . Mean durations were analysed using nonparametric Anova analysis with Dunn correction for multiple comparisons . Variability in the data was analysed using Levene’s test for equality of variance . DOI: http://dx . doi . org/10 . 7554/eLife . 10473 . 008 The mechanism for alteration of NF-κB responses between the late G1- and S-phases of the cell cycle was sought . Previous studies had suggested that E2-Factor-1 ( E2F-1 ) could physically associate with RelA , and/or its major dimer partner p50 ( Kundu et al . , 1997; Tanaka et al . , 2002; Lim et al . , 2007 ) . E2F-1 is the key transcriptional regulator of the cell cycle transition between G1- and S-phase ( Tsantoulis and Gorgoulis , 2005 ) , where its expression is highest . In the presence of ectopically-expressed EGFP-E2F-1 , we observed a reduction in the activity of a NF-κB-regulated luciferase reporter ( Figure 4A ) . Moreover , the ability of NF-κB to induce endogenous mRNA levels of IκBα and IκBε was impaired in cells co-expressing EGFP-E2F-1 and RelA-DsRedxp , compared to cells expressing RelA-DsRedxp alone ( Figure 4B ) . E2F-1 target gene transcription was also impaired by RelA expression , as indicated by a reduction in the activity of a Cyclin E luciferase reporter ( Figure 4C ) and in the mRNA level of E2F-1 itself ( Figure 4D ) . These data support the reciprocal and coordinated control of transcription by E2F-1 and NF-κB . 10 . 7554/eLife . 10473 . 009Figure 4 . Physical and functional interaction between NF-κB and E2F-1 systems . ( A ) NF-κB-dependent transcription was assessed by luciferase reporter assay ( NF-luc ) , in SK-N-AS cells ( n=3 , +/- s . d ) expressing EGFP-E2F-1 , RelA-dsRedxp or both . ( B ) IκBα and IκBε mRNA levels in SK-N-AS cells ( n=3 , +/- s . d ) following transient expression of EGFP-E2F-1 , RelA-DsRedxp or both . ( C ) E2F-1-dependent transcription as assessed by luciferase reporter assay ( CyclinE-luc ) , in SK-N-AS cells ( n=3 , +/- s . d ) expressing EGFP-E2F-1 , RelA-dsRedxp or both . ( D ) E2F-1 mRNA levels in SK-N-AS cells ( n=3 , +/- s . d ) transiently transfected with RelA-dsRedxp . ( E ) Representative SK-N-AS cells transiently expressing EGFP-E2F-1 ( green ) , RelA-dsRedxp ( red ) , both fluorescent fusion proteins at different levels , or EGFP-E2F-1 , RelA-dsRedxp and IκBα-AmCyan ( blue ) . DOI: http://dx . doi . org/10 . 7554/eLife . 10473 . 00910 . 7554/eLife . 10473 . 010Figure 4—figure supplement 1 . E2F-1 modulates NF-κB dynamics in the absence of stimulus in SK-N-AS cells . ( A ) Time-lapse confocal microscopy of representative SK-N-AS cells transiently transfected with RelA-dsRedxp and EGFP-E2F-1 . ( B ) Trajectories of three representative cells expressing different levels of EGFP-E2F-1 . ( C ) Correlation between RelA-dsRedxp T½ nuclear occupancy ( NO ) time and EGFP-E2F-1 T½ nuclear degradation time , based on data in ( A ) . ( D ) Recapitulation of the observed dynamics with an in silico model for physical interaction between RelA ( NFkB ) and E2F-1 ( E2F ) ( E ) Correlation between NF-κB nuclear occupancy time and nuclear E2F-1 degradation time , based on data in ( D ) ( n= 30 cells ) . DOI: http://dx . doi . org/10 . 7554/eLife . 10473 . 01010 . 7554/eLife . 10473 . 011Figure 4—figure supplement 2 . E2F-1 modulates NF-κB dynamics in the absence of stimulus in HeLa cells . ( A ) Representative HeLa cells transiently transfected with combinations of RelA and E2F-1 fluorescent fusion proteins . ( B ) Time-lapse confocal microscopy of representative HeLa cells transiently transfected with RelA-dsRedxp and EGFP-E2F-1 . ( C ) Trajectories of three representative cells expressing different levels of EGFP-E2F-1 . ( D ) Correlation between RelA-dsRedxp T½ nuclear occupancy ( NO ) time and EGFP-E2F-1 T½ nuclear degradation time , based on data in ( C ) ( n=20 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 10473 . 011 In transient transfection experiments , a predominantly cytoplasmic localization of RelA-DsRedxp was observed when expressed alone , whereas in cells co-expressing EGFP-E2F-1 , both proteins were predominantly nuclear ( Figure 4E ) . In addition we also found that the steady-state cytoplasmic localisation of RelA was restored in cells transiently expressing IκBα-AmCyan in addition to EGFP-E2F-1 and RelA-dsRedxp . These data suggest the hypothesis that IκBα and E2F-1 may compete for the same binding site on RelA , with IκBα perhaps having the higher affinity . Time-series experiments in both SK-N-AS and HeLa cells showed that a decrease in EGFP-E2F-1 expression over time was associated with a re-localization of RelA-DsRedxp from the nucleus to the cytoplasm ( for SK-N-AS cells , Figure 4—figure supplement 1A–B; for HeLa cells , Figure 4—figure supplement 2A–C ) . Quantitative analysis showed a strong correlation between the EGFP-E2F-1 decay half-life and the delay in RelA-DsRedxp translocation back into the cytoplasm ( for SK-N-AS cells , Figure 4—figure supplement 1C; for HeLa cells , Figure 4—figure supplement 2D ) . Initial mathematical modelling of this interacting system ( for details of the model see Appendix Section B ) was able to recapitulate the main features of the observed correlation between E2F-1 levels and RelA localization in silico ( Figure 4—figure supplement 1D–E ) . These data supported a direct interaction between E2F-1 and RelA . Therefore , the physical interactions between E2F-1 and NF-κB proteins in cells were investigated . Co-localization of E2F-1 and RelA had previously been shown through fluorescence imaging experiments ( see Figure 5A ) . A clear physical interaction between fluorescently labelled E2F-1 and RelA in the nucleus of living cells was evident using Förster Resonance Energy Transfer ( FRET ) , in conjunction with acceptor photobleaching as a qualitative indicator of intermolecular interaction ( Figure 5D ) , and Fluorescence Cross-Correlation Spectroscopy ( FCCS ) ( Figure 5C ) . 10 . 7554/eLife . 10473 . 012Figure 5 . Interaction of E2F-1 with RelA . ( A ) Representative cell demonstrating co-localisation of E2F1-EGFP and RelA-dsRedxp upon transient transfection . ( B ) Co-Immunoprecipitation of E2F-1 with RelA pull down in HeLa cells synchronized in late G1 ( HeLa cells used for this experiment due to their greater ease of synchronization ) . ( C ) FCCS assay between transiently transfected EGFP-E2F-1 and RelA-dsRedxp ( red line ) or empty-dsRedxp ( blue line ) fluorescent fusion proteins in single live SK-N-AS cells ( +/- s . e . m based on 10 measurements from 10+ cells per condition ) . ( D ) Qualitative FRET assay between transiently transfected ECFP-E2F-1 and RelA-EYFP fluorescent fusion proteins in live SK-N-AS cells . First negative control between IkB-ECFP and EYFP-E2F1 , and second negative control between free ECFP and EYFP fluorophores expressed in an SK-N-AS cell ( shown are average ECFP and EYFP signals ( +/- s . e . m based on 20 cells per condition normalised to pre-bleach intensity . p . b . indicates the time point at which photo-bleaching occurred ) . DOI: http://dx . doi . org/10 . 7554/eLife . 10473 . 012 In order to further support the interaction between the endogenous proteins , we used co-immunoprecipitation ( Co-IP ) of endogenous E2F-1 and RelA in HeLa cells that had been synchronized in late G1 , when E2F-1 levels were at their peak ( Figure 5B ) . These data confirmed a physical interaction between E2F-1 and RelA , in agreement with previous studies ( Tanaka et al . , 2002; Lim et al . , 2007; Garber et al . , 2012 ) . We were not able to observe a positive co-IP in asynchronous cells ( see Appendix 1—figure 4 ) , suggesting that this interaction was only detectable in HeLa cells at G1/S when E2F-1 was at its highest level . Considered together , all of these different measurements support a significant interaction between these proteins . These data suggest the hypothesis that the interaction between RelA and E2F-1 in the nucleus of G1/S cells , which have been subjected to an inflammatory stimulus , may coordinate differential regulation of NF-κB target gene transcription . In order to understand and further investigate the dynamic behaviour of TNF-α-mediated NF-κB activation in the presence of E2F-1 ( at the G1-S transition ) , an ordinary differential equation-based mathematical model of the NF-κB system ( Ashall et al . , 2009 ) was extended to include the interaction with E2F-1 ( see Appendix Section B ) . In this model , E2F-1 was assumed to compete with IκBα for binding to free NF-κB , but had no effect on the localization of RelA bound to IκBα . Simulations ( of nuclear NF-κB levels over time from transfection experiments ) using this model , supported the hypothesis that E2F-1 might temporally control the duration of RelA nuclear occupancy through a combination of binding to RelA in the nucleus and inhibition of RelA-dependent IκBα transcription ( as suggested by data shown in Figure 4 ) . E2F-1 degradation could allow NF-κB to re-activate IκBα , which in turn could restore RelA to a cytoplasmic localization . When the initial mathematical model was used to simulate the effect of E2F-1 on the responsiveness of NF-κB to TNFα , the in silico simulations predicted that TNFα would induce immediate oscillations of free RelA ( Figure 6A ) . In contrast , time-lapse live cell imaging of SK-N-AS cells stimulated with TNFα , showed that in cells expressing RelA-DsRedxp and EGFP-E2F-1 ( which initially had nuclear RelA-DsRedxp ) , there was a delay before the onset of oscillations ( Figure 6B and D ) . The length of this refractory period was on average ~4-fold longer than the peak1:peak2 timing in cells expressing RelA-DsRedxp alone ( Figure 6B and [Ashall et al . , 2009] ) . Altered model structures were investigated in order to resolve this discrepancy between experimental data and model predictions . One of the simplest altered models predicted that an E2F-1 target gene might stabilize IκBα ( keeping NF-κB in the cytoplasm during S-phase [Figure 6C] ) . In support of this prediction , TNFα treatment of SK-N-AS cell populations transiently expressing EGFP-E2F-1 and RelA-DsRedxp led to reduced levels of phospho-S536-RelA and stabilized levels of IκBα ( Figure 6E ) . Simulations of the response to TNFα from the revised model were consistent with the observed delay in oscillations in single cells expressing ectopic EGFP-E2F-1 ( Figure 6B and D ) and also with the inhibition or delay in the response during S-phase , but not during G1 or G2 ( Figure 2C ) . Candidates for the E2F-1-regulated component ( s ) predicted by the revised model were therefore sought . 10 . 7554/eLife . 10473 . 013Figure 6 . Mathematical modelling predicts an additional key component for NF-κB - cell cycle interactions: E2F-4 identified as a putative candidate . ( A ) Model simulations of RelA-dsRedxp dynamics when co-expressed with EGFP-E2F-1 in cells treated with TNFα . ( B ) Dynamics analysed in representative SK-N-AS cells treated with 10 ng/ml TNFα expressing RelA-dsRedxp and EGFP-E2F-1 ( C ) Model simulation of experimental conditions in B , incorporating interactions between NF-κB complexes and a putative E2F-1-induced target protein , subsequently proposed as E2F-4 . ( D ) Analysis of average timing to second peak of NF-κB translocation following TNFα treatment in SK-N-AS cells expressing RelA-dsRedxp alone or with EGFP-E2F-1 ( n=20 cells per condition , error bars show s . d . ) ( E ) Assessment of the extent of RelA Ser536 phosphorylation ( p-RelA ) , E2F-4 and IκBα stability by western blot compared to cyclophilin A ( cyclo A ) amounts in SK-N-AS cells either untreated or treated with 10 ng/ml TNFα and expressing combinations of either untagged or fluorescent RelA-dsRedxp and EGFP-E2F-1 . ( F ) Western blot of E2F-1 and E2F-4 in synchronized HeLa cells , where t=0 is late G1-phase . DOI: http://dx . doi . org/10 . 7554/eLife . 10473 . 013 Previous studies had shown strong structural homology between E2F-1 and other E2F family members ( Tsantoulis and Gorgoulis , 2005 ) . E2F-4 is a transcriptional target of E2F-1 ( Xu et al . , 2007 ) and can be cytoplasmic during S-phase ( Lindeman et al . , 1997 ) . E2F-4 ( together with E2F family members ) was therefore considered as a prospective candidate . We confirmed that ectopic expression of E2F-1 in cells resulted in increased E2F-4 expression , consistent with E2F-4 being a transcriptional target of E2F-1 in these cells ( Figure 6E ) . The profile of E2F-4 expression was found to be delayed relative to that of E2F-1 in the cell cycle , peaking in S-phase in synchronized HeLa cells ( Figure 6F ) . To further confirm the role of E2F-4 in the suppression of RelA translocation following TNFα treatment during S-phase , the physical and functional interactions between E2F-4 and RelA proteins in cells were investigated . When transiently expressed in either HeLa or SK-N-AS cells , both proteins were located in the cytoplasm ( Figure 7A ) . Following TNFα treatment , the timing of RelA-DsRedxp translocation to the nucleus in both cell lines was delayed relative to the level of the fluorescent signal from EGFP-E2F-4 ( Figure 7B for dynamic profiles and Figure 7—figure supplement 1 for analysis ) . The physical interaction of endogenous E2F-4 and RelA proteins was supported by Co-IP from HeLa cells synchronized in S-phase ( Figure 7C ) . No pull-down was observed in cells synchronised in late G1 phase ( see Appendix 1—figure 4 ) . This is the cell cycle stage when E2F-1 , but not E2F-4 is at its peak expression level . This interaction was confirmed by acceptor photo bleaching FRET and FCCS data obtained from cells transiently expressing ECFP-E2F-4 and RelA-EYFP ( for FRET ) or RelA-dsRedxp and EGFP-E2F-4 ( for FCCS ) fluorescent fusion proteins ( Figure 7D and E respectively ) . These data suggested that members of the E2F family have differing , but functionally linked , roles in the regulation of NF-κB dynamics . The observed dynamics could be represented by a mathematical model that recapitulates data ( Figure 6C ) from live cell imaging of the transient expression of the appropriate fluorescent fusion proteins ( Figure 5A and 7A , for details of modelling see Appendix Section B ) . 10 . 7554/eLife . 10473 . 014Figure 7 . E2F-4 directly interacts with NF-κB and perturbs RelA dynamics in response to TNFα stimulation . ( A ) Single cell trajectories from groups of HeLa cells expressing RelA-dsRedxp and different levels of EGFP-E2F-4 showing the dynamics of RelA-dsRedxp after 10 ng/ml TNFα treatment ( n=60 cells ) . ( B ) HeLa cells synchronized in S-phase , co-immunoprecipitated with anti-RelA antibody and probed for E2F-4 . Also shown are IgG negative controls and whole cell lysate unsynchronized positive control ( ctrl ) . ( C ) Representative SK-N-AS cells transiently transfected with RelA-dsRedxp and EGFP-E2F-4 . ( D ) FRET assay in live SK-N-AS cells expressing ECFP-E2F-4 and RelA-EYFP fluorescent fusion proteins ( shown are average ECFP and EYFP signals ( +/- s . e . m ) based on 20 cells per condition normalised to pre-bleach intensity . p . b . indicates the point of photo-bleaching ) . ( E ) FCCS assay in cells transiently expressing EGFP-E2F-4 and RelA-dsRedxp ( red line ) or dsRedxp ( blue line ) fluorescent proteins in single live SK-N-AS cells ( +/- s . e . m based on 10 measurements in each of 10+ cells per condition ) . DOI: http://dx . doi . org/10 . 7554/eLife . 10473 . 01410 . 7554/eLife . 10473 . 015Figure 7—figure supplement 1 . Analysis of RelA-dsRedxp dynamics in HeLa and SK-N-AS cells co-expressing EGFP-E2F-4 following TNFα stimulation . The effects of different EGFP-E2F-4 expression levels on the amplitude and timing of the first peak of RelA translocation in HeLa and SK-N-AS cells treated with 10 ng/ml and 30 pg/ml TNFα , respectively . These data indicate how ectopically expressed EGFP-E2F-4 can inhibit the translocation of RelA-dsRedxp in response to TNFα . DOI: http://dx . doi . org/10 . 7554/eLife . 10473 . 015 The majority of experiments described above utilised transient expression of the E2F and RelA fusion proteins driven from a CMV promoter in a plasmid vector . Previous data had suggested that RelA fusion proteins expressed in a knock-in mouse are functional and fusion protein expression does not perturb the system ( De Lorenzi et al . , 2009 ) . Our transcription analyses ( Figure 4 ) suggested that E2F-1 N- and C-terminal fusion proteins also retained functional activity . However , as E2F proteins are normally expressed at specific stages of the cell cycle , ectopic expression from a strong constitutive promoter could give rise to out-of-context expression at inappropriate stages of the cell cycle ( i . e . for E2F-1 , stages other than late G1 and early S-phase ) . Therefore , expression of fusion proteins from these vectors might potentially show interactions that are not physiologically relevant . An additional complication in these experiments was that exogenous expression of E2F-1 ( but not E2F-4 ) fluorescent fusion protein from a CMV promoter caused apoptosis when transfected alone . Interestingly this effect was rescued by co-expression with RelA . To further validate the functional link between the E2F and RelA proteins , we sought to achieve more physiologically relevant levels and timing of the fluorescent fusion protein expression . To this end , stable HeLa cell lines were generated , with integrated Bacterial Artificial Chromosomes expressing E2F-1-Venus and RelA-DsRedxp under the control of their natural human gene promotors and associated regulatory elements ( see Appendix Section C ) . HeLa cells were chosen for this study based on their more consistent cell cycle timing ( between cells ) compared to SK-N-AS cells ( as shown in Figure 2—figure supplement 1 ) . Stable cell lines were generated with a human E2F-1-Venus BAC construct , and showed the same pattern of synthesis and degradation of a transiently expressed FUCCI reporter for SCF ( SKP-2 ) activity , indicating normal cell cycle progression ( see Figure 8—figure supplement 1 ) . All viable clones had relatively low expression of the E2F-1-Venus BAC , further suggesting that E2F-1 over-expression was detrimental to cell survival . Following the generation of these stable clones , a single clone was selected for the integration of a RelA-DsRedxp BAC into this cell line . This generated a dual stable clone of E2F-1-Venus and Rel-A-DsRedxp ( termed C1-1 ) . This clonal cell line showed a slight increase ( ~8% ) in mean cell cycle length ( with similar cell-to-cell variability ) comparable with wild type HeLa ( see Appendix 1—figure 7 ) . Similar to wild type cells , TNFα treatment in the C1-1 cell line increased the variability in cell cycle timing compared to that of resting cells . The slight change in mean cell cycle duration ( ~20 hr ) in the dual BAC stable clonal cell line C1-1 was taken into account for inference of the dynamics of RelA-DsRedxp translocation at different cell cycle phases . The profile of E2F-1-Venus expression was used for assignment of the cell cycle stage at the time of stimulation cells based upon the time of peak E2F-1-Venus expression ( Figure 8—figure supplement 2 ) . This provided an alternative and faster method of virtual synchronisation to that used in Figure 2 , allowing the assignment of G1 , S and G2 phases to the data from the simulated BAC stable cells . The level of RelA translocation ( Figure 8B ) was then quantified for cells from each cell cycle phase . In agreement with data from the transiently transfected HeLa and SK-N-AS cells ( Figure 2 , Figure 2—figure supplements 2 and 3 ) , the cells treated in late G1/S-phase showed higher amplitude RelA nuclear translocations , whereas Cells treated in S-phase showed a statistically significant suppression in S-phase RelA translocation compared to cells in early G1- or G2-phases ( Figure 8 and Figure 8—figure supplement 3 ) . 10 . 7554/eLife . 10473 . 016Figure 8 . Effect of cell cycle timing on RelA-dsRedXP translocation in dual BAC HeLa cells ( C1-1 line ) that co-express E2F-1-Venus fusion protein . ( A ) Selected images from time-lapse experiment of dual BAC transfected HeLa stable clone 1-1 showing translocation of RelA-dsRedXP and E2F-1-Venus expression at different cell cycle phases . Cells were treated with 10 ng/ml TNFα . ( B ) Analysis of the dynamics of initial RelA-dsRedxp translocation in cells ordered at specific cell cycle times with respect to the peak of E2F-1 expression ( n = 128 ) . Data were analysed using nonparametric Anova analysis with Dunn correction for multiple comparisons . Red lines indicate mean normalised amplitude of NF-κB nuclear translocation for different cell cycle phases , and the population average ( dotted red line ) . Analysis of nuclear RelA occupancy was assessed in virtually synchronised C 1-1 cells , based on time from cell division and relative to peak E2F-1-Venus expression level . RelA-dsRedxp localization was visualized to allow quantification of translocation , following treatment with 10 ng/ml TNFα . The dotted black line shows the spline fitted level of E2F1 at different times and cell cycle stages ( see also Figure 8—figure supplement 1 below ) . Statistical analysis showed a difference between G1 vs S , and G2 vs S with respect to distribution of amplitude of the RelA translocation response . ( C ) RelA-dsRedxp dynamics following 10 ng/ml TNFα treatment in asynchronous cells ( left panel ) and cells virtually synchronised into G1 , G1/S , S and G2 phases . The data for each cell was normalised to the amplitude ( N:C ratio ) at t = 0 min . DOI: http://dx . doi . org/10 . 7554/eLife . 10473 . 01610 . 7554/eLife . 10473 . 017Figure 8—figure supplement 1 . Virtually synchronized HeLa C 1-1 cells . Normalised E2F-1-Venus expression at the time of TNFα stimulation of C1-1 cells ( data also shown in Figure 8B ) . E2F-1-Venus expression was normalised to its peak expression . The time axis represents the time of TNFα stimulation relative to the peak of E2F-1-Venus for each cell , where time 0 is the peak of E2F-1-Venus expression . Positive times indicate stimulation after the peak of E2F-1-Venus expression and negative values indicate stimulation events before the peak of E2F-1-Venus expression . The black line shows a spline interpolation of the level of E2F-1-Venus expression . Cell cycle phases were estimated based on measured the E2F1 profile and average cell cycle timing . DOI: http://dx . doi . org/10 . 7554/eLife . 10473 . 01710 . 7554/eLife . 10473 . 018Figure 8—figure supplement 2 . Physiological and functional expression of E2F-1-Venus in stable BAC-transduced HeLa cells . ( A ) HeLa cells stably expressing an E2F-1-Venus fluorescent fusion protein from a 5KB endogenous E2F-1 promoter ( Green ) , transiently transfected with a FUCCI reporter for SCF ( SKP-2 ) activity ( Orange ) . Showing the profile of E2F-1 over two consecutive cell cycles ( one parent and two daughter cells ) , with a peak in late G1 . E2F-1 levels dropped during S-phase consistent with rapid rise in SCF ( SKP-2 ) activity and a loss of FUCCI fluorescence . ( B ) Representative cell from the E2F-1-Venus and RelA-DsRedxp stably transfected population of Hela cells through one full cell cycle . DOI: http://dx . doi . org/10 . 7554/eLife . 10473 . 01810 . 7554/eLife . 10473 . 019Figure 8—figure supplement 3 . Analysis of the expression of E2F-1-Venus and RelA-DsRedxp translocation in single C1-1 HeLa cells stimulated with 10ng/ml TNFα at different cell cycle phases . Grey line shows the E2F-1-Venus expression level plotted agains the right y-axis . The red , green blue and orange lines show the timecourse of RelA-dsRedxp localization in exemplar cells in the G1 , G1/S , S and G2 phases respectively plotted against the left y-axis . The black vertical line represents the point at which cells were treated with TNFα . DOI: http://dx . doi . org/10 . 7554/eLife . 10473 . 01910 . 7554/eLife . 10473 . 020Figure 8—figure supplement 4 . Expression and interaction of RelA-dsRedxp and E2F-1-Venus . ( A ) Western blot of RelA and α-tubulin levels in dual BAC stable C1-1 and WT HeLa showing exogenous expression of RelA-dsRedxp . ( B ) Western blot of E2F-1 and cyclophilin A levels in dual BAC stable C1-1 and WT HeLa cells expressing the E2F-1-Venus BAC ( C ) Total fluorescent molecules per cell for E2F1-Venus at peak expression and RelA-dsRedxp in unstimulated cells ( data obtained from Fluorescent Correlation Spectroscopy measurements , and calculated using volume estimates from z-stacked WT HeLa in suspension ) . ( D ) FCCS mean correlation curves ( +/- s . e . m ) between E2F-1-Venus and RelA-dsRedxp ( red line , n=46 ) for TNFα treated BAC stable cells . A comparison to transient empty-dsRedxp is shown ( blue line , n=15 ) ( E ) Kd determination results using a scatter plot and linear regression ( Theil-Sen estimator ) . The slope of the regression gives the Kd value . DOI: http://dx . doi . org/10 . 7554/eLife . 10473 . 020 Expression of the RelA-DsRedxp and E2F-1-Venus fusion proteins in the stable cell line was quantified through molecular counting of fluorophores via FCS ( Figure 8—figure supplement 4 ) . This gave an estimate of 310 , 000 ± 120 , 000 molcules of RelA-DsRedxp per cell . This figure was comparable to previous molecular estimates using FCS that had been obtained in stable cell lines generated using lentivirus ( Bagnall et al . , 2015 ) , and previous estimates of RelA concentration using analytical chemistry ( Martone et al . , 2003; Zhao et al . 2011 ) . RelA showed an approximate ratio of 3:1 ectopic to endogenous expression based on quantitative analysis of western blot data ( see Figure 8—figure supplement 4A ) . By contrast , FCS analysis suggested that E2F-1-Venus expression was lower ( 24 , 000 ± 9100 molecules of E2F-1-Venus per cell ) . Western blot analysis ( Figure 8—figure supplement 4B ) suggested that there was an approximate ratio of 10:1 endogenous to ectopic levels ) . This might suggest selective pressure during cloning , as over-expression of E2F-1 has been reported to compromise cell viability ( Crosby and Almasan , 2004 ) . The apparent selective pressure against higher E2F-1 fusion protein expression was also in agreement with our own data that suggested that transient exogenous expression of E2F-1 fusion protein ( but not E2F-4 ) alone caused apoptosis , but that this was rescued by co-expression of RelA . In the same manner observed with low EGFP-E2F1 expression from transient co-expression , the more physiological levels of E2F-1-Venus expression in the stably transfected cells suggested that RelA-DsRedxp remained predominantly cytoplasmic in unstimulated cells . The interaction between E2F-1-Venus and RelA-DsRedxp following TNFα stimulation was measured by Fluorescence Cross-Correlation Spectroscopy ( FCCS ) . A strong cross-correlation was confirmed in the nucleus ( Figure 8—figure supplement 4D ) indicating that the interaction uncovered by transient transfection with plasmids was not an artefact of over-expression , but was contextually relevant in relation to the cell cycle and RelA activation . Analysis of the dissociation constant ( by FCCS ) for the RelA-DsRedxp and E2F-1-Venus binding in the nucleus of TNFα–stimulated cells suggested a dissociation constant ( Kd ) of 12 nM ( Figure 8E ) . The stable and physiological co-expression of E2F-Venus and RelA-DsRedxp facilitated fluorescently labelled proteins to be observed over the course of a full cell cycle . Cells were virtually synchronized as previously described following stimulation with 10 ng/ml TNFα , and translocation of RelA-DsRedxp was plotted against the nuclear expression of E2F-1-Venus ( Figure 8—figure supplement 2 ) . We also investigated the consequences of knocking down both E2F-1 and E2F-4 using siRNA . Imaging experiments showed E2F-1 knockdown did not prevent cell cycle progression , and did not affect the heterogeneity of population response upon TNFα stimulation ( data not shown ) , perhaps indicating compensation by other E2F family members . In addition , our mathematical model predicted that knocking down E2F-1 might not substantially affect the repression of the NF-κB response in S-phase , which was instead predicted to be due to the effect of E2F-4 expression . However , knock-down of E2F-4 was found to be lethal to cells ( Crosby and Almasan , 2004 ) preventing time lapse analysis . A key additional consideration is the overlapping roles of other E2F family members , which makes knock-down of individual E2F proteins unpredictable , due to potentially co-operative and/or redundant functions . Biological timing plays a key role in the encoding and decoding of biological information . Of particular interest is the role of biological oscillators , which can have very different cycle periods . A key question is how they may interact to robustly control essential biological processes . Here , we propose a reciprocal relationship between two oscillators , NF-κB signalling and the cell cycle . TNFα stimulation in S-phase showed a suppressed and delayed translocation of RelA , with no observable perturbation to cell cycle timing . In contrast , stimulation in late G1 showed strong translocation of RelA ( Figure 2 ) and led to significant lengthening of the cell cycle ( Figure 3 ) . These data suggest that cells use the G1/S checkpoint to prioritize between inflammatory signalling and the onset of DNA replication prior to cell division ( see schematic diagram in Figure 9 ) . The presence of a mechanism for prioritization between the important processes of cell proliferation and inflammation suggests that an inflammatory response during DNA replication might be detrimental to the cell . 10 . 7554/eLife . 10473 . 021Figure 9 . Schematic representation of NF-κB – E2F interactions . ( A ) Predicted mechanisms for NF-κB interaction with E2F proteins over the G1/S transition ( B ) Model simulations of single cell behaviour . DOI: http://dx . doi . org/10 . 7554/eLife . 10473 . 021 The data showing that TNFα stimulation alters cell cycle timing in a cell cycle phase-dependent manner is intriguing ( Figure 3 ) . However , our data do not identify a specific mechanism by which TNFα may regulate cell cycle length . The observation that the effect of TNFα stimulation on cell cycle lengthening appears to be specific to G1/S- rather than S-phase suggests that this may occur by delaying transition through the G1/S checkpoint . One hypothesis is that this might occur through NF-κB modulation of E2F family transcriptional activity . At the same time , the system is more complex as NF-κB is known to regulate the expression of other key cell cycle regulating proteins . Important examples include Cyclin D ( Guttridge et al . , 1999; Hinz et al . , 1999 , Sée et al . , 2004 ) , and p21waf1/cip1 ( Basile et al . 2003 ) . Therefore , there is undoubtedly a more complex set of interactions between NF-κB and the control of cell proliferation and cancer ( Perkins and Gilmore , 2006 ) . As well as a number of studies that suggest a physical interaction between E2F and NF-κB proteins ( Kundu et al . , 1997; Chen et al . , 2002; Tanaka et al . , 2002 , Shaw et al . , 2008; Palomer et al . , 2011 ) , there have been a few previous studies that have suggested that this interaction might have functional importance . Araki et al . described an NF-κB-dependent mechanism for growth arrest mediated by a dual mechanism . They suggested that E2F-1-dependent transcription was inhibited by IKK activation and that E2F-4 was phosphorylated directly by IKK resulting in increased activity of the E2F-4/p130 repressor complex ( Araki et al . , 2008 ) . Their study did not assume direct interactions between the E2F and Rel proteins and did not take into account protein dynamics . Nevertheless , their conclusions are very complementary to the present study . Another study by Tanaka et al . focused on the combined role of E2F-1 and c-MYC in the inhibition of NF-κB activity ( Tanaka et al . , 2002 ) . This study demonstrated interactions between E2F-1 and both RelA and p50 . Rather than focusing on cell division , their study showed that inhibition of RelA activity by E2F-1 resulted in increased apoptosis . Since both the NF-κB and E2F families of transcription factors have important roles in the control of apoptosis ( Phillips and Vousden , 2001; Kucharczak et al . , 2003; Crosby and Almasan , 2004 ) , it is therefore interesting to speculate that the levels of different E2F proteins at different cell cycle stages may regulate cell fate decision making in collaboration with signalling systems such as NF-κB . One important conclusion of the current study is the physical interaction of RelA with E2F-1 and E2F-4 proteins . It is however not necessary to assume strong binding and sequestration into different cellular compartments . Instead , control of cross-talk could be a consequence of mutual control of gene expression . We provide some data that suggests that E2F-1 and IκBα may compete for binding to RelA ( see Figure 4E ) . We suggest that control may be achieved through repression of the IκBα feedback loop ( and perhaps other negative feedbacks , such as A20 ) . However , it might be that other genes are differentially activated through the combined action of these transcription factors . In support of this , Garber et al . performed a study in dendritic cells where they studied a panel of transcription factors by ChIP-Seq following LPS stimulation . Their data suggested that E2F-1 and RelA are common transcription factor pairs that were bound together at a large set of functionally important gene promoters ( see data in Figure 3B of Garber et al . , 2012 ) . It therefore seems likely that these proteins mutually regulate patterns of transcriptional activity , controlling the expression of downstream feedback genes , cell proliferation and apoptosis . We describe a mechanism for E2F-1 that suggests competition with IκBα for NF-κB binding . This was effectively described by the model ( see also Figure 9 ) , and was used to predict the role for an E2F-1 target gene , upregulated in S-phase . Our data support E2F-4 as a candidate for this E2F-1 target gene . It should be noted however , that the E2F family of proteins may all play a role in this complex system . A surprising characteristic of E2F-4 is its predominantly cytoplasmic localisation in some cell types . As a result , we were unable to perform a competition localisation experiment ( as for E2F-1 , Figure 4E ) . We cannot therefore comment on whether E2F-4 also competes with IκBα for RelA binding . Therefore , the model ( both mathematical model and schematic model in Figure 9 ) encode E2F-4 binding as a ternary complex to RelA and IκBα together . We stress that this is only one possible mechanism , but we have used this formulation since it is the simplest model that is consistent with all of our data . As described by Araki et al . ( see above ) there may be other components involved such as IKK-mediated E2F-4 phosphorylation ( Araki et al . , 2003 ) . Functional and context-dependent coupling between dynamic cellular processes ( such as the cell cycle , the circadian clock [Yang et al . , 2010; Bieler et al . , 2014; El Cheikh et al . , 2014] , or p53 [Toettcher et al . , 2009] ) is emerging as a common theme in intracellular signalling ( Ankers et al . , 2008; White and Spiller , 2009; Spiller et al . , 2010 ) . The present study has characterized a dynamic and functional interaction between NF-κB and the cell cycle systems , which each oscillate with different periods . Coupling between cellular processes ( e . g . at the G1/S commitment point ) can have contrasting effects on cell fate . Such temporal communication between processes represents a way for cells to gate their biological signals and coordinate and prioritize cell fate decisions in response to changes in their environment . In a wider context , understanding how ( and when ) these dynamic interactions occur could yield important therapeutic targets for fields such as cancer chronotherapy ( Choong et al . , 2009; Lévi et al . , 2010 ) . Human recombinant TNFα was supplied by Calbiochem ( UK ) . Tissue culture medium was supplied by Invitrogen ( UK ) and Fetal Bovine Serum ( FBS ) was from Harlan Seralab ( UK ) . All other chemicals were supplied by Sigma ( UK ) unless stated otherwise . All plasmids were propagated using E . coli DH5α and purified using Qiagen Maxiprep kits ( Qiagen , UK ) . NF-κB-Luc ( Stratagene , UK ) contains five repeats of an NF-κB-sensitive enhancer element upstream of the TATA box , controlling expression of luciferase . Luciferase reporter CyclinE-Luc was obtained from Peggy Farnham ( University of Wisconsin-Madison , USA ) . EGFP-E2F-1 and EGFP-E2F-4 contain the Enhanced Green Fluorescent Protein ( EGFP ) gene ( Invitrogen , UK ) fused to the N-terminal ends of the human E2F-1 and E2F-4 gene fragments ( kind gifts from Emmanuelle Trinh , BRIC , Denmark ) . Similarly , ECFP-E2F-1 and ECFP-E2F-4 contain the Enhanced Cyan Fluorescent Protein ( ECFP ) gene ( Invitrogen , UK ) RelA-DsRedxp contain the optimised DsRed Express protein ( DsRedxp ) gene ( Clontech , CA ) fused to the c-terminal end of human RelA gene ( described previously in Nelson et al . ( 2002 ) . RelA-EYFP contain Enhanced Yellow Fluorescent protein ( EYFP ) gene ( Invitogen , UK ) fused to the C-terminal end of human RelA gene . SK-N-AS neuroblastoma ( cat . no . 94092302 ) and HeLa cervical carcinoma ( Cat . No . 93021013 ) cell lines were obtained from European Collection of Authenticated Cell Cultures ( ECACC ) . Cells were cultured and frozen down to form a low passage working stock . Subsequent working stocks were used for no more than 10 passages . Working stocks were screened to ensure the absence of mycoplasma every 3 months using LookOut Mycoplasma PCR Detection Kit ( Cat . No . D9307 Sigma , UK ) . For confocal fluorescence microscopy and immuno-cytochemistry , SK-N-AS and HeLa cells were plated on 35 mm glass-bottom dishes ( Iwaki , Japan and Greiner , Germany ) at 1x105 cells per dish in 3 ml medium . HeLa cells were plated at 5x104 cells per dish in 3 ml medium . 24 hr post-plating , the cells were transfected with the appropriate plasmid ( s ) using Fugene 6 ( Boehringer Mannheim/Roche , Germany ) . The optimized ratio of DNA:Fugene 6 used for transfection of HeLa or SK-N-AS cells was 2 µg DNA with 4 µl Fugene 6 and 0 . 8 µg DNA with 1 . 2 µl Fugene 6 respectively . For Co-IP assays , SK-N-AS cells were plated on 100 mm tissue culture dishes ( Corning , USA ) at 4 . 5x106 cells per dish in 10 ml medium . For western blotting , semi-quantitative and quantitative PCR , HeLa and SK-N-AS cells were plated on 60 mm tissue culture dishes ( Corning , USA ) at 5x105 and 1x106 cells respectively per dish in 5 ml medium . 24 hr post-plating , 2 mM Thymidine was added to the culture medium . Following a 19 hr incubation , cells were washed and fresh medium added . Following a 9 hr incubation , 2 mM Thymidine was again added to the culture medium and the cells incubated for a further 16 hr . Cells were then washed and fresh media added . Following release from Thymidine block , the G1/S-synchronized cells were either imaged or incubated ( at 37°C , 5% CO2 ) for the indicated duration prior to cell lysis or fixation . For confocal fluorescence microscopy , the cells were treated in-situ between imaging acquisitions after an indicated pre-treatment incubation period ( usually 24 hr post-transfection ) . For western blotting and q-PCR experiments , the cells were treated with TNFα 24 hr post-plating . The cells were imaged either immediately after treatment , or incubated ( at 37°C , 5% CO2 ) for the indicated duration prior to cell lysis or fixation . Confocal microscopy was carried out as described ( Nelson et al . , 2004 ) using either 20x Fluar 0 . 8 NA or 63x Planapochromat 1 . 4 NA objectives . CellTracker ( Shen et al . , 2006; Du et al . , 2010 ) was used for data extraction . For RelA fusion proteins , mean fluorescence intensities were calculated for each time point for both nucleus and cytoplasm then nuclear:cytoplasmic ( N:C ) fluorescence intensity ratios were determined . For time lapse microscopy , a modified version of the Autofocus macro ( an improved version of the Autotimeseries macro [Rabut and Ellenberg , 2004] ) was used . The cell cycle duration and G1/S timing of SK-N-AS and HeLa cells was analysed using live-cell imaging of successive cell divisions to determine typical cell cycle duration . In addition , the cell cycle dynamics were quantified expressing Fluorescence Ubiquitin-based Cell Cycle Indicators ( FUCCI , [Sakaue-Sawano et al . , 2008] ) ( Figure 2—figure supplement 1 ) . The crossing point in fluorescent levels from FUCCI markers of APC and SCF E3 ubiqutin ligase was used as an indication of G1/S transition in the cells ( Figure 2—figure supplement 1B ) . Mitosis to mitosis timings were determined in non-transfected cells , as well as in cells transfected with RelA-dsRedXP and the dual BAC cell line ( Appendix 1—figure 7 ) For the BAC cell line that expressed E2F-1-Venus it was only possible to use the single SCF FUCCI G1 vector ( due to fluorescent protein spectral overlap ) . Cells were imaged for ~30 hr prior to TNFα treatment in order to capture each cell passing through mitosis . The timing of TNFα treatment relative to mitosis for each cell was then calculated . Events following TNFα treatment ( i . e . the dynamics of RelA-DsRedxp translocation , or cell cycle duration ) could then be correlated to inferred cell cycle phase at the point of treatment . Dual BAC cell lines were imaged for an entire cell cycle . Cells were aligned based upon normalised peak amplitude of E2F-1-Venus , and virtually synchronised based upon alignment of peak E2F-1 expression and the relative timing of TNFα stimulation . Cell cycle boundaries were inferred through characterization of cell cycle progression through transfection of FUCCI G1 phase marker construct ( Figure 2—figure supplement 1 ) . HeLa cells were cultured in 100 mm dishes . Following trypsinization , and resuspension in 1 ml of medium the cells were stained by addition of 250 μl of 50 μg/ml propidium iodide , 0 . 15% TritonX-100 , and 150 μg/ml RNase A before analysis in an Altra flow cytometer ( Beckman Coulter ) . FRET was carried out using a Zeiss LSM510 with 'META' spectral detector mounted on an Axiovert 100S microscope with a 63x Planapochromat , 1 . 4 NA oil-immersion objective ( Zeiss ) . ECFP and EYFP ( Karpova et al . , 2003 ) were excited with 458 nm laser light , emitted fluorescence was collected in 8 images each separated by 10 nm between 467 nm and 638 nm in lambda scanning mode . Separation of ECFP and EYFP fluorescence spectra was carried out using the linear unmixing algorithms of the Zeiss LSM510 software ( Zeiss ) , using reference spectra taken from cells expressing the ECFP or EYFP fusion proteins alone or untransfected cells . The fluorescence spectrum was separated into ECFP , EYFP and background signals . FRET was assayed by acceptor ( EYFP ) photo-bleaching . Bleaching was accomplished using 50 iterations of 514 nm laser light with no attenuation from the acousto-optical tuneable filter ( AOTF ) . FCS and FCCS was carried using either a Zeiss LSM780 or Zeiss 710 with Confocor 3 mounted on an AxioObserver Z1 microscope with a 63x C-apochromat , 1 . 2 NA water-immersion objective . Zen 2010B software was used for data collection and analysis . EGFP fluorescence was excited with 488nm laser light and emission collected between 500 and 530 nm . DsRed-express was excited with 561nm laser light and emission collected between 580 and 630 nm . The protocols as outlined in Kim et al . ( Kim et al . , 2007 ) were followed , with 10 x 10 s runs used for each measurement . FCS was used to quantify the total number of fluorescent molecules per cell as previously described ( Bagnall et al . , 2015 ) . The confocal volume had previously been estimated at 0 . 59 ± 11 fL ( mean ± SD ) using Rhodamine 6G of known diffusion rate , and WT HeLa cells in suspension were imaged by confocal microscopy to give volume estimates of 1420 ± 490 fL and 6110 ± 3580 fL for nucleus and cytoplasm respectively . ( For FCCS controls see Appendix Section E ) . HeLa cells synchronized at G1/S or S-phase were washed with room temperature PBS and lysed with modified RIPA buffer ( 50 mMTris-HCl pH 7 . 4 , 150 mM NaCl , 1 mM EDTA , 1% NP-40 ) including a 1:100 dilution of Protease Inhibitor cocktail ( Sigma , UK ) , PMSF and phosphatase inhibitor ( Phos Stop , Roche ) . Immunoprecipitation was carried out using Immunoprecipitation kit-Dynabeads Protein G ( Invitrogen ) with anti-RelA antibody ( #3034 , Cell Signaling , MA , USA ) . The samples were analyzed by western blotting using anti-E2F-1 ( Cell Signaling , #3742 ) or anti E2F-4 ( Santa Cruz , C-20 sc-866 ) antibodies . The RNeasy Mini Kit ( Invitrogen , UK ) was used to extract mRNA from the cells following manufacturer’s instructions , using the primers: IκBα left TGGTGTCCTTGGGTGCTGAT right GGCAGTCCGGCCATTACA , IκBε left GGACCCTGAAACACCGTTGT right CCCCAGTGGCTCAGTTCAGA , E2F-1 left TGCAGAGCAGATGGTTATGG right TATGGTGGCAGAGTCAGTGG , cyclophilin A left GCTTTGGGTCCAGGAATG right GTTGTCCACAGTCAGCAATGGT . Luciferase reporter assay were carried out as described in White et al . ( 1990 ) , using a LUMIstar plate reading luminometer ( BMG , Germany ) . HeLa cells were prepared using combinations of the above techniques , typically involving synchronization and/or TNFα stimulation of cells seeded at appropriate density into 35 mm glass-bottomed dishes . Dishes were subsequently washed three times with PBS and fixed with 1 ml 4% paraformaldehyde for 15 min . Dishes were then washed three times with PBS , and ‘blocked’ to prevent non-specific antibody binding with the addition of 1–2 ml of 1% BSA , 0 . 1% Triton X-100 ( in PBS ) from 20 min up to overnight . The primary antibody ( or antibodies for dual-staining ) , dissolved in Ab Buffer ( 1% BSA , 0 . 1% Triton X-100 in PBS ) , were added to the dishes for 60/90 min at a 1:2000 dilution . Dishes were then washed ( 3x1 ml ) with Ab buffer for 10 min . Secondary Antibody ( s ) were subsequently added to the dishes ( Cy3-anti-mouse , 1:200 dilution ( Sigma ) , FITC Rabbit , 1:200 [AbCam] ) for 30/45 min respectively , prior to 3 sequential washes of PBS blocking buffer ( described above ) . Following the addition of fluorescent secondary antibodies , dishes were covered in aluminium foil and left in 2 ml PBS prior to imaging . Whole cell lysates were prepared at the indicated times after stimulation . Membranes were probed using the following antibodies: anti-IκBα ( #9242 , Cell Signaling , MA ) , anti-RelA ( #3034 , Cell Signaling , MA ) , anti-phospho-RelA ( Ser 536 ) ( #3031 , Cell Signaling , MA ) , anti-IκBα ( #9242 , Cell Signaling , MA ) , anti-E2F-1 ( #KH-95 , Millipore Biotechnology , USA ) , anti-E2F-4 ( sc-866 , Santa Cruz ) , α-Tubulin Antibody ( #2144 Cell Signaling , MA ) , and anti-cyclophilin A ( #07–313 , Millipore Biotechnology , USA ) .	Investigating how cells adapt to the constantly changing environment inside the body is vitally important for understanding how the body responds to an injury or infection . One of the ways in which human cells adapt is by dividing to produce new cells . This takes place in a repeating pattern of events , known as the cell cycle , through which a cell copies its DNA ( in a stage known as S-phase ) and then divides to make two daughter cells . Each stage of the cell cycle is tightly controlled; for example , a family of proteins called E2 factors control the entry of the cell into S phase . “Inflammatory” signals produced by a wound or during an infection can activate a protein called Nuclear Factor-kappaB ( NF-κB ) , which controls the activity of genes that allow cells to adapt to the situation . Research shows that the activity of NF-κB is also regulated by the cell cycle , but it has not been clear how this works . Here , Ankers et al . investigated whether the stage of the cell cycle might affect how NF-κB responds to inflammatory signals . The experiments show that the NF-κB response was stronger in cells that were just about to enter S-phase than in cells that were already copying their DNA . An E2 factor called E2F-1 –which accumulates in the run up to S-phase – interacts with NF-κB and can alter the activity of certain genes . However , during S-phase , another E2 factor family member called E2F-4 binds to NF-κB and represses its activation . Next , Ankers et al . used a mathematical model to understand how these protein interactions can affect the response of cells to inflammatory signals . These findings suggest that direct interactions between E2 factor proteins and NF-κB enable cells to decide whether to divide or react in different ways to inflammatory signals . The research tools developed in this study , combined with other new experimental techniques , will allow researchers to accurately predict how cells will respond to inflammatory signals at different points in the cell cycle .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "cell", "biology", "computational", "and", "systems", "biology" ]	2016	Dynamic NF-κB and E2F interactions control the priority and timing of inflammatory signalling and cell proliferation
Regeneration following tissue damage often necessitates a mechanism for cellular re-programming , so that surviving cells can give rise to all cell types originally found in the damaged tissue . This process , if unchecked , can also generate cell types that are inappropriate for a given location . We conducted a screen for genes that negatively regulate the frequency of notum-to-wing transformations following genetic ablation and regeneration of the wing pouch , from which we identified mutations in the transcriptional co-repressor C-terminal Binding Protein ( CtBP ) . When CtBP function is reduced , ablation of the pouch can activate the JNK/AP-1 and JAK/STAT pathways in the notum to destabilize cell fates . Ectopic expression of Wingless and Dilp8 precede the formation of the ectopic pouch , which is subsequently generated by recruitment of both anterior and posterior cells near the compartment boundary . Thus , CtBP stabilizes cell fates following damage by opposing the destabilizing effects of the JNK/AP-1 and JAK/STAT pathways . As development proceeds in multicellular organisms , cells become more restricted in developmental potential . This results initially from the expression of lineage-specific transcription factors and is then stabilized by heritable chromatin states that restrict accessibility of the transcriptional machinery to subsets of genes ( Britten and Davidson , 1969; Levine , 2010; Allis and Jenuwein , 2016 ) . The relative stability of these ‘epigenetic landscapes’ is thought to protect cells from patterns of gene expression that are inappropriate to that particular lineage . The paradigm that progressive restrictions in cell fate during development are unidirectional and irreversible was challenged by the demonstration that nuclei from differentiated cells could be reprogrammed to a more naïve state either by transferring them into an enucleated fertilized egg ( Gurdon , 1962 ) or by expressing a combination of transcription factors ( Takahashi and Yamanaka , 2006 ) . Indeed , pluripotent stem cells have been derived from highly specialized cell types including neurons ( Kim et al . , 2011 ) and lymphocytes ( Loh et al . , 2009 ) . Even earlier , studies of Drosophila imaginal discs by Ernst Hadorn and colleagues showed that cells determined to one fate could switch to a very different fate . During embryogenesis , groups of cells at particular locations are specified to form particular imaginal discs ( e . g . genital disc , wing disc , eye-antennal disc ) depending on their location in the embryo ( Cohen , 1993 ) . Although these fates are determined during embryogenesis , disc cells do not differentiate to form adult structures until many days later , when metamorphosis occurs . To investigate the stability of the determined state , Hadorn’s group transplanted imaginal disc fragments into abdomens of female adult flies where regeneration occurred . Most often , the tissue generated was appropriate for the implanted imaginal disc . However , at low frequency , regeneration resulted in tissue that was appropriate for a different disc ( e . g . leg disc tissue generated from a fragment of genital disc ) . Hadorn termed this phenomenon transdetermination ( a switch from one determined state to another ) ( reviewed in [Hadorn , 1968; McClure and Schubiger , 2007; Worley et al . , 2012] ) . While transdetermination has been studied for years , many aspects are still mysterious . First , the fate change does not happen in a single cell , but rather , initiates in a small group of cells ( Gehring , 1967; Hadorn et al . , 1970; Maves and Schubiger , 1998 ) . How a group of cells coordinately changes fate is not known . Second , certain fate changes seemed more likely than others . For example , genital-to-leg transformations or leg-to-wing transformations were far more frequent than transformations in the opposite direction . Third , transdetermination can be viewed as an aberrant form of regeneration that occurs following tissue damage and correlates with increased local proliferation ( Sustar and Schubiger , 2005 ) . Increased JNK activity , which is observed following damage ( Bosch et al . , 2005; Mattila et al . , 2005; Bergantiños et al . , 2010; Fan et al . , 2014 ) , reduces Polycomb-mediated repression and increases the frequency of transdetermination ( Lee et al . , 2005 ) as well as other types of fate changes ( Herrera and Morata , 2014; Schuster and Smith-Bolton , 2015 ) by mechanisms that have not been elucidated . Finally , not all regions within a given disc have the same probability of undergoing a change of fate; Gerold Schubiger and colleagues described a region in the leg disc called the ‘weak point’ that has drastically higher rates of leg-to-wing transdetermination ( Schubiger , 1971; Maves and Schubiger , 1995 ) . Our understanding of cell fate plasticity during regeneration is still at an early stage . Based on gene expression changes in regenerating discs , mutations in several candidate genes examined were shown to change the frequency of transdetermination , but the underlying mechanisms are not known ( Klebes et al . , 2005; McClure et al . , 2008 ) . To systematically screen for mutations that increase the frequency of cell fate changes during regeneration , we used a genetic system developed in our laboratory that reproducibly ablates a specific portion of the wing pouch and then allows it to regenerate ( Smith-Bolton et al . , 2009 ) . We identified mutations in the gene encoding the transcriptional co-repressor , C-terminal Binding Protein ( CtBP ) , and show that CtBP opposes the destabilization of cell fates during regeneration caused by JNK and the JAK/STAT pathway . The increased frequency of ectopic wings in some genetic backgrounds suggests that there are genes whose normal function includes the stabilization of cell fates during damage and regeneration . To identify such genes , we systematically screened 226 stocks bearing deletions of the third chromosome for their ability , when heterozygous , to increase the frequency of ectopic wings following egr-mediated damage and regeneration ( Figure 1H ) . The frequency of adults with ectopic wings ( EW ) was scored for each deletion and graphed based on relative position of the deletion in the genome ( Figure 1I ) . Taken together , these deletions covered approximately 90% of the euchromatic portion of chromosome 3 . Of these 226 deletions tested , two were lethal , 91 elicited no EWs , 93 elicited the formation of EWs at a low penetrance ( <10% ) , 28 showed an intermediate penetrance ( >10% , <25% ) and 12 showed a high frequency of EWs ( >25% ) ( Figure 1J ) . Deletions from different deletion collections , Bloomington Stock Center ( BSC ) , DrosDel Project ( ED ) and Exelixis ( Exel ) each had a similar breakdown of classes ( Figure 1K ) . The score for each deletion tested is in ( Supplementary file 1 ) . We focused on the 87D region , since multiple overlapping deletions allowed us to narrow the region of interest to a small number of genes . We then tested the available alleles and identified the responsible gene as C-terminal Binding Protein ( CtBP ) . A small deletion ( Df ( 3R ) Exel8157 ) that spans CtBP caused EWs in 47% of adults ( Figure 2A ) . A hypomorphic CtBP allele , CtBP03463 , and a loss-of-function allele , CtBP87De-10 ( which we will henceforth refer to by its protein truncating mutation CtBPQ229* ) caused EWs in 22% and 42% of adults , respectively ( Figure 2A ) . Mock ablation , with only either 1 ) rn-GAL4 , tub-GAL80ts or 2 ) tub-GAL80ts , UAS-egr did not yield EWs ( Figure 2A ) . EWs were attached to the scutellum and were often unable to flatten and appeared round ( Figure 2B ) . The highest frequency of EWs occurred when the genetic damage was initiated on day seven after egg lay ( AEL ) in CtBPQ229/+ mutants , while no EWs were observed in wild type on all days tested ( Figure 2—figure supplement 1 ) . CtBP was originally identified as a protein that interacted with the C-terminus of the adenovirus E1A protein ( Boyd et al . , 1993 ) . It has subsequently been shown to function as a transcriptional co-repressor in mammals and in Drosophila ( Nibu et al . , 1998; Zhang and Levine , 1999 ) . While mammals have two paralogs of CtBP , reviewed in ( Chinnadurai , 2003 ) , the Drosophila genome contains a single CtBP gene . CtBP itself does not bind DNA . Rather , it associates with multiple DNA-binding proteins that possess a PxDLS motif , or related motif , including Hairy ( Poortinga et al . , 1998 ) , Krüppel ( Nibu et al . , 1998 ) and Brinker ( Hasson et al . , 2001 ) . The Drosophila CtBP gene has multiple spliced isoforms that are predicted to generate seven different proteins , which are usually grouped as short ( 379–386 aa ) or long ( 473–481 aa ) ( Figure 2C ) . To confirm that the increased frequency of EWs was in fact due to the disruption of CtBP , we used CRISPR/Cas9 to generate CtBP alleles on an isogenized Oregon-R third chromosome that did not cause EWs following damage ( 0% , n = 1892 adults ) ( Figure 2D ) . We generated two CRISPR guide RNAs that target regions spanning codons shared between all predicted protein isoforms , located at codon 148 and 334 ( Figure 2C ) . The alleles recovered include a likely hypomorph , CtBPN148PY ( three base pair insertion ) , which induced EWs in approximately 11% of adults and a suspected null , CtBP148Δ2 ( frame shift inducing two base pair deletion ) , which induced EWs in approximately 20% of adults ( Figure 2D ) . When compared to the isogenic parent chromosome , CtBP-/+ increases the frequency of EWs . Ectopic pouches were observed in CtBP-/+ wing discs following damage and recovery ( Figure 2E ) . To investigate what was triggering the formation of EWs , we focused on the early stages of ectopic pouch formation . Wingless ( Wg ) plays a key role in specifying the wing pouch in early development and is also upregulated during regeneration ( Smith-Bolton et al . , 2009 ) . We compared Wg expression and apoptosis following damage in wild type ( Figure 3A–C ) and CtBPQ229/+ ( Figure 3D–F ) discs . At 0 hr of recovery , both wild type and CtBPQ229/+ discs had apoptotic cells and cellular debris ( which stain for DCP-1 ) and high Wg expression around the damaged wing pouch . In the wild type discs , the normal notum Wg stripe was maintained ( Figure 3A’–C’ ) . In contrast , in ~45% of CtBPQ229/+ discs ( n > 50 ) , there was a spot of increased Wg expression along the posterior edge of the notum , along with a disruption of the notum Wg stripe ( Figure 3D’ ) . Over the next 24 hr to 48 hr of recovery and regeneration , ~40% of CtBPQ229/+ wing discs showed a domain of Wg expression accompanying morphological changes associated with the outgrowth of a wing pouch from the posterior edge of the notum ( Figure 3E , F ) . Note that apoptosis was also detected in this region after 24 hr and 48 hr of recovery ( Figure 3E’ , F’ ) . During regeneration , wg expression is upregulated through a damage-dependent wg enhancer , which is JNK-dependent ( Schubiger et al . , 2010; Harris et al . , 2016 ) . We found that this damage-dependent wg enhancer ( BRV-B-GFP ) drives reporter expression at the area of fate change in a subset of CtBPQ229/+ discs , but not in wild type ( Harris et al . , 2016 ) , following rnts>egr damage ( Figure 3G , H ) . In addition , the removal of one copy of the damage-dependent wg enhancer , which is deleted in the wg1 allele ( Schubiger et al . , 2010 ) , reduced the frequency of ectopic wings ( Figure 3I ) . This is consistent with the ectopic pouch forming in a similar manner to the regenerating wing pouch ( Harris et al . , 2016 ) . It is known that ectopic wing pouches can be generated by the constitutive expression of UAS-wg along the anteroposterior compartment boundary in the absence of damage ( Ng et al . , 1996 ) . We determined that ptcts>wg expression initiated during the second instar triggered ectopic wings , but expression initiated during the third instar did not for both wild type and CtBPQ229/+ ( Figure 3—figure supplement 1 ) . This is in contrast to the fate change triggered by rnts>egr in the third instar and suggests that Wg alone is insufficient to reprogram cell fates as the disc matures . In agreement with this finding , the overexpression of rnts>wg alone , or co-expression of rnts>egr and >wg did not enhance the frequency of ectopic wings ( Figure 3J ) . Together , this suggests that other signaling pathways are critical for triggering ectopic wings at this point in development . Since the BRV-B wg enhancer is activated by the JNK pathway , we examined other indicators of increased JNK signaling: AP-1-RFP ( Chatterjee and Bohmann , 2012 ) and Matrix Metalloprotease 1 ( MMP1 ) protein ( Page-McCaw et al . , 2003 ) expression . At 0 hr of recovery , both AP-1-RFP and MMP1 were expressed robustly around the egr-ablated wing pouch , and in many CtBPQ229/+ discs , at a second location in the notum ( Figure 3—figure supplement 2 ) . AP-1-RFP activity was observed in an ectopic pouch at 48 hr of recovery ( Figure 3—figure supplement 2 ) , suggesting that JNK activity at the secondary site may be important for triggering ectopic wing pouch formation in CtBP-/+ mutants . Previous studies found that dilp8 was highly upregulated during leg-to-wing transdetermination ( Klebes et al . , 2005 ) and also in regenerating imaginal discs ( Katsuyama et al . , 2015; Skinner et al . , 2015; Harris et al . , 2016 ) . dilp8 encodes a secreted protein that is generated in response to abnormalities in tissue growth , likely in response to increased AP-1 activity , and Dilp8 expression delays pupariation ( Colombani et al . , 2012; Garelli et al . , 2012; Colombani et al . , 2015; Garelli et al . , 2015; Vallejo et al . , 2015; Jaszczak et al . , 2016 ) . A Minos insertion in the dilp8 locus , dilp8MI00727 , functions as a transcriptional reporter of dilp8 via the expression of GFP , which we will refer to as dilp8-GFP . Under normal undamaged conditions , there was very low expression of dilp8-GFP in the wing disc ( Figure 3K ) . As expected , damage with rnts>egr resulted in high dilp8-GFP expression around the damaged and regenerating wing pouch ( Figure 3L ) . In a subset of CtBPQ229/+ discs there was a second spot of dilp8-GFP expression in the notum at 0 hr of recovery ( Figure 3M ) , which largely overlapped with the area of increased Wg expression . After 72 hr of recovery , dilp8-GFP was still detectable in a morphologically distinct outgrowth , which by this stage expressed Wg in the characteristic pattern of a mature wing pouch ( Figure 3N ) . At 0 hr of recovery , dilp8-GFP expression in the notum overlapped with MMP1 expression ( Figure 3—figure supplement 2 ) , which is consistent with JNK activity contributing to dilp8-GFP expression . dilp8-GFP in the notum was expressed in cells that still expressed the hinge and notum marker Teashirt ( Tsh ) and before the pouch marker Nubbin ( Nub ) was detected ( Figure 3—figure supplement 3 ) suggesting that it may be an early marker of an impending fate change in notum cells . To determine if early dilp8-GFP expression in the notum was predictive of ectopic pouch formation , we quantified its frequency at different recovery time points in wild type and CtBPQ229/+ discs and the frequency of EWs per side in adults ( Figure 3O ) . The frequency of secondary dilp8-GFP spots was relatively constant across time points in CtBPQ229/+ discs and absent in wild-type discs . In addition , the frequency of EW per adult heminotum roughly matched the frequency of dilp8-GFP spots in discs . Therefore , in this system , the secondary spot of dilp8-GFP expression in the notum is likely an early predictor of future ectopic pouch and ectopic wing blade formation . Do the events that trigger ectopic wings occur independently between the two wing imaginal discs within one larva ? We recorded the number of CtBPQ229/+ adults with ectopic wings absent ( 63% ) , on one side ( 23% ) and on both sides ( 14% ) ( Figure 3—figure supplement 4 ) . Slightly more adults had bilateral ectopic wings than predicted for completely independent events . This could be due to identical timing of the discs or long-range signals ( circulating factors ) that simultaneously promote , or are permissive for , fate changes in both discs . In support of the idea of possible circulating factors promoting a permissive environment for ectopic pouch formation , we found that as animals homozygous for the loss-of-function allele , dilp8MI00727 , do not regenerate well ( as seen in [Skinner et al . , 2015] ) and also lacked ectopic wings ( EW ) ( Figure 3—figure supplement 5 ) . Conversely , we found that the addition of UAS-dilp8 enhanced the frequency of EWs , without triggering EWs on its own ( Figure 3—figure supplement 5 ) . Together this suggests that a developmental delay is important for EW formation . To determine if the ectopic pouch arises from a single cell changing its fate or from multiple cells , we generated marked clones of three varieties by hsFLP-induced recombination during the damage period and observed the clones following the growth of the ectopic pouch . The marked clones were arranged radially in the ectopic pouch yet only occupied a small fraction of it ( estimated to be 10% ) ( Figure 4A ) . Thus , the ectopic pouch was generated either from multiple cells simultaneously changing fate or by the progressive recruitment of additional cells . To determine if the cells that give rise to the ectopic pouch had originated in the notum , we lineage-marked cells using the lexA system . This allowed us to use the GAL4/UAS system to ablate the pouch and to permanently label cells that expressed the lexA lines with lexAOp-FLP , Ubi<stop<GFPnls . R76B02-lexA was expressed primarily in the notum in the absence of damage ( Figure 4—figure supplement 1 ) . Following regeneration , a large number of notum cells were GFP labeled in the control discs ( Figure 4B ) and a significant proportion of the cells in the ectopic pouch was GFP labeled in CtBP-/+ discs ( Figure 4C ) . Therefore , the cells that gave rise to the ectopic pouch once expressed R76B02-lexA , which indicates that these cells were once notum tissue , but had undergone a fate change to generate the ectopic pouch . Labeling with two additional lexA lines also supported the conclusion that a fate change occurred from notum to ectopic pouch ( Figure 4—figure supplement 1 ) . A key transcription factor in the development of the notum is Mirror ( Mirr ) ( Diez del Corral et al . , 1999 ) . The boundary between the notum and the hinge is located at a prominent fold at the edge of the mirr-lacZ expression domain ( Wang et al . , 2016 ) ( Figure 4D ) . We observed the ectopic Wg expression in the notum within the mirr-lacZ expression domain ( Figure 4E ) , and evidence of a new notum/hinge boundary in the ectopic growth ( Figure 4F ) . In addition , removing one copy of mirr increased the frequency of ectopic wings in both wild type and CtBP mutant backgrounds ( Figure 4G ) , suggesting that reducing the expression of notum-specifying factors increases the probability of damage-induced fate change , especially when CtBP levels are also reduced . The ectopic pouch was usually composed of both anterior ( A ) and posterior ( P ) cells , yet the initial events appeared to occur near the posterior edge of the disc , raising the possibility that posterior cells might have changed to an anterior identity . Indeed , when we examined dilp8-GFP expression in discs following ablation , with the A and P compartments visualized , we found that the entire cluster of dilp8-GFP expressing cells was usually located posterior to the compartment boundary ( Figure 4H ) . For discs at 0 hr of recovery with dilp8-GFP or AP-1-RFP expression in the notum , the expressing cells were scored as P in 17/20 discs , as A in 2/20 discs and as both A and P in 1/20 discs . Then , during the growth of the ectopic pouch , dilp8-GFP expression was observed in both A and P cells ( Figure 4I , J ) . When we labeled individual cells two days prior to the ablation and examined the clones that they generated , we found that even large clones in the ectopic pouch did not cross the compartment boundary ( n = 7 ) ( Figure 4K ) . Therefore , the ectopic pouch was likely generated from both A and P precursor cells . Taken together , these experiments show that the ectopic pouch arose from a group of cells in the notum that is composed of both A and P cells . These cells collectively underwent a notum-to-pouch fate change while preserving their compartmental identities . The appearance of the dilp8 expression first in the P compartment and later in the A compartment suggests that the initial fate change recruited cells from across the compartment boundary to generate the ectopic pouch . We have shown that JNK signaling was increased in the notum at an early stage in the development of an ectopic pouch . Moreover , our system is based on the targeted expression of Egr , which is the Drosophila Tumor-Necrosis Factor-α ( TNF-α ) ligand that activates JNK-dependent apoptosis ( Igaki et al . , 2002 ) in the wing pouch . Thus JNK-signaling may function in multiple ways to promote the notum-to-pouch fate change . To test if JNK signaling was required , we reduced JNK signaling either throughout the disc or only within the GAL4-expressing domain . The removal of one copy of the gene encoding the JNK kinase , hemipterous ( hep ) , reduced the frequency of ectopic wings ( Figure 5A ) . Blocking JNK signaling autonomously in the rn-GAL4 domain with the expression of a dominant negative form of JNK ( >JNKDN ) or of the AP-1 transcription factor Fos ( >FosDN ) completely suppressed the formation of ectopic wings ( Figure 5A ) , and also reduced the ablation of the wing pouch ( Figure 5—figure supplement 1 ) . Thus , JNK/AP-1 signaling was needed for the generation of ectopic wings , either through JNK-mediated apoptosis or other JNK-mediated signaling events . We found , however , that ablation of the pouch alone was not sufficient to produce ectopic wings , as the expression of the pro-apoptotic gene reaper ( rpr ) did not produce ectopic wings , even in a CtBPQ229/+ background ( Figure 5B ) . Although rnts>rpr leads to a more effective ablation of the wing pouch than rnts>egr , it does not activate JNK to the same levels ( Smith-Bolton et al . , 2009; Harris et al . , 2016 ) . We tested if cell-autonomous methods of increasing JNK activity would lead to ectopic wings by expressing either a wild-type or constitutively active form of hep ( rnts>hepwt , rnts>hepCA ) ( Figure 5B ) . Ectopic wings were only observed at very low frequency with rnts>hepCA , which could conceivably be the result of cell non-autonomous JNK activation via Egr expression ( Pérez-Garijo et al . , 2013 ) . Reducing the amount of apoptosis by shortening the ablation period from 40 hr to 24 hr reduced the frequency of EW for rnts>egr from ~45% to ~11% in CtBPQ229/+ , and did not produce any EWs for either rnts>rpr or rnts>hepCA ( Figure 5—figure supplement 2 ) . Egr signals through the receptor Grindelwald ( Grind ) to activate JNK signaling ( Andersen et al . , 2015 ) . Overexpression of the intracellular domain of Grind ( rnts>grindICD ) induced apoptosis , yet did not produce EWs ( Figure 5B ) . However , expression of a secreted form of Egr ( rnts>ecto-egr ) ( Narasimamurthy et al . , 2009 ) , caused EWs at a slightly higher frequency than the full-length ( membrane-tethered ) form ( Figure 5B ) . Thus , while the ablation of the pouch by Egr was necessary for the formation of EWs , some local spread of Egr also appeared necessary . In order to determine where egr must be expressed to trigger ectopic pouch formation , we closely examined the expression domain of rn-GAL4 with G-TRACE and detected expression in the wing pouch , a subset of the myoblasts and , at low levels , in scattered notum cells ( Figure 5—figure supplement 3 ) . Experiments with other GAL4 drivers determined that the expression of egr in the pouch was critical to the formation of ectopic wings and that the expression of egr in the notum or myoblasts was not sufficient ( Figure 5—figure supplement 3 ) . Egr signaling activates both JNK signaling and apoptosis . While it is difficult to separate high-JNK activity from apoptosis , reducing apoptosis either by removing one copy of the pro-apoptotic genes rpr , hid , and grim ( deleted in Df ( 3L ) XR38 ) , or by co-expressing a dominant-negative form of an effector caspase ( >DroncDN ) did not affect EW frequency ( Figure 5—figure supplement 4 ) . Co-expression of the anti-apoptotic gene UAS-p35 with rnts>egr , still generated EWs in the adults ( Figure 5—figure supplement 4 ) . Conversely , increasing apoptosis by co-expression of rnts>ecto-egr and rnts>rpr suppressed the formation of ectopic wings . This suggests that a population of cells that experienced increased JNK signaling without dying quickly was needed . Such cells in the pouch could be a source of diffusible factors that promote cell fate re-specification in the notum . We observed evidence for cells that experience Egr expression and yet survive in both the regenerated wing pouch and the ectopic pouch; cells forming the ectopic pouch may activate rn-GAL4 as they adopt a wing pouch fate ( Figure 5—figure supplement 5 ) . CtBP expression was uniform in undamaged and damaged wing discs ( Figure 5—figure supplement 6 ) . Rescue was achieved when additional CtBP was expressed throughout the disc , but not when only expressed in the wing pouch ( Figure 5—figure supplement 6 ) . In addition , the knockdown of CtBP in the rn-GAL4 domain during damage did not result in ectopic wings ( Figure 5—figure supplement 6 ) . Therefore CtBP likely functions outside the rn-GAL4 domain , possibly in the notum , to prevent damage-induced ectopic wings . Does a reduction in CtBP gene dosage affect JNK signaling ? CtBP mutant clones in wing imaginal discs autonomously upregulated AP-1 targets AP-1-GFP ( Figure 5G ) , dilp8-GFP ( Figure 5H ) , and MMP1-lacZ ( Figure 5I ) but did not cause detectable apoptosis or elevated levels of MMP1 protein ( Figure 5—figure supplement 7 ) . Interestingly , the activation of AP-1-GFP and dilp8-GFP was not uniform within the mutant clones or between clones , and clones in the hinge often showed higher levels of activation . Thus , reducing CtBP protein levels could enhance JNK signaling especially in some regions of the disc . Since the experiments with manipulations of the JNK pathway pointed to diffusible factors that were produced in response to JNK signaling , we examined the JAK/STAT pathway . Expression of several members of the Unpaired ( Upd ) family of cytokine ligands are promoted by JNK signaling ( Bunker et al . , 2015 ) . The co-expression of UAS-upd1 with rnts>egr led to a dramatic increase in the frequency of ectopic wings in both wild type and CtBPQ229/+ genetic backgrounds ( Figure 6A ) , up to ~17% in wild type and ~90% in CtBPQ229/+ . After regeneration , the wing imaginal discs that expressed rnts>egr and >upd1 showed a high percent ( 85% ) of large ectopic pouches ( n = 13 ) ( Figure 6B ) . During normal development , JAK/STAT signaling has been reported to occur throughout the wing imaginal disc at early stages and then becomes localized to the hinge region during L2/L3 ( Ayala-Camargo et al . , 2013; Hatini et al . , 2013; Johnstone et al . , 2013 ) . As shown by the reporter STAT-DGFP , which encodes a destabilized GFP , the activity of JAK/STAT signaling was significantly higher in early L3 discs than late L3 discs ( Figure 6C , D ) . After ablation of the pouch using rnts>egr in wild-type discs , STAT-DGFP expression was observed immediately surrounding the ablated pouch ( Figure 6E ) , as observed previously , ( La Fortezza et al . , 2016 ) , with very little STAT-DGFP expression in the notum . In contrast , in many CtBPQ229/+ discs , a patch of STAT-DGFP expression was detected in the notum ( Figure 6F ) , suggesting that JAK/STAT signaling may play a role in inducing a cell fate change . STAT-DGFP expression was often observed around sites of MMP1 expression . However , we observed that in some CtBPQ229/+ discs , only STAT-DGFP expression was detected , without a spot of MMP1 expression , suggesting that JAK/STAT activity in the notum can occur independently of ectopic JNK activation ( Figure 6—figure supplement 1 ) . If JAK/STAT activation occurs in the notum following damage to the pouch , where is the ligand being produced ? It is known that Unpaired-family ligands are upregulated following damage to discs ( Pastor-Pareja et al . , 2008 ) and during regeneration ( Katsuyama et al . , 2015; Santabárbara-Ruiz et al . , 2015; La Fortezza et al . , 2016 ) . We used the reporter upd3-lacZ ( Bunker et al . , 2015 ) , which is not expressed at high levels in undamaged late L3 discs , to investigate where Upd3 ligand is being produced ( Figure 6G ) . At 0 hr of recovery , upd3-lacZ expression was observed in the region of the ablated pouch in both wild type ( Figure 6H ) and CtBPQ229/+ discs ( Figure 6I ) , but not in the notum , even though STAT-DGFP expression was detected there ( Figure 6I’ ) . Since ligands of the Upd family are known to diffuse considerable distances , it is possible that the Upd ligands were generated by tissue damage and JNK activation in the pouch region and then diffused to the notum and activated JAK/STAT signaling . Indeed , the expression of UAS-upd1 in the myoblasts , which are basal to the notum epithelium , caused the activation of the JAK/STAT pathway in the notum and the disruption of the notum Wg stripe , but not the formation of an ectopic pouch ( Figure 6—figure supplement 2 ) . A similar disruption of the notum Wg stripe was observed following rnts>egr damage ( Figure 3D ) . Thus disruption of the notum Wg stripe appears insufficient to trigger ectopic pouch formation and other events seem necessary . If JAK/STAT activation is needed to form ectopic wings , then a decrease in JAK/STAT activity may suppress the frequency of ectopic wings . The removal of one copy of the JAK/STAT transcription factor Stat92E ( FRT82B Stat92E85C9 ) significantly decreased the frequency of ectopic wings ( EWs ) when compared to FRT82B ( which had a background rate of EWs ) in both a wild type and CtBPQ229/+ genetic background ( Figure 6J ) . The reduction in the levels of the Upd-family ligands , by removal of one copy of either upd2 or upd3 , or both upd2 and upd3 significantly reduced the frequency of ectopic wings ( Figure 6K ) . In addition , this suppression was completely rescued by the addition of UAS-upd1 , which indicates that the different Upd-family ligands are acting similarly in this situation . Together , this strongly indicates that the levels of JAK/STAT signaling can modulate the frequency of the notum-to-pouch cell fate change . Does CtBP modulate the JAK/STAT pathway ? First , we tested if upd3 expression was altered in CtBP-/- clones and found no change under undamaged conditions ( Figure 6—figure supplement 3 ) . Then we tested the STAT-DGFP reporter and found a slight increase in STAT signaling , although only in young discs and in regions where the pathway is normally active ( Figure 6—figure supplement 4 ) . Thus reducing CtBP levels may make tissues more responsive to the JAK/STAT pathway . The term ‘weak point’ has been used to describe regions prone to fate change or transdetermination , ( reviewed in [McClure and Schubiger , 2007] ) . This might be because specific parts of the disc are more sensitive than others to the effects of Upd family ligands . Consistent with this notion , the expression of UAS-upd1 along the A-P compartment boundary specifically activated dilp8-GFP expression in the notum , in a similar location to the observed fate change in the notum ( Figure 7B ) . In addition , this same area in the notum was specifically responsive to UAS-upd1 produced by random GAL4-expressing clones throughout the disc ( Figure 7C ) , as shown by strong dilp8-GFP expression in this notum region ( Figure 7D–F ) . dpp>upd1 only activated STAT-DGFP in specific regions of the notum and wing pouch , although there was an increase in apoptosis throughout the wing disc and adults showed notum defects consistent with a loss of proper notum patterning ( Figure 7—figure supplement 1 ) . Taken together , these experiments indicate that there is a region in the notum of the wing disc that is specifically and highly responsive to Upd ligands . However , the addition of UAS-upd1 without activation of JNK did not seem to be sufficient to trigger a fate change suggesting that JNK signaling and JAK/STAT signaling function together to promote re-specification of cell fates . Activation of two pathways , JNK/AP-1 and JAK/STAT , appears to be necessary for promoting cells in the notum to adopt a pouch fate . How does ablation of the pouch promote the activation of these pathways in the notum ? Neither ablation of the pouch using rpr nor activation of the Egr-receptor Grind in the pouch suffices . Thus , Egr is necessary and it either activates additional Grnd-independent pathways , or , more likely , activation of the Egr pathway in cells outside the rn-GAL4 expression domain is necessary . This could be both to activate JNK/AP-1 signaling locally around the damaged pouch and at a greater distance in the notum . The local signal to the hinge cells surrounding the wing pouch may produce the upd ligands , as these cells normally express upd1 . Egr may also signal directly to the notum , as a cleaved secreted ligand or from membrane fragments of cellular debris , much of which is trapped apically . In support of this idea , a secreted form of Egr is more effective than the membrane tethered form at inducing ectopic wings . Since reducing CtBP function increases AP-1 activity , the notum of CtBP-/+ discs will likely be more responsive to Egr signaling , leading to the activation of downstream AP-1 targets , such as dilp8 and wg . STAT-DGFP expression in the notum is increased in CtBP heterozygotes following ablation of the pouch . Increased JNK activity in the dying cells of the pouch would be expected to generate increasing levels of Upd-family ligands since they are upregulated in regenerating discs ( Katsuyama et al . , 2015; Santabárbara-Ruiz et al . , 2015; La Fortezza et al . , 2016 ) and regulatory elements of these genes have AP-1-binding sites ( Bunker et al . , 2015 ) . Consistent with their production by dying cells is the observation that more rapid killing of pouch cells by co-expression of rpr reduces notum-to-pouch transformations . Increased STAT activity in the notum reduces notum-specific patterns of gene expression and promotes a hinge-like fate ( Ayala-Camargo et al . , 2013 ) . Consistent with this , we found that reducing gene dosage of mirr , a homeodomain-containing protein expressed in the notum , increased the frequency of ectopic wings . A recent study reported the generation of ectopic wings simply by expression of UAS-upd1 in the notum ( without directed tissue damage ) ( Recasens-Alvarez et al . , 2017 ) . However , this study induced expression in the second larval instar ( day 5 at 18°C ) and did not report if expression of UAS-upd1 at later time points could also induce ectopic wings . Of note , we have found that expression of high levels of UAS-upd1 generates considerable levels of apoptosis . Conceivably , the generation of ectopic wings in that report could therefore also have resulted from the combined activation of the JAK/STAT and AP-1 pathways . Imaginal discs have been hypothesized to have ‘weak points’ , where transdetermination occurs most often following physical fragmentation ( Schubiger , 1971 ) , ( reviewed in [McClure and Schubiger , 2007] ) . Other cell fate transformations are also more likely to occur in particular regions of the discs ( Salzer and Kumar , 2010 ) . For leg-to-wing transdetermination in the prothoracic leg disc , the weak point is located where two different morphogens , Wg and Dpp , are co-expressed ( Johnston and Schubiger , 1996; Maves and Schubiger , 1998 ) . In the wing disc , Dpp is expressed along the anteroposterior ( AP ) compartment boundary and Wg , is expressed along part of the dorsoventral ( DV ) compartment boundary through the wing pouch . The AP and DV boundaries likely intersect at two locations – one in the middle of the wing pouch and in the notum near the region where we observe ectopic pouch formation . In addition , the notum Wg stripe in the L3 disc generates an additional location where both Wg and Dpp are expressed . Therefore , it is possible that similar to the weak point in the leg disc , the presence of both these morphogens makes cells more prone to change fate , especially following damage . This region of the notum could be predisposed to an increase in STAT activity . When UAS-upd1 is expressed using dpp-GAL4 , STAT-DGFP expression is not observed along the entire GAL4 domain . Rather , it is mostly restricted to the pouch , and importantly , a region of the notum . These regions are more responsive to Upd-family ligands than other parts of the disc and the area in the notum is likely the weak point for notum-to-pouch fate change . Recent studies have described regions in wing discs that are tumor hot-spots ( Tamori et al . , 2016 ) and are resistant to apoptosis ( Verghese and Su , 2016 ) and found that they also have elevated JAK/STAT activity . Thus , parts of the disc that have elevated STAT activity may be more capable of surviving damage and then re-specifying cell fates . During normal development , the initial specification of the wing pouch occurs towards the end of the first larval instar . At this time STAT is active through most of the wing disc and wg is expressed in a patch in the anterior and ventral portion of the disc . Subsequently , wg expression expands from its initial region of expression when cells surrounding the nascent pouch are recruited to a pouch fate by a feed-forward mechanism that is fueled by Wg ( Wu and Cohen , 2002; Zecca and Struhl , 2007b , 2010 ) . In the absence of Wg expression , the pouch is not specified . In our system , wg expression from the damage-responsive enhancer is important for triggering ectopic wings . However , we found that additional Wg ( rnts>egr , >wg ) , if at all , decreased the frequency of ectopic wings and that Wg alone was incapable of transforming a more mature notum to an ectopic pouch; therefore JAK/STAT and AP-1 activity appear necessary . Based on our observations and those of others , JAK/STAT activity can reduce the expression of notum-specific genes ( Ayala-Camargo et al . , 2013 ) . This likely transforms that part of the notum to a less committed state . Thus , a region with elevated STAT activity leads to relatively uncommitted notum cells and within this region , the activation of AP-1 promotes Wg expression via a damage-responsive enhancer , which may create conditions similar to the initial specification of the wing pouch at earlier stages of development ( Figure 7D ) . Once the ectopic pouch has been specified , the nascent pouch could turn on the rn-GAL4 driver , even briefly , prior to the downshift . Then egr expression would occur resulting in localized apoptosis and JNK activation . The surrounding cells would activate AP-1 and STAT and also express wg as occurs in the regeneration blastema surrounding rnts>egr ablated tissue . This would then have the effect of reinforcing the conditions that are necessary for notum-to-pouch fate change and the generation of a new wing pouch . While this paper was under review , another paper was published that gives support to damage-induced fate changes relying on JAK/STAT and Wg signaling ( Verghese and Su , 2017 ) . We have shown that clones of cells lacking CtBP function have elevated levels of AP-1 activity and , in a few cases in specific portions of the disc , increased STAT activity . CtBP is a co-repressor that interacts with a number of transcriptional repressors via a PxDLS motif including Brinker ( Hasson et al . , 2001 ) and Hairy ( Poortinga et al . , 1998 ) and hence multiple CtBP-containing complexes could be relevant for this role of CtBP . Intriguingly , some of the isoforms of the Drosophila Fos ortholog , Kayak , which heterodimerizes with Jun to form the AP-1 transcription factor , have PxDLS motifs . While AP-1 has mostly been studied as a transcription factor that activates gene expression upon phosphorylation of Jun by JNK , in the absence of sufficient activating signals , it may repress inappropriate expression of AP-1 target genes via CtBP . In this way , CtBP could buffer cell fates from destabilization caused by fluctuations in the level of JNK activity . The stocks that were used in this study include: rn-GAL4 , tub-GAL80ts , rn-GAL4 , tub-GAL80ts , UAS-egr and rn-GAL4 , tub-GAL80ts , UAS-rpr ( Smith-Bolton et al . , 2009 ) , hh-GAL4 , ptc-GAL4 , tub-GAL80ts , nub-GAL4; tub-GAL80ts , UAS-egr/TM6B-GAL80 , dpp-GAL4 ( BL:1553 ) , R15B03-GAL4 ( BL:49261 ) , R76A01-GAL4 ( BL:46953 ) ( Pfeiffer et al . , 2008 ) , G-TRACE ( BL:28280 , BL:28281 ) and Ubi<stop<GFPnls ( BL:32250 , BL:32251 ) ( Evans et al . , 2009 ) , Act5C>FRT . CD2>GAL4 , UAS-RFP ( BL:30558 ) , vgQE-RFP ( Zecca and Struhl , 2007a ) , hs-FLP; Act<stop<lacZnls , Ubi<stop<GFPnls ( Worley et al . , 2013 ) , UAS-CtBP ( Bhambhani et al . , 2011 ) , Genomic CtBP construct ( gCtBP ) ( Zhang and Arnosti , 2011 ) , UAS-ecto-eiger ( Narasimamurthy et al . , 2009 ) , UAS-grindICD ( Andersen et al . , 2015 ) , MMP1-LacZ ( Uhlirova and Bohmann , 2006 ) , AP-1-GFP and AP-1-RFP reporters ( Chatterjee and Bohmann , 2012 ) , upd3-lacZ ( Bunker et al . , 2015 ) , upd3 . 1-lacZ ( Jiang et al . , 2011 ) , BRV-B-GFP ( Harris et al . , 2016 ) , UAS-DroncDN ( UAS-DroncC318S ) ( Hawkins et al . , 2000 ) , FRT82B , FRT82B CtBPQ229* and FRT82B Stat92E85C9 . Stocks obtained from Bloomington stock center include: CtBP87De-10 ( BL:1663 ) , CtBP03463 ( BL:11590 ) , UAS-CtBPRNAi ( BL:31334 , BL:32889 ) , mirr1486 ( BL:23935 ) , mirr1825 ( BL:23928 ) , hepr75 ( BL:6761 ) , UAS-FosDN ( UAS-Fra . Fbz , BL:7214 ) , UAS-JNKDN ( UAS-bskK53R , BL:9311 ) , UAS-p35 ( BL:5072 , 5073 ) , dIlp8MI00727 ( BL:33079 ) , UAS-wg ( BL:5918 , BL:5919 ) , UAS-hepCA ( BL:6406 ) , lexAOp-FLP ( BL:55820 ) , R81E08-lexA ( BL:54377 ) , R76B02-lexA ( BL:54118 ) , R76B06-lexA ( BL:54225 ) , R15C03-lexA ( BL:52490 ) , 10XSTAT92E-GFP ( BL:26197 ) and 10XSTAT92E-DGFP ( BL:26199 ) ( Bach et al . , 2007 ) , Df ( 3R ) Exel8157 ( BL:7973 ) , and upd2Δ ( BL:55727 ) , upd2Δupd3Δ ( BL:55729 ) , upd3Δ ( BL:55728 ) ( Osman et al . , 2012 ) . Genetic ablation experiments were conducted as described in ( Smith-Bolton et al . , 2009 ) . Briefly , eggs were collected on grape plates and animals were density controlled by picking 55 L1 larvae into vials with yeast paste . Shifts were conducted on particular days ( day 3 - day 9 ) AEL from an 18°C incubator to 30°C incubator . Cultures were shifted on day 7 AEL for 40 hr unless otherwise noted . Experimental replicates were done on separate days . Adults were scored as ( 1 ) No EW for no ectopic wings or ( 2 ) EW for either one or two ectopic wings , except in ( Figure 3 ) , where adults were scored for having zero , one , or two ectopic wing ( s ) . Graphs show average frequency of EWs for experiments done on separate days . Error bars show standard deviations between the experiments . To generate CtBP mutant clones , hs-FLP;;FRT82B RFPnls was crossed to FRT82B CtBPQ229* and vials were heat shocked at 37°C for 1 hr to induce mitotic clones . To generate randomly marked clones , hs-FLP;;rnts>egr/TM6B , Tub-GAL80 was crossed to Act5C<stop<lacZnls , Ubi<stop<GFPnls; CtBPQ229*/TM6B and the cultures were staged as described above . A brief 10 min 37°C heat-shock was used to generate flip-out clones at a relatively low density . Without a 37°C heat shock , the 30°C temperature shift for the genetic ablation experiments did not produce hs-FLP induced clones , which we also noted in ( Smith-Bolton et al . , 2009 ) . Imaginal discs were fixed in 4% paraformaldehyde for 15 min , washed and permeabilized in phosphate-buffered saline solution with 0 . 1% Triton X-100 , and blocked in either 10% Normal Goat Serum or Normal Donkey Serum depending on staining . The following antibodies were used from Developmental Studies Hybridoma Bank ( DSHB ) : mouse anti-Wg ( 1:100 , 4D4 ) ; mouse anti-Mmp1 ( 1:100 , a combination of 14A3D2 , 3A6B4 and 5H7B11 ) ; mouse anti-En ( 1:10 , 4D9 ) ; rat anti-Ci ( 1:10 , 2A1 ) ; mouse anti-Cut ( 1:200 , 2B10 ) ; mouse anti-Ptc ( 1:50 , Apa-1 ) ; and mouse anti-Ubx ( 1:20 , FP3 . 38 ) . The following antibodies were gifted: rat anti-Pdm2 ( Chris Doe ) ; rat anti-Twist ( 1:1000 , Eric Wieschaus ) ; and mouse anti-Nub and rabbit anti-Tsh ( Stephen Cohen ) . The following antibodies are from commercial sources: goat anti-dCtBP ( 1:50 , dN-20 Santa Cruz Biotechnology , Dallas TX ) ; rabbit anti-DCP-1 ( 1:250 , Cell signaling ) ; rabbit anti-GFP ( 1:500 , Torrey Pines Laboratories , Secaucus , NJ ) ; chicken anti-GFP ( 1:500 , ab13970 Abcam , Cambridge , UK ) ; and rabbit anti-β-galactosidase ( 1:1000 , #559762; MP Biomedicals , Santa Ana , CA ) . Secondary antibodies were from Cell Signaling . Nuclear staining was by DAPI ( 1:1000 , Cell Signaling ) . Images were obtained on either a Leica TCS or a Zeiss LSM 700 . Images were processed using ImageJ ( Fiji ) ( Schindelin et al . , 2012 ) . All scale bars are 100 μm . For scoring discs with dilp8-GFP expression in the notum or ectopic pouch at different time points of recovery , in ( Figure 3 ) , images were collected from discs from biological replicates ( cultures started on separate days ) and imaged on Zeiss Axio Imager M1 without staining with anti-GFP . Images were collected and then scored for the presence or absence of a second spot of dilp8-GFP expression . We generated CRISPR guide RNAs to regions conserved in all CtBP isoforms , using the method described in FlyCas9 system ( Kondo and Ueda , 2013 ) . We generated stable transgenetics using the U6-promoter driving the guide RNAs inserted at the attP40 landing site . The guide RNAs were optimized to work in our wtiso genetic background , as we sequenced the predicted regions guide RNA binding , detected single nucleotide polymorphisms as compared to the FlyBase consensus , and incorporated these changes into the guide RNA sequence . We used the transgenic services provided by BestGene ( Chino Hills , CA ) . We generated the following stocks to generate CRISPR mutations on particular third chromosomes: Sp/Cyo , nos-Cas9; wtiso and Sp/Cyo , nos-Cas9; rn-GAL4 , UAS-eiger , tub-GAL80ts , which were crossed to U6-CtBP-guideRNA; TM2/TM6B in order to recover CRISPR generated mutations on particular third chromosomes . Single males carrying possible CtBP alleles were tested for the failure to complement viability when crossed to a deletion that spans the CtBP gene , Df ( 3R ) Exel8157 . The newly generated CtBP alleles were then sequenced to determine the nucleotide change near the guide RNA cut site .	Some animals are more able to replace damaged tissue than others . A salamander , for example , can re-grow an amputated limb but a mouse or human cannot . After damage or injury certain types of cells are lost and need to be replaced by cells that are left behind . The remaining cells – or new cells that develop from them – must change their characteristics to better resemble the lost cells . This property , known as plasticity , needs to be controlled tightly . Excessive plasticity can result in forming tissues that are completely inappropriate for that location in the animal . The fruit fly Drosophila melanogaster can be used to investigate plasticity during regeneration . Fruit fly larvae contain structures known as imaginal discs that can regenerate if damaged . Occasionally , when the imaginal discs regenerate , they produce the wrong kind of tissue . Worley et al . set out to look for genes that would normally prevent such mistakes . Their search began with looking for flies with mutations that caused regeneration to go awry following damage . Specifically , Worley et al . looked for mutant flies that grew extra wings after a structure was damaged that would normally only generate a single wing . Once such flies had been found , further experiments were used to narrow down the search and confirm which gene was mutated . This approach revealed that flies with mutations in the gene for a protein called CtBP ( which is short for C-terminal binding protein ) made more errors during regeneration and commonly regenerated inappropriate structures such as an extra wing . Importantly , mammals have very similar genes , but few researchers had previously studied if they also play a role in regeneration . Worley et al . went on to show that CtBP dampens the activity of two signaling pathways ( namely the JNK/AP-1 pathway and the JAK-STAT pathway ) , both of which promote plasticity . Thus , when CtBP levels are reduced , there is excessive plasticity . These findings implicate CtBP as a regulator of plasticity during regeneration . This is an important first step in thinking of strategies that would allow researchers to guide and reshape the development of tissues during regeneration .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "stem", "cells", "and", "regenerative", "medicine" ]	2018	CtBP impedes JNK- and Upd/STAT-driven cell fate misspecifications in regenerating Drosophila imaginal discs
The export of mRNA from nucleus to cytoplasm requires the conserved and essential transcription and export ( TREX ) complex ( THO–UAP56/DDX39B–ALYREF ) . TREX selectively binds mRNA maturation marks and licenses mRNA for nuclear export by loading the export factor NXF1–NXT1 . How TREX integrates these marks and achieves high selectivity for mature mRNA is poorly understood . Here , we report the cryo-electron microscopy structure of the human THO–UAP56/DDX39B complex at 3 . 3 Å resolution . The seven-subunit THO–UAP56/DDX39B complex multimerizes into a 28-subunit tetrameric assembly , suggesting that selective recognition of mature mRNA is facilitated by the simultaneous sensing of multiple , spatially distant mRNA regions and maturation marks . Two UAP56/DDX39B RNA helicases are juxtaposed at each end of the tetramer , which would allow one bivalent ALYREF protein to bridge adjacent helicases and regulate the TREX–mRNA interaction . Our structural and biochemical results suggest a conserved model for TREX complex function that depends on multivalent interactions between proteins and mRNA . Eukaryotic protein-coding mRNA is matured in the nucleus before it is exported and translated in the cytoplasm . During maturation , mRNA is capped , spliced , and poly-adenylated to form fully processed and packaged messenger ribonucleoprotein complexes ( mRNPs ) ( Köhler and Hurt , 2007; Singh et al . , 2015; Heath et al . , 2016; Stewart , 2019; Xie and Ren , 2019 ) . The transcription and export ( TREX ) complex is recruited during transcription ( Heath et al . , 2016; Viphakone et al . , 2019 ) to maturing mRNPs through mRNA and protein interactions at the mRNP 5’-end , splice junctions , and mRNP 3’-end ( Cheng et al . , 2006; Merz et al . , 2007; Gromadzka et al . , 2016; Shi et al . , 2017 ) . TREX subsequently loads the global mRNA-export factor NXF1–NXT1 , to license mRNAs for export ( Strässer and Hurt , 2001; Köhler and Hurt , 2007; Hautbergue et al . , 2008; Taniguchi and Ohno , 2008 ) . By chaperoning the nascent mRNA , TREX also inhibits the formation of harmful DNA-RNA hybrids , called R-loops , and protects genome integrity ( Luna et al . , 2019; Pérez-Calero et al . , 2020 ) . The TREX complex is found in all eukaryotes and contains the multi-subunit THO complex , the DEXD-box RNA helicase UAP56/DDX39B ( yeast Sub2 ) , and an RNA export adapter such as ALYREF ( yeast Yra1 ) ( Strässer et al . , 2002 ) . The human THO complex comprises six subunits , THOC1 , −2 , –3 , −5 , –6 , and −7 , of which four have known counterparts in the yeast Saccharomyces cerevisiae ( Sc ) : THOC1 ( yeast Hpr1 ) , −2 ( yeast Tho2 ) , −3 ( yeast Tex3 ) , and −7 ( yeast Mft1 ) ( Heath et al . , 2016; Mitchell et al . , 2019 ) . Additional TREX interactors include SARNP ( yeast Tho1 ) , the mammalian protein ZC3H11A and ALYREF-like proteins UIF , LUZP4 , POLDIP3 , and CHTOP ( Dufu et al . , 2010; Heath et al . , 2016 ) . In this study we focus on the conserved TREX complex ( Heath et al . , 2016; Xie and Ren , 2019 ) : THO–UAP56/DDX39B–ALYREF , and hereafter refer to UAP56/DDX39B as UAP56 . In the current model of mRNA export , the THO complex is recruited to the mRNP and delivers UAP56 , which subsequently ‘clamps’ the mRNA . THO–UAP56 binds the export adapter ALYREF and together they promote loading of the export factor NXF1–NXT1 onto mRNA ( Hautbergue et al . , 2008; Köhler and Hurt , 2007; Strässer and Hurt , 2001; Taniguchi and Ohno , 2008 ) . However , the structural basis for TREX activities remains poorly understood despite active study ( Peña et al . , 2012; Pérez-Alvarado et al . , 2003; Ren et al . , 2017; Shi et al . , 2004; Xie and Ren , 2019 ) . Here , we present the cryo-electron microscopy ( cryo-EM ) structure of the human THO–UAP56 complex at 3 . 3 Å resolution . We resolved the THO architecture , its binding mode to UAP56 , and can explain mutations linked to human disease ( Heath et al . , 2016 ) . Our results show that THO–UAP56 adopts a tetrameric 28-subunit architecture , suggesting how TREX may bind multiple mRNA and mRNP regions at the same time . Taken together , the data reveal a conserved mechanism for TREX function that depends on multivalent protein–mRNA interactions . The THO–UAP56 28-subunit tetramer consists of two asymmetric dimers , 1 and 2 , each comprising two seven-subunit THO–UAP56 monomers , A and B . Using this nomenclature , the tetramer contains monomers 1A , 1B , 2A , and 2B ( Figures 1b , c and 2 ) . Each monomer is partitioned into two regions , one formed by THOC1 , −2 , –3 , and UAP56 and the second by THOC5 , −6 , and −7 . THOC1 , −2 , –3 , and UAP56 adopt the same architecture in all monomers , whereas THOC5 , −6 , –7 assume different conformations to assemble the dimer via THOC5 and THOC7 and the tetramer via THOC5 and THOC6 . THOC5 and THOC7 form a parallel coiled coil , whose C-terminal ends from monomer A and B meet in a four-helix bundle , yielding the dimerization region ( Figures 2 and 3a , b ) . The dimer is further stabilized by contacts between the N-terminal THOC5 tandem RWD ( tRWD-N ) domains from monomers A and B that pack against each other . The THOC5–THOC7 coiled coil is interrupted by a ‘hinge’ in each monomer , which generates three mobile sub-regions in the 14-subunit THO–UAP56 dimer: monomer A , monomer B , and the above mentioned dimerization region ( Figures 2 and 3a ) . Notably , the THOC5–THOC7 coiled coil measures ~ 350 Å across the human dimer , suggesting that distal positioning of monomers A and B may play a role in TREX function . The THO–UAP56 tetramer assembles from dimer 1 and 2 via two interfaces that involve the THOC5 and THOC6 subunits . The first interface is formed by homotypic interactions between the THOC5 tRWD-C domains from monomers 1A and 2A . The second is formed by the THOC6 β-propeller from monomers 1A and 2A that interact with the THOC1 , −5 , and −7 subunits from monomers 2B and 1B , respectively ( Figure 3a–c ) . THOC6 thereby associates with the neighboring dimer and causes THOC1 and the THOC5–THOC7 coiled coil to adopt different positions ( Figure 3c ) . This interaction stabilizes the tetramer and at the same time leads to the asymmetry within the THO–UAP56 dimer ( Figure 3c ) . In agreement with this architecture , the recombinant THO complex lacking THOC6 assembles into a dimer rather than a tetramer , shown by its slowed migration in sucrose density gradients ( Figure 2—figure supplement 1a ) . Deletion of the THOC5 and THOC7 subunits leads to loss of THOC6 , as expected , and results in a monomeric THOC1/2/3 complex ( Figure 2—figure supplement 1a ) . Nine residues in THOC6 are known to be mutated in human disease ( Heath et al . , 2016 ) , all of which map to the THOC6 β-propeller . These mutations are predicted to destabilize the β-propeller fold or the THOC5–THOC7 interaction and would thereby disrupt THO tetramerization ( Figure 2—figure supplement 1b ) . Indeed , four mutations were previously shown to reduce nuclear THOC6 levels and perturb THOC6 interaction with the THO complex ( Mattioli et al . , 2019 ) . The core of the THO–UAP56 monomer is formed by the THOC2 subunit , which comprises an extended helical repeat with five distinct domains that we name ‘anchor’ , ‘bow’ , ‘MIF4G’ , ‘stern’ , and the disordered ‘charged domain’ ( CD , truncated for recombinant expression ) ( Peña et al . , 2012; Figures 1a and 4a ) . The N-terminal THOC2 anchor helices bind α-helices THOC5 α3 and THOC7 α2 and connect via a disordered linker to the THOC2 bow domain . The THOC2 bow is sandwiched by the helical THOC1 ‘dock’ domain ( Figure 3b ) . Hence , the THOC1 , −5 , and −7 subunits jointly bind THOC2 , and THOC1 orients the extended and mobile THOC2 helical repeat ( bow , MIF4G , stern ) ( Figures 1c and 2 , Figure 1—figure supplements 1d and 3c ) . The THOC2 bow and adjacent MIF4G domain bind the THOC3 β-propeller ( Figure 4a ) , and the THOC2 MIF4G domain makes additional , extensive contacts to the UAP56 helicase ( Figures 1a and 4a ) . UAP56 contains two ATPase lobes , RecA1 and RecA2 . We observed density for the UAP56 RecA2 lobe , but not RecA1 , which remains mobile in the absence of RNA and ATP ( Shi et al . , 2004 ) . Consistent with the structure , THO oligomeric state did not impact UAP56 binding in a pulldown experiment ( Figure 2—figure supplement 1c ) , whereas point mutations in the THOC2 MIF4G domain at the UAP56 interface did impair UAP56 binding ( Figure 2—figure supplement 1d , e ) . The THOC2 MIF4G domain adjoins the curved THOC2 stern domain , which in turn connects to the disordered CD implicated in nucleic acid binding ( Peña et al . , 2012 ) . Near the THOC2 stern , we observed weak density for the THOC1 C-terminus ( Figure 4a ) , which can bind the mRNA-export factor in yeast and humans ( Hobeika et al . , 2007; Viphakone et al . , 2012 ) . The close positioning of UAP56 , the THOC1 C-terminus , and the THOC2 CD suggests that this region is involved in export factor loading . Supporting this model , we observed that the recombinant core THOC1/2/3–UAP56 complex is sufficient to bind the NXF1–NXT1 export factor in vitro ( Figure 2—figure supplement 1f ) . Previous data showed that the isolated THOC5 subunit can also bind NXF1–NXT1 in vitro ( Katahira et al . , 2009; Viphakone et al . , 2012 ) . However , THOC5 did not contribute to NXF1–NXT1 binding in the context of multi-subunit THO–UAP56 complexes ( Figure 2—figure supplement 1f ) , consistent with its distant location from the putative export factor loading site in the THO–UAP56 structure ( Figures 1c and 4a ) . The THO–UAP56 tetramer is ~260 Å long , ~290 Å high , and ~150 Å wide . Subunits involved in mRNA and export factor binding are located at the ends of the complex ( THOC1 , −2 , –3 , and UAP56 ) , away from those involved in oligomerization ( THOC5 , −6 , and −7 ) ( Figures 1c and 2 , Figure 1—figure supplement 6a ) . Whereas all THO subunits are essential genes ( Dempster et al . , 2019 ) , the requirement for THOC5 , −6 , and −7 suggests that THO oligomerization may be needed for normal function . The extended shape and flexibility of the THO–UAP56 tetramer ( Figure 4—figure supplement 1d ) could allow for multiple interactions with spatially distant mRNA regions and mRNP maturation marks , such as the cap-binding complex or exon-junction complex ( Cheng et al . , 2006; Merz et al . , 2007 ) and facilitate NXF1–NXT1 loading from multiple sites . ALYREF ( yeast Yra1 ) is part of the conserved TREX complex ( Chávez et al . , 2000; Strässer et al . , 2002 ) , is required for transfer of mRNA to the export factor NXF1–NXT1 ( Hautbergue et al . , 2008; Taniguchi and Ohno , 2008 ) , and is essential for viability in yeast ( Strässer and Hurt , 2000; Stutz et al . , 2000 ) and humans ( Dempster et al . , 2019 ) . ALYREF contains two UAP56-binding motifs ( UBMs ) at its N- and C-terminus that can separately bind UAP56 ( Hautbergue et al . , 2009; Figure 4—figure supplement 1a , b ) . Between the UBMs is an RG-rich region , followed by a central RNA-recognition motif ( RRM ) domain , and a second RG-rich region ( Gromadzka et al . , 2016 ) . To gain insight into ALYREF function , we prepared a homology model of UAP56 bound to RNA , an ATP-analog , and the C-terminal ALYREF UBM ( C-UBM ) based on a yeast crystal structure ( Ren et al . , 2017 ) . We then superimposed this model onto the human THO–UAP56 tetramer structure via the four UAP56 RecA2 lobes ( Figure 4b , Figure 4—figure supplement 1d , e ) . The resultant model reveals how ALYREF may function in the TREX complex bound to mRNA . First , the model suggests that a single ALYREF protein may reach across the two THO dimers to bind two juxtaposed UAP56 helicases . The ALYREF N- and C-UBMs would each bind one UAP56 RecA1 lobe ( Hautbergue et al . , 2009 ) , located 20 Å apart , from THO–UAP56 monomers 1A and 2B ( or 2A and 1B ) ( Figure 4b ) . Alternatively , ALYREF could bridge two UAP56 proteins within one dimer , which are ~200 Å apart , or possibly multiple THO–UAP56 tetramers . Bridging by ALYREF can explain why deletion of either N- or C-UBM leads to growth defects in yeast ( Strässer and Hurt , 2001; Zenklusen et al . , 2001 ) and suggests that the UBMs and associated RG-rich domains contribute to the avidity of TREX–mRNA binding , consistent with biochemical data ( Hautbergue et al . , 2008; Ren et al . , 2017 ) . Furthermore , RNA binding by UAP56 ( monomers 1A and 2A ) would orient the mobile RecA1 lobe and project conserved negatively charged residues of the interacting ALYREF UBM towards a conserved patch of positively charged residues of the juxtaposed UAP56 RecA2 lobe ( from monomers 1B and 2B ) ( Figure 4b ) . This RecA2 surface is frequently bound by regulators in other DEXD-box helicases ( Linder and Jankowsky , 2011; Buchwald et al . , 2013; Sharif et al . , 2013; Mathys et al . , 2014; Figure 4—figure supplement 1g ) and suggests how a single UBM could be sufficient for partial ALYREF function ( Strässer and Hurt , 2001; Zenklusen et al . , 2001; Golovanov et al . , 2006; Gromadzka et al . , 2016 ) . Notably , these two models of ALYREF function are not mutually exclusive and collectively suggest that the ALYREF UBMs bridge UAP56 helicases , which may correctly position other ALYREF regions for export factor loading . The functionally related ALYREF-like proteins contain one ( LUZP4 , POLDIP3 , UIF ) or two ( CHTOP ) closely spaced UBMs ( Gromadzka et al . , 2016; Hautbergue et al . , 2009; Viphakone et al . , 2015; Figure 4—figure supplement 1c ) , and thus are unlikely to bridge two helicase RecA1 lobes in the THO–UAP56 structure ( Figure 4—figure supplement 1c ) . Nevertheless , we speculate that the THO–UAP56 tetramer increases the affinity of mRNA-export adaptors to their target mRNAs through avidity , by providing multiple UAP56 binding sites in close proximity . UAP56 is a member of the DEXD-box RNA helicase family . These helicases function as ‘clampases’ that couple ATP hydrolysis to RNA release and can be regulated by interacting proteins ( Linder and Jankowsky , 2011 ) . Biochemical data from yeast ( Ren et al . , 2017 ) and humans ( Taniguchi and Ohno , 2008; Shen et al . , 2007 ) show that ALYREF weakly promotes UAP56 ATPase activity , and the yeast THO complex further stimulates this ( Ren et al . , 2017 ) . We observed that the human THOC1/2/3 core complex is sufficient to promote UAP56 ATPase activity in vitro , consistent with our structure and other MIF4G–DEXD-box helicase systems ( Hilbert et al . , 2011; Montpetit et al . , 2011; Figure 4—figure supplement 2a , b ) . Taken together , we propose that THO and ALYREF regulate the activity of two bridged UAP56 helicases within the TREX–mRNA complex to control export factor loading onto mRNA ( Hautbergue et al . , 2008; Taniguchi and Ohno , 2008 ) . To gain insights into conservation of the TREX complex , we re-analyzed a previously reported 6 . 0 Å resolution crystal structure of the six-subunit yeast Sc THO–Sub2 complex ( Ren et al . , 2017; Figure 4—figure supplement 3 , Materials and methods ) . The yeast structure was proposed to be a six-subunit monomer , within which a single Tex1 ( THOC3 ) and a single Sub2 ( UAP56 ) subunit were assigned , but not Tho2 , Hpr1 , Thp2 , or Mft1 . Surprisingly , our revised yeast model revealed that the crystal contained a 12-subunit THO–Sub2 dimer , which could better explain the published electron density and phosphotungsten cluster positions ( Figure 4—figure supplement 3a–c , Video 2 ) . We could assign two copies of Tho2 , Hpr1 , the Thp2–Mft1 coiled coil , and place one additional copy of the Tex1 and Sub2 subunits based on ( i ) comparisons to the human THO–UAP56 structure , ( ii ) the known homologies of yeast Hpr1 , Tho2 , Tex1 , and Mft1 with human THOC1 , −2 , –3 , and −7 , and ( iii ) the predicted Thp2 coiled coil ( Söding et al . , 2005; Figure 4—figure supplement 3d , Materials and methods ) . The revised model allowed us to identify Thp2 and confirm Mft1 as the respective yeast homologs of the THOC5 and THOC7 coiled coils . This indicates an unexpectedly high degree of structural conservation of the five-subunit THO complex monomer ( THOC1 , −2 , –3 , −5 , –7 ) and its dimerization via coiled coils . Notably , the THO–Sub2 model shows that the two Sub2 helicases are located ~80 Å apart . Both bind to their respective Tho2 MIF4G domain in monomers A and B , analogous to the human THOC2–UAP56 interaction . Thus , also in yeast a single Yra1 could bridge two Sub2 helicases via its N- and C-UBM regions , mirroring human ALYREF ( Figure 4—figure supplement 3d ) . Our results suggest a unified model for TREX function through multimerization of the THO–UAP56 complex ( Figure 4c ) . THO–UAP56 forms a dimer in yeast ( Figure 4—figure supplement 3 ) and a constitutive tetramer in humans , mediated by the additional THOC5 tRWD domain and THOC6 subunit ( Figure 1 , Figure 1—figure supplement 1f–h , Figure 4—figure supplement 3f ) . In humans , THO–UAP56 tetramerization may have evolved in response to an increased complexity in gene architecture and mRNP composition ( Singh et al . , 2015 ) . The conserved THO–UAP56 multimerization , extended architecture , and flexibility ( Figure 1—figure supplement 1b–d , Figure 4—figure supplement 1d ) indicate that one complex can bind multiple mRNP maturation marks and mRNA regions simultaneously . Subsequent to THO–UAP56 recruitment to the mRNP , a single ALYREF ( or yeast Yra1 ) molecule could bridge two juxtaposed UAP56–mRNA complexes and , potentially , multimerize itself through its RG-rich regions ( Gromadzka et al . , 2016 ) . In the assembled TREX–mRNP complex UAP56 ATPase activity is stimulated and , together with ALYREF or alternative export adaptors , may load the export factor NXF1–NXT1 from multiple sites within the complex . Taken together , the data support a model where requirements for both the selectivity and efficiency of mRNP recognition and mRNA-export licensing are fulfilled through multivalent interactions between proteins and mRNA . For heterologous co-expression of the human 6-subunit THO complex in insect cells , the open reading frames ( ORFs ) of THOC1 , −2 residues 1–1203 , −3 , –5 , −6 , –7 with an N-terminal 10xhistidine tag on THOC2 , an N-terminal 3xV5 tag on THOC1 , and a TwinStrepII tag on THOC3 were cloned into a modified pACEBac1 vector ( Geneva Biotech ) by Golden Gate cloning ( Engler et al . , 2008 ) . THOC2 was truncated for co-expression at its disordered C-terminus ( residues 1204–1593 ) for an improved biochemical behavior ( Figure 1—figure supplement 1a ) . The human UAP56 ORF was cloned into a pOPINB vector with an N-terminal 6xhistidine tag for expression in E . coli . For endogenous purification of THO–UAP56 from human K562 cells , the THOC1 cDNA and a C-terminal 3C-AID-GFP tag were cloned into a lentiviral vector backbone ( Addgene plasmid #31485 ) , yielding a plasmid containing pRRL-SFFV-THOC1-3C-AID-GFP . The human heterodimer NXF1–NXT1 ORFs with an N-terminal 10xhistidine-3xflag tag on NXF1 and , separately , the human NUP214 FG-repeat ( residues 1916–2033 ) ORF with an N-terminal Maltose-binding protein fusion and C-terminal 10xhistidine tag were cloned into modified pACEBac1 vectors as done for the THO complex . Recombinant THO complex was co-expressed in insect cells using a plasmid containing all six subunits ( THOC1 , −2 residues 1–1203 , −3 , –5 , −6 , –7 ) . The THO plasmid was electroporated into DH10EMBacY cells to generate bacmids ( Trowitzsch et al . , 2010 ) that were then transfected into Spodoptera frugiperda Sf9 cells to generate a V0 virus . The V0 virus was further amplified in Sf9 cells to yield V1 virus . For protein expression , 750 mL Trichoplusia ni Hi5 cells at a density of 1 × 106 cells were infected with 1 mL V1 virus and harvested after 3 days by centrifugation . Harvested cells were resuspended in buffer A ( 50 mM Tris-HCl pH 8 . 0 , 300 mM NaCl , 5% ( w/v ) glycerol , 20 mM imidazole , 1 mM dithiothreitol ( DTT ) , 0 . 5 mM PMSF , cOmplete EDTA-free protease inhibitor cocktail ( Roche ) ) and lysed by sonication . The lysate was clarified by ultracentrifugation and the supernatant loaded on a HisTrap HP 5 mL column ( GE Healthcare ) , equilibrated in buffer B ( 10 mM Tris-HCl pH 8 . 0 , 300 mM NaCl , 5% ( w/v ) glycerol , 20 mM imidazole , 1 mM DTT ) . The column was washed with a linear gradient from 20 mM to 100 mM imidazole and eluted with buffer B containing 350 mM imidazole . Peak fractions were diluted 1:1 with buffer B lacking NaCl and imidazole and applied to anion exchange using a HiTrapQ HP 5 mL column ( GE Healthcare ) , equilibrated in buffer C ( 25 mM HEPES pH 7 . 9 , 150 mM NaCl , 5% ( w/v ) glycerol , 2 . 5 mM DTT ) . The complex was eluted with a linear gradient of buffer C from 150 to 1000 mM NaCl . Fractions containing the THO complex were concentrated , loaded on a HiLoad 16/600 Superdex 200 pg column ( GE Healthcare ) equilibrated in buffer D ( 25 mM HEPES pH 7 . 9 , 250 mM NaCl , 5% ( w/v ) glycerol , 1 mM TCEP ) . The purified THO complex was concentrated to 4 mg mL−1 , flash frozen and stored at −80°C . Variant THO complexes , lacking subunits THOC6 or containing mutations , were purified as described above . The THOC1/2/3 complex was purified as above , except that buffer A and B contained 500 mM NaCl , the complex was eluted from the HisTrap HP 5 mL column using a linear gradient from 0–50% with buffer B containing 500 mM NaCl and 250 mM Imidazole , and that buffer D lacked TCEP . The identity of all THO subunits was confirmed by mass spectrometry . Full-length human UAP56 was expressed in E . coli BL21 DE3 RIL cells grown in autoinduction media at 37°C for 16 hr . Cells were lysed by sonication in buffer E ( 25 mM HEPES pH 7 . 9 , 500 mM NaCl , 5% ( w/v ) glycerol , 20 mM imidazole , 1 mM DTT , 0 . 5 mM PMSF , 0 . 1% ( v/v ) Tween-20 , cOmplete EDTA-free protease inhibitor cocktail ) . The supernatant was clarified by centrifugation and loaded on a HisTrap HP 5 mL column , equilibrated in buffer F ( 25 mM HEPES pH 7 . 9 , 500 mM NaCl , 5% ( w/v ) glycerol , 20 mM imidazole , 1 mM DTT ) . The column was washed with buffer F and eluted with a linear gradient from 20 to 250 mM imidazole . The 6xhistidine tag was cleaved by PreScission protease at 4°C overnight , and UAP56 was then diluted to 50 mM NaCl using buffer F lacking NaCl and imidazole and applied to anion exchange using a HiTrapQ HP 5 mL , equilibrated in buffer G ( 25 mM HEPES pH 7 . 9 , 50 mM NaCl , 5% ( w/v ) glycerol , 2 . 5 mM DTT ) . The protein was eluted with a linear gradient from 50 to 400 mM NaCl . Fractions containing UAP56 were concentrated and loaded on a HiLoad 16/600 Superdex 75 pg column , equilibrated in buffer H ( 25 mM HEPES pH 7 . 9 , 100 mM NaCl , 5% ( w/v ) glycerol , 2 . 5 mM DTT ) . Purified UAP56 was concentrated to 11 mg mL−1 , flash frozen and stored at −80°C . To reconstitute the THO–UAP56 complex , the THO complex was mixed with a two-fold molar excess of UAP56 and incubated at 20°C for 30 min in buffer I ( 25 mM HEPES pH 7 . 9 , 100 mM NaCl , 5% ( w/v ) glycerol , 1 mM TCEP ) . We used GraFix ( Kastner et al . , 2008 ) to obtain a homogenous and crosslinked THO–UAP56 complex for EM studies . 90 µg of THO–UAP56 complex were loaded onto a 4 mL 10–40% ( w/v ) sucrose gradient in buffer J ( 25 mM HEPES pH 7 . 9 , 50 mM KCl , 1 mM TCEP , 0–0 . 05% glutaraldehyde ) and ultracentrifuged at 91 , 100 x g in a SW60 Ti swing bucket rotor ( Beckman ) for 20 hr 30 min at 4°C . Peak fractions were pooled and the crosslinking reaction was quenched for 15 min using a final concentration of 50 mM Lysine . The sample was then buffer-exchanged using Zeba Spin 7K MWCO Desalting Columns that were equilibrated in buffer K ( 25 mM HEPES pH 7 . 9 , 50 mM KCl , 2% ( w/v ) glycerol , 1 mM TCEP ) . The sample was concentrated using a Vivaspin 500 centrifugal concentrator ( MWCO 100 kDa ) and immediately used for EM studies . NXF1–NXT1 virus generation and co-expression in insect cells were performed as for the THO complex ( see above ) . Cells were harvested by centrifugation and resuspended in buffer L ( 50 mM HEPES pH 7 . 9 , 300 mM NaCl , 10% ( w/v ) glycerol , 1 mM DTT , cOmplete EDTA-free protease inhibitor cocktail ) before sonication . The cell lysate was ultracentrifuged and the supernatant applied to a HisTrap HP 5 mL column , equilibrated in buffer M ( 25 mM HEPES pH 7 . 9 , 300 mM NaCl , 10% ( w/v ) glycerol , 30 mM imidazole , 1 mM DTT ) . The column was washed with 45 mM imidazole and eluted with buffer M containing 300 mM imidazole . Peak fractions were diluted 1:2 with buffer M lacking imidazole and containing 5% ( w/v ) glycerol and 10 mM NaCl , and were loaded on a HiTrap Heparin HP 5 mL column ( GE Healthcare ) , equilibrated in buffer N ( 25 mM HEPES pH 7 . 9 , 50 mM NaCl , 5% ( w/v ) glycerol , 1 mM DTT ) . The complex was eluted with a linear gradient of buffer N from 50 to 1000 mM NaCl . Fractions containing the NXF1–NXT1 heterodimer were concentrated to 1 mg mL−1 , flash frozen , and stored at −80°C . The NUP214 FG-repeat ( residues 1916–2033 ) was expressed in insect cells as done for the THO complex above . The harvested cells were lysed by sonication in buffer A , and applied to a HisTrap HP 5 mL column , equilibrated in buffer B . The column was washed with buffer B and then with buffer B containing 600 mM NaCl . The protein was eluted in buffer B containing 300 mM imidazole and dialyzed overnight against 1 L buffer C containing 200 mM NaCl . The dialyzed protein was loaded onto a HiLoad 16/600 Superdex 200 pg column , equilibrated in buffer C containing 200 mM NaCl . Fractions containing the NUP214 FG-repeat protein were pooled , concentrated to 14 mg mL−1 , flash frozen , and stored at −80°C . For endogenous purification of THO–UAP56 , lentiviral particles carrying the THOC1-3C-AID-GFP construct were generated using Lenti-X cells ( Takara ) by polyethylenimine transfection ( Polysciences ) of the viral plasmid and helper plasmids pCMVR8 . 74 ( Addgene plasmid #22036 ) and pCMV-VSV-G ( Addgene plasmid #8454 ) , according to standard procedures . K562 ( DSMZ ) cells were infected at limiting dilutions and GFP-positive cells were isolated using a BD FACSAria III cell sorter ( BD Biosciences ) . Viral integration was confirmed by immunoblotting for THOC1 and GFP ( Figure 1—figure supplement 1f ) . To prepare nuclear extract ( NE ) , 15 L of human K562 cells were grown to a density of 1 × 106 cells mL−1 at 37°C , stirred at 80 rpm and 5% CO2 . The NE was prepared as previously described ( Mayeda and Krainer , 1999 ) and dialyzed against buffer O ( 20 mM HEPES , pH 7 . 9 , 100 mM KCl , 20% ( w/v ) glycerol , 0 . 2 mM EDTA , 2 mM DTT ) . Lenti-X ( Takara ) and K562 ( DSMZ ) cells tested negative for mycoplasma . NE from wild type and THOC1-3C-AID-GFP containing human K562 cells was diluted 1:1 with buffer P ( 40 mM HEPES pH 7 . 9 , 15 mM KCl , 6 mM MgCl2 , 2 mM DTT ) . For immunoprecipitation , the NE was first incubated for 1 hr at 4°C with or without 2 µg Benzonase per mL NE , as indicated in Figure 1—figure supplement 1f , and then incubated on a rotating wheel for 2 . 5 hr at 4°C with GFP-Trap Agarose resin ( Chromotek ) that was previously equilibrated with buffer Q ( 20 mM HEPES pH 7 . 9 , 100 mM KCl , 2 mM MgCl2 , 8% ( w/v ) glycerol , 0 . 05% ( v/v ) Igepal CA-630 , 1 mM DTT ) . After five washes , the beads were eluted by boiling and the samples were applied to SDS–PAGE , transferred onto a PVDF membrane ( ThermoScientific ) and probed with anti-THOC1 primary antibody ( HPA019096 , Merk , dilution 1:500 ) . Goat anti-mouse IgG-HRP ( W4021 , Promega , dilution 1:10000 ) was used as a secondary antibody . Antibody detection was performed with Amersham ECL Select Western Blotting Detection Reagent ( GE Healthcare ) and a ChemiDoc MP imaging system ( Bio-Rad Laboratories ) . The THOC1-3C-AID-GFP K562 NE was treated with 1 µg Benzonase per mL NE for 12 hr at 4°C , to release THO–UAP56 complexes from the pellet fraction and substantially increase the total yield . THOC1 was immunoprecipitated as described above ( Western blot ) and eluted for 2 hr at 4°C with 3C PreScission Protease . The eluate was loaded onto a 10–50% w/v sucrose step gradient containing 0–0 . 05% glutaraldehyde in buffer T ( 20 mM HEPES pH 7 . 9 , 100 mM KCl , 2 mM MgCl2 , 2 mM TCEP ) , and centrifuged for 16 hr at 105 , 600 x g in a SW60 Ti rotor ( Beckman coulter ) . Fractions containing the THO–UAP56 complex were quenched for 15 min using a final concentration of 50 mM lysine , pooled , concentrated in a 0 . 5 mL 100 kDa MWCO Amicon concentrator ( Sigma ) and immediately used for EM grid preparation . We measured steady-state UAP56 ATPase activity using a NADH-coupled ATPase assay , essentially as described ( Montpetit et al . , 2012 ) . The assay was set up at the final concentrations of 5 U/mL rabbit muscle pyruvate kinase , Type III ( Sigma-Aldrich ) , 5 U/mL rabbit muscle L-lactic dehydrogenase , Type XI ( Sigma-Aldrich ) , 500 µM phosphoenolpyruvate and 50 µM NADH . Reactions were prepared in buffer R ( 25 mM HEPES pH 7 . 9 , 100 mM NaCl , 25 mM KCl , 10 mM MgCl2 , 5% ( w/v ) glycerol , 1 mM ATP ) and contained the indicated combinations of proteins and RNA at the following concentrations: 2 µM UAP56 , 2 µM THOC1/2/3 , 10 µM poly-uridine 15 nucleotide RNA . The decay of NADH emission signal was monitored over time at 37°C in a PHERAstar FS ( BMG LABTECH ) , using a 0 . 03–100 µM NADH dilution series as calibration standard . Average UAP56 ATPase rates from triplicate experiments were calculated from linear slopes of NADH decay as hydrolyzed molecules of ATP s−1 per enzyme . To test whether THO complex oligomeric state influences the THO–UAP56 interaction , we immobilized equimolar amounts of the recombinant THOC1/2/3 complex ( 3 . 3 µg ) , THO∆THOC6 complex ( 4 . 5 µg ) and THO complex ( 5 µg ) via the 10xhistidine tag on THOC2 on magnetic nickel beads ( Promega ) together with 1 µg recombinant human UAP56 ( two-fold molar excess over THO ) for one hour at 4°C in buffer S ( 25 mM HEPES pH 7 . 9 , 50 mM KCl , 5% ( w/v ) glycerol , 20 mM imidazole , 0 . 01% ( v/v ) Igepal CA-630 ) . The beads were washed five times with 1 mL buffer S , and proteins were eluted for one hour at 20°C using 500 mM imidazole in buffer S and visualized by SDS-PAGE ( Coomassie blue ) ( Figure 2—figure supplement 1c ) . To probe the THOC2–UAP56 interface , 5 µg of THO complex or three variants ( THOM1: THOC2 Y551A , K554S , R555S , K558S; THOM2: THOC2 K589A , Y590S , N592A; THOM3: combined THOM1 and THOM2 mutants ) were immobilized , incubated with 1 µg UAP56 , washed in buffer T ( 25 mM HEPES pH 7 . 9 , 100 mM KCl , 5% ( w/v ) glycerol , 20 mM imidazole , 0 . 01% ( v/v ) Igepal CA-630 ) , eluted and visualized by SDS-PAGE ( Coomassie blue ) ( Figure 2—figure supplement 1e ) . To assess binding between the NXF1–NXT1 export factor and the THO complex , we immobilized 4 µg MBP-tagged NUP214 FG-repeat peptide on amylose resin ( New England Biolabs ) together with NXF1–NXT1 ( 1:1 molar ratio ) in buffer U ( 20 mM HEPES pH 7 . 9 , 100 mM NaCl , 10% ( w/v ) glycerol , 0 . 1% ( v/v ) Igepal CA-630 ) . Each binding reaction additionally contained equimolar amounts of THO , THO∆THOC6 , and THOC1/2/3 complex relative to NXF1–NXT1 and these were incubated together at 4°C on a rotating wheel . The beads were washed three times with 1 mL buffer U , and eluted for 1 hr on ice in buffer U containing 12 mM maltose . The elutions were visualized by SDS-PAGE ( Coomassie blue ) . Each pulldown was repeated in triplicates . For negative stain EM imaging of the endogenous THO–UAP56 complex , copper grids were coated with a ~ 5 nm homemade carbon film and glow-discharged . 4 µL sample was applied to the grid and incubated for 1 min . The grid was blotted and washed four times with 4 µL distilled water , stained for 1 min in 4 µL 2% ( w/v ) uranyl-acetate solution and blotted until dry . 1660 micrographs were acquired using SerialEM ( Mastronarde , 2005 ) on a FEI Tecnai G2 20 transmission electron microscope ( Eagle 4 k HS CCD camera ) operated at 200 keV at a nominal magnification of 50 , 000x ( 2 . 21 Å pixel−1 ) and a defocus range of –1 µm to –1 . 5 µm . 60938 particles were picked and extracted with a 2562 pixel box size using WARP 1 . 07 ( Tegunov and Cramer , 2019 ) and transferred to RELION 3 . 1 ( Scheres , 2012 ) for 2D classification using default settings . 4 µL concentrated and crosslinked THO–UAP56 complexes were applied to glow-discharged Cu R1 . 2/1 . 3 300 mesh holey carbon grids ( Quantifoil ) . Grids were blotted at 4°C and 70% humidity and plunged into liquid ethane using a Leica EM GP . Cryo-EM data was recorded using SerialEM ( Mastronarde , 2005 ) in two sessions ( data sets 1 and 2 ) on a FEI Titan Krios G3i operated at 300 keV , equipped with Gatan K3 direct electron detector . Datasets 1 ( 7482 movies ) and 2 ( 18821 movies ) were acquired with a defocus range of –0 . 4 to –3 . 7 µm at a nominal magnification of 105 , 000x ( 0 . 86 Å pixel−1 ) . The camera was operated in ‘super-resolution’ mode ( 0 . 43 Å pixel−1 ) with an exposure time of 3 . 8 s and 33 frames per micrograph , a dose rate of 9 . 73 e– pixel−1 s−1 and a total dose of 50 e– Å−2 . Movie alignment with 4 × 6 patches , dose-weighting , and contrast transfer function ( CTF ) parameter estimation were all carried out in WARP 1 . 07 ( Tegunov and Cramer , 2019 ) for the separate datasets 1 and 2 . Automated particle picking was performed with a re-trained neural network in WARP 1 . 07 ( Tegunov and Cramer , 2019 ) , yielding 414 , 082 and 1 , 156 , 183 particles for datasets 1 and 2 , respectively . The particles were then extracted , normalized , and Fourier cropped to 1 . 34 Å pixel−1 with a 4402 pixel box size in RELION 3 . 1 ( Scheres , 2012 ) . The gold-standard Fourier shell correlation ( FSC ) ( 0 . 143 criterion ) was used to determine resolution , and B-factors were estimated and applied in RELION 3 . 1 ( Scheres , 2012 ) . The initial 3D model of the THO–UAP56 complex was determined from the first 30 , 000 particles from dataset one with an ab initio refinement in cryoSPARC 2 . 0 ( Punjani et al . , 2017 ) using default parameters and two classes . The resultant class one was filtered to 60 Å and used as reference for separate 3D refinements in RELION 3 . 1 ( Scheres , 2012 ) of all particles from each dataset , in order to align all particle images to a common reference . To increase data set size , we separately extracted two THO–UAP56 tetramer units from each octamer using the particle extraction parameters as above . This was possible since the octamer apparently does not show particle orientations with two tetramers on top of each other ( Figure 1—figure supplement 1c ) , yielding 828 , 164 and 2 , 312 , 366 tetramer units for datasets 1 and 2 , respectively . Subsequent 3D classification was carried out without image alignment to identify homogenous particle groups ( Figure 1—figure supplement 2 ) . To identify high-quality tetramer units we classified each dataset in 3D into eight classes using a soft-edged mask in the shape of the THO–UAP56 complex monomers 1A and 2B generated with the volume eraser in UCSF Chimera ( Pettersen et al . , 2004 ) and RELION 3 . 1 ( Scheres , 2012 ) . 3D class 8 ( Round 1a ) and class 7 ( Round 1b ) were selected for their excellent density quality , resulting in 195 , 098 high-quality tetramer units for subsequent processing , resulting in the final density maps A-E ( Figure 1—figure supplements 2–4 ) . First , we carried out a 3D refinement of the high-quality tetramer data set with the THO–UAP56 monomer 1A/2B mask , yielding a density ( map B ) with an overall resolution of 3 . 3 Å and a B-factor of −132 Å2 . To better resolve the connecting density to THO–UAP56 monomers 1B and 2A , we prepared a soft mask enveloping the complete tetramer , yielding a refinement from the same particles to 3 . 9 Å and a B-factor of −176 Å2 ( map A ) . THO–UAP56 monomer 1A THOC2 , −3 , –5 , −7 and UAP56 regions remained poorly resolved , and were subjected to a focused refinement with a soft-edged mask surrounding these parts , to a resolution of 4 . 6 Å and a B-factor of −235 Å2 ( map D ) . THO–UAP56 monomer 1B THOC5 and THOC6 as well as the four-helix bundle connecting to monomers 1A and 1B were improved in a focused refinement of this region to resolution of 4 . 7 Å and a B-factor of −244 Å2 ( map E ) . To locally improve the density of THO–UAP56 monomer 2B THOC2–THOC3–UAP56 , we performed an additional round of 3D classification ( Round 2 ) yielding a subset of particles that could be subsequently refined to a resolution of 4 . 7 Å and a B-factor of −160 Å2 ( map C ) . We used ResMap ( Kucukelbir et al . , 2014 ) to estimate the local resolution ( Figure 1—figure supplement 3c ) and performed 3D variability analysis of the high-quality tetramer units ( 6 . 2% of the total data ) in cryoSPARC 2 . 0 ( Punjani et al . , 2017; Figure 4—figure supplement 1d ) . We prepared a composite THO–UAP56 complex model by combining the cryo-EM densities A-E ( Figure 1—figure supplements 1d , 2 and 3 ) . The structure was manually built in COOT ( Emsley and Cowtan , 2004 ) using THOC2 MIF4G domain and THOC3 homology models and the previously determined crystal structure of the UAP56 RecA2 lobe ( Shi et al . , 2004 ) . The model coordinates were refined into the respective sharpened maps B and D in PHENIX ( Adams et al . , 2010 ) using the phenix . real_space_refine routine , applying secondary structure and rotamer restraints . We first built monomers 1A and 2B . THOC12B ( residues 10–392 ) , THOC22B ( residues 164–288 ) , THOC52B ( residues 47–227 ) , THOC72B ( residues 22–181 ) , THOC51A ( tRWD ) , and THOC61A were modeled into map B . The THOC22B anchor helices were putatively assigned and modeled as poly-alanine with unknown register . To build the THOC21A MIF4G domain ( residues 535–687 ) we first prepared a homology model of the CWC22 MIF4G-like domain crystal structure ( Buchwald et al . , 2013 ) ( PDB ID 4C9B ) , and then fitted this into map D and manually adjusted and extended the model to THOC21A residue 288 . The THOC21A C-terminal residues 688–1175 ( stern ) were assigned based on density connectivity in map D , and were modeled as poly-alanine owing to a lower local resolution of ~5–6 Å ( Figure 1—figure supplement 3c ) . Note that the connectivity between THOC21A residues 688–895 remains uncertain . The THOC31A homology model was generated with MODELLER ( Webb and Sali , 2017 ) based on the WDR61 crystal structure ( Xu and Min , 2011 ) ( PDB ID 3OW8 ) , rigid-body fitted into map D , and refined in real space using the Relax protocol in ROSETTA ( Tyka et al . , 2011 ) . The human UAP561A RecA2 structure ( Shi et al . , 2004 ) ( PDB ID 1XTJ ) was rigid-body fitted into map D and the interface with THOC2 was adjusted . To extend monomer 2B , THOC21A ( residues 240–1175 ) –THOC31A–UAP561A were superimposed on THOC22B as a rigid body , and this fit is in agreement with map C . To extend the monomer 1A , refined THOC12B , THOC22B ( residues 164–287 ) , THOC52B ( residues 47–227 ) , and THOC72B were fitted into map D as rigid bodies . Further , THOC51A ( tRWD ) , and THOC61A were rigid body fitted into map B , into the density of monomer 1B , and adjusted at the THOC51A–THOC51B tRWD domain interface . To generate the complete THO–UAP56 complex model the refined monomer 1A , monomer 1B THOC5 tRWD and THOC6 , and monomer 2B models were fitted into map A into the symmetry related positions of monomers 2A , 2B , and 1B , respectively , using COOT ( Emsley and Cowtan , 2004 ) . The THOC51A/1B–THOC71A/1B four-helix bundle was modeled as poly-alanine into maps A and E and refined , and then copied as a rigid body into the THOC52A/2B–THOC72A/2B position . ADP refinement was carried out in a composite THO–UAP56 map , generated from the individual maps A-E using phenix . combine_focused_maps ( Adams et al . , 2010 ) . The final model comprises 28 proteins . Based on our THO–UAP56 model , we revised the 6 . 0 Å resolution model of the chimeric yeast THO–Sub2 ( PDB ID 5SUQ ) crystal structure ( Figure 4—figure supplement 3a , Supplementary file 1 ) . We first defined a putative THO–Sub2 monomer A model that comprised residues 8–5289 ( chain M ) from the crystal structure . We then superimposed this monomer A model onto residues 6170–8550 ( chain M ) in COOT ( Emsley and Cowtan , 2004 ) , showing an excellent fit to the electron density and revealing the presence of a second monomer B . Four pieces of additional evidence validate this re-interpretation of the yeast crystal structure . First , placement of monomer A into the monomer B position could explain unmodeled density ( Figure 4—figure supplement 3b ) . Second , the positions of phosphotungsten are consistent with a dimer architecture: M8701 and M8702 are now bound to equivalent positions in monomer A and B; N8701 binds the monomer B Sub2 RecA2 lobe in an equivalent position to A501 ( Figure 4—figure supplement 3b ) . Third , weak density is observed in the location of the second Sub2 RecA2 lobe . Fourth , completion of the yeast THO-Sub2 dimer model with the monomer B Tho2 Stern , Tex1 and Sub2 fits into the asymmetric unit ( Figure 4—figure supplement 3c ) and is compatible with the observed crystal packing . The Tho2 C-terminus , corresponding to the human THOC2 Stern , is not resolved in monomer B . This might be due to the flexibility observed between Tho2 MIF4G and Stern domains ( joint 2 , Figure 4—figure supplement 3d ) and the lack of stabilizing crystal packing contacts that are present in monomer A . We also see no density for monomer B Tex1 , however , based on the 1:1 stoichiometry of Tex1 ( or THOC3 ) in yeast ( Ren et al . , 2017 ) and human THO complexes ( Figure 1 , Figure 1—figure supplement 1 ) , its presence is likely . Overall , minor differences remain between the yeast and human THO–UAP56 monomer models that may stem from species differences and/or the unique THO–Sub2 architecture captured in the crystal ( Ren et al . , 2017 ) . Figures were made with UCSF Chimera X ( Goddard et al . , 2018 ) and PyMol ( https://www . pymol . org ) .	The DNA of human and other eukaryotic cells is stored inside a compartment called the nucleus . DNA carries the genetic code and provides a blueprint for all of the cell’s proteins . However , protein production occurs outside the nucleus , in the main body of the cell . To transmit genetic information from one compartment to the other , the DNA sequences are first transcribed into another molecule called messenger RNA , or mRNA for short . Once made , mRNA exits the nucleus and enters the cell’s main body to encounter the machinery that translates its sequence into a protein . Before mRNA can exit the nucleus , it must first undergo a series of modifications , which result in the mRNA molecule being successively bound to specific proteins . Once mRNA has passed through these steps , it is recognized by the transcription-and-export complex , or TREX for short , which is comprised of several proteins . When TREX binds to mRNA , it adds on a final protein which allows the mRNA molecule to be transported out of the nucleus . However , it remained unclear how TREX selects the completed mRNA-protein complexes that are ready for export while at the same time recognizing the wide variety of mRNA molecules produced by cells . Now , Pühringer and Hohmann et al . have identified the first three-dimensional structure of the core of the human TREX complex using a technique called cryo-electron microscopy . This revealed that the seven proteins of the TREX core assemble into a large complex that has four copies of each protein . The structure suggests that TREX can bind to mRNA and its attached proteins in various ways . These different binding arrangements may help the complex select which mRNA molecules are fully modified and ready to be exported . The structure also sheds light on how mutations in this complex can lead to diseases such as Beaulieu–Boycott–Innes syndrome ( BBIS ) . This work will help guide future research into the activity of TREX , including how its structure changes when it binds to mRNA and deposits the final transport protein . Identifying these structures will make it easier to design experiments that target specific aspects of TREX activity and provide new insights into how these complexes work .	[ "Abstract", "Introduction", "Results", "and", "discussion", "Materials", "and", "methods" ]	[ "chromosomes", "and", "gene", "expression", "structural", "biology", "and", "molecular", "biophysics" ]	2020	Structure of the human core transcription-export complex reveals a hub for multivalent interactions
The concerted evolution of morphological and behavioral specializations has compelling examples in ant castes . Unique to ants is a marked divergence between winged queens and wingless workers , but morphological specializations for behaviors on the ground have been overlooked . We analyzed thorax morphology of queens and workers in species from 21 of the 25 ant subfamilies . We uncovered unique skeletomuscular modifications in workers that presumably increase power and flexibility of head–thorax articulation , emphasizing that workers are not simply wingless versions of queens . We also identified two distinct types of queens and showed repeated evolutionary associations with strategies of colony foundation . Solitary founding queens that hunt have a more worker-like thorax . Our results reveal that ants invest in the relative size of thorax segments according to their tasks . Versatility of head movements allows for better manipulation of food and objects , which arguably contributed to the ants’ ecological and evolutionary success . A detailed understanding of morphology is of prime importance to elucidate how organisms evolved and operate in nature . This is especially so in an era of increasingly sophisticated developmental genetic analysis , as the correct interpretation of molecular data depends largely on the precise characterization of morphological structures ( e . g . , Prud’homme et al . , 2011 vs Yoshizawa , 2011; Mikó et al . , 2012 ) . During development , the relative investment in the growth of different body parts ( e . g . , allocation of nutritional resources to somatic vs germ tissues ) will determine adult morphologies , and thus influence an organism’s ecology ( Nijhout and Emlen , 1998; Emlen , 2001 ) . Morphology interacts very closely with behavior in shaping phenotypic evolution ( Baldwin , 1896; Simpson , 1953; Robinson and Dukas , 1999 ) . On the one hand , changes in behavior will often influence the environment in which organisms are selected , leading to modifications of morphology ( Wcislo , 1989; Crispo , 2007 ) . On the other hand , morphological specializations can open the potential for further behavioral change ( West-Eberhard , 2003 ) . Specializations that associate morphology and behavior have compelling examples in insect polyphenisms , where alternative morphologies result from environmental regulation of development and are typically associated with distinct behavioral repertoires ( Beldade et al . , 2011; Simpson et al . , 2012 ) . For example , horned and hornless male beetles produced as a result of nutritional plasticity have different reproductive tactics ( guarding vs sneaking access to females in nests [Moczek and Emlen , 2000] ) . In many social insects , differential feeding leads to the production of distinct queen and worker castes , each with characteristic morphology and behavior underlying reproductive vs maintenance functions within the colony ( Wheeler , 1986; Beldade et al . , 2011 ) , and increasing colony performance as a whole ( Oster and Wilson , 1978 ) . Among the social Hymenoptera , ants are an extreme case of caste polyphenism , because queens are usually winged and workers are always wingless ( Wilson , 1971; Hölldobler and Wilson , 1990 ) . Flight allows queens to disperse from the natal nests before they start new colonies , while the lack of wings in workers is thought to facilitate the exploitation of ground habitats and cramped spaces ( Hölldobler and Wilson , 1990 ) . The presence and operation of wings is tightly associated with the morphology of the thorax . In the typical thorax of modern flying insects , the first segment ( T1 ) bears no dorsal appendages , while the second ( T2 ) and third ( T3 ) each bear a pair of wings ( Snodgrass , 1935 ) . Because of this , studies of morphological specializations of the insect thorax have focused on the wing-bearing segments T2 and T3 . The relative size of these segments varies widely across insect orders , but tends to be conserved within ( Dudley , 2002 ) . Surprisingly , the entire thoracic skeletomuscular architecture of ant castes , including the T1 segment that forms the articulation with the head , has been neglected , from both functional and comparative perspectives . In this study , we use a phylogenetically broad comparative approach , involving queens and workers from species representing 21 of the 25 extinct and extant ant subfamilies , to investigate external morphology and internal anatomy in the context of caste-specific specialized behaviors . Our analysis reveals a unique modification of the thoracic architecture in worker ants , presumably connected with their powerful head and mandibles , and uncovers two types of thoracic configurations in queens , associated to different strategies for the foundation of new colonies . We assessed the relative sizes and configuration of the dorsal plates that form the thoracic exoskeleton in 265 queens and workers belonging to 11 species in five major ant subfamilies ( Table 1 ) . For each caste of each species , we measured the length of T1 , T2 , and T3 of 5–17 individuals from museum collections . Our morphometric analyses showed that in ant queens , both T1 and T3 are reduced relative to T2 , which makes up most of the thorax ( Figure 1A ) . This conforms to the typical proportions in insects where flight is powered exclusively by large wing muscles inside T2 ( Snodgrass , 1935; Dudley , 2002 ) ( e . g . , Diptera , Hymenoptera , and Lepidoptera ) . In contrast , in ant workers , T1 is greatly enlarged and forms a significant portion of the thorax , while T2 is reduced ( illustrative SEM image in Figure 2 ) . T3 is absent dorsally in workers of most species but , when T3 is distinguishable , the T3/T2 ratio does not differ between castes . In contrast , the ratio between T1 and T2 clearly discriminates workers and queens . Rather than just showing an overall reduction in T2 , consistent with their lack of wings , worker ants have a T1/T2 ratio reversed in relation to queens ( Figure 1A; SEM images in Figure 2 ) . The difference between castes in this ratio depends on the species ( Linear model: interaction Species x Caste , df = 10 , F= 68 . 3 , p<0 . 00001 ) but it is always greater in workers than in queens ( Linear Model: holding the factor Species constant , factor Caste , df = 1 , F= 8975 . 3 , p<0 . 00001 ) . Visual inspection of an extended sample of species ( Table 2 ) from 21 of the 25 ant subfamilies ( including the extinct taxon Sphecomyrminae† ) confirmed the universality of these relative length differences . Castes of all species with specimens available show a differential investment in the growth of T1 and T2 . T2 was larger than T1 in queens of all 52 species examined , and T1 was larger than T2 in workers of all 111 species examined ( Table 2; examples in Figure 1—figure supplement 2 ) . 10 . 7554/eLife . 01539 . 003Table 1 . Ant species studied for morphometrics and/or internal anatomyDOI: http://dx . doi . org/10 . 7554/eLife . 01539 . 003MorphometricsDissectionsSubfamilySpeciesqwqwAmblyoponinaeAmblyopone australis6822DolichoderinaeTapinoma simrothi––610EctatomminaeEctatomma ruidum––35FormicidaeLasius niger151524Oecophylla smaragdina––25Polyrhachis laboriosa13538MyrmeciinaeMyrmecia simillima––24Nothomyrmecia macrops––14MyrmicinaeCarebara vidua5311Cataulacus wasmanni151533Leptothorax pergandei131513Messor barbarus––38Monomorium pharaonis151524Monomorium subopacum––23Pogonomyrmex barbatus151745PonerinaeBrachyponera lutea151535Harpegnathos saltator––24Neoponera apicalis712410PseudomyrmecinaeTetraponera aethiops111546q = number of queens examined; w = number of workers examined . Generic placement of Brachyponera lutea and Neoponera apicalis reflects the new reclassification of species within the former paraphyletic genus Pachycondyla ( Schmidt CA , Shattuck SO , The higher classification of the ant subfamily Ponerinae [Hymenoptera: Formicidae] , with a review of ponerine ecology and behavior . Under review ) . 10 . 7554/eLife . 01539 . 004Figure 1 . Variation in length of first ( T1 ) and second ( T2 ) thoracic segments in ants shows characteristic differences depending on caste and species . ( A ) Relative lengths of T1 and T2 ( left ) show clear differences between queens and workers for 11 ant species . T3 ( right ) constitutes a small portion of the total length of the thorax in both queens and workers and , when present ( when T3/TL > 0 . 0 ) , is indistinguishable between castes . Numbers correspond to sample sizes and are equal for both panels ( Table 1 ) . ( B ) Gradient of investment in neck strength vs flight/storage musculature sorts individuals into three categories . Queens fall into two discrete categories based on the relative lengths of T1 and T2 . While the use of T1/T2 in ( A ) emphasizes the distinction between workers and queens and within species variation , T1/TL and T2/TL in ( B ) enables the distinction between queen types across species with large differences in body size . Measurements and ratios are available in the Dryad data repository under DOI doi: 10 . 5061/dryad . d62p2/1 ( Keller et al . , 2014 ) . Species codes: A . aus = Amblyopone australis; B . lut = Brachyponera lutea; C . vid = Carebara vidua; C . was = Cataulacus wasmanni; L . nig = Lasius niger; L . per = Leptothorax pergandei; M . pha = Monomorium pharaonis; N . api = Neoponera apicalis; P . bar = Pogonomyrmex barbatus; P . lab = Polyrhachis laboriosa; T . aet = Tetraponera aethiops . DOI: http://dx . doi . org/10 . 7554/eLife . 01539 . 00410 . 7554/eLife . 01539 . 005Figure 1—figure supplement 1 . Measurements used in this study . The length of the first ( T1 = pronotum ) , second ( T2 = mesonotum ) and third ( T3 = metanotum ) dorsal thoracic plates was measured along the dorsal midline . Total thoracic length ( TL ) was measured as the diagonal length in profile from the anterior-most point of the first thoracic segment to the posterior-most point of the third thoracic segment ( also known as Weber's length ) . For each of the specimens measured , images of dorsal and profile views are available in the Dryad data repository under DOI doi: 10 . 5061/dryad . d62p2/2 . Note that the total length of the thorax ( TL ) is always greater than the sum of the lengths of the dorsal thoracic plates ( T1 to T3 ) , because in ants ( as in most Hymenoptera ) the first abdominal segment ( A1 = propodeum ) is fused dorsally to the thorax and occupies most of the posterior part of the mesosomal region . A2 = second abdominal segment . Scale bars = 1 . 0 mm . DOI: http://dx . doi . org/10 . 7554/eLife . 01539 . 00510 . 7554/eLife . 01539 . 006Figure 1—figure supplement 2 . Differences in length proportion of thoracic segments among castes in nine representative species from different subfamilies . T2 is always larger than T1 in queens ( top ) , while T1 is larger than T2 in workers ( bottom ) . ( A ) Aneuretus simoni ( Aneuretinae ) ; ( B ) Discothyrea testacea ( Proceratiinae ) ; ( C ) Ectatomma tuberculatum ( Ectatomminae ) ; ( D ) Myopopone castanea ( Amblyoponinae ) ; ( E ) Myrmecia chasei ( Myrmeciinae ) ; ( F ) Myrmica emeryana ( Myrmicinae ) ; ( G ) Pseudoponera stigma ( Ponerinae ) ; ( H ) Pseudomyrmex gracilis ( Pseudomymecinae ) ; ( I ) Tapinoma erraticum ( Dolichoderinae ) . White-black-white on thick bars equals length of T1 , T2 , and T3 respectively . Note that T3 has no distinguishable dorsum in workers of most species . Scale bars upper left , 1 mm . All images by April Nobile/antweb . org . DOI: http://dx . doi . org/10 . 7554/eLife . 01539 . 00610 . 7554/eLife . 01539 . 007Figure 2 . Skeletomuscular specialization of queens and workers in ants . The dorsal plate of T1 is always enlarged in workers relative to queens ( left column; multiple individuals from 52 genera examined , Table 2 ) . Queens can either ( A ) have a reduced T1 and huge T2-associated wing muscles ( represented here by Oecophylla smaragdina ) , or ( B ) show a slightly enlarged T1 and associated neck muscles ( represented here by Neoponera apicalis ) . T1 , T2 , and T3 , first , second and third thoracic segments respectively; A1 , first abdominal segment . Workers of N . apicalis lack a discernible T3 . Internally ( right column ) , the wing muscles in queens ( red ) fill most of the thoracic cavity , while the T1 muscles ( blue ) are narrow and close to the thoracic wall . In all workers examined ( see Table 1 for list of species and sample sizes ) , the T1 notopleural muscles ( np , dark blue ) that support the anteroventral plates ( yellow ) fill the anterior portion of the cavity . The dorsal cervical muscles ( dc , light blue; see also Figure 2—figure supplement 1B ) that in winged queens originate at the anterior phragma and pull the head up at contraction , show a shifted position in workers . In the absence of phragma , these muscles originate at the dorsal boundary between T1 and T2 . Rather than being short and thin , they form long and thick bundles that stretch the entire length of the enlarged T1 cavity to their place of insertion on the back of the head ( Figure 2—figure supplement 2 ) . Figure 1—supplement 1 has photos of many more species of ‘reduced T1’ and ‘intermediate T1’ species for comparison of external thoracic morphology . DOI: http://dx . doi . org/10 . 7554/eLife . 01539 . 00710 . 7554/eLife . 01539 . 008Figure 2—figure supplement 1 . Thoracic musculature in queen and worker ants . ( A ) Sagittal section of a queen ( Polyrhachis laboriosa ) reveals the thoracic cavity filled by the longitudinal ( lw ) and dorsoventral ( dw ) indirect wing-muscles ( he , head; T1 , pronotum; T2 , mesonotum; T3 metanotum; A1 , first abdominal segment; A2 , second abdominal segment; ap , anterior phragma ) . ( B ) Anterior view of the queen's T2 shows the thin dorsal cervical muscle pair ( dc ) that originates at the anterior phragma ( ap ) . ( C ) Removing the wing muscles and the dorsal plates of T1 and T2 exposes the notopleural muscle pair ( np ) inside the anterior part of the thorax ( left column is dorsal view , right column is profile view; tissues are stained with methylene blue ) . These muscles are thin and narrow in queens ( first row P . laboriosa , second row Neoponera apicalis ) . Equivalent muscles in ant workers are hypertrophied , and fill the T1 cavity completely ( third row , N . apicalis ) . Scale bar = 1 mm . DOI: http://dx . doi . org/10 . 7554/eLife . 01539 . 00810 . 7554/eLife . 01539 . 009Figure 2—figure supplement 2 . Internal anatomical adaptations in ant workers for powerful head movement . ( A ) One dorsal pair of prothoracic muscles ( dc , dorsal cervical ) traverses the enlarged workers' T1 cavity and pulls the head up at contraction; the expanded prothoracic endosternum ( pe ) is the origin for two pairs of muscles ( dorsal and ventral ) that move the head up-and-down . ( B ) Skeletal preparations of 18 species ( Table 1 ) revealed that workers show enlargement of the endosternum , an internal skeletal structure that branches inside T1 for attachment of muscles that , in bees , power up-and-down movement of the head ( left column is profile view , right column is frontal view; represented by Neoponera apicalis; scale bar , 500 µm ) . While in queens , the T1 endosternum has an upper face ( up ) perpendicular to its basal stalk ( ba ) , in workers the upper face rises almost parallel to the basal stalk and has a larger surface for the attachment of the muscles that pull the head . This modification of the endosternum in workers is only possible because the complete absence of wing muscles that occur in this caste leaves the thoracic cavity with sufficient space for the expansion of T1 internal structures . In queens ( as is the case in all castes of honey bees ) , the perpendicular orientation of the endosternal face is necessary for the occurrence of the longitudinal wing muscles across the thoracic cavity . DOI: http://dx . doi . org/10 . 7554/eLife . 01539 . 00910 . 7554/eLife . 01539 . 010Table 2 . List of species surveyed for relative length of thoracic segmentsDOI: http://dx . doi . org/10 . 7554/eLife . 01539 . 010FAMILY/subfamilyspeciesqueenworkerMuseumVoucher codeMuseumVoucher codeFORMICIDAE AenictinaeAenictus vaucheri/binghamiMSNGCASENT0903754AMNHRAK0094 AgroecomyrmecinaeTatuidris tatusiaDADCCASENT0178881BMNHRAK0001 AmblyoponinaeAdetomyrma spAMNHRAK0003 AmblyoponinaeAmblyopone australisANICCASENT0172213AMNHRAK0005 AmblyoponinaeAmblyopone mercovichiMCZRAK0006 AmblyoponinaeApomyrma stygiaMNHNCASENT0101445MCZRAK0083 AmblyoponinaeConcoctio concentaMCZRAK0011 AmblyoponinaeMyopopone castaneaANICCASENT0172069AMNHRAK0012 AmblyoponinaeMystrium spCASCCASENT0104559CASCCASENT0076622 AmblyoponinaeOnychomyrmex doddiAMNHRAK0014 AmblyoponinaePrionopelta punctulataANICCASENT0172312AMNHRAK0016 AmblyoponinaeStigmatomma armigeraAMNHRAK0004 AmblyoponinaeStigmatomma pallipesABSCASENT0103553MCZRAK0009 AmblyoponinaeStigmatomma plutoMCZRAK0010 AmblyoponinaeXymmer muticusMCZRAK0007 AneuretinaeAneuretus simoniANICCASENT0172259MCZRAK0074 CerapachyinaeAcanthostichus serratulusAMNHRAK0095 CerapachyinaeCerapachys nitidulusRAKCRAK127AMNHRAK0096 CerapachyinaeCerapachys doryloidesAMNHRAK0097 CerapachyinaeCylindromyrmex brevitarsusJTLCCASENT0610653AMNHRAK0098 CerapachyinaeSimopone schoutedeniAMNHRAK0099 DolichoderinaeDolichoderus bispinosusALWCCASENT0173835ALWCCASENT0173833 DolichoderinaeIridomyrmex lividusANICCASENT0172066ANICCASENT0172041 DolichoderinaeLeptomyrmex pallensAMNHRAK0075 DolichoderinaeTapinoma erraticumCASCCASENT0173200AMNHRAK0078 DolichoderinaeTechnomyrmex albipesCASCCASENT0060419AMNHRAK0079 DorylinaeDorylus conradti/helvolusMSNGCASENT0903712AMNHRAK0100 EcitoninaeCheliomyrmex morosusAMNHRAK0101 EcitoninaeEciton hamatumJTLCINBIOCRI001283500AMNHRAK0103 EcitoninaeLabidus coecusAMNHRAK0102 EctatomminaeEctatomma tuberculatumJTLCJTLC000014186AMNHRAK0017 EctatomminaeGnamptogenys annulataAMNHRAK0018 EctatomminaeGnamptogenys striatulaMIZACASENT0178660AMNHRAK0019 EctatomminaeGnamptogenys bufonisMCZRAK0020 EctatomminaeGnamptogenys minutaMCZRAK0021 EctatomminaeRhytidoponera metallicaANICCASENT0172346ANICCASENT0172345 EctatomminaeTyphlomyrmex pusillusMIZACASENT0178662AMNHRAK0023 EctatomminaeTyphlomyrmex rogenhoferiAMNHRAK0024 FormicinaeFormica sp . ( fusca group ) CASCCASENT0173171AMNHRAK0080 FormicinaeLasius flavusCASCCASENT0173149UCDCCASENT0005406 FormicinaeOecophylla smaragdinaCASCCASENT0173644AMNHRAK0082 FormicinaePolyergus spRAKCRAK0129RAKCRAK0130 FormicinaePolyrhachis revoiliCASCCASENT0403971CASCCASENT0227558 HeteroponerinaeAcanthoponera minorAMNHRAK0025 HeteroponerinaeHeteroponera brouniMCZRAK0128AMNHRAK0026 HeteroponerinaeHeteroponera relictaAMNHRAK0027 LeptanillinaeLeptanilla swaniAMNHRAK129AMNHRAK0084 LeptanilloidinaeLeptanilloides erinys/biconstrictaUCDCCASENT0234616AMNHRAK0104 MartialinaeMartialis heurekaMZSPCASENT0106181 MyrmeciinaeMyrmecia gulosaCASCCASENT0103309CASCCASENT0103310 MyrmeciinaeNothomyrmecia macropsAMNHRAK0086 MyrmicinaeAphaenogaster fulvaCASCCASENT0104857CASCCASENT0103585 MyrmicinaeCarebara viduaCASCCASENT0260121CASCCASENT0010803 MyrmicinaeCataulacus wasmanniCASCCASENT0498338CASCCASENT0498558 MyrmicinaeLeptothorax pergandeiMCZRAK0125MCZRAK0126 MyrmicinaeManica rubidaAMNHRAK0090 MyrmicinaeMessor barbarusRAKCRAK0123RAKCRAK0124 MyrmicinaeMetapone madagascaricaCASCCASENT0004524MCZRAK0093 MyrmicinaeMonomorium pharaonisABSCASENT0104094ABSCASENT0104095 MyrmicinaeMyrmica wheeleriMCZCASENT0102860MCZCASENT0102862 MyrmicinaePogonomyrmex uruguayensisRAJCCASENT0172689RAJCCASENT0103054 ParaponerinaeParaponera clavataRAKCRAK0122AMNHRAK0028 PonerinaeAnochetus mayriABSCASENT0103555MCZCASENT0003324 PonerinaeAsphinctopone silvestriiMCZRAK0031 PonerinaeBelonopelta deletrixMCZRAK0032 PonerinaeBothroponera pachydermaAMNHRAK0054 PonerinaeBrachyponera croceicornisAMNHRAK0051 PonerinaeCentromyrmex brachycolaUCDCCASENT0178343AMNHRAK0033 PonerinaeCryptopone gilvaCASCCASENT0006055AMNHRAK0034 PonerinaeDiacamma ceylonenseAMNHRAK0035 PonerinaeDinoponera lucidaAMNHRAK0036 PonerinaeDolioponera fustigeraMCZRAK0037 PonerinaeEmeryopone buttelreepeniMCZRAK0038 PonerinaeHagensia marleyiMCZRAK0053 PonerinaeHarpegnathos saltatorAMNHRAK0039 PonerinaeHypoponera sp1 . AMNHRAK0040 PonerinaeLeptogenys ( Leptogenys ) sp . 1AMNHRAK0041 PonerinaeLeptogenys ( Lobopelta ) sp . 2AMNHRAK0042 PonerinaeLeptogenys podenzanaiMCZRAK0043 PonerinaeLoboponera obeliscataAMNHRAK0044 PonerinaeLoboponera vigilansAMNHRAK0045 PonerinaeMyopias chapmaniANICCASENT0172094ANICCASENT0172093 PonerinaeNeoponera apicalisALWCCASENT0103060AMNHRAK0048 PonerinaeNeoponera villosaAMNHRAK0058 PonerinaeOdontomachus bauriCASCCASENT0172630AMNHRAK0030 PonerinaeOdontoponera transversaBMNHCASENT0900664AMNHRAK0047 PonerinaeOphthalmopone berthoudiMCZRAK0049 PonerinaePachycondyla crassinodaAMNHRAK0050 PonerinaeCryptopone guianensisMCZRAK0052 PonerinaePseudoneoponera porcataAMNHRAK0055 PonerinaePseudoponera stigmaAMNHRAK0056 PonerinaePaltothyreus tarsatusAMNHRAK0057 PonerinaePhrynoponera gabonensisAMNHRAK0059 PonerinaePlatythyrea punctataABSCASENT0104429AMNHRAK0060 PonerinaePlatythyrea turneriMCZRAK0061 PonerinaePlectroctena strigosaAMNHRAK0062 PonerinaePonera alphaMCZRAK0063 PonerinaePonera pennsylvanicaCASCCASENT0006086AMNHRAK0064 PonerinaePsalidomyrmex procerusAMNHRAK0065 PonerinaeSimopelta oculataMCZRAK0066 PonerinaeStreblognathus peetersiAMNHRAK0067 PonerinaeThaumatomyrmex atroxAMNHRAK0068 ProceratiinaeDiscothyrea oculataAMNHRAK0069 ProceratiinaeDiscothyrea testaceaABSCASENT0103848AMNHRAK0070 ProceratiinaeProceratium croceumABSCASENT0104440AMNHRAK0071 ProceratiinaeProceratium pergandeiAMNHRAK0072 ProceratiinaeProbolomyrmex guineensisAMNHRAK0073 PseudomyrmecinaePseudomyrmex gracilisABSCASENT0103779AMNHRAK0087 PseudomyrmecinaeTetraponera aethiopsAMNHRAK0088 PseudomyrmecinaeTetraponera attenuataCASCCASENT0217587AMNHRAK0089 Sphecomyrminae†Sphecomyrma freyi†AMNHAMNH NJ-943SCOLIIDAEScolia nobilitataAMNHRAK0121VESPIDAEMetapolybia cingulataAMNHRAK0120Information on museum holdings and voucher codes for queens and workers . ABS , Archbold Biological Station; ALWC , Alexander Wild Collection; AMNH , American Museum of Natural History; ANIC , Australian National Insect Collection; BMNH , British Museum of Natural History; CASC , California Academy of Science; DADC , David A . Donoso Collection; JTLC , Jack Longino Collection; MCZ , Museum of Comparative Zoology ( Harvard ) ; MIZA , Museo del Instituto de . Zoología Agrícola ( Venezuela ) ; MNHN , Muséum national d’Histoire naturelle; MSNG , Natural History Museum , Genoa; MZSP , Museu de Zoologia Universidade de São Paulo; RAJC , Robert Johnson Collection; RAKC , Roberto Keller Collection; UCDC; University of California Davis . † denotes extinct taxa . To infer the functional significance of the caste-specific external thoracic configurations , we performed a comparative analysis of the internal skeletomuscular system in queens and workers . We dissected 144 individuals from 19 species belonging to eight subfamilies ( Table 1 ) and analyzed both muscle ( extent of attachment ) and skeletal elements . Our dissections showed that the length of the thoracic segments in dorsal view is a reflection of the volume of the muscles associated with each segment . In the same way that the large T2 of queens is indicative of the presence of large wing muscles , the large T1 in workers reflects the enlargement of muscles in this segment ( Figure 2 , Figure 2—figure supplement 1 , Figure 2—figure supplement 2 ) . Studies in other insects established that homologous T1-associated skeletomuscular elements are involved in the head-thorax articulation or neck ( Snodgrass , 1935 , 1956; Hartenstein , 2006 ) . In queens of all 19 ant species dissected ( Table 1 ) , the neck-associated muscles were short and thin , traversing the narrow space of T1 between the head and the anterior phragma ( cuticular invagination ) of T2 where the wing muscles attach ( Figure 2 , Figure 2—figure supplement 1A ) . This configuration of neck elements is similar to that of female honey bees irrespective of caste ( Snodgrass , 1956 ) , and Drosophila fruit flies ( Hartenstein , 2006; McQuilton et al . , 2012 ) . In contrast , in ant workers , the expansion of T1 and the lack of both anterior phragma and wing muscles result in a larger anterior cavity that contains neck muscles and skeletal pieces in a unique configuration . The most striking muscular difference between ant castes concerns one of the notopleural pairs of muscles that originate dorsally on T1 ( np in Figure 2 , right column; Figure 2—figure supplement 1C–H ) . The main function of homologous muscles in honey bees is to carry the plates that support the head and serve to move it sideways or rotate it ( Snodgrass , 1956 ) . In ant queens , where most of the thoracic cavity is filled by the wing muscles ( as is the case in all castes of honey bees ) , these muscles are narrow and close to the thoracic wall . Our dissections revealed that the equivalent muscles in ant workers are hypertrophied , and fill the wider T1 cavity completely . Ant workers also show important differences in internal skeleton associated with T1 ( Figure 2—figure supplement 2B ) . This skeletomuscular configuration highlights the increased strength of the workers’ neck that powers head movements . Even though queens invest mostly in the thoracic segment used for flight ( T2 ) , our morphometric data showed that queens of different species fall into two discrete categories based on the relative investment into T1 . When plotting the normalized length of T1 vs T2 for 130 queens measured ( Table 1 ) , we can discriminate two clusters of species ( Figure 1B; where workers of all species form a third cluster ) . For five of the 11 species in Table 1 , queens have a reduced T1 , almost not visible in dorsal view ( Figure 2A ) . The other six species form a category with queens having an intermediate T1 , corresponding to enlarged T1 muscles ( Figure 2B ) . To investigate the evolution of queen thoracic configurations across the ant phylogeny , we focused on a total of 54 ant species ( those in Table 2 for which queens were available for measurements ) representing 21 subfamilies , as well as two species of wasps from different families as outgroups ( Table 3 ) . Queens were scored as belonging to the categories ‘reduced T1’ ( 22 species , all with T1/T2 < 0 . 14 ) or ‘intermediate T1’ ( 32 species , all with T1/T2 > 0 . 28 ) , as seen in Figure 1B . This information was combined with a well-established ant phylogeny ( Brady et al . , 2006; Moreau et al . , 2006 ) and we used parsimony and maximum likelihood ( ML ) methods to reconstruct ancestral character states ( Figure 3 ) . Our analysis showed that an ‘intermediate T1’ in queens arose in the common ancestor to all ants ( ML proportional likelihood = 0 . 800 ) , and that there were multiple transitions to the ‘reduced T1’ ( Figure 3 ) . This reduction seems to have evolved convergently in at least four major ant lineages . Transitions back to an ‘intermediate T1’ are rare and more recent events , being restricted to the genera Polyergus within subfamily Formicinae , and Cataulacus and Metapone within subfamily Myrmicinae . In contrast , the universal occurrence of a hypertrophied T1 in workers ( including the primitive fossil Sphecomyrma† ) supports a single origin of this novel thoracic configuration in the common ancestor of all ants ( Figure 3 ) . 10 . 7554/eLife . 01539 . 011Table 3 . Queen thoracic morphology and type of colony foundation across antsDOI: http://dx . doi . org/10 . 7554/eLife . 01539 . 011SubfamilyGenusT1/T2 in queensT1 in queensColony foundingReferencesAenictinaeAenictus2 . 742intermediatefission ( Gotwald and Cunningham-van Someren , 1976 ) AgroeconomyrmecinaeTatuidris0 . 111reducedunknownAmblyoponinaeAmblyopone0 . 382intermediatenon-claustral ( Haskins and Haskins , 1951 ) AmblyoponinaeApomyrma0 . 338intermediateunknownAmblyoponinaeMyopopone0 . 453intermediatenon-claustral ( Ito , 2010 ) AmblyoponinaeMystrium0 . 454intermediatenon-claustral ( Molet et al . , 2009 ) AmblyoponinaePrionopelta0 . 514intermediatenon-claustral ( Ito and Billen , 1998 ) AneuretinaeAneuretus0 . 096reducedclaustral ( Wilson et al . , 1956 ) CerapachyinaeCerapachys0 . 364intermediateunknown ICF + fission ( Brown , 1975 ) CerapachyinaeCylindromyrmex0 . 454intermediatenon-claustral ( Delabie and Reis , 2000 ) DolichoderinaeDolichoderus0 . 061reducedunknownDolichoderinaeIridomyrmex0 . 071reducedclaustral ( Hölldobler and Carlin , 1985 ) DolichoderinaeTapinoma0 . 111reducedclaustral ( Kannowski , 1959 ) DolichoderinaeTechnomyrmex0 . 071reducedclaustral ( Yamauchi et al . , 1991 ) DorylinaeDorylus0 . 372intermediatefission ( Kronauer et al . , 2004 ) EcitoninaeEciton0 . 469intermediatefission ( Schneirla , 1949 ) EctatomminaeEctatomma0 . 325intermediatenon-claustral ( Dejean and Lachaud , 1992 ) EctatomminaeGnamptogenys0 . 331intermediatenon-claustral ( ‡ ) EctatomminaeRhytidoponera0 . 363intermediatenon-claustral ( Ward , 1981 ) EctatomminaeTyphlomyrmex0 . 504intermediateunknownFormicinaeFormica0 . 076reducedclaustral ( Stille , 1996 ) FormicinaeLasius0 . 053reducedclaustral ( Stille , 1996 ) FormicinaeOecophylla0 . 066reducedclaustral ( Hölldobler and Wilson , 1978 ) FormicinaePolyergus0 . 323intermediatenon-claustral† ( Mori et al . , 1995 ) FormicinaePolyrhachis0 . 072reducedclaustral and non-claustral ( Lenoir and Dejean , 1994 ) HeteroponerinaeHeteroponera0 . 485intermediatenon-claustral ( § ) LeptanillinaeLeptanilla2 . 685intermediatefission ( Masuko , 1990 ) LeptanilloidinaeLeptanilloides3 . 021intermediatefission ( Donoso et al . , 2006 ) MartialinaeMartialisn/aunknownunknownMyrmeciinaeMyrmecia0 . 485intermediatenon-claustral ( Haskins and Haskins , 1950 ) MyrmicinaeAphaenogaster0 . 117reducedclaustral ( Lubertazzi , 2012 ) MyrmicinaeCarebara0 . 072reducedclaustral ( Robertson and Villet , 1989 ) MyrmicinaeCataulacus0 . 494intermediateunknownMyrmicinaeLeptothorax0 . 090reducedclaustral ( Keller and Passera , 1989 ) MyrmicinaeMessor0 . 110reducedclaustral and non-claustral ( Brown , 1999 ) MyrmicinaeMetapone0 . 428intermediateunknownMyrmicinaeMonomorium0 . 132reducedclaustral ( Bolton , 1986 ) MyrmicinaeMyrmica0 . 071reducedclaustral and non-claustral ( Brown and Bonhoeffer , 2003 ) MyrmicinaePogonomyrmex0 . 097reducedclaustral and non-claustral ( Johnson , 2002 ) ParaponerinaeParaponera0 . 086reducednon-claustral ( # ) PonerinaeAnochetus0 . 367intermediatenon-claustral ( Brown , 1978 ) PonerinaeCentromyrmex0 . 493intermediatenon-claustral ( Dejean and Fénéron , 1996 ) PonerinaeCryptopone0 . 533intermediatenon-claustral ( Peeters , 1997 ) PonerinaePonera0 . 356intermediatenon-claustral ( Kannowski , 1959 ) PonerinaeMyopias0 . 282intermediatenon-claustral ( Peeters , 1997 ) PonerinaeOdontomachus0 . 411intermediatenon-claustral ( Brown , 1976 ) PonerinaeOdontoponera0 . 524intermediatenon-claustral ( Peeters , 1997 ) PonerinaePachycondyla0 . 385intermediatenon-claustral ( Peeters , 1997 ) PonerinaePlatythyrea0 . 417intermediatenon-claustral ( Peeters , 1997 ) ProceratiinaeDiscothyrea0 . 093reducednon-claustral and claustral ( Dejean and Dejean , 1998 ) ProceratiinaeProceratium0 . 095reducednon-claustral ( ¶ ) PseudomyrmecinaePseudomyrmex0 . 479intermediatenon-claustral ( * ) PseudomyrmecinaeTetraponera0 . 558intermediatenon-claustral ( ** ) Sphecomyrminae†Sphecomyrma†n/aunknownunknownOUTGROUPS ScoliinaeScolia0 . 087reducednon-social ( †† ) PolistinaeMetapolybia0 . 074reducedfission ( †† ) Queen thoracic morphology and type of colony foundation across ants . The wasp taxa Scolia and Metapolybia are included as outgroups . species with wingless queens . † denotes extinct taxa . †Polyergus is an obligatory social parasite of Formica spp . ‡John Lattke , personal communication . §Rodrigo Feitosa , personal communication . #Haskins CP , Enzmann EV ( 1937 ) Studies of certain sociological and physiological features in the Formicidae . Ann NY Acad Scien 37:97-162; Michael Breed , personal communication . ¶Fuminori Ito , personal communication; Keiichi Masuko , personal communication . Philip Ward , personal communication . ††James M Carpenter , personal communication . 10 . 7554/eLife . 01539 . 012Figure 3 . Phylogenetic reconstruction reveals a single origin of a hypertrophied T1 in workers and multiple independent origins of ‘reduced’ T1 in queens . The latter is associated with modifications in modes of colony foundation . Tree branches and tree background are colored for queen morphology and founding behavior respectively , according to the parsimony ancestral reconstruction . Typical queen-worker dimorphism shown to the right to illustrate ratio T1/T2 ( not to scale ) . Species with wingless queens are marked with an asterisk . Phylogeny was pruned from Moreau et al . ( 2006 ) . Placement of Sphecomyrma† and Martialis after Grimaldi et al . ( 1997 ) and Rabeling et al . ( 2008 ) , respectively . Metapolybia and Scolia wasps are included as outgroups . Data on the species are analyzed , and their morphology and type of colony founding behavior are summarized in Table 3 . Numbers correspond to major taxonomic groups within Formicidae after Ward ( 2007 ) : 1 , Sphecomyrminae†; 2 , Leptanillinae; 3 , Martialinae; 4 , Proceratiinae; 5 , Amblyoponinae; 6 , Paraponerinae; 7 , Agroecomyrmecinae; 8 , Ponerinae; 9 , dorylomorphs; 10 , myrmeciomorphs; 11 , dolichoderomorphs; 12 , ectaheteromorphs; 13 , Formicinae; 14 , Myrmicinae . DOI: http://dx . doi . org/10 . 7554/eLife . 01539 . 012 Out of the two morphological categories of queens we identified , one is closer to workers in size of T1 vs T2 ( Figure 1B ) . Similarly for behavior , it is known that queens in some species go through a worker-like phase after they mate and shed their wings . In several lineages , lone founding queens regularly forage outside the nest ( they are ‘non-claustral’ ) , and can hunt and carry large prey to feed the first generation of worker larvae ( Haskins , 1970; Peeters , 1997 ) . During several weeks , these non-claustral queens behave much like workers . This contrasts with the vast majority of ants , where founding queens are confined to the nest ( they are ‘claustral’ ) and , instead of foraging , use their metabolic reserves to feed their first brood ( Hölldobler and Wilson , 1990; Wheeler and Buck , 1996 ) . To test the hypothesis that the morphological classes associate with the behavioral classes , we compiled data on mode of colony foundation for the 54 ant species in our tree ( Figure 3; Table 3 ) . We could find information for a total of 45 species: 25 non-claustral , 15 claustral , and 4 with dependent colony foundation ( i . e . colony fission , when queens are never alone ) . Unfortunately , there are no data for some of the putative early lineages ( the fossil Sphecomyrma [Grimaldi et al . , 1997] , and the two rare subfamilies Leptanillinae and Martialinae [Borowiec et al . , 2011] ) . Using parsimony and ML methods , we established that non-claustral behavior is the most likely ancestral condition ( Figure 3; ML proportional likelihood = 0 . 919 ) . Claustral colony foundation has evolved at least twice independently , with reversals to non-claustral foundation occurring sporadically within some genera . Our reconstruction supports colony fission as a secondary shift among ants ( Cronin et al . , 2013 ) . Next , we performed a Concentrated Changes test ( Maddison , 1990 ) to investigate the phylogenetic correlation between queen thoracic morphology and founding behavior . We found strong support for correlated evolution ( p=0 . 027 , calculated by simulation of 100 , 000 actual changes with two gains and four loses ) : all queens with an ‘intermediate T1’ are non-claustral founders , whereas two of four independent origins of queens with a ‘reduced T1’ coincide with shifts to claustral foundation ( Figure 3; clades 11 and 13+14 ) . A reversal in morphology to ‘intermediate T1’ corresponds to a modified claustral behavior in Polyergus ( clade 13 ) which parasitizes colonies of Formica , hence Polyergus queens need to fight to invade the host colonies ( Trager , 2013 ) . However , sporadic reversals to non-claustral founding have been reported for a few species ( Lenoir and Dejean , 1994; Brown , 1999; Johnson , 2002; Brown and Bonhoeffer , 2003 ) that according to our morphological survey are not accompanied by reversals in queen morphology ( Figure 3 ) . Modeling suggests that such facultative reversals to non-claustral behavior are likely to occur in cases of increased resource availability ( Johnson , 2002; Brown and Bonhoeffer , 2003 ) . We did not observe changes in the T1/T2 ratio in lineages that secondarily evolved colony fission , even though this mode of colony foundation is known to co-occur with wing-loss in queens ( Cronin et al . , 2013 ) . This suggests that , despite being wingless , in the absence of the selective pressures related to worker-like foraging ( as in non-claustral queens ) or of the need for storing metabolic reserves as flight muscles ( as in claustral queens ) , queens in those lineages maintain the ancestral T1/T2 ratio ( e . g . dorylomorph clade in Figure 3 ) . The ecological dominance of ants in terrestrial ecosystems is unparalleled in the animal kingdom ( Wilson , 1971; Folgarait , 1998 ) . Because no other group of social insects reaches equivalent levels of adaptive radiation and species-richness ( Hölldobler and Wilson , 1990 ) , it seems that factors in addition to social behavior and division of labor promoted ant diversification . The evolutionary success of ants is indisputably associated with a strong divergence between queens and workers . A caste of flightless workers specialized in non-reproductive activities is unique among social Hymenoptera . However , rather than being just simplified , wingless versions of the queen , the thorax of ant workers has its own specialization . Relative to the thoracic morphology of queens , which is typical of species of flying insects , worker ants have an unusually large T1 and T1-associated muscles , which provide superior strength and mobility to the neck controlling head movements . Control of the head is of great importance for ant workers , which in some species singly hunt and carry prey up to 30-90 times their weight ( Dejean , 2011; Dejean et al . , 2012 ) . Among insects , ant carrying behavior is unique in that workers lift their load off the ground . Many other insects can move relatively large objects , but by dragging ( e . g . , spider wasps ) or rolling them ( e . g . , dung beetles ) on the ground , or holding them while flying ( e . g . , robber flies ) . Biomechanical studies on grass-cutting ants have shown that workers perform controlled head movements at the neck articulation when transporting large objects ( Moll et al . , 2010 ) . Precise head movements are essential to reduce displacement of the center of mass , and retain stability while carrying objects many times the workers’ weight and length . Our finding that worker ants differ from queens and other flying insects in the configuration and size of the T1-associated muscles suggests that ants can achieve this biomechanical feat by virtue of their specialized neck musculature . This represents a striking structural innovation , differentiating ant workers from the typical flying insects , which had not been recognized until now . Their distinctive internal skeletomuscular modifications presumably enhance their behavior as flightless foragers and heavy-load transporters . We propose that the modified T1 was an innovation that helped ants to use their heads and mandibles in novel ways , and hence exploit a broader spectrum of trophic resources . Compared to social bees and wasps ( Hölldobler and Wilson , 1990 ) , where worker morphology is constrained by the requirements of a winged thorax , mandibular morphology and function have specialized enormously across ant lineages ( Paul and Gronenberg , 1999 ) , in parallel with their much greater diversification of foraging habits . Our analysis also showed that queens fall in two distinct anatomical types that evolved in association with the two strategies of independent colony founding . Foraging activity during independent foundation is high in non-claustral species vs absent in claustral species . Non-claustral queens have a T1 that is closer in size to that of workers , while claustral queens , which do not go through a worker-like phase , have a much more reduced T1 . Unfortunately , biomechanical data of neck strength in queens are difficult to obtain because they are evasive and , especially in claustral species , cannot be induced to carry objects . Claustral queens have an enlarged T2 relative to non-claustral queens , reflecting the existence of massive wing muscles ( Figure 2A ) . A correlation between increased wing muscle mass and claustral behavior has been suggested before: these larger muscles do not function to enhance flight , rather they are a solution for storing amino acids that are essential for feeding the first generation of workers without outside foraging ( Jones et al . , 1978; Peeters , 2012 ) . We speculate that , during the acquisition of claustral behavior , the decrease in foraging activity lessened the constraint on the size of T1 , thus allowing T2 to expand and accommodate larger wing muscles as metabolic reserves . Differences in the nesting habits of queens , such as excavating a nest vs nesting in pre-existing cavities , might also impose variable muscle requirements . However , this type of behavioral differences occurs across species in a scattered pattern that does not match the anatomical categories we revealed . There are examples of nest excavating by queens with ‘intermediate’ ( e . g . , Amblyopone ) and ‘reduced’ T1 ( e . g . , Pogonomyrmex ) , and of nesting in pre-existing cavities by ‘intermediate’ ( e . g . , Tetraponera ) as well as ‘reduced’ T1 ( e . g . , Leptothorax ) species . While data on queen morphology is readily accessible from museum collections for many species , knowledge about their founding behavior remains sparse . There is no published information in many important genera , possibly because this requires field observations of behavior at an appropriate time of the year . Our findings provide a means of predicting colony foundation strategy from the morphology of the queen thorax , and thus guide field research on particular species of interest . For example , within the subfamily Myrmicinae ( clade 14 in Figure 3 ) , the genera Cataulacus and Metapone show independent reversals to an ‘intermediate T1’ in queens , suggesting that colony foundation is not claustral as in closely related genera . Importantly , the phylogenetic component of our correlation provides a powerful tool to infer the ecology of extinct clades for which behavioral observations are impossible . For example , we lack data on queens of two early lineages , the extinct Sphecomyrma† and the enigmatic Martialis , but based on our reconstructions we can predict that they will have an ‘intermediate T1’ and behave non-claustrally . Our finding that the ratio of the lengths of T1 and T2 is inverted between queens and workers suggests that a morphological trade-off was at play in determining the relative size of these two segments . It is likely that T1 can become hypertrophied only at the expense of a reduced , non-functional T2 . Indeed , our anatomical analysis showed that some of the internal modifications of T1 in workers are only possible in conjunction with a complete absence of wing muscles . Conversely , only queens with a highly reduced T1 have an expanded T2 that constitutes most of the thoracic dorsum ( queens with intermediate T1 are also intermediate for T2 , see Figure 1B ) . This morphological trade-off between adjacent body segments can occur due to competition for metabolic resources during pre-adult development ( Nijhout and Emlen , 1998 ) . It is possible that the functional cost of enlarging T2 ( reserves for colony founding ) at the expense of T1 ( reduced neck strength and work performance ) , occurred when founding behavior gradually shifted to claustral , with a decreased need to forage outside the nest . We compared the thorax of queens and workers across multiple species representing all major ant lineages . First , we measured the length of the thoracic segments and entire thorax in a sample of individuals belonging to 11 different species from five ant subfamilies ( Table 1 ) . Second , we used a database of scanning electronic micrographs ( Keller , 2011 ) and an online database of light microscopy images ( http://www . antweb . org ) to further assess the extent of the taxonomic distribution of the traits of interest . We inspected the external thorax of workers belonging to 110 ant species , and of queens belonging to a subset of 47 species where this caste is known or available ( listed in Table 2 ) . This taxon sampling represents all 21 extant subfamilies with the exception of the dorylomorph subfamily Aenictogitoninae ( for which only males have been formally described ) , and includes the extinct subfamily Sphecomyrminae† . Lastly , we analyzed internal thoracic anatomy by dissecting multiple individuals from 19 representative species ( Table 1 ) . Our sample of queens included individuals both before and after the phase of muscle reabsorption , as assessed by the shedding of their wings . For the quantitative characterization of the thorax , we took dorsal and lateral photographs of pinned specimens ( Museum of Comparative Zoology , Harvard University ) with a JVC digital camera mounted on a Leica MZ16 binocular microscope ( images are deposited in the Dryad data repository under DOI doi: 10 . 5061/dryad . d62p2/2 ) . We then measured ( ImageJ , http://rsb . info . nih . gov/ij ) the dorsal length of the first ( T1 ) , second ( T2 ) , and third ( T3 ) thoracic segments along the midline , and the total thoracic length ( TL ) as the diagonal length in profile from the anterior-most point of T1 to the posterior-most point of T3 ( Figure 1—figure supplement 1; measurements are available in the Dryad data repository under http://dx . doi . org/10 . 5061/dryad . d62p2/1 ) . For the analysis of internal anatomy we performed muscle preparations using specimens fixed in either 80% ethanol or 4% formaldehyde , and sectioning their thoraces in sagittal and parasagittal planes or disarticulating the plates of the thorax . Muscle preparations were stained in 0 . 2% methylene blue ( Sigma-Aldrich ) to increase contrast against other tissues . We also performed skeletal preparations by disarticulating specimens with overnight digestion of soft tissues in 10% KOH , and kept in 90% ethanol for inspection . When necessary ( i . e . , lightly pigmented specimens ) , skeletal preparations were stained in 70% ethanol saturated solution of Chlorazol Black E ( Sigma-Aldrich ) . All analyses were performed with R ( R Development Core Team , 2008 ) . Residuals of the models have been checked for normality and equality of variance , and data have been transformed when necessary . To compare the relative investment in the thoracic plates 1 and 2 between castes , we performed a linear model ( LM ) constructed as √ ( T1/T2 ) ∼ species caste . T1/T2 is the ratio of the thoracic plate 1 length ‘T1’ over the length of the thoracic plates 2 ‘T2’ . ‘*’ indicated that the effects were tested for both main factors as well as interaction . We scored the queens of our 56 exemplar species as either ‘reduced’ or ‘intermediate’ . We divided the length of T1 by the length of T2 , and determined a cut-off index equal to 0 . 25 based on our previous morphometric analysis . We then assigned ‘reduced’ to queens falling below the cut-off value and ‘intermediate’ for queens falling above it . Missing data ( i . e . , the unknown queens of Sphecomyrma freyi† , and Martialis eureka ) ( Grimaldi et al . , 1997; Rabeling et al . , 2008 ) were coded as ‘ ? ’ . For the modes of colony foundation we assigned states for ‘non-claustral’ , ‘claustral’ and ‘fission’ based on records from the scientific literature ( Table 3 ) . Unknown mode of colony foundation was coded as ‘ ? ’ . Data in Table 3 correspond to a single queen for each of the species in Figure 1 ( except Brachyponera lutea because its exact phylogenetic position within the subfamily Ponerinae remains undetermined ) and from 44 more species ( listed in Table 2 ) . Character evolution was reconstructed under parsimony using WinClada ( Nixon , 2002 ) and under maximum likelihood ( ML ) using Mesquite ( Maddison and Maddison , 2012 ) , under the Mk1 model , ( Lewis , 2001 ) . Ambiguous optimizations under parsimony were resolved using DELTRAN . This algorithm gave results closer to the ML analysis than did the ACCTRAN parsimony algorithm . Tree topology with branch lengths was pruned from Moreau et al . ( 2006 ) . We implemented the concentrated changes test ( Maddison , 1990 ) , using MacClade ( Maddison and Maddison , 2005 ) , to test for a correlation between modes of colony foundation and queen morphology . This test calculates the probability that changes in a binary character along the phylogeny are distributed randomly on the branches defined by a second binary character . We therefore transformed our data on behavior and morphology to binary characters by pruning out the branches with fission and wingless queens , since both traits always co-occur in the phylogeny .	The size and shape of an animal , known as its morphology , often reflect the actions it can perform . A grasshopper’s long legs , for example , are well suited to hopping , whilst the streamlined body of a dolphin helps swimming through water . These specialized features result from the interplay between morphology and behavior during evolution . A change in morphology can make new behaviors possible , which can then expose the animal to new environments and selective pressures that , in turn , can lead to further changes in morphology . The interplay between morphology and behavior is particularly interesting in social insects such as ants . Queens and workers within an ant colony have a similar set of genes , but they have dramatically different morphologies and very different roles within the colony . Queens are responsible for reproduction , and are larger and have wings , which allow them to fly and establish a new colony away from where they were born . Workers are smaller and lack wings , and they devote themselves to building the nest , feeding the young larvae and protecting the colony . This marked morphological divergence , unique to ants , has fascinated researchers for more than a century . However , most studies have focused on the presence or absence of wings and have overlooked the interactions between morphology and the actions performed on the ground . Like all insects , an ant’s body is divided into three parts: the head , the thorax ( to which the legs and wings are attached ) , and the abdomen . Now , Keller et al . have examined the shape of the thorax in many species of ants and found that workers are not just smaller wingless versions of queens: rather , the architecture of their thorax is unique among species of flying insects . The front end of the worker thorax is greatly enlarged and is filled by strong neck muscles that power the head and its jaws , and allow workers to hunt and carry prey many times their own weight . Keller et al . also identified two distinct types of queens and went on to show that these two shapes evolved in association with the two types of strategy that lone queens use to found new colonies . In species where queens convert their own wing muscles into the food for the first generation of workers , the wing muscles are much enlarged and the neck segment is extremely reduced . In species where queens hunt to feed the new colony , the wing and neck muscles are more balanced in size . As such , for those ant species where very little is known about how new colonies are founded , Keller et al . show that we can use the shape of the queen’s thorax to help predict this behavior . Taken together , the results of Keller et al . show that female ants invest in the relative size of the different segments of the thorax in a way that reflects their behavior as adults . These adaptations partly explain why ants have been so extraordinarily successful in nature , and underscore the importance of carefully analyzing an organism’s form to fully understand its biology .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "ecology", "evolutionary", "biology" ]	2014	Evolution of thorax architecture in ant castes highlights trade-off between flight and ground behaviors
Arterial occlusive diseases are major causes of morbidity and mortality . Blood flow to the affected tissue must be restored quickly if viability and function are to be preserved . We report that disruption of the mixed-lineage protein kinase ( MLK ) - cJun NH2-terminal kinase ( JNK ) signaling pathway in endothelial cells causes severe blockade of blood flow and failure to recover in the murine femoral artery ligation model of hindlimb ischemia . We show that the MLK-JNK pathway is required for the formation of native collateral arteries that can restore circulation following arterial occlusion . Disruption of the MLK-JNK pathway causes decreased Dll4/Notch signaling , excessive sprouting angiogenesis , and defects in developmental vascular morphogenesis . Our analysis demonstrates that the MLK-JNK signaling pathway is a key regulatory mechanism that protects against ischemia in arterial occlusive disease . Ischemic stroke , myocardial infarction and peripheral artery disease result from arterial occlusion that blocks blood flow leading to severe tissue ischemia and necrosis . To prevent loss of tissue viability and function , blood flow to the affected tissue must be restored quickly . Collaterals are artery-to-artery or arteriole-to-arteriole interconnections that can bypass an occlusion by providing an alternative route for blood flow to the affected tissue that restores tissue homeostasis and limits tissue damage ( Antoniucci et al . , 2002; Schaper , 2009; Faber et al . , 2014; Simons and Eichmann , 2015 ) . Indeed , clinical outcome in patients with arterial occlusion depends on the presence of an adequate collateral circulation and animal models of arterial occlusion provide strong evidence for the critical importance of the extent of the native ( pre-existing ) collateral circulation in restoring blood perfusion and limiting ischemic sequelae following arterial occlusion . Adequate restoration of blood flow depends on collateral artery size , number , and the pattern of connectivity , but also on functional adaptation to changes in blood flow . Following arterial occlusion , more blood flow is diverted to the collateral circulation and this increased flow and sheer stress in collateral arteries initiates a number of processes that result in the outward remodeling ( arteriogenesis ) of these vessels into efficient conductance arteries ( Heil et al . , 2006; Schaper , 2009; van Royen et al . , 2009; Simons and Eichmann , 2015 ) . Collateral artery remodeling involves multiple cellular processes , including endothelial cell activation and proliferation , monocyte/macrophage recruitment and smooth muscle cell proliferation , all of which contribute to increased collateral artery diameter , including increased thickness of the tunica media . These structural and functional adaptations depend on the presence of the native collateral circulation . Although the collateral circulation is crucially important for the protective response to arterial occlusive diseases , little is known about the cellular and morphogenetic processes , or the molecular factors and mechanisms , that contribute to native collateral artery formation ( Antoniucci et al . , 2002; Schaper , 2009; Faber et al . , 2014; Simons and Eichmann , 2015 ) . However , studies of leptomenengial ( or pial ) collateral arteries in the brain have provided significant insight . Murine pial collaterals are established during embryonic development with some remodeling and maturation continuing postnatally . The process of native collateral artery formation during embryogenesis has been termed collaterogenesis and involves a number of molecules including , platelet-endothelial cell adhesion molecule 1 ( PECAM1 ) ( Chen et al . , 2010 ) , gap junction protein , connexin37 ( Cx37 ) ( Fang et al . , 2011 , 2012 ) , prolyl hydroxylase domain-containing protein 2 ( PHD2 ) ( Takeda et al . , 2011 ) , endothelial nitric oxide synthase ( eNOS ) ( Dai and Faber , 2010 ) , chloride intracellular channel 4 ( CLIC4 ) ( Chalothorn et al . , 2009 ) , and Synectin ( Moraes et al . , 2013 ) . Signaling pathways that have been reported to contribute to collaterogenesis include NF-κB ( Tirziu et al . , 2012 ) , VEGF ( Chalothorn et al . , 2007; Lucitti et al . , 2012 ) , and the Dll4 – Notch pathway ( Cristofaro et al . , 2013 ) . The purpose of this study was to examine the role of the c-Jun NH2-terminal kinase ( JNK ) ( Davis , 2000 ) , a signaling pathway that has been reported to play major roles in angiogenic responses ( Jiménez et al . , 2001; Ennis et al . , 2005; Medhora et al . , 2008; Uchida et al . , 2008; Guma et al . , 2009; Shen et al . , 2010; Kaikai et al . , 2011; Ma et al . , 2012; Du et al . , 2013; Salvucci et al . , 2015 ) . Our approach was to study the effect of compound gene disruption in endothelial cells to prevent JNK signaling in mice . We did not find that JNK signaling was required for angiogenesis in vitro or in adult mice , but JNK signaling is required for proper vascular morphogenesis and the normal formation of collateral arteries in muscle . We examined angiogenic responses of control and E3KO MLEC in vitro . JNK was not activated by hypoxia or VEGF and both control and E3KO MLEC mounted similar responses to hypoxia and VEGF ( Figure 1—figure supplement 3 ) . Tubulogenesis assays in matrigel demonstrated no differences between control and E3KO MLEC and no differences between control and E3KO mice were detected in VEGF-induced microvessel sprouting from collagen-embedded aortic rings ( Figure 1—figure supplement 4A , B ) . We also found no differences in proliferation or migration between control and E3KO MLEC ( Figure 1—figure supplement 4C–E ) . To assess angiogenesis in vivo , we examined laser-induced injury of the eye; no differences in choroidal neovascularization between control and E3KO mice were observed ( Figure 1—figure supplement 5A , B ) . Similarly , we found no differences in tumor angiogenesis between control and E3KO mice ( Figure 1—figure supplement 5C–F ) . Collectively , these data demonstrate that JNK in endothelial cells is not required for angiogenesis in vitro or in vivo in adult mice . To test the role of endothelial JNK in the response to arterial occlusion , we performed unilateral femoral artery ligation ( FAL ) on control and E3KO mice . This procedure causes hypoxia in the calf muscles that stimulates angiogenesis , but the proximal adductor muscles experience little or no hypoxia because of blood flow redistribution by the native collateral circulation ( Deindl et al . , 2001 ) ( Figure 1—figure supplement 1B ) . We ligated the femoral artery ( FA ) between the proximal caudal femoral artery ( PCFA ) and the popliteal artery ( PA ) ( Kochi et al . , 2013 ) ; this is a mild version of the FAL procedure ( Figure 1B ) . Laser Doppler imaging revealed ~80% decreased blood perfusion to the ligated limbs of control mice; blood perfusion was restored to ~60% of the contralateral limbs by day 3 ( Figure 1C , D ) . These mice did not exhibit major hallmarks of ischemia ( Figure 1E–G ) . In contrast , E3KO mice showed complete blockade of blood flow following occlusion ( Figure 1C , D ) leading to severe necrosis ( Figure 1E–G ) . Ligation of the FA more proximally at its origin ( a more severe form of FAL ) also demonstrated increased blood flow blockade and failure to recover in both E2KO and E3KO mice ( Figure 1—figure supplement 6A–C ) . In contrast , no post-FAL phenotype was detected in mice with JNK1 or JNK2-deficiency alone ( Figure 1—figure supplement 6D ) or in mice with JNK1 plus JNK2-deficiency in hematopoietic cells or muscle ( Figure 1—figure supplement 6G–I ) . These data suggest that JNK1/2 in endothelial cells play a key role in the response to arterial occlusion . Consistent with this conclusion , coronary artery occlusion caused significantly greater mortality of E3KO mice compared with control mice ( Figure 1—figure supplement 6E , F ) . The defect in blood flow of E3KO mice post-FAL could be mediated by cardiovascular dysfunction , but no changes in blood pressure , heart rate , or echocardiographic measurements of cardiac function were detected ( Figure 1—figure supplement 7A–C ) . Moreover , contraction and endothelium-dependent relaxation responses in aortic explants from E3KO and control mice were similar ( Figure 1—figure supplement 7D ) . These data indicate that neither cardiovascular dysfunction nor defective vasodilatory responses contribute to the post-FAL phenotype of E3KO mice . The early and severe blood perfusion blockade in E3KO mice post-FAL suggests a defect in collateral artery function . Two highly stereotypic superficial arteries ( gracilis collaterals ) extend along the gracilis muscle in the medial aspect of the thigh ( Figure 1B ) . Gracilis collaterals were identified as two lumenized continuous arteries that connected the PCFA to the saphenous artery ( SA ) ( Figure 1H ) and expanded radially during the post-FAL response ( Figure 1H , I ) . In contrast , these arteries were abnormal in E3KO mice; arteries emerged from the PCFA and SA ( Figure 1H ) , but were thin and branched into multiple smaller vessels forming a disorganized network ( Figure 1H ) . Micro-computed tomography ( μCT ) analysis confirmed reduced collateral artery size and continuity in the limbs of E3KO mice ( Figure 1—figure supplement 8A ) . These collateral artery defects may contribute to decreased blood flow and increased hypoxia in post-FAL E3KO mice , despite no overall reduction in muscle vascularization or macrophage recruitment ( Figure 1—figure supplement 8B–E ) . We tested whether disruption of genes that encode other JNK pathway components caused a similar post-FAL phenotype . The MLK group of MAP3K causes activation of the JNK pathway by a Rac1/Cdc42-dependent mechanism ( Gallo and Johnson , 2002; Kant et al . , 2011 ) . Gene expression analysis demonstrated that Map3k10 and Map3k11 were the most highly expressed members of this group in endothelial cells ( Figure 2A ) . Indeed , Map3k10-/- Map3k11-/- MLEC exhibited reduced phosphorylation of the JNK substrate cJun compared with control MLEC ( Figure 2B ) . We therefore examined the post-FAL response of Map3k10-/- Map3k11-/- mice . Similar to E3KO mice , MLK-deficient mice showed increased blood flow blockade , failure of blood flow restoration by day 3 post-FAL , and necrosis ( Figure 2C–G ) . Moreover , we found abnormal gracilis collateral arteries in Map3k10-/- Map3k11-/- mice ( Figures 2H and 3A ) . These data demonstrate that the MLK-JNK signaling pathway in endothelial cells is important for collateral artery patterning and the post-FAL response . 10 . 7554/eLife . 18414 . 021Figure 2 . Severe ischemic injury in MLK2/3-deficient mice after femoral artery ligation . ( A ) The expression of members of the MLK protein kinase family ( MLK1 , MLK2 , MLK3 & MLK4 ) in primary MLEC cultures was examined by measurement of Map3k9 , Map3k10 , Map3k11 , and BC021891 mRNA by RNA-seq analysis ( mean fragments per kilobase of transcript per million mapped reads ( FPKM ) ± SEM; n = 6 ) . ( B ) Primary wild-type ( WT ) and Map3k10-/- Map3k11-/- MLEC cultures were treated without and with 100 ng/ml bFGF and examined by immunoblot analysis by probing with antibodies to pSer63-cJun , cJun , JNK , Cdh5 , and GAPDH . ( C ) Representative laser doppler images showing blood perfusion ( high perfusion red , no perfusion dark blue ) in the hindlimbs of WT and Map3k10-/- Map3k11-/- mice mice prior to unilateral FA ligation ( Pre-FAL ) and post-FAL . ( D ) Quantitation of hindlimb blood flow demonstrated that MLK2/3-deficient mice exhibited significantly increased blood perfusion blockade and no recovery by day 3 post-FAL compared with control mice ( mean ± SEM; n = 7 ) . ( E ) Representative images of mouse paws on Day 4 post-FAL . Lig . , ligated; Unlig . , contralateral unligated . ( F , G ) Quantitation of ischemic ( F ) and movement ( G ) scores for mice on Day 4 post-FAL ( n = 9 ) . ( H ) Representative whole mount preparations of the medial surface of Microfil-filled adductor muscle vasculature isolated from day 4 post-FAL hindlimbs and contralateral unligated limbs . Source data are included as Figure 2—source data 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 18414 . 02110 . 7554/eLife . 18414 . 022Figure 2—source data 1 . Source data for Figure 2 . This file contains raw source data used to make the graphs presented in Figure 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 18414 . 02210 . 7554/eLife . 18414 . 023Figure 3 . MLK2/3-deficient mice exhibit defects in native collateral artery formation . ( A ) Representative confocal images ( n = 5 mice ) of control and MLK2/3-deficient whole mount P6 adductor muscle vasculature stained with antibodies to endomucin ( capillary and venous vasculature , red ) and SMA ( arterial and venous smooth muscle , green ) . Gracilis collateral arteries in WT mice , but not MLK2/3-deficient mice , interconnect the PCFA to the SA . ( B ) Representative stereomicroscope images of P6 whole mount abdominal muscle stained with an antibody to SMA ( green ) . Arteriole-to-arteriole arcades are indicated ( red arrows ) . The abdominal muscle vasculature of Map3k10-/-Map3k11-/- mice shows very few arteriole-to-arteriole interconnections . Quantitation reveals significantly reduced arteriolar arcade numbers in Map3k10-/-Map3k11-/- mice compared to WT mice ( mean ± SEM; n = 5 mice ) . Source data are included as Figure 3—source data 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 18414 . 02310 . 7554/eLife . 18414 . 024Figure 3—source data 1 . Source data for Figure 3 . This file contains raw source data used to make the graphs presented in Figure 3 . DOI: http://dx . doi . org/10 . 7554/eLife . 18414 . 024 The abnormal gracilis collateral arteries in MLK and JNK-deficient mice suggested that the JNK pathway could be important for collateral artery function , but these observations may also reflect a required role for the MLK-JNK pathway during development . To distinguish between these possiblities , we established mice with tamoxifen-inducible gene ablation . We found that compound Mapk8/9/10 gene ablation in adult mice did not alter the post-FAL response ( Figure 4A–D ) . In contrast , compound Mapk8/9/10 gene ablation in embryos prior to FAL in adult mice caused increased blood flow blockade post FAL and failure to recover ( Figure 4E–G ) . These data demonstrate that the post-FAL phenotype of E3KO mice is caused by an early developmental defect . 10 . 7554/eLife . 18414 . 025Figure 4 . Endothelial JNK is not required for the arteriogenic response of gracilis collaterals in adult mice . ( A ) Timeline of tamoxifen administration to induce disruption of Mapk8LoxP and Mapk9LoxP alleles in the vascular endothelium of adult mice prior to FAL and analysis of blood flow by laser doppler imaging . ( B ) Primary MLEC cultures prepared from mice treated without and with tamoxifen were examined by immunoblot analysis by probing with antibodies to JNK and αTubulin . The data are representative of two independent MLEC isolations per group ( 2~3 mice used per cell preparation ) . ( C ) Quantitation of laser doppler analysis of limb blood flow demonstrated no significant differences ( p>0 . 05 ) in blood perfusion blockade and recovery over 28 days post-FAL between tamoxifen-treated endothelial JNK-deficient mice and tamoxifen-treated control mice ( mean ± SEM; n = 5~10 ) . ( D ) Microfil perfusion of adductor muscle vasculature demonstrated the presence of similar gracilis collateral arteries in JNK-deficient and control mice and similar collateral artery remodeling at day 28 post-FAL . The images are representative of 5~8 mice per group . ( E ) Timeline of tamoxifen administration to induce disruption of Mapk8LoxP and Mapk9LoxP alleles in the vascular endothelium during embryonic development prior to analysis of FAL in adults and examination of blood flow by laser doppler imaging . ( F ) Quantitation of laser doppler analysis of limb blood flow demonstrated significantly enhanced blood perfusion blockade in adult mice with embryonic endothelial JNK-deficiency compared with control mice ( mean ± SEM; n = 5~6 mice per group ) . ( G ) The adductor muscle vasculature of two E16 . 5 embryos obtained from a pregnant female mouse that was treated with tamoxifen at 12 . 5 dpc was examined by confocal microscopy . The Rosa26mTmG genetic background allows detection of Cre-mdiated recombination in vascular endothelial cells ( green ) . The data presented are representative of six mice examined . Source data are included as Figure 4—source data 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 18414 . 02510 . 7554/eLife . 18414 . 026Figure 4—source data 1 . Source data for Figure 4 . This file contains raw source data used to make the graphs presented in Figure 4 . DOI: http://dx . doi . org/10 . 7554/eLife . 18414 . 026 The pial collateral circulation that interconnects the distal branches of the middle cerebral and the anterior cerebral arteries has been studied ( Chalothorn and Faber , 2010; Lucitti et al . , 2012 ) , but the formation of collateral arteries in muscle is unclear . We found that the gracilis collaterals in post-natal day 6 ( P6 ) and P0 control mice interconnected the PCFA and the SA ( Figure 5A , B ) . Dil perfusion analysis demonstrated that the arteries had lumens ( Figure 5A , B ) and were fully covered by smooth muscle cells at P6 ( Figure 5A; SMA ) , but not at P0 ( Figure 5B; SMA ) . In contrast , gracilis collaterals in E3KO mice were not formed at P6 or P0 ( Figure 5A , B ) . Individual vessels did emerge from the PCFA and the SA , but instead of interconnecting to form collaterals , these vessels branched off into multiple smaller caliber vessels ( Figure 5A , B; Dil Perfusion ) that lacked smooth muscle coverage at P6 ( Figure 5A; SMA ) and appeared to continue into the capillary circulation ( Figure 5A , B; Dil Perfusion ) . Similarly , analysis of the abdominal muscle arterial circulation in P0 pups revealed numerous arteriolar arcades ( direct arteriole-to-arteriole interconnections ) in control mice , but this arterial patterning was significantly reduced in E3KO mice ( Figure 5C ) . These defects in gracilis collateral and abdominal arteriolar arcade development were also detected in MLK-deficient mice ( Figure 3A , B ) . 10 . 7554/eLife . 18414 . 027Figure 5 . Endothelial JNK-deficient mice display abnormal native collateral arteries . ( A , B ) Representative confocal images ( n = 7 mice ) of whole mount adductor muscle vasculature reveals SMA-covered gracilis collateral arteries in P6 control mice , but not JNK-deficient mice ( A ) . Confocal imaging of Dil perfused P6 adductor muscle vasculature ( n = 5 mice ) demonstrates distinct gracilis collaterals interconnecting the PCFA to the SA in control mice . Vessels emerging from the PCFA and the SA in E3KO mice do not fully interconnect , but branch into smaller vessels . At P0 , gracilis collaterals were not SMA-covered , but were perfused with Dil in control mice ( B ) . The analogous vessels in E3KO mice did not interconnect , but branched extensively into smaller vessels ( B ) ( n = 5 mice ) . ( C ) Representative stereomicroscope images of Dil-perfused abdominal muscle arterial vasculature of control and JNK-deficient P0 mice . Arteriole-to-arteriole arcades ( indicated by red arrows ) were quantitated ( mean ± SEM; n = 3 mice ) . ( D ) Representative confocal images ( n = 3~4 mice ) of whole mount adductor muscles of control and JNK-deficient E16 . 5 embryos showing GFP-labeled vascular endothelial cells . ( E ) Representative confocal images ( n = 3~5 mice ) of control and JNK-deficient E16 . 5 embryo adductor muscle vasculature immunostained for Endomucin ( Emcn , red ) and isolectinB4 ( iB4 , green ) . Prominent vessels emerging from the PCFA and SA are indicated with white arrowheads . Source data are included as Figure 5—source data 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 18414 . 02710 . 7554/eLife . 18414 . 028Figure 5—source data 1 . Source data for Figure 5 . This file contains raw source data used to make the graphs presented in Figure 5 . DOI: http://dx . doi . org/10 . 7554/eLife . 18414 . 02810 . 7554/eLife . 18414 . 029Figure 5—figure supplement 1 . Intimate association of gracilis collaterals and peripheral nerves in adductor muscles . Confocal microscopy of a whole mount adductor muscle stained with antibodies to SMA ( green ) and Neurofilament-M ( red ) illustrates the close association of gracilis collateral arteries with peripheral nerves . DOI: http://dx . doi . org/10 . 7554/eLife . 18414 . 029 To gain insight into the mechanism that might account for these defects in collateral artery patterning/maturation in E3KO mice , we examined the vasculature in whole mount preparations of adductor muscles in embryonic day 16 . 5 ( E16 . 5 ) embryos . While large caliber vessels , including the FA/SA and PCFA , were established ( Figure 5D , E ) , distinct collaterals directly interconnecting the PCFA and SA were not formed at E16 . 5 and the gracilis muscle was covered by a capillary plexus ( Figure 5D , E ) . Prominent vessels do emerge from the PCFA and the SA ( Figure 5E ) , but did not extend along the gracilis muscle as distinct collaterals; instead , these vessels branched and appeared to continue into the capillary plexus ( Figure 5E ) . Gracilis collaterals may therefore form through a plexus intermediate . Remodeling of vessels within this plexus likely leads to the formation of collateral arteries in close apposition to nerve fibres ( Figure 5—figure supplement 1 ) . This process of maturation appears to start at the two distal ends , where the future gracilis collaterals emerge from the PCFA and the SA ( Figure 5E , arrowheads ) and continue toward the middle of the muscle; a pattern of remodeling that likely reflects the blood flow characteristics of these vessels ( Meisner et al . , 2013 ) . Studies of E16 . 5 E3KO embryos demonstrated that the gracilis muscle capillary plexus was hyperbranched , denser , and disorganized with vessels of variable width that elaborated more filopodia than control embryos ( Figure 5D , E ) . This observation suggests that the failure of collateral artery formation in E3KO mice may be caused by defective sprouting angiogenesis that initially generates a hyperbranched , denser , and more chaotically organized plexus that fails to properly remodel . To test whether JNK-deficiency caused endothelial cell hypersprouting , we examined retinal vascular development during the early postnatal period because this is a well-characterized system that enables analysis of sprouting angiogenesis in a vascular plexus that initially extends from the center towards the periphery of the retina in two dimensions ( Eilken and Adams , 2010 ) . Analysis of retinal flatmounts from P6 E3KO mice demonstrated significantly reduced radial extension of the vascular plexus ( Figure 6A–C , L ) . Closer examination demonstrated higher vascular density in the growing angiogenic front of the mutant retinas compared to littermate control mice ( Figure 6D–G , H , J , M ) . Vascular extension in the retina occurs through the coordinated interaction , migration , and proliferation of endothelial tip and stalk cells together with non-endothelial cells , including pericytes , that stabilize the vascular plexus . We found no differences in vessel pericyte coverage ( Figure 6—figure supplement 1 ) ; however , the angiogenic front of the mutant retinas included larger numbers of tip cells and a larger number of filopodia ( Figure 6H–K , N , O ) . Similar data were obtained from analysis of E2KO and Map3k10-/- Map3k11-/- mice ( Figure 6—figure supplements 2 and 3 ) . These data confirm that the MLK-JNK signaling pathway restrains excessive endothelial cell sprouting during developmental angiogenesis . 10 . 7554/eLife . 18414 . 030Figure 6 . Abnormal retinal vascular development associated with excessive sprouting of tip cells in endothelial JNK-deficient mice . ( A–C ) Collages of confocal images of P6 whole mount retinas stained with isolectinB4 ( iB4 ) show reduced vascular extension in JNK-deficient retinas ( B , C ) compared with littermate control retinas ( A ) . The images are representative of 17~31 retinas examined for each genotype . ( D–K ) Higher magnification reveals increased vascular density ( D–G , H , J ) , increased numbers of tip cells ( yellow asterisks , H & J ) and increased numbers of filopodia ( red dots , I and K ) at the vascular front region of JNK-deficient retinas compared to littermate control retinas . A , artery; V , vein . ( L–O ) The vascularized retinal area ( L ) , vascular density within angiogenic front regions outlined in panels E & G ( M ) , tip cell number ( N ) , and filopodia number ( O ) is presented ( mean ± SEM; n = 17~31 ( panel l ) ; n = 6~10 ( panels M–O ) . Source data are included as Figure 6—source data 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 18414 . 03010 . 7554/eLife . 18414 . 031Figure 6—source data 1 . Source data for Figure 6 . This file contains raw source data used to make the graphs presented in Figure 6 . DOI: http://dx . doi . org/10 . 7554/eLife . 18414 . 03110 . 7554/eLife . 18414 . 032Figure 6—figure supplement 1 . NG2+ pericyte coverage of the P6 retinal vasculature . Confocal microscopy of whole mount retinal vasculature stained with an antibody to the pericyte marker NG2 ( red ) and isolectinB4 ( green ) demonstrated no obvious differences in vessel pericyte coverage in retinas from endothelial JNK-deficient mice compared to littermate control mice . The data presented are representative of images obtained from 4 mice per group . DOI: http://dx . doi . org/10 . 7554/eLife . 18414 . 03210 . 7554/eLife . 18414 . 033Figure 6—figure supplement 2 . Abnormal retinal vascular development associated with excessive sprouting in endothelial JNK1/2-deficient mice . ( A–C ) Collages of confocal images of P6 whole mount retinas stained with iB4 show reduced vascular extension in JNK1/2-deficient retinas ( B , C ) compared with littermate control retinas ( A ) . The images are representative of 4~9 retinas examined for each genotype ( ≥4 mice per group ) . ( D–K ) Higher magnification reveals increased vascular density ( D–G , H , J ) , increased numbers of tip cells ( yellow asterisks , H & J ) and increased numbers of filopodia ( red dots , I and K ) at the vascular front region of JNK1/2-deficient retinas compared to littermate control retinas . A , artery; V , vein . ( L–O ) The vascularized retinal area ( L ) , vascular density within angiogenic front regions outlined in panels e & g ( M ) , tip cell number ( N ) , and filopodia number ( O ) is presented ( mean ± SEM; n = 4~9 ( panel l ) ; n = 4~8 ( panels M–O ) ( ≥4 mice per group ) . Source data are included as Figure 6—figure supplement 2—source data 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 18414 . 03310 . 7554/eLife . 18414 . 034Figure 6—figure supplement 2—source data 1 . Source data for Figure 6—figure supplement 2 . This file contains raw source data used to make the graphs presented in Figure 6—figure supplement 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 18414 . 03410 . 7554/eLife . 18414 . 035Figure 6—figure supplement 3 . Abnormal retinal vascular development associated with excessive sprouting in Map3k10-/- Map3k11-/- mice . ( A–C ) Collages of confocal images of P6 whole mount retinas stained with isolectinB4 show reduced vascular extension in MLK2/3-deficient retinas ( B , C ) compared with littermate control retinas ( A ) . The images are representative of 5 mice examined for each genotype . ( D–K ) Higher magnification reveals increased vascular density ( D–G , H , J ) , increased numbers of tip cells ( yellow asterisks , H & J ) and increased numbers of filopodia ( red dots , I and K ) at the vascular front region of MLK2/3-deficient retinas compared to littermate control retinas . A , artery; V , vein . ( L–O ) The vascularized retinal area ( L ) , vascular density within angiogenic front regions outlined in panels E & G ( M ) , tip cell number ( N ) , and filopodia number ( O ) is presented ( mean ± SEM; n = 5 mice per group ) . Source data are included as Figure 6—figure supplement 3—source data 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 18414 . 03510 . 7554/eLife . 18414 . 036Figure 6—figure supplement 3—source data 1 . Source data for Figure 6—figure supplement 3 . This file contains raw source data used to make the graphs presented in Figure 6—figure supplement 3 . DOI: http://dx . doi . org/10 . 7554/eLife . 18414 . 036 To examine the molecular mechanisms that contribute to hypersprouting caused by defects in the MLK-JNK pathway , we examined gene expression by primary endothelial cells isolated from control and E3KO mice . We found 781 genes that were differentially expressed between E3KO and control endothelial cells . Gene ontology analysis demonstrated significant enrichment for several biological processes , including vascular development and morphogenesis , and we identified 64 differentially expressed genes that might contribute to vascular defects ( Figure 7—figure supplement 1 ) . These genes included components of the Notch signaling pathway; for example , Dll4 , Hey1 , Hes1 , and Lfng ( Figure 7A , B ) . This may be significant because the Notch pathway plays a major role during developmental angiogenesis , including tip/stalk cell specification and endothelial cell sprouting ( Roca and Adams , 2007; Phng and Gerhardt , 2009 ) . Indeed , the hypersprouting defects observed in E2KO , E3KO and Map3k10-/- Map3k11-/- mice resemble those previously reported for mice with reduced Dll4/Notch signaling ( Hellström et al . , 2007; Suchting et al . , 2007; Benedito et al . , 2009 ) , including Lfng-/- mice ( Benedito et al . , 2009 ) . Moreover , Dll4-/+ mice display perturbations in collateral artery formation and , like Notch1-/+ mice ( Takeshita et al . , 2007 ) , show reduced recovery of blood perfusion in models of vascular occlusion ( Cristofaro et al . , 2013 ) . To test whether JNK-deficiency regulates Notch signaling in endothelial cells , we examined the effect of angiogenic cytokines that can induce expression of the Notch ligand Dll4 and engage the Notch signaling pathway . Treatment of endothelial cells with VEGF or bFGF caused increased expression of Dll4 protein and promoted accumulation of the Notch intracellular domain ( NICD ) . However , these responses were suppressed in JNK-deficient endothelial cells ( Figure 7C , D ) . Indeed , reduced Dll4 protein expression by E3KO endothelial cells was observed in vitro and in vivo ( Figure 7B–E ) . These data indicate that JNK can promote Notch signaling in endothelial cells by regulating Dll4 expression . 10 . 7554/eLife . 18414 . 037Figure 7 . Reduced Dll4 / Notch signaling in the JNK-deficient vascular endothelium . ( A ) Quantitative RT-PCR analysis of Notch pathway genes revealing reduced expression in JNK-deficient primary endothelial cells compared with control cells ( mean ± SEM; n = 4 ) . The data shown are representative of the results obtained with three independent primary endothelial cell preparations . ( B ) Dll4 expression by control and JNK-deficient primary endothelial cells was examined by Immunofluorescence analysis ( mean ± SEM; n = 10 ) . ( C ) Control and JNK-deficient primary endothelial cells treated without and with 100 ng/ml VEGF ( 16 hr ) were examined by immunoblot analysis by probing with antibodies to Dll4 , Notch intracellular domain ( NICD ) , and αTubulin . The data are representative of experiments performed using two independent endothelial cell preparations . ( D ) Endothelial cells treated without and with 100 ng/ml bFGF were examined by immunoblot analysis by probing with antibodies to Dll4 , NICD , pSer63-cJun , cJun , pJNK , JNK , Cdh5 and GAPDH . The data are representative of experiments performed using two independent endothelial cell preparations . ( E ) Confocal immunofluorescence analysis of P6 whole mount retinas immunostained for Dll4 ( red ) , isolectinB4 ( iB4 , green ) , and Hoechst ( DNA , blue ) demonstrates that JNK-deficiency causes reduced expression of Dll4 at the angiogenic vascular front compared with retinas from littermate control mice ( mean ± SEM; n = 42~44 ) . Source data are included as Figure 7—source data 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 18414 . 03710 . 7554/eLife . 18414 . 038Figure 7—source data 1 . Source data for Figure 7 . This file contains raw source data used to make the graphs presented in Figure 7—figure supplement 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 18414 . 03810 . 7554/eLife . 18414 . 039Figure 7—figure supplement 1 . RNA-Seq analysis of differentially expressed genes between control and JNK-deficient endothelial cells . ( A ) Heatmap of the 781 differentially expressed genes ( FPKM > 2; absolute log2 fold change > 0 . 5; n = 3; q < 0 . 05 ) between MLEC cultures isolated from endothelial JNK-deficient mice and control MLEC cultures . JNK-deficiency caused similar numbers of upregulated and downregulated genes . ( B ) Gene ontology analysis of the differentially expressed genes demonstrated significant enrichment for several biological processes . ( C ) Differentially expressed genes related to vascular development , morphogenesis , and function were grouped in several categories and are presented as a heatmap . Genes are displayed with highest upregulation top and highest downregulation bottom within each category . Source data are included as Figure 7—figure supplement 1—source data 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 18414 . 03910 . 7554/eLife . 18414 . 040Figure 7—figure supplement 1—source data 1 . Source data for Figure 7—figure supplement 1 . This file contains raw source data used to make the graphs presented in Figure 7—figure supplement 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 18414 . 040 Studies of the development of leptomeningeal ( or pial ) collaterals that interconnect the medial , anterior and posterior cerebral artery trees in the brain have provided evidence that these collaterals form during embryonic development starting at ~E13 . 5 as sprout-like extensions of endothelial cells from arterioles of existing cerebral artery trees ( Lucitti et al . , 2012 ) . These nascent vessels appear to course above the pial capillary plexus and fuse with an arteriole from an adjacent arterial tree ( Lucitti et al . , 2012 ) . By E15 . 5 , a portion of these collaterals have acquired expression of the arterial marker EphrinB2 ( Chalothorn and Faber , 2010 ) . Pial collateral density peaks ~E18 . 5 and is followed by extensive remodeling , maturation and pruning that continues postnatally , achieving adult form and density by P21 ( Chalothorn and Faber , 2010 ) . The process of collateral artery formation during embryonic development has been termed collaterogenesis ( Chalothorn et al . , 2009; Faber et al . , 2014 ) . Little is known about the development of the collateral arteries in muscle . Our studies focused on the development of the gracilis collateral arteries . We found that a vascular plexus is formed between the proximal caudal femoral artery ( PCFA ) and the saphenous artery ( SA ) at ≤ E16 . 5 before gracilis collaterals are detected at ≤ P0 and covered with smooth muscle at ≤ P6 ( Figure 5 ) . This observation suggests that gracilis collaterals may be formed through a plexus intermediate by selection and maturation within a pre-formed capillary network that separates adjacent arteries . The presence of nerve fibres may contribute to this process ( Figure 5—figure supplement 1 ) . Developmental defects in gracilis collateral development were observed in mice with JNK deficiency in endothelial cells . The vascular plexus between adjacent arteries at E16 . 5 in JNK-deficient mice is denser and more chaotically organized ( Figure 5 ) . Highly variable vessel thickness and increased numbers of filopodia are also evident in the JNK-deficient vasculature ( Figure 5 ) . Similar defects were observed in the developing retinal vasculature ( Figure 6 ) . At P0 , continuous and distinct gracillis collateral vessels interconnecting the PCFA to the SA were fully formed in control mice , but the analogous vessels in the gracillis muscle of JNK-deficient mice were defective with extensive branching into smaller vessels that appeared to enter the capillary circulation . These observations indicated that an endothelial cell patterning defect may account for the absence of gracillis collateral arteries in the JNK-deficient mice . Hypersprouting represents one aspect of this patterning defect ( Figure 6 ) and reduced Dll4 – Notch signaling may significantly contribute to this phenotype ( Figure 7 ) . The formation of properly organized vascular networks is essential for function and requires the coordinated interaction of numerous factors and signaling pathways that regulate diverse cellular processes . VEGF signaling promotes endothelial cell survival , proliferation and motility while Dll4–Notch signaling suppresses VEGF signaling , in part , by regulating VEGFR expression . These interactions between VEGF and Notch signaling underly the specification of endothelial cell phenotypes during angiogenesis , including highly motile tip cells that extend numerous filopodia and trailing stalk cells with low motility that form the lumen of nascent tubules ( Roca and Adams , 2007; Phng and Gerhardt , 2009; Eilken and Adams , 2010; Jakobsson et al . , 2010 ) . The proper specification and interplay of tip and stalk cells is essential for the orchestration of sprouting angiogenesis that mediates expansion of vascular networks . Moreover , VEGF and Notch signaling are also critical for formation of the native collateral circulation ( Cristofaro et al . , 2013 ) . Studies of retinal vascular development demonstrate that MLK – JNK signaling regulates tip cell identity , and filopodia dynamics ( Figure 6 ) . Consequently , defects in the MLK – JNK signaling pathway cause excessive sprouting angiogenesis characterized by an increased number of tips and filopodia , increased vascular density , and decreased expression of the Notch ligand Dll4 at the angiogenic front ( Figure 7 ) . Excessive sprouting angiogenesis may contribute to the critical requirement for MLK – JNK signaling during native collateral artery development ( Figure 5 ) and this phenotype may result from promotion of Dll4 expression by MLK – JNK signaling ( Figure 7 ) . Indeed , it is established that defects in Notch signaling , including loss of Dll4 ( Hellström et al . , 2007 ) , Notch1 ( Hellström et al . , 2007 ) , Lfng ( Benedito et al . , 2009 ) , RBP-J ( Dou et al . , 2008; Izumi et al . , 2012 ) or chemical inhibition of γ-secretase ( Hellström et al . , 2007 ) , cause excessive endothelial cell sprouting during retinal vasculature development . Moreover , Dll4+/- mice display defects in blood perfusion restoration following femoral artery ligation ( Cristofaro et al . , 2013 ) . These data support the conclusion that the decreased Dll4 expression caused by JNK-deficiency in the vascular endothelium promotes excessive sprouting angiogenesis that disrupts the development of muscle collateral vessels ( Figure 8 ) . 10 . 7554/eLife . 18414 . 041Figure 8 . Schematic illustration of the role of the MLK-JNK pathway in vascular morphogenesis , formation of native collateral arteries , and the femoral artery ligation ( FAL ) model of hindlimb ischemia . DOI: http://dx . doi . org/10 . 7554/eLife . 18414 . 041 Our study points to an important role for JNK-mediated regulation of Dll4 – Notch signaling and vascular morphogenesis , and identifies a MLK – JNK signaling axis that is critical for native collateral artery formation . Disruption of this pathway causes defective collateral artery formation that results in severe blood perfusion blockade and tissue injury following arterial occlusion . Our study provides insight into the mechanism that controls muscle collaterogenesis , which is critically important for the response to arterial occlusive disease . Moreover , our analysis suggests that human germ-line mutations in JNK pathway genes may contribute to an increased risk for poor outcomes in patients following exposure to an ischemic insult . C57BL/6J mice ( RRID:IMSR_JAX:000664 ) , B6 . SJL-Ptprca Pepcb/BoyJ mice ( RRID:IMSR_JAX:002014 ) , B6 . 129P2-Lyz2tm1 ( cre ) Ifo/J ( also known as Lyz2-Cre mice and ΦCtrl mice; RRID:IMSR_JAX:004781 ) ( Clausen et al . , 1999 ) , B6 . FVB-Tg ( Cdh5-cre ) 7Mlia/J mice ( RRID:IMSR_JAX:006137 ) ( Alva et al . , 2006 ) , B6 . Cg-Tg ( Vav1-cre ) A2Kio/J mice ( RRID:IMSR_JAX:008610 ) ( de Boer et al . , 2003 ) , B6 . FVB ( 129S4 ) Tg ( Ckmm-cre ) 5Khn/J mice ( RRID:IMSR_JAX:006475 ) ( Brüning et al . , 1998 ) , and B6 . 129 ( Cg ) -Gt ( ROSA ) 26Sortm4 ( ACTB-tdTomato , -EGFP ) Luo/J mice ( also known as Rosa26mTmG mice ) ( RRID:IMSR_JAX:007676 ) ( Muzumdar et al . , 2007 ) were obtained from The Jackson Laboratories ( Bar Harbor , ME ) . Tg ( Cdh5-cre/ERT2 ) 1Rha mice ( RRID:IMSR_TAC:13073 ) ( Wang et al . , 2010 ) were provided by Prof . Ralf H . Adams . We have previously described Mapk8LoxP/LoxP , Mapk9LoxP/LoxP , Mapk8-/- ( RRID:IMSR_JAX:004319 ) , Mapk9-/- ( RRID:IMSR_JAX:004321 ) , Mapk10-/- ( RRID:IMSR_JAX:004322 ) , ΦKO mice ( Lyz2-Cre+ Mapk8LoxP/LoxP Mapk9LoxP/LoxP ) , and Map3k10-/- Map3k11-/- mice ( RRID:MGI:5296041 ) ( Dong et al . , 1998; Yang et al . , 1998; Kuan et al . , 2003; Das et al . , 2007; Kant et al . , 2011; Han et al . , 2013 ) . We generated the following mice: E3KO ( Cdh5-Cre+/- Mapk8LoxP/LoxP Mapk9LoxP/LoxP Mapk10-/- ) EfCtrl ( Cdh5-Cre-/- Mapk8LoxP/LoxP Mapk9LoxP/LoxP Mapk10-/- ) ECtrl ( Cdh5-Cre+/- Mapk8+/+ Mapk9+/+ Mapk10-/- ) E2KO ( Cdh5-Cre+/- Mapk8LoxP/LoxP Mapk9LoxP/LoxP ) ELoxP ( Cdh5-Cre-/- Mapk8LoxP/LoxP Mapk9LoxP/LoxP ) EWT ( Cdh5-Cre+/- Mapk8+/+ Mapk9+/+ ) E2KO:mTmG ( Cdh5-Cre+/- Mapk8LoxP/LoxP Mapk9LoxP/LoxP Rosa26mTmG+/- ) EmTmG ( Cdh5-Cre+/- Rosa26mTmG+ ) iE3KO ( Cdh5-Cre/ERT2+/- Mapk8LoxP/LoxP Mapk9LoxP/LoxP Mapk10-/- ) iEfCtrl ( Cdh5-Cre/ERT2-/- Mapk8LoxP/LoxP Mapk9LoxP/LoxP Mapk10-/- ) iECtrl ( Cdh5-Cre/ERT2+/- Mapk8+/+ Mapk9+/+ Mapk10-/- ) iE2KO:mTmG ( Cdh5-Cre/ERT2+/- Mapk8LoxP/LoxP Mapk9LoxP/LoxP Rosa26mTmG+/- ) ELoxP:mTmG ( Cdh5-/CreERT2-/- Mapk8LoxP/LoxP Mapk9LoxP/LoxP Rosa26mTmG+/- ) iEmTmG ( Cdh5-Cre/ERT2+/- Mapk8+/+ Mapk9+/+ Rosa26mTmG+/- ) H2KO ( Vav1-Cre+/- Mapk8LoxP/LoxP Mapk9LoxP/LoxP ) HLoxP ( Vav1-Cre-/- Mapk8LoxP/LoxP Mapk9LoxP/LoxP ) HWT ( Vav1-Cre+/- Mapk8+/+ Mapk9+/+ ) M2KO ( Ckm-Cre+/- Mapk8LoxP/LoxP Mapk9LoxP/LoxP ) MWT ( Ckm-Cre+/- Mapk8+/+ Mapk9+/+ ) All mice used in this study were backcrossed ( ≥ ten generations ) to the C57BL/6J strain . The mice were housed in a specific pathogen-free facility accredited by the American Association for Laboratory Animal Care . The animal studies were approved by the Institutional Animal Care and Use Committees of the University of Massachusetts Medical School , Tufts University School of Medicine , and Brigham and Women’s Hospital . PCR assays with genomic DNA and the amplimers 5’-TTACTGACCGTACACCAAATTTGCCTGC-3’ and 5’-CCTGGCAGCGATCGCTATTTTCCATGAGTG-3’ were used to detect the Cre+ allele ( 450 bp ) . The amplimers 5’CCTCAGGAAGAAAGGGCTTATTTC-3’ and 5’-GAACCACTGTTCCAATTTCCATCC-3’ detected the Mapk8+ allele ( 1550 bp ) , the Mapk8LoxP allele ( 1 , 095 bp ) , and the Mapk8Δ allele ( 395 bp ) . The amplimers 5’-GTTTTGTAAAGGGAGCCGAC-3’ and 5’-CCTGACTACTGAGCCTGGTTTCTC-3’ were used to detect the Mapk9+ allele ( 224 bp ) and the Mapk9LoxP allele ( 264 bp ) . The amplimers 5’-GGAATGTTTGGTCCTTTAG-3’ , 5’-GCTATTCAGAGTTAAGTG-3’ , and 5’-TTCATTCTAAGCTCAGACTC-3’ were used to detect the Mapk9LoxP allele ( 560 bp ) and the Mapk9Δ allele ( 400 bp ) . The amplimers 5’-CCTGCTTCTCAGAAACACCCTTC-3’ , 5’-CGTAATCTTGTCACAGAAATCCCATAC-3’ and 5’-CTCCAGACTGCCTTGGGAAAA-3’ were used to detect the Mapk10+ allele ( 437 bp ) and the Mapk10-allele ( 250 bp ) . The amplimers 5’-CTCTGCTGCCTCCTGGCTTCT-3’ , 5’-CGAGGCGGATCACAAGCAATA-3’ and 5’-TCAATGGGCGGGGGTCGTT-3’ were used to detect the mTmG allele ( 250 bp ) and the WT allele ( 330 bp ) . The amplimers 5’-CCTGGTTCTCACTGGGACAACAG-3’ , 5’-GTCACATCCACTTTCCTGGGC-3’ , and 5’-CGCCTTCTATCGCCTTCTTGA-3’ detected the Map3k10+ allele ( 500 bp ) and the Map3k10- allele ( 600 bp ) . The amplimers 5’-AGCAAACTCCGAGCAAGGGAC-3’ , 5’-GGCTAAACCAGAACTCAAGCGTG-3’ , and 5’-GTAGAAGGTGGCGCGAAGGG-3’ were used to detect the Map3k11+ allele ( 160 bp ) and the Map3k11- allele ( 280 bp ) . Male mice ( 6–8 wk old ) were treated with tamoxifen ( Sigma-Aldrich ( St . Louis , MO ) ) dissolved in 2% ethanol / 98% sunflower seed oil by intraperitoneal injection ( 1 mg/mouse ) 5 times ( at 48 hr intervals ) . Mouse embryos were treated by oral gavage of Cre- female mice at 12 . 5 days post coitus ( dpc ) with 3 mg tamoxifen dissolved in 2% ethanol / 98% sunflower seed oil . The tamoxifen-treated pups were delivered by C-section at ~ 19 . 5 dpc and transferred to foster mothers . Unilateral FAL and laser doppler imaging was performed using 10–14 week old male mice as previously described ( Limbourg et al . , 2009; Craige et al . , 2011 ) with the following modifications . Two ligation protocols were performed . In one protocol we ligated the femoral artery at its origin . The second protocol involved ligation of the femoral artery between the proximal caudal femoral artery and the popliteal artery as well as ligation of the superficial epigastric artery . The second ligation schema allows for more blood flow to be diverted to the gracillis collateral circulation . Quantitative scores for ischemia and movement post-FAL were performed as described ( Chalothorn et al . , 2007 ) . The aortic ring assays were performed in collagen as previously described ( Baker et al . , 2012 ) . Choroidal neovascularization was induced in mice using a 532 nm laser as previously described ( Cashman et al . , 2011 ) . Four laser spots/eye were applied . The eyes were harvested 7 days post-injury , fixed in 4% paraformaldehyde at 4°C overnight and eyecups dissected and subjected to wholemount immunofluorescence analysis using a TCS SP2 Leica confocal microscope . One million congenic B16F10 melanoma cells ( American Type Culture Collection Cat# CRL-6475 , RRID:CVCL_0159 ) were injected subcutaneously on both flanks of mice . Tumors were harvested 2 weeks later , weighed , imaged using a Zeiss Stereo Discovery V12 stereomicroscope , fixed in 4% paraformaldehyde ( 4°C , 12 hr ) , dehydrated sequentially in 15% and 30% sucrose in PBS , and embedded in Optical Cutting Temperature ( OCT ) prior to preparing frozen sections ( 10 µm ) . Sections were allowed to dry at room temperature , rehydrated in PBS , blocked and permeabilized in 10% normal donkey serum , 0 . 1% Triton X-100 in PBS for 1 hr at RT and incubated with primary antibodies; mouse anti-smooth muscle actin ( 1:500 , Sigma-Aldrich Cat# F3777 , RRID:AB_476977 ) and rat anti-CD31 ( 1:50 , BD Biosciences ( San Jose , CA ) Cat# 558736 , RRID:AB_397095 ) in 1% BSA PBS for 2 hr at RT . Sections were washed 3 × 5 min each with PBS and incubated with Alexa Fluor 546-goat anti mouse and Alexa Fluor 488-goat anti rat antibodies in 1% BSA PBS for 1 hr at RT . Following washing as above , DNA was stained with DAPI , sections mounted in FluoromountG ( Southern Biotech ( Birmingham , AL ) ) and imaged on a TCS SP2 Leica confocal microscope . Myocardial infarction studies were done at the Partners Cardiovascular Physiology Core at Brigham and Women’s Hospital , as previously described ( Bauer et al . , 2011; Li et al . , 2011 ) . Briefly , adult male mice were anesthetized by IP injection of a mixture of ketamine ( 40 mg/kg ) and xylazine ( 10 mg/kg ) , intubated , and mechanically ventilated . Following thoracotomy , the pericardium was removed , and the proximal left coronary artery was permanently occluded with an intramural stitch . Echocardiography ( Vevo 2100 , VisualSonics Inc ) was performed at the Partners Cardiovascular Physiology Core at Brigham and Women’s Hospital as previously described ( Bauer et al . , 2011 ) . Two-dimensional and M-mode echocardiographic images were obtained from lightly sedated ( 1% isoflurane in oxygen ) mice and recorded . M-mode images were obtained from the parasternal short-axis view at the level of the papillary muscles and used for measurements . Blood pressure and heart rate measurements were done on 10–14 week old male mice using a noninvasive computerized tail cuff system ( BP-2000 , VisiTech Systems ) . Mice were trained for 1 week , and then systolic and diastolic blood pressure and heart rate were recorded as the mean of at least 16 successful measurements over 1 week . Aortas were harvested from mice , flushed and cleaned of periaortic fat as described ( Baker et al . , 2012 ) , cut into 2 mm long rings and equilibrated in Opti-MEM containing penicillin/streptomycin overnight at 37°C . Contraction and relaxation responses were measured using a 6-mL vessel myograph ( Danish Myo Technology ) as previously described ( 36 ) with the following modifications . Arterial contraction in response to increasing doses of phenylephrine ( Phe ) was recorded and expressed as percent of maximum contraction obtained in response to incubation in K-PSS ( 60 mM potassium-containing physiologic salt solution [mM: NaCl 130 , KCl 4 . 7 , KHPO4 1 . 18 , MgSO4 1 . 17 , CaCl2 1 . 6 , NaHCO3 14 . 9 , dextrose 5 . 5 , CaNa2/EDTA 0 . 03] ) . Vasorelaxation in response to increasing doses of acetylcholine was recorded following pre-contraction with Phe ( 10–6 M ) . Hindlimbs were scanned in air aligned axially on a Scanco ( Wayne , PA ) µCT 40 at 70kVp , 114 µA and a resolution of 10 µm . The region of interest ( ROI ) included the entire hindlimb . To obtain the bone/vasculature overlay image , a contour around the entire ROI was utilized and segmented to include all soft and hard tissue . A second contour of the same ROI with the bone removed was also performed and segmented . The segmentation parameters included the values 0 . 8 Gauss sigma , 1 . 0 Gauss support , and a threshold of 212–1000 ( density range of 500 mg of HA/cm3 ) . The two segmented files were overlaid using Scanco’s IPL Transparency program and a false color image of the resulting file was created using the 3D Display program . Mice were anesthetized ( 150 mg/kg ketamine and 13 mg/kg xylazine ) and treated by intravenous injection of 400 U heparin prior to thoracotomy . The right atrium was severed and the mice were maximally vasodilated by infusing , via the left ventricle , 30 ml normal saline containing 1 g/l adenosine , 4 mg/l papaverine , and 100 μg/ml heparin followed by 15 ml 2% formalin and ~0 . 5 ml uncatalyzed blue Microfil ( Flow Tech ) to help visualize the abdominal aorta during cannulation . Mice were then transected just below the diaphragm , the abdominal aorta was cannulated ( Mc-28 , Braintree Scientific ( Braintreet , MA ) ) and the vasculature perfused using a syringe gun ( IGSET-3510 , Medco ) with ~3 ml undiluted catalyzed blue Microfil or 10 ml of a warm 50% Bismuth ( prepared as described [Simons , 2008] ) / 7% gelatin in normal saline . The aorta and vena cava were then clamped and the perfusate was allowed to polymerize for at least an hour ( 4°C ) before the hindlimbs were harvested , the skin removed , and the limbs placed in 10% formalin . We dissected the medial surface of adductor muscles of fixed , dehydrated ( 70 and 100% ethanol ) , and Microfil-perfused hindlimbs . The tissue was cleared in methyl salicylate ( Sigma ) and imaged with a stereomicrosope ( Ziess ) . The Dil solution was prepared as previously described ( Li et al . , 2008 ) , but with the addition of a filtration ( 40 μm ) step to remove undissolved particles . P0 or P6 pups were euthanized by isofluorane inhalation , decapitated , and immediately perfused via the left ventricle with 3 or 5 ml , respectively , of Dil solution using a 10 ml syringe and 27 gauge needle and/or the thoracic aorta using a micro cannula ( Mc-28 , Braintree Scientific ) . Pups were then rinsed with PBS and fixed/stored in 4% paraformaldehyde ( PFA ) at 4°C until dissected . Eyes were fixed in 4% paraformaldehyde ( RT , 1 hr or 4°C , 12 hr ) and retinas were dissected as previously described ( Pitulescu et al . , 2010 ) . P6 pups were perfused with PBS via the left ventricle ( and E16 . 5 and P0 pups were rinsed in PBS ) prior to fixation in 4% paraformaldehyde ( 4°C , 12 hr ) . Using a steromicroscope , the mice were transected below the diaphragm and a mid-sagital incision was performed to separate the two hindlimbs and the associated abdominal musculature . The skin and associated adipose tissue was carefully removed and the abdominal muscles isolated via incisions at their attachment to the pelvis and vertebral column . The entire medial surface of the hindlimb adductor muscles was harvested en block via dissection 1–2 mm around the saphenous , femoral and proximal caudal femoral arteries . Muscle tissues were then either cleared sequentially ( 70% and 90% glycerol/PBS , at least 5 hr each ) and mounted in 90% glycerol/PBS for direct visualization of GFP / Dil or processed for immunofluorescence analysis . A similar procedure was used to examine adult muscle . Muscles were blocked and permeabilized in 1% BSA , 0 . 5% Triton X-100 PBS ( 12 hr , 4°C ) . Tissues were equilibrated by washing 3 × 10 min each with Pblec buffer ( 1% Triton X-100 , 1 mM CaCl2 , 1 mM MgCl2 , and 1 mM MnCl2 in PBS pH 6 . 8 ) and incubated with biotinylated Griffonia simplifolica isolectin B4 ( iB4 , 1:25 , Vector Labs ( Burlingame , CA ) in Pblec buffer . Antibodies were diluted in 1% BSA , 1% normal donkey serum ( NDS ) , 1% Triton X-100 PBS and muscle samples were incubated in antibody solution for two days at 4°C . We used the following primary antibodies: FITC-conjugated smooth muscle actin ( 1:500 , Sigma-Aldrich Cat# F3777 , RRID:AB_476977 ) , goat anti-endomucin ( 1:100 , R & D Systems ( Minneapolis , MN ) Cat# AF4666 , RRID:AB_2100035 ) and mouse anti-Neurofilament-M ( 1:100 , Developmental Studies Hybridoma Bank ( University of Iowa ) , Cat# 2H3 , RRID:AB_531793 ) . The samples were washed 3 × 20 min each with 0 . 5% BSA , 0 . 5% Triton X-100 in PBS at room temperature ( RT ) . Fluorescence detection using secondary antibodies was performed by incubation with Alexa Fluor-488-conjugated streptavidin ( 1:100 , Invitrogen ( Carlsbad , CA ) ) and/or Alexa Fluor-546-conjugated donkey anti-goat or donkey anti-mouse antibodies ( 1:200 , Invitrogen ) in 1% BSA , 1% NDS , 1% Triton X-100 in PBS overnight at 4°C . Samples were washed 3 × 20 min each with 0 . 5% BSA , 0 . 5% Triton X-100 in PBS and once with PBS at RT and then cleared sequentially ( 70% and 90% glycerol/PBS , at least 5 hr each ) and mounted in 90% glycerol/PBS . Whole mount retina ( Pitulescu et al . , 2010 ) and ( retinal pigment epithelium ( RPE ) /choroid/sclera ) ( Cashman et al . , 2011 ) staining was performed as previously described . Samples were stained with biotinylated or Alexa Fluor-488- conjugated iB4 ( 1:25 , Vector Labs ) , rabbit anti-NG2 ( 1:200 , EMD Millipore ( Billerica , MA ) Cat# AB5320 , RRID:AB_91789 ) , or goat anti-Dll4 ( 1:100 , R & D Systems Cat# MAB1389 , RRID:AB_2092985 ) . Fluorescence detection was performed using Alexa Fluor-488-conjugated streptavidin , Alexa Fluor-546 or 633-conjugated secondary antibodies and Alexa Fluor 546-conjugated Phalloidin ( Invitrogen ) . DNA was stained with 1 µM 4 , 6'-diamidino-2-phenylindole ( DAPI ) or 10 μg/ml Hoechst ( both from Invitrogen ) in PBS for 10 min at RT and retinas and ( RPE/choroid/sclera ) were mounted in FluoromountG ( Southern Biotech ) . Mice were anesthetized ( 150 mg/kg ketamine and 13 mg/kg xylazine ) and treated by intravenous injection of 400 U heparin prior to thoracotomy . The right atrium was severed and the mice were maximally vasodilated by infusing , via the left ventricle , 20 ml normal saline containing 1 g/l adenosine , 4 mg/l papaverine and 100 μg/ml heparin followed by 10 ml 2% formalin . The skin was removed and entire hindlimbs were fixed in 10% formalin ( RT , 24 hr ) . Calf and adductor muscles were dissected en block from fixed hindlimbs , dehydrated and embedded in paraffin . Cross sections ( 7 μm ) were prepared and subjected to antigen retrieval using 1x antigen unmasking solution ( Vector Labs ) . The sections were blocked and permeabilized in 10% normal goat serum , 0 . 1% Triton X-100 in PBS for ( RT , 1 hr ) and incubated with Alexa Fluor 488-conjugated IsolectinB4 ( 1:25 , Vector Labs ) and primary antibodies , mouse anti-smooth muscle actin ( 1:500 , Sigma-Aldrich Cat# A5228 , RRID:AB_262054 ) and rat anti-CD31 ( 1:50 , BD Biosciences Cat# 558736 , RRID:AB_397095 ) in 1% BSA in PBS ( RT , 2 hr ) . Sections were washed 3 × 5 min each with PBS and incubated with Alexa Fluor 546-goat anti mouse and Alexa Fluor 488-goat anti rat antibodies in 1% BSA PBS for 1 hr at RT . Following washing as above , DNA was stained with DAPI , sections mounted in FluoromountG ( Southern Biotech ) and imaged on a TCS SP2 Leica confocal microscope . Whole mount muscle and retinal vasculature imaging was done on a Zeiss stereomicroscope or a TCS SP2 Leica confocal microscope . Maximum projection confocal images of the adductor muscle vasculature were generated from z-stacks ( 30–300 μm , 1–10 μm step size depending on specimen size , staining and objective used ) acquired starting at the medial surface of the adductor muscle specimens . To visualize large areas of the vasculature on the confocal microscope , a tile-scanning technique was employed whereby multiple overlapping ( 20–30% overlap ) maximum projection images were acquired with a 10x or 20x objective and a composite image was constructed by arraying the individual images in Photoshop . Quantification of vascularized area in whole mount retinas was done from fluorescence stereomicroscopic images using ZEN software ( Zeiss ) . Retinal angiogenic front vascular density , endothelial sprouts and filopodia were quantitated using ImageJ and maximum projection confocal images acquired with a 10x , 20x and 63x objective respectively . The number of tip cells and filopodia was measured as described ( Pitulescu et al . , 2010 ) . Primary MLEC cultures were prepared as described ( Kuhlencordt et al . , 2004 ) with minor modifications . Briefly , lungs were harvested aseptically , rinsed in Dulbecco's modified eagle medium ( DMEM ) , cut into small pieces and digested ( 1 hr , 37°C ) in 1 . 7 mg/ml collagenase ( Worthington ) . Lung digests were triturated by pipetting repeatedly through a 10 ml pipette fitted with a 1 ml pipette tip , passed through a 40 µm filter , and the cells obtained were cultured ( 2 days ) on gelatin-coated plates in MLEC medium ( 20% fetal bovine serum , 38% DMEM , 38% Ham's F-12 supplemented with 100 μg/mL endothelial cell growth supplement ( ECGS , Biomedical Technologies ) , 4 mM L-glutamine , 100 μg/mL heparin , and 1% penicillin/streptomycin ( Life Technologies ) . Endothelial cells were isolated by selection with rat anti-mouse intercellular adhesion molecule 2 ( ICAM2 ) antibody ( BD Biosciences Cat# 553326 , RRID:AB_394784 ) -coupled sheep anti-rat Ig magnetic beads ( Invitrogen ) and cultured for an additional 3–4 days . The cells were then subjected to a second round of selection using magnetic beads and then cultured for an additional 2 days before studies were performed . The purity of the primary endothelial cell cultures was examined by staining live cells with 1 , 1'-dioctadecyl - 3 , 3 , 3' , 3'-tetramethyl-indocarbocyanine perchlorate acetylated low-density lipoprotein ( Dil-Ac-LDL; BT-209; Alfa Aesar ( Ward Hill , MA ) ) and by staining with a PE-conjugated antibody to Cdh5 ( 1:50 , eBioscience Cat# 12-1441-82 , RRID:AB_1907346 ) . The staining was examined by fluorescence microscopy and flow cytometry . The 5-ethynyl-2'-deoxyuridine ( EdU ) incorporation assays were performed by incubation of cell cultures with 10 μM EdU ( 6 hr ) . The cells were processed for detection of EdU incorporation using the Click-iT EdU Alexa Fluor 488 Imaging Kit according to the manufacturer’s instructions ( Invitrogen ) . Confluent MLEC cultures in DMEM/F12 supplemented with 1% FBS were stimulated with 100 ng/ml VEGF-A or bFGF ( Peprotech ) or incubated in 1% O2 . Immunofluorescence analysis was done using cells fixed with 4% paraformaldehyde ( RT , 15 min ) . The cells were washed ( 3 × 10 min each ) with PBS and subsequently incubated in permeabilizaton/blocking buffer ( 10% normal goat serum ( NGS ) or NDS ( depending on the species of secondary antibody used ) , 0 . 1% Triton X-100 ( 1 hr , RT ) , and then incubated with primary antibodies , including PE-conjugated rat anti-Ki-67 ( 1:200 , eBioscience Cat# 12-5698-82 , RRID:AB_11150954 ) , mouse anti-αTubulin ( 1:500 , Sigma-Aldrich Cat# T5168 , RRID:AB_477579 ) , rat anti-Cdh5 ( 1:50 , BD Biosciences Cat# 550548 , RRID:AB_2244723 ) and goat anti-Dll4 ( 1:100 , R & D Systems Cat# MAB1389 , RRID:AB_2092985 ) in 1% BSA , 0 . 1% Triton X-100 ( 4°C , 12 hr ) and washed ( 3 × 10 min each ) with PBS . Fluorescence detection using secondary antibodies was performed by incubation ( RT , 2 hr ) with appropriate Alexa Fluor- 488 , 546 or 633-conjugated secondary antibodies ( 1:200 , Invitrogen ) . The cells were then washed ( 3 × 10 min each ) with PBS . DNA was stained with 4 , 6'-diamidino-2-phenylindole ( DAPI ) or Hoechst ( 1 μM , Invitrogen ) . The cells were mounted in FluoromountG ( Southern Biotech ) and imaged on a TCS SP2 Leica confocal microscope . Fluorescence was quantitated using ImageJ software . Primary MLECs ( 1 × 105 cells ) in 0 . 5% FBS DMEM/F12 were seeded in 8 well chamberslides ( BD Biosciences ) layered with 300 μl polymerized growth factor reduced matrigel ( BD Biosciences ) and incubated at 37°C for 8 hr . Tubular networks were imaged using a Zeiss inverted microscope . Confluent monolayers of primary MLECs in 96 well plates were simultaneously scratched using a 96-pin wound making tool ( WoundMaker , Essen Bioscience ) , rinsed twice with media and wound closure was monitored by automated live cell imaging on an IncuCyte ZOOM system ( Essen Bioscience ( Ann Arbor , MI ) ) using a 10x objective . The area between the edges of the wound in images taken at different time intervals was quantitated using ImageJ . To isolate RNA from tissues , mice were perfusion cleared with PBS via the left ventricle . Hindlimb adductor and calf skeletal muscles were harvested en block , snap frozen in liquid nitrogen , and then pulverized into a powder using a CryoPREP impactor ( Covaris ( Woburn , MA ) ) . Total RNA was extracted with TRIzol ( Life Technologies ) and was purified using an RNeasy kit ( Qiagen ) . RNA from cells and other tissues homogenized in RLT buffer ( Qiagen ) was isolated using the RNeasy kit . We used purified RNA to prepare cDNA using The High Capacity Reverse Transcription Kit ( Life Technologies ) . The expression of mRNA was examined by quantitative PCR analysis using a Quantstudio PCR system ( Life Technologies ) . TaqMan assays were used to quantitate Cdh5 ( Mm00486938_m1 ) , Dll4 ( Mm00444619_m1 ) , Emr1 ( Mm00802529_m1 ) , Hes1 ( Mm01342805_m1 ) , Hey1 ( Mm00468865_m1 ) , Lfng ( Mm00456128_m1 ) , Pecam1 ( Mm01242584_m1 ) , Slc2a1 ( Mm00441480_m1 ) and Vegfa ( Mm01281449_m1 ) . The relative mRNA expression was normalized by measurement of the amount of 18S RNA in each sample using TaqMan assays ( catalog number 4308329; Life Technologies ) . RNA was isolated using the RNeasy kit ( Qiagen ) . RNA quality ( RIN > 9 ) was verified using a Bioanalyzer 2100 System ( Agilent Technologies ) . Total RNA ( 10 μg ) from independent MLEC isolations ( lungs from 4 mice per isolation ) was used for the preparation of each RNA-seq library by following the manufacturer’s instructions ( Illumina ) . Three independent libraries were examined for each condition . The cDNA libraries were sequenced by Illumina Hi-Seq with a paired-end 40-bp format . Reads from each sample were aligned to the mouse genome ( UCSC genome browser mm10 build ) using TopHat2 ( Kim et al . , 2013 ) . The average number of aligned reads per library was > 20 , 000 , 000 . Endothelial cell gene expression was quantitated as fragments per kilobase of exon model per million mapped fragments ( FPKM ) using Cufflinks ( Trapnell et al . , 2010 ) . Differentially expressed genes were identified using the Cufflinks tools Cuffmerge and Cuffdiff . Differentially expressed genes were defined as those genes that were expressed ( Fragments Per Kilobase of exon per Million fragments mapped [FPKM] > 2 ) ; absolute log2-fold change > 0 . 5; q ≤ 0 . 05 . Gene ontology was examined by Kyoto Encyclopedia of Genes and Genome ( KEGG ) pathway analysis ( Kanehisa et al . , 2012 ) with the Database for Annotation , Visualization and Integrated Discovery ( DAVID ) ( Huang et al . , 2009 ) . Bone marrow ( BM ) was harvested by flushing tibias and femurs from at least three 10–12 week old mice with ice cold PBS . Erythrocytes were disrupted by incubation of BM in ACK lysing buffer ( Life Technologies ) . The BM cells were then resuspended in PBS and passed through a 100 μm filter . Cells were counted and mixtures of test BM cells from the indicated genotypes were prepared by mixing test BM cells expressing the CD45 . 2 allele with competitor BM cells from B6 . SJL-Ptprca Pepcb/BoyJ mice expressing the CD45 . 1 allele at a 20 test:80 competitor cell ratio . 1 × 106 total BM cells were intravenously injected via the tail vein into lethally irradiated ( 11 Gy ) 10–12 week old CD45 . 1/CD45 . 2 heterozygous female mice . Transplanted mice were maintained on antibiotic water for the first two weeks post transplantation . Blood was harvested via the retroorbital sinus using heparinized capillary tubes and EDTA-coated vials at 5 and 20 weeks post transplantation and subjected to flow cytometry analysis . CBC analysis was done using a HemaTrue hematology analyzer ( Heska ( Loveland , CO ) ) by the Department of Animal Medicine , University of Massachusetts Medical School . Blood was washed in PBS , stained with live/dead fixable blue dead cell staining kit ( Invitrogen ) , washed in PBS and blocked in 2% FBS-PBS / 0 . 02% sodium azide plus Fc-block ( Anti-CD16/32 antibody 1:200 , BD Biosciences ) . Surface antigens were detected by incubation for 30 min at 4°C with conjugated antibodies including CD45 . 1-eFluor 450 ( eBioscience Cat# 48-0453-82 , RRID:AB_1272189 ) , CD45 . 2-FITC ( BD Biosciences Cat# 553772 , RRID:AB_395041 ) , CD3e-APC ( BD Biosciences Cat# 561826 , RRID:AB_10896663 ) , CD19-APC-H7 ( BD Biosciences Cat# 560143 , RRID:AB_1645234 ) , CD11b-PE ( BD Biosciences Cat# 562287 , RRID:AB_11154216 ) and GR1-Alexa Fluor 700 ( BioLegend Cat# 108422 , RRID:AB_2137487 ) . Following washing with 2% FBS-PBS 0 . 02% sodium azide , red cells were lysed and leukocytes fixed by incubating in lyse/fix solution ( BD Biosciences ) . Cells were washed with PBS and analyzed on an LSR-II cytometer ( Becton Dickenson ) . Data were processed using FlowJo Software ( Tree Star ) . Cell extracts were prepared using Triton lysis buffer ( 20 mM Tris at pH 7 . 4 , 1% Triton X-100 , 10% glycerol , 137 mM NaCl , 2 mM EDTA , 25 mM β-glycerophosphate , 1 mM sodium orthovanadate , 1 mM phenylmethylsulfonylfluoride , 10 µg/mL of aprotinin plus leupeptin ) . Extracts ( 20–50 μg of protein ) were examined by protein immunoblot analysis by probing with antibodies to Cdh5 ( BD Biosciences Cat# 550548 , RRID:AB_2244723 ) , cJun ( Cell Signaling Technology Cat# 9165L , RRID:AB_2129578 ) , pSer63-cJun ( Cell Signaling Technology ( Danvers , MA ) Cat# 9261S , RRID:AB_2130162 ) , Dll4 ( R&D Systems Cat# MAB1389 , RRID:AB_2092985 ) , pERK ( Cell Signaling Technology Cat# 4370S , RRID:AB_2281741 ) , ERK ( Cell Signaling Technology Cat# 4695P , RRID:AB_10831042 ) , GAPDH ( Santa Cruz Biotechnology Cat# sc-365062 , RRID:AB_10847862 ) , JNK ( BD Biosciences Cat# 554285 , RRID:AB_395344 ) , pJNK ( Cell Signaling Technology Cat# 4668P , RRID:AB_10831195 ) , cleaved Notch1 ( NICD ) ( Cell Signaling Technology Cat# 4147S , RRID:AB_2153348 ) , and αTubulin ( Sigma-Aldrich Cat# T5168 , RRID:AB_477579 ) . Immune complexes were detected using the Odyssey infrared imaging system ( LI-COR Biotechnology ( Lincoln , NE ) ) . Differences between groups were examined for statistical significance with an unpaired Student's test with equal variance or a log-rank ( Mantel-Cox ) test for determining significance of Kaplan-Meier survival curves . All studies were performed with at least three biological replicates . The number of biological replicates is stated in the figure legends .	A blocked artery can have serious health consequences . For example , heart attacks and strokes are caused by such blockages . Artery blockages are harmful because tissues and organs are supplied with oxygen carried in blood cells and without adequate oxygen they may suffer damage or even die . The body has back up arteries or collateral arteries that help to mitigate the damage caused by a blood flow blockage . These arteries normally do not deliver blood to tissues . However , if an artery becomes blocked , these arteries can provide new and efficient routes of blood flow that can bypass the blockage . How well patients fair after an artery blockage depends on them having a working system of collateral arteries . These collateral vessels must quickly spring into action and even widen to accommodate adequate blood flow . However , little is known about how the collateral arteries are formed . Previous studies have suggested that an enzyme called cJun NH2-terminal kinase ( or JNK for short ) was essential for the formation of new blood vessels . Now , Ramo et al . show that while JNK is essential for the formation of collateral arteries during development , it is not required for the formation of blood vessels in adults . In the experiments , mice were genetically engineered to lack genes encoding two types of JNK in the cells that line the blood vessels . When these mice became adults they were still able to produce new blood vessels , just like normal mice . But the genetically engineered mice were missing the healthy collateral arteries that mice normally have in their muscles . Next , Ramo et al . blocked arteries in the muscles of mice and found that mice without JNK suffered worse injuries than normal mice . Further experiments showed that JNK helps to regulate genes that control blood vessel formation during early development . Without JNK , developing blood vessels grow into poorly organized networks instead of forming normal collateral arteries . The experiments suggest that humans who have mutations in the genes for JNK or related genes may be more susceptible to the harmful effects of artery blockages in the muscles . More studies are now needed to test this hypothesis .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "developmental", "biology", "cell", "biology" ]	2016	Suppression of ischemia in arterial occlusive disease by JNK-promoted native collateral artery development
Neuronal circuits' ability to maintain the delicate balance between stability and flexibility in changing environments is critical for normal neuronal functioning . However , to what extent individual neurons and neuronal populations maintain internal firing properties remains largely unknown . In this study , we show that distributions of spontaneous population firing rates and synchrony are subject to accurate homeostatic control following increase of synaptic inhibition in cultured hippocampal networks . Reduction in firing rate triggered synaptic and intrinsic adaptive responses operating as global homeostatic mechanisms to maintain firing macro-stability , without achieving local homeostasis at the single-neuron level . Adaptive mechanisms , while stabilizing population firing properties , reduced short-term facilitation essential for synaptic discrimination of input patterns . Thus , invariant ongoing population dynamics emerge from intrinsically unstable activity patterns of individual neurons and synapses . The observed differences in the precision of homeostatic control at different spatial scales challenge cell-autonomous theory of network homeostasis and suggest the existence of network-wide regulation rules . Neural circuits achieve an ongoing balance between flexibility and stability to enable plastic adaptations to environmental changes , while maintaining neuronal activity in a stable regime over extended timescales . The balance between stability and plasticity has rarely been addressed in the past due to the technical challenge of monitoring the activity of the same neurons over extended timescales ( Lütcke et al . , 2013 ) . Recently , great efforts have been made to establish a system for the monitoring of neuronal activity at long timescales in the hippocampus of freely moving mice ( Ziv et al . , 2013 ) . That study , performed by monitoring Ca2+ dynamics in thousands of CA1 cells over weeks through a miniature head-mounted microscope , revealed a remarkable degree of instability in the coding of space: only 25% of cells with place fields at one recording session exhibited the same properties 5 days later . Importantly , the diverse activity patterns at the single-neuron level gave rise to largely invariant spatial representations at the population level . Such results raise the question of whether inter-neuronal dynamics of spiking patterns stems from intrinsic variability of the hippocampal network and its constituent neurons , or whether they reflect extrinsic changes in hippocampal modulation by higher-order brain structures , adaptations to environmental changes , or subtle changes in animal behavior states . Diverse homeostatic negative feedback systems operate to stabilize ongoing spiking properties in neuronal populations around a predefined ‘set point’ in face of constant environmental changes ( Turrigiano and Nelson , 2004; Davis , 2006; Marder and Goaillard , 2006 ) . Extensive research lead to compelling evidence on a wide repertoire of possible homeostatic mechanisms , including adaptations of synaptic strength , changes in excitation–inhibition balance , and modulation of intrinsic excitability ( Turrigiano , 2011 ) . Despite progress in understanding the homeostatic mechanisms that underlie the stability of network firing properties , several key questions remain open . First , what are the basic properties of neural networks that are subjected to homeostatic control ? Second , are homeostatic control systems equally precise at the level of individual neurons and neuronal populations ? Third , what is the trigger of synaptic homeostatic mechanisms ? And finally , how do compensatory changes in synaptic strength affect the network's functions ? To address these questions , long-term measurements of spiking activity over several days from the exact same neuron or population of neurons are required . While long-term monitoring of spiking activity cannot be reliably performed from the same neurons in vivo due to technical limitations , greater stability can be achieved in vitro . Therefore , we combined long-term extracellular spike recordings at the population and single-neuron levels using micro-electrode arrays ( MEAs ) and calcium imaging , together with intracellular patch-clamp measurements of synaptic responses and imaging of synaptic activity . We utilized cultured neuronal networks that represent an excellent experimental tool to study internal variability vs stability in a highly controlled environment . Here , we describe the basic relationships between ongoing spiking properties of individual neurons , population dynamics , and neuronal adaptive mechanisms . While the majority of homeostatic mechanisms have been studied following complete pharmacological blockade of spikes or AMPAR-mediated excitatory postsynaptic currents ( EPSCs ) , we decided to examine long-term effects of neuromodulation on firing properties of the network . Specifically , we examined how use-dependent synaptic inhibition via G-protein-coupled receptors , a widespread mechanism of neuromodulation in the central nervous system ( CNS ) , affects firing properties of the network over long timescales . As a perturbation , we chose a suppression of synaptic activity via widely expressed GABAB receptors ( GABABRs ) using a selective GABABR agonist , baclofen . GABABRs mediate presynaptic inhibition via inhibition of presynaptic calcium transients ( Wu and Saggau , 1995; Laviv et al . , 2010 ) . Therefore , we first conducted experiments to determine the dose–response of baclofen at the level of presynaptic terminals utilizing FM1-43 dye in primary hippocampal cultures ( Abramov et al . , 2009 ) . As GABABR-mediated inhibition is frequency-dependent ( Varela et al . , 1997; Kreitzer and Regehr , 2000; Ohliger-Frerking et al . , 2003 ) , we quantified the effects of baclofen for two types of input: low-frequency single spikes ( 30 stimuli at 1 Hz ) and high-frequency spike bursts ( 30 stimuli delivered as 6 bursts; each burst contains 5 stimuli at 100 Hz ) , while maintaining the mean spiking rate constant at 1 Hz ( Figure 1A ) . As expected , baclofen inhibits synaptic vesicle release , displaying preferential effects during low-frequency stimulation ( Figure 1A ) . Quantitatively , IC50 for single spikes was ∼23-fold lower than for bursts ( 0 . 33 and 7 . 6 μM , respectively , Figure 1B ) . As a result , short-term facilitation was enhanced in a wide range of baclofen concentrations ( 0 . 01–10 μM , Figure 1C ) , indicating a conversion of synapses to a high-pass filter mode . Similar results were obtained using voltage-clamp experiments in acute hippocampal slices ( Figure 1—figure supplement 1A , B ) . 10 . 7554/eLife . 04378 . 003Figure 1 . Quantifying GABABR-mediated synaptic inhibition at synaptic , single neuron and network levels . ( A ) Representative ΔF images obtained by single ( 30 APs@0 . 2 Hz ) and burst ( 30 APs@6 bursts , each burst contained 5 APs; inter-spike-interval , 10 ms; inter-burst-interval , 5 s ) stimulations before and 10 min after 10 µM baclofen application . Scale bar: 5 μm . ( B ) Dose–response curve for the inhibitory presynaptic effect of baclofen during single vs burst stimulation patterns . Note , shift in the apparent IC50 from 0 . 33 µM in single stimulation ( n = 7–12 ) to 7 . 6 µM for burst stimulation type ( n = 8–10 ) . ( C ) Short-term plasticity detected by FM method ( calculated as Sburst/Ssingle ) was in the range of 2 . 0 ± 0 . 01 to 2 . 3 ± 0 . 01 for baclofen concentrations in the range of 0 . 1–10 µM ( n = 10–14 ) comparing to plasticity of control ( 1 . 4 ± 0 . 07 , n = 13 ) . ( D ) Top: image of dissociated hippocampal culture plated on MEA . Black circles at the end of the black lines are the recording electrodes . Scale bar: 200 µm . Bottom: representative trace of recording from one MEA channel . Spikes are detected based on a set threshold ( blue dashed line ) . Inset is a zoom to one detected spike . Scale bars: 8 µV , 35 ms ( 6 ms for insert ) . ( E ) Example of spike sorting for one channel . Bottom: each waveform is represented in principal component space forming three distinct clusters . Top: mean waveforms for each cluster . Scale bar: 10 µV , 1 ms . ( F ) Representative raster plots of MEA recording before and immediately after application of 10 µM baclofen ( *p < 0 . 0001 , unpaired , two-tailed Student's t-test ) . ( G ) Acute effect of baclofen on firing rate at the single-unit level . Left: 1 µM baclofen ( n = 5 , 260 units ) . Right: 10 µM baclofen ( n = 5 , 314 units ) . ( H ) Acute effect of baclofen on the mean firing rate of the population ( the same data as in G ) . Error bars represent SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 04378 . 00310 . 7554/eLife . 04378 . 004Figure 1—figure supplement 1 . Properties of baclofen-induced changes in synaptic dynamics and single-unit firing in hippocampal neurons . ( A ) EPSC ( −60 mV holding potential ) traces evoked by low-frequency stimulation and by high-frequency spike bursts under control and 10 min after 10 µM baclofen application in acute hippocampal slices . Scale bars: Top: 20 pA , 20 ms; Bottom: 50 pA , 50 ms . ( B ) Baclofen increases short-term synaptic facilitation calculated based on EPSC measurements ( 1 µM—n = 5 , p < 0 . 001; 10 µM—n = 4 , p < 0 . 001 ) in acute hippocampal slices . ( C ) Distribution of per unit firing rate during 1 hr of control MEA recording ( 490 units ) . Firing rates are heavily skewed towards low frequencies . ( D ) Representative raster plots of MEA recording before and immediately after application of 1 µM baclofen . ( E ) Per unit correlation between baseline firing rates and the percent change in firing rates after 2 days of control recording showing significant negative correlation ( Spearman r = −0 . 29 , p < 0 . 001 , 467 units ) . ( F ) Per unit correlation between baseline firing rates and the percent change in firing rates after 2 days in the presence of baclofen showing significant negative correlation ( Spearman r = −0 . 47 , p < 0 . 001 , 311 units ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04378 . 004 Next , we examined how use-dependent blockade of synaptic transmission affects firing properties of hippocampal neurons and networks . For this purpose , we grew high-density hippocampal cultures on MEAs for ∼3 weeks . Each MEA contains 59 recording electrodes with each electrode capable of recording the activity of several adjacent neurons ( Figure 1D ) . Spikes were detected and analyzed using principal component analysis to obtain well-separated single units ( Figure 1E ) . Unit firing rates were highly heterogeneous and skewed towards low frequencies ( Figure 1—figure supplement 1C ) as has been previously reported in vivo ( Mizuseki et al . , 2012 ) . As expected , baclofen caused a fast and robust drop in single-unit firing rates ( Figure 1F , G ) , leading to a reduction of the population mean firing rate to 25 ± 4% ( 1 µM baclofen , see raster plots in Figure 1—figure supplement 1D ) and to 1 . 2 ± 0 . 4% ( 10 µM baclofen ) of baseline values ( Figure 1H ) . Thus , we can use this system to study how chronic changes in the GABABR-mediated neuromodulation impact properties of individual synapses , single neurons , and neural networks . To assess how GABABR activation affects firing properties of the population at long timescales , we measured spiking activity during a baseline recording period and for 2 days following the application of 10 μM baclofen . As expected , baclofen caused a fast and pronounced drop in firing rates to 1 . 2 ± 0 . 4% of baseline values ( Figure 2A , C , D ) . We hypothesized that if the network has an intrinsically regulated firing rate ‘set point’ , as suggested by theories of homeostatic plasticity , we would see an increase in firing rates back to baseline values after administration of baclofen . Indeed , firing rates gradually returned to baseline values , reaching 57 ± 8% and 103 ± 14% of baseline after 1 and 2 days , respectively ( Figure 2C , D ) . Notably , the average firing rates of networks under control conditions showed changes of similar magnitude during two recording days ( without baclofen application , 113 ± 9% of baseline after 2 days; Figure 2B–D ) . Moreover , the characteristic log-normal distributions of single-unit firing patterns were essentially identical before and 2 days after baclofen application ( Figure 2E , p = 0 . 66; Kolmogorov–Smirnov test ) . Importantly , baclofen remains potent after over 2 days of exposure to recording conditions ( Figure 2—figure supplement 1A ) . Additionally , washout of baclofen following 2 days caused a significant increase in firing rate , indicating sustained activity of both baclofen and the GABABR-induced G-protein-mediated signaling ( Figure 2—figure supplement 1B–D ) . Moreover , the magnitude of the GABABR-mediated presynaptic inhibition , as measured by FM dyes , was not reduced following washout ( Figure 2—figure supplement 1E , F ) , indicating that GABABRs did not undergo desensitization under our experimental conditions . Taken together , these results demonstrate a robust and precise compensatory response at the level of population averages , confirming the idea that neuronal networks grown in vitro ( Turrigiano et al . , 1998 ) , similar to the circuits in vivo ( Hengen et al . , 2013; Keck et al . , 2013 ) , have the ability to homeostatically regulate their mean firing rate . 10 . 7554/eLife . 04378 . 005Figure 2 . Homeostatic regulation of firing rates is more precise at the network , than the single-unit level . ( A ) Analysis of the firing rate of each unit in a representative MEA experiment over the course of 2 days of recording in the presence of 10 µM baclofen . Representative units that precisely returned ( green ) , increased ( red ) , and decreased ( blue ) relative to baseline are highlighted . ( B ) Analysis of the firing rate of each unit in a representative MEA experiment over the course of 2 days of recording under control conditions . Representative units that didn't change ( green ) , increased ( red ) , and decreased ( blue ) relative to baseline are highlighted . ( C ) Mean firing rate of 48 hr control ( grey , n = 7 ) and baclofen ( blue , n = 5 ) MEA recordings . 3 hr of baseline rate are shown for baclofen experiments . ( For clarity , only every other hour is shown . ) Error bars represent SEM . ( D ) Statistical comparison of the representative time points ( the same data as in C ) . Error bars represent SEM . ( p < 0 . 0001 , baclofen compared to baseline; all control hours were not significantly changed; repeated-measures ANOVA with Bonferroni's multiple comparison test . ) ( E ) Distribution of unit firing rates ( log scale ) during baseline and after 2 days in the presence of baclofen . ( F ) Per unit correlation between baseline firing rates and firing rates after 2 days: Left: control ( n = 7; 490 units ) ; Right: in the presence of baclofen ( n = 5; 314 units ) . Colors represent units that significantly increased ( red ) , decreased ( blue ) , or remained stable ( green ) as determined by bootstrapping ( see ‘Materials and methods’ for details ) . Note log scale of both axes . ( G ) Summary of data in F ( p < 0 . 05; unpaired , two-tailed Student's t-test ) . Error bars represent SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 04378 . 00510 . 7554/eLife . 04378 . 006Figure 2—figure supplement 1 . Prolonged exposure to baclofen does not cause reduction in the sensitivity of synapses and neurons to the GABABR-mediated inhibition . ( A ) Baclofen ( 10 µM ) was added to a culture dish for 2 days . The medium from this dish was applied to a new MEA . The reduction of firing rate by pre-incubated baclofen was 4 ± 1% ( n = 50 , p < 0 . 0001 ) , similar to the one observed by fresh baclofen ( Figure 1G ) . ( B ) Washout of baclofen causes an increase in mean firing rate relative to both baseline and after 2 days baclofen incubation ( relative to baseline: 306% ± 67 , relative to 2 days baclofen: 296% ± 38; n = 77 units ) . ( C ) Per unit correlation between firing rates before and after washout of baclofen . Colors represent units that significantly increased ( red ) , decreased ( blue ) , or remained stable ( green ) as determined by bootstrapping . Note log scale of both axes . Same units as in ( B ) . ( D ) Summary of data in ( C ) . ( E ) Experimental protocol used for determining acute effect of baclofen after 2 days incubation with 10 µM baclofen and subsequent washout . ( F ) Baclofen ( 1 µM ) decreased FM staining to 52 ± 5% ( n = 18 ) and 49 ± 8% ( n = 8 ) relative to control with and without 2 days incubation with baclofen , respectively ( p = 0 . 89 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04378 . 00610 . 7554/eLife . 04378 . 007Figure 2—figure supplement 2 . Firing rate homeostasis is not precise at the level of multi-units . ( A ) Per unit correlation between baseline firing rates and firing rates after 2 days: Left: control ( n = 7; 192 multi-units ) ; Right: in the presence of baclofen ( n = 5; 133 multi-units ) . Colors represent units that significantly increased ( red ) , decreased ( blue ) , or remained stable ( green ) as determined by bootstrapping ( see ‘Materials and methods’ for details ) . Note log scale of both axes . ( B ) Summary of data in ( A ) ( unpaired , two-tailed Student's t-test ) . Error bars represent SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 04378 . 00710 . 7554/eLife . 04378 . 008Figure 2—figure supplement 3 . Effect of GABA uptake inhibitor on mean firing rate . Mean firing rate is decreased by 10 μM SKF89976A , a selective GAT-1 blocker , and fully recovers after 2 days ( n = 46 units ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04378 . 00810 . 7554/eLife . 04378 . 009Figure 2—figure supplement 4 . MEA analyses are robust over different parameters . ( A ) Histogram detailing the relationship between the mean firing rate of 20 min time segments and the mean firing rate of the full hour they represent . No representative time segment is more than 20% different than the full hour and 90% are within 10% ( n = 792 segments ) . ( B ) CV was calculated over 8 hr using time segments between 10 s and 1 hr . There was no significant difference when time segments greater than 10 min were used . ( C , D , E ) Per unit MFR during baseline and after 2 days of baclofen incubation were compared using bootstrapping statistics . No differences were observed when using different durations of time segments ( C , p > 0 . 3 ) , different bin sizes ( D , p > 0 . 9 ) , or different numbers of iterations ( E , p > 0 . 9 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04378 . 009 After establishing that mean firing rates of hippocampal networks are precisely restored even after a pronounced , ongoing perturbation , we asked whether this stability is maintained at the level of the individual neurons that comprise the network . First , we analyzed firing rates of individual units during 2 days of recording under control condition . While a high correlation was observed between firing rates at the baseline and after 2 days ( Spearman r = 0 . 8 , p < 0 . 0001 , n = 490 ) , a large fraction of units did not return to their baseline values ( Figure 2F , left ) . Only 23 ± 3% of units remained significantly unchanged during 2 days of control recordings , while 77 ± 3% changed their firing rates significantly ( Figure 2G ) . This phenomenon became even more pronounced after the application of baclofen , with only 11 ± 2% of units returning to the baseline , while 89 ± 2% were significantly changed ( Figure 2F right and 2G ) . The change in firing rate after 2 days of recordings was negatively correlated with the initial firing rate of the same unit under both control ( Figure 1—figure supplement 1E; Spearman r = −0 . 29 , p < 0 . 0001 , n = 490 ) and baclofen ( Figure 1—figure supplement 1F; Spearman r = −0 . 47 , p < 0 . 0001 , n = 305 ) conditions , due to a larger relative variability of low firing units . Notably , when analyzing multiunit recordings per electrode , firing rates after 2 days of recording displayed statistically significant change from the baseline in ∼80% of the cases ( Figure 2—figure supplement 2A , B ) . To ensure that per unit variability is not due to a possible movement of neurons in and out from recording electrodes , we measured the movement of individual neurons over the course of 2 days and found that the mean movement of neurons ( 1 . 6 ± 0 . 1 μM ) was negligible in comparison to the size of the electrode ( 30 µM ) and neuron's cell body ( 11 . 4 ± 0 . 3 µM ) . These results confirm that the observed variability at the level of single units does not result from imperfect spike sorting or neuronal mobility . Taken together , these data indicate the existence of significant dynamics at the level of the individual hippocampal neurons in mature hippocampal cultures , operating under the constraint that the overall firing rates do not change . The observed instability of firing rates at the individual neuron level becomes even more pronounced following the perturbation . Thus , the same mean firing rate of the population arises from variable firing rates of individual neurons . As fluorescent calcium sensors are widely used to image neural activity , we performed somatic calcium imaging in visually identified individual neurons to verify whether the stabilization of population mean firing rates emerges from variable firing rates of single neurons following chronic perturbation . We utilized the genetically encoded Ca2+ indicator GCaMP6f that displays single-action-potential sensitivity and broad dynamic range ( Chen et al . , 2013 ) . An AAV-based gene delivery system under the synapsin promoter was used to express GCaMP6f in hippocampal neurons . GCaMP6f-expressing neurons displayed spontaneous Ca2+ dynamics typically consisting of transients of varying amplitudes corresponding to single spikes and spike bursts ( Figure 3A ) . We first established the relationship between the calcium transients and spiking activity . Single spikes induced robust calcium transients ( 24 . 6 ± 1 . 4% ΔF/F; Figure 3—figure supplement 1A ) . Although the exact number of spikes could not be accurately predicted from the size of the calcium transients , high-frequency bursts consisting of 2–10 spikes at 100 Hz induced significantly higher ΔF/F peak amplitudes as compared with single spikes and the average ΔF/F linearly correlated with the number of spikes ( Figure 3—figure supplement 1B , C ) . Thus , the ΔF/F peak amplitude of calcium transients is approximately proportional to the number of spikes that triggered it . 10 . 7554/eLife . 04378 . 010Figure 3 . Calcium imaging confirms more precise homeostatic regulation of firing rates at the network , than the single-unit level . ( A ) Representative traces ( ΔF/F ) showing the same 16 neurons before , 10 min after 10 µM baclofen application , and after 2 days in the presence of baclofen . Bar scales: 50% ΔF/F , 5 s . ( B ) Pseudo-color coded image showing cultured neurons expressing GCaMP6f . Scale bar: 50 µm . ( C ) Summary of the mean rate ( averaged peak amplitude per min ) before ( green ) , after 10 min ( black ) and following 2 days ( purple ) of baclofen application . Population mean rates were restored after 2 days ( n = 4 , 192 neurons ) . ( D ) Distribution of neuron ΔF/F rates ( log scale ) during baseline and after 2 days in the presence of baclofen . ( E ) Per neuron correlation between baseline firing rates and firing rates after 2 days in the presence of baclofen ( n = 4; 192 neurons ) . Colors represent neurons that significantly increased ( red ) , decreased ( blue ) , or remained stable ( green ) as determined by bootstrapping . Note log scale of both axes . ( F ) Summary of data in E ( 54 . 3% ± 6 cells were significantly changed ) . Error bars represent SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 04378 . 01010 . 7554/eLife . 04378 . 011Figure 3—figure supplement 1 . Calcium imaging using GCaMP6f sensor in cultured hippocampal neurons . ( A ) Fluorescence change in response to one action potential ( AP ) . Single sweeps ( grey ) and averages of 10 sweeps ( blue ) are overlaid . Scale bars: 10% ΔF/F , 500 ms . ( B ) Fluorescence change ( average of 10 sweeps ) in response to 1—10 APs ( 3 trials ) . Scale bars: 200% ΔF/F , 2 s . ( C ) Peak fluorescence change as a function of number of APs ( normalized to the peak fluorescence changed induced by 1 AP ) . A linear relationship between ΔF/F and number of APs ( slope of linear fit is 1 . 928 ± 0 . 078 , r2 = 0 . 99 , n = 5 ) . Recurrent activity was suppressed with 20 µM DNQX and 50 µM AP5 in ( A–B ) . ( D ) Example of cultured neurons before and 2 days following baclofen incubation show no notable change in cell appearance . Scale bar: 50 µm . ( E ) Baseline firing rates 1 hr and after 2 days of control recording show stability with no evidence of phototoxicity ( 62 neurons ) . ( F ) Per unit correlation between baseline ΔF/F mean rates and the percent change in ΔF/F rates after 2 days in the presence of baclofen showing significant negative correlation ( Spearman r = −0 . 33 , p < 0 . 0001 , 192 neurons ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04378 . 011 To examine the stability of single neurons and neuronal populations at extended timescales , we performed time-lapse imaging of somatic Ca2+ dynamics during a baseline recording period , after acute baclofen application and for 2 days following baclofen application . Each imaging session had a duration of 20 min , fairly representing firing stability of several hours ( Figure 2—figure supplement 4 ) . Analysis of Ca2+ dynamics in 192 neurons shows that 10 μM baclofen blocked Ca2+ transients acutely , while 48 hr after baclofen incubation the mean Ca2+-transient amplitude returned to the baseline level ( Figure 3A , C ) . Notably , the distributions of single-neuron Ca2+-transient sizes were skewed towards low-activity levels ( Figure 3D ) as seen for electrophysiological measurements of spiking activity , indicating that the population of visually selected neurons does not represent a frequency-dependent bias . In addition , the distributions were indistinguishable before and 2 days after baclofen application ( Figure 3D ) . Despite stability of Ca2+ dynamics at the population level , 54 ± 6% of neurons did not return to their baseline values 2 days after baclofen application ( Figure 3E , F ) . The change in Ca2+-transient amplitudes after 2 days of baclofen incubation was negatively correlated with the initial amplitudes at the same neuron ( Figure 3—figure supplement 1F ) . Thus , calcium imaging in identified neurons confirms the electrophysiological data , strengthening our conclusion regarding the stabilization of population dynamics despite imprecise homeostatic compensations at the level of single neurons . We next investigated the persistence of temporal firing patterns in hippocampal networks . We utilized electrophysiological recordings by MEAs that enable superior temporal resolution compared to calcium imaging . Figure 4A presents raster plots during periods of baseline , 4 hr and 2 days following baclofen application in a single experiment . We first ensured that there was no significant change in firing synchrony , as estimated by the fraction of spikes participating in network-bursts and burst duration ( Figure 4B , see ‘Materials and methods’ for burst detection ) , under control conditions . Baclofen transiently increased population firing synchrony by increasing the fraction of spikes participating in bursts , burst duration ( Figure 4B , C ) , and number of spikes per network burst ( Figure 4—figure supplement 1A ) . The observed GABABR-mediated increase in firing synchrony returned to baseline values after 14 hr ( Figure 4B , C ) . These results indicate that the population-burst pattern , similar to the population firing rate , undergoes a precise homeostatic compensation . 10 . 7554/eLife . 04378 . 012Figure 4 . Temporal firing patterns are homeostatically regulated . ( A ) Representative raster plot of MEA recording before , 4 hr and 2 days after application of baclofen . ( B ) Baclofen causes a transient increase in fraction of spikes participating in network bursts ( hours 2–4 , p < 0 . 01; hours 4–10 , p < 0 . 001; hour 12 , p < 0 . 01; hour 14 , p < 0 . 05; repeated-measures ANOVA with Bonferroni's multiple comparison test , compared to baseline values ) . The number of spikes that are part of network bursts was divided by the total number of spikes . ( C ) Baclofen causes a transient , short-lived , increase in duration of network bursts ( hours 2–4 , p < 0 . 001; hour 6 , p < 0 . 05; repeated-measures ANOVA with Bonferroni's multiple comparison test , compared to baseline values ) . ( D ) Baclofen causes a transient increase in fraction of spikes participating in single-unit bursts ( hours 2–4 , p < 0 . 01; hours 4–12 , p < 0 . 001; hour 14 , p < 0 . 01; repeated-measures ANOVA with Bonferroni's multiple comparison test , compared to baseline values ) . The number of spikes that are part of single-unit bursts was divided by the total number of spikes of that unit . ( E ) Baclofen causes a transient , short-lived , increase in duration of single-unit bursts ( hours 2 , p < 0 . 001; repeated-measures ANOVA with Bonferroni's multiple comparison test , compared to baseline values ) . ( F ) Per unit correlation of fraction of spikes in single-unit bursts between baseline and 2 days after baclofen application: Top: control ( n = 7; 279 units ) ; Bottom: in the presence of baclofen ( n = 5; 234 units ) . Colors represent units that significantly increased ( red ) , decreased ( blue ) , or remained stable ( green ) as determined by bootstrapping ( see ‘Materials and methods’ for details ) . Note log scale of both axes . ( G ) Summary of data in ( D ) ( p < 0 . 05 , unpaired , two-tailed Student's t-test ) . Error bars represent SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 04378 . 01210 . 7554/eLife . 04378 . 013Figure 4—figure supplement 1 . Number of spikes in network- and single-unit bursts in the presence of baclofen . ( A ) Baclofen causes a transient increase in the number of spikes per network burst ( hours 2–4 , p < 0 . 001; repeated-measures ANOVA with Bonferroni's multiple comparison test , compared to baseline values ) . ( B ) Baclofen causes a transient increase in the number of spikes per unit burst ( hour 2 , p < 0 . 001; repeated-measures ANOVA with Bonferroni's multiple comparison test , compared to baseline values ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04378 . 01310 . 7554/eLife . 04378 . 014Figure 4—figure supplement 2 . Single-unit burst characteristics are stable across a large range of thresholds . ( A ) Single-unit burst data for experiments from Figure 4D–G after using a burst threshold of 100 Hz , minimum three spikes per burst . ( B ) Single-unit burst data for experiments from Figure 4D–G after using a burst threshold of 20 Hz , minimum three spikes per burst . DOI: http://dx . doi . org/10 . 7554/eLife . 04378 . 014 We then asked whether the observed stability of population firing synchrony results from stability of spike patterns at the single-unit level . Here too , we ensured that there was no change in single-unit bursts under control conditions ( Figure 4D , E and Figure 4—figure supplement 1B ) . Interestingly , following baclofen application , the average fraction of spikes participating in single-unit bursts returned to baseline levels with dynamics very similar to those of network bursts ( Figure 4D ) . However , single-unit burst analysis per individual unit shows a high degree of variability . Under control conditions , 42 ± 2% of units significantly changed the fraction of spikes in bursts , while 56 ± 2% remained unchanged . 2 days after baclofen application , 63 ± 8% were significantly changed , while only 37 ± 8% of units remained unchanged ( Figure 4F , G ) . Notably , bursts characteristics were not significantly changed , outside of the first 2 hr following baclofen application ( Figure 4E and Figure 4—figure supplement 1B ) . These data indicate that firing synchrony of spontaneous activity is maintained at the network level , despite large changes in burst patterns at the single-unit level . The inhibition–excitation ( I/E ) ratio constitutes an important factor in firing rate homeostasis ( Liu , 2004; Maffei et al . , 2004; Maffei and Turrigiano , 2008 ) . To assess the effect of chronic increase in the GABABR activity on the I/E ratio , we isolated the spontaneous excitatory and inhibitory currents ( sEPSCs , sIPSCs , respectively ) at the same cell based on the reversal potentials of AMPAR-mediated excitatory and GABAAR-mediated inhibitory currents ( Figure 5A , B ) . We then calculated the integrated excitatory and inhibitory conductances ( GE , GI , respectively; see ‘Materials and methods’ ) . 10 . 7554/eLife . 04378 . 015Figure 5 . Dynamics of I/E ratio following chronic GABABR-mediated inhibition . ( A ) Image of patched neuron . Alexa-fluor 488 ( 10 µM ) was added to the patch pipette for imaging . Scale bar: 20 µm . ( B ) Representative traces of sEPSCs ( −65 mV holding potential , bottom ) and sIPSCs ( +10 mV holding potential , top ) for control , 4 hr and 2 days preincubation with baclofen . Measurements of sEPSCs following baclofen pre-incubation were done in the presence of baclofen . ( C ) Mean integrated excitatory conductance ( GE ) in control ( n = 32 ) , following acute ( n = 10 ) , 4 hr ( n = 11 ) and 2 days ( n = 16 ) baclofen application . Excitatory conductance is completely restored following 2 days of exposure to baclofen . Error bars represent SEM . ( p < 0 . 05 , *p < 0 . 0001; Kruskal–Wallis test with Dunn's multiple comparison test ) . ( D ) Mean integrated inhibitory conductance ( GI , same cells as B ) . Inhibitory conductance is completely restored as well following 2 days of exposure to baclofen . Error bars represent SEM . ( p < 0 . 001; Kruskal–Wallis test with Dunn's multiple comparison test ) . ( E ) Mean I/E ratio per neuron ( same cells as in C , D ) . The I/E ratio of each cell was calculated and the resulting ratios were averaged . p < 0 . 05 , p < 0 . 01 , p < 0 . 001; one-way ANOVA with Dunnett's multiple comparison test . Error bars represent SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 04378 . 015 As expected , acute application of baclofen almost completely blocked both sEPSCs and sIPSCs , from 0 . 40 ± 0 . 07 and 0 . 87 ± 0 . 1 nS to 0 . 02 ± 0 . 006 and 0 . 06 ± 0 . 02 nS for GE and GI , respectively ( Figure 5C , D ) . After 4 hr of incubation with baclofen , GE and GI partially recovered to 0 . 16 ± 0 . 07 nS and 0 . 61 ± 0 . 2 nS , respectively ( Figure 5C , D ) , suggesting a faster rate of GI homeostatic regulation . Finally , after 2 days in the presence of baclofen , the network showed full recovery with GE reaching 0 . 44 ± 0 . 1 nS and GI reaching 1 . 02 ± 0 . 2 nS ( p > 0 . 8 , Figure 5C , D ) . As a result , I/E ratio was transiently increased from 3 . 1 ± 0 . 4 to 9 . 9 ± 2 . 8 following 4 hr baclofen incubation ( p < 0 . 001 ) , returning to the baseline levels following 2 days of baclofen ( 3 . 04 ± 0 . 5 , p > 0 . 5; Figure 5E ) . These data show that I/E balance of the network is tightly regulated , supporting homeostatic restoration of spontaneous spiking activity of the network even under a constant increase in the GABABR activity . Having observed a recovery of the firing properties at the population level , we next investigated which homeostatic mechanisms were underlying these changes . To accomplish this , we recorded mEPSCs from hippocampal neurons under control conditions and at different time points following baclofen application ( Figure 6A ) . While acute baclofen application did not have a significant effect on mEPSCs ( Figure 6—figure supplement 1 ) as has been reported previously ( Lei and McBain , 2003 ) , chronic baclofen incubation triggered gradual changes in mEPSC frequency and amplitude . We did not observe a significant increase in mEPSC amplitude 4 hr after baclofen application ( 24 . 7 ± 1 . 3 and 27 . 7 ± 3 . 3 pA in control and 4 hr after baclofen , respectively , p = 0 . 46 ) . However , a 1 . 25-fold increase to 30 . 8 ± 2 . 2 pA ( p < 0 . 05 ) was detected 2 days following baclofen application ( Figure 6B , C ) . This was coupled with twofold increase in the frequency of mEPSCs from 2 . 1 ± 0 . 7 to 4 . 3 ± 1 . 5 Hz ( p < 0 . 05 ) following 4 hr baclofen application and a large 4 . 8-fold increase to 10 ± 3 . 2 Hz ( p < 0 . 01 ) following 2 days ( Figure 6D , E ) . These data indicate that both pre- and postsynaptic modifications of quantal excitatory synaptic transmission occur following chronic , use-dependent inhibition of the evoked synaptic activity . 10 . 7554/eLife . 04378 . 016Figure 6 . Chronic GABABR-mediated inhibition triggers an increase in mEPSC frequency and amplitude . ( A ) Representative traces of mEPSCs for control , 4 hr and 2 day incubations in baclofen . Scale bar: 40 pA , 200 ms . Measurements of mEPSCs were done in the presence of baclofen . ( B ) Cumulative histograms of mEPSC amplitudes in control ( n = 30 ) and following 4 hr ( n = 11 ) and 2 days ( n = 15 ) of incubation with baclofen . The mean of mEPSC amplitude increased from 25 . 4 pA in control to 27 . 7 and 30 . 8 pA following 4 hr and 2 days of baclofen application , respectively . ( C ) Summary of data in B . Mean mEPSC amplitude is significantly elevated 1 . 25-fold ( p < 0 . 05 ) only after 2 days in baclofen . Error bars represent SEM . p < 0 . 05; one-way ANOVA with Dunnett's multiple comparison test . ( D ) Cumulative histogram of mEPSC inter-event intervals showing a gradual shift to smaller values from control through 4 hr baclofen to 2 days baclofen incubation ( the same experiments as in C ) . ( E ) Summary of data in D . mEPSC frequency is increased twofold after 4 hr ( p < 0 . 05 ) and 4 . 6-fold after 2 days ( p < 0 . 01 ) incubation in baclofen . p < 0 . 05 , p < 0 . 01; one-way ANOVA with Dunnett's multiple comparison test . Error bars represent SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 04378 . 01610 . 7554/eLife . 04378 . 017Figure 6—figure supplement 1 . Baclofen does not affect mEPSC frequency and amplitude acutely . ( A ) mEPSC amplitude is not effected by acute application of baclofen ( Cnt: 25 . 4 ± 1 . 5 pA , n = 31; acute Bac: 24 . 4 ± 2 . 9 , n = 14 , p = 0 . 72 , unpaired , two-tailed Student's t-test ) . ( B ) mEPSC frequency is slightly , but not significantly , lowered by acute application of baclofen ( Cnt: 2 . 1 ± 0 . 7 Hz , n = 31; acute Bac: 1 . 7 ± 0 . 6 Hz , n = 14 , p = 0 . 7 , unpaired , two-tailed Student's t-test ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04378 . 017 In addition to synaptic homeostatic mechanisms , modulation of intrinsic excitability is an important facet of neuronal adaptation ( Desai et al . , 1999; Kim and Tsien , 2008; Maffei and Turrigiano , 2008 ) . To examine the effect of chronic GABABR activation on intrinsic electrophysiological properties of hippocampal neurons , we incubated cultures with 10 µM baclofen for 4 hr and 2 days . We then elicited action potentials ( APs ) in response to increasing somatic current injections ranging from −40 to +180 pA ( F–I curves ) in the presence of postsynaptic receptor blockers . While there was no difference in F–I curves between control and 4 hr baclofen incubation , 2 days of baclofen incubation caused a significant leftward shift of the curve ( Figure 7A , B ) . Additionally , resting membrane potential ( RMP , Figure 7C ) was −64 . 8 ± 1 . 7 mV in untreated neurons , became more depolarized already after 4 hr ( −58 . 2 ± 2 . 2 , p < 0 . 05 ) and further depolarized after 2 days ( −55 . 4 ± 1 . 5 mV , p < 0 . 001 ) . Finally , input resistance ( Rin ) , determined by the voltage response to increasing somatic current injections ranging from −80 to −20 pA , showed a tendency towards larger values after 4 hr and was significantly increased following 2 days baclofen incubation ( from 233 ± 14 to 373 ± 59 MΩ , p < 0 . 01; Figure 7D , E ) . These data show that , in addition to synaptic modifications , increased intrinsic excitability contributes to the homeostatic restoration of network firing properties following chronic GABABR-mediated inhibition . 10 . 7554/eLife . 04378 . 018Figure 7 . Intrinsic excitability is increased in response to activity suppression . ( A ) Representative traces of voltage responses evoked by 20 pA step of current injections after control , 4 hr and 2 days baclofen incubation , elicited from RMP ( scale bars: 40 mV , 100 ms ) . ( B ) F–I relationship after control , 4 hr and 2 days baclofen incubation . After 2 days incubation there is a significant leftward shift of the curve showing greater excitability ( control , n = 18; 4 hr , n = 18; 2 days , n = 16; p < 0 . 05 ) following long-term GABABR activation . ( C ) RMP is depolarized after short baclofen incubation ( control , n = 19; 4 hr , n = 21; 2 days , n = 14 ) . ( D ) I–V curve ( same cells as C ) . ( E ) Rin is significantly increased following 2 day baclofen incubation ( same cells as C ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04378 . 018 To assess whether the increase in mEPSC frequency is paralleled by modifications in the evoked synapse release probability , we quantified basal synaptic vesicle release at different time points following baclofen application utilizing FM1-43 dye ( Abramov et al . , 2009 ) . To this end , we quantified the total amount of releasable fluorescence at each bouton ( ΔF ) and the density of FM-positive puncta per image ( D ) following stimulation ( 30 stimuli at 1 Hz ) in the presence of 10 μM FM1-43 . Our results demonstrate a 1 . 5- and 1 . 8-fold increase in ΔF across synaptic populations following 4 hr and 2 days , respectively ( p < 0 . 001 , Figure 8A , B ) . To confirm that prolonged incubation with baclofen increases synaptic vesicle exocytosis , the total pool of recycling vesicles was stained by maximal stimulation ( 600 stimuli at 10 Hz ) and subsequently destained by 1 Hz stimulation . The destaining rate constant ( measured as 1/τdecay , where τdecay is an exponential time course ) increased by ∼36% following 2 days of baclofen incubation ( p < 0 . 01; Figure 8—figure supplement 1 ) . Thus , an increase in the release probability of hippocampal boutons constitutes an adaptive mechanism stabilizing firing properties of hippocampal network . 10 . 7554/eLife . 04378 . 019Figure 8 . Short-term synaptic plasticity is not preserved in networks with similar firing properties . ( A ) Cumulative histogram of fluorescence intensity of FM stained puncta following 30 stimuli given at 1 Hz . ( ΔFsingle control , n = 15; 6099 puncta; 4 hr , n = 15 , 6269 puncta; 2 days , n = 13; 5551 puncta . ) ( B ) Summary of mean ΔFsingle from ( A ) . Mean ΔFsingle is increased already after 4 hr of incubation with baclofen and remains high after 2 days of baclofen incubation . *p < 0 . 001; one-way ANOVA with Tukey's multiple comparison test . ( C ) Experimental protocol used for STP experiments . ( D ) Representative images of FM1-43 staining from STP experiments . Note the increase in fluorescence intensity after 1 Hz stimulation following baclofen incubation . Scale bar: 5 µm . ( E ) The mean burst-to-single ratio of S is significantly decreased following baclofen incubation ( control , n = 15; 4 hr , n = 15; 2 days , n = 13; p < 0 . 0001 ) . p < 0 . 001; one-way ANOVA with Tukey's multiple comparison test . Error bars represent SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 04378 . 01910 . 7554/eLife . 04378 . 020Figure 8—figure supplement 1 . Synaptic vesicle exocytosis evoked by 1 Hz stimulation is increased after 2 days baclofen incubation . ( A ) FM1-43 destaining rate curves of 50 synapses under control and 2 days after 10 μM baclofen application . ( B ) Averaged destaining rate constants in control ( n = 5 ) and in cultures pretreated for 2 days with baclofen ( n = 6 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04378 . 020 Given an inverse correlation between release probability and short-term synaptic facilitation ( Debanne et al . , 1996; Dobrunz and Stevens , 1997 ) , homeostasis of mean firing rate is expected to be paralleled by concurrent changes in short-term synaptic plasticity . The total presynaptic strength within a given region of the hippocampal network ( S ) can be estimated as the product of ΔF and D ( S = ΔF × D ) . The magnitude and the sign of short-term presynaptic plasticity ( Sburst/Ssingle ) were calculated by dividing the total number of vesicles recycled due to bursts by the number of vesicles recycled by a similar number of single spikes in the same population of synapses ( Figure 8C ) . Indeed , short-term synaptic facilitation was decreased from 1 . 9- to 1 . 2-fold after 4 hr ( p < 0 . 001 , Figure 8D , E ) and was completely abolished following 2 days ( p < 0 . 001 , Figure 8D , E ) of incubation with baclofen . These results demonstrate that short-term synaptic plasticity is not preserved under homeostasis of population firing properties . Having established a relationship between the use-dependent blockade of synaptic transmission via GABABRs and spontaneous firing properties , we asked whether use-independent blockade of synaptic transmission triggers similar effects . For this purpose , we blocked fast excitatory synaptic transmission by using the AMPA receptor ( AMPAR ) blocker CNQX . While previous studies demonstrated that chronic AMPAR blockade increases both the frequency and the amplitude of mEPSCs ( Thiagarajan et al . , 2002 ) already 4 hr after application of the antagonist ( Jakawich et al . , 2010 ) , its effect on firing properties of the network has not been explored yet . CNQX ( 10 µM ) caused a reduction in the population mean firing rate to 52 ± 12% of baseline values immediately after application ( p < 0 . 01; Figure 9B ) . Application of NMDAR and AMPAR blockers together resulted in ∼80% of firing rate inhibition ( Figure 9—figure supplement 1 ) , indicating that NMDARs contribute ∼30% to spontaneous spiking activity . The residual ∼20% may arise from electrical coupling via gap junctions ( Hormuzdi et al . , 2001 ) , intrinsically bursting neurons ( Yue and Yaari , 2004 ) , ephaptic effects , and/or rebound spiking following inhibition . While CNQX triggered a compensatory increase in synapse release probability already 4 hr after its application ( Figure 9F , G and ( Jakawich et al . , 2010 ) ) , the population mean firing rate remained largely uncompensated , staying at 40 ± 5% of baseline 2 days following the perturbation ( Figure 9A , B ) . Likewise , there was a significant shift in the single-unit firing rates towards lower frequencies after 2 days ( p < 0 . 001 , Kolmogorov–Smirnov test; Figure 9C ) . When examining changes at the individual unit level , we found that only 16 ± 8% of units remained significantly unchanged during 2 days of the perturbation , while 71 ± 8% decreased their firing rates ( p < 0 . 001 , Figure 9D , E ) . It is noteworthy that CNQX and AMPAR-mediated signaling remain active even after 2 days incubation as evidenced by the increase in network and single unit firing rate following washout ( Figure 9—figure supplement 2 ) . 10 . 7554/eLife . 04378 . 021Figure 9 . Effects of chronic AMPAR blockade on spontaneous network firing . ( A ) Representative raster plot of MEA recording before and 2 days after application of CNQX . ( B ) Changes in mean firing rate following 10 µM CNQX application utilizing MEA recordings ( n = 4 ) . 3 hr of baseline rates are shown . There is an immediate and prolonged reduction of firing rate ( p < 0 . 001 for all hours compared to baseline; one-way ANOVA with Bonferroni's multiple comparison test ) . For clarity , only every other hour is shown . Error bars represent SEM . ( C ) Distribution of unit firing rates ( log scale ) during baseline and after 2 days in the presence of CNQX . ( D ) Per unit correlation between baseline firing rates and firing rates after 2 days in the presence of CNQX ( n = 4 , 128 units ) . Colors represent units that significantly increased ( red ) , decreased ( blue ) , or remained stable ( green ) as determined by bootstrapping ( see ‘Materials and methods’ for details ) . Note log scale of both axes . ( E ) Summary of data in ( D ) . p < 0 . 001; one-way ANOVA with Tukey's multiple comparison test . ( F ) Cumulative histogram of fluorescence intensity of FM1-43 stained puncta following 30 stimuli given at 1 Hz ( ΔFsingle control , n = 9 , 3631 puncta; 2 days CNQX , n = 9 , 5251 puncta ) . ( G ) Summary of mean ΔFsingle from ( F ) . Mean ΔFsingle is increased already after 4 hr of incubation with baclofen and remains high after 2 days of baclofen incubation . p < 0 . 001; one-way ANOVA with Tukey's multiple comparison test . ( H ) Normalized fraction of spikes in network bursts ( the same experiments as in B ) . There was a significant reduction from the fourth hour onward ( p < 0 . 001; repeated-measures ANOVA with Bonferroni's multiple comparison test ) . ( I ) Normalized fraction of spikes in single-unit bursts ( the same experiments as in B ) . There was a significant reduction from the sixth hour onward ( hours 6–10 , p < 0 . 05; hours 12–36 , p < 0 . 01; hours 36–48 , p < 0 . 001; repeated-measures ANOVA with Bonferroni's multiple comparison test ) . Error bars represent SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 04378 . 02110 . 7554/eLife . 04378 . 022Figure 9—figure supplement 1 . Effect of AMPAR and NMDAR blockers on mean firing rate measured by MEA in hippocampal cultures . Addition of 10 µM CNQX together with 50 µM AP5 reduced mean firing rate by 80% . DOI: http://dx . doi . org/10 . 7554/eLife . 04378 . 02210 . 7554/eLife . 04378 . 023Figure 9—figure supplement 2 . CNQX washout reveals an increase in the mean firing rate . ( A ) Washout of CNQX causes an increase in MFR relative to both baseline and after 2 days CNQX incubation ( relative to baseline: 212% ± 28 , relative to 2 days CNQX: 442% ± 67; n = 51 units ) . ( B ) Per unit correlation between firing rates before and after washout of CNQX . Colors represent units that significantly increased ( red ) , decreased ( blue ) , or remained stable ( green ) as determined by bootstrapping . Note log scale of both axes . Same units as in ( A ) . ( C ) Summary of data in ( C ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04378 . 02310 . 7554/eLife . 04378 . 024Figure 9—figure supplement 3 . Characteristics of network- and single-unit bursts in the presence of CNQX . ( A ) The mean number of spikes in each network burst is reduced by CNQX with no return to baseline ( hour 6 , p < 0 . 05; hour 8 , p < 0 . 01; hours 10–48 , p < 0 . 001; repeated-measures ANOVA with Bonferroni's multiple comparison test , compared to baseline values ) . ( B ) CNQX causes an acute increase in duration of network bursts ( p < 0 . 001; repeated-measures ANOVA with Bonferroni's multiple comparison test , compared to baseline values ) followed by a slow , non-significant reduction in duration . ( C ) The mean number of spikes in single-unit bursts was not changed by CNQX ( p > 0 . 05 , repeated-measures ANOVA with Bonferroni's multiple comparison test , compared to baseline values ) . ( D ) CNQX causes a small non-significant increase in duration of single-unit bursts ( repeated-measures ANOVA with Bonferroni's multiple comparison test , compared to baseline values ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04378 . 024 Analysis of firing pattern shows that CNQX caused a permanent reduction in the fraction of spikes participating in population bursts ( Figure 9A , H ) , as well as in the number of spikes comprising population bursts ( Figure 9—figure supplement 3A ) , reflecting a reduction in firing synchrony . Population-burst duration , however , showed a brief increase followed by a gradual , non-significant , decrease ( Figure 9—figure supplement 3B ) . A similar effect was observed at the single-unit level regarding the fraction of spikes participating in single-unit bursts , while burst properties were not significantly affected ( Figure 9I and Figure 9—figure supplement 3C , D ) . Taken together , these results suggest that use-independent postsynaptic blockade of excitatory synaptic transmission induces firing rate reduction with concomitant desynchronization of population firing . Although blockade of AMPAR-mediated excitatory synaptic transmission had a lower impact on firing rates than frequency-dependent synaptic inhibition via GABABRs ( Figure 2E ) , this effect cannot be efficiently compensated in the network . To relate the network's behavior to properties of its single components , we integrated , in this study , recordings with extracellular MEA , the intracellular patch-clamp recordings and high-resolution functional imaging at the single-synapse level . In particular , this system enables us to examine whether homeostatic mechanisms operating at a cell-autonomous level are sufficient to confer population firing stability ( Burrone et al . , 2002; Goold and Nicoll , 2010; Turrigiano , 2012 ) or whether homeostasis operates at the level of network-average properties . Our results show that population spiking rates and patterns are intrinsically stable not only under basal conditions but also following profound activity perturbations . Despite a transient GABABR-mediated blockade of population firing rate with a simultaneous increase in firing synchrony , population activity was precisely restored to a ‘set point’ level after a period of 2 days . Interestingly , restoration of firing synchrony displays a faster kinetics than rebound of mean rates . In contrast to the observed firing macro-stability , single-unit behavior appears to be extremely dynamic . Only 23% of units displayed stable firing rates without perturbation , and this fraction was reduced to 11% following the perturbation . A complementary method , based on long-term recordings of somatic Ca2+ dynamics , reflecting firing rates , in identified neurons showed that 46% of neurons returned to their baseline firing rates ( Figure 3 ) . Thus , two independent methodologies suggest that the majority of neurons display unstable firing rates even at extended timescales , confirming scale-invariant rate dynamics observed in a previous study ( Gal et al . , 2010 ) . Thus , stability , as well as the compensatory feedback response , is greater at the population compared to the single-unit level . Moreover , quantal excitatory synaptic strength and intrinsic excitability were significantly affected by the perturbation and , thus , differ between the baseline and rebound periods . These results highlight the idea that similar network properties may arise from multiple configurations of individual components ( Prinz et al . , 2004; Marder and Goaillard , 2006 ) . Most importantly , the observed differences in the precision of homeostatic regulation at different spatial scales suggest that population firing homeostasis is more than a sum of single-neuron adaptive responses , implying the existence of network-wide regulation rules ( Maffei and Fontanini , 2009 ) . What is the trigger of compensatory synaptic responses ? A variety of synaptic compensation mechanisms have been found following pharmacological or genetic perturbations in hippocampal and cortical neuronal cultures ( Turrigiano et al . , 1998; Burrone et al . , 2002; Thiagarajan et al . , 2002 , 2005; Branco et al . , 2008; Jakawich et al . , 2010 ) , following sensory deprivation ( Maffei et al . , 2004; Hengen et al . , 2013; Keck et al . , 2013 ) or in a more physiological context of sleep ( Vyazovskiy et al . , 2008; Lanté et al . , 2011 ) . Interestingly , synaptic homeostasis hypothesis proposed by Tononi and colleagues ( Tononi and Cirelli , 2003 , 2014 ) states that population firing synchrony , a hallmark of slow-wave sleep , contributes directly to homeostatic synaptic scaling . To examine if changes in population synchrony per se can trigger adaptive responses , we compared two perturbations producing similar ( inhibitory ) effects on the mean firing rate , while differentially affecting firing synchrony . For this purpose , we used baclofen that increases firing synchrony ( Figure 4 ) and CNQX that induces firing desynchronization ( Figure 9H , I ) . Both perturbations trigger the same types of synaptic adaptation at a comparable timescale: increase in the amplitude and the frequency of mEPSCs and in release probability ( Figures 6 and 9F , G and ( Thiagarajan et al . , 2002; Jakawich et al . , 2010 ) ) . Based on these results , we may conclude that ( 1 ) firing rates and patterns are independently regulated; ( 2 ) homeostatic systems generally sense a drop in spiking rates to induce an adaptive increase in excitatory synaptic transmission . Thus , our results in cultured neural networks don't support a causal relationship between firing synchrony and homeostatic synaptic response , suggesting that other factors , occurring during slow-wave sleep , may play a role in down-scaling of synapses in behaving animals . Why network firing rates and patterns were not compensated following the AMPAR blockade requires future investigation . On the one hand , complete silencing of excitatory drive might exceed the capacity of the homeostatic system to compensate . On the other hand , the drop in the firing rate following baclofen application was even more pronounced; nevertheless , firing rate homeostasis was achieved . Similarly , blockade of GABA uptake via GAT-1 transporter resulted in a transient blockade of firing rates that were precisely compensated 2 days following the perturbation ( Figure 2—figure supplement 3 ) . We can exclude the necessity of population bursts for induction of a proper profile of adaptive response since other treatments , such as ACh exposure , induce a reduction in synchrony that can be successfully compensated ( Kaufman et al . , 2012 ) . Thus , our results indicate that adaptations at the level of intrinsic neuronal excitability and inhibitory drive are not sufficient to compensate the blockade of AMPAR-mediated excitatory drive . It is conceivable that AMPAR blockade might perturb the relationship between the firing rate and the intracellular Ca2+ concentration , causing a failure in the regulatory system ( O'Leary et al . , 2014 ) . It is generally assumed that experience-dependent modifications of synaptic strength or intrinsic excitability are associated with functional changes in the network performance . Nevertheless , under some circumstances , synaptic and intrinsic modifications may have little effect on network functional properties ( Prinz et al . , 2004; Thirumalai et al . , 2006 ) . For example , networks utilizing uniform and multiplicative forms of postsynaptic scaling as a main compensatory response to the perturbation may preserve learning rules and information content transferred between neurons ( Turrigiano et al . , 1998; Turrigiano and Nelson , 2004 ) . While changes in neuromodulation via GABABRs caused a profound increase in synapse release probability , intrinsic excitability , mEPSC frequency and amplitude , spontaneous spiking activity of the network returned to the baseline level during 2 days in the constant presence of the perturbation . However , the sensitivity of synaptic population to bursts was profoundly modified: while acute GABABR activation by baclofen shifts synapses towards high-pass filters ( Figure 1A–C ) , chronic baclofen application results in complete abolishment of short-term synaptic plasticity ( Figure 8D , E ) , potentially reducing discrimination of input patterns by synaptic mechanisms . The reduction in the selectivity of synapses to afferent input may represent a trade-off between population firing stability and synaptic metaplasticity ( Abraham and Bear , 1996; Thiagarajan et al . , 2007 ) . Thus , robust homeostatic control of ongoing population dynamics may coexist with unstable short-term synaptic plasticity . In summary , our results suggest that invariant population mean rate and temporospatial coherence of spontaneous spiking can emerge from highly diverse combinations of synaptic strength and intrinsic neuronal properties . The observed micro-instability of individual neurons was truly intrinsic , taking place in a highly-controlled environment , irrespective of changes in experience , behavioral states , and interactions with higher-order supervising circuits . While firing macro-stability is robustly and accurately maintained by homeostatic control systems in the face of perturbations and uncertainties , the ability of synapses to discriminate input patterns was sacrificed . Thus , impairments of short-term synaptic plasticity and working memory functions , characterizing initial phases of numerous brain disorders , may be the tradeoff resulting from the system's efforts to maintain phenotypic stability of spontaneous population firing patterns . It remains to be seen whether hippocampal short-term synaptic plasticity can be maintained without sacrificing critical parameters of population dynamics . Primary cultures of CA3–CA1 hippocampal neurons were prepared from newborn BALB/c mice on postnatal days 0–2 , as described ( Slutsky et al . , 2004 ) . The experiments were performed in 15–22 DIV cultures . All animal experiments were approved by the Tel Aviv University Committee on Animal Care . Cultures were plated on MEA plates containing 59 TiN recording and one internal reference electrodes ( Multi Channel Systems [MCS] , Germany ) . Electrodes are 30 µm in diameter and spaced 500 µm apart . Data were acquired using a MEA1060-Inv-BC-Standard amplifier ( MCS ) with frequency limits of 5000 Hz and a sampling rate of 10 kHz per electrode mounted on an Olympus IX71 inverted microscope . Recordings were carried out under constant 37°C and 5% CO2 conditions , identical to incubator conditions . Raw data were filtered , offline , at 200 Hz using a Butterworth high-pass filter . Spikes cutouts were then detected , offline , using MC Rack software ( MCS ) based on a fixed threshold set to between 4–5 standard deviations from noise levels . In order to ensure the veracity of the detected spikes , TTX was added at the end of some experiments , resulting in an immediate loss of all spiking activity . Only the first 20 min of each hour were used for all analyses in order to reduce computation time . We showed that 20 min can reliably represent the MFR of a full-hour by comparing the MFR of 792 20 min segments to the MFR of the full hours represented by those segments . We found that only 9 . 6% of segments were more than 10% different from their full hour and none were more than 20% different ( Figure 2—figure supplement 4A ) . Furthermore , we calculated the coefficient of variation ( CV ) for 104 units over the course of 8 consecutive hr using different bin sizes . The CV represents the variability of MFRs within each unit over a given time-frame . We found that small bin sizes resulted in significantly higher CVs compared to 1 hr bins while bins larger than 10 min were not significantly different from bins of a full hour ( CV = 121 . 3% ± 8 . 4 , 53 . 4% ± 4 . 1 , 25 . 1% ± 2 , 19 . 2% ± 1 . 4 , 14 . 8% ± 1 . 1 , 13 . 5% ± 1 , 11 . 60% ± 0 . 8 , for bins of 10 s , 1 , 5 , 10 , 20 , 30 , and 60 min , respectively; Figure 2—figure supplement 4B ) . This is an indication that , while there is always some intrinsic variability in per unit MFR , 20 min is an accurate representation of a full hour . Spike cutouts were then transferred to Offline Sorter ( Plexon Inc . , Dallas , Texas , USA ) for spike sorting . Spikes were plotted in 2-D or 3-D principal component ( PC ) space and unit clusters were semi-automatically detected using K-means clustering algorithm followed by template sorting . Clusters were then manually inspected to insure stability throughout experiment . Only clusters that fulfilled the following requirements were considered units and used for analysis: ( 1 ) there was no spiking during the absolute refractory period . ( 2 ) The clusters were well defined relative to other clusters from the same electrode throughout the entire experiment . ( 3 ) There were no sudden jumps in cluster location on PC axes . ( 4 ) The cluster is not centered around the origin of the PC axes . ( 5 ) Auto-correlation histograms showed a distinct peak at t ≠ 0 . A cluster was classified as ‘multiunit’ if the autocorrelogram lacked a clear refractory period . All analysis was performed using custom-written scripts in MATLAB ( Mathworks , Natick , Massachusetts , USA ) . Network mean firing rates were calculated by averaging the mean firing rates of all units for a given time-point . For comparison of single unit parameters , we checked whether the difference in values between two time-points for a given unit significantly deviated from a null ( zero centered ) distribution using a bootstrapping method ( see ‘bootstrap . m’ in Source code 1 ) . The two time segments to be compared were divided into 1 min bins which were then randomly shuffled 10 , 000 times into two groups . The differences between the means of the two randomly shuffled groups produced a null-distribution . The real difference was significant if it fell outside of the 95% confidence interval of the null-distribution . The length of time segments , size of bins , and number of iterations had no effect on bootstrapping results ( Figure 2—figure supplement 4C–E ) . For population-burst detection and analysis , we used the following algorithm ( see ‘network_burst_analysis . m’ in Source code 1 ) . We start with a set of spike time sequences acquired from a group of single units . The sequences are then projected onto a single timeline to produce an ordered time sequence . Θ=∪α , i{tiα} Here , tiα is the ith spike inside the αth unit sequence . Next , we estimate the local spike density f^ ( t ) . To reduce binning issues , we use Parzen kernel density estimation ( KDE ) approach ( Parzen , 1962 ) :f^ ( t ) = ∑iρσ ( t−ti ) , where we choose a Gaussian kernel . ρσ ( t ) = ( 2πσ2 ) −1/2exp ( −t2/2σ2 ) . The timeline is discretized with sampling intervals of 50 ms; for the kernel , we set σ2=0 . 5 . The resulting spike density function peaks strongly around potential burst locations , where the spike density is especially high . We use this fact as a cue to detect network bursts—every peak higher than a certain threshold is considered a candidate burst . Spike density threshold is estimated as the 95% percentile for a random Poisson process producing the same quantity of spikes . The timelines for each unit and thus for the entire network are divided into hourly intervals; we assume that throughout each interval the physiological state of the network and thus the spike statistics do not change . Thus , a different threshold is calculated for each hourly interval . Since we are looking for collective behavior modes , as a concluding filtering stage , we eliminate candidates generated by a small number of units . A unit is considered active if during the candidate burst it contributed at least a predefined number of spikes ( usually 2–3 ) . Then , for each hourly interval , a histogram of number of active units in a burst is calculated . The threshold for candidate burst elimination is set at the predefined percentile . In practice , choosing any value between 20–30% produces very similar results . For detection of bursts at the single-unit level , bursts were defined as 3 or more spikes at a minimum of 50 Hz ( see ‘single_unit_bursts . m’ in Source code 1 ) . We analyzed data for all the combinations of parameters ranging from a minimum of 2 , 3 , and 4 spikes at frequencies of ≥20 , 50 , and 100 Hz and found no difference in the qualitative results showing that this analysis is robust over a wide range of definitions ( Figure 4—figure supplement 2 ) . Hippocampal cultures were infected by AAV2/1-Syn-GCaMP6f at 5–8 day in vitro . The experiments were performed 8–12 days post-infection when expression reached stable levels . Time-lapse images were acquired at 37°C and CO2 controlled environment , using Nikon Eclipse Ti microscope with air 20× objective lens ( NA = 0 . 45 ) , controlled via iQ software ( Andor , UK ) . Time-lapse images were collected at 20 Hz ( 1920 × 1080 pixels; 624 × 351 μm ) using Neo sCMOS camera ( Andor ) and binned 4 × 4 pixels . The light source was AMH-2000 metal-halide lamp . The excitation wavelength was 485/20 nm , the emission band-pass filter was 525/50 nm . Imaging parameters were optimized to minimize photobleaching and phototoxicity , while preserving sufficient signal-to-noise ratio and temporal resolution . Light intensity was kept constant in all measurements during the 2 days of the experiment . Images were taken each for 2 min with 5 min intervals for 10 cycles , thus sampling 20 min of activity over a 65 min period . We saw no discernible changes in neurons' morphologies or in mean Ca2+-transient amplitudes within or across sessions under our imaging conditions ( Figure 3—figure supplement 1D , E ) . For AP-evoked signals ( Figure 3—figure supplement 1A , B ) , synaptic blockers ( 50 μM AP5 and 20 μM CNQX ) were used to block recurrent activity . Signal intensity was measured using imageJ software by selecting ROIs over the soma of neurons expressing GCaMP6f while avoiding the nucleus . The fluorescence time course of each cell was measured by averaging all pixels within the ROI . Only cells that were in the same estimated focal plane and visually separated from neighboring cells and surrounding neuropil were used . For long-term GCaMP6f imaging , baseline fluorescence images of multiple sessions were inspected manually , and only the cells that could be clearly identified in all imaged sessions in the focal plane were included in the analysis . ΔF/F was quantified for each cell as change in fluorescence divided by baseline fluorescence measured 1 s before the spontaneous signal or stimulation . The sum of amplitudes of all the events was divided by the recording time to estimate the rate of Ca2+ transients . Experiments were performed at room temperature in a recording chamber on the stage of FV300 inverted confocal microscope ( Olympus , Japan ) . Extracellular Tyrode solution contained ( in mM ) : NaCl , 145; KCl , 3; glucose , 15; HEPES , 10; MgCl2 , 1 . 2; CaCl2 , 1 . 2; pH adjusted to 7 . 4 with NaOH . Whole-cell patches were recorded using the following intracellular solution ( in mM ) : Cs-MeSO3 , 120; HEPES , 10; NaCl , 10; CaCl2 , 0 . 5; Mg2+–ATP , 2; Na3GTP , 0 . 3; EGTA , 10 for mEPSC and E/I experiments; pH adjusted to 7 . 25 with NaOH . Serial resistance was not compensated . For mEPSCs recordings , Tetrodotoxin ( TTX; 1 μM ) , amino-phosphonopentanoate ( AP-5; 50 μM ) , and gabazine ( 30 μM ) were added to the Tyrode solution . For measurement of excitation/inhibition ( E/I ) balance , sEPSCs and sIPSCs were isolated in the same cell based on reversal potentials of GABAAR-mediated and AMPAR-mediated currents , respectively . When corrected for the liquid junction potential , the reversal potential for E ( VE ) was close to 10 mV and VI was close to −65 mV , close to the predicted value for the intracellular and extracellular solutions used in the present study . For intrinsic excitability measurements , the following intracellular solution was used ( in mM ) : K-gluconate 120; KCl 10; HEPESs 10; Na-phosphocreatine 10; ATP-Na2 4; GTP-Na 0 . 3; MgCl2 0 . 5 . Recordings were done in the presence of synaptic blockers ( in µM: 25 DNQX , 50 APV , and 10 bicuculline ) . Frequency vs current intensity curves were plotted by measuring the average rate of action potentials in current-clamp during 500-ms long depolarizing steps of increasing intensity; a small DC current was injected to maintain membrane potential at −65 mV in between depolarizations . Input resistance ( Rin ) was measured by calculating the slope of the voltage change in response to increasing current injections from −80 pA to −20 pA in 20 pA increments . Access resistance was between 5–15 MΩ . Neurons were excluded from the analysis if RMP was >−55 mV , serial resistance was >15 MΩ , and Rin was <80 MΩ or if any of these parameters changed by >20% during the recording . Signals were recorded using a MultiClamp 700B amplifier , digitized by DigiData1440A ( Molecular Devices , Sunnyvale , California , USA ) at 10 kHz , and filtered at 2 kHz . Electrophysiological data were analyzed using MiniAnalysis ( Synaptosoft , Decatur , Georgia , USA ) for mEPSCs and in pClamp10 ( Molecular Devices ) for sEPSCs and sIPSCs . The integrated conductances GE and GI were calculated according to the following equations GE=∫0tsEPSCt ( VM−VErev ) and GI=∫0tsIPSCt ( VM−VIrev ) . Activity-dependent FM1-43 styryl dye was used to estimate basal synaptic vesicle recycling and exocytosis . Action potentials were elicited by passing 50 mA constant current for 1 ms ( ∼50% above the threshold for eliciting action potential ) through two platinum wires , separated by ∼7 mm , and close to the surface of the coverslip . The extracellular Tyrode solution contained non-selective antagonist of ionotropic glutamate receptors ( kynurenic acid , 0 . 5 mM ) to block recurrent neuronal activity . Synaptic vesicles were loaded with 10 μM FM1-43 . FM loading and unloading were done using protocols described previously ( Abramov et al . , 2009 ) . The fluorescence of individual synapses was determined from the difference between images obtained after staining and after destaining ( ΔF ) . To estimate vesicle recycling/release during low frequency stimulation , we quantified: ( i ) ΔF signal for staining by 30 action potentials at a rate of 1 Hz stimulation; ( ii ) FM destaining rate during 1 Hz stimulation following staining of boutons by maximal stimulation . For FM- ( + ) puncta detection , ΔF images have been analyzed ( only the puncta exhibiting ≥90% destaining were subjected to analysis ) . Detection of signals has been done as described ( Abramov et al . , 2009 ) . FM1-43 and Advasep-7 were purchased from Biotium ( Hayward , California , USA ) , TTX from Alamone Labs ( Israel ) , baclofen , CNQX , and AP-5 from Tocris ( UK ) . Error bars shown in the figures represent standard error of the mean ( SEM ) . The number of experiments is defined by n . Student's paired t-tests were used in all the experiments where the effect of baclofen was tested in the same cell/synapse ( p < 0 . 05; p < 0 . 01; *p < 0 . 001 ) , unless otherwise noted . Unpaired t-tests were used to compare different populations of synapses . One-way ANOVA Kruskal–Wallis non-parametric test was used to compare several populations of synapses . Nonparametric Spearman's test has been used for correlation analysis . For comparison of mEPSC amplitude or frequency under different conditions , 200 mEPSCs were randomly selected for each cell and pooled for each condition . A Kolmogorov–Smirnov ( K–S ) test was used to compute differences in distributions across the pooled datasets .	The human brain contains more than 80 billion neurons , which are organised into extensive networks . Changes in the strength of connections between neurons are thought to underlie learning and memory: neuronal networks must therefore be sufficiently stable to allow existing memories to be stored , while remaining flexible enough to enable the brain to form new memories . Evidence suggests that the stability of neuronal networks is maintained by a process called homeostasis . If properties of the network—such as the average firing rate of all the neurons—deviate from a set point , changes occur to return the network the original set point . However , much less is known about the effects of homeostasis at the level of individual neurons within networks: do their firing rates also remain stable over time ? Slomowitz , Styr et al . have now addressed this question by recording the activity of neuronal networks grown on an array of electrodes . Applying a drug that inhibits neuronal firing caused the average firing rate of the networks to decrease initially , as expected . However , after 2 days , homeostasis had restored the average firing rate to its original value , despite the continued presence of the drug . By contrast , the individual neurons within the networks behaved differently: on day 2 almost 90% of neurons had a firing rate that was different from their original firing rate . Similar behavior was seen when Slomowitz , Styr et al . studied the degree of synchronization between neurons as they fire: the average value for the network returned to its original value , but this did not happen at the level of individual neurons . Surprisingly , however , the ability of the network to undergo short-lived changes in average strength of the connections between neurons—which is thought to support short-term memory—was not subject to homeostasis . This suggests that the loss of short-term memory that occurs in many brain diseases may be an unfortunate consequence of the efforts of neuronal networks to keep their average responses stable .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "neuroscience" ]	2015	Interplay between population firing stability and single neuron dynamics in hippocampal networks
Drosophila male courtship is controlled by the male-specific products of the fruitless ( fruM ) gene and its expressing neuronal circuitry . fruM is considered a master gene that controls all aspects of male courtship . By temporally and spatially manipulating fruM expression , we found that fruM is required during a critical developmental period for innate courtship toward females , while its function during adulthood is involved in inhibiting male–male courtship . By altering or eliminating fruM expression , we generated males that are innately heterosexual , homosexual , bisexual , or without innate courtship but could acquire such behavior in an experience-dependent manner . These findings show that fruM is not absolutely necessary for courtship but is critical during development to build a sex circuitry with reduced flexibility and enhanced efficiency , and provide a new view about how fruM tunes functional flexibility of a sex circuitry instead of switching on its function as conventionally viewed . Drosophila male courtship is one of the best understood innate behaviors in terms of genetic and neuronal mechanisms ( Dickson , 2008; Yamamoto and Koganezawa , 2013 ) . It has been well established that the fruitless ( fru ) gene and its expressing neurons control most aspects of such innate behavior ( Ito et al . , 1996; Manoli et al . , 2005; Ryner et al . , 1996; Stockinger et al . , 2005 ) . The male-specific products of the P1 promoter of the fru gene ( fruM ) are expressed in ~2000 neurons , which are inter-connected to form a sex circuitry from sensory neurons to motor neurons ( Cachero et al . , 2010; Lee et al . , 2000; Manoli et al . , 2005; Stockinger et al . , 2005; Usui-Aoki et al . , 2000; Yu et al . , 2010 ) . fruM function is necessary for the innate courtship behavior and sufficient for at least some aspects of courtship ( Baker et al . , 2001; Demir and Dickson , 2005; Manoli et al . , 2005 ) . Thus , the study of fruM function in controlling male courtship serves as an ideal model to understand how innate complex behaviors are built into the nervous system by regulatory genes ( Baker et al . , 2001 ) . Although fruM serves as a master gene controlling Drosophila male courtship , we recently found that males without fruM function , although did not court if raised in isolation , were able to acquire at least some courtship behaviors if raised in groups ( Pan and Baker , 2014 ) . Such fruM-independent but experience-dependent courtship acquisition requires another gene in the sex determination pathway , the doublesex ( dsx ) gene ( Pan and Baker , 2014 ) . dsx encodes male- and female-specific DSX proteins ( DSXM and DSXF , respectively ) ( Burtis and Baker , 1989 ) , and DSXM is expressed in ~700 neurons in the central nervous system ( CNS ) , the majority of which also express fruM ( Rideout et al . , 2010; Robinett et al . , 2010 ) . It has been found that the fruM and dsxM co-expressing neurons are required for courtship in the absence of fruM function ( Pan and Baker , 2014 ) . Thus fruM-expressing neurons , especially those co-expressing dsxM , control the expression of courtship behaviors even in the absence of FRUM function . Indeed , although the gross neuroanatomical features of the fruM-expressing circuitry are largely unaffected by the loss of fruM ( Manoli et al . , 2005; Stockinger et al . , 2005 ) , detailed analysis revealed morphological changes of many fruM-expressing neurons ( Cachero et al . , 2010; Kimura et al . , 2005; Kimura et al . , 2008; Mellert et al . , 2010 ) . Recent studies further reveal that FRUM specifies neuronal development by recruiting chromatin factors and changing chromatin states , and also by turning on and off the activity of the transcription repressor complex ( Ito et al . , 2012; Ito et al . , 2016; Sato et al . , 2019a; Sato et al . , 2019b; Sato and Yamamoto , 2020 ) . That FRUM functions as a transcription factor to specify development and/or physiological roles of certain fruM-expressing neurons , and perhaps the interconnection of different fruM-expressing neurons to form a sex circuitry raises important questions regarding when fruM functions and how it contributes to the sex circuitry ( e . g . , how the sex circuitry functions differently with different levels of FRUM ) , especially in the background that fruM is not absolutely necessary for male courtship ( Pan and Baker , 2014 ) . To at least partially answer these questions , we temporally or spatially knocked down fruM expression and compared courtship behavior in these males with that in wild-type males or fruM null males and revealed crucial roles of fruM during a narrow developmental window for the innate courtship toward females . We also found that the sex circuitry with different fruM expression has distinct function such that males could be innately heterosexual , homosexual , bisexual , or without innate courtship but could acquire such behavior in an experience-dependent manner . Thus , fruM tunes functional flexibility of the sex circuitry instead of switching on its function as conventionally viewed . To specifically knockdown fruM expression , we used a microRNA targeting fruM ( UAS-fruMi at attp2 or attp40 ) and a scrambled version as a control ( UAS-fruMiScr at attp2 ) as previously used ( Chen et al . , 2017; Meissner et al . , 2016 ) . Driving the fruM microRNA by fruGAL4 specifically knocked down mRNA of fruM , but not the common form of fru ( Figure 1—figure supplement 1A–C ) . We firstly tested male courtship without food in the behavioral chamber . Knocking down fruM in all the fruGAL4-labeled neurons eliminated male courtship toward females ( courtship index [CI] , which is the percentage of observational time that males displayed courtship , is nearly 0 ) ( Figure 1A ) , consistent with previous findings that fruM is required for innate male–female courtship ( Demir and Dickson , 2005; Pan and Baker , 2014 ) . As fruGAL4 drives expression throughout development and adulthood ( Figure 1—figure supplement 1D–K ) , we set out to use a temperature-dependent tub-GAL80ts transgene to restrict UAS-fruMi expression ( e . g . , at 30°C ) at different developmental stages . We raised tub-GAL80 ts/+; fruGAL4/UAS-fruMi flies at 18°C ( permissive for GAL80ts that inhibits GAL4 activity ) and transferred these flies to fresh food vials every 2 days . In this way , we generated tub-GAL80 ts/+; fruGAL4/UAS-fruMi flies at nine different stages from embryos to adults and incubated all flies at 30°C to allow fruM knockdown for 2 days , then placed all flies back to 18°C until courtship test ( Figure 1B ) . We found that males with fruM knocked down at stage 5 for 2 days , matching the pupation phase , rarely courted ( CI < 10% ) , and none successfully mated , while males with fruM knocked down near this period ( stages 4 and 6 ) showed a partial courtship or mating deficit , and males with fruM knocked down at earlier or later stages showed strong courtship toward females and successful mating ( Figure 1C , D ) . To validate efficiency of fruM knockdown during specific developmental periods , we generated an antibody against FruM as well as a V5 knock-in into the fru gene ( fruV5 ) to visualize FruM expression . Both tools successfully labeled male-specific FruM proteins ( Figure 1—figure supplement 2 ) , and there is almost perfect overlap of the two markers ( Figure 1E , G ) . Note that the anti-FruM antibody also labeled several pairs of false-positive neurons in both wild-type and fruM mutants ( Figure 1—figure supplement 2 ) , indicating the strong but not perfect specificity of this antibody ( Figure 1—figure supplement 2B–D ) . To test whether 2 day heat shock at 30°C is sufficient to knockdown fruM expression , we dissected brains of tub-GAL80ts/UAS-fruMi; fruGAL4/fruV5 males immediately after 2 day heat shock at stage 5 or 7 and found that anti-V5 and anti-FruM signals were both dramatically decreased , such that only a small fraction of neurons could be weakly labeled; in contrast , control males with the same age but raised at 18°C have regular anti-V5 and anti-FruM signals ( Figure 1E–H ) . These results indicate that induction of fruM microRNA during development for 2 days could effectively knockdown fruM expression . As induced fruM microRNA may not be degraded immediately and has longer effect , we further tested to how much extent such knockdown effect may last . Thus , we dissected brains of adult tub-GAL80ts/UAS-fruMi; fruGAL4/fruV5 males that have been heat shocked for 2 days at different developmental stages ( from stages 1 to 9 ) and found that males that have been heat shocked at earlier stages ( from stages 1 to 5 ) still have strong FruM expression ( Figure 1—figure supplement 3A–F ) , suggesting effective restore of FruM expression after transferring at 18°C . However , males that have been heat shocked at later stages ( stages 6–9 ) have obviously reduced FruM expression ( Figure 1—figure supplement 3G–J ) , suggesting a partial restore of FruM expression , probably due to prolonged fruM microRNA effect . Note that knocking down fruM expression at these later stages has partial ( stage 6 ) or no effect ( other stages ) on male courtship , comparing with fruM knockdown at stage 5 that almost eliminated male courtship . Together these results indicate a critical developmental period during pupation ( from late larvae at stage 5 to early pupas at stage 6 ) where fruM is required for adult male courtship toward females . We reasoned that fruM function during pupation may be involved in neuronal development for circuit construction . Thus we set out to examine the morphology of a subset of fruM-positive gustatory receptor neurons ( GRNs ) innervating the ventral nerve cord ( VNC ) in tub-GAL80ts/UAS-mCD8GFP; fruGAL4/UAS-fruMi males that have been heat shocked for 2 days in different developmental stages , as it has been found that fruM is required for the male-specific midline crossing of these GRNs ( Mellert et al . , 2010 ) . We found that these GRNs were only labeled in males that have been heat shocked after stage 4 , probably because these GRNs were developed after stage 4 ( Figure 1—figure supplement 4A–C ) , consistent with a previous study ( Mellert et al . , 2012 ) . Interestingly , we found that all males heat shocked at stage 5 for 2 days showed defect of midline crossing in these GRNs , and 60% of males heat shocked at stage 6 for 2 days showed defect of midline crossing , while all males heat shocked after stage 6 showed regular midline crossing ( Figure 1—figure supplement 4C , D ) . Males heat shocked for 4 days during adulthood also have regular midline crossing ( Figure 1—figure supplement 4C , D ) . These results clearly showed a critical developmental period during pupation where FruM functions to ensure regular development of GRNs and enable innate male courtship toward females . As knocking down fruM at stage 9 when flies were newly eclosed did not affect male courtship ( CI > 80% ) and mating success ( Figure 1C , D ) , we further tested the role of fruM in adulthood using different approaches . We set out to express the female-specific transformer ( traF ) gene ( Baker and Ridge , 1980; McKeown et al . , 1988 ) to feminize all fruGAL4 labeled neurons , in addition to the above fruM RNAi experiments . We express UAS-traF or UAS-fruMi in all the fruGAL4-labeled neurons specifically during adulthood for 4 days before test ( see procedure above each figure ) for single-pair male–female , male–male , and male chaining ( in groups of eight males ) behaviors . We found that overexpression of traF in all fruGAL4 labeled neurons during adulthood for 4 days did not affect male–female courtship ( Figure 2A ) , but slightly increased male–male ( Figure 2B ) and male chaining behaviors ( Figure 2C ) . Furthermore , knocking down fruM in all fruGAL4-labeled neurons during adulthood for 4 days did not affect male–female ( Figure 2A ) or male–male courtship ( Figure 2B ) , but significantly increased male chaining behaviors ( Figure 2C ) . We also checked FruM expression in males that have been heat shocked for 4 days during adulthood using anti-V5 and anti-FruM antibodies , and found that FruM expression was almost eliminated , while control males have regular FruM expression ( Figure 2D , E ) . These results indicate that although fruM function during adulthood is dispensable for female-directed courtship , it is involved in inhibiting male–male courtship behaviors . Thus , FruM has distinct functions during development and adulthood for male courtship behaviors . The above results indicate crucial roles of fruM during pupation for female-directed courtship in adult males . We reasoned that fruM function during pupation may specify the construction of courtship circuitry and affects female-directed courtship as well as other courtship behaviors , especially given our previous findings that fruM null males were able to acquire courtship behavior after group-housing ( Pan and Baker , 2014 ) . Thus , we set out to compare courtship behaviors in males with distinct fruM expression modes , such as with wild-type fruM , systemic low level of fruM , spatially low level of fruM , or completely without fruM function . We tested one-time single-pair male–female and male–male courtship ( single housed before test ) as well as male chaining in groups of eight males over 3 days on food for better comparison of these courtship assays , as courtship by fruM null males largely depends on food presence ( Pan and Baker , 2014 ) . We found that male–male courtship in fruM knocked down males is higher if tested on food , consistent with a courtship promoting role by food ( Grosjean et al . , 2011; Pan and Baker , 2014 ) , while courtship in wild-type males on food or without food is not changed in our assays ( Figure 3—figure supplement 1 ) . We found that wild-type males performed intensive courtship behavior toward virgin females ( CI > 80% ) and rarely courted males ( CI ~0 ) ( Figure 3A ) . Furthermore , these control males did not show any chaining behavior after grouping from 3 hr to 3 days ( ChI = 0 ) ( Figure 3B ) . In striking contrast , fruM null mutant males rarely courted either females or males ( Figure 3C , Figure 3—figure supplement 2A , C , and E ) ; however , these males developed intensive chaining behavior after grouping for 1–3 days ( Figure 3D , Figure 3—figure supplement 2B , D , and F ) . These observations replicated previous findings that there exists a fruM-independent experience and dsxM-dependent courtship pathway ( Pan and Baker , 2014; Figure 3E ) . To compare behavioral differences by fruM null males and fruM RNAi knocked down males that have systemic low level of fruM , we firstly quantified to how much extent the microRNA against fruM ( UAS-fruMi at attp40 ) worked . We found that the fruM mRNA level was reduced to ~40% of that in control males ( Figure 3F ) . Interestingly , while males with fruM knocked down in all fruM neurons rarely courted females ( CI ~5% , Figure 3G ) , they displayed a high level of male–male courtship behavior ( CI > 50% , Figure 3G ) and constantly high level of male chaining ( Figure 3H ) , dramatically different from fruM null males . These results reveal distinct roles of low fruM ( RNAi ) and high fruM ( wild-type ) in regulating male–male and male–female courtship ( Figure 3I ) . To further reveal the role of fruM expression patterns in determining male courtship modes , we tried to spatially knockdown fruM expression using a simple way: fruM in brain and fruM outside brain . We used Otd-Flp expressing FLP specifically in the central brain ( Asahina et al . , 2014 ) to divide fruGAL4 expression ( Figure 3—figure supplement 3A ) into two parts: fruM- and Otd-positive neurons ( specifically in brain ) in Otd-Flp/UAS-mCD8GFP; fruGAL4/tub>GAL80> males ( Figure 3—figure supplement 3B ) and fruM-positive but Otd-negative neurons ( theoretically outside brain , but still with few in brain ) in Otd-Flp/UAS-mCD8GFP; fruGAL4/tub>stop>GAL80 males ( Figure 3—figure supplement 3C ) . We also checked GFP expression in peripheral nervous system in these males and found a few GFP-positive cells in antennae and forelegs in Otd-Flp/UAS-mCD8GFP; fruGAL4/+ males , but rare expression in Otd-Flp/UAS-mCD8GFP; fruGAL4/tub>stop>GAL80 or Otd-Flp/UAS-mCD8GFP; fruGAL4/tub>GAL80> males ( Figure 3—figure supplement 3D , E ) . Thus , we successfully divided fruGAL4 expression into two categories: one with GAL4 expressed in fru+Otd+ neurons in brain and the other with GAL4 expressed in fru+Otd− neurons outside brain . We then used the above intersectional strategy to specifically knockdown fruM expression in or outside brain . To validate such strategy , we used anti-V5 to visualize FruM expression in these males ( together with fruV5 ) and found effective , if not perfect , knockdown of FruM expression spatially ( Figure 3—figure supplement 3F–I ) . We found that males with fruM knocked down specifically in brain had a reduced level of courtship toward females ( CI = 56 . 61 ± 5 . 86% ) , but their sexual orientation was not changed as they courted males in a much lower level ( CI = 15 . 94 ± 3 . 26% , Figure 3J ) . Furthermore , males with fruM knocked down in brain showed low male chaining behavior initially but increasing levels of chaining behavior over 1–3 days ( ChI [3 hr] = 9 . 35 ± 5 . 40% , ChI[3d] = 68 . 82 ± 5 . 53% , Figure 3K ) . Knocking down fruM only in a subset of male-specific P1 neurons driven by P1-splitGAL4 in the brain that are important for courtship initiation ( Clowney et al . , 2015; Kallman et al . , 2015; Kimura et al . , 2008; Pan et al . , 2012; Wu et al . , 2019 ) failed to decrease male–female courtship or induce male chaining behavior ( Figure 3—figure supplement 4A , B ) . These results indicate that fruM function in brain promotes male–female courtship and inhibits acquisition or progression of the experience-dependent chaining behavior ( Figure 3L ) . In contrast , males with fruM knocked down outside brain showed equally intensive male–female and male–male courtship ( CI [male–female] = 85 . 62 ± 1 . 42% , CI [male–male] = 82 . 89 ± 2 . 76% , Figure 3M ) , indicating an inhibitory role of fruM in these neurons for male–male courtship ( Figure 3O ) . These males performed a high level of male chaining behavior initially ( ChI [3 hr] = 92 . 90 ± 3 . 08% ) , but decreased levels of chaining behavior over 1–3 days ( ChI [3d] = 20 . 01 ± 3 . 75% , Figure 3N ) , consistent with the above finding that fruM function in the brain which is intact in these males inhibits acquisition or progression of male chaining behavior ( Figure 3L ) . Knocking down fruM in a subset of gustatory receptor neurons expressing ppk23 that respond to female-specific pheromones ( Lu et al . , 2012; Thistle et al . , 2012; Toda et al . , 2012 ) mildly enhanced male–male courtship but did not induce male chaining behavior ( Figure 3—figure supplement 4C , D ) , suggesting a moderate role of fruM in these neurons for inhibiting male–male courtship , although its roles in these neurons during development or adulthood were not yet discriminated . Taken together , the above results demonstrate distinct roles of fruM expression during a critical developmental period for the manifestation of courtship behaviors and adulthood for inhibiting male–male courtship ( Figure 4A ) , and further reveal that different fruM expression levels and patterns determine courtship modes , indicative of functional flexibility of the fruM-expressing sex circuitry tuned by fruM function ( Figure 4B ) . Previous findings show that fruM expression commences at the wandering third-instar larval stage , peaks at the pupal stage , and thereafter declines but does not disappear after eclosion ( Lee et al . , 2000 ) , which suggests that fruM may function mainly during development for adult courtship behavior despite of no direct evidence . Here we temporally knocked down fruM expression in different developmental stages for 2 days and found that males with fruM knocked down during pupation rarely courted , while males with fruM knocked down during adulthood courted normally toward females . This is the first direct evidence that fruM is required during development but not adulthood for female-directed courtship behavior . A caveat of these experiments is that while fruM expression is effectively knocked down upon 2 day induction of fruM microRNA , it is not restored acutely after transferring to permissive temperature , although it is restored in adulthood if induction of fruM microRNA was performed at earlier stages ( stages 1–5 ) . Such a caveat does not compromise the above conclusion as knocking down fruM during pupation ( stage 5 ) almost eliminated male courtship while knocking down at later stages have minor or no effect on male courtship . Consistent with these behavioral findings , knocking down fruM during stages 5 and 6 , but not later stages , results in developmental defect in the gustatory receptor neurons innervating VNC . In addition to the role of fruM during development to specify female-directed courtship , we also found a role of fruM during adulthood in suppressing male–male courtship , as males with fruM knocked down or tra overexpressed during adulthood displayed enhanced male–male courtship or male chaining behaviors . Note that a previous study found that removal of transformer 2 ( tra2 ) specifically during adulthood using a temperature sensitive tra2 allele induced 8 of 96 females to show male-type courtship behaviors ( Belote and Baker , 1987 ) , which suggests that expression of FRUM and DSXM ( by removal of tra2 function in females ) during adulthood is sufficient to masculinize CNS to some extent and induce a small fraction of females to display male-like courtship behaviors . Recent studies also found that fruM expression in the Or47b-expressing olfactory receptor neurons as well as their neuronal sensitivity depend on social experiences during adulthood ( Hueston et al . , 2016; Sethi et al . , 2019 ) . Based on all these findings , we propose that fruM expression during pupation is crucial for neuronal development and reconstruction of adult sex circuitry that allows innate courtship toward females , and its expression during adulthood may be activity dependent in at least some neurons and modulates some aspects of courtship ( e . g . , inhibits male–male courtship ) . Thus , there are at least two separate mechanisms that fruM contributes to the sex circuitry , one during a critical developmental period to build the female-directed innate courtship into that circuitry , and the other during adulthood to modulate neuronal physiology in an experience-dependent manner . Most importantly , we revealed striking flexibility of the fly sex circuitry by manipulating fruM expression . We listed four cases with fruM manipulation here for comparison: ( 1 ) males with a sex circuitry having wild-type fruM function have innate heterosexual courtship , as they court readily toward females , but do not court males no matter how long they meet; ( 2 ) males with a sex circuitry having no fruM function lose the innate courtship ability , but have the potential to acquire courtship toward males , females , and even other species in an experience-dependent manner; ( 3 ) males with a sex circuitry having limited fruM expression ( e . g . , 40% ) have innate homosexual courtship , as they court readily toward other males , but rarely court females; ( 4 ) males with a sex circuitry having limited fruM expression outside brain ( but intact fruM expression in brain ) are innately bisexual , as they court equally toward females or males . Although previous studies found that different fruM alleles ( e . g . , deletions , inversions , or insertions related to fru ) showed very different courtship abnormalities ( Anand et al . , 2001; Villella et al . , 1997 ) , it was very hard to link fruM function to the flexibility of sex circuitry and often seen as allele-specific or background-dependent phenotypes . Our study using relatively simple genetic manipulations that generate dramatical different courtship behaviors promoted us to speculate a different view about the role of fruM: instead of simply being a master gene that controls all aspects of male courtship , fruM is not absolutely necessary for courtship , but changes the wiring of the sex circuitry during development such that the sex circuitry may function in very different ways , ranging from innately heterosexual , homosexual , bisexual , to largely experience-dependent acquisition of the behavior . Such flexibility of the sex circuitry is tuned by different fruM expression , such that changes of fruM regulatory regions during evolution would easily select a suitable functional mode of the sex circuitry . Flies were maintained at 22 or 25°C in a 12 hr:12 hr light:dark cycle . Canton-S flies were used as the wild-type strain . Other stocks used in this study include the following: fruGAL4 ( Stockinger et al . , 2005 ) , fruV5 ( this study ) , UAS-fruMi ( attp40 ) , UAS-fruMi ( attp2 ) , and UAS-fruMiScr ( attp2 ) ( Meissner et al . , 2016 ) , fruLexA , fru4-40 , fruAJ96u3 , and fruSat15 ( Pan and Baker , 2014 ) , ppk23-GAL4 ( Thistle et al . , 2012 ) , P1-splitGAL4 ( R15A01-AD; R71G01-DBD ) ( Zhang et al . , 2018 ) , and Otd-Flp ( Asahina et al . , 2014 ) . UAS-traF ( BL#4590 ) , tub-GAL80ts ( BL#7019 ) , tub>GAL80> ( BL#38881 ) , and tub>stop>GAL80 ( BL#39213 ) were from Bloomington Drosophila Stock Center . fruV5 was generated by fusing V5 tag in frame with the start codon of fruP1 . To generate the fruV5 knock-in line , two gRNAs ( gRNA1: 5′-GCCATTAGTGTCGCGGTGCG-3′; gRNA2: 5′-GCGGCCGCGCGAGTCGCCGC-3′ ) against fru were inserted into pCFD4 vector ( Addgene #49411 ) to induce DNA break near the start codon of fruP1 . Then , ~2 . 1 kb 5′ homologous arm was incorporated into the 5′ MCS of pHD-DsRed ( Addgene #51434 ) through Gibson assembly ( digested with NheI and NdeI ) . To insert V5 tag after the start codon of fruP1 , ~2 . 4 kb 3′ homologous arm was divided into two fragments and amplified separately . These two fragments including the V5 sequence were then subcloned into the 3′ MCS of pHD-DsRed ( containing the above 5′ homologous arm ) through Gibson assembly ( digested with BglII and XhoI ) . The modified pCFD4 and pHD-DsRed plasmids were injected into vas-cas9 embryos . Successful knock in was selected by 3xP3-DsRed ( DsRed-positive eyes ) and confirmed by PCR followed by sequencing . The verified knock-in line was balanced and crossed to hs-Cre flies to remove the 3xP3-DsRed marker . The rabbit polyclonal antibody against FruM was generated by ABclonal ( Wuhan , China ) . In brief , the fragment of fru gene encodes the N-terminal 101 amino acids , starting with MMATSQDYFG and ending in SPRYNTDQGA , was cloned into expression vector pET-28a ( Sigma–Aldrich , #69864 ) . The 101 amino acids are only present in male-specific Fru proteins ( FruM ) from fruP1 . A SUMO-tagged FruM fusion antigen was synthesized from bacteria , purified , and used to immunize a rabbit . The anti-FruM antibody was affinity purified . For the single-pair courtship assay , the tester males and target flies ( 4–8 days old ) were gently aspirated into round two-layer chambers ( diameter: 1 cm; height: 3 mm per layer ) and were separated by a plastic transparent barrier that was removed ~30 min later to allow courtship test . Courtship index ( CI ) , which is the percentage of observation time a fly performs any courtship step , was used to measure courtship to female targets or between two males . Paired male–male courtship used two males of the same genotype but focused on the male fly that first initiated courtship ( courtship of the initiator to the other ) . All tester flies were single housed if not otherwise mentioned . Each test was performed for 10 min . For male chaining assay , tester males ( 4–8 days old ) were loaded into large round chambers ( diameter: 4 cm; height: 3 mm ) by cold anesthesia . Tests were performed daily for four consecutive days ( 3 hr after grouping as day 0 , then days 1–3 ) . For chaining behavior in Figure 2C , flies were only tested after grouping together for 3 days . Chaining index ( ChI ) , which is the percentage of observation time at least three flies engaged in courtship together , was used to measure courtship in groups of eight males . To generate males with fruM knocked down only for 2 days during development or adulthood , we raised tub-GAL80 ts/+; fruGAL4/UAS-fruMi flies at 18°C and transferred these flies to fresh food vials every 2 days . In this way , we generated tub-GAL80 ts/+; fruGAL4/UAS-fruMi flies at nine different stages from embryos ( stage 1 ) to newly eclosed adults ( stage 9 ) , with wandering larvae at stage 5 and early pupas at stage 6 . We then transferred all these flies to a 30°C incubator allowing fruM knockdown for 2 days , then placed all flies back to 18°C until courtship test at adult . Total RNA was extracted from ~15 male flies with TRIzol ( 15596026 , Invitrogen ) , according to the manufacturer’s instructions . The cDNA was synthesized using Prime Script reagent kit ( 18091050 , Invitrogen ) . Quantitative PCR was performed on LightCycler 96 Real-Time PCR System ( Roche ) using AceQ qPCR SYBR Green Master Mix ( Q121-02 , Vazyme ) . Actin was used as control for normalization . The primers used were as follows: Actin ( forward: 5′- CAGGCGGTGCTTTCTCTCTA-3′; reverse: 5′-AGCTGTAACCGCGCTCAGTA-3′ ) , fru P1 promotor ( forward: 5′-GTGTGCGTACGTTTGAGTGT-3′; reverse: 5′-TAATCCTGTGACGTCGCCAT-3′ ) , and fru P4 promotor ( forward: 5′-TGTATAGCGGCAACTGAACC-3′; reverse: 5′-CCGGTCAAATTTGTGGGATG-3′ ) . We dissected brains and ventral nerve cords of males in defined developmental stages ( e . g . , Figure 1E–H ) or 5–7 days old males in Schneider’s insect medium ( Thermo Fisher Scientific , Waltham , MA ) and fixed in 4% paraformaldehyde in phosphate-buffered saline ( PBS ) for 30 min at room temperature . After washing four times in 0 . 5% Triton X-100% and 0 . 5% bovine serum albumin [BSA] in PBS ( PAT ) , tissues were blocked in 3% normal goat serum ( NGS ) for 60 min , then incubated in primary antibodies diluted in 3% NGS for ~24 hr at 4°C , washed ( 4× 15 min ) in PAT at room temperature , and incubated in secondary antibodies diluted in 3% NGS for ~24 hr at 4°C . Tissues were then washed ( 4× 15 min ) in PAT and mounted in Vectorshield ( Vector Laboratories , Burlingame , CA ) for imaging . Primary antibodies used were rabbit anti-FruM ( 1:200; this study ) , mouse anti-V5-Tag:DyLight550 ( 1:500; MCA1360D550GA , Bio-Rad , Hercules , CA ) , rabbit anti-GFP ( 1:1000; A11122 , Invitrogen , Waltham , MA ) , and mouse anti-Bruchpilot ( 1:50; nc82 , Developmental Studies Hybridoma Bank , Iowa City , IA ) . Secondary antibodies used were donkey anti-mouse IgG conjugated to Alexa 555 ( 1:500 , A31570 , Invitrogen ) and donkey anti-rabbit IgG conjugated to Alexa 488 ( 1:500 , A21206 , Invitrogen ) . Samples were imaged at 10× or 20× magnification on a Zeiss 700 confocal microscope and processed with ImageJ . Experimental flies and genetic controls were tested at the same condition , and data are collected from at least two independent experiments . Statistical analysis is performed using GraphPad Prism and indicated inside each figure legend . Data presented in this study were first verified for normal distribution by D’Agostino–Pearson normality test . If normally distributed , Student’s t test is used for pairwise comparisons , and one-way ANOVA is used for comparisons among multiple groups , followed by Tukey’s multiple comparisons . If not normally distributed , Mann–Whitney U test is used for pairwise comparisons , and Kruskal–Wallis test is used for comparisons among multiple groups , followed by Dunn’s multiple comparisons .	Innate behaviors are behaviors that do not need to be learned . They include activities such as nest building in birds and web spinning in spiders . Another behavior that has been extensively studied , and which is generally considered to be innate , is courtship in fruit flies . Male fruit flies serenade potential mates by vibrating their wings to create a complex melody . This behavior is under the control of a gene called ‘fruitless’ , which gives rise to several distinct proteins , including one that is unique to males . For many years , this protein – called FruM – was thought to be the master switch that activates courtship behavior . But recent findings have challenged this idea . They show that although male flies that lack FruM fail to show courtship behaviors if raised in isolation , they can still learn them if raised in groups . This suggests that the role of FruM is more complex than previously thought . To determine how FruM controls courtship behavior , Chen et al . have used genetic tools to manipulate FruM activity in male flies at different stages of the life cycle and distinct cells of the nervous system . The results revealed that FruM must be present during a critical period of development – but not adulthood – for male flies to court females . However , FruM strongly influences the type of courtship behavior the male flies display . The amount and location of FruM determines whether males show heterosexual , homosexual or bisexual courtship behaviors . Adult flies with lower levels of FruM show an increase in homosexual courtship and a decrease in heterosexual courtship . These findings provide a fresh view on how a master gene can generate complex and flexible behaviors . They show that fruitless , and the FruM protein it encodes , work distinctly at different life cycles to modify the type of courtship behavior shown by male flies , rather than simply switching courtship behavior on and off . Exactly how FruM acts within the fruit fly brain to achieve these complex effects requires further investigation .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "short", "report", "neuroscience" ]	2021	fruitless tunes functional flexibility of courtship circuitry during development
Linking interindividual differences in psychological phenotype to variations in brain structure is an old dream for psychology and a crucial question for cognitive neurosciences . Yet , replicability of the previously-reported ‘structural brain behavior’ ( SBB ) -associations has been questioned , recently . Here , we conducted an empirical investigation , assessing replicability of SBB among heathy adults . For a wide range of psychological measures , the replicability of associations with gray matter volume was assessed . Our results revealed that among healthy individuals 1 ) finding an association between performance at standard psychological tests and brain morphology is relatively unlikely 2 ) significant associations , found using an exploratory approach , have overestimated effect sizes and 3 ) can hardly be replicated in an independent sample . After considering factors such as sample size and comparing our findings with more replicable SBB-associations in a clinical cohort and replicable associations between brain structure and non-psychological phenotype , we discuss the potential causes and consequences of these findings . The early observations of inter-individual variability in human psychological skills and traits have triggered the search for defining their correlating brain characteristics . Studies using in-vivo neuroimaging have provided compelling evidence of a relationship between human skills and traits and brain morphometry that were further influenced by individuals’ years of experience , as well as level of expertise . More subtle changes were also shown following new learning/training ( Draganski et al . , 2004; Taubert et al . , 2011 ) , hence further demonstrating dynamic relationships between behavioral performance and brain structural features . Such observations quickly generated a conceptual basis for growing number of studies aiming to map subtle inter-individual differences in observed behavior such as personality traits ( Nostro et al . , 2017 ) , impulsivity traits ( Matsuo et al . , 2009 ) or political orientation ( Kanai et al . , 2011 ) to normal variations in brain morphology ( for review see Genon et al . , 2018; Kanai and Rees , 2011 ) . Altogether , these studies created an empirical background supporting the assumption that the morphometry of the brain in humans is related to the wide spectrum of aspects observed in human behavior . Such reports on structural brain behavior ( SBB ) associations may not only have important implications in psychological sciences and clinical research ( Ismaylova et al . , 2018; Kim et al . , 2015; Luders et al . , 2013; Luders et al . , 2012; McEwen et al . , 2016 ) , but also possibly hold an important key for our understanding of brain functions ( Genon et al . , 2018 ) and thus concern many research fields including basic cognitive neuroscience . Yet , along with the general replication crisis affecting psychological sciences ( Button et al . , 2013; De Boeck and Jeon , 2018; Open Science Collaboration , 2015 ) , replicability of the previously reported SBB-associations were also questioned recently . In particular , ( Boekel et al . , 2015 ) in a purely confirmatory replication study , picked on few specific previously reported SBB-associations . Strikingly , for almost all the findings under scrutiny , they could not find support for the original results in their replication attempt . In another study we demonstrated lack of robustness of the pattern of correlations between cognitive performance and measures of gray matter volume ( GMV ) in a-priori defined sub-regions of the dorsal premotor cortex in two samples of healthy adults ( Genon et al . , 2017 ) . In particular we found a considerable number of SBB-associations that were counterintuitive in their directions ( i . e . , higher performance related to lower gray matter volume ) . Furthermore , subsampling revealed that for a given psychological score , negative correlations with GMV were as likely as positive correlations . Although our study did not primarily aim to address the scientific qualities of SBB , it revealed , in line with Boekel et al . ( 2015 ) , that a replication issue in SBB-associations could seriously be considered . However , ringing the warning bell of a replication crisis would be premature since these previous studies have approached replicability questions within very specific contexts and methods and using small sample sizes ( Muhlert and Ridgway , 2016 ) . In particular , Boekel et al . and Genon et al . ’s studies were performed by focusing on a-priori defined regions-of-interest ( ROIs ) . However , several SBB studies are commonly performed in groups of dozens of individuals , using an exploratory setting employing a mass-univariate approach . Thus , the null findings of the two questioning studies could be related to the focus and averaging of GMV within specific regions-of-interest , as suggested by Kanai ( 2016 ) and discussed in Genon et al . ( 2017 ) . In stark contrast with this argument , in whole-brain mass-univariate exploratory SBB studies , the multitude of statistical tests that is performed ( as the associations are tested for each voxel , separately ) likely yield many false positives . Directly addressing this limitation , several strategies for multiple comparison correction have been proposed to control the rate of false positives ( Eklund et al . , 2016 ) . We could hence assume that the high number of multiple tests and general low power of neuroimaging studies , combined with the flexible analysis choices ( Button et al . , 2013; Poldrack et al . , 2017; Turner et al . , 2018 ) represent critical factors likely to lead to the detection of spurious and not replicable associations . Characterization of spatial consistency of findings across neuroimaging studies is often performed with meta-analytic approaches , pooling studies investigating similar neuroimaging markers in relation to a given behavioral function or condition . However , in the case of SBB , the heterogeneity of the behavioral measures and the large proportion of apriori-ROI analyses complicate the application of a meta-analytic approach . Illustrating these limitations , previous meta-analyses have focused on specific brain regions and capitalized on a vast majority of ROI studies . For example , ( Yuan and Raz , 2014 ) have focused on SBB within the frontal lobe based on a sample made of approximately 80% of ROI studies . Given these limitations of meta-analytic approaches for the SBB literature , an empirical evaluation of the replicability of the findings yielded by an exploratory approach is crucially needed to allow questioning the replicability of exploratory SBB studies . Thus in the current study , we empirically examined replicability rates of SBB-association over a broad range of psychological scores , among heathy adults . In order to avoid the criticisms raised regarding the low sample size in Boekel et al . ’s study , we used an openly available dataset of a large cohort of healthy participants and assessed replication rate of SBB-associations using both an exploratory as well as a confirmatory approach . While in the recent years multivariate methods are frequently recommended to explore the relationship between brain and behavior ( Cremers et al . , 2017; Smith and Nichols , 2018 ) , SBB-association studies using these approaches remain in minority . The mass-univariate approach is still the main workhorse tool in such studies , not only due to its historical precedence and its wide integration in common neuroimaging tools , but also possibly owing to more straightforward interpretability of the detected effects ( Smith and Nichols , 2018 ) . The current study , therefore , focused on the assessment of replicability of SBB-associations using the latter approach . In particular , we first identified ‘significant’ findings with an exploratory approach based on mass-univariate analysis , searching for associations of GMV with psychometric variables across the whole brain . Here a linear model was fit between inter-individual variability in the psychological score and GMV at each voxel . Inference was then made at cluster level , using a threshold-free cluster enhancement approach ( Smith and Nichols , 2009 ) . We then investigated the reproducibility of these findings , across resampling , by conducting a similar whole-brain voxel-wise exploratory analysis within 100 randomly generated subsamples of individuals ( discovery samples ) . Each of these 100 discovery subsamples ( of the same size ) were generated by randomly selecting apriori-defined number of individuals ( e . g . 70% ) from the original cohort under study . In order to empirically investigate spatial consistency of significant results from these 100 exploratory analyses , an aggregate map characterizing the spatial overlap of the significant findings across all discovery samples was generated . This map denotes the frequency of finding a significant association between the behavioral score and gray matter volume , at each voxel , over 100 analyses and thus provides information about replicability of ‘whole brain exploratory SBB-associations’ for each behavioral score . Conceptually , this map gives an estimate of the spatial consistency of the results that one could expect after re-running 100 times the same SBB study across similar samples . Additionally , for each of the 100 exploratory analyses , we assessed the replicability of SBB-associations using a confirmatory approach ( i . e . ROI-based approach ) . For each of the 100 discovery samples , we generated a demographically-matched test pair sample from the remaining participants of the main cohort . Average GMV within regions showing significant SBB-association in the initial exploratory analysis , that is ROIs , are calculated among the demographically-matched independent sample and their association with the same psychological score was compared between the discovery and matched-replication sub-samples ( see Materials and methods for more details ) . Confirmatory replication is commonly used in the literature ( Boekel et al . , 2015; Genon et al . , 2017; Open Science Collaboration , 2015 ) , nevertheless , there is no single standard defined for evaluating the replication success . Therefore , here , we assessed the replication rate of SBB , for three different definitions of successful replication in the confirmatory analyses: 1- Successful replication of the direction of association , only; 2- Detection of significant ( p<0 . 05 ) association in the same direction as the exploratory results; While the first definition is arguably too lenient and may result in many very small correlation coefficients defined as successful replication , it is frequently used as a qualitative measure of replication and may be used to characterize the possible inconsistency of the direction of associations ( that was observed in our previous study [Genon et al . , 2017] ) . In addition it could be used as a complement for the possible limitation of the second definition , namely the possibility of declaring many replications that fell just short of the bright-line of p<0 . 05 as failed replication . 3- lastly , in line with previous studies and the reproducibility literature , we included the Bayes Factors ( BF ) to quantify evidence that the replication sample provided in favor of existence or absence of association in the same direction than in the discovery subsample ( Boekel et al . , 2015 ) . In other words , when compared to standard p-value methodology , here hypothesis testing using BF enables additional quantification of the evidence in favor of the null hypothesis , that is evidence for the absence of a correlation; see Materials and methods for more details . If the replication issue of SBB associations can be objectively evidenced , this naturally opens the questions of the accounting factors . Here , we considered proximal explanatory factors , in particular at the measurements and analysis level , but also in relation to the object level , that is , in relation to the nature itself of variations in brain structure and psychometric scores in healthy individuals . One main proximal factor that is almost systematically blamed is small sample size . In line with replication studies in other fields ( e . g . Cremers et al . , 2017; Turner et al . , 2018 ) , we thus here investigated the influence of sample size and replication power on the reproducibility of SBB-associations . More specifically for every phenotypic score under study we repeated both whole brain exploratory and ROI-based confirmatory replication analyses using three sample sizes ( see Materials and methods for more details ) to assess how sample size influences replication rate of SBB . Furthermore , for the successfully replicated effects , we also investigated existence of a positive relationship between the effect size of exploratory and confirmatory analyses . Finally , in order to promote discussion on the underlying reality which is aimed to be captured by SBB in the framework of the psychology of individual differences , we included as benchmarks non-psychological phenotypical measures , that is age and body-mass-index ( BMI ) , and extended our analysis to a clinical sample , where SBB-associations are expected to enjoy higher biological validity . For this purpose , a subsample of patients drawn from Alzheimer's Disease Neuroimaging Initiative ( ADNI ) database were selected , in which replicability of structural associations of immediate-recall score from Rey auditory verbal learning task ( RAVLT ) ( Schmidt , 1996 ) was assessed ( see Materials and methods ) . Due to availability of the same score within the healthy cohort , this later analysis is used as a ‘conceptual’ benchmark . According to the scientific literature , associations between psychological phenotype ( cognitive performance and psychological trait ) and local brain structure are not uncommon ( Kanai and Rees , 2011 ) . However , in our exploratory analyses , when looking at a range of psychological variables , significant associations with GMV were very rare . It is worth noting that here by having a-priori fixed analysis design and inference routines , we aimed to avoid ‘fishing’ for significant findings ( Gelman and Loken , 2014 ) . Flexible designs and flexible analyses routines ( Simmons et al . , 2011 ) as well as p-hacking ( John et al . , 2012 ) are considered as inappropriate but frequent research practices ( Poldrack et al . , 2017 ) . Based on our findings of infrequent significant SBB-associations , we could assume that flexible analyses routines , p-hacking and most importantly publication bias ( Dwan et al . , 2013 ) have contributed to the high number of significant SBB-reports in the literature . When considering potential impacts of biased SBB-reports on our confidence of psychological measures , as well as our conception and apprehension of brain-behavior relationships and psychological interindividual differences , we would strongly argue for null findings reports . Such reports would contribute to a more accurate and balanced apprehension of associations between differences in psychological phenotype and brain morphometric features , but it would also help to progressively disentangle factors that mediate or modulate the relationship between brain structure and behavioral outcomes . In addition to the low likelihood of finding ‘any’ significant SBB-association using the exploratory approach , clusters that do survive the significance thresholding did not often overlap in different subsamples . Furthermore , the probability of spatial overlap further dropped as the number of participants in the subsamples decreased ( Figure 1 ) . Putting this finding in light of the literature brings two main hypotheses . First , from the conceptual level , we could hypothesize that the pattern of correlation between a psychological measure is by nature spatially diffuse at the brain level . Psychological measures aim to conceptually articulate behavioral functions and processes , thus , in most cases , they have not been developed to identify specific localized brain functions . Following this philosophical segregation between psychological sciences and neurosciences , it is now widely acknowledged that there is no one-to-one mapping between behavioral functions and brain regions ( Anderson , 2016; Genon et al . , 2018; Pessoa , 2014 ) . Instead , mapping a psychological concept to brain features usually result in a diffuse brain spatial pattern with small effect sizes ( Bressler , 1995; Poldrack , 2010; Tononi et al . , 1998 ) . From this axiom , we can expect that several studies conducted in small samples ( specifically after rigorous corrections for multiple comparisons ) are likely to each capture a partial and minor aspect of the whole true association pattern , resulting in a poor replication rate for each individual study ( i . e . high type II error ) . Alternatively , a more parsimonious hypothesis is a methodological one questioning the truth or validity of the found significant associations hence considering them as spurious ( i . e . type I error ) . Psychological and MRI measurements are both relatively indirect estimations of respectively , behavioral features and brain structural features and thus are susceptible to noise . Correlations in small samples in the presence of noise for both type of variables is likely to produce spurious significant results ( Loken and Gelman , 2017 ) by fitting a correlation or regression between random noise in both variables . Thus , the pattern of poor spatial consistency of SBB findings could result either from factors at the object of study level , that is the relationship between brain and behavior , or , from factors at the measurement and analysis level . While the latter hypothesis is more parsimonious , one argument for the former hypothesis comes from the relatively substantial replications by-sign observed in our confirmatory analyses , of three top behavioral scores ( see Figure 2 ) . If the significant SBB findings would be purely driven by noise in the data , we would expect them to show purely random signs across resampling , which was not the case ( but also see Figure 2—figure supplement 1 for example of scores with lower replicability and higher inconsistent associations across resampling ) . Therefore , it is actually likely that both hypotheses hold true and that the spatial variability of significant SBB findings result from both , factors at the analyses levels and factors at the object level , potentially interacting together . It is worth noting that similar complexity and uncertainty have been described for task-based functional associations studies ( Cremers et al . , 2017; Turner et al . , 2018 ) . In particular , Cremers et al . ( 2017 ) using simulated and empirical data demonstrated that task-based functional activations have a generally weak and diffuse pattern . Therefore , these authors concluded that most whole-brain analyses in small samples , specifically when combined with stringent correction for multiple comparison , to control the false positive rates , would most likely frequently overlook global meaningful effects and depict results with poor replicability ( type II error ) . Relatedly , in the present study , higher spatial extent and lower consistency of significant findings in smaller samples in Figure 1 , also suggest higher number of spurious associations ( type I error ) in smaller samples ( due to winners curse [Button et al . , 2013; Forstmeier and Schielzeth , 2011] ) than in larger samples . These factors , added to the complexity of human behavior , renders the objective of capturing covariations with psychometric variables in brain structure locally particularly challenging . For that reason , in exploratory studies whose aim is to identify brain structural features correlating with a specific ( set of ) psychological variable ( s ) , a multivariate approach could be advised ( Habeck et al . , 2010; McIntosh and Mišić , 2013 ) . As mentioned earlier , like all methods , multivariate analyses have their own limitations: in particular , the ensuing difficulty of interpretability of the revealed pattern . While some authors argue either for one or the other approach , the use of these approaches are far from being mutually exclusive ( Moeller and Habeck , 2006 ) . Combining both approaches in small datasets indeed revealed that the results of the univariate approach reflect the ‘tip of the iceberg’ of the behavior’s brain correlates , whose spatial extent are more comprehensively captured with the multivariate analysis , but interpretability is facilitated by the use of univariate analyses; for example ( Genon et al . , 2016; Genon et al . , 2014 ) . Thus , to partially address the previously described concerns of small and spatially diffuse effects at the brain level in exploratory whole-brain-behavior study , we here recommend for the future studies to combine a univariate and a multivariate approach . Although it does not provide any protection against the influence of noise that may affect both approaches , this solution may help to reduce the false negatives . ROI-based analysis further highlighted that significant associations , which have been discovered when starting with a psychological measure and searching within the whole brain for a significant association ( i . e . ‘evidenced in an exploratory study’ ) , show poor replicability ( using significance and Bayes factor criteria , but also using a similar sign criterion for most psychometric scores; For example , see Figure 2—figure supplement 1 and Figure 4—figure supplement 1 ) in a confirmatory ROI-based study ( in line with what was previously shown by Boekel et al . , 2015 ) . These findings thus call for a general acknowledgment of the uncertainty and fragility of exploratory findings and the need for out of sample confirmatory replications to provide evidence about the robustness of the reported effects ( Ioannidis , 2018; Tukey , 1980 ) . Another clear finding of our study is the overestimation of the effect size in the exploratory approach ( Kriegeskorte et al . , 2010 ) , specifically in smaller samples ( see marginal distributions of the x- and y-axis in Figure 3 ) . For the majority of the psychological scores , in the ROI-based approach , we failed to find a clear association between effect size in the discovery and replication samples . Instead , we observed a rather high estimated statistical power for replication ( due to an inflated effect size estimation [Ioannidis , 2008] ) , despite very low actual rate of replicated effects in the independent samples . These findings are particularly important when considering the current research context , in which power analyses are encouraged to justify the allocation of financial and human investment in specific future researches . Prospective studies with power analyses are frequently proposed , where power is computed based on the findings from previous exploratory analyses in a small sample ( Albers and Lakens , 2018 ) . An inflated effect size estimation from the exploratory study results in an unreliable high power , which in turn lead to confidence in prospective studies to find relevant findings and hence in the allocation and possibly waste of ( frequently public ) resources ( Albers and Lakens , 2018; Poldrack et al . , 2017 ) . Nevertheless , this provocative conclusion does not imply that SBB studies should be banished to hell . Our conclusion here mainly concerns the study of association between variations at standard psychological measures and variations in measures of gray matter in ‘small’ samples of healthy individuals . Our results further show that different types of SBB exploratory studies should not be epistemologically all put in the same pot . In support for this argument , in ADNI sample , despite the additional confounding effect of different scanners and/or scanning parameters due to the multi-site nature of the cohort , associations between immediate-recall score and GMV were relatively stable . Compared to associations of the same measure of verbal learning performance within the healthy population ( see supplementary Figure 1 ) , these results highlight the superior reliability of SBB-associations that are defined in a clinical context . These findings have important conceptual implications . From an epistemological and conceptual point of view , our comparative investigation suggests that the object of study matters in the replicability of SBB . Searching for correlation between variations in cognitive performance and GMV in healthy adults , on one hand , and in neurodegenerative patients , on the other hand , appear as two different objects of study , with different replicability rates . While several SBB results in healthy population are likely to be spurious ( see Supplementary file 2 ) , it seems that SBB in clinical population are more likely to capture true relationships . Thus , maybe the conceptual objective itself should be questioned: should we expect the association between normal psychological phenotype , in particular cognitive performance , in healthy population to be substantially driven by local brain macrostructure morphology ? Brain structure can certainly not be questioned as the primary substrates of behavior and more than a century of lesion studies recalls this primary principle to our attention ( Broca , 1865; Scoville and Milner , 1957 ) , but this does not imply that ‘normal’ variations at standard psychological tests can be related to variations in markers of local brain macrostructure . Our results suggest that reliable answer to this important question requires substantially big samples ( bigger than those used here ) and independent replications . The sample size and related power issues hold a central position in the current discussions of the replication crisis in behavioral sciences , as well as in neuroimaging studies ( Button et al . , 2013; Ioannidis , 2005; Lilienfeld , 2017; Munafò et al . , 2017; Open Science Collaboration , 2015 ) . Higher power is defined as increased probability of finding effects that are genuinely true . Furthermore , high power experiments have higher positive predictive values ( PPV ) of the claimed effects ( i . e . probability that the claimed effect reflects a true effect ) . They also result in less exaggerated effects sizes when a true effect is discovered ( Button et al . , 2013 ) . As such , in the discovery sample , by increasing the sample size , the correlation coefficients get closer to their real value and their PPV increases . However , in the current study , as the number of participants in the main sample is limited , the size of the discovery and their matched replication samples are dependent on each other . Therefore , for each behavioral measure , larger discovery samples have smaller replication counterparts . These smaller replication samples have in turn lower power to find the true effects and have lower PPV . However , in splits with larger replication samples , as the discovery sample gets smaller , apart from the lower PPV , the estimated correlation coefficients are possibly more exaggerated ( e . g . due to winner’s curse ) ( Cremers et al . , 2017 ) and thus the power of the replication would be over-estimated . This is a limitation which complicates the interpretation of the relationship between the calculated replication power and the actual rate of replicability of associations in the present study . We hoped that the use of a large cohort of healthy individuals as our main cohort would result in large enough discovery and test cohorts and thus minimize the impact of above-mentioned limitation . However , large discrepancies between the rate of ‘significant’ within-split replicability and the a-priori estimated replication power , as we observed in the ROI-based confirmatory analyses , confirms an exaggerated power estimation in most of our analyses and thus highlights the insufficiency of the size of the discovery and replication samples . Thus overall , these findings suggest that samples consisting of ~200–300 participants have in reality still low power to identify reliable SBB-associations among healthy participants . However , the sample size of SBB studies is usually substantially smaller . Figure 5 depicts the distribution of sample sizes ( log-scale ) of published studies examining GMV in human participants with the standard voxel-based morphometry approach across previous years ( BrainMap data [Vanasse et al . , 2018] ) . SBB studies in healthy adults also fall under this general trend . Based on our current work , we would argue that the probability of finding spurious or inconclusive results and exaggerated effect size estimations in these studies is thus quite high ( Albers and Lakens , 2018; Schönbrodt and Perugini , 2013; Yarkoni , 2009 ) . In addition , to underscore the importance of the sample size , our analyses and results further show that the size of the replication sample also matters when examining the replicability of a previous SBB findings . This is an obvious factor that has been frequently neglected in the discussions about replication crisis . Yet , while many replication studies straightforwardly blame the sample size of the original studies , it is important to keep in mind that a replication failure might also come from a too small sample size of the replication study ( Muhlert and Ridgway , 2016 ) . When interpreting our results , it should be noted that , in order to keep large sample sizes for the exploratory replication analyses , the discovery subsamples were not necessarily designed to be independent from each other . Considering this limitation , the poor spatial consistency of the whole brain exploratory associations that we observed for almost all the behavioral scores is hence even more alarming . As discussed earlier , another indirect limitation of the limited size of the selected cohort is the dependence between the size of the discovery and their matched replication sub-samples . This limitation prevents us to state strong conclusions about the relationship between the calculated replication power and the actual rate of replicability . Overall , these acknowledged limitations raise the need for even larger sample sizes for such investigations . Recent advancements through data collection from much larger number of participants , such as UK-biobank ( Miller et al . , 2016 ) are promising opportunities for overcoming these limitations in future replication studies . Moreover , the generalizability of our results are partly limited to our methodological choices such as the computation of volumetric markers of brain structure ( as opposed to surface-based markers ) , the size of the smoothing kernel , and the use of a priori-defined ROIs in the replication sample . Future studies should therefore investigate to which extend our replicability rates are reproduced with different data preprocessing pipelines and analyses approaches . Overall , our work and review of the recent and concomitant replication literature in related fields demonstrate that several improvements could be recommended to get more accurate insight on the relationship between psychological phenotype and brain structure and to progressively answer open questions . Importantly , our recommendations and suggestions concern different levels of SBB researches: the dataset level , the analyses level , as well as at the post-publication and replication level . At the dataset level , our study pointed out the need for big data samples to identify robust associations between psychological variables and brain structure , with sample size of at least several hundreds of participants . It should be acknowledged that this conclusion is easier to achieve than to implement in research practice . Nevertheless , large scale cohort datasets from healthy adult populations , such as eNKI used in the current study , human connectome project ( HCP ) ( WU-Minn HCP Consortium et al . , 2013 ) and UK-biobank ( Miller et al . , 2016 ) are now openly available , hence facilitating endeavor in that direction . At the analysis level , we recommend the combined use of multivariate analyses , for comprehensive assessment of the spatial extent of associations and , univariate analyses , to facilitate interpretability , when studying brain structural correlates of psychological measures . Furthermore , we emphasis on the importance of independent confirmatory replications to provide evidence about the robustness of the effects . Finally , at the post-analysis level , we concluded from our observations that publication of null findings should be more encouraged . In addition to directly shaping a more objective picture of SBB-associations , these null-reports could contribute to new quantitative approaches . In particular , meta-analyses of published literature ( Vanasse et al . , 2018 ) would clearly benefit from such unbiased reports of null findings . Sharing raw data would undoubtedly improve the problem of low statistical power , but if not possible , sharing the unthresholded statistical maps ( e . g . through platforms such as Neurovault [Gorgolewski et al . , 2015] ) could also be a significant scientific contribution . In addition to directly contribute to our understanding of brain-behavior relationship , such efforts would open up new possibilities for estimating the correct size and extent of effects by integrating unthresholded statistical maps in the estimation of the effects sizes throughout the brain . Thus , we could hope that sharing initiatives will also contribute indirectly to more valid and insightful SBB studies in the remote future and hence to a better allocation of resources . Healthy adults’ data were selected from the enhanced NKI ( eNKI ) Rockland cohort ( Nooner et al . , 2012 ) . Data collection received ethics approval through both the Nathan Klein Institute and Montclair State University . Written informed consent was obtained from all participants . We focused only on participants for which good quality T1-weighted scans was available along with timewise-corresponding psychological data . Exclusion criteria consisted of alcohol or substance dependence or abuse ( current or past ) , psychiatric illnesses ( eg . Schizophrenia ) and current depression ( major or bipolar ) . Furthermore , we excluded participants with missing information on important confounders ( age , gender , education ) or bad quality of structural scans after pre-processing , resulting in a total sample of 466 healthy participants ( age: 48 ± 19 , 153 male ) . Replicability of SBB-associations within the clinical sample was investigated using a subsample drawn from the Alzheimer's Disease Neuroimaging Initiative ( ADNI ) database , which was launched in 2003 as a public–private partnership and led by Principal Investigator Michael W Weiner . The primary goal of ADNI has been to test whether serial magnetic resonance imaging ( MRI ) , positron emission tomography ( PET ) , other biological markers , and clinical and neuropsychological assessment can be combined to measure the progression of mild cognitive impairment ( MCI ) and early Alzheimer’s disease ( AD ) . For up-to-date information , see www . adni-info . org . We used the baseline measurements from 371 patients ( age: 71 ± 7 , 200 male; 47 with significant memory complaint , 177 early MCI , 85 late MCI and 62 AD patients ( defined based on ADNI diagnostic criteria , see http://adni . loni . usc . edu/wp-content/themes/freshnews-dev-v2/documents/clinical/ADNI-2_Protocol . pdf ) , in whom anatomical brain scans had been acquired in a 3Tesla scanner ( from 39 different sites ) . The imaging data of the eNKI cohort were all acquired using a single scanner ( Siemens Magnetom TrioTim , 3 . 0 T ) . T1-weighted images were obtained using a MPRAGE sequence ( TR = 1900 ms; TE = 2 . 52 ms; voxel size = 1 mm isotropic ) . ADNI , on the other hand , is a multisite dataset . Here we selected a subset of this data , which has been acquired in a 3 . 0 T scanner ( baseline measurements from ADNI2 and ADNI GO cohort ) from 39 different sites; see http://adni . loni . usc . edu/methods/documents/ for more information . Both datasets were preprocessed using the CAT12 toolbox ( Gaser and Dahnke , 2016 ) . Briefly , each participant’s T1-weighted scan was corrected for bias-field inhomogeneities , then segmented into gray matter ( GM ) , white matter ( WM ) , and cerebrospinal fluid ( CSF ) ( Ashburner and Friston , 2005 ) . The segmentation process was further extended for accounting for partial volume effects ( Tohka et al . , 2004 ) by applying adaptive maximum a posteriori estimations ( Rajapakse et al . , 1997 ) . The gray matter segments were then spatially normalized into standard ( MNI ) space using Dartel algorithm ( Ashburner , 2007 ) and further modulated . The modulation was performed by scaling the normalized gray matter segments for the non-linear transformations ( only ) applied at the normalization step . While this procedure ignores the volume changes due to affine transformation , it allows preserving information about individual differences in local gray matter volume . In other words , it re-introduces individual differences in local gray matter volume removed in the process of inter-subject registration and normalization . Finally modulated gray matter images were smoothed with an isotropic gaussian kernel of 8 mm ( full-width-half-maximum ) . SBB-associations are commonly derived in an exploratory setting using a mass-univariate approach , in which a linear model is used to fit interindividual variability in the psychological score to GMV at each voxel . Inference is then usually made at cluster level , in which groups of adjacent voxels that support the link between GMV and the tested score are clustered together . Replicability of thus-defined associations could be assessed by conducting a similar whole-brain voxel-wise exploratory analysis in another sample of individuals and comparing the spatial location of the significant findings that survive multiple comparison correction , between the two samples . Alternatively , replicability could be assessed , using a confirmatory approach , in which only regions showing significant SBB-association in the initial exploratory analysis , that is regions of interest ( ROIs ) , are considered for testing the existence of the association between brain structure and the same psychological score in an independent sample . The latter procedure commonly focuses on a summary measure of GMV within each ROI and tests for existence of the SBB-association in the direction suggested by the initial exploratory analysis . Thus this approach circumvents the need for multiple comparison correction and therefore increases the power of replication . Here we assessed replicability of associations between each behavioral measure and gray mater structure , using both approaches: the whole brain replication approach and the ROI replication approach , which are explained in details in the following sections .	All human brains share the same basic structure . But no two brains are exactly alike . Brain scans can reveal differences between people in the organization and activity of individual brain regions . Studies have suggested that these differences give rise to variation in personality , intelligence and even political preferences . But recent attempts to replicate some of these findings have failed , questioning the existence of such a direct link , specifically between brain structure and human behavior . This had led some disagreements whether there is a general replication crisis in psychology , or if the replication studies themselves are flawed . Kharabian Masouleh et al . have now used brain scans from hundreds of healthy volunteers from an already available dataset to try to resolve the issue . The volunteers had previously completed several psychological tests . These measured cognitive and behavioral aspects such as attention , memory , anxiety and personality traits . Kharabian Masouleh et al . performed more than 10 , 000 analyzes on their dataset to look for relationships between brain structure and psychological traits . But the results revealed very few statistically significant relationships . Moreover , the relationships that were identified proved difficult to replicate in independent samples . By contrast , the same analyzes demonstrated robust links between brain structure and memory in patients with Alzheimer's disease . They also showed connections between brain structure and non-psychological traits , such as age . This confirms that the analysis techniques do work . So why did the new study find so few relationships between brain structure and psychological traits , when so many links have been reported previously ? One possibility is publication bias . Researchers and journals may be more likely to publish positive findings than negative ones . Another factor could be that that most studies use too few participants to be able to reliably detect relationships between brain structure and behavior , and that studies with 200 to 300 participants are still too small . Therefore , future studies should use samples with many hundreds of participants , or more . This will be possible if more groups make their data available for others to analyze . Researchers and journals must also be more willing to publish negative findings . This will help provide an accurate view of relationships between brain structure and behavior .	[ "Abstract", "Introduction", "Discussion", "Materials", "and", "methods" ]	[ "neuroscience" ]	2019	Empirical examination of the replicability of associations between brain structure and psychological variables
Nuclear factor kappa B ( NF-κB ) -mediated transcription is an important mediator for cellular responses to DNA damage . Genotoxic agents trigger a 'nuclear-to-cytoplasmic' NF-κB activation signaling pathway; however , the early nuclear signaling cascade linking DNA damage and NF-κB activation is poorly understood . Here we report that Src-associated-substrate-during-mitosis-of-68kDa/KH domain containing , RNA binding , signal transduction associated 1 ( Sam68/KHDRBS1 ) is a key NF-κB regulator in genotoxic stress-initiated signaling pathway . Sam68 deficiency abolishes DNA damage-stimulated polymers of ADP-ribose ( PAR ) production and the PAR-dependent NF-κB transactivation of anti-apoptotic genes . Sam68 deleted cells are hypersensitive to genotoxicity caused by DNA damaging agents . Upregulated Sam68 coincides with elevated PAR production and NF-κB-mediated anti-apoptotic transcription in human and mouse colon cancer . Knockdown of Sam68 sensitizes human colon cancer cells to genotoxic stress-induced apoptosis and genetic deletion of Sam68 dampens colon tumor burden in mice . Together our data reveal a novel function of Sam68 in the genotoxic stress-initiated nuclear signaling , which is crucial for colon tumorigenesis . Nuclear factor kappa B ( NF-κB ) is a crucial transcription factor in a variety of pathophysiological conditions ( Dietz and Bahr , 2004; Grilli et al . , 1993; Harhaj and Dixit , 2012; Hayden and Ghosh , 2008; Natoli , 2010; Smale , 2011; Vallabhapurapu and Karin , 2009; Wan and Lenardo , 2010; Wertz and Dixit , 2010; Wertz et al . , 2004 ) . NF-κB responds to genotoxic threats ( e . g . DNA damaging agents and γ-irradiation ) via the activation of the inhibitor of NF-κB kinase ( IKK ) and NF-κB liberation from IκB proteins , similar to the canonical pathway activated by external stimuli ( Janssens et al . , 2005; Perkins , 2007; Scheidereit , 2006; Wu and Miyamoto , 2007 ) . NF-κB signaling pathway has emerged as an important mediator for cellular responses to DNA damage , in particular NF-κB-conferred anti-apoptotic transcription facilitates the cell 'escape' from the lethal effects of DNA damage ( Janssens et al . , 2005; Perkins , 2007; Scheidereit , 2006; Wu and Miyamoto , 2007 ) and initiates cell cycle checkpoint control to promote cellular recovery from damage ( McCool and Miyamoto , 2012; Miyamoto , 2011 ) . Besides ataxia telangiectasia mutated ( ATM ) and IKKγ , two known crucial regulators of the genotoxic stress-activated NF-κB signaling pathway ( Li et al . , 2001; Piret et al . , 1999 ) , poly ( ADP-ribose ) polymerase 1 ( PARP1 ) was recently revealed to be indispensable for the signaling cascade that links nuclear DNA damage recognition to cytoplasmic IKK activation ( Stilmann et al . , 2009 ) . Sequential post-translational modifications , including phosphorylation , ubiquitination and SUMOylation , of these signaling regulators are critical for NF-κB activation following DNA damage ( Huang et al . , 2003; Mabb et al . , 2006; Wu et al . , 2006 ) , in particular , PARP1-catalyzed poly ( ADP-ribosyl ) ation ( PARylation ) has emerged as a vital means for rapid assembly of the signaling complexes that are critical for DNA damage-initiated NF-κB activation ( Mabb et al . , 2006; Stilmann et al . , 2009 ) . Although these studies have considerably advanced our understanding of the cellular response to DNA damage , the genotoxic stress-initiated ‘‘nuclear-to-cytoplasmic’’ NF-κB signaling pathway remains poorly understood , in particular the early signaling networks linking DNA lesion recognition in the nucleus to subsequent activation of IKK and liberation of NF-κB in the cytoplasm . Sam68 ( Src-associated substrate during mitosis of 68 kDa , also named KH domain containing , RNA binding , signal transduction associated 1 [KHDRBS1] , and encoded by KHDRBS1 gene ) , an RNA-binding protein that preferentially resides in the nucleus , plays versatile functions in an increasing number of cellular processes ( Bielli et al . , 2011; Cheung et al . , 2007; Fu et al . , 2013; Glisovic et al . , 2008; Henao-Mejia et al . , 2009; Huot et al . , 2012; Iijima et al . , 2011; Lukong and Richard , 2003; Matter et al . , 2002; Paronetto et al . , 2009; Rajan et al . , 2008a , 2008b; Ramakrishnan and Baltimore , 2011; Richard , 2010; Sette , 2010; Yang et al . , 2002 ) . Through its KH ( heteronuclear ribonucleoprotein particle K homology ) domain , Sam68 is capable of binding single- and double-stranded DNA in addition to RNA ( Lukong and Richard , 2003 ) . Of note , Sam68 was identified as a PAR-binding protein in alkylating agent treated cells ( Gagne et al . , 2008 ) and a putative substrate of ATM , ATM and Rad3-related ( ATR ) , and DNA-dependent protein kinase ( DNA-PK ) ( Beli et al . , 2012 ) , which suggests that Sam68 could be an important molecule in the cellular response to DNA damage . Although emerging evidence suggests the involvement of Sam68 in multiple signaling pathways , it has not been extensively investigated yet whether Sam68 , an almost strictly nuclear protein , participates in the signal communication network of nuclear-initiated signaling pathways . Moreover , aberrant expression of Sam68 has been acknowledged in multiple cancers and elevated Sam68 expression correlates with tumor progression and poor prognosis in cancer patients ( Chen et al . , 2012; Liao et al . , 2013; Song et al . , 2010; Zhang et al . , 2009 ) . Overexpression of Sam68 has been proposed as a prognostic marker ( Chen et al . , 2012; Liao et al . , 2013; Song et al . , 2010; Zhang et al . , 2009 ) , however , the precise function of Sam68 in cancer development and survival remains obscure . Here we report that Sam68 is an important regulator in genotoxic stress-initiated early signaling in the nucleus , which leads to NF-κB activation . Sam68 deletion diminishes DNA damage-stimulated PARP1 activation and PAR production , as well as the PAR-dependent NF-κB signaling and transactivation of an array of anti-apoptotic genes . As a consequence , Sam68 knockout cells are hypersensitive to genotoxicity caused by γ-irradiation and DNA damaging chemicals . Moreover , downregulation of Sam68 substantially sensitizes human colorectal cancer cells to spontaneous and genotoxic stress-induced cell death and retards colon tumor growth and survival in genetically susceptible Apcmin716/+ mice . Hence our data reveal a crucial function of Sam68 in the genotoxic stress-initiated 'nuclear to cytoplasmic' NF-κB transactivation and the involvement of Sam68 in the development and survival of colon cancer . To examine the potential role of Sam68 in nuclear-initiated NF-κB activation , we first compared the genotoxic stress-induced NF-κB signaling in immortalized wild-type and Sam68 knockout ( KO ) mouse embryonic fibroblasts ( MEFs ) . As expected , Camptothecin ( CPT ) , a DNA-damaging chemical that inhibits DNA topoisomerase I ( Stilmann et al . , 2009 ) , stimulated a rapid degradation of IκBα , a prerequisite for NF-κB liberation and transactivation , in a dose- and time-dependent manner in wild-type MEFs ( Figure 1A–B and Figure 1—figure supplement 1A ) . In contrast , DNA damage-induced IκBα degradation was remarkably attenuated in Sam68 KO MEFs ( Figure 1A–B ) and MEFs without poly ( ADP-ribose ) polymerase 1 ( PARP1 ) ( Figure 1—figure supplement 1B ) , a recently identified key nuclear regulator of DNA damage-induced NF-κB activation ( Stilmann et al . , 2009 ) . Moreover , CPT treatment triggered remarkable nuclear translocation of NF-κB in wild-type MEFs ( Figure 1C ) ; additionally , the derived nuclear extracts formed high-affinity binding complexes with immunoglobin ( Ig ) κB double-stranded DNA , which was further confirmed by cold oligonucleotide competition and super shift assays ( Figure 1D ) . In striking contrast , NF-κB nuclear accumulation and binding capacity to Ig κB DNA were almost abolished in the CPT-stimulated Sam68 KO MEFs ( Figure 1C–D ) . To ascertain that Sam68 deficiency solely results in the impaired genotoxic stress-induced NF-κB activation in MEFs , we examined the NF-κB signaling in wild-type MEFs with Sam68 knockdown by small interference RNAs ( siRNAs ) and Sam68 KO MEFs supplemented with exogenous Sam68 . CPT-triggered IκBα degradation was substantially reduced in Sam68-specific siRNA-expressing wild-type MEFs , compared to scrambled non-specific siRNA-transfected cells ( Figure 1—figure supplement 1C ) , whereas ectopic expression of green fluorescent protein ( GFP ) tagged Sam68 , but not GFP alone , markedly restored CPT-induced IκBα degradation in Sam68 KO MEFs ( Figure 1—figure supplement 1D–E ) . Together , these results suggest that Sam68 could execute an essential function in the nuclear-initiated NF-κB signaling pathway . 10 . 7554/eLife . 15018 . 003Figure 1 . Sam68 is required for the nuclear-initiated NF-κB signaling in response to DNA damage . ( A and B ) Whole cell lysates from wild-type ( WT ) and Sam68 knockout ( KO ) mouse embryonic fibroblasts ( MEFs ) treated with indicated concentrations of CPT for 2 hr ( A ) or 10 μM of CPT for indicated periods ( B ) were immunoblotted ( IB ) for IκBα and Sam68 , with β-actin as a loading control . Right , the IκBα levels , normalized to β-actin and untreated controls , were quantified from three independent experiments . ( C ) Cytosolic and nuclear fractions derived from WT and Sam68 KO MEFs stimulated with 10 μM of CPT for indicated periods were IB for indicated proteins . Caspase-3 ( Casp3 ) and PARP1 served as loading controls and cytosolic and nuclear markers , respectively . Right , the p65 levels in the nucleus , normalized to PARP1 and untreated controls , were quantified from three independent experiments . ( D ) Nuclear extracts of WT and Sam68 KO MEFs treated with ( + ) or without ( − ) CPT ( 10 μM , 2 hr ) were analyzed by EMSA with 32P-labeled immunoglobin ( Ig ) κB or OCT1 oligonucleotides . In some cases , EMSA was performed in the presence of 100-fold unlabeled Ig κB or OCT1 oligonucleotide competitors ( lanes 5–6 ) or p65 antibody ( Ab ) ( lane 7 ) . Ig κB DNA binding complexes are labeled with filled triangles , and the supershifted band and nonspecific band are labeled with an open triangle and an asterisk , respectively . ( E ) WT and Sam68 KO MEFs were stimulated with 10 μM of CPT for indicated periods , and immunoprecipitants ( IP ) with IKKγ antibody were immunoblotted for indicated proteins . ( F and G ) Whole cell lysates ( Input ) from WT and Sam68 KO MEFs stimulated with 10 μM of CPT for indicated periods were IB directly or after IP with PARP1 antibody ( F ) or Sam68 antibody ( G ) for indicated proteins . Data are representative of at least three independent experiments . Results in ( A , B and C ) are expressed as mean and s . e . m . p<0 . 05 , p<0 . 01 , *p<0 . 001 by Student’s t tests . DOI: http://dx . doi . org/10 . 7554/eLife . 15018 . 00310 . 7554/eLife . 15018 . 004Figure 1—figure supplement 1 . Sam68 is critical for genotoxic stress-induced IκBα degradation . ( A ) Wild-type ( WT ) mouse embryonic fibroblasts ( MEFs ) , pretreated with DMSO or 10 μM of MG132 for 2 hr , were stimulated with 10 μM of CPT for indicated periods . Whole cell lysates were derived and immunoblotted ( IB ) for IκBα , with β-actin as a loading control . ( B ) Whole cell lysates from WT , Sam68 knockout ( Sam68 KO ) , and PARP1 knockout ( PARP1 KO ) MEFs treated with ( + ) or without ( − ) 10 μM of CPT for 2 hr were IB for IκBα , PARP1 , and Sam68 , with β-actin as a loading control . ( C ) WT MEFs transiently transfected with either non-specific control ( si-NC ) small interference RNA ( siRNA ) or siRNA targeting Sam68 ( si-Sam68 Sam68 ) were treated with ( + ) or without ( − ) 10 μM of CPT for 2 hr , and whole cell lysates were derived and IB for IκBα and Sam68 , with β-actin as a loading control . ( D ) Schematic diagram of the full-length ( residues 1–443 ) and truncated mutant ( ΔC lacks residues 347–443 ) of Sam68 fused with GFP . The hnRNP K homology ( KH ) domain and nuclear localization signal ( NLS ) are labeled in red and blue , respectively . Immunofluorescence micrographs of MEFs transiently transfected with GFP or indicated GFP fusion proteins , with nuclei counterstained by DAPI . Scale bar , 10 μm ( bottom ) . ( E ) WT and Sam68 KO MEFs transiently transfected with GFP or indicated GFP fusion proteins were stimulated with ( + ) or without ( − ) CPT ( 10 μM , 2 hr ) , and whole cell lysates were derived and IB for IκBα and GFP , with β-actin as a loading control . ( F ) WT and Sam68 KO MEFs transiently transfected with GFP or GFP-Sam68 fusion protein were stimulated with 10 μM of CPT for indicated time periods . Nuclear fractions were derived and IB for indicated proteins . Caspase-3 ( Casp3 ) and PARP1 served as loading controls and cytosolic and nuclear markers , respectively . p-p65 , Ser536 phosphorylated p65 . DOI: http://dx . doi . org/10 . 7554/eLife . 15018 . 00410 . 7554/eLife . 15018 . 005Figure 1—figure supplement 2 . Sam68 complexes with PARP1 and IKKγ during the cellular response to genotoxic stress . ( A ) Scheme of Sam68 depletion of in CPT-stimulated wild-type mouse embryonic fibroblasts ( MEFs ) . MEFs were stimulated with 100 μM of CPT for 0 , 20 , and 60 min . Under the Sam68 depletion condition , cell lysates were precleared with anti-Sam68 antibody ( IP: Sam68 ) in order to remove all the Sam68-containing complexes . The remaining lysates ( Fraction Sam68 depleted ) were immunoprecipitated ( IP ) with an anti-PARP1 , and the interaction between PARP1 and IKKγ was evaluated by immunoblot ( IB ) . Under the condition without Sam68 depletion , cell lysates were directly IP with anti-PARP1 or anti-Sam68 antibody . ( B ) Identical aliquots of indicated fractions were IB for indicated proteins , with β-actin as a loading control . ( C ) The indicated fractions from either Sam68-depleted or –undepleted condition were IP with anti-PARP1 . The immunoprecipitants were separated and IB for PARP1 and IKKγ . ( D ) The Sam68-undepleted whole cell lysates were IB directly or after IP with anti-Sam68 antibody for the indicated proteins . DOI: http://dx . doi . org/10 . 7554/eLife . 15018 . 005 Sam68 is an almost strictly nuclear protein; its nuclear import is conferred by a nuclear localization signal ( NLS ) in the C-terminus ( Fu et al . , 2013; Ishidate et al . , 1997; Lukong and Richard , 2003 ) . Consistently , the ΔC ( deletion of C-terminal NLS ) mutant of Sam68 switches its preferred subcellular localization to the cytoplasm , in contrast to the strict nuclear accumulation observed in wild-type Sam68 ( Figure 1—figure supplement 1D ) . Sam68 KO MEFs reconstituted with wild-type Sam68 , but not GFP control , displayed restored CPT-triggered NF-κB activation signaling including IκBα degradation , phosphorylation of p65 , and p65 nuclear accumulation ( Figure 1—figure supplement 1E–F ) . In contrast , supplementing with the Sam68 ( ΔC ) mutant failed to restore CPT-triggered IκBα degradation ( Figure 1—figure supplement 1E ) , which indicates an indispensable role of the Sam68 nuclear function in DNA damage-initiated NF-κB activation . Indeed , CPT treatment-induced SUMOylation of IKKγ , a pivotal post-translational modification on IKKγ in the nucleus that subsequently leads to activation of IKKβ and liberation of NF-κB in the cytoplasm ( Huang et al . , 2003; Mabb et al . , 2006 ) , was substantially attenuated in Sam68 KO MEFs in comparison to wild-type controls ( Figure 1E ) . Moreover , assembly of the PARP1-IKKγ signal complex has been established as a prerequisite for cytoplasmic NF-κB activation in the cellular response to genotoxic stresses ( Gibson and Kraus , 2012; McCool and Miyamoto , 2012; Miyamoto , 2011 ) . As expected , PARP1 complexed with IKKγ following CPT treatment , whereas there was no detectable PARP1-IKKγ interaction in the absence of genotoxic stress in wild-type MEFs ( Figure 1F ) . In striking contrast , the CPT-induced PARP1-IKKγ interaction was abolished in Sam68 KO MEFs ( Figure 1F ) . Moreover , strong inducible interactions among Sam68 , PARP1 , and IKKγ were observed in wild-type MEFs , at 20 min after CPT treatment ( Figure 1F–G ) . Of note , immune-depletion of Sam68 using antibodies abolished the CPT-induced PARP1-IKKγ interaction in wild-type MEFs ( Figure 1—figure supplement 2 ) , suggesting that PARP1 interacts with Sam68 and IKKγ simultaneously . Hence our results suggest that Sam68 participates in the early nuclear signaling cascade in DNA damage-initiated NF-κB activation . Beyond its indispensable role in DNA repair ( Gibson and Kraus , 2012 ) , emerging evidence demonstrates that PARP1-mediated PARylation is one of the most crucial post-translational modifications orchestrating DNA damage-initiated NF-κB signaling ( Stilmann et al . , 2009 ) . The inducible Sam68-PARP1 interaction following genotoxic stress led us to examine whether Sam68 impacts DNA break-stimulated PAR synthesis . As expected , CPT treatment induced a wide array of PARylated proteins peaking at 20 min post stimulation in wild-type MEFs , which were diminished by one-hour pretreatment with two independent PARP inhibitors , Olaparib and PJ-34 , thus confirming the PAR specificity ( Figure 2A and Figure 2—figure supplement 1A ) . Strikingly , genotoxic stress-induced PAR synthesis was markedly dampened in Sam68 KO MEFs ( Figure 2A and Figure 2—figure supplement 1A ) . Consistently , genotoxic stress-induced PARylation of the known PARP1 substrates including PARP1 itself , NBS1 , and Ku70 was also substantially attenuated in Sam68 KO MEFs ( Figure 2—figure supplement 1B ) . PAR that is synthesized after DNA damage can be rapidly degraded by PAR glycohydrolase ( PARG ) , therefore we examined the possibility that PARG is hyper-activated in the absence of Sam68 , thus degrading the formed PAR chains immediately after genotoxic stress . However , down-regulation of PARG by siRNA in Sam68 KO MEFs failed to restore CPT-induced PAR production to a similar level to that in wild-type MEFs ( Figure 2—figure supplement 2 ) , which suggests that the attenuated PAR formation in the absence of Sam68 is not caused by a rapid degradation of PAR chains . Furthermore , CPT and γ-irradiation stimulated PAR production was markedly reduced in primary colonic epithelial cells ( CECs ) derived from Khdrbs1-/- ( Sam68 knockout ) mice , compared to Khdrbs1+/- ( Sam68 heterozygote ) controls ( Figure 2B–D ) , thus supporting the critical function of Sam68 in controlling DNA damage-stimulated PAR production . 10 . 7554/eLife . 15018 . 006Figure 2 . Sam68 facilitates PARP1-catalyzed PARylation in response to DNA damage . ( A ) Wild-type ( WT ) and Sam68 knockout ( KO ) mouse embryonic fibroblasts ( MEFs ) pretreated with Olaparib ( 10 μM ) , PJ-34 ( 10 μM ) , or DMSO for 1 hr , were stimulated with 10 μM of CPT for indicated periods , and whole cell lysates were immunoblotted ( IB ) for indicated proteins , with β-actin as a loading control . ( B and C ) Primary colonic epithelial cells ( CECs ) isolated from Khdrbs1+/- and Khdrbs1-/- mice were stimulated with 10 μM of CPT ( B ) or 10 Gy of γ-irradiation ( IR ) ( C ) . Whole cell lysates were derived at indicated periods post stimulation and IB as in ( A ) . Right , the PAR levels , normalized to β-actin and untreated controls , were quantified from three independent experiments . ( D ) Immunofluorescence micrographs of PARylated proteins ( PAR ) in CECs treated as in ( C ) , with nuclei counterstained by DAPI . Scale bar , 10 μm . Percentage of CECs ( >100 cells from 5–8 random fields ) with PAR staining was quantified ( right ) . ( E ) WT and Sam68 KO MEFs were γ-irradiated ( IR ) at 10 Gy and the chromatin , soluble ( Sol . fr . ) , and insoluble ( Ins . fr . ) subcellular fractions were derived at indicated time points post γ-irradiation and IB for Sam68 , Caspase-3 ( Casp-3 ) , and Histone H3 ( H3 ) . ( F ) WT MEFs were stimulated with 10 μM of CPT for indicated periods and the chromatin fractions were derived and IB as in ( E ) . Right , the Sam68 and PARP1 levels in the chromatin fractions , normalized to H3 and untreated controls , were quantified from three independent experiments . ( G ) Whole cell lysates from WT and Sam68 KO MEFs stimulated with 10 μM of CPT for 20 min , were IB directly or after immunoprecipitation ( IP ) with PARP1 antibody for indicated proteins . ( H ) Recombinant PARP1 protein was incubated in reaction buffer containing damaged DNA or with purified GST or GST-Sam68 protein in the presence or absence of NAD+ or the PARP1 inhibitor PJ-34 , as indicated . The reaction mixture was separated by SDS-PAGE and subjected to IB with the PAR , PARP1 , and Sam68 antibodies . Results in ( B , C , D and F ) are expressed as mean and s . e . m . ns , non-significant difference and , p<0 . 05 , **p<0 . 001 by Student’s t tests . Data are representative of at least three independent experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 15018 . 00610 . 7554/eLife . 15018 . 007Figure 2—figure supplement 1 . Sam68 deletion attenuates genotoxic stress-stimulated PARylation . ( A ) Wild-type ( WT ) and Sam68 knockout ( KO ) mouse embryonic fibroblasts ( MEFs ) pretreated with PJ-34 ( 10 μM ) or DMSO for 1 hr , were stimulated with 10 μM of CPT for indicated periods , and whole cell lysates were immunoblotted ( IB ) for indicated proteins , with β-actin as a loading control . ( B ) WT and Sam68 KO MEFs were γ-irradiated ( IR ) at 10 Gy . Whole cell lysates ( Input ) derived at 0 or 5 min post γ-irradiation were IB directly or after immunoprecipitation ( IP ) with PAR antibody for indicated proteins . DOI: http://dx . doi . org/10 . 7554/eLife . 15018 . 00710 . 7554/eLife . 15018 . 008Figure 2—figure supplement 2 . Down-regulation of PARG does not restore DNA damage-induced PAR production in Sam68 KO MEFs . Wild-type ( WT ) and Sam68 knockout ( Sam68 KO ) mouse embryonic fibroblasts ( MEFs ) transfected with non-specific control ( si-NC ) or PARG-specific ( si-PARG ) small interference RNA , as indicated , were stimulated with 100 μM of CPT for 0 , 5 , and 20 min . Whole cell lysates were derived and immunoblotted ( IB ) for indicated proteins , with β-actin as a loading control . The down-regulation of PARG in indicated cells was evaluated by semi-quantitative RT-PCR ( Right ) . DOI: http://dx . doi . org/10 . 7554/eLife . 15018 . 00810 . 7554/eLife . 15018 . 009Figure 2—figure supplement 3 . Sam68 functions upstream of PARP1 in cellular response to genotoxic stresses . ( A ) Wild-type mouse embryonic fibroblasts ( MEFs ) , pretreated with PJ-34 ( 10 μM ) or DMSO for 1 hr , were γ-irradiated ( IR ) at 10 Gy . The chromatin fractions derived at indicated time points post γ-irradiation were immunoblotted ( IB ) for PARP1 and Sam68 , with Histone H3 ( H3 ) as a loading control . ( B ) Wild-type ( WT ) and Sam68 knockout ( KO ) MEFs expressing indicated GFP or GFP-fusion proteins were IR at 10 Gy . Whole cell lysates were derived at indicated time points post γ-irradiation and IB for the indicated proteins . DOI: http://dx . doi . org/10 . 7554/eLife . 15018 . 00910 . 7554/eLife . 15018 . 010Figure 2—figure supplement 4 . Sam68-stimulated PARP1 PARylation is DNA dependent . ( A ) Recombinant PARP1 protein was incubated in reaction buffer containing NAD+ and damaged DNA as indicated . The reaction mixture was separated by SDS-PAGE and subjected to immunoblot ( IB ) with the PAR and PARP1 . ( B ) Recombinant PARP1 protein was incubated in reaction buffer containing damaged DNA or with purified GST or GST-Sam68 protein in the presence or absence of NAD+ or the PARP1 inhibitor PJ-34 , as indicated . The reaction mixture was separated by SDS-PAGE and subjected to IB with the PAR , PARP1 , and Sam68 antibodies . ( C ) Recombinant proteins were incubated in reaction buffer in the presence or absence of damaged DNA and NAD+ , as indicated . The reaction mixture was separated by SDS-PAGE and IB for indicated proteins . DOI: http://dx . doi . org/10 . 7554/eLife . 15018 . 010 Given the evidence that PARP1 is rapidly recruited to DNA damage sites ( Krishnakumar and Kraus , 2010 ) , we performed chromatin fractionation assays to examine the possibility that Sam68 can be recruited to DNA lesions . Remarkably , Sam68 was enriched in chromatin fractions in MEFs following γ-irradiation ( Figure 2E ) and CPT treatment ( Figure 2F ) , and the kinetics of Sam68 enrichment was similar to that of PARP1 on damaged chromatin ( Figure 2F ) , which supports the inducible interaction between Sam68 and PARP1 in the early DNA damage signaling ( Figure 1F–G ) . Moreover , Sam68 and PARP1 were still substantially enriched in chromatin fractions in the PARP1-inhibited MEFs following γ-irradiation ( Figure 2—figure supplement 3A ) , indicating that genotoxic stress-induced Sam68 enrichment in chromatin is not dependent on PAR formation . We further examined whether ectopic expression of PARP1 could rescue the defect in DNA damage-induced PAR formation caused by Sam68 loss . As expected , supplementing exogenous GFP-Sam68 , compared to the GFP control , largely restored γ-irradiation-stimulated PAR production in Sam68 KO MEFs ( Figure 2—figure supplement 3B ) . In contrast , PARP1 overexpression did not rescue the PAR chain formation in response to γ-irradiation under a Sam68 deleted condition ( Figure 2—figure supplement 3B ) , which suggests that Sam68 is required for DNA damage-triggered PARP1 activation and is consistent with Sam68 being an essential upstream activator of PARP1 . DNA-dependent PARP enzymes , i . e . PARP1 and PARP2 , are activated during the cellular response to DNA damage , of which PARP1 is the most robust enzyme that catalyzes >90% of PAR formation in cells ( Krishnakumar and Kraus , 2010 ) . In contrast to the genotoxic stress-induced Sam68-PARP1 interaction ( Figure 1F–G and Figure 1—figure supplement 2D ) , no detectable interaction between Sam68 and PARP2 was observed under either unstimulated or damaged conditions ( Figure 1—figure supplement 2D ) . Provided the crucial funciton of PARP1 in DNA damage responses , we therefore examined whether Sam68 impacts PARP1-conferred PARylation following DNA damage . In line with the vigorous PAR production in the presence of genotoxic stress ( Figure 2A ) , immunoprecipitated PARP1 was associated with Sam68 and various PARylated target proteins in CPT-treated wild-type MEFs ( Figure 2G ) . In contrast , the PARylated species were dramatically reduced despite an equal amount of PARP1 immunoprecipitated from Sam68 KO MEFs was used ( Figure 2G ) . Conversely , supplementing with GFP-tagged Sam68 , in comparison to a GFP control , markedly augmented the DNA damage-triggered PAR synthesis in CPT-treated Sam68 KO MEFs ( Figure 2—figure supplement 3B ) , further supporting the key function of Sam68 in controlling DNA damage-initiated PARylation . To assess the direct impact of Sam68 on PARP1-catalyzed PAR production , we utilized recombinant PARP1 and Sam68 proteins in in vitro PARylation assays . DNA damage-activated PARP1 auto-modified itself with the addition of PAR moieties in the presence of nicotinamide adenine dinucleotide ( NAD+ ) and DNA , as indicated by the formation of PARylated PARP1 species ( Figure 2H and Figure 2—figure supplement 4A ) and a corresponding reduction in unmodified PARP1 protein ( Figure 2H and Figure 2—figure supplement 4B , compare lane 4 with lane 3 ) . Strikingly , incubation of recombinant Sam68 protein , but not GST control , with PARP1 dramatically boosted PAR production , which was paralleled by a sharp loss of unmodified PARP1 protein ( Figure 2H and Figure 2—figure supplement 4B , compare lane 5 with lane 4 ) . Of note , the amount of Sam68 protein from the boosted PARylation reaction remained at a comparable level ( Figure 2H , compare lane 5 with lane 7 ) , thus ruling out the possibility that Sam68 is a substrate of PARP1 in vitro and indicating that Sam68 harbors a stimulatory function for DNA-dependent PARP1 activity . In the absence of PARP1 , neither GST nor GST-Sam68 protein exhibited PARylation activity ( Figure 2H , lanes 1 , 2 , 6 , 7 , and 10 ) , indicating that Sam68 per se does not possess the enzymatic activity to transfer ADP-ribosyl polymers . Moreover , in the absence of damaged DNA , incubation of Sam68 and PARP1 failed to form detectable PAR chains from supplemented NAD+ ( Figure 2—figure supplement 4C ) , which suggests that Sam68 stimulates DNA-dependent PARP1 activation and subsequent PAR production . Using various Sam68 truncates ( Figure 3A ) , we sought to understand the key domain ( s ) in Sam68 needed for its interaction with PARP1 . We detected the association of the full-length , ΔC , and ΔKH truncated Sam68 to PARP1 , but not GFP control ( Figure 3B and Figure 3—figure supplement 1A ) . In contrast , deletion of the N-terminal amino acids 1–102 ( ΔN ) of Sam68 almost abolished the association of Sam68 to PARP1 ( Figure 3B and Figure 3—figure supplement 1A ) , suggesting that the N-terminal residues are critical for the Sam68-PARP1 interaction . Moreover , our pull-down assays using recombinant proteins demonstrated a direct Sam68-PARP1 interaction ( Figure 3C ) . In contrast to the strong association between PARP1 and full-length Sam68 , ΔN truncated Sam68 protein barely interacted with PARP1 ( Figure 3C ) , which further supports the critical role of N-terminus of Sam68 for the Sam68-PARP1 interaction . To examine the functional importance of the Sam68-PARP1 interaction , we compared DNA damage-stimulated NF-κB signaling in Sam68 KO MEFs supplemented with full-length or ΔN mutant Sam68 . Transient transfection of full-length Sam68 , but not GFP control , significantly restored genotoxic stress-induced PARylation ( Figure 3D and Figure 3—figure supplement 1C ) , assembly of the PARP1-IKKγ signaling complex ( Figure 3D ) , cytoplasmic degradation of IκBα , and nuclear translocation of p65 ( Figure 3—figure supplement 1C ) , consistent with our previous observation ( Figure 1F–G , 2G , and Figure 1—figure supplement 1E–F ) . In contrast , despite full-length Sam68 and Sam68 ( ΔN ) truncate sharing strict nuclear localization ( Figure 3E ) , ectopic expression of Sam68 ( ΔN ) mutant failed to restore the DNA damage-triggered PARylation , PARP1-IKKγ signal complex assembly , IκBα degradation , and p65 nuclear accumulation ( Figure 3D and Figure 3—figure supplement 1B–C ) . Moreover , we carried out in vitro PARylation assays using recombinant Sam68 and Sam68 ( ΔN ) proteins , to examine the role of the Sam68-PARP1 interaction for Sam68-stimulated PARP1 activation . Consistent with our previous observation ( Figure 2H ) , incubation of Sam68 with PARP1 substantially enhanced PAR formation in the presence of damaged DNA and NAD+ ( Figure 3—figure supplement 1D , compare lane 3 with lane 7 ) . However , the stimulatory effect of Sam68 on PARP1 activation and PARylation under the same condition was dramatically impeded by Sam68 ( ΔN ) protein ( Figure 3—figure supplement 1D , compare lane 7 with lane 11 ) , suggesting that the Sam68-PARP1 interaction is essential for Sam68 to stimulate PARP1 activation . Therefore our results demonstrate that the N-terminus of Sam68 is important for the association between Sam68 and PARP1 and the Sam68-PARP1 interaction is critical for DNA damage-induced PARylation and PAR-dependent NF-κB signaling . 10 . 7554/eLife . 15018 . 011Figure 3 . N-terminus of Sam68 is crucial for the Sam68-PARP1 interaction and genotoxic stress-induced NF-κB signalosome assembly . ( A ) Schematic diagram of Sam68 protein ( residues 1–443 ) , full-length or indicated mutants ( ΔN lacks residues 1–102 , ΔC lacks 347–443 , and ΔKH lacks 165–224 ) fused with GFP . KH , The hnRNP K homology ( KH ) domain and nuclear localization signal ( NLS ) are labeled in black and orange , respectively . ( B ) Whole cell lysates from HEK293T cells expressing indicated GFP or GFP-fusion proteins were IB directly or after IP with GFP antibody for indicated proteins . ( C ) Whole cell lysate ( Input ) containing recombinant PARP1 were IB directly or after pulldown with indicated GST or GST-fusion proteins for indicated proteins . ( D ) Sam68 KO MEFs expressing GFP , GFP-Sam68 , or GFP-Sam68 ( ΔN ) proteins were stimulated with 10 μM of CPT for indicated periods , and the derived whole cell lysates ( WCL ) were IB directly or after IP with PARP1 antibody for indicated proteins . ( E ) Immunofluorescence micrographs of Sam68 KO MEFs expressing GFP ( Vehicle ) , GFP-Sam68 ( Sam68 ) , or GFP-Sam68 ( ΔN ) proteins , with nuclei counterstained by DAPI . Scale bar , 10 μm . Data in ( B–E ) are representative of at least three independent experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 15018 . 01110 . 7554/eLife . 15018 . 012Figure 3—figure supplement 1 . The N-terminus-mediated Sam68-PARP1 interaction is critical for DNA damage-induced PARylation and NF-κB activation . ( A ) Whole cell lysates ( WCL ) from HEK293T cells expressing GFP or indicated GFP-fusion proteins were immunoblotted ( IB ) directly or after immunoprecipitation ( IP ) with GFP antibody for indicated proteins . ( B ) Sam68 knockout ( KO ) mouse embryonic fibroblasts ( MEFs ) expressing GFP , GFP-Sam68 , or GFP-Sam68 ( ΔN ) proteins were mock- or γ-irradiated ( IR ) at 10 Gy for 5 min and whole cell lysates were derived and IB for indicated proteins , with β-actin as a loading control . ( C ) Sam68 KO MEFs expressing GFP , GFP-Sam68 , or GFP-Sam68 ( ΔN ) proteins were stimulated with 10 μM of Camptothecin ( CPT ) for indicated periods , and cytosolic and nuclear fractions were derived and IB for indicated proteins . Caspase-3 ( Casp3 ) and PARP1 served as loading controls and cytosolic and nuclear markers , respectively . ( D ) Indicated recombinant proteins were incubated in reaction buffer containing damaged DNA in the presence and absence of NAD+ and PARP1 inhibitor PJ-34 . The reaction mixture was separated and IB for indicated proteins . Data are representative of at least three independent experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 15018 . 012 To explore the relevance of Sam68 in primary cells , we examined the DNA damage-initiated NF-κB activation in CECs , known for their DNA damage-hypersensitivity , from Khdrbs1+/- and Khdrbs1-/- mice . CPT- and γ-irradiation-induced IκBα degradation rapidly occurred in Khdrbs1+/- CECs , but was significantly attenuated in Khdrbs1-/- cells ( Figure 4A and Figure 4—figure supplement 1A ) . In parallel , γ-irradiation and CPT triggered p65 nuclear translocation in Khdrbs1+/- CECs , whereas the nuclear accumulation of p65 was nearly abolished in Khdrbs1-/- cells ( Figure 4B–C and Figure 4—figure supplement 1B ) . Moreover , in Khdrbs1+/- CECs , γ-irradiation stimulated the SUMOylation of IKKγ and migration of IKKγ from the nucleus to the cytoplasm , both required signaling events for cytoplasmic NF-κB liberation ( McCool and Miyamoto , 2012; Miyamoto , 2011 ) ; however , such signaling events was dampened in γ-irradiated Khdrbs1-/- cells ( Figure 4C–D ) . These results , together with our findings that Sam68 deletion diminishes DNA damage-initiated PARylation in CECs ( Figure 2B–D ) , suggest that Sam68 is essential in the nuclear-initiated NF-κB signaling in mouse primary cells . 10 . 7554/eLife . 15018 . 013Figure 4 . Sam68 deficiency attenuates NF-κB-mediated anti-apoptotic transcription in mouse colonic epithelial cells ( CECs ) under genotoxic stresses and sensitizes CECs to death . ( A ) Whole cell lysates from primary Khdrbs1+/- and Khdrbs1-/- CECs treated with 10 Gy of γ-irradiation ( IR ) for indicated periods were immunoblotted ( IB ) for IκBα and Sam68 , with β-actin as a loading control . Right , the IκBα levels , normalized to β-actin and untreated controls , were quantified from three independent experiments . ( B ) Isolated Khdrbs1+/- and Khdrbs1-/- CECs were treated with 10 Gy of IR for indicated periods , and fixed cells were stained for p65 and nuclei and subjected to immunofluorescence micrographs . Percentage of CECs ( >100 cells from 5–8 random fields ) with nuclear p65 staining was quantified . ( C ) Cytosolic and nuclear fractions derived from Khdrbs1+/- and Khdrbs1-/- CECs treated as in ( B ) were IB for indicated proteins . Cytosolic Caspase-3 ( Casp3 ) and PARP1 served as loading controls and cytosolic and nuclear markers , respectively . ( D ) Whole cell lysates from Khdrbs1+/- and Khdrbs1-/- CECs stimulated with 10 Gy of IR for indicated periods , were IB for indicated proteins after immunoprecipitation ( IP ) with IKKγ antibody . ( E ) Total RNA was extracted from Khdrbs1+/- and Khdrbs1-/- CECs at indicated time points following IR ( 10 Gy ) and mRNA profiles of Birc3 , Bcl2l1 , Xiap , and Actb were analyzed by semi-quantitative RT-PCR . Right , the relative expression levels of Birc3 , Bcl2l1 and Xiap , normalized to Actb and untreated controls , were quantified from three independent experiments . ( F ) Khdrbs1+/- and Khdrbs1-/- CECs were γ-irradiated with indicated doses for 6 hr , and whole cell lysates were derived and IB for indicated proteins . c-Casp3 , cleaved Caspase-3 . The full-length and cleaved PARP1 are indicated by a black triangle and a red triangle , respectively; the two species of Bcl-XL proteins are labeled by open triangles . ( G ) Immunofluorescence micrographs of c-Casp3 in CECs treated as in ( F ) , with nuclei counterstained by DAPI . Scale bar , 10 μm . Percentage of CECs ( >100 cells from 5–8 random fields ) with c-Casp3 staining was quantified ( right ) . ( H ) Khdrbs1+/- and Khdrbs1-/- CECs were treated as in ( F ) , and live cells following γ-irradiation were counted using a particle counter and normalized to the un-irradiated controls . Results in ( A , B , E , G and H ) are expressed as mean and s . e . m . ns , non-significant difference; p<0 . 05; p<0 . 01; p<0 . 001 by Student’s t tests . Data are representative of at least three independent experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 15018 . 01310 . 7554/eLife . 15018 . 014Figure 4—figure supplement 1 . Sam68 deletion attenuates genotoxic stress-induced NF-κB signaling cascade in primary mouse cells . ( A ) Whole cell lysates from isolated Khdrbs1+/- ( Sam68 heterozygote ) and Khdrbs1-/- ( Sam68 knockout ) colonic epithelial cells ( CECs ) treated with indicated concentrations of Camptothecin ( CPT ) for 2 hr were immunoblotted ( IB ) for IκBα and Sam68 , with β-actin as a loading control . ( B ) Nuclear fractions derived from isolated Khdrbs1+/- and Khdrbs1-/- CECs stimulated with 25 μM of CPT for indicated periods were IB for indicated proteins . DOI: http://dx . doi . org/10 . 7554/eLife . 15018 . 01410 . 7554/eLife . 15018 . 015Figure 4—figure supplement 2 . Sam68 deficiency attenuates DNA damage-triggered NF-κB-mediated expression of anti-apoptotic molecules . ( A ) Wild-type ( WT ) and Sam68 knockout ( KO ) mouse embryonic fibroblasts ( MEFs ) were γ-irradiated ( IR , 10 Gy ) and total RNA was extracted at indicated time points . The mRNA profiles of Birc3 , Xiap , and Actb were analyzed by semi-quantitative RT-PCR . ( B ) WT and Sam68 KO MEFs were γ-irradiated ( IR , 10 Gy ) for indicated periods and whole cell lysates derived and IB for indicated proteins , with β-actin as a loading control . DOI: http://dx . doi . org/10 . 7554/eLife . 15018 . 015 NF-κB-mediated transcription of a panel of anti-apoptotic molecules is an important factor for cell fate determination after DNA damage ( Chen et al . , 2015; Kim et al . , 2005; Stilmann et al . , 2009 ) . Indeed , mRNA levels of Birc3 ( encoding cellular inhibitor of apoptosis protein-1 , cIAP1 ) , Bcl2l1 ( encoding B-cell lymphoma-like-1 , Bcl-XL ) , and Xiap ( encoding X-linked inhibitor of apoptosis protein , XIAP ) were elevated in Sam68 sufficient CECs and MEFs post γ-irradiation ( Figure 4E and Figure 4—figure supplement 2A ) . However , γ-irradiation-induced transcription of these genes was attenuated in Sam68 KO CECs and MEFs ( Figure 4E and Figure 4—figure supplement 2A ) , in line with attenuated nuclear-initiated NF-κB signaling ( Figure 4A–D ) . Moreover , XIAP and Bcl-XL protein levels were enhanced in Khdrbs1+/- CECs ( Figure 4F ) and wild-type MEFs ( Figure 4—figure supplement 2B ) following γ-irradiation , whereas such induction did not occur in Khdrbs1-/- CECs and MEFs ( Figure 4F and Figure 4—figure supplement 2B ) . In contrast , cleavage of PARP1 and Caspase-3 , two known biochemical hallmarks for apoptosis , were elevated in the γ-irradiated Khdrbs1-/- , but not Khdrbs1+/- CECs ( Figure 4F–G ) , mirroring the inefficient anti-apoptotic gene expression in Khdrbs1-/- CECs ( Figure 4E–F ) . Consistently , more Khdrbs1-/- CECs underwent cell death in response to γ-irradiation than Khdrbs1+/- controls , as reflected by an irradiation dose-dependent cell loss ( Figure 4H ) . Thus our data suggest that Sam68 deficiency diminishes nuclear-initiated NF-κB signaling , thus dampening NF-κB-mediated anti-apoptotic gene transcription and promoting cells to undergo cell death . It is widely accepted that massive intrinsic DNA damage occurs during rapid DNA replication and proliferation in cancer cells ( Hanahan and Weinberg , 2011; Rouleau et al . , 2010 ) , and that the dysregulation of NF-κB and apoptosis play crucial roles in cancer development and progression ( Townson et al . , 2003; Zubair and Frieri , 2013 ) . We therefore examined the relevance of Sam68 and DNA damage-initiated NF-κB signaling in Apcmin716/+ mice , a mouse model for human colon cancer ( Wu et al . , 2009 ) . Colon adenomas spontaneously developed in Apcmin716/+ mice , as conveyed by staining with hematoxylin/eosin and the colon cancer marker β-catenin ( Figure 5A ) . Interestingly , Sam68 levels were elevated in colon tumors , compared to adjacent normal tissue , from the tumor-laden Apcmin716/+ mice ( Figure 5B and D ) . Moreover , the enhanced Sam68 expression coincided with elevated levels of PAR production , phosphorylated p65 ( phosphor-p65 , indicative of NF-κB activation ) , and Bcl-XL ( anti-apoptotic transcriptional target of NF-κB ) in colon tumors ( Figure 5B–D ) . Similarly , in tissue samples derived from colon adenocarcinoma patients ( Figure 5E and Figure 5—source data 1 ) , Sam68 levels were substantially elevated in 16 ( 94 . 1% ) out of 17 patients , when compared to adjacent normal tissue from the same patient ( Figure 5G–H ) . Moreover , anti-apoptotic molecules Bcl-XL and XIAP were both upregulated at the transcriptional ( 5 [100%] out of 5 patients ) and translational ( 16 [94 . 1%] out of 17 patients ) levels in colon tumors , in comparison to normal tissue controls ( Figure 5F–H ) . Furthermore , the elevated Sam68 levels positively correlated with increased PAR and phospho-p65 levels in human colon cancer samples ( Figure 5H–I ) . Therefore these correlative results suggest that elevated Sam68 levels could facilitate PAR synthesis and PAR-dependent NF-κB signaling/transactivation of anti-apoptotic genes to counter intrinsic DNA damage in human and mouse colon tumor cells . 10 . 7554/eLife . 15018 . 016Figure 5 . Sam68 , PAR , and NF-κB-mediated anti-apoptotic transcription are elevated in mouse and human colon cancers . ( A and E ) Hematoxylin and eosin ( H&E ) staining and β-catenin immunohistochemistry of colon sections from tumor-loaded Apcmin716/+ mice ( A ) and tissue sections of colon tumor or adjacent normal colon tissue from human cancer patients ( E ) . Scale bars , 200 μm . N , normal tissue; T , tumor tissue . ( B , C and G ) Immunofluorescence micrographs of indicated proteins in colon sections from tumor-loaded Apcmin716/+ mice ( B , C ) or Normal and Tumor tissue derived from human colon cancer patients ( G ) , with nuclei counterstained by DAPI . Scale bars , 100 μm . ( D and H ) Colonic epithelial cells were isolated from normal ( N ) or tumor ( T ) colon tissue from tumor-loaded Apcmin716/+ mice ( D ) or normal ( N ) and tumor ( T ) tissue derived from human colon cancer patients ( Pt . ) ( H ) and whole cell lysates were derived and immunoblotted for indicated proteins , with β-actin as a loading control . The two species of Bcl-XL proteins are labeled by open triangles . ( F ) Relative mRNA levels of BCL2L1 and XIAP , normalized to ACTB , from normal ( N ) and tumor ( T ) tissue derived from human colon cancer patients ( Pt . ) . ( I ) Linear regression analysis of the levels of Sam68 protein versus PAR and phosphorylated p65 in CECs from normal ( blue ) and tumor ( red ) tissue derived from human colon cancer patients . AU , arbitrary unit . Results in ( F ) are expressed as mean and s . e . m . ns , non-significant difference; p<0 . 05; p<0 . 01; p<0 . 001 by Student’s t tests . Data in ( A–H ) are representative of at least three independent experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 15018 . 01610 . 7554/eLife . 15018 . 017Figure 5—source data 1 . Surgical colorectal cancer ( CRC ) and polyp metadata . Listed are the metadata of the surgical CRC and polyp from human patients , which have been used for this research . NA , not available . DOI: http://dx . doi . org/10 . 7554/eLife . 15018 . 017 To assess the impact of Sam68 in colon tumor development and survival , we examined colon adenoma development in Apcmin716/+ mice in the presence and absence of Sam68 . Apcmin716/+; Khdrbs1+/- mice , compared to wild-type controls , spontaneously developed adenomas in the cecum and distal colon at 3 months of age , as conveyed by whole mount methylene blue staining ( Figure 6A ) and showed substantial increases in the size and load of tumors ( Figure 6B ) . In contrast , colon tumor development , as reflected by both tumor size and tumor load , in Apcmin716/+; Khdrbs1-/- mice was dramatically reduced; albeit genetic deletion of Sam68 in Apcmin716/+ mice did not significantly affect tumor number ( Figure 6B ) . These results illustrate an essential role of Sam68 in the colon tumor growth and survival in Apcmin716/+ mice . 10 . 7554/eLife . 15018 . 018Figure 6 . Sam68 plays a critical protective role for the survival of mouse and human colon cancers . ( A ) Methylene blue ( MB ) staining of the colons ( with cecum , proximal colon [PC] , distal colon [DC] , and anus indicated ) derived from 3-month old Apcmin716/+; Khdrbs1+/- and Apcmin716/+; Khdrbs1-/- mice . Red arrows indicate colon tumors . Scale bar , 1 cm . ( B ) Quantification of tumor number , tumor size , and tumor load in the colons from Apcmin716/+; Khdrbs1+/- ( n = 6 ) and Apcmin716/+; Khdrbs1-/- mice ( n = 3 ) following MB staining . ( C ) HCT8 and HCT116 cells were transfected with nonspecific control ( si-NC ) or Sam68-specific ( si-Sam68 ) small interference RNAs . 72 hr later , whole cell lysates were derived and immunoblotted ( IB ) for indicated proteins , with β-actin as a loading control . ( D ) Immunofluorescence micrographs of Bcl-XL and PARylated ( PAR ) proteins in the si-NC and si-Sam68 transfected HCT116 cells , with nuclei counterstained by DAPI . Scale bar , 20 μm . ( E ) Percentage of HCT116 cells ( >100 cells from 5–8 random fields ) with Bcl-XL and PAR staining was quantified . ( F ) HCT116 cells transfected with indicated siRNAs as in ( C ) were stimulated with indicated doses of Camptothecin ( CPT ) for 6 hr . Whole cell lysates were derived and IB for indicated proteins , with β-actin as a loading control . c-Casp3 , cleaved Caspase-3 . The full-length and cleaved PARP1 are indicated by a black triangle and a red triangle , respectively; the two species of Bcl-XL proteins are labeled by open triangles . ( G ) HCT8 cells transfected with indicated siRNAs as in ( C ) were left untreated ( UT ) or stimulated with 10 μM of CPT for 12 hr , followed by propidium iodide ( PI ) /Annexin V staining and flow cytometry analysis . ( H ) Percentages of apoptotic ( PI- Annexin V+ ) HCT8 cells treated as in ( G ) were quantified . Results in ( B , E , and H ) are expressed as mean and s . e . m . ns , non-significant difference; p<0 . 05; p<0 . 01; p<0 . 001 by Student’s t tests . Data in ( A and C–H ) are representative of at least three independent experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 15018 . 01810 . 7554/eLife . 15018 . 019Figure 6—figure supplement 1 . Sam68 knockdown and PARP inhibition attenuate PAR synthesis and PAR-dependent NF-κB transactivation in human colon cancer cell lines . ( A ) T84 cells were transfected with nonspecific control ( si-NC ) or Sam68-specific ( si-Sam68 ) small interference RNAs . 72 hr later , whole cell lysates were derived and immunoblotted ( IB ) for indicated proteins , with β-actin as a loading control . ( B ) Immunofluorescence micrographs of Bcl-XL and PARylated ( PAR ) proteins in the si-NC and si-Sam68 transfected HCT8 cells , with nuclei counterstained by DAPI . Scale bar , 20 μm . ( C ) Percentage of HCT8 cells ( >100 cells from 5–8 random fields ) with Bcl-XL and PAR staining was quantified . Results in ( C ) are expressed as mean and s . e . m . The p values are calculated by Student’s t tests . ( D ) HCT116 cells were transfected with si-NC or si-Sam68 small interference RNAs . 72 hr later , cells were subjected to 10 Gy of γ-irradiation ( IR ) and harvested at the indicated time periods post IR . Whole cell lysates were derived and IB for indicated proteins , with β-actin as a loading control . p-ATM , Ser1981 phosphorylated ATM; p-p65 , Ser536 phosphorylated p65 . DOI: http://dx . doi . org/10 . 7554/eLife . 15018 . 01910 . 7554/eLife . 15018 . 020Figure 6—figure supplement 2 . Sam68 knockdown sensitizes human colon cancer cells to genotoxic stress-induced apoptosis . ( A–C ) HCT116 and HCT8 cells were transfected with nonspecific control ( si-NC ) or Sam68-specific ( si-Sam68 ) small interference RNAs . 72 hr later , HCT116 ( A ) and HCT8 ( B ) cells were stimulated with indicated doses of Camptothecin ( CPT ) for 6 hr and HCT116 cells were γ-irradiated with indicated doses ( C ) . Whole cell lysates were derived and immunoblotted ( IB ) for indicated proteins , with β-actin as a loading control . c-PARP1 , cleaved PARP1; c-Casp3 , cleaved Caspase-3 . The two species of Bcl-XL proteins are labeled by open triangles ( A ) . The full-length and cleaved PARP1 are indicated by a black triangle and a red triangle , respectively ( B , C ) . ( D ) HCT116 cells expressing si-NC or si-Sam68 siRNA were transfected with either FLAG vehicle control or FLAG-IKKβ ( SSEE ) plasmid . 18 hr later , the cells were stimulated with 10 μM of CPT for 6 hr , and whole cell lysates were derived and IB for indicated proteins , with β-actin as a loading control . c-Casp3 , cleaved Caspase-3 . The full-length and cleaved PARP1 are indicated by a black triangle and a red triangle , respectively; the two species of Bcl-XL proteins are labeled by open triangles . The left of this panel was duplicated from the blots in Figure 6F to illustrate the rescuing effect of ectopic expression of IKKβ ( SSEE ) . DOI: http://dx . doi . org/10 . 7554/eLife . 15018 . 020 We further examined whether Sam68 is essential for human colon cancer cell survival , knowing that Sam68 levels were elevated in colon cancer from human patients ( Figure 5G–I ) . Indeed , knockdown of Sam68 by siRNAs significantly reduced PAR levels in human colon cancer-derived HCT8 , HCT116 , and T84 cell lines ( Figure 6C–E and Figure 6—figure supplement 1A–C ) . Moreover , p65 phosphorylation and Bcl-XL expression were substantially attenuated in Sam68 knockdown cancer cells ( Figure 6C–E and Figure 6—figure supplement 1A–C ) . Furthermore , genotoxic stress-dependent NF-κB activation , as indicated by p65 phosphorylation and ATM phosphorylation ( Wu et al . , 2006 ) ( Figure 6—figure supplement 1D ) and subsequent Bcl-XL expression ( Figure 6F ) were both significantly attenuated in Sam68 knockdown cancer cells . These results thus suggest a critical role of Sam68 in DNA damage-initiated and PAR-dependent NF-κB transactivation . Along with reduced anti-apoptotic transcription , Sam68 knockdown sensitized cancer cells to CPT- or γ-irradiation-induced apoptosis , as conveyed by the boosted cleavage of PARP1 and Caspase-3 ( Figure 6F and Figure 6—figure supplement 2A–C ) . Of note , ectopic expression of IKKβ ( SSEE ) , which constitutively activates NF-κB , substantially rescued the attenuated Bcl-XL expression and the DNA damage-induced apoptosis , as evidenced by reduced PARP1 and Caspase-3 cleavage , in Sam68 down-regulated HCT116 cells ( Figure 6—figure supplement 2D ) , which supports that Sam68 knockdown affects genotoxic stress-induced NF-κB transactivation . Consistently , in contrast to controls , Sam68 knockdown triggered colon cancer cells to undergo spontaneous apoptosis and dramatically sensitized cancer cells to CPT- or γ-irradiation-induced cell death , as conveyed by Annexin-V staining ( Figure 6G–H ) . These results thus demonstrate that downregulation of Sam68 lessens colon tumor development in Apcmin716/+ mice and sensitizes human colon cancer cells to genotoxic stress-induced apoptosis , in line with the indispensible role of Sam68 in the nuclear-initiated PARylation , NF-κB activation , and anti-apoptotic transcription in mouse and human colon cancer cells . To ascertain whether Sam68 deletion reduces colon tumor formation as a result of the defect in PARP1 activation and PARylation , we assessed the impact of the PARP inhibitor , Olaparib , on phosphor-p65 and Bcl-XL levels and colon tumor development in Apcmin716/+ mice . 8-week-old Apcmin716/+ mice ( when visible colon adenomas start to form ) were utilized to assess whether inhibiting PAR production and PAR-dependent NF-κB signaling and anti-apoptotic transcription prevents adenoma formation . Of note , five continuous daily intraperitoneal injections of Olaparib , in comparison to vehicle control , substantially reduced PAR production and levels of phosphorylated p65 and Bcl-XL in CECs derived from Apcmin716/+ mice ( Figure 7A–B ) , indicative of the impact of PARP1 inhibition on PAR formation and NF-κB transactivation in vivo . Consistent with our observation ( Figure 6A–B ) , the vehicle-treated Apcmin716/+ mice spontaneously developed adenomas in the cecum and distal colon ( Figure 7C–D ) . In contrast , a 4-week treatment with Olaparib significantly retarded colon tumor development in Apcmin716/+ mice ( Figure 7C–D ) . PARP1 inhibition by Olaparib , compared to the control , substantially reduced the tumor size and tumor load , albeit had no statistically significant impact on tumor number was observed in the tumor-laden Apcmin716/+ mice ( Figure 7E ) . Thus genetic deletion of Sam68 and PARP1 inhibition exhibited similar effects on reducing colon tumor development in Apcmin716/+ mice , which supports the essential roles of Sam68 and PARP1 in mouse tumor growth in vivo . 10 . 7554/eLife . 15018 . 021Figure 7 . PARP1 inhibition reduces colon tumor development in mice and sensitizes human colon cancer cells to undergo apoptosis . ( A ) PARP1 inhibition in Apcmin716/+ mice in vivo . 8-week-old Apcmin716/+ mice were intraperitoneally injected with vehicle control or Olaparib ( 50 mg/kg ) once daily for 5 days , followed by euthanization and further analysis . ( B ) Colon epithelial cells ( CECs ) were isolated from vehicle- or Olaparib-treated Apcmin716/+ mice , treated as in ( A ) , and whole cell lysates were derived and immunoblotted ( IB ) for indicated proteins , with β-actin as a loading control . ( C ) A schematic of the experimental timeline for the impact of PARP1 inhibition on colon tumor development in vivo in Apcmin716/+ mice . 8-week-old Apcmin716/+ mice were intraperitoneally injected vehicle control or Olaparib ( 50 mg/kg , once daily for 5 days × 4 weeks ) . Mice were euthanized to analyze tumor development in the colon . ( D ) Methylene blue ( MB ) staining of the colons derived from 12-week old Apcmin716/+ mice , post 4-week vehicle control or Olaparib treatment , as illustrated in ( C ) . ( E ) Quantification of tumor number , tumor size , and tumor load in the colons from Apcmin716/+ mice treated with vehicle control ( n = 8 ) and Olaparib ( n = 10 ) , following MB staining . ( F ) HCT116 cells were treated with indicated concentration of Olaparib for 72 hr . Whole cell lysates were derived and IB for indicated proteins , with β-actin as a loading control . ( G ) HCT116 cells were treated with Olaparib as in ( F ) , and subjected to flow cytometry analysis of propidium iodide ( PI ) /Annexin V staining . Percentages of apoptotic ( PI- Annexin V+ ) cells as indicated were quantified . Results in ( E and G ) are expressed as mean and s . e . m . ns , non-significant difference; p<0 . 05; p<0 . 01; p<0 . 0001 by Student’s t tests . Data in ( B , D , F , and G ) are representative of at least three independent experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 15018 . 02110 . 7554/eLife . 15018 . 022Figure 7—figure supplement 1 . PARP1 inhibition and down-regulation of PARP1 and NF-κB triggers human cancer cells undergo apoptosis . ( A ) HCT116 cells were treated with indicated concentration of PJ-34 for 72 hr . Whole cell lysates were derived and immunoblotted ( IB ) for indicated proteins , with β-actin as a loading control . The two species of Bcl-XL proteins are labeled by open triangles . ( B ) HCT116 cells , pretreated with 10 μM of Olaparib or vehicle control for 1 hr , were stimulated with ( + ) or without ( − ) 10 μM of Camptothecin ( CPT ) for 6 hr . Whole cell lysates were derived and IB for Bcl-XL , with β-actin as a loading control . The two species of Bcl-XL proteins are labeled by open triangles . ( C ) HCT116 cells treated as in ( A ) , were stained by propidium iodide ( PI ) /Annexin V , followed by flow cytometry analysis . ( D ) HCT8 , HCT116 , and T84 cells were treated with PJ-34 as in ( A ) , and subjected to flow cytometry analysis of PI/Annexin V staining . Percentages of apoptotic ( PI- Annexin V+ ) cells as indicated were quantified from three representative experiments . ( E , F , and G ) HCT116 were transfected with nonspecific control ( si-NC ) , Sam68-specific ( si-Sam68 ) ( E ) , PARP1-specific ( si-PARP1 ) ( F ) , or p65-specific ( si-p65 ) ( G ) small interference RNAs . 72 hr later , whole cell lysates were derived and IB for indicated proteins , with β-actin as a loading control . Right , HCT116 cells transfected as in ( E , F , and G ) were stained by PI/Annexin V , followed by flow cytometry analysis . Percentages of apoptotic ( PI- Annexin V+ ) cells were quantified from three representative experiments . Results in ( D–G ) are expressed as mean and s . e . m . , p<0 . 001 by Student’s t tests . DOI: http://dx . doi . org/10 . 7554/eLife . 15018 . 022 We sought to examine whether down-regulation of PARP1 or NF-κB transactivation executes a similar function as down-regulation of Sam68 on human colon cancer cell survival . Similar to Sam68 knockdown ( Figure 6C–E and Figure 6—figure supplement 1 ) , PARP1 inhibition by Olaparib or PJ-34 treatment significantly reduced the basal PAR , phospho-p65 , and Bcl-XL levels in HCT116 cells ( Figure 7F and Figure 7—figure supplement 1A ) as well as genotoxic stress-induced Bcl-XL expression ( Figure 7—figure supplement 1B ) . In line with the attenuated NF-κB activation and anti-apoptotic transcription , Olaparib or PJ-34 treatment triggered colon cancer cells to undergo spontaneous apoptosis , as conveyed by Annexin-V staining ( Figure 7G and Figure 7—figure supplement 1C–D ) , which mirrors the impact of Sam68 knockdown on the survival of these cell lines ( Figure 6G–H ) . Moreover , PARP1 knockdown by siRNAs , similar to Sam68 knockdown ( Figure 7—figure supplement 1E ) , substantially reduced PAR production and levels of phosphorylated p65 and Bcl-XL in HCT116 cells ( Figure 7—-figure supplement 1F ) . In contrast , p65 knockdown attenuated only the levels of phosphorylated p65 and Bcl-XL , without affecting PAR production ( Figure 7—figure supplement 1G ) , supporting the PAR-dependent NF-κB transactivation of anti-apoptotic genes . Importantly , down-regulation of PARP1 or p65 sensitized colon cancer cells to undergo spontaneous apoptosis ( Figure 7—figure supplement 1F–G ) , which mirrors the effects of PARP1 inhibition ( Figure 7G and Figure 7—figure supplement 1C–D ) and Sam68 knockdown ( Figure 6G–H and Figure 7—figure supplement 1E ) . Therefore , our results demonstrate the Sam68-PARP1-NF-κB-anti-apoptotic gene axis plays a crucial function for colon cancer survival . The NF-κB signaling pathway remains a very attractive avenue for pharmacological intervention , given its crucial function in human health and disease , particularly inflammatory diseases and cancers . In spite of the widespread use of chemotherapy and radiotherapy in current-day cancer treatments , the genotoxic stress-induced nuclear NF-κB signaling pathway that leads to NF-κB transactivation is still less defined , than NF-κB activation initiated from cell membrane stimuli ( e . g . immune receptors ) . Herein , we report that Sam68 is a novel regulator participating in the early cellular responses to DNA damage; it does this by orchestrating the signaling cascade that links DNA lesion recognition in the nucleus to NF-κB liberation and activation in the cytoplasm ( Figure 8 ) . Following genotoxic stress , sophisticated cellular networks consisting of a variety of molecules and post-translational modifications , collectively termed as DNA damage responses ( DDR ) , are crucial for cell-cycle checkpoint control , DNA repair , transcription , and apoptosis ( Jackson and Bartek , 2009 ) . Among these , DNA damage-initiated NF-κB signaling and transactivation of an array of anti-apoptotic molecules are pivotal in facilitating cells to 'escape' from the lethal effects of DNA damage ( McCool and Miyamoto , 2012; Miyamoto , 2011 ) . Our results demonstrate that Sam68 deletion impairs inducible IκBα degradation , NF-κB liberation , and subsequent anti-apoptotic gene expression in cells under genotoxic stress . Sam68 interacts with several established molecules including PARP1 and IKKγ in the nuclear-initiated NF-κB signaling pathway and facilitates assembly of the NF-κB activation signaling complex . More importantly , Sam68 plays an indispensable role in DNA damage-triggered PARP1 activation and PAR synthesis , thus controlling the PAR-dependent signaling complex assembly . This is distinct from leukemia related protein 16 ( LRP16 ) , a recently reported regulator of DNA damage-induced NF-κB activation , which binds to PARP1 and IKKγ in a PAR-dependent manner and therefore functions downstream of PARP1 and PAR production ( Wu et al . , 2015 ) . Previously , PARP1 was proposed to be dispensable for ATM activation ( Stilmann et al . , 2009 ) . Our results suggest that Sam68 is important for the activation of PARP1 , as well as , ATM in DDR . In addition to the currently illustrated major impact of Sam68 on the PARP1/PAR-dependent signaling pathway that leads to NF-κB activation , Sam68 may contribute to NF-κB activation via a direct or indirect effect on the ATM-involved signaling cascade during the cellular response to DNA damage . Our results underscore a previously unknown function of Sam68 , a versatile protein that preferentially resides in the nucleus , in the early nuclear signaling cascade following DNA damage . Furthermore , we previously reported that Sam68 is important for the promoter selectivity and transcriptional specificity of NF-κB in the nucleus ( Fu et al . , 2013 ) . Sam68 could work at both initiating of NF-κB activation and controlling the NF-κB transactivation potential in the DNA damage-induced NF-κB signaling pathway; thus playing a critical role in the genotoxic stress-initiated “nuclear to cytoplasmic to nuclear” NF-κB activation . Of note , tumor necrosis factor/tumor necrosis factor receptor ( TNF/TNFR ) signaling was revealed to be important for a feed-forward response to DNA damage , supported by DNA damage-induced phosphorylation of several well-known components of TNF/TNFR signaling pathway including TRAF2 , p62 , RIP1 , and CYLD ( Beli et al . , 2012 ) . Moreover , Sam68 has been proposed to be crucial for the recruitment of RIP1 to the TNF receptor in TNF-triggered NF-κB signaling ( Ramakrishnan and Baltimore , 2011 ) . The involvement of Sam68 in both TNF/TNFR signaling and genotoxic stress-induced NF-κB signaling indicates that it could be a crucial molecule conferring the crosstalk between TNF/TNFR signaling pathway and DNA damage responses . 10 . 7554/eLife . 15018 . 023Figure 8 . Schematic model representation of Sam68 functioning as an early signaling molecule in genotoxic stress-initiated NF-κB signaling pathway . DOI: http://dx . doi . org/10 . 7554/eLife . 15018 . 023 We demonstrated that , compared to Khdrbs1+/- cells , Khdrbs1-/- CECs are hypersensitive to genotoxic stress , since Sam68 deletion abolishes the DNA damage-initiated NF-κB signaling and attenuates the inducible expression of anti-apoptotic genes . These results , along with the previous report that NF-κB-mediated anti-apoptotic transcription executes a protective function to CECs following γ-irradiation ( Egan et al . , 2004 ) , highlight the pathophysiological relevance of the nuclear-initiated NF-κB signaling and transactivation in colonic cell survival from environmental acute DNA damage . Of note , nuclear-initiated NF-κB signaling plays a key role in cellular responses to intrinsic DNA damage , especially damage that occurs during rapid DNA replication and proliferation in cancer cells ( Hayakawa et al . , 2009; Horst et al . , 2009; Kim et al . , 2010; Koh et al . , 2011; Kojima et al . , 2004; Onizawa et al . , 2009; Pouyet et al . , 2010; Shaked et al . , 2012; Stark et al . , 2007; Wan et al . , 2011; Williams et al . , 2008 ) . Aberrant NF-κB activation and elevated expression of NF-κB target genes , in particular those encoding anti-apoptotic molecules , have been acknowledged as key factors facilitating colon cancer survival and development ( Song et al . , 2014; Townson et al . , 2003; Zubair and Frieri , 2013 ) . Here we report that the levels of Sam68 , phospho-p65 ( indicative of NF-κB activation ) , anti-apoptotic molecules Bcl-XL and XIAP are all elevated in colon tumors in comparison to adjacent normal tissue derived from genetically susceptible Apcmin716/+ mice and human colon cancer patients . Of note , the positive correlation between Sam68 levels and PAR levels , as well as phospho-p65 levels in human colon cancers suggests that Sam68 could be essential for the development and survival of colon cancer , considering the pivotal function of Sam68 in orchestrating the intrinsic DNA damage-initiated NF-κB signaling and transactivation . In support of this notion , knockdown of Sam68 , PARP1 , and p65 in human colon cancer cells significantly reduces the basal and genotoxic stress-induced PAR production , NF-κB activation , and expression of anti-apoptotic molecules Bcl-XL and XIAP , thus leading to spontaneous and DNA damage-induced apoptosis in Sam68-downregulated colon cancer cells . Moreover , genetic deletion of Sam68 and inhibition of PARP1 markedly reduces the development and survival of colon tumors in Apcmin716/+ mice , further supporting the pivotal role of Sam68-conferred PAR-dependent NF-κB activation in colon tumorigenesis . In spite of the crucial role of PARP1 in DNA damage-induced NF-κB activation , DNA repair , and other cellular responses , the precise mechanisms of the activation and regulation of PARP1 remains elusive . We show here that Sam68 deficiency significantly attenuates DNA damage-induced PARP1 activation and PAR production , which suggests that Sam68 , as an early signaling regulator , governs the genotoxic stress-stimulated PARP1 activity . In particular , the PAR-dependent NF-κB signaling cascade is dampened in Sam68 deleted cells , as well as PARP1 knockout cells ( Stilmann et al . , 2009 ) . Moreover , the reduction in anti-apoptotic gene expression and increase in genotoxic stress-induced apoptosis are observed in Sam68 knockout cells and PARP1 knockout cells ( Stilmann et al . , 2009 ) , in line with the impeded PAR-dependent NF-κB signaling in these cells . Furthermore , Sam68 knockout and PARP1 inhibition both attenuates colon tumor development in Apcmin716/+ mice . Such similarity in the phenotypes of Sam68- and PARP1-deficient cells and animals in response to genotoxic stresses further supports the notion that Sam68 is a crucial regulator of PARP1 in cellular response to genotoxic stress . Elevated Sam68 levels correlate with tumor progression and poor prognosis in multiple cancer patients and overexpression of Sam68 has been proposed as a prognostic marker ( Chen et al . , 2012; Liao et al . , 2013; Song et al . , 2010; Zhang et al . , 2009 ) ; however , the significance of Sam68 in tumorigenesis is still obscure . Here we report that Sam68 knockdown markedly sensitizes colon cancer cells to genotoxic stress-induced cell death and Sam68 knockout substantially retards colon tumor burden and survival in Apcmin716/+ mice , which highlights the pivotal function of Sam68 in tumor development and survival . Importantly , we establish proof-of-concept showing that manipulation of Sam68 sensitizes colon cancer to DNA damage-induced apoptosis . As a key early signaling regulator at the proxy of the nuclear-initiated NF-κB signaling pathway , Sam68 could provide a novel target for therapeutics for cancers and other human diseases associated with impaired DNA damage responses . The human patient study was approved by the Johns Hopkins Institutional Review Board . All samples were obtained in accordance with the Health Insurance Portability and Accountability Act ( HIPAA ) . All animal experiments were performed according to protocol number MO13-H349 , approved by the Johns Hopkins University’s Animal Care and Use Committee and in direct accordance with the NIH guidelines for housing and care of laboratory animals . Colon tumors ( adenomas and cancers ) and paired normal tissues were collected from patients undergoing surgery at Johns Hopkins Hospital , as described previously ( Dejea et al . , 2014 ) . Patients who received pre-operative radiation and/or chemotherapy or with a personal history of colitis-associated colon cancer were excluded . Khdrbs1-/- ( Sam68 knockout ) mice and their gender-matched littermate Khdrbs1+/- ( Sam68 heterozygote ) mice ( occasionally substituted with gender-matched littermate Khdrbs1+/+ [Sam68 wild-type] mice when Khdrbs1+/- ones were lacking , and referred as Khdrbs1+/- alone for simplicity ) were produced using heterozygous breeding pairs , as previously described ( Huot et al . , 2012 ) . Apcmin716/+ mice expressing a mutant gene encoding an adenomatous polyposis coli protein truncated at amino acid 716 were described previously ( Su et al . , 1992; Wu et al . , 2009 ) . Mice were maintained in a specific pathogen-free facility and fed autoclaved food and water ad libitum . The following mouse embryonic fibroblasts ( MEFs ) were obtained from other institutions: wild-type and Sam68 knockout ( KO ) MEFs ( Richard et al . , 2005 ) from Stephan Richard ( McGill University , Canada ) and PARP1 KO MEFs ( Tong et al . , 2001 ) from Zhao-Qi Wang ( Fritz Lipmann Institute , Germany ) , respectively . HEK293T , HCT8 , HCT116 , and T84 cell lines were purchased from ATCC ( Manassas , VA ) and the identities have been authenticated by short tandem repeat DNA profiling . All cells described above were regularly tested for mycoplasma contamination . Cells were cultured in DMEM medium containing 10% fetal calf serum , 2 M glutamine , and 100 U/ml each of penicillin and streptomycin . Antibodies used were: IκBα , Sam68 , p65 , IKKγ , GFP , PARP1 , PARP2 , GST from Santa Cruz Biotechnology ( Dallas , TX ) ; β-actin from Sigma-Aldrich ( St . Louis , MO ) ; PAR from Trevigen ( Gaithersburg , MD ) ; ATM , phospho-ATM , PARP1 , phospho-p65 , cleaved Caspase-3 , Caspase-3 , Ku70 , Histone H3 from Cell Signaling Technology ( Danvers , MA ) ; XIAP form BD Biosciences ( San Jose , CA ) ; α-tubulin from EMD Millipore ( Billerica , CA ) ; Bcl-XL , cIAP1 , NBS1 from GeneTex ( Irvine , CA ) ; SUMO1 , kindly provided by Dr . M . Matunis ( Johns Hopkins University ) . 4-[ ( 3-[ ( 4-cyclopropylcarbonyl ) piperazin-4-yl]carbonyl ) -4-fluorophenyl]methyl ( 2H ) phthalazin-1-one ( Olaparib ) and N- ( 6-oxo-5 , 6-dihydrophenanthridin-2-yl ) -N , N-dimethylacetamide-HCl ( PJ-34 ) were purchased from Fisher Scientific ( Pittsburgh , PA ) and Enzo Life Sciences ( Farmingdale , NY ) , respectively . Camptothecin ( CPT ) , MG132 , and 4' , 6-diamidino-2-phenylindole ( DAPI ) were obtained from Sigma-Aldrich . Recombinant PARP1 protein was obtained from Trevigen . The FLAG , FLAG-IKKβ ( SSEE ) , GFP , GFP-Sam68 , GFP-Sam68 ( ΔC ) , GFP-Sam68 ( ΔN ) , GFP-Sam68 ( ΔKH ) , GST , GST-Sam68 , and GST-Sam68 ( ΔN ) constructs were described previously ( Fu et al . , 2013 ) . Mouse Sam68 siGENOME SMARTpool siRNA ( catalog number M-065115-01 ) was purchased from Thermo Scientific ( Waltham , MA ) . Human Sam68 and p65 siRNAs were described previously ( Fu et al . , 2013 ) . Human PARP1 and PARG siRNAs were purchased from Santa Cruz Biotechnology . Transient transfection of siRNA or plasmids into MEFs was performed using Lipofectamine 2000 or Lipofectamine RNAiMAX ( Life Technologies , Grand Island , NY ) according to the manufacturer's instructions . Subcellular fractionation was performed by differential centrifugation as previously described ( Wan et al . , 2007 ) . EMSAs were carried out as described ( Wan et al . , 2007 ) , and the reaction mixture was resolved on 6% DNA retardation gel ( Life Technologies ) in 0 . 25 × TBE buffer , and dried gels were visualized in a Fujifilm image reader FLA-7000 ( Fujifilm Life Science , Valhalla , NY ) . Immunoprecipitation and immunoblot assays were conducted as previously described ( Hodgson et al . , 2015 ) . In brief , cells were harvested and lysed on ice by 0 . 4 ml of lysis buffer ( 50 mM Tris-HCl [pH 8 . 0] , 150 mM NaCl , 1% NP-40 and 0 . 5% sodium deoxycholate , 1 × complete protease inhibitor cocktail [Roche Applied Science , Indianapolis , IN] ) for 30 min . The lysates were centrifuged at 10 , 000 ×g at 4°C for 10 min . The protein-normalized lysates were subjected to immunoprecipitation by adding 10 mg/ml of the appropriate antibody , 30 μl of protein G-agarose ( Roche Applied Science ) , and rotating for more than 2 hr in the cold room . The precipitates were washed at least four times with cold lysis buffer followed by a separation by SDS-PAGE under reduced and denaturing conditions . The resolved protein bands were transferred onto nitrocellulose membranes and probed by the Super Signaling system ( Thermo Scientific ) according to the manufacturer's instructions , and imaged using a FluorChem E System ( Protein Simple , Santa Clara , CA ) . Immunofluorescence microscopy was performed as previously described ( Hodgson et al . , 2015 ) . Briefly , cells were fixed with 4% paraformaldehyde in PBS and then mounted onto slides by Cellspin . After a 5-min permeabilization with 0 . 05% Triton X-100 in PBS , the fixed cells were stained with appropriate primary antibodies for 1 hr , and FITC- or AlexaFluor 594-conjugated secondary antibodies ( Life Technologies ) for 1 hr together with 1 µg/ml of DAPI for 5 min at 25°C . The slides were then rinsed with PBS three times and cover mounted for fluorescence microscopy . Cells were harvested at indicated time points after γ-irradiation , and cell pellets were resuspended in the NETN buffer ( 20 mM Tris–HCl [pH 8 . 0] , 100 mM NaCl , 1 mM EDTA , and 0 . 5% NP-40 ) and incubated on ice for 20 min . Supernatant after 3000 ×g for 10 min were collected as soluble fraction . Pellets were recovered and resuspended in 0 . 2 M HCl on ice for 30 min , and sonicated for 10 sec to release chromatin-bound proteins , and then the soluble fractions were neutralized with 1 M Tris–HCl ( pH 8 . 5 ) and collected as chromatin fraction , and the pellets were collected as insoluble fraction for further analysis , as described previously ( Liu et al . , 2013; Wu et al . , 2011 ) . Total RNA was isolated using Trizol reagent ( Life Technologies ) and treated with the TURBO DNA-free Kit ( Life Technologies ) to remove residual genomic DNA . Complementary DNA was synthesized using qScript cDNA SuperMix Kit ( Quanta Biosciences , Gaithersburg , MD ) according to the manufacturer's instructions . Gene specific products were amplified using MyTaq Rad Mix ( Bioline USA , Taunton , MA ) in a multiple conventional and gradient Veriti Thermal Cycler ( Life Technologies ) with the following primers: Birc3-f , 5'-GAAACCATTTGGCGTGTTCT-3'; and Birc3-r , 5'-TGGATCGCAATGATGATGTC -3'; Bcl2l1-f , 5'-AATGAACTCTTTCGGGATGGAG-3’; and Bcl2l1-r , 5'- CCAACTTGCAATCCGACTCA-3’; Xiap-f , 5'-CCATGTGTAGTGAAGAAGCCAGAT-3'; and Xiap-r , 5'-TGATCATCAGCCCCTGTGTAGTAG -3'; Actb-f , 5'-CACATCAAGAAGGTGGTG-3'; and Actb-r , 5'-TGTCATACCAGGAAATGA-3' . In vitro PARylation assays using recombinant His-PARP1 or immunoprecipitated endogenous PARP1 from MEFs were performed as previously described ( Zaniolo et al . , 2007 ) . Briefly , PARP1 protein or immunoprecipitant was incubated for 20 min or 2 min at 30°C with GST or GST-Sam68 in a standard assay mixture containing 100 mM Tris-HCl ( pH 8 . 0 ) , 10 mM MgCl2 , 10% ( v/v ) glycerol , 1 . 5 mM DTT , 10 μg/ml activated DNA ( sonicated ) and 200 μM NAD+ . The reaction was terminated by the addition of SDS sample buffer ( Life Technologies ) , and the boiled samples were subjected to SDS-PAGE . When indicated , the PARP inhibitor PJ-34 was added to the reaction mixture at a final concentration of 1 μM for 15 min prior to the reaction . Colonic epithelial cells ( CECs ) were isolated from mice as previously described ( Hodgson et al . , 2015 ) . Briefly , after euthanizing mice , the entire colon was removed under aseptic conditions and washed twice with ice-cold PBS . After dividing the colon into 2–3 mm long fragments and transferring them into chelating buffer ( 27 mM trisodium cirtcrate , 5 mM Na2HPO4 , 96 mM NaCl , 8 mM KH2PO4 , 1 . 5 mM KCl , 0 . 5 mM DTT , 55 mM D-sorbitol , 44 mM sucrose , 6 mM EDTA , 5 mM EGTA [pH 7 . 3] ) for 15 min at 4°C , CECs were then dislodged by repeated vigorous shaking . Tissue debris was removed by a 70-μm cell-strainer ( Fisher Scientific , Suwanee , GA ) and CECs were harvested by centrifugation at 4°C . The viability of CECs was confirmed by trypan blue staining and isolated CECs were cultured at 37°C for 1 hr for recovery , followed by indicated treatment . The γ-irradiation on primary mouse cells and cell lines were performed using a 137Caesium source ( dose rate 8 Gy/min ) . After euthanizing mice , the colons were removed under aseptic conditions , washed once with ice-cold PBS , the terminal 0 . 5-cm piece of the colon was fixed in 10% buffered formalin for 24 hr , embedded in paraffin and 5-micron sections were cut and processed for Hematoxylin and Eosin ( H&E ) staining . For immunohistology , after euthanizing mice , the entire colons were excised under aseptic conditions and frozen in optimal cutting temperature ( O . C . T . ) media ( Tissue-Tek , Elkhart , In ) . 5-micron frozen sections were cut using a Microm HM 550 Cryostat ( Thermo Scientific ) , collected on coated slides , fixed in paraformaldehyde , washed with PBS , and blocked with appropriate sera in PBS . After incubating with appropriate antibodies , sections were washed and incubated with fluorescence dye-conjugated second antibodies and 1 µg/ml of DAPI . Stained sections were washed and mounted under a coverslip using Fluoro-gel with Tris Buffer ( Electron Microscopy Sciences , Hatfield , PA ) and examined using an Axio Observer fluorescence microscope ( Zeiss , Oberkochen , Germany ) . The visualization and quantification of colon adenomas in mice were conducted as previously described ( Wu et al . , 2009 ) . Briefly , mice were sacrificed at 3 months of age . Colon tissue was excised , cleaned with cold PBS , opened longitudinally , fixed in 10% neutral buffered formalin ( 3 . 7% formaldehyde , 1 . 2% methanol , 6 . 5 g/l sodium phosphate dibasic , 4 . 0 g/l sodium phosphate monobasic ) at 25°C overnight , and stained with 0 . 2% ( w/v ) methylene blue solution . The adenomas were quantified and sized under dissecting scope . Average tumor size and tumor load per individual mouse were determined by averaging diameters of all tumors present and summing the areas of all tumors presented in a given mouse , as previously described ( Grivennikov et al . , 2012 ) . All statistical analysis was performed using GraphPad Prism version 6 . 0 ( GraphPad Software , La Jolla , CA ) . Standard errors of means ( s . e . m . ) were plotted in graphs . Significant differences were considered: ns , non-significant difference; at p<0 . 05; at p<0 . 01; * at p<0 . 001; **** at p<0 . 0001 by unpaired Student’s t-test .	Cells use signaling pathways to detect and respond to harmful conditions by switching on genes that keep the cell healthy . One important pathway is the nuclear factor kappa B ( NF-κB ) signaling pathway , which is activated by many stimuli . These stimuli may come from infections from outside the cell or may originate inside the cell , as seen for DNA damage caused by irradiation , chemicals or rapid DNA replication in cancer cells . Most of a cell’s DNA is located in the cell nucleus . However , NF-κB proteins are normally located outside the nucleus , in the cell’s cytoplasm . Damage to DNA triggers a signal from the nucleus to the cytoplasm . This signal activates the NF-κB proteins , which move into the nucleus and turn on genes that help the cell to recover from the damage . These genes include those that prevent the cell from self-destructing . In one step of the NF-κB activation process , chain-like molecules called polymers are made from a compound called poly ( ADP-ribose ) , or PAR for short . However , few other details are known about how the damaged DNA in the nucleus signals to the cytoplasm . A protein called Sam68 , which is found in the cell nucleus , has been linked to DNA damage signaling . Fu , Sun et al . now present evidence that suggests that if mouse cells lack Sam68 , they do not produce PAR polymers in response to DNA damage . In addition , these cells could not trigger the PAR-dependent signaling cascade that is essential for activating NF-κB and for turning on the protective genes . Consequently , cells that lacked Sam68 were extremely sensitive to agents that cause DNA damage , such as chemicals and irradiation . The NF-κB pathway is regulated incorrectly in some cancers , but is also activated by DNA damage caused by cancer treatments . Therefore , Fu , Sun et al . also explored the role of Sam68 in cancer . Reducing the levels of Sam68 made human colon cancer cells more likely to self-destruct when they were exposed to DNA-damaging agents . Furthermore , removing Sam68 from mice that spontaneously grow colon cancer caused their tumors to develop more slowly than mice that retained Sam68 in their cells . Overall , the findings presented by Fu , Sun et al . suggest that Sam68 regulates the signal from the nucleus to the cytoplasm that activates NF-κB proteins in response to DNA damage . Sam68 also appears to be important for helping colon cancer cells grow and survive . Future challenges will be to understand how Sam68 regulates the production of the PAR polymer in this response and to explore whether Sam68 can be targeted for treating cancer .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "cancer", "biology" ]	2016	Sam68/KHDRBS1 is critical for colon tumorigenesis by regulating genotoxic stress-induced NF-κB activation
Single-cell RNA-sequencing is revolutionising our understanding of seemingly homogeneous cell populations but has not yet been widely applied to single-celled organisms . Transcriptional variation in unicellular malaria parasites from the Plasmodium genus is associated with critical phenotypes including red blood cell invasion and immune evasion , yet transcriptional variation at an individual parasite level has not been examined in depth . Here , we describe the adaptation of a single-cell RNA-sequencing ( scRNA-seq ) protocol to deconvolute transcriptional variation for more than 500 individual parasites of both rodent and human malaria comprising asexual and sexual life-cycle stages . We uncover previously hidden discrete transcriptional signatures during the pathogenic part of the life cycle , suggesting that expression over development is not as continuous as commonly thought . In transmission stages , we find novel , sex-specific roles for differential expression of contingency gene families that are usually associated with immune evasion and pathogenesis . Malaria is caused by unicellular eukaryotic parasites from the Plasmodium genus . These organisms have a complex life cycle comprising many different developmental stages . In the blood of infected patients , asexual intra-erythrocytic replication of parasites is solely responsible for pathogenesis , whilst sexual stages , termed gametocytes , are the only stage capable of transmission to the next host via the mosquito vector . These distinct life stages have been extensively investigated using transcriptomic approaches ( Otto et al . , 2010; Bozdech et al . , 2003a; López-Barragán et al . , 2011; Llinás et al . , 2006; Hall et al . , 2005; Lasonder et al . , 2016; Otto et al . , 2014 ) , but this has been largely at a population level . Little is known about how individual cells vary within stages . Single-cell RNA-seq ( scRNA-seq ) produces transcriptomic profiles for multiple individual cells . This has allowed the decomposition of cell populations ( Haber et al . , 2017 ) , uncovered previously unknown cell types ( Grün et al . , 2015 ) and enhanced our understanding of developmental pathways ( Mohammed et al . , 2017 ) . Several scRNA-seq methods with different attributes have now been described ( Ziegenhain et al . , 2017 ) , with some providing depth – a good representation of full length transcripts ( Picelli et al . , 2013 ) from tens or hundreds of cells – and others providing breadth , with poorer representation of transcriptomes but from a much greater number of cells ( Macosko et al . , 2015 ) . scRNA-seq promises powerful new examinations of unicellular organisms , especially those that are difficult to obtain in large numbers or are not amenable to in vitro cultivation . A number of important questions in malaria biology will benefit from single-cell technology . For instance , what are the transcriptional switches in individual parasites that drive phenotypes such as commitment to the sexual development pathway ( Sinha et al . , 2014; Kafsack et al . , 2014 ) , parasite sequestration ( Tembo et al . , 2014 ) and immune evasion ( Scherf et al . , 2008 ) . A recent study ( Poran et al . , 2017 ) demonstrated the use of a high-throughput , low-coverage scRNA-seq technique ( Drop-seq [Macosko et al . , 2015] ) to identify a signature of sexual commitment in Plasmodium . Here , we use a lower throughput ( fewer cells ) , but higher coverage ( both more genes detected and more of each gene’s length detected via full-length transcript sequencing ) approach to examine transcriptional dynamics of the parasite during the blood stages in both the most popular rodent model parasite ( P . berghei ) and the most deadly human malaria parasite ( P . falciparum ) . We show that this method is highly effective at capturing transcriptional variation associated with different parasite stages and cell cycle states , and we also uncover previously unknown aspects of the parasite's progression through its asexual cycle and in its sexual stages . The greatest coverage of genes in mammalian cells using scRNA-seq has been achieved with the Smart-Seq2 protocol ( Picelli et al . , 2013 ) . In this method , cells are sorted by FACS into individual wells , followed by full-length cDNA generation using a viral reverse transcriptase . This mediates the addition of a triple cytosine overhang to the 3′ end to the first strand cDNA that allows the annealing of a strand switching oligonucleotide for second strand synthesis and direct cDNA amplification by PCR . This plate-based approach tends to result in detection of more transcripts from more genes than other approaches ( Svensson et al . , 2017 ) . Furthermore , it is a full-length transcript method , providing information about transcript structure , allowing deconvolution of splicing variants and inference about the strand of origin ( Wu et al . , 2015 ) . Initially , we trialled the standard version of the Smart-seq2 protocol ( Picelli et al . , 2013 ) on sorted , Plasmodium falciparum-infected single red blood cells ( Figure 1A ) , adjusting only the number of PCR cycles ( 30 rather than 18 ) to account for the relatively low RNA content of protozoan cells . However , on average , only 10% of reads mapped to genes in the parasite genome and more than half of these mapped to rRNA genes ( Figure 1B ) . To improve yield , we tested the impact of: removing the anchoring base from the oligo ( dT ) and varying length of the oligo ( dT ) primer ( 20 vs 30 ) ; changing the reverse transcription enzymes ( SuperScriptII , SuperScriptIV , SMARTMMLV , and SmartScribe ) ; and varying the number of amplification cycles ( 25 or 30 ) . We generated libraries for pools of 10 sorted late stage P . falciparum cells and tested the abundance of transcripts from the msp-1 gene by quantitative RT-PCR . A longer , unanchored oligo ( dT ) primer ( T30 ) significantly improved yield and SuperScript II and SMARTScribe were the highest yielding reverse transcriptases ( Figure 1C ) . Amplification for 25 and 30 cycles appeared to give equivalent results ( Figure 1C ) . To understand the impact of these permutations on transcriptome sequence complexity , we sorted individual P . falciparum cells and generated single-cell transcriptome libraries using the dT30 oligo , either the SuperScript II or SmartScribe enzymes and either 25 or 30 cycles of PCR ( Figure 1D ) . Significantly more genes were detected , with dramatically reduced rRNA contamination , using the SMARTScribe enzyme ( Figure 1D; Table 1 ) . Given equivalent results for 25 or 30 cycles , we opted to use the lower number of cycles for all subsequent experiments . Two potential sources of contamination are important to consider in scRNA-seq experiments . First , single-sorted cells could actually comprise multiple cells , resulting in a hybrid signal that adds noise to downstream analyses . Second , ambient RNA from lysed cells in the cell suspension could be transferred along with intact cells into each well . To evaluate these potential sources of contamination , we flow-sorted individual parasites from a mixture of fluorescently-labelled GFP P . falciparum ( Pf ) and mCherry P . berghei ( Pb ) late-stage parasites into a 96-well plate ( Figure 2—figure supplement 1 ) . We then prepared and sequenced transcriptome libraries for each cell . The reads were mapped to a combined reference of both genome sequences . No evidence for doublet events was found ( Figure 2A ) and , for each cell , the vast majority of reads ( 98 . 1% for P . berghei , 99 . 4% for P . falciparum ) mapped uniquely to the genome of the expected species ( Figure 2A ) . The few transcripts that mapped to the wrong genome were those most highly expressed in the other species and most likely to be picked up from the solution ( Figure 2—figure supplement 1B , C ) . A very low number of individual ambient transcripts were detected ( Figure 2—figure supplement 1D , E ) . Only 15 of 3566 transcripts detected in P . berghei cells were from P . falciparum , and none of these were differentially expressed , suggesting they will not affect our downstream analysis . Having established the reliability of the protocol , we generated 188 single-cell transcriptomes of mixed asexual and sexual ( gametocyte ) blood-stage parasites of the rodent malaria model P . berghei . After filtering to remove transcriptomes with fewer than 25 , 000 total reads and fewer than 1000 detected genes ( with at least one read ) , 144 high-quality transcriptomes remained . We then removed genes unless they had at least ten reads in each of five or more cells . In total , we detected expression of 4579 genes: over 90% of genes in the P . berghei genome . From each cell , we identified expression from , on average , 1981 genes ( ~33% ) , similar to the proportion of transcriptomes detected in mammalian single-cell experiments ( Treutlein et al . , 2014 ) ( Figure 2B ) . We also generated single-cell transcriptomes for the human malaria parasite P . falciparum and processed them using the same filtering procedure as for P . berghei . This resulted in 191 high-quality single-cell transcriptomes ( of 237 total ) for sexual stages , with an average 2090 genes detected , and 161 high-quality single-cell transcriptomes ( of 174 total ) for asexual stages , with an average 1712 genes detected ( Figure 2B ) . We used the P . berghei dataset to explore biases in the representation of transcripts sequenced with our protocol . First , we checked if some regions were overrepresented amongst our transcript sequences due to preferential amplification of less AT-rich sequences by PCR . Second , because the reverse transcriptase ought to process a complete mRNA in order to produce cDNA , we determined whether there was a bias against long genes . In fact , neither GC content ( Figure 2—figure supplement 2 ) nor gene length ( Figure 2C ) had an impact on transcript detection . In the case of many long genes , the lack of a length-bias could be due to the sequencing of mRNA fragments , rather than full-length sequences . This suggests that the Smart-seq2 protocol is susceptible to internal priming by oligo-d ( T ) ( as described in [Nam et al . , 2002] ) and template-switching at the exposed 5’ ends of mRNA fragments . The benefit of this is that we are able to assay transcription levels of long and short genes with similar accuracy . Many RNA-seq approaches display a signal bias towards the 5’ or 3’ end of transcripts and in our data , a slight 5’ bias was detected that might also reflect binding of oligo ( dT ) to internal polyA-rich regions of transcripts ( Figure 2D ) . Having developed and assessed our protocol for sequencing single-cell transcriptomes , we next determined whether different parasite stages could be resolved among the 144 P . berghei mixed blood stage transcriptomes . Using a combination of Principal Components Analysis ( PCA ) , k-means clustering using SC3 ( Kiselev , 2016 ) , and comparison to bulk transcriptome datasets ( Otto et al . , 2014; Hoo et al . , 2016 ) , we classified each cell as male , female , or asexual ( Figure 3A ) . Classification of cells is an important step in the analysis of single-cell transcriptome data but classifying all cells in a particular dataset can be a challenge . For Plasmodium , the availability of a variety of published bulk RNA-seq and microarray datasets enabled us to determine the approximate life stage of each cell . For P . berghei , we used a microarray dataset ( Hoo et al . , 2016 ) that examined the 24 hr asexual cycle at 2-hr intervals and an RNA-seq dataset ( Otto et al . , 2014 ) that included samples at three asexual timepoints ( rings , trophozoites and schizonts ) as well as mixed sex gametocytes . For each cell , we compared the list of genes ranked by expression level to those of each sample from the above data sets , picking the best-correlated time point . Male and female gametocytes were differentiated by examining marker genes from cell clusters made using SC3 ( Kiselev et al . , 2017 ) . We established a manually annotated consensus classification for each cell based on the above analyses . Some cells appeared to have intermediate transcriptomes between asexuals and gametocytes and these were labelled as outliers . These may result from co-infected individual red blood cells . The accuracy of our classification was strongly supported by established stage-specific markers ( Figure 3B; Figure 3—figure supplement 1 ) . Moreover , the confirmed absence of contaminating parasites of other life-cycle stages enabled us to determine a new , longer list of stage-specific markers ( Supplementary file 1 ) . We conducted similar analyses for two P . falciparum samples composed of asexual and sexual stages . Because they originated from two distinct pure samples , their classification was more straightforward and both sets of cells ( asexual and sexual ) correlated as expected with previously published bulk datasets ( Otto et al . , 2010; López-Barragán et al . , 2011; Lasonder et al . , 2016 ) ( Figure 3—figure supplement 2 ) . Plasmodium asexual development is replicative , yet it does not follow canonical eukaryotic cell cycle progression and although checkpoints are believed to exist , they have not been characterized ( Gerald et al . , 2011 ) . Bulk RNA-seq studies monitoring transcriptional patterns along the complete asexual cycle of both human and rodent malaria parasite species have consistently revealed a continuous cascade of transcription initiation ( Hoo et al . , 2016; Bozdech et al . , 2003b ) similar to that seen in other eukaryotes ( Spellman et al . , 1998 ) . Although these analyses have used synchronised parasite populations that allow reasonably tight windows of expression to be assayed , their resolution has been limited by surveying pools of cells within each expression window that can differ in developmental progression by several hours . Single-cell RNA-seq allows unsynchronised populations to be sampled , from across large parts of the cycle , and the order of cells in the cycle to be identified using pseudotime analysis ( Trapnell et al . , 2014 ) . Pseudotime analysis orders cells into developmental trajectories by identifying cells with transcriptomes that are most similar to each other and placing those closest to each other in order . To reconstruct the latter part of the asexual development cycle , we first used M3Drop ( Andrews and Hemberg , 2016 ) to identify genes that varied between the asexual cells . This tool takes account of the large number of zero values ( drop outs ) in the data that are due to the low capture rate inherent in single-cell approaches . We then used these genes to compare each transcriptome and carry out a pseudotime analysis with Monocle 2 ( Trapnell et al . , 2014 ) . This enabled us to place each P . berghei and P . falciparum asexual cell along a developmental trajectory . The cell orderings determined by pseudotime analysis were highly concordant with previously published transcriptional studies of the developmental time course ( Otto et al . , 2010; López-Barragán et al . , 2011; Otto et al . , 2014; Hoo et al . , 2016 ) ( Figure 4A , B , Figure 4—figure supplement 1A , B ) . This demonstrates that single Plasmodium cells from an unsynchronised pool can be ordered by their transcriptional signatures to accurately derive a transcriptional map of development in the late asexual cycle ( Figure 4C , Figure 4—figure supplement 1C ) . In stark contrast to the smooth transitions observed previously in bulk time course experiments ( Bozdech et al . , 2003a; Hoo et al . , 2016 ) , we observed abrupt changes in gene expression during the cell cycle of both P . berghei and P . falciparum ( Figure 4C , Figure 4—figure supplement 1 ) . Whereas a continuous cascade of transcription initiation along the asexual cycle can be seen in bulk RNA-seq data , single-cell data clearly revealed an abrupt transition in expression for the same genes ( Figure 4—figure supplement 2 ) . We also analysed recently published P . falciparum Drop-seq data ( Poran et al . , 2017 ) and observed a similar pattern ( Figure 4—figure supplement 3 ) . Step-wise progression in the cycle represents a departure from the common view and suggests a previously hidden transcriptional pattern , conserved across Plasmodium parasites . Nascent strand bulk RNA-seq had already called into question the cascading nature of transcription initiation in the asexual cycle ( Lu et al . , 2017 ) . We suspect that averaging across slightly asynchronous life cycle stages in bulk RNA-seq studies has previously masked the true nature of transitions along the asexual cell cycle . Individual parasites do not proceed along an incremental path of transcriptional change , but instead generally appear to undergo transcriptional shifts , turning on or shutting down expression of a whole repertoire of genes simultaneously . While these transcriptional modules appear to be rapidly turned on and off during development , they can overlap and cells may express two modules at once . A k-means analysis in pseudotime identified three clusters of genes ( Trapnell et al . , 2014 ) for each species ( Figure 4C , Figure 4—figure supplement 1 , Supplementary file 2 ) . Cluster 1 in P . berghei ( equivalent to cluster 2 in P . falciparum; Figure 4—figure supplement 1 ) was enriched for protein dynamics and energy metabolism including many ribosomal subunits , proteasome subunits and ATPases ( Figure 4C ) . Cluster 2 in P . berghei ( equivalent to cluster 3 in P . falciparum ) was associated with the rhoptry secretory organelle , including ron2 , ron4 , ron5 , ron12 , rop14 , rap1 and rap2/3 . Cluster 3 in P . berghei was enriched for the microneme secretory organelle and the inner membrane complex , including sub2 , ama1 , ripr , imc1c , imc1e , imc1f , imc1g , imc1m and isp3 . This latter cluster was not captured in P . falciparum . These clusters may represent discrete transcriptional modules that underlie parasitic cell cycle checkpoints during the transition from a metabolically active , fast growing trophozoite to a budding multinucleated schizont . We note that two essential ApiAP2 transcription factors ( Figure 4—figure supplement 4 ) were associated with equivalent gene expression clusters in both species: PBANKA_1453700 ( PF3D7_1239200 ) with the early cluster ( 1 ) and PBANKA_0939100 ( PF3D7_1107800 ) with the late cluster ( 2 ) , implicating them as potential regulators of these modules . Like many other cell types ( Spellman et al . , 1998; Kowalczyk et al . , 2015 ) , the point at which Plasmodium parasites are within their cell cycle dominates the transcriptional variation observed within a genetically clonal population . However , there are also genes that vary independently of the cell cycle including clonally variant gene families , which are found largely in the subtelomeric regions of the genome ( Rovira-Graells et al . , 2012 ) . A unique chromatin environment is thought to allow switching between expression of different members of gene families and this mechanism allows parasite populations to adapt to the host immune system ( var genes ) ( Scherf et al . , 2008 ) , establish chronic infection ( pir genes ) ( Scherf et al . , 2008 ) and vary red blood cell invasion pathways ( p235 ) ( Preiser et al . , 1999 ) . Because they enable the parasite to adapt to unexpected environments , members of these multigene families have been termed contingency genes ( Reid , 2015 ) . There is also evidence for variation in expression in response to nutrient sensing ( Mancio-Silva et al . , 2017 ) and to a variety of chemical interventions ( Hu et al . , 2010 ) . We used a regression approach to identify genes that vary independently of the cell cycle ( scLVM ) ( Buettner et al . , 2015 ) by removing cell-cycle-dependent variation from P . falciparum asexual cells . To train this method , we used genes that varied in pseudotime ( i . e . the cell cycle ) . We found that the first two latent factors of the expression data were driven by the cell cycle , each explaining at least 5% of variation in cell cycle genes ( Figure 5—figure supplement 1 ) . After adjusting for these , we identified 56 genes in P . falciparum asexual cells that showed residual variation ( Figure 5A; Supplementary file 2 ) . Unlike clonally variant genes identified in previous work ( Rovira-Graells et al . , 2012 ) , these 56 genes were not located in subtelomeric regions . The products of these genes were involved in nucleosome assembly , the proteasome and vacuolar acidification , suggesting a role in controlling gene expression through transcription initiation , protein stability and protein localisation . The expression patterns of the 56 genes were not correlated , as might have been expected if they were part of a coordinated transcriptional response , such as a stress response . We therefore investigated whether the observed expression pattern resulted from variations in steady-state mRNA levels due to intermittent expression of these genes , followed by rapid mRNA decay . From a published dataset of mRNA half-lives in the asexual cycle , we found that these genes actually have moderately longer than average half lives ( Figure 5B ) . This suggests that the variability of these genes was more likely to be driven by variable transcription initiation than by rapid decay . We found that these genes are more conserved in evolution than expected by chance ( p=2 . 2e-16 ) , and that that this is not simply because they tend to be highly expressed ( Figure 5C ) . Intriguingly , 22 of these 56 genes are also variably expressed genes in the sexual stages , suggesting an intrinsic variability across the life cycle ( Supplementary file 3E ) . Furthermore , similar types of genes were variable in P . berghei sexual stages ( Supplementary file 1A , C ) , but we were unable to identify many cell-cycle-independent variable genes in P . berghei asexual cells , perhaps due to too few cells examined . It is yet to be seen whether the volatile expression of these genes is also reflected in protein abundance . Surprisingly , the most variably expressed genes in sexual stages were those from contingency gene families: var in P . falciparum and pir in P . berghei ( Figure 6; Supplementary file 4 ) . Contingency gene families are extremely evolutionarily labile and different species have different repertoires ( Reid , 2015 ) . Between P . falciparum and P . berghei , there is no evidence of homology between these families and while many are known or assumed to play a role in host–parasite interactions , the extent to which they might perform overlapping functions in the two species is unclear . Little is known about the role of these families in sexual stages and although transcriptional variation has not been observed , expression has ( Florens et al . , 2002 ) and suggests a role for contingency genes in transmission . Several important parts of transmission might require contingency genes encoding cell surface proteins . First , mature gametocytes are found in the blood and are thus susceptible to attack by the host adaptive immune system in much the same way as P . falciparum rings or P . berghei rings and trophozoites . Second , it has been suggested that gametocytes may cluster in order to make transmission more reliable and this might require antigenically variable cell surface proteins ( Pichon et al . , 2000 ) . Finally , after transmission , gametes face a complex and hostile environment in the mosquito midgut where male gametes must rapidly find females , which they do at rates that are difficult to explain without invoking non-random movement such as chemotaxis ( Lawniczak and Eckhoff , 2016 ) . Our data revealed that males and females are very different in their expression of contingency gene families . In P . berghei male gametocytes , we observed significant variability of a set of pir genes ( Otto et al . , 2014 ) ( p=0 . 014; Figure 6—figure supplement 1A; Supplementary file 4 ) , whose protein products have previously been identified in male gametes ( Talman et al . , 2014 ) , indicating a potential role in fertilization . This raises the intriguing possibility that variation in expression of these genes could impact male/female interactions during fertilization . We found no female-specific pir genes , instead , females showed transcriptional variation in members of subtelomeric multigene families fam-a and fam-b ( Figure 6A; Supplementary file 4 ) . In P . falciparum , the var genes are critical for establishing chronic infections through cytoadherence and antigenic variation ( Scherf et al . , 2008 ) . Rather than finding significant variation in males , as expected from our findings in P . berghei , it was females that showed transcriptional variation within the var genes ( p=0 . 0006; Figure 6B ) . In asexual parasites , expression of two different non-coding var transcripts is common and is involved in maintaining the mutually exclusive var gene expression that is essential for their immune evasion role ( Amit-Avraham et al . , 2015; Guizetti and Scherf , 2013 ) . They are both transcribed from a bidirectional promoter within the single var intron . This means that the presence of coding var transcripts in gametocyte transcriptomes can be assessed by identifying intron-spanning reads . We found that within any single female cell , only a single var gene had reads supporting correct splicing , suggesting that mutually exclusive expression of var genes occurs in sexual stages , as it does in asexual parasites ( Guizetti and Scherf , 2013 ) . The coding var transcripts were always from internal var gene clusters , often with the upsC class of promoters , distinct from the subtelomeric var genes seen in asexual stages , with upsB and upsA promoters ( Figure 6B; Figure 6—figure supplement 1B ) . Single male gametocytes were not represented well in this study , so instead we examined previously published bulk male and female gametocyte RNAseq data ( Lasonder et al . , 2016 ) for male var gene expression . Male gametocytes only ever showed mRNA from a single var gene , var2csa , known for its importance in pregnancy related malaria ( Figure 6B; Supplementary file 4 ) . This gene has also been proposed as an important regulator of var gene expression switching ( Mok et al . , 2008 ) . Our novel observation that gametocytes show significant sex-specific variation in expression of large multigene families , hitherto known for their importance in asexual stages , suggests that their evolution and function may also be driven by sexual stage biology . Plasmodium does not have sex chromosomes and the genetic underpinning of sexual dimorphism is very poorly understood . To explore the regulation of sexual dimorphism , we examined sex-specific expression of transcription factors in both species and conducted a co-expression analysis in males and females . We observed a marked , conserved sex-specific pattern of TF expression ( Figure 6B , Figure 6—figure supplement 2 ) . Interestingly , one female-specific TF in particular ( ap2-o ) has been previously shown to have a female function and is likely to have a role in differentiating male and female forms ( Modrzynska et al . , 2017 ) . We have established an optimised protocol for generating single-cell transcriptome sequences of Plasmodium parasites with power to identify not only different cell types but also to explore potential functional variation from one cell to another . This protocol enables evaluation of full length transcripts , something required for evaluating the complex transcriptional patterns we observed for var genes but which is not currently possible with 3’ tag-based approaches ( Poran et al . , 2017 ) . Furthermore , this method also has the advantage of providing information on nearly three times as many genes per cell compared to Drop-seq evaluations of the same species ( ~1900 on average here vs ~650 on average for Drop-seq ) ( Poran et al . , 2017 ) . Future malaria studies will greatly benefit from the availability of both ( i ) low-coverage droplet-based methods allowing for a large number of cells to be analysed and ( ii ) high-coverage full-length transcript methods , allowing high-definition , focused analysis of flow cytometry sorted cells . During the optimisation of our protocol for Plasmodium parasites , we identified several decisive steps and permutable reagents that when modified were key determinants of transcriptome quality . We hope that this optimisation framework may assist in extending full-length transcript scRNA-seq to a much wider range of diverse eukaryotic cell types . As well as establishing a new tool , our study has made several new observations about Plasmodium biology . First , we used single-cell data to produce high-resolution surveys of schizogony and observed sharp transcriptional transitions over the asexual life cycle , which was previously thought to be a continuous process . The intracellular cycle of Plasmodium is complex , consisting of several rounds of endomitotic DNA replication followed by a final synchronised cytokinesis . Although checkpoints are most likely required to ensure timeliness of complex cellular events , such as assembly of the red cell invasion machinery , they have not yet been identified ( Gerald et al . , 2011 ) . We speculate that the sharp transitions we have observed correspond to such checkpoints . Although we found clues as to the possible underlying regulatory architecture , the true regulators remain to be confirmed . A second major finding of our study was unexpected cell-to-cell variation in gene expression . Most genes are known to vary during the asexual , blood stage cell cycle with a single peak of expression ( Bozdech et al . , 2003a ) . Some genes in subtelomeric regions are known to vary independently of the cell cycle , by switching on and off in individual parasites . These include the multigene families of contingency genes known to be involved in sequestration and chronic infection ( var and pir ) . But unexpectedly we found another class of genes varying independently of the cell cycle both in cycling and arrested cells . We found that , unlike contingency genes , they were highly conserved between species and the same types of genes were variable in parasite species infecting both humans and rodents . One could speculate that this is due to noisier signals associated specifically with some cellular function for which it is beneficial to relax transcriptional control . Generating variation in a population of many millions of closely related parasites occupying an ever varying host environment may be a bet-hedging strategy favouring success of at least some of the members of this population . Finally , because our approach was able to dissect both male and female gametocyte transcriptomes , and assess expression of multigene families , we were able to discover an unexpected sex-specificity in expression of several multigene families . Especially intriguing is that these families are known to encode extracellular proteins involved in host-parasite interactions in asexual blood stages . They could have similar host interactive functions not yet described for sexual stages or have uncharted roles in sexual behaviour of the parasite . In the mammalian host , they might be involved in sequestration of mature gametocytes in the peripheral vasculature , as an immune evasion strategy or to aid in transmission through a mosquito bite . The sex-specific nature of the expression of var and pir genes could also indicate a possible role in fertilisation in the mosquito midgut . Single-cell RNA-seq will have many applications for malaria parasites . Surveying parasites directly from patient samples in natural infections will undoubtedly lead to new understandings of the genes underlying important phenotypes . In addition to aspects addressed here , it may be particularly powerful for addressing the following problems: ( i ) analysis of small samples from nonculturable life-cycle stages or Plasmodium species that cannot yet be cultured such as the prevalent human parasite P . vivax , ( ii ) discovery of rare/undescribed cells states , ( iii ) characterisation of the effect of genetic alterations to generate high-dimensional phenotypes for many mutants in parallel ( Bushell et al . , 2017 ) , and ( iv ) examination of cell-to-cell variability in the face of drugs and vaccines . The constitutively mCherry-expressing P . berghei ANKA line , clone RMgm-928 ( Khan et al . , 2013 ) , was propagated in a female 6- to 8-week-old Theiler’s original outbred mouse supplied by Envigo UK . Parasites were purified from an overnight ( 20 hr ) 50 mL culture of 1 mL of infected blood using a 55% Histodenz cushion ( SIGMA ) , following an established schizont purification protocol detailed elsewhere ( Gomes et al . , 2015 ) . Purified late stages ( asexual and sexual ) were pelleted at 450 g for 3 min and incubated with 500 µL of RNALater ( ThermoFisher ) for 5 min , and further diluted into 3 mL of 1x PBS prior to cell sorting . All animal researches were conducted under licenses from the UK Home Office and used protocols approved by the ethics committee of the Wellcome Sanger Institute . An early passage of 3D7-HTGFP ( strain MR4-1029 ) , a transmissible GFP-expressing P . falciparum strain , ( no more than three expansions from frozen stock since initial cloning ) , ( Talman et al . , 2010 ) , was maintained in O-negative red blood cells obtained from the NHSBT , using RPMI 1640 culture medium ( GIBCO ) supplemented with 25 mM HEPES ( SIGMA ) , 10 mM D-Glucose ( SIGMA ) , 50 mg/L hypoxanthine ( SIGMA ) , 10% human serum ( obtained locally in accordance with ethically approved protocols ) , and gassed using a mix containing 5% O2 , 5% CO2 and 90% N2 . Parasites were highly synchronised using two consecutive cycles of Percoll-Sorbitol treatment ( Kutner et al . , 1985 ) . Late asexual parasites ( trophozoites and schizonts ) were purified on a cushion of 63% Percoll ( GE Healthcare ) . Stage V gametocytes were obtained using standard gametocyte culturing ( Fivelman et al . , 2007 ) and purified magnetically with an LS column ( Ribaut et al . , 2008 ) ( Miltenyi Biotec ) . Following purification of each stage , all P . falciparum parasites were pelleted at 800 g for 5 min , incubated with 500 µL of RNALater ( ThermoFisher ) for 5 min , and further diluted into 3 mL of 1x PBS prior to cell sorting . Parasitaemia was determined by Giemsa–stained thin blood smear . Four microlitres of lysis buffer ( 0 . 8% of RNAse-free Triton-X ( Fisher ) in nuclease-free water ( Ambion ) ) , UV-treated for 30 min with a Stratalinker UV Crosslinker 2400 at 200 , 000 µJ/cm2 , 2 . 5 mM dNTPs ( Life Technologies ) , 2 . 5 µM of oligo ( dT ) ( Non-Anchored 30 bp: 5’-AAGCAGTGGTATCAACGCAGAGTACT ( x30 ) −3’; Anchored 30 bp: 5’-AAGCAGTGGTATCAACGCAGAGTACT ( x30 ) VN-3’;Non-Anchored 20 bp: 5’-AAGCAGTGGTATCAACGCAGAGTACT ( x20 ) −3’; Anchored 20 bp: 5’-AAGCAGTGGTATCAACGCAGAGTACT ( x20 ) VN-3’IDT; see Table 1 for detail ) and 2U of SuperRNAsin ( Life Technologies ) were dispensed into each well of the recipient RNAse-free 96-well plate ( Abgene ) immediately prior to the sort and kept on ice . In the first experiment , only 2 µL of lysis buffer were used but the observed cell-capture efficiency was very poor so the volume was increased . Cell sorting was conducted on an Influx cell sorter ( BD Biosciences ) with a 70 µm nozzle . Parasites were sorted by gating on single-cell events and on GFP ( P . falciparum ) or mCherry ( P . berghei ) fluorescence . A non-sorted negative control well and a positive 100-cell control well were included in every plate alongside single cells . Sorted plates were spun at 200 G for 10 s and immediately placed on dry ice . Cells in plates were incubated at 72°C for 3 min . A reverse transcription master mix was added to the samples containing 1 µM of LNA-oligonucleotide ( 5’-AGCAGTGGTATCAACGCAGAGTACATrGrG+G-3’; Exiqon ) , 6 µM MgCl2 , 1M Betaine ( Affymetrix ) , 1X reverse transcription buffer , 50 µM DTT , 0 . 5 U of SuperRNAsin , and 0 . 5 µL of reverse transcriptase ( Table 1 ) . The total volume of the reaction was 10 µL . The plate was incubated using the following programme: 1 cycle of 42°C for 90 min; 10 cycles ( 42°C/2 min , 50°C/2 min ) ; 1 cycle of 70°C for 15 min . Samples were then supplemented with 1X KAPA Hotstart HiFi Readymix and 2 . 5 µM of the ISO SMART primer ( Picelli et al . , 2013 ) and incubated using the following cycling programme 1 cycle of 98°C for 3 min; 25 or 30 cycles ( 98°C/20 s , 67°C/15 s , 72°C/6 min ) ; 1 cycle of 72°C for 5 min ( Table 1 ) . Samples were then purified with 1X Agencourt Ampure beads ( Beckman Coulter ) in a Zephyr G3 SPE Workstation ( Perkin Elmer ) according to the manufacturer’s recommendation . Amplified cDNA was eluted in 10 µL nuclease-free water . Details of different permutations of the protocol tested during the optimisation process are given in Table 1 . The quality of a subset of amplified cDNA samples was monitored with the high-sensitivity DNA chip on an Agilent 2100 Bioanalyser . Samples were verified by qPCR using LightCycler 480 SYBR Green I Master and MSP-1 primers at a concentration of 0 . 4 µM ( Forward: 5’-TCCCAATCAGGAGAAACAGAAG-3’; Reverse: 5’-GATGGTTGTGTTGGTGGTAATG-3’ ) , on a Roche Lightcycler 480 II . Reactions were incubated according to the following cycling programme: one cycle , 95°C/10 min; 45 cycles ( 98°C/20 s , 58°C/10 s , 68°C/30 s ) . Transcripts were quantified with the absolute quantification method using a standard dilution . Libraries were prepared using the Nextera XT kit ( Illumina ) according to manufacturer recommendations . 96 or 384 different index combinations were used to allow multiplexing during sequencing . After indexing , libraries were pooled for clean-up at a 4:5 ratio of Agencourt Ampure beads ( Beckman Coulter ) . Quality of the libraries was monitored with the high sensitivity DNA chip on an Agilent 2100 Bioanalyser . Empty-well controls and single cells were pooled separately from 100-cell controls and loaded proportionally to their expected cell content for sequencing on an Illumina MiSeq or HISeq 4000 . The original Smart-seq2 protocol with the Superscript II enzyme and the original oligo ( dT ) with an anchoring base was run with 30 PCR cycles of preamplification on 10 samples . The samples included a single no-cell control , five single P . falciparum gametocytes , two 10-cell controls and two 100-cell controls . These were multiplexed , along with three samples each of individual human lung carcinoma cells ( A549 ) and sequenced on a single MiSeq run with 150 bp paired end reads . To test the effect of different reverse transcriptase enzymes and different numbers of PCR cycles , we sequenced P . falciparum schizont libraries prepared using the SmartScribe enzyme ( Clontech ) or the SuperScript II enzyme ( Thermofisher ) for each of six single cells , one 100-cell control and two no-cell controls , using 25 cycles of PCR in each case . Samples were multiplexed on a single MiSeq run and sequenced as 150 bp paired end reads . To determine whether single-cell samples might be contaminated with either additional cells or RNA from lysed cells , individual mCherry P . berghei ( RMgm-928 [Khan et al . , 2013] ) and GFP P . falciparum ( Talman et al . , 2010 ) schizonts were mixed in a 1:1 ratio , inactivated with RNAlater fixation and then sorted . A multiplex library was prepared comprising 32 single P . berghei schizonts , two 100-cell P . berghei schizont controls , one no-cell control , 40 single P . falciparum schizonts and two 100-cell P . falciparum schizont controls . These libraries were sequenced as a multiplex pool on a single MiSeq run with 150 bp paired end reads . The P . berghei mixed blood stage samples comprised 182 single-cells of P . berghei , plus four no-cell controls and six 100-cell controls . These were multiplexed with another 192 samples not analysed in this work and sequenced on a single HiSeq 4000 lane using HiSeq v4 with 75 bp paired end reads . The P . falciparum gametocyte samples were sequenced as three multiplexed pools of 84 , using the same chemistry . Three technical duplicate samples were excluded from analysis . The P . falciparum asexual samples were sequenced as two pools of 96 , each on one Illumina HiSeq 2500 lane using HiSeq v4 chemistry with 75 bp paired-end reads . Each batch of 96 samples contained three 100-cell controls . The second batch ( lane 7 ) contained six samples of stage I gametocytes and six samples of stage II gametocytes , each with a single 100-cell control . These were not included in the analysis , leaving 176 single-cell samples . All sequencing experiments were processed in the following way . CRAM files of reads were acquired from the WTSI core pipeline , converted to BAM using samtools-1 . 2 view -b , sorted using samtools sort –n , converted to fastq using samtools-1 . 2 bam2fq and then deinterleaved ( Li et al . , 2009 ) . Nextera adaptor sequences were trimmed using trim_galore -q 20 -a CTGTCTCTTATACACATCT --paired --stringency 3 --length 50 -e 0 . 1 ( v0 . 4 . 1 ) . HISAT2 ( v2 . 0 . 0-beta ) ( Kim et al . , 2015 ) indexes were produced for the P . falciparum v3 ( http://www . genedb . org/Homepage/Pfalciparum ) or P . berghei v3 ( Fougère et al . , 2016 ) genome sequences , downloaded from GeneDB ( Logan-Klumpler et al . , 2012 ) , using default parameters ( October 2016 ) . Trimmed , paired reads were mapped to either genome sequence using hisat2 --max-intronlen 5000 p 12 . For the dual sort experiment , we mapped against a combined reference , allowing us to exclude reads that map to both genomes . SAM files were converted to BAM using samtools-1 . 2 view –b and sorted with samtools-1 . 2 sort . GFF files were downloaded from GeneDB ( October 2016 ) and converted to GTF files using an in-house script . All feature types ( mRNA , rRNA , tRNA , snRNA , snoRNA , pseudogenic_transcript and ncRNA ) were conserved , with their individual ‘coding’ regions labelled as CDS in every case for convenience . Where multiple transcripts were annotated for an individual gene , only the primary transcript was considered . Reads were summed against genes using HTSeq: htseq-count -f bam -r pos -s no –t CDS ( v0 . 6 . 0 [Anders et al . , 2015]; ) . HTSeq excludes multimapping reads by default ( -a 10 ) . This means that reads mapping ambiguously to similar genes from the same family , are not considered in our analysis . For downstream analysis ( excluding examination of rRNA counts ) , transcripts not included in the GeneDB cDNA sequence files were excluded . The raw read counts for P . berghei mixed blood stages , P . falciparum gametocytes and P . falciparum asexual stages are presented in Supplementary file 5 . To determine the useful yield of different RNA amplification protocols ( summarised in Table 1 ) , we classified resulting reads into those mapping to rRNA genes , other genes , unmapped or ambiguous ( falling into more than one category ) . We concentrated here on rRNA because we had observed that this was a particular problem . To do this we began with HISAT2 BAM files produced as described above . Total read pairs were all the unique read pair identifiers . Ribosomal RNA reads were counted using bedtools intersect ( v2 . 17 . 0 [Quinlan and Hall , 2010]; ) to find the overlap of unique read pair ids with rRNA features . Other coding reads were counted in the same way , but looking for overlap with all other features . Unmapped reads were identified using samtools view -f 0 × 8 ( v1 . 2 ) and extracting unique read pair identifiers . Where a read pair occurred in more than one of these lists , it was counted as ambiguous . We compared the library complexity of different iterations of our protocol in order to determine whether more reads resulted in more complexity , or simply more reads from the same genes , perhaps due to large numbers of PCR cycles . Different sequencing runs had very different library sizes and so we downsampled the data . To maximise the number of cells included , while also allowing a reasonable number of reads per cell , we chose to downsample to 50 , 000 reads per cell . To do this , 50 , 000 counts from HTSeq were randomly sampled for each cell . Counts associated with protein coding genes were enumerated and genes were called as detected if there were at least 10 reads mapping to them . Different library preparation and sequencing protocols exhibit different biases in representation of GC/AT-rich sequences and 5` or 3` transcript ends . In order to assess such biases , we took an approach of using the mapped RNA-seq data to identify fragments of genes that were expressed and examined the coverage of genes by these fragments . The reason for doing this , rather than looking at coverage depth was that we had noticed that genes often did not have full coverage , particularly when very long or expressed at a low level . This suggests that , although we would expect Smartseq-2 to amplify full length transcripts , in some cases only partial transcripts survived the full protocol . We used Stringtie ( v1 . 2; default options [Pertea et al . , 2016]; ) to call expressed fragments from our HISAT2 BAM files . We then looked for Stringtie transcript features overlapping each mRNA feature in our reference annotation . Where multiple Stringtie transcripts overlapped each other , these were merged . We then determined , for each gene , the exonic sequence covered by the merged Stringtie transcripts . The length , GC content and relative start and end of these regions was calculated . Observed GC content was compared against the GC content for the whole coding region . Each relative position along a coding sequence ( 0–100 ) covered by a fragment was incremented for each fragment covering it . The coverage of each relative position for each gene was then normalised between 0 and 1 based on the highest coverage across that coding sequence . To examine the effect of gene length , we compared the length distribution of all 4943 P . berghei genes used in our initial analysis to the 4579 which passed our filtering criteria ( having at least 10 reads in at least five cells ) . Reads for the dual sort samples were mapped as above , but to a combined reference of both parasites , enabling reads that map equally well to both genomes to be discarded as their origin could not be determined . Read counts were converted to FPKMs and transcripts with an FPKM >= 10 were counted as expressed . We used these data to show that no well contained more than one cell , that is wells with good data ( a large number of expressed genes ) never had similar numbers of genes from both species . Furthermore , no good wells contained a large number of genes from the incorrect species . To explore whether contaminating genes were similar in different wells , we compared P . falciparum genes identified in wells with a P . berghei cell sorted into them and vice versa between wells . Similarity was calculated as the number of common contaminating genes with an FPKM >= 10 , divided by the average number of contaminating genes between the two wells . Each cell contained very few contaminating genes . This was higher for P . falciparum contamination of P . berghei than P . berghei contamination of P . falciparum , suggesting P . falciparum cells contribute more to extracellular RNA in the medium . Different cells shared very few contaminating transcripts , but the more commonly occurring contaminants were also more highly expressed in their cells of origin . To examine the effect of contamination on downstream experiments , we filtered P . berghei transcriptomes to excluded those with fewer than 10 , 000 reads and fewer than 500 genes , leaving 16 cells out of 32 . We then excluded genes which were not present with at least five read counts in at least two cells . Of 3566 transcripts detected in P . berghei cells ( with at least five read counts in at least two cells ) , 15 were from P . falciparum . The most highly expressed of these were two ribosomal RNA genes - PF3D7_0532000 and PF3D7_0726000 . There were also other ribosomal RNA genes , histones and several known highly expressed genes such as MSP1 , S-antigen , 60S ribosomal protein L6-2 and ETRAMP2 . We then used M3Drop ( min . genes = 500 , MT threshold FDR = 0 . 01 ) to determine whether there were any variable genes across the samples . We found only four , none of which were from P . falciparum . The three main datasets ( Pb mixed , Pf asex , Pf sex ) were processed using Scater v1 . 0 . 4 ( McCarthy et al . , 2016 ) . Firstly we removed genes with no counts in any cell , and the control cells ( 100 cell pools ) . We then removed cells with a total of less than or equal to 25000 read counts and/or less than 1000 genes with at least one read . Subsequently we removed genes that did not have at least 10 reads in 5 cells . For the P . berghei dataset , this resulted in 144/183 cells and 4579 unique genes detected across all cells . For the P . falciparum gametocyte dataset , there were 191/238 cells and 4454 unique genes after filtering and for the P . falciparum asexual dataset 161/180 cells and 4387 unique genes . The counts were then normalised using scran ( Lun et al . , 2016 ) ( v1 . 0 . 3 ) . Normalisation is required due to technical variation between samples due to , for example , variable sequencing depth and capture efficiency . Single-cell RNA-seq read count data contain many zeroes compared to bulk RNA-seq data . These are caused by drop out of low expressed genes or variation between cells and reduce the accuracy of normalisation methods designed for bulk RNA-seq data . Scran uses a pooling approach to reduce these zeroes . Furthermore , it allows an initial clustering of the data and normalisation within these clusters ( e . g . cell types ) , prior to a final normalisation step across the whole dataset . This is particularly useful for our P . berghei data , where the asexual , male and female gametocyte cells differ greatly in their expression patterns . The initial clustering step was performed with the scran function quickCluster ( minimum size = 30 ) . This resulted in three clusters representing the asexual , male and female gametocyte populations . The computeSumFactors function was run using these clusters , with sizes = 20 and positive = TRUE . All downstream analyses were performed with the scran normalised data except where stated . For P . falciparum gametocytes , the computeSumFactors function was run with sizes = 15 . For P . falciparum asexual stages , we set min . size = 20 for quickCluster and the computeSumFactors function was run with sizes = 10 . For some applications , it is necessary to normalise the data by transcript length . For instance , when comparing ranked gene expression values to reference data for determining life cycle stage of a cell . We therefore normalised the scran values by taking the exponent ( 2x ) , multiplying by 1000 and dividing by the cDNA length , determined from the GeneDB cDNA FASTA file ( coding sequence only , no UTRs ) . This is similar to the FPKM calculation , except the library size normalisation is already accomplished . We refer to these values as l-scran , for length-normalised scran values . We used several bulk RNA-seq data sets to assign a life cycle stage to each cell . For P . berghei asexual stages , we used both microarray data from Hoo et al . ( 2016 ) that captures the 24 hr asexual development cycle at 2-hr resolution and RNA-seq data from Otto et al . , 2014 which captures different distinct stages ( Otto et al . , 2014 ) . In the Hoo microarray experiment , Cy5 was used to label each time point while Cy3 was used to label a pool of all samples . The ‘F635 Median - B635’ values are the difference in Cy5 intensity between the median foreground and the median background . This intensity value is related to the actual expression level and these are the values we used . Their data were generated using the P . berghei v2 genome assembly , so we remapped their probe sequences against v3 using HISAT2 ( default parameters; [Kim et al . , 2015] ) . We then used htseq-count -a 200 f sam -r name -s no to identify the genes to which the probes mapped ( cut -f1 , 21 probes_berghei_htseq . sam \| grep PBANKA \| grep -v ambiguous >probes_berghei . map ) . We then used the GPR files provided from ArrayExpress ( Parkinson et al . , 2005 ) ( accession GSE80015 ) and the probe map to produce a table of percentile ranks for each gene in each condition . RNA-seq reads from the Otto et al . dataset were downloaded from the ENA ( PRJNA212241 ) . They were mapped to P . berghei ANKA v3 transcript sequences using Bowtie2 v2 . 2 . 9 ( -a -X 800; [Langmead and Salzberg , 2012] ) and eXpress v1 . 5 . 1 ( Roberts and Pachter , 2013 ) . The resulting read counts were converted to FPKM . Single-cell gene expression values were converted to length normalised scran values ( l-scran ) , as described above , in order to produce more accurate rank expression levels for our scRNA-seq data . We compared each single-cell expression profile against each reference data set . To reduce noise , genes that do not vary greatly between conditions in the reference data were removed . For the P . berghei 24 hr intraerythrocytic developmental cycle reference data ( Hoo et al . , 2016 ) , genes were only included if their expression profile had a mean rank of greater than 30 and less than 70 and standard deviation in rank across samples of greater than 3 . Genes from the query dataset with l-scran <3 were also removed . A minimum of 100 remaining genes common to both the reference and query profiles were required to calculate a correlation between them . The Spearman rank correlation was used in order make the microarray and RNA-seq datasets more comparable . The best correlation of a single-cell expression profile with a reference expression profile was taken as the stage prediction for that single-cell . As new data ( e . g . single-cell analysis of timepoints across the full , synchronised erythrocytic development cycle ) become available , benchmarking staging algorithms will become feasible . Bulk RNA-seq data to classify P . berghei males and females directly was not available . Therefore , we used bulk RNA-seq data ( Otto et al . , 2014 ) that includes mixed-sex gametocyte samples , after converting the profiles to v3 using previous id annotation from PlasmoDB ( Aurrecoechea et al . , 2017 ) . To determine distinct groups of single-cells based on their expression patterns , we used the clustering tool SC3 ( Kiselev , 2016 ) . We used the combined Euclidean , Pearson and Spearman distance , plus the combined PCA and spectral transformation . For the P . berghei dataset the optimal k was 3 ( average silhouette width = 0 . 99 ) , with four being nearly as good ( average silhouette width = 0 . 97 ) . We found that the additional cluster split the asexual parasites into trophozoites and schizonts , while both k values retained the male and female gametocytes as separate clusters . However , there was still extensive variation within these clusters so we further investigated this by excluding asexual cells and clustering again . With this reduced dataset we were able to get a new , robust clustering with k = 3 ( width = 0 . 99 ) . Here , outliers from both the putative male and female clusters clustered together , exclusive of the core of male and female clusters . Markers suggested that six of these outlier cells possessed both male genes and asexual genes , while a single cell possessed both female genes and asexual genes . It is possible that these cells are early gametocytes , committed schizonts or cells doubly infected with both asexual and sexual parasites . These were excluded from further analysis . The markers function of SC3 ( AUROC threshold 0 . 85 , p-value threshold 0 . 01 ) was used on the initial clustering , with k = 3 , to identify novel markers for asexuals , males and females ( Supplementary file 1 ) . Bulk RNA-seq data from Otto et al . ( 2010 ) and López-Barragán et al . ( 2011 ) were used to classify 161 P . falciparum asexual stage cells . RNA-seq reads from Otto et al . ( 2010 ) for the 36 bp Illumina libraries only , were downloaded from the European Nucleotide Archive ( accession ERX001048 ) . They were mapped to the P . falciparum 3D7 genome sequence using HISAT2 v2 . 0 . 0-beta ( Kim et al . , 2015 ) and reads were counted using HTSeq v0 . 6 . 0 ( Anders et al . , 2015 ) . Read counts were then converted into FPKM for subsequent analysis . RNA-seq reads from López-Barragán et al . ( 2011 ) were downloaded from the European Nucleotide Archive ( accession SRX105940 ) and mapped to P . falciparum 3D7 transcript sequences using Bowtie2 v2 . 2 . 9 ( -a -X 800; [Langmead and Salzberg , 2012] ) and eXpress v1 . 5 . 1 ( Roberts and Pachter , 2013 ) . The resulting read counts were converted to FPKM . The López-Barragán et al . ( 2011 ) was used as the consensus prediction , the prediction included six stage II gametocytes which were removed from further pseudotime analysis ( n = 155 ) . Data from Lasonder et al . ( 2016 ) was used to classify P . falciparum gametocyte cells by sex . Raw count data was downloaded from the Gene Expression Omnibus ( Barrett et al . , 2013 ) ( accession GSE75795 ) and converted to FPKM . Data from Young and colleagues ( Young et al . , 2005 ) , was used to classify P . falciparum cells along the gametocyte development time course ( days 1 , 2 , 3 , 6 , 8 , 12 ) . For this dataset profiles of ranks were downloaded from PlasmoDB . The Lasonder data ( Lasonder et al . , 2016 ) highlighted five male cells , with the rest called as females . The Young data ( Young et al . , 2005 ) suggested that all the cells were at a consistent stage of development ( eight days ) , although resolution is lacking at the most relevant timepoints , between eight and twelve days . The best classification of each cell based on each of the bulk datasets used above is listed in Supplementary file 5 . Within the 54 P . berghei cells identified as asexual , 275 genes were found to be variable using M3Drop ( Andrews and Hemberg , 2016 ) ( raw count input , False Discovery Rate <= 0 . 01 ) . L-scran expression values for this subset of genes were used to order the cells in pseudotime using Monocle 2 ( Trapnell et al . , 2014 ) ; specifically the reduceDimension ( ) and orderCells ( num_path = 2 ) functions were used to derive the ordering of the cells . Monocle 2 identified a single cell state and the cells were ordered in a single trajectory ( Figure 2a ) . The Monocle 2 package was further used to cluster genes in pseudotime ( k = 3 ) with the clusterGenes ( ) function; the Nbclust package was used to define the optimal number of clusters . We looked for enrichment of Gene Ontology terms within the three clusters identified , using topGO ( Alexa et al . , 2006 ) ( summarised in Figure 2c ) . For the P . falciparum 155-cell dataset , 360 genes were found to be variable genes with M3Drop ( raw count input , False Discovery Rate <= 0 . 01 ) ( Andrews and Hemberg , 2016 ) , Monocle 2 identified two branches defining three possible trajectories , although 2 of these were minor ( States 2 and 5 in Figure 4—figure supplement 1B ) . Cells ordered in these minor trajectories did not seem to correlate with known biological markers , such as sexual commitment markers ( ap2-g and gdv-1 ) and these cells were removed from all further analyses . The pseudotime analysis was repeated on the main trajectory of cells ( 125 cells ) . The Monocle 2 package was further used to cluster genes in pseudotime ( k = 3 ) with the clusterGenes ( ) function; the Nbclust package was used to define the optimal number of clusters . We looked for enrichment of Gene Ontology terms within the three clusters identified , using topGO ( Alexa et al . , 2006 ) . To determine whether the same set of genes displayed different patterns across development in bulk and single-cell RNA-seq experiments we made direct comparisons between these two approaches . After ordering the P . berghei asexual cells by pseudotime , genes were ordered by their peak of expression based on linear ( i . e . not logged ) , length-normalised scran expression values . To do this , expression value data , ordered by pseudotime , were normalised , then Fourier transformed , sorting transcripts according to the phase of the most prominent frequency . Signal-to-noise ( S/N ) ratios were calculated for each transformed signal and normalised with respect to the maximum achievable value for the dataset . Transforms with a normalised S/N of less than 0 . 1 were excluded from the results as lacking evidence of periodicity . The Hoo et al . dataset ( Hoo et al . , 2016 ) were treated in the same way , but initially ordered by time point of collection rather than pseudotime and using intensity values as described above . This resulted in 1141 ordered genes for our single cell data and 2612 genes for the Hoo data ( Hoo et al . , 2016 ) . There were 651 shared genes , which were used to compare the two datasets ( Figure 4—figure supplement 2A , B ) The P . falciparum asexual cells were ordered differently . We used the Otto P . falciparum asexual development cycle time course data ( Otto et al . , 2010 ) as a reference . These data were processed as described above . We used the Fourier transform approach described above , with a normalised S/N ratio of 0 . 5 to identify 4517 genes from the Otto et al . dataset ( Otto et al . , 2010 ) . We then identified 336 genes common to this list and the list of 361 differentially expressed genes identified across the 155 single cells . This different approach for P . falciparum was taken because the window of time captured by our single cells was too narrow to identify cycling genes using the Fourier approach ( all the normalised S/N ratios were very low e . g . <0 . 05 ) . We then generated heatmaps for the the bulk and single cell datasets , with the genes ordered by their peak time in the Otto et al . dataset ( Otto et al . , 2010 ) in both cases ( Figure 4—figure supplement 2C , D ) . The Plasmodium Drop-seq data ( Poran et al . , 2017 ) was loaded with the package Seurat version 2 . 1 ( Butler and Satija , 2017 ) , QC steps and normalisation were performed with the exact same parameters as in the released code ( https://github . com/KafsackLab/scRNAseq-Malaria ) . 8581 AP2-DD ( on and off ) cells from three time points ( 30 , 36 , 42 hr post invasion ) were subsetted and ordered in pseudotime with Monocole 2 ( Trapnell et al . , 2014 ) . Orthologous P . falciparum and P . berghei ApiAP2 transcription factors were mapped within the single cell data and ordered according to expression pattern in both species . The set of ApiAP2 expressed in each P . berghei cell type ( >10 cells ) was isolated and used to generate a co-expression correlation network using the Hmisc package ( clustfunc - method = compete; distfunc - method=‘euclidean’ ) . The edges with a Pearson coefficient superior to 0 . 4 were used to represent the network with Cytoscape ( Prefuse force directed layout setting ) . Essentiality data was from Modrzynska et al . , 2017 . An average co-expression correlation of the 14 TFs expressed in asexuals ( Figure 4—figure supplement 4B ) with each gene in the clusters identified was calculated , and the TFs were each associated with the cluster with which they had the greatest correlation coefficient if it was greater than 0 . 4 . P . berghei trophozoite and schizont single cells were processed using scLVM v0 . 99 . 2 ( Buettner et al . , 2015 ) . Counts were normalised by size factors . Technical noise was estimated using a log fit and 1485 variable genes were called using getVariableGenes with default parameters . To fit the latent cell cycle factor , we used the 2104 genes identified as differentially expressed in pseudotime using Monocle with corrected p-value<=0 . 01 . We were aware from our PCA analysis that the first two principal components were likely driven by the cell cycle , and found using the latent factor analysis that only the first two factors explained at least 5% of variation in cell cycle genes . Therefore , we refitted the model with two latent factors . We identified only four genes that had >= 50% of their variance attributable to biological noise after correction , for example the variation in their expression is largely not driven by the cell cycle or by technical noise . These appeared to be driven by a small number of potential outlier cells , which had not been identified in previous analyses . A larger sample of P . berghei asexual cells would be required to better address cell-cycle-independent gene expression variation in this species . The P . falciparum asexual stage single-cell RNA-seq data were filtered as previously and processed with scLVM as above . We used 416 genes identified as differentially expressed in pseudotime using Monocle with corrected p-value<=0 . 01 . We were aware from our PCA analysis that the first two latent factors were likely driven by the cell cycle , so we refitted the model with two latent factors . We identified 56 genes that had >= 50% of their variance attributable to biological noise after correction , e . g . the variation in their expression is largely not driven by the cell cycle or by technical noise . We used TopGO ( Alexa et al . , 2006 ) to identify GO terms enriched amongst these genes , with parameters as elsewhere . We looked for correlations amongst the cell-cycle independent genes by fitting a linear mixed model ( LMM ) as suggested in the scLVM vignette ( https://github . com/PMBio/scLVM/blob/master/R/tutorials/scLVM_vignette . Rmd ) . We found no significant correlations with a correlation coefficient >= 0 . 5 or <=−0 . 5 . We hypothesised that these variable transcripts might be more rapidly degraded than others , making them appear more sporadically expressed in steady state mRNA . We looked at the distributions of half life for the 56 variable transcripts in P . falciparum . We took the half life estimates per oligo from the supplementary material of ( Shock et al . , 2007 ) . We then used the oligo to gene mapping from ( Bozdech et al . , 2003b ) and converted the ids into the new PF3D7_ prefix gene ids using one-to-one orthologues from PlasmoDB ( Aurrecoechea et al . , 2017 ) . For each life stage , we compared the distribution of half-life values in the whole sample to those of the 56 variable genes . We had half-life measurements for between 34 and 39 transcripts , depending on stage . We used the Kolmogorov-Smirnov test ( two-sided ) to determine differences between the distributions ( Figure 5B ) Protein sequences for 4290 1:1 orthologous genes in P . falciparum and P . berghei were retrieved from PlasmoDB ( Aurrecoechea et al . , 2017 ) and aligned with Muscle ( Edgar , 2004 ) . Within alignments , a substitution score was calculated for each amino acid position based on the BLOSUM62 substitution matrix ( Henikoff and Henikoff , 1992 ) . The conservation score for each gene corresponds to the mean of the scores of all the amino acids in the protein the genes encodes ( MalariaGEN Plasmodium falciparum Community Project , 2016 ) . To examine gene expression variation within life stages , we used the filtered datasets and considered asexual , male , and female cells separately for each species . For P . berghei , M3Drop ( Andrews and Hemberg , 2016 ) was used to determine gene expression variability amongst cells with FDR <= 0 . 01 . We identified 115 variable genes in P . berghei females , 73 in males , and 275 in asexuals . In P . falciparum , we found 360 variable genes in asexuals , and 448 variable genes in females . We were not able to analyse variability in P . falciparum males because we found only five of them . To examine functional classes enriched amongst variable genes , we used topGO with the weight01 algorithm , the Fisher statistic , node size = 5 and False Discovery Rate >= 0 . 05 ( Alexa et al . , 2006 ) . Gene ontology terms for P . berghei and P . falciparum genes were extracted from GeneDB EMBL files . Multigene families in P . berghei do not have associated Gene Ontology ( GO ) terms and so we used ad hoc hypergeometric tests to look at their enrichment . We found that pir genes were enriched amongst variable genes in males ( 5 of 135 , hypergeometric test , p=0 . 014 ) , but there were none in females . We found that for P . falciparum females , enriched GO terms included modulation by symbiont of host erythrocyte and cytoadherence to microvasculature , mediated by symbiont protein . These terms refer to the 14 var genes found amongst the variable genes ( 14 of 60 genes , hypergeometric test , p=0 . 0006 ) . The genes that show significant expression variation in each of these categories are found in Supplementary file 2 and 3 , along with their functional enrichment analyses . It is known that antisense transcripts are expressed from a bidirectional promoter in the intron of each var gene ( Guizetti and Scherf , 2013 ) . Our protocol does not preserve information about which strand is transcribed . Therefore , finding that reads map to either exon of a var gene does not provide evidence that it is functionally expressed . In order to identify sense transcripts , we looked for reads mapping over the single intron of each var gene . These reads , which include both exons , must originate from mRNA transcripts and thus not from antisense transcripts beginning within the intron . Initially we identified reads from the HISAT2 mappings which overlapped annotated var genes using bedtools intersect ( Quinlan and Hall , 2010 ) . From the resulting BAM file we selected those reads that included an N in the CIGAR string , indicating a split read . We then identified the var gene each read overlapped and whether it was split exactly over the intron . We called expression for a var gene where there were at least two reads mapping over the intron . SRA files for Lasonder et al . ( 2016 ) were downloaded from GEO ( SRR2981459 , SRR2981460 , SRR2981461 ) . These were converted into FASTQ format using fastq-dump ( v 2 . 3 . 5 ) . The reads were mapped to the P . falciparum 3D7 references genome using HISAT2 ( --rna-strandness R --max-intronlen 5000 p 12 [Kim et al . , 2015] ) . Reads were counted against features using htseq-count ( -f bam -r pos -s reverse -t CDS; [Anders et al . , 2015] ) . We wanted to look more generally at how multigene family expression and variation between cells differ between males and females and between species . To determine DE genes between males and females in the Lasonder P . falciparum bulk RNA-seq data , we used EdgeR ( Robinson et al . , 2010 ) . This tool is however not appropriate for single-cell data as it does not take account of drop outs . Therefore , for the P . berghei single-cell data we collated the single-cell data to form three replicates of pseudo-bulk transcriptomes for males and three for females to reduce drop outs and simulate bulk data . This was done by summing each single-cell gene expression profile to either replicate 1 , 2 or 3 until we had run out of single cells . We went on to look for genes that were more highly expressed in males or females using EdgeR and were from multigene families . For the female bulk data , we saw evidence of sense transcription for a similar set of var genes to the female single cells . Those which we did not observe in the single cell data were also upsC1 types . We saw no evidence for sense expression of var genes in our five male single cells . However , in the Lasonder bulk RNA-seq male sample MG2 , the var gene var2csa ( PF3D7_1200600 ) had 13 reads overlapping the PF3D7_1200600 var2csa intron . Perl , R and C++ code for various analyses are available at https://github . com/adamjamesreid/Plasmodium-single-cell-RNA-seq ( Reid , 2018; copy archived at https://github . com/elifesciences-publications/Plasmodium-single-cell-RNA-seq ) . The single-cell RNA-seq reads are available from the European Nucleotide Archive ( accession ERP021229 ) and ArrayExpress ( accession E-ERAD-611 ) . Raw read counts and metadata including classifications for each cell are also presented in Supplementary files 5 and 6 .	Malaria is a life-threatening disease that affects hundreds of millions of people every year and causes around 500 , 000 deaths , mostly among young children . The disease is caused by Plasmodium parasites , which have a complex life cycle that involves different stages in different hosts . During mosquito bites , the parasites can be transmitted to people where they spend part of their life cycle inside red blood cells . Inside these cells , they can multiply rapidly and eventually burst the blood cells , which causes some of the symptoms of the disease . The parasite also produces sexual stages , which can be passed on to the next mosquito that feeds on the host . Scientists have been studying these different stages to better understand how the parasites manage to evade the human immune system so successfully . Most of the research has looked at how genes differ between large pools of parasites , but this approach hides important differences between individual parasites . Understanding variation and how individual parasites behave could help to develop new and effective drugs and vaccines for malaria . Now , Reid et al . used a technique called single-cell RNA sequencing , which allowed them to hone in on individual genes within a single parasite . This revealed hidden patterns in the way the parasites use their genes across the life cycle . When the parasite is developing inside a red blood cell , distinct groups of genes turn on simultaneously and are later switched off together . Reid et al . found clues about the genes that might be controlling these groups . The experiments also showed that a set of genes previously thought to be involved solely in evading the immune system is also important for the transition from human to mosquito . A next step will be to see if single-cell RNA sequencing technology could be used to reveal more about the basic biology of the parasite and how it resists drugs or evades the immune system . In the future , this may help to develop drugs that interfere with the synchronisation of these groups of genes to disrupt the parasite’s development and stop it from causing the disease . The genes involved in transmission between hosts could be another promising drug target , and one day , may help to eliminate the disease .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "cell", "biology", "tools", "and", "resources" ]	2018	Single-cell RNA-seq reveals hidden transcriptional variation in malaria parasites
Measures of lung function are heritable , and thus , we sought to utilise genetics to propose drug-repurposing candidates that could improve respiratory outcomes . Lung function measures were found to be genetically correlated with seven druggable biochemical traits , with further evidence of a causal relationship between increased fasting glucose and diminished lung function . Moreover , we developed polygenic scores for lung function specifically within pathways with known drug targets and investigated their relationship with pulmonary phenotypes and gene expression in independent cohorts to prioritise individuals who may benefit from particular drug-repurposing opportunities . A transcriptome-wide association study ( TWAS ) of lung function was then performed which identified several drug–gene interactions with predicted lung function increasing modes of action . Drugs that regulate blood glucose were uncovered through both polygenic scoring and TWAS methodologies . In summary , we provided genetic justification for a number of novel drug-repurposing opportunities that could improve lung function . Optimal lung ( pulmonary ) function is vital for the ongoing maintenance of homeostasis , with reduced pulmonary function associated with a marked increase in the risk of mortality ( Vasquez et al . , 2017; Young et al . , 2007 ) . This is particularly critical due to the considerable number of disorders for which diminished pulmonary function is a clinical hallmark . For instance , chronic obstructive pulmonary disease ( COPD ) , characterised by an irreversible limitation of airflow , is one of the leading causes of death worldwide ( Quaderi and Hurst , 2018 ) . Pulmonary manifestations are also common amongst disorders not directly classified as respiratory conditions , including diabetes ( Pitocco et al . , 2012; Walter et al . , 2003 ) , congenital heart disease ( Alonso-Gonzalez et al . , 2013 ) , and inflammatory bowel disease ( Ji et al . , 2016; Yilmaz et al . , 2010 ) . Bacterial and viral infection , such as Streptococcus pneumoniae , Mycobacterium tuberculosis , influenza , and coronaviruses , also cause severe declines in respiratory function . In order to better manage the spectrum of respiratory disorders , there is a desperate need for new interventions , including those that can be targeted to an individual’s heterogeneous risk factors . While the development pathway for new compounds is difficult , there are likely to be opportunities for precision repurposing of existing drugs to enhance lung function and improve patient outcomes . Spirometry measures of pulmonary function have been shown to display significant heritability both in twin designs and genome-wide association studies ( GWAS ) ( Palmer et al . , 2001; Ingebrigtsen et al . , 2011; Shrine et al . , 2019 ) . Genomics may reveal clinically relevant insights into the biology underlying lung function , and thus , could be leveraged for drug repurposing . We sought to interrogate the genomic architecture of three spirometry indices to propose drug-repurposing candidates which could be used to improve lung function: forced expiratory volume in 1 s ( FEV1 ) , forced vital capacity ( FVC ) , and their ratio ( FEV1/FVC ) . Firstly , we assessed each lung function trait for evidence of genetic correlation with biochemical traits that could be pharmacologically modulated , followed by models to investigate whether there was evidence of causation . The previously developed pharmagenic enrichment score ( PES ) framework was then implemented to identify druggable pathways enriched with lung function-associated variation and calculate pathway-specific polygenic scores ( PGS ) to prioritise individuals who may benefit from a repurposed compound which interacts with the pathway ( Reay et al . , 2020 ) . A transcriptome-wide association study ( TWAS ) of FEV1 and FVC was also undertaken to reveal genes which could be targeted by existing drugs that may increase pulmonary function . Finally , we considered the repurposing candidates proposed by these strategies in the context of three respiratory viruses ( severe acute respiratory syndrome coronavirus 2 [SARS-CoV2] , influenza [H1N1] , and human adenovirus [HAdV] ) , specifically analysing the interactions between viral and human proteins . An overview schematic of this study is detailed in Figure 1 and Figure 1—figure supplement 1 . We assessed genetic correlation between three pulmonary function measurements ( FEV1 , FVC , and FEV1/FVC ) and 172 GWAS summary statistics of European ancestry using bivariate linkage disequilibrium score regression ( LDSC ) ( Bulik-Sullivan et al . , 2015; Zheng et al . , 2017 ) . A number of clinically significant traits displayed significant genetic correlation with FEV1 , FVC , and/or FEV1/FVC after correcting for the number of tests performed ( p<2 . 9×10−4 , Figure 2a , Supplementary file 1a–c ) . FVC had the largest number of genetic correlations which surpassed Bonferroni correction ( N = 35 ) , followed by FEV1 and FEV1/FVC for which 25 and 8 traits survived multiple testing correction , respectively . The trait most significantly correlated with both FEV1 and FVC was waist circumference – FEV1: rg = −0 . 19 , SE = 0 . 02 , p=5 . 71×10−20; FVC: rg = −0 . 24 , SE = 0 . 02 , p=9 . 54×10−33 – whilst asthma demonstrated the most significant correlation with FEV1/FVC ( rg = −0 . 35 , SE = 0 . 05 , p=3 . 49×10−12 ) . Interestingly , there was evidence of genetic correlation between measures of lung function and circulating levels of both metabolites and hormones . This is notable as these molecules can be pharmacologically modulated , potentially informing novel therapeutic strategies and drug-repurposing opportunities to improve lung function . Significant genetic correlations were observed with four metabolites ( fasting glucose , high-density lipoprotein [HDL] , triglycerides , and urate ) and two hormones ( fasting insulin and leptin ) for at least one measure of lung function ( Table 1 ) . The genetic correlations observed between lung function measures and metabolite/hormone traits may be clinically actionable; however , a significant estimate of genetic correlation does not imply causality ( O'Connor and Price , 2018 ) . In response , we constructed a latent causal variable ( LCV ) model to estimate mean posterior genetic causality proportion ( GCP^ ) for each metabolite or hormone trait and the lung function measure with which it is genetically correlated ( Figure 2b , Supplementary file 1d ) . The LCV method assumes that a latent variable mediates the genetic correlation between two traits and tests whether this latent variable displays stronger correlation with either of the traits . We used the recommended threshold for partial genetic causality of \|GCP\| > 0 . 6 as this has been demonstrated in simulations to appropriately guard against false positives ( O'Connor and Price , 2018 ) . There was strong evidence of partial genetic causality of fasting glucose on FVC: GCP^ = 0 . 77 , SE = 0 . 15 , P H0:GCP = 0 = 1 . 32 × 10−56 . Importantly , the posterior mean GCP estimate for the relationship between fasting glucose and FVC remained strong ( \|GCP\| > 0 . 6 ) using a fasting glucose GWAS additionally adjusted for body mass index ( BMI ) : GCP^ = 0 . 63 , SE = 0 . 22 , p=1 . 67×10−56 . The LCV model constructed between fasting glucose and FEV1 did not surpass the threshold of \|GCP\| > 0 . 6 we use to designate partial genetic causality; however , it was directionally consistent with the fasting glucose to FVC estimate and closely approaches this threshold , FEV1: GCP^ = 0 . 57 , SE = 0 . 18 , P H0:GCP = 0 = 7 . 18 × 10−12 . A strong posterior GCP estimate was observed for urate and FVC ( GCP^=0 . 73 ) , although the relatively low heritability z score as calculated by the LCV framework ( z < 7 ) may lead to a biased estimate . As a result , the relationship between urate and FVC should be treated with caution and further study would be needed to replicate this finding in a urate GWAS with a more precise heritability estimate . The estimate between HDL cholesterol and FEV1/FVC ( GCP^=0 . 59 , SE = 0 . 26 , P H0:GCP = 0 = 4 . 12 × 10−7 ) was also close to the GCP threshold but we do not denote this as strong evidence of genetic causality given the 0 . 6 threshold was not exceeded . There was no strong evidence of genetic causality between any of the remaining LDSC-prioritised hormone or metabolite traits and FEV1 , FVC , or FEV1/FVC . As it was the most significant LCV model , the causal effect of fasting glucose on FEV1 and FVC was further investigated utilising a Mendelian randomisation ( MR ) approach . MR differs from an LCV model as it exploits genome-wide significant variants as genetic instrumental variables ( IVs ) to calculate a causal estimate of an exposure ( fasting glucose ) on an outcome ( lung function ) . Given genetic correlation may bias MR due to pleiotropy , we implement MR here as a validation of the LCV results as it uses different set of statistical parameters and assumptions; however , the estimates derived from MR should be viewed cautiously in light genetic correlation which exists between fasting glucose and lung function ( O'Connor and Price , 2018 ) . We selected 32 genome-wide significant variants associated with glucose in approximate linkage equilibrium as IVs ( p<5×10−8 , r2 < 0 . 001 ) to ensure that variants were both rigorously associated with the exposure and independent from one another . A 1 mmol/L increase in fasting glucose was associated with a −0 . 088 ( 95% confidence interval [CI]: −0 . 17 , −0 . 01 ) standard deviation decline in FVC using an inverse variance weighted ( IVW ) estimator with multiplicative random effects . Similarly , elevated fasting glucose was also shown to have a negative effect on FEV1: βIVW = −0 . 096 ( 95% CI: –0 . 18 , –0 . 01 ) . The IVW estimate for fasting glucose was only nominally significant for both FVC and FEV1 ( p=0 . 033 and 0 . 023 , respectively ) , with relatively wide confidence intervals to approach zero , and thus , the estimate should be treated with appropriate caution . We implemented a number of sensitivity analyses to test the rigour of our causal estimate of the effect of fasting glucose on lung function ( Figure 2—figure supplement 1 , Supplementary file 1e–g ) . Firstly , we obtained an analogous , and statistically significant , causal estimate using the weighted median method ( FVC: βWeighted median = −0 . 09 [95% CI: –0 . 16 , –0 . 04] , p=1 . 87×10−3 , FEV1: βWeighted median = −0 . 07 [95% CI: –0 . 13 , –0 . 01] , p=0 . 025 ) . The weighted median method relaxes the assumption that all IVs must be valid , as described elsewhere ( Bowden et al . , 2016 ) . An MR–Egger model was then constructed , which includes a non-zero intercept term which can be used as a measure of unbalanced pleiotropy ( Bowden et al . , 2015 ) . The causal estimate using MR–Egger was in the same direction for FEV1 and FVC; however , it was non-significant ( FVC: βMR Egger = –0 . 13 [95% CI: −0 . 30 , 0 . 04] , p=0 . 148 , FEV1: βMR Egger = –0 . 12 [95% CI: −0 . 30 , 0 . 06] , p=0 . 21 ) . Importantly , the MR–Egger intercept was not significantly different from zero in the FEV1 or FVC model , indicating no evidence of unbalanced pleiotropy . This was supported by a non-significant global test of pleiotropy implemented as part of the MR-Pleiotropy Residual Sum and Outlier ( MR PRESSO ) framework ( Supplementary file 1f; Verbanck et al . , 2018 ) . Furthermore , we evaluated whether there was any evidence of reverse causation , that is , FEV1 and FVC exerting a causal effect on fasting glucose using the MR Steiger directionality test , with our observed direction of causation from glucose to lung function supported . Finally , we successively recalculated the IVW causal estimate for the effect of fasting glucose on FEV1 and FVC by removing one IV at a time in a ‘leave-one-out’ analysis ( Supplementary file 1g; Burgess et al . , 2017 ) . An analogous causal estimate was derived regardless of which IV was removed; however , there were five IVs ( FEV1 model = two outlier single nucleotide polymorphisms ( SNPs ) , FVC model = four outlier SNPs [two outlier SNPs shared] ) for which the estimate was marginally non-significant after exclusion ( maximum p=0 . 11 , IVW with multiplicative random effects ) . We then used a phenome-wide association approach to demonstrate that these five SNPs were ( i ) annotated to genes with important roles in glycaemic homeostasis and ( ii ) were almost exclusively associated with glycaemic traits or diabetes ( Supplementary file 1g–l ) . As a result , we concluded that these IVs did not likely represent horizontal pleiotropy , which would bias the causal estimate , but instead were biologically salient IVs with large effects ( Supplementary file 1g ) . Whilst smoking status ( ever vs . never smoked ) was a covariate in the lung function GWAS , we sought to assess whether the relationship between blood glucose and lung function could be driven by residual effects of smoking . There was a significant genetic correlation between the number of cigarettes smoked per day and fasting glucose ( rg = 0 . 16 , SE = 0 . 043 ) , although this was not observed with the ‘ever vs . never smoked’ phenotype ( rg = 0 . 007 , SE = 0 . 039 ) . However , an LCV model constructed for fasting glucose and cigarettes smoked per day did not indicate evidence of genetic causality in contrast to the glucose/lung function models: GCP^ = −0 . 47 , SE = 0 . 33 , P H0:GCP = 0 = 0 . 25 . The MR IVs for glucose were further checked for association with either ‘ever vs . never smoked’ and ‘cigarettes per day’ , with none of the IVs demonstrating any association with either smoking phenotype at a genome-wide ( p<5×10−8 ) or suggestive ( p<1×10−5 ) significance threshold ( Supplementary file 1m , n ) . Moreover , we investigated the possibility that our results may be impacted by collider bias given the lung function GWAS we utilised was phenotypically covaried for smoking status . We leveraged a smaller UK Biobank GWAS of FEV1 and FVC from the Neale Lab that did not adjust for smoking ( N = 272 , 338 ) . The posterior mean GCP and IVW estimates were in the same direction and relatively analogous for both spirometry measures to that observed using the larger GWAS covaried for smoking status , with no apparent evidence that the negative relationship between glucose and lung function was influenced by smoking as a collider variable . In summary , these data suggested that there is an effect of fasting glucose on lung function beyond what is directly attributable to a residual impact of smoking . We aimed to further expand drug-repurposing opportunities for lung function using the PES approach ( Reay et al . , 2020 ) . Briefly , PES aims to implement genetically informed drug repurposing with PGS calculated using genetic variants specifically within druggable pathways ( Figure 3a ) . In the context of this study , individuals with a depleted PES for lung function ( lower genetically predicted lung function ) mapped to pathways with known drug targets may specifically benefit from drugs which modulate these pathways . Firstly , we performed gene-set association of FEV1 and FVC using a collection of high-quality gene-sets from the molecular signatures database ( MSigDB ) . These sets contain at least one gene which is modulated by an approved pharmacological agent ( NSets = 1030 ) . The FEV1/FVC phenotype is less directly interpretable in this context , given that it is used primarily as a diagnostic tool rather than as a quantitative measure , and thus , we focused on repurposing candidates for FEV1 and FVC individually . Variants were annotated to genes using genomic proximity , with both conservative and liberal upstream and downstream boundary definitions . Gene-set association using the FEV1 and FVC GWAS was undertaken at each PT with both conservative and liberal genic boundaries . If a gene-set was significant at multiple PT , the most significantly associated PT was retained . The conservative genic boundaries only yielded one druggable gene-set enriched with FEV1-associated variants after multiple testing correction ( q < 0 . 05 ) : signalling events mediated by the Hedgehog family – β = 0 . 973 , SE = 0 . 2 , p=9 . 3×10−7 , PT <0 . 5 , NGenes = 22 ( Supplementary file 2a ) . There were no gene-sets with known drug targets using conservative genic-boundaries which survived multiple testing correction ( false discovery rate ( FDR ) < 0 . 05 ) for association with FVC . Extending the genic boundaries to capture more regulatory variation ( liberal boundaries ) uncovered more druggable gene-sets ( Supplementary file 2b ) . Specifically , there were seven and nine unique gene-sets which survived correction for FEV1 and FVC , respectively ( q < 0 . 05 , Table 2 ) . It should be noted that there were two pathways related to Hedgehog signalling; however , as these were from different annotation sources and had a different number of genes , we considered them separately . A number of biological processes were encompassed by these prioritised gene-sets , such as cancer ( pathways in cancer , basal cell carcinoma ) , transforming growth factor ( TGF ) -β superfamily signalling ( TGF-β signalling pathway , bone morphogenetic protein [BMP] receptor signalling , activin receptor-like kinase [ALK] in cardiac myocytes ) , and cardiac function ( dilated cardiomyopathy ) . For each candidate PES gene-set , we performed computational drug selection to identify approved compounds predicted to modulate the enriched pathway . Firstly , we investigated U . S . Food and Drug Administration ( FDA ) -approved pharmacological agents with a statistically significant overrepresentation of target genes in each of these sets ( NOverlap ≥3 , q < 0 . 05 ) . Drugs which target ( i ) multiple gene-set members and ( ii ) more genes than expected by chance were assumed to be particularly relevant for a biological pathway . There were six such gene-sets from the PES candidates which survived multiple testing correction enriched with the targets of an FDA-approved compound ( pathways in cancer , dilated cardiomyopathy , class B/2 [secretin family receptors] , circadian clock , extension of telomeres , and extracellular matrix ( ECM ) /ECM-associated proteins , Supplementary file 2c ) – notable drugs included the anti-mineralocorticoid spironolactone , antihyperglycaemic compounds ( rosiglitazone , pramlintide ) , antihypertensives ( e . g . verapamil and felodipine ) , antineoplastic agents ( e . g . bexarotene and sunitinib ) , and nutraceuticals ( zinc , vitamin E , and doconexent ) . Each compound was annotated with its Anatomical Therapeutic Chemical ( ATC ) classification; the most common first-level ATC code amongst these compounds was antineoplastic and immunomodulating agents ( L , N = 16 ) , followed by cardiovascular system ( C , N = 15 ) , and alimentary tract and metabolism ( A , N = 12; Figure 3b ) . Each of these compounds was subjected to expert curation by a pharmacist in relation to side effects and prior literature evidence as detailed in Supplementary file 2d ( Figure 3—figure supplement 1 ) . Single drug–gene matching was undertaken for remaining PES candidate gene-sets lacking an approved compound with statistically overrepresented target , retaining drug–gene interactions with at least two lines of evidence from Drug–Gene Interaction Database ( DGIdb ) ( Supplementary file 2e–p ) . In order to test the phenotypic relevance of FEV1 and FVC PES profiles , we utilised an independent genotyped cohort from the Hunter Community Study ( HCS , N = 1804 ) . Firstly , we constructed a genome-wide PGS for FEV1 and FVC at six different p-value thresholds ( Supplementary file 3a ) . The optimum FEV1 genetic score explained approximately 6 . 4% of the variance in FEV1 measured in the HCS cohort , whilst the FVC PGS explained approximately 5 . 7% of variance in FVC . Each of the seven PES profiles were tested for association with FEV1 and/or FVC both with and without adjustment for genome-wide PGS . Four of the PES considered had at least a nominally significant association with their respective spirometry measure ( p<0 . 05 , Table 3 , Supplementary file 3b ) , whilst three survived correction for the number of tests ( p<7 . 14×10−3 ) . The variance explained by the significant PES was between 0 . 4 and 0 . 7% , with the number of independent SNPs in these scores ranging from 76 to 16 , 390 . We then constructed a model which was adjusted for genome-wide PGS at the same PT as the PES and found that only the class B/2 secretin family receptor FVC PES remained nominally significant ( β = 0 . 047 , SE = 0 . 022 , p=0 . 038 , Figure 3c , Supplementary file 3c ) , although we acknowledge this association does not survive correction for the seven tests performed . This PES did not display any association with smoking status in this cohort ( β = −0 . 014 , SE = 0 . 047 , p=0 . 758 ) , whilst the signal remained nominally significant upon removing HCS participants with self-reported respiratory illness ( N = 1433 , β = 0 . 052 , SE = 0 . 025 , p=0 . 042 ) . Furthermore , there was a significant depletion of FVC within the 10th percentile ( low genetically predicted FVC ) of the class B/2 secretin receptor family FVC PES in the HCS cohort , with the odds of being in the lowest decile decreasing by around 20% per standard deviation increase in FVC ( OR = 0 . 80 [95% CI: 0 . 68 , 0 . 93] , p=4 . 7×10−3 ) . All of the PES tested demonstrated small albeit significant correlations with genome-wide PGS at the same pT in the HCS cohort , with the exception of the extracellular matrix PES for which the correlation was relatively large ( r = 0 . 33 , Figure 3—figure supplement 2 ) . The higher correlation in this gene-set was probably due to the large number of genes involved ( >1000 ) . Interestingly , there was still a number of individuals with high genetically predicted lung function using a genome-wide PGS ( 90th percentile of HCS cohort ) but low genetically predicted lung function using one of the PES ( 10th percentile ) . Specifically , 12 . 17% and 12 . 05% of the HCS participants in the 90th percentile PGS for FVC and FEV1 respectively had a depleted PES ( 10th percentile , low predicted lung function by PES ) . Taken together , this suggests that pathway-based PGS provide distinct biological insights for some individuals with otherwise high genetic load of lung function increasing alleles , although the association between the class B/2 secretin family receptor PES and FVC after covariation for PGS did not survive multiple testing correction , and thus , these data require replication . The correlation between the expression of genes within each pathway encompassed by the PES and the PES profiles themselves could provide further support for their biological impact . We investigated the association between lung function PES and gene expression using RNA sequencing ( RNAseq ) on transformed lymphoblastoid cell lines ( LCL ) from 357 European individuals for which phase 3 whole-genome sequencing data was available from the 1000 genomes project ( Figure 3d , Supplementary file 3d–j; Lappalainen et al . , 2013 ) . We identified a significant association between the FVC PES class B/2 [secretin family receptors] and the expression of WNT3 using a strict FDR threshold q < 0 . 05 ( t = −3 . 53 , p=4 . 71×10−4 , q = 0 . 028 ) ; a more lenient FDR cut-off ( q < 0 . 1 ) yielded two more significant PES–gene expression correlations: FVC circadian clock PES and PPARA: t = −3 . 23 , p=1 . 37×10−3 , q = 0 . 07; FVC pathways in cancer and HSP90AB1: t = 3 . 72 , p=2 . 33×10−4 , q = 0 . 066 . Expression of WNT3 and PPARA was not associated with genome-wide PGS at the same p-value threshold ( p=0 . 63 and 0 . 29 ) , whilst the PGS exhibited a weaker , nominal relationship with HSP90AB1 ( p=0 . 04 ) . The remaining four PES tested ( FEV1 or FVC ) all demonstrated at least one nominal , uncorrected association ( p<0 . 05 ) . The observed effects of PES on gene expression at the population level were subtle; this is not surprising as each PES profile will encompass heterogenous variants for each individual , and thus , impacts on gene expression may be greater within specific genomic contexts . We performed a TWAS of the three lung function measures using SNP weights from lung and blood tissue . TWAS leverages models of genetically regulated expression to test for a correlation between predicted expression and a phenotype ( Gusev et al . , 2016 ) . Models of imputed expression derived from cis-eQTLs are generated from genes for which expression displays significant cis-heritability , that is , a significant genetic contribution to expression variance . We aimed to identify genes for which increased or decreased expression was associated with increased lung function and had approved compounds available which could improve lung function based on their mechanism of action ( Figure 4a ) . For instance , if increased expression of a gene was associated with improved lung function , then an agonist of that gene may be clinically useful or vice versa in the case of decreased expression . Using a Bonferroni threshold for the number of genes tested in lung and blood individually , we identified a number of transcriptome-wide significant genes as follows – FEV1: NGenes [Lung]=232 , NGenes [Whole blood]=201; FVC: NGenes [Lung]=222 , NGenes [Whole blood]=167 ( Supplementary file 3k–n , Figure 4b ) . The number of significant genes remained very similar using a more conservative threshold for Bonferroni correction that accounted for all genes in both tissues ( p<3 . 6×10−8 ) , which is conservative due to correlation between imputed models ( Supplementary file 3k–n ) . Transcriptome-wide associated genes were only retained if they were not also associated with a smoking phenotype to minimise residual smoking-related confounding . Specifically , we tested whether predicted expression of the genes which survived correction in the FEV1 or FVC TWAS was associated with smoking behaviour ( ‘ever vs . never smoked’ and ‘cigarettes per day’ ) in a TWAS using SNP weights from lung , blood , and two brain regions implicated in nicotine addiction ( dorsolateral prefrontal cortex and nucleus accumbens; Supplementary file 3m–v; Goldstein and Volkow , 2011; Scofield et al . , 2016 ) . We searched each of these significant genes in the DGIdb v3 . 0 . 2 to ascertain compounds which may improve lung function based on the direction of effect from the TWAS analyses . In accordance with the PES analyses , FEV1/FVC was not directly considered and we focused on FEV1 and/or FVC-associated genes which could be pharmacologically modulated ( Supplementary file 3w ) . Four candidate genes were identified satisfying tier one criteria: PPARD , ADORA2B , KCNJ1 , and AMT . For instance , decreased expression of potassium channel gene KCNJ1 was associated with FVC ( ZTWAS = −4 . 60 ) , and this channel can be inhibited by approved compounds such as the antidiabetic drug glimepiride . There were an additional seven genes with tier 2 investigational targets: PYGB , PIK3C2B , LINGO1 , APH1A , OPRL1 , MST1R , and ACVR2B . Probabilistic finemapping of these transcriptome-wide significant regions using a multi-tissue reference panel was then performed to prioritise whether these genes are likely causal at that locus . A credible set with 90% probability of containing the causal gene was computed for each locus utilising the marginal posterior inclusion probability ( PIP ) calculated from the observed TWAS statistics . We did not proceed with finemapping the PPARD locus due to its proximity to the defined boundaries of the major histocompatibility complex ( MHC ) region . Two FEV1-associated genes with tier 1 and/or tier 2 drug interactions , AMT and PYGB , were included in the credible set with a PIP >0 . 9 or nearing that threshold . Tetrahydrofolate is a co-factor for AMT ( ZTWAS = 5 . 96 , PIP = 0 . 893 , whole blood SNP weights ) , which has been previously implicated as having a beneficial effect on lung function . PYGB ( ZTWAS = −6 . 98 , PIP = 0 . 999 , lung SNP weights ) encodes a protein involved in glycogenolysis and can be putatively inhibited by the new exploratory treatment for respiratory failure , sivelestat ( Figure 4c ) . We acknowledge that the interaction between PYGB and sivelestat was derived from two public databases curated by DGIdb v . 3 . 0 . 2 ( Supplementary file 3w ) , and appropriate caution should be exercised in interpreting this relationship given that PYGB is not the primary target of sivelestat . Interestingly , there is evidence of a high-confidence biological interaction between PYGB and the gene that encodes the principal target of sivelestat via the STRING database ( ELA2 , neutrophil elastase ) . In addition , we tested a more conservative Bernoulli prior for each causal indicator ( p=1×10−5 ) but this only had a negligible effect on the PIP for either AMT ( PIP = 0 . 87 ) or PYGB ( PIP = 0 . 994 ) . Whilst there is a plausible role for AMT in respiratory biology ( aminomethyltransferase , involved in glycine cleavage ) , it should be noted that decreased predicted expression of AMT also trended towards the Bonferroni threshold for a significant association with smoking status ( ZTWAS = −4 . 33 , p=1 . 46×10−5 ) , although this was weaker for the cigarettes per day phenotype ( ZTWAS = −2 . 97 , p=2 . 94×10−3 ) . As a result , the association of this region with FEV1 should be treated cautiously until its biological relevance can be clarified to ensure that this signal is not driven by a residual effect of smoking . Respiratory viruses are an important contributor to acute , and potentially fatal , declines in lung function . We sought to investigate whether our proposed drug-repurposing candidates for lung function may also exhibit antiviral properties against these pathogens . The host–virus interactome was analysed for three respiratory viruses to perform computational drug repurposing – SARS-CoV2 , H1N1 , and the HAdV family ( Supplementary file 4a–c; Gordon et al . , 2020; Watanabe et al . , 2014; Martinez-Martin et al . , 2016 ) . Specifically , human proteins which are predicted to interact with virally expressed proteins ( ‘prey proteins’ ) were investigated to identify those which could be inhibited by existing drugs to potentially disrupt the progression of infection . Approved inhibitors or antagonists of proteins in each respective host–virus interactome were sourced using DGIdb and compared to our candidate compounds for lung function from the PES approach . Furthermore , we investigated the reported drug-label side-effect frequencies of each of these overlapping pharmacological agents and retained only candidates with no commonly reported ( >1% frequency ) respiratory adverse effects . There were three inhibitors of human proteins with evidence of interaction with a viral protein that also targeted a gene which was a member of a PES candidate gene-set . Vorinostat ( HDAC2 inhibitor ) and aminocaproic acid ( PLAT inhibitor ) both inhibited a SARS-CoV2 ‘prey protein’ and targeted a gene within the pathways in cancer and extracellular matrix ( ECM ) /ECM-associated proteins PES pathways , respectively . Similarly , ruxolitinib inhibits the influenza prey protein JAK1 , a part of the pathways in cancer gene-set . We caution that these pathways are quite broad in the biology that they encompass , and , as a result , the relevance of these drug–gene interactions to the pathways of interest warrants further study . We demonstrated using multiple lines of evidence a putative relationship between increased fasting blood glucose and lung function; therefore , we investigated whether any of the host–viral interactome members were enriched within biological pathways involved in glycaemic homeostasis . Interestingly , there was an overrepresentation of SARS-CoV2 ‘prey proteins’ amongst four gene-sets related to glucose metabolism , along with insulin and glucagon signalling pathways ( Table 4 ) . Fourteen SARS-CoV2 ‘prey proteins’ were members of at least one of these gene-sets , with a greater number of interactions amongst these genes than expected by chance ( p=4 . 42×10−12 , Supplementary file 4d ) . We outline evidence for the potential role of these viral prey genes in glycaemic homeostasis in Supplementary file 4d . These data support emerging evidence that SARS-CoV2-infected patients with hyperglycaemia are at higher risk of morbidity and mortality ( Kumar et al . , 2020 ) . None of the glycaemic ‘prey proteins’ were direct target of antidiabetic compounds; however , 57% of these proteins had a high-confidence protein–protein interaction with antidiabetic target gene ( Supplementary file 4e–f ) . For instance , GNB1 putatively binds with a SARS-CoV2 non-structural proteins ( Nsp7 ) that forms the part of the replicase/transcriptase complex , whilst this protein also demonstrated evidence of interacting with 15 proteins modulated by an antidiabetic compound , such as GLP1R , which is the primary target of GLP-1 analogues , including exenatide . Pharmacological interventions which seek to control blood glucose may have positive implications both in terms of improving baseline lung function and reducing the risk of adverse consequences after SARS-CoV2 exposure . We revealed candidate drug-repurposing opportunities to potentially improve pulmonary function and provide the means for aligning their application in individuals that carry a high relative burden of variants associated with their function . Through this process we identify glycaemic interventions in particular as being potentially beneficial in the context of respiratory infection . Our study suggests a causal relationship between blood glucose and lung function using a genome-wide ( LCV ) and IV ( MR ) approach , whilst downregulation of the glycogen phosphorylase PYGB was also associated with FEV1 after probabilistic finemapping of TWAS loci . These data support previous literature suggesting that declines in pulmonary function are overrepresented amongst individuals with diabetes and correlates with poor glycaemic control ( Walter et al . , 2003; Davis et al . , 2004; Gutiérrez-Carrasquilla et al . , 2019; van den Borst et al . , 2010 ) ; a phenomenon which has also been reported in non-diabetics ( McKeever et al . , 2005; Barrett-Connor and Frette , 1996 ) . There are a number of pathophysiological mechanisms postulated to underlie this relationship , including fibrosis mediated by hyperglycaemia-accelerated epithelial-to-mesenchymal transition ( Talakatta et al . , 2018 ) and aberrant inflammatory responses to dysglycaemia ( Mohanty et al . , 2000; Sun et al . , 2014 ) . Importantly , our data extends on these previous observational studies to provide novel evidence for a causal relationship . Respiratory sequalae after infection may also be significantly affected by dysregulation of glycaemic control . Acute hyperglycaemia is associated with a significant increase in morbidity and mortality amongst non-diabetic community-acquired pneumonia ( CAP ) patients , which further supports its utility as a treatment target ( Lepper et al . , 2012; Jensen et al . , 2017; Kornum et al . , 2007; McAlister et al . , 2005 ) . Notably , even patients with mild hyperglycaemia ( serum glucose 6–10 . 99 mmol/L ) have a purported elevated risk of death at 90 days following CAP diagnosis ( Lepper et al . , 2012 ) , whilst the association between type 2 diabetes and poor pneumonia outcomes appears to be driven by glycaemic control ( McAlister et al . , 2005 ) . Inflammation is likely to be an important component of glycaemic-influenced adverse effects; for instance , the intracellular carbohydrate O-linked β-N-acetylglucosamine has been recently linked to influenza-associated cytokine storms ( Wang et al . , 2020 ) . Future work should focus on the relevance of glycaemic biology to specific respiratory illnesses like asthma and COPD . Our findings supported the relevance of glycaemia to respiratory infection through demonstrating that proteins which putatively interact with the SARS-CoV2 virus were overrepresented in glycaemic pathways . Whilst the viral prey proteins we identified as members of glycaemic pathways were not the direct targets of antihyperglycaemic agents , some interact with these compounds , although the biological saliency of these interactions warrants future investigation . The presence of a viral–prey protein interaction also does not necessarily support its essentiality in the viral life cycle , and further data are needed to support this . Furthermore , the viral prey proteins overrepresented in the glycaemic pathways were mostly genes such as nucleoporins and cAMP-dependent protein kinases which have pleiotropic regulatory roles spanning a number of biological systems . It would also be interesting to further explore the relationship between the genetic architecture of fasting glucose and expression of these SARS-CoV2 prey proteins . These data taken together support the utility of managing blood glucose in the clinical improvement of respiratory outcomes . Targeted drug application and repurposing is by its very nature confounded by biological heterogeneity amongst individuals . This is likely particularly true in the case of complex traits as their polygenic genetic architecture provides the substrate for each individual to display a unique profile of trait-associated variation . In the second stream of this study , we stratified the polygenic architecture of lung function into a series of druggable pathways to provide a framework for pathway-specific genetic scores we designate the PES . We suggest that leveraging inter-individual genetic heterogeneity in this way will improve the precision application of novel drug repurposing . A number of interesting drug-repositioning candidates had overrepresented targets amongst the candidate PES gene-sets . For example , magnesium sulfate had enriched targets in the dilated cardiomyopathy PES and has previously shown promise as a repurposing candidate to improve pulmonary function in asthma ( Okayama , 1987; Hossein et al . , 2016 ) . Using an independent cohort , several PES profiles tested explained a small , but significant , percentage of variance in FEV1 and/or FVC . The class B/2 secretin family receptors score for FVC was noteworthy given that it remained nominally significant after an adjustment for genome-wide PGS . However , this did not survive multiple testing correction , and thus , further replication is needed to confirm this signal . Interestingly , this gene-set features a number of proteins involved with glycaemic homeostasis , including antidiabetic drug targets glucagon-like peptide receptor 1 ( GLP1R ) and amylin receptors ( RAMP1 , RAMP2 , and RAMP3 ) . While all of the PES demonstrated significant correlation with genome-wide PGS , in the majority of cases it was small ( r < 0 . 2 ) , suggesting that most of these functionally relevant foci of genomic risk in lung function GWASs were relatively independent of the total PGS . Importantly , we still identified individuals with high genetically predicted lung function using a genome-wide PGS but observed low predicted lung function with a pathway-specific PES . This was supported by the observed correlation between the PES and related mRNA expression which was distinct from a genome-wide PGS . Collectively , these data are consistent with the hypothesis that important treatment-related biology could be captured at a pathway level for individuals with or at risk of respiratory illness . The specific data from this study require future replication and validation in independent cohorts in order to provide greater support to our observed relationships between PES and spirometry measures . In addition , further study is warranted to dissect the signals encompassed by pathway-specific PGS , particularly in light of what would be observed amongst other pathways without drug targets or in gene-sets associated with other related traits . Taken together , our approach provides template for genetically informed precision drug repositioning to improve lung function . The clinical implementation in its most basic form would involve common variant genotyping using a commercial SNP array followed by imputation and lung function PES-based stratification of treatment options . This would be combined with other biochemical exposure measures , such as fasting glucose , that are causal risk factors and have approved treatments . To illustrate the clinical implementation of our strategy , we generated a schematic representation of individual heterogeneity in biochemical and genetic components of risk in lung function and related them to candidates for precision drug repositioning ( Figure 5 ) . We envisage that our approach to variant and exposure risk stratification can be applied more broadly to identify and implement precision drug repositioning in a range of complex traits . Whilst there are some potential confounds in the use of GWAS data for causal inference via both LCV models and MR , such as measurement error , population stratification , and horizontal pleiotropy , we are confident that the relationship between glycaemia and lung function presented in this study is robust given the multiple lines of support . Replicated , well-powered randomised controlled trials , however , are needed to fully resolve the clinical benefit of repurposing antihyperglycaemic compounds to improve lung function and in the context of viral infection . We also acknowledge that the direction of suitable pharmacological intervention is not inherently clear , such that an agonist or antagonist of genes within a pathway implicated by the PES approach is an important consideration ( Reay et al . , 2020 ) . Careful curation of the proposed repurposing candidates will therefore be critical , particularly in the context of pulmonary traits where a variety of currently approved compounds have adverse respiratory effects . We suggest that TWAS could be utilised to help overcome these issues by identifying druggable genes which are members of candidate PES gene-sets for which a clinically beneficial impact on expression can be predicted . These candidate genes derived from TWAS could in future be explored further using well-powered cohorts with genetic and transcriptomic that recorded spirometry measures . Interestingly , we also saw some evidence of cross-talk between heritable risk at genes associated with lung function and fasting glucose , with the downregulation of the glycogen phosphorylase PYGB ( associated with FEV1 ) observed through the probabilistic finemapping of TWAS loci . This study demonstrated a variety of methods for which genomic data could be utilised to propose drug-repurposing candidates , ranging from approaches which exploit genome-wide variant effects to the identification of candidate clinically significant drug–gene interactions . Lung function is a particularly relevant phenotype to study in this context as its aetiology is influenced by a variety of complex biological factors , and it is a significant contributor to global morbidity and mortality . Genetics-informed approaches will likely be increasingly useful to target novel respiratory interventions and reposition existing compounds . In future , genetics-based methods could be integrated with other clinical information to further enhance precision drug repurposing , whilst further consideration could be given to experimental compounds to enhance the number of repurposing opportunities . Our data strongly supported the efficacy of antihyperglycaemic compounds as repurposing candidates which could act as the impetus for further clinical investigation via randomised controlled trials . We obtained GWAS summary statistics for FEV1 , FVC , and their ratio from a meta-analysis of the UK Biobank sample with the SpiroMeta consortium cohorts as outlined extensively elsewhere ( N = 400 , 102 ) ( Shrine et al . , 2019 ) . Phenotypes were adjusted for age , age2 , sex , height , smoking status ( ever vs . never smoked ) , and genotyping array before the residuals were subjected to rank inverse-normal transformation . This GWAS was performed using European ancestry individuals . Bivariate LDSC regression was performed between each lung function trait and a variety of GWAS as implemented by LDhub v1 . 9 . 3 ( Zheng et al . , 2017 ) . Lung function summary statistics were cleaned ( ‘munged’ ) prior to LDSC using munge_sumstats . py and merged with common HapMap3 SNPs excluding the MHC region due to its LD complexity , as is usual practice ( Bulik-Sullivan et al . , 2015 ) . We retained estimates of genetic correlation ( rg ) for GWAS ( N = 172 ) with European ancestry and a heritability Z value >4 , as calculated by LDhub . When a phenotype had multiple GWAS , the GWAS with largest sample size was retained . The Bonferroni method was utilised for multiple testing correction with the significance threshold set as p<2 . 9×10−4 ( α = 0 . 05/172 ) . A heatmap was constructed using the ComplexHeatmap package ( Gu et al . , 2016 ) . LCV models were constructed between each measure of lung function which displayed a significant genetic correlation with a hormone or metabolite trait . The RunLCV . R and MomentFunctions . R scripts were leveraged to perform these analyses ( https://github . com/lukejoconnor/LCV ) . The LCV framework assumes that a latent variable , L , mediates the genetic correlation between two traits ( trait 1 , trait 2 ) and uses the mixed fourth moments of the bivariate effect size distribution to estimate the mean posterior GCP as described in detail by O'Connor and Price , 2018 . The GCP estimate quantifies the magnitude of genetic causality between the two traits . GCP values range from −1 to 1 ( full genetic causality ) ; within these limits , positive values indicate greater partial genetic causality of trait 1 on 2 , and vice versa for negative values . All traits were munged prior to LCV analyses , with only HapMap3 SNPs ( minor allele frequency [MAF] >0 . 05 ) outside the MHC region retained in accordance with the LDSC analyses . We utilised the baseline 1000 genomes phase 3 LD scores for HapMap3 SNPs ( MHC excluded ) . A two-sided t-test was used to assess whether the estimated GCP was significantly different from zero . We investigated the causal effect of fasting glucose on both FEV1 and FVC using two-sample MR . MR is underpinned by the use of genetic variants as IVs , with the random inheritance of these IVs as per Mendel’s laws facilitating the use of IVs to perform causal inference between an exposure and outcome , providing a series of assumptions are met ( Burgess et al . , 2017 ) . We defined IVs as independent variants which are associated with fasting glucose using the traditional GWAS genome-wide significance threshold ( p<5×10−8 , r2 <0 . 001 , palindromic SNPs removed ) . A different GWAS of fasting glucose was utilised for MR than for LDSC and LCV . Scott et al . performed a replication of ~66 , 000 Illumina CardioMetabochip variants following the Manning et al . GWAS for which more complete summary statistics were available , and thus , the latter was included in the LDhub catalogue instead of the former ( Manning et al . , 2012; Scott et al . , 2012 ) . We required only genome-wide significant SNPs for MR; therefore , the Scott et al . CardioMetabochip replication was more suitable as this was a larger sample size than the Manning et al . GWAS . Fasting glucose data for GWAS were obtained from either plasma or whole blood of non-diabetic individuals of European ancestry and corrected to plasma levels ( N = 133 , 310 , unit of effect = mmol/L ) ( Scott et al . , 2012 ) . Our primary MR model was an IVW effect model with multiplicative random effects ( Burgess et al . , 2013 ) . Further , we implemented a weighted median model which takes the median of the ratio estimates ( as opposed to the mean in the IVW model ) , such that upweighting was applied to ratio estimates with greater precision ( Bowden et al . , 2016 ) . An MR–Egger model was then constructed; an adaption of Egger regression wherein the exposure effect is regressed against the outcome with an intercept term added to represent the average pleiotropic effect ( Bowden et al . , 2015 ) . In addition , we examined evidence of reverse causality by using the MR Steiger directionality test ( Hemani et al . , 2017 ) . We also tested whether the Egger intercept is significantly different from zero as a measure of unbalanced pleiotropy . In addition , heterogeneity amongst the IV ratio estimates was quantified using Cochran’s Q statistic , given that horizontal pleiotropy may be one explanation for significant heterogeneity . A global pleiotropy test was also implemented via the MR PRESSO framework ( Verbanck et al . , 2018 ) . Leave-one-out analyses were then performed to assess whether causal estimates are biased by a single IV , which may indicate the presence of outliers , and the sensitivity of the estimate to said outliers . However , outliers may not necessarily be evidence of horizontal pleiotropy . We performed a phenome-wide association study for each of these ‘outlier’ SNPs using summary data collated by GWAS atlas v20191115 to assess evidence of horizontal pleiotropy , that is , acting through non-glycaemic pathways to influence lung function ( Watanabe et al . , 2019 ) . All MR analyses were performed in R version 3 . 6 . 0 using the TwoSampleMR v0 . 4 . 25 and MRPRESSO v1 . 0 packages . We investigated whether a residual effect of smoking could confound the link between glucose and lung function . Firstly , we selected two well-powered GWAS of smoking behaviours: ever vs . never smoked ( N = 385 , 013 ) ( Watanabe et al . , 2019 ) and cigarettes smoked per day ( N = 263 , 954 ) ( Liu et al . , 2019 ) . Genetic correlation between these two smoking phenotypes and fasting glucose was estimated as described above , followed by the construction of an LCV model . The MR IVs utilised for fasting glucose were also checked for association with each smoking GWAS . We also probed whether there could be an effect of collider bias in the event smoking does indeed exert an effect on fasting glucose . To this end the LCV and MR analyses was repeated for fasting glucose using smaller UK Biobank GWAS of FEV1 and FVC from the Neale Lab as it was not adjusted for smoking ( N = 272 , 338 , http://www . nealelab . is/uk-biobank ) . We implemented gene-set association using MAGMA method ( MAGMA v1 . 06b ) , with some customisations to the framework to identify candidate PES gene-sets ( Reay et al . , 2020; de Leeuw et al . , 2015 ) . These gene-sets became the basis to calculate pathway-specific polygenic scores ( PES ) . MAGMA aggregates SNP-wise p-values for trait association into a gene-based p-value and , thereafter , tests whether a set of genes is more strongly associated with the phenotype than all other genes . Gene-based test statistics were calculated analogous to Brown’s method , which is applicable to dependent p-values with known covariance ( as common SNPs display through the phenomenon of linkage disequilibrium [LD] , which can be quantified at a population level ) . p-Value thresholding ( PT ) was utilised for the gene test statistic calculation; four p-values were selected: all SNPs , PT <0 . 5 , PT <0 . 05 , and PT <0 . 005 , meaning only SNPs below these thresholds were included in the gene-based model . We argue that distinct biological processes in individuals may only be captured when the optimal spectrum of polygenic variation is included in the model . A variety of PT could be utilised; for simplicity , we selected the four p-values thresholds described , as per our previous work ( Reay et al . , 2020 ) . We mapped variants to 18297 autosomal genes in hg19 assembly defined by NCBI and obtained from the MAGMA website – genes within the MHC were removed due to the complexity of LD within this region . The 1000 genomes phase 3 European reference panel was utilised to define LD for input into MAGMA . Genic boundaries were extended to capture regulatory variation , with both conservative and liberal upstream and downstream boundary definition implemented . An extension of 5 kb upstream of the gene and 1 . 5 kb downstream was the conservative construct , whilst a larger 35 kb upstream and 10 kb downstream was the liberal construct . Boundaries were longer upstream of the gene in both instances to capture more promoter-related variation , as is usual practice ( Wray et al . , 2018; Kunkle et al . , 2019; Reay and Cairns , 2020 ) . Genic p-values were transformed to Z-scores with the probit function for input into the gene-set association model . Competitive gene-set association was undertaken by a linear regression model whereby genic Z-scores are the outcome and confounders including gene size and genic minor allele count included as covariates . When these models are constructed at different PT , this approach constitutes testing whether the gene-set is more associated than the other genes , for which test statistics were calculated only including SNPs below the threshold . We selected pathways that survived multiple testing correction for an enrichment of lung function-associated variation relative to all other genes at that threshold by applying correction via the Benjamini–Hochberg ( BH ) method ( FDR < 0 . 05 ) to all thresholds combined . These associations can be interpreted based on the p-value threshold for the model , for example , at gene-set which survives FDR correction that includes only variants which displayed a nominally significant univariable association with lung function ( p<0 . 05 ) is indicative of a set of genes that are more associated with lung function than all other genes with at least one SNP that had p<0 . 05 in the GWAS . The BH approach was implemented rather than Bonferroni as several gene-sets will be tested multiple times at different p-value thresholds , and thus , the assumption of independence underlying Bonferroni correction likely means this would be overly conservative . In summary , gene-based test statistics were constructed at four different p-value thresholds , whereby only SNPs below the said threshold were included in the gene-based test statistic . Thereafter , competitive gene-set association is conducted for each druggable pathway at the different thresholds , with the null hypothesis being that the druggable pathway is no more associated with the trait ( enriched with association ) than all other genes for which gene-based p-values could be calculated by virtue of having an SNP below the threshold annotated to it . The concept underlying this is that distinct pathways may be enriched with common variants at differing levels of the polygenic signal , for example , a model including all SNPs will identify gene-set enriched with association relative to all other genes , whilst a less polygenic model , like a threshold of p<0 . 05 , will capture gene-sets enriched with association relative to genes with at least one SNP mapped to it with a univariate association p<0 . 05 . We defined gene-sets with known drug targets by sourcing hallmark and canonical ( BioCarta , KEGG , PID , and Reactome ) from the Molecular Signatures Database ( MSigDB ) ( Liberzon et al . , 2015 ) and retaining those with at least one gene with a high-confidence interaction with at least one approved pharmacological agent ( TClin genes ) , as annotated using the Target Central Resource Database ( TCRD v6 . 1 , NGenes = 613 ) ( Oprea et al . , 2018 ) . We tested each candidate PES gene-set for overrepresentation of DrugBank compound targets using WebGestaltR v0 . 4 . 2 ( Liao et al . , 2019 ) . Compounds were retained for each pathway if they survived FDR correction ( q < 0 . 05 ) and were FDA approved . Single-drug gene matching was performed using the DGIdb v . 3 . 02 , with a minimum of three lines of supporting evidence the criterion for selection ( Cotto et al . , 2018 ) . The list of FDA-approved DrugBank compounds which were overrepresented targets in a PES candidate gene-set was reviewed by a pharmacist to prioritise potential useful compounds for lung function . A total of eight topical compounds were excluded . The remaining 55 oral and/or parenteral compounds were investigated for lung function-related adverse events ( including all of dyspnoea , abnormal breath sounds , decreased respiratory rate , orthopnoea , shallow breathing , respiratory distress , respiratory depression , or any other related term ) , other alarming adverse events , important precautions , black-box warnings , or any contraindication that might prohibit the drug use in our study population . These data were reviewed for each compound using the following databases: drugs . com , Medscape , SIDER v4 . 1 , and the summaries of each product’s characteristics . We also searched for articles that discussed either an improvement or worsening in the lung functions for each compound along with the allowed paediatric age use . The drugs were then categorised into one of five categories ( Figure 3—figure supplement 1 , Supplementary file 2d ) . Level 1 was assigned for an oral or parenteral formulation , with no documented respiratory side effects and with positive evidence of prior use for lung function in the literature . Level 2 was assigned for an oral or parenteral formulation , with no documented or rare ( <1% ) respiratory side effects and with/without positive evidence but no negative evidence of prior use for lung function in the literature . Level 3 was assigned to an oral or parenteral formulation , with common ( 1–15% ) respiratory side effects and with/without positive evidence but no negative evidence of prior use for lung function in the literature . Level 4 compounds were those oral or parenteral formulations with very common ( 16–50% ) respiratory side effects or other alarming adverse effects unrelated to respiratory function , without positive evidence but with/without negative evidence of prior use for lung function in the literature . Finally , level 5 was assigned when the drug was associated with a serious adverse event ( including a black-box warning or an absolute contraindication ) . We defined the model to calculate PES profiles for individuals as follows ( Equation 1 ) . Consider j SNPs for i individuals , wherein the SNPs are those physically mapped to genes which are members of a candidate PES gene-set ( m ) . Let β^j denote the statistical effect size for each variant from the GWAS , multiplied by its dosage Gij . The SNPs included were those below the p-value threshold utilised to discover the gene-set . ( 1 ) PESi= ∑i=1mβ^jGij We averaged these scores by the number of SNPs carried by each individual and scaled them using the scale ( ) function in R . PES profiles were generated in all instances by first filtering the GWAS summary statistics for common variants ( MAF >0 . 01 ) within the genic boundaries of variants which comprise the PES gene-set . The genic boundaries were extended using the liberal or conservative configuration , dependent on which boundary definition was utilised in the gene-set association for that pathway . PRSice v2 . 2 . 12 calculated the respective PES , along with genome-wide PGS ( using the same additive model but genome wide ) for FEV1 and FVC ( Choi and O'Reilly , 2019 ) . We utilised an independent , genotyped cohort for which spirometry measures were recorded to investigate the phenotype relevance of PES profiles for lung function . Participants were drawn from the HCS , a population-based cohort of individuals aged between 55 and 85 years , predominantly of European ancestry and residing in Newcastle , New South Wales , Australia . All work was conducted in accordance with ethics committee approvals . Consenting participants completed a series of questionnaires , attended a clinic visit , and provided blood samples . Individuals were recruited by random selection from the New South Wales State electoral roll with detailed recruitment and data collection methods for the HCS described elsewhere ( McEvoy et al . , 2010 ) . Participants provided blood samples from which DNA was extracted and genotyped using the Affymetrix Axiom Kaiser array . Quality control excluded SNPs with genotype call rate of <0 . 95 , deviation from Hardy–Weinberg equilibrium ( p<1×10−6 ) or MAF of <0 . 01 . The input for relatedness testing and removal of population outliers were autosomal , common ( MAF >0 . 05 ) , physically genotyped SNPs in relative linkage equilibrium ( r2 <0 . 02 ) , with regions of long-range LD removed , as is usual practice ( Price et al . , 2008 ) . We used PLINK 1 . 9 to retain only unrelated individuals ( pi_hat >0 . 185 ) , with one participant from each related pair blinded to phenotype information . Population outliers were determined by performing principal component analysis ( PCA ) using PLINK 1 . 9 . We clustered individuals in the HCS with the first two principal components from each 1000 genomes phase 3 superpopulation using k-means clustering . Thereafter , we conservatively excluded any HCS individual with a first or second principal component above or below the maximum or minimum 1000 genomes European values for these eigenvectors . PCA was repeated in the filtered European ancestry HCS subset such that eigenvectors could be used as downstream covariates . Imputation to the Haplotype Reference Consortium panel involved a series of steps and additional data clean up , reference lift over to the hg19/GRCh37 , and data submission to the Michigan imputation server , as specified in the submission guidelines ( Loh et al . , 2016; Das et al . , 2016 ) . Post-imputation quality control was as follows: imputation R2 >0 . 8 , MAF >0 . 01 , and missingness <0 . 02 . We retained common variants ( MAF >0 . 01 ) with high imputation quality ( R2 >0 . 8 ) . Spirometry data from the HCS was then processed by selecting individuals with non-missing FEV1 and FVC . We utilised the maximum FEV1 and FVC from four attempts and fitted a linear model which covaried for sex , age , age2 , height , height2 , smoking status , self-reported asthma status , and self-reported bronchitis/emphysema status . The phenotype for association testing was residuals from these models transformed via inverse-rank normalisation ( Blom transformation ) using the RNOmni package . We tested the association between a genome-wide PGS for FEV1 and FVC ( PT <1 , 0 . 5 , 0 . 05 , 0 . 005 , 5 × 10−5 , 5 × 10−8 ) with their respective transformed spirometry indices adjusted for the first five SNP-derived principal components using PRSice v2 . 2 . 12 . Similarly , the association between each of the PES profiles with an overrepresentation of FDA-approved drug targets and FEV1 and/or FVC was investigated using the same approach; however , we only constructed the PES at the p-value for which it demonstrated the strongest association signal after multiple testing correction in the GWAS . We further adjusted each of these models for genome-wide PGS at the same PT for which the PES was calculated . We obtained RNAseq normalised read counts ( PEER normalised RPKM ) for 23723 genes which survived QC in the Geuvadis dataset ( https://www . ebi . ac . uk/arrayexpress/experiments/E-GEUV-1/files/analysis_results/ ? ref=E-GEUV-1 ) . The Geuvadis project performed RNAseq on transformed LCL for participants in the 1000 genomes project ( Lappalainen et al . , 2013 ) . We retained 357 European individuals in this dataset for which phase 3 sequencing data was available from the 1000 genomes . The association between normalised mRNA expression for genes part of the candidate gene-set and each PES was tested using a linear model , adjusted for sex , the first three SNP-derived principal components , and genome-wide PGS at the same PT utilised to calculate the PES . Multiple testing correction was applied for the number of genes in each set via the BH method using the p . adjust ( ) function . A TWAS of each lung function measure was performed using the FUSION software ( Gusev et al . , 2016 ) . SNP weights were derived for genes with a significant contribution of cis acting SNPs to expression variability ( cis-h2p<0 . 01 ) using lung and whole blood RNAseq GTEx v7 data . A transcriptome-wide significant gene was defined by accounting for the number of genes with models of genetically regulated expression in lung and whole blood , respectively – lung: p<6 . 43×10−6 ( α = 0 . 05/7776 ) ; whole blood: p<8 . 32×10−6 ( α = 0 . 05/6007 ) . A more conservative threshold could be applied which corrects for all models in both tissues ( p<3 . 6×10−6 ) ; however , given the correlation between models and the discovery nature of this study , we chose the more liberal correction threshold . We excluded genes within the MHC region due to its LD complexity . Furthermore , we subjected two smoking behaviour phenotypes to TWAS to uncover associations which could be driven by residual effects of smoking . This is inherently conservative as it is possible that genes associated with both lung function and smoking behaviours could exhibit pleiotropic effects; however , as we wish to define drug targets relevant to lung function , the exclusion of these shared genes is warranted . The smoking phenotypes were ‘ever vs . never smoked’ and ‘cigarettes smoked per day’ , and TWAS was performed using lung and blood for consistency , along with SNP weights from the dorsolateral prefrontal cortex and nucleus accumbens , as these brain regions have been implicated in nicotine addiction . Genes which survived the above were searched using DGIdb , with the following criteria utilised to define gene-target pairs , where the drug mode of action matched the sign of the TWAS Z value: ( i ) tier 1 – FDA-approved compound with at least two lines of evidence for interacting with the target gene; ( ii ) tier 2 – investigational compound ( not FDA approved ) with at least two lines of evidence for interacting with the target gene . The TWAS Miami plots were generated using an edited version of the TWAS-plotter . V1 . 0 . R script ( https://github . com/opain/TWAS-plotter ) . A Bayesian method FOCUS was utilised to finemap the TWAS associations which could be therapeutically useful ( tier 1 or 2 ) ( Mancuso et al . , 2019 ) . Given observed TWAS statistics , the marginal posterior inclusion probability ( PIP ) was calculated and subsequently used to compute a credible set with 90% probability ( ρ ) of containing the causal gene ( ci=1 ) . As FOCUS allows the null model to be predicted as a possible member of the credible set , we excluded any genes for which that occurred . The credible set ( S ) was defined by summing normalised PIP such that ρ was exceeded , sorting the genes and then including those genes until at least ρ of the normalised-posterior mass is explained ( Equation 2 ) . ( 2 ) S {Gene1 , … , Genek}=∑i=1kPIP ( ci=1\|ZTWAS ) ≥ ρ The Bernoulli prior for each causal indicator was set as the default p = 1×10−3 , with a default prior variance for effects at causal genes set as 40 ( nσc2=40 ) . Previous work has demonstrated that FOCUS-computed PIPs were robust to different specified prior variances ( Mancuso et al . , 2019 ) ; however , we further utilised a more conservative prior of p = 1×10−5 to assess the effect on the PIP calculated for candidate druggable genes . In all instances , we utilised a multi-tissue panel obtained from FOCUS GitHub repository which combines GTEx v7 SNP-weights with other FUSION TWAS weights ( https://github . com/bogdanlab/focus/wiki , GTEx v7 with METSIM , CMC , YFS , and NTR ) . The marginal TWAS Z to use for finemapping for each locus was selected in the tissue for which the gene was found to be associated via the FUSION TWAS methodology ( lung or blood ) , if available , otherwise by predictive accuracy ( cross-validated R2 ) . We selected three respiratory viruses for which host–viral protein interaction data was previously published: SARS-CoV2 , H1N1 , and the HAdV family . The host–SARS-CoV2 interactome was defined using affinity-purification mass spectrometry ( NGenes = 332 , MiST score ≥0 . 7 , a SAINTexpress BFDR ≤0 . 05 ) ( Gordon et al . , 2020 ) . We selected 91 proteins which both interact with viral proteins expressed by influenza ( mass spectrometry ) and siRNA-mediated downregulation-reduced viral replication in cultured cells by at least three log10 units while retaining >80% cell viability ( Watanabe et al . , 2014 ) . Finally , the HAdV–host interactome was defined using a protein microarray platform ( NGenes = 24 ) , which encompasses 20 viral proteins encoded by five HAdV species ( Martinez-Martin et al . , 2016 ) . We investigated approved inhibitors or antagonists of these genes using DGIdb as described above in the PES candidate gene-set drug repurposing section . The sets of genes which interact with viral proteins for each virus ( ‘viral prey proteins’ ) were subjected to overrepresentation analysis using the GENE2FUNC function of FUMA ( Watanabe et al . , 2017 ) . We selected gene-sets which survived multiple testing correction ( q < 0 . 05 ) , which contained at least one of the following key terms related to glycaemic biology: glucose , insulin , diabetes , or glucagon . Further , we investigated whether there was a significant overrepresentation of interactions amongst these viral prey proteins overlapping a glycaemic pathway using STRING v11 . 0 ( Szklarczyk et al . , 2019 ) . We assembled a list of antidiabetic drug targets by searching compounds annotated with the level 2 ATC code A10 ( drugs used in diabetes ) in DGIdb , retaining drug–gene interactions with two or more lines of evidence . The interactions between these drug target proteins and the glycaemic SARS-CoV2 prey proteins were investigated once more using STRING , with only interactions scoring >0 . 75 considered .	Chronic respiratory disorders like asthma affect around 600 million people worldwide . Although these illnesses are widespread , they can have several different underlying causes , making them difficult to treat . Drugs that work well on one type of respiratory disorder may be completely ineffective on another . Understanding the biological and environmental factors that cause these illnesses will allow them to be treated more effectively by tailoring therapies to each patient . Reduced lung function is a factor in respiratory disorders and it can have many genetic causes . Studying the genes of patients with reduced lung function can reveal the genes involved , some of which may already be targets of existing drugs for other illnesses . So , could a patient’s genetics be used to repurpose existing drugs to treat their respiratory disorders ? Reay et al . combined three methods to link genetics and biological processes to the causes of reduced lung function . The results reveal several factors that could lead to new treatments . In one example , reduced lung function showed a link to genes associated with high blood sugar . As such , treatments used in diabetes might help improve lung function in some patients . Reay et al . also developed a scoring system that could predict the efficacy of a treatment based on a patient’s genetics . The study suggests that COVID-19 infection could be affected by blood sugar levels too . Chronic respiratory disorders are a critical issue worldwide and have proven difficult to treat , but these results suggest a way to identify new therapies and target them to the right patients . The findings also support a connection between lung function and blood sugar levels . This implies that perhaps existing diabetes treatments – including diet and lifestyle changes aimed at reducing or limiting blood sugar – could be repurposed to treat respiratory disorders in some patients . The next step will be to perform clinical trials to test whether these therapies are in fact effective .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "genetics", "and", "genomics" ]	2021	Genetic association and causal inference converge on hyperglycaemia as a modifiable factor to improve lung function
Formation of a regularly branched blood vessel network is crucial in development and physiology . Here we show that the expression of the Notch ligand Dll4 fluctuates in individual endothelial cells within sprouting vessels in the mouse retina in vivo and in correlation with dynamic cell movement in mouse embryonic stem cell-derived sprouting assays . We also find that sprout elongation and branching associates with a highly differential phase pattern of Dll4 between endothelial cells . Stimulation with pathologically high levels of Vegf , or overexpression of Dll4 , leads to Notch dependent synchronization of Dll4 fluctuations within clusters , both in vitro and in vivo . Our results demonstrate that the Vegf-Dll4/Notch feedback system normally operates to generate heterogeneity between endothelial cells driving branching , whilst synchronization drives vessel expansion . We propose that this sensitive phase transition in the behaviour of the Vegf-Dll4/Notch feedback loop underlies the morphogen function of Vegfa in vascular patterning . The formation and maintenance of adequately branched and hierarchically organised networks of blood vessels is critical for all aspects of tissue growth and physiology ( Potente et al . , 2011 ) . How the endothelial cells lining blood vessels collectively determine whether to form a new vessel branch or expand existing vessels is currently not understood . Similarly , the morphogenic and cellular principles underlying the chaotic vascular network formation in disease scenarios remain unclear . Recent work established that Dll4/Notch signalling between the activated endothelial cells functions to amplify stochastic differences in expression levels of the vascular endothelial growth factor receptors ( Vegfr ) , ultimately establishing tip cells bearing high Vegfr expression , and inhibited stalk cells , bearing lower levels of signalling receptors ( Jakobsson et al . , 2010; Hellström et al . , 2007; Suchting et al . , 2007; Phng and Gerhardt , 2009; Lobov et al . , 2007 ) . The Dll4/Notch pathway thereby establishes a lateral-inhibition feedback loop with the Vegf-Vegfr pathway , controlling branching frequency by balancing the number of new tip cells with the number of stabilizing stalk cells ( Phng and Gerhardt , 2009 ) . Genetic mosaic experiments in 3D embryoid body sprouting assays and in vivo illustrated tip and stalk cells regularly change position as endothelial cells dynamically compete for the tip position ( Jakobsson et al . , 2010 ) . Every 3–5 hr , the tip cell is exchanged , leading to continuously changing cellular neighbourhood relationships . Computational modelling of the feedback loop illustrated that it should suffice to pattern new sprouts , and comprise a robust mechanism for the establishment of a regular branching pattern ( Bentley et al . , 2008; 2009 ) . Recently integrated modelling and experiments further identified that this Notch/VEGF feedback drives differential VE-cadherin dynamics at individual endothelial junctions contributing to cell rearrangement behaviour ( Bentley et al . , 2014a ) . The meeting of new neighbours itself propagates iterative lateral-inhibition . Whether and how this activity might affect Notch signalling in individual cells over time remains unclear . In other cell systems , Notch signalling is highly dynamic , and Notch target genes oscillate both in single cells through autonomous feedback regulation , as well as in cell collectives ( Masamizu et al . , 2006 ) . The latter is best studied during formation of somites from the presomitic mesoderm , in which Notch activity synchronizes intrinsic oscillatory gene regulation between neighbouring mesoderm cells ( Kageyama et al . , 2007; Jiang , 2000; Hubaud and Pourquié , 2013 ) . In the absence of Notch signalling , cells eventually drift out of synchrony , finally leading to loss of somite patterning . Dll4 over-expression in tumour cells in a glioblastoma model caused substantial vessel enlargement at the expense of branching ( Li et al . , 2007 ) . Constitutive genetic endothelial activation of Notch4 leads to embryonic vessel diameter increase and arterio-venous shunt formation , further suggesting that controlled Notch activity plays a role in vessel diameter regulation and adequate network remodelling ( Uyttendaele et al . , 2001 ) . Collectively , existing studies show that reduced Notch signalling in endothelial cells promotes a highly branched network with small calibre vessels whereas increased Notch activity promotes a sparse network of large calibre vessels . Similarly , Vegfa shows strong dosage dependent effects on vascular patterning . Like Dll4 , Vegfa is genetically haplo-insufficient , and overexpression causes embryonic lethality ( Miquerol et al . , 2000; Carmeliet et al . , 1996 ) . Surprisingly , despite the extensive body of work on Vegf and Dll4/Notch , our understanding of the principles and mechanisms that underlie these exquisitely dose sensitive effects on vascular patterning have hardly progressed beyond phenomenology . This may in part be because of the difficulties in analysing Vegf and Dll4/Notch signalling in a quantitative and dynamic manner , especially in vivo . Here , we developed in vitro and in vivo analysis of Dll4 mRNA , protein and gene expression reporter dynamics under normal and pathological Vegfa stimulation , identifying a phase transition in the Dll4 dynamics that determines whether new vessels branch or expand . Computational modelling previously predicted that the Vegf-Dll4/Notch-Vegfr feedback loop normally establishes salt-and-pepper patterning between endothelial cells to regulate tip/stalk specification , but under elevated Vegfa levels , simulations predicted that this feedback loop would switch to drive the cells to collectively fluctuate their Dll4 levels in contiguous clusters , unable to stabilize into a heterogeneous pattern ( Bentley et al . , 2009 ) . This highlights how the non-linear feedback involved in Vegf/Notch signalling can make it extremely hard to intuit how perturbation conditions , such as elevated Vegf , will impact on dynamics . Importantly , clear experimental evidence for the predicted dynamics and changing behaviours has been difficult to obtain . Further more , the computational models contain a limited parameter set , thus simplifying the complexity , potentially missing critical modifiers . Such modifiers may not only be molecular components , but also effects that originate from differences in cell shape and geometries , as these can trigger changes to signalling pathway dynamics ( Bentley et al . , 2009; 2014b ) . In the present study , we therefore chose to combine and compare refined computational models that reflect the experimental assays and their endothelial geometries and integrate specific experimental assays and computational modelling throughout . Using high Vegfa levels in embryoid body assays , intraocular injection of Vegfa , the oxygen induced retinopathy model of ischemia driven ocular neovascularization , and finally syngenic mouse glioblastoma tumours , we present evidence for local Notch-dependent synchronization of Dll4 dynamics leading to vessel expansion whilst disrupting branching . In order to gain first experimental insight into the dynamic behaviour of Dll4/Notch signalling under normal versus elevated Vegf conditions , we performed a time course experiment on endothelial monolayers . We collected mRNA from endothelial monolayers treated with either 50 ng/ml Vegfa 164 ( normal ) or 1 µg/ml Vegfa 164 ( high ) ( Figure 1e–i ) . We monitored dll4 mRNA levels by qPCR over a period of 9 and 24 hr post-stimulation . High Vegfa consistently induced fluctuations with high amplitude and several peaks ( Figure 1f , i ) , which given the population based measurement indicates the cells are fluctuating in relative synchrony . Lomb-Scargle analysis ( Dequéant et al . , 2006 ) showed that the dominant periodicity in each dataset was 5–6 hr . The modest and varying degree of confidence in this analysis however suggests that these dynamic patterns in vitro are inherently noisy . Under normal Vegfa levels , dll4 mRNA showed an unexpected low-amplitude rise and decline , but then remained relatively unchanged ( Figure 1e ) . We had hypothesized these conditions should permit a stabilized salt and pepper pattern , manifested as a stable population level of dll4 . To investigate whether this observation contests our current working model of Vegf/Notch feedback or if it can be explained simply by the different geometric nature of cells in a monolayer compared to a sprouting system impacting on the ensuing signalling dynamics , we simulated the Vegf/Notch feedback in a 2D monolayer geometry ( see Materials and methods ) . This model revealed that the restricted ability of cells to change shape in a monolayer ( without filopodia extensions and Vegf gradients ) destabilizes the salt and pepper pattern; they instead fluctuate between salt-and-pepper pattern-like states and then near-uniform Notch activation across the cells ( Figure 1a , b and Video 1 ) . Simulation of higher levels of Vegfa in this simple model led to high amplitude , synchronized oscillations of Dll4 ( Figure 1c , d and Video 2 ) . 10 . 7554/eLife . 12167 . 003Figure 1 . Endothelial Dll4 mRNA levels fluctuate dynamically in a Vegfa and Notch dependent manner . ( a–d ) Simulation of Dll4 dynamics in a monolayer of 20×20 cells using the memAgent-Spring model under normal Vegfa and high Vegfa ( 20x normal level ) . ( a ) The total Dll4 level across all cells fluctuated overtime with a regular periodicity even though normal levels ( b ) generated a transient salt and pepper pattern with many non-adjacent cells high in Dll4 at high points in the fluctuation and few in the low phase . ( c ) High Vegf caused the cells to synchronise with the two phases of behavior showing either ( d ) most adjacent cells low , or most high in Dll4 . Dll4 level low to high are represented purple-green . ( e–i ) Quantitative real time PCR analysis of dll4 mRNA levels in bEND5 cell monolayer . Cells were starved for four hours with serum-depleted medium and then stimulated with medium supplemented with either 50 ng/ml ( e ) , 1 μg/ml ( f , i ) , 0 Vegf ( g ) , or 1 μg/ml Vegf and 50 µM DAPT ( h ) . Cell lysates were collected every hour for the times indicated in the graphs . Values represent means ± S . D of technical replicates . DOI: http://dx . doi . org/10 . 7554/eLife . 12167 . 00310 . 7554/eLife . 12167 . 004Video 1 . Monolayer simulation normal Vegf . DOI: http://dx . doi . org/10 . 7554/eLife . 12167 . 00410 . 7554/eLife . 12167 . 005Video 2 . Monolayer simulation high Vegf . DOI: http://dx . doi . org/10 . 7554/eLife . 12167 . 005 To confirm that the fluctuations observed in vitro are indeed Notch regulated , we utilized the gamma-secretase inhibitor DAPT , a potent inhibitor of Notch signalling ( Hellström et al . , 2007 ) , which completely abolished the fluctuations of dll4 levels under high Vegfa ( Figure 1g , h ) . Taken together these results suggest that high Vegfa levels synchronize contiguous endothelial cells in their fluctuating expression of Dll4 , leading to dynamic and recurring fluctuations that require Notch activity . To gain insight into single cell versus population dynamics of Dll4 in vascular sprouting conditions , we generated a novel dynamic fluorescent reporter for Dll4 expression ( Figure 2 ) . Given the lack of detailed knowledge on the transcriptional regulation of Dll4 , we used a BAC clone containing the entire genomic locus of mouse Dll4 ( RP23_46P4 , BACPAC CHORI ) and replaced the stop codon in exon 11 with a viral self cleaving P2A peptide sequence ( Hsiao et al . , 2008 ) , followed by a destabilized version of the yellow fluorescent protein Venus ( Li , 1998; Corish and Tyler-Smith , 1999 ) and a selection cassette flanked by loxP sites by recombineering ( Yu et al . , 2000 ) ( Figure 2—figure supplement 1a , b ) . We selected several ES cell lines carrying one , two or more copies of the transgene either replacing the endogenous Dll4 allele ( knock-in ) , or integrated in distant sites . Heterozygous knock-in ES cells and those carrying one distant copy of the Dll4-P2A-dVenus construct gave rise to high degree chimeras and germ line transmission , effectively establishing colonies of viable and fertile mice . dll4 mRNA and protein levels appeared unaffected by the P2A-dVenus cassette ( not shown ) . ES cells and mice carrying one additional copy showed expectedly 1 . 5 times the normal dll4 mRNA levels ( Figure 2—figure supplement 1d ) . Studying dVenus protein expression and Dll4 protein expression in ES cells and in mouse retinas from this strain showed perfect correlation between cells expressing dVenus and Dll4 protein ( Figure 2a–d , g , h ) . 10 . 7554/eLife . 12167 . 006Figure 2 . Dynamic and stable genomic reporters of Dll4 expression show differential distributions . ( a–f ) Representative confocal overview ( a , c , e ) and high magnification ( b , d , f ) images of vascular sprouts in WT ( a , b ) , 3Dll4-dVenus ( c , d ) , 3Dll4-Emerald ( e , f ) embryoid bodies immunolabelled with antibodies specific to Dll4 ( white in a , c , e and red in b , d , f ) and GFP ( green ) . Cell nuclei labeled with DAPI ( white; only shown in b , d , f ) . ( g–l ) Isolectin B4 ( blue ) , Dll4 ( red in g–i ) , anti-GFP ( green ) and endothelial nuclear marker ERG ( red in j–l ) protein staining in representative overview tile-scan ( g–i ) and high magnification ( j–l ) images of whole-mounted WT ( g , j ) , 3Dll4-dVenus ( h , k ) and 3Dll4-Emerald ( i , l ) P5 retinas . DOI: http://dx . doi . org/10 . 7554/eLife . 12167 . 00610 . 7554/eLife . 12167 . 007Figure 2—figure supplement 1 . Generation of dynamic dVenus and stable Emerald Dll4 reporters . DOI: http://dx . doi . org/10 . 7554/eLife . 12167 . 00710 . 7554/eLife . 12167 . 008Figure 2—figure supplement 2 . ( g–i ) Quantification of sprout density ( g ) , sprout diameter ( h ) , and nuclei number ( i ) in WT , 3Dll4-dVenus and 3Dll4-Emerald embryoid bodies ( See Materials and methods for details ) . Values represent means ± S . D . n= number of individual embryoid bodies analysed . p values calculated using two-tailed unpaired t-test . ( p–s ) Quantifications of radial expansion ( p ) , vascular density ( q ) , branching points ( r ) and sprouts number ( s ) in WT , 3Dll4-dVenus and 3Dll4-Emerald P5 retinas ( See Materials and methods for details ) . Values represent means ± S . D . n=number of mice analysed . p values calculated using a two-tailed , unpaired t test . DOI: http://dx . doi . org/10 . 7554/eLife . 12167 . 008 Using the same strategy , we also generated ES cells and mice carrying the more stable and thus less dynamic fluorescent reporter Emerald for Dll4 ( Figure 2—figure supplement 1c ) . Both reporters have a fast maturation time ( Nagai et al . , 2002; Day and Davidson , 2009 ) ; however , dVenus has an extremely short half-life ( 15 min ) , while Emerald has a half-life of 24–26 hr ( Li , 1998; Corish and Tyler-Smith , 1999 ) . Like Dll4-P2A-dVenus , also Dll4-P2A-Emerald was expressed in sprouting endothelium , and generated ES cells and viable reporter mice with no changes in sprouting , branching and retinal patterning ( Figure 2e–i ) . However , unlike dVenus , the distribution of Emerald protein was more widespread than Dll4 protein or dll4 mRNA . Both in the embryoid body ES cell sprouting assay ( EB ) ( Jakobsson et al . , 2007 ) and in the postnatal mouse retina , staining for dVenus showed the typical salt-and-pepper distribution pattern , with high levels in tip cells , and low levels in neighbouring stalk cells ( Hellström et al . , 2007 ) . The long sprouts in EBs showed strong expression at the tip , followed by one or two cells with low expression , and then again single cells with high expression ( Figure 2d ) . This pattern matches the principles of lateral inhibition thought to govern Dll4/Notch signalling in endothelial cells ( Hellström et al . , 2007; Suchting et al . , 2007; Phng and Gerhardt , 2009; Lobov et al . , 2007 ) . Emerald however was present at high levels in both Dll4 protein positive and Dll4 protein negative cells ( Figure 2f ) , suggesting that Dll4 protein half-life is considerably shorter than that of Emerald . Also in the retina , Emerald protein highlighted large stretches of the vascular front with little difference between neighbouring cells , illustrating that most cells at the front experienced a phase of Dll4 expression , and that the reporter accumulates over time ( Figure 2i , l ) . The salt-and-pepper distribution of dVenus and widespread distribution of Emerald provide the first evidence for dynamic phases of Dll4 expression in sprouting angiogenesis in vivo . Remarkably , these results indicate that the salt-and-pepper distribution is not fixed , but dynamic , consistent with the idea of continued and iterative competition for the tip position via Dll4/Notch signalling ( Jakobsson et al . , 2010 ) . Given the known haplo-insufficiency in Dll4 mutant mice ( Gale et al . , 2004 ) , the homozygous viability and lack of vascular defects in mice and ES cells carrying the Dll4-P2A-dVenus or Dll4-P2A-Emerald alleles indicate that the targeted Dll4 allele is fully functional . Furthermore , the distant integration leading to slightly higher Dll4 levels nevertheless fully recapitulates endogenous and functional Dll4 expression patterns . Only the ES cell line , and mouse line , carrying an additional distant integration of Dll4 ( 3Dll4 ) showed sufficient dVenus levels to allow detection without antibody staining . Therefore , we used this line to study the effects of normal and high Vegfa levels on endothelial cell behaviour and sprouting , and for dynamic studies of Dll4 expression ( Figures 3 , 4 ) . 10 . 7554/eLife . 12167 . 009Figure 3 . In vitro high Vegfa or Dll4 levels lead to a switch from branching and elongation to sprout expansion . ( a-d ) Dll4 ( white in a , c; red in b , d ) and reporter protein ( anti-GFP , green ) staining in representative confocal overview ( a , c ) and high magnification ( b , d ) images of vascular sprouts in 3Dll4-dVenus ( a , b ) and 3Dll4-Emerald ( c , d ) embryoid bodies cultured under high Vegf concentration . Cell nuclei are labeled with DAPI ( white; only shown in b , d ) . ( e , f ) Representative confocal overview ( e ) and high magnification ( f ) images of 7Dll4-dVenus embryoid bodies immunolabelled with antibodies specific to Dll4 ( white in e , red in f ) and GFP ( green ) . Cell nuclei labeled with DAPI ( white; only shown in f ) . ( g-i ) Quantification of sprout density ( g ) , sprout diameter ( h ) , and nuclei number ( i ) in WT , 3Dll4-dVenus , 7Dll4-Venus and 3Dll4-Emerald embryoid bodies treated with either normal ( N ) or high Vegfa levels ( HV ) . ( See Materials and methods for details ) . Values represent means ± S . D . n= number of individual embryoid bodies analysed . P values calculated using two-tailed unpaired t-test . DOI: http://dx . doi . org/10 . 7554/eLife . 12167 . 00910 . 7554/eLife . 12167 . 010Figure 3—figure supplement 1 . Simulated Dll4 dynamics with 2 , 3 and 7 copies of Dll4 . DOI: http://dx . doi . org/10 . 7554/eLife . 12167 . 01010 . 7554/eLife . 12167 . 011Figure 3—figure supplement 2 . In vitro synchronization of cell signaling and behavior driven by Dll4 overexpression is Notch dependent . DOI: http://dx . doi . org/10 . 7554/eLife . 12167 . 01110 . 7554/eLife . 12167 . 012Figure 4 . Time-lapse analysis of Dll4 reporter levels identify Dll4/Vegf dose- dependent shift from ‘individual cell’ to ‘synchronous cell’ signalling and behavior . Start ( a , c , e , g , i , k , m , o , q ) and end ( a’ , c’ , e’ , g’ , i' , k’ , m’ , o’ , q’ ) point of confocal time-lapse acquisitions of 3Dll4-dVenus ( a-f ) and 7Dll4-dVenus ( g-l ) embryoid bodies cultured in normal Vegf conditions ( 50 ng/ml ) , and 3Dll4-dVenus embryoid bodies treated with high Vegf concentration ( 500 ng/ml ) ( m-r ) . A heatmap intensity range color was used to represent dVenus expression levels . ( See supplementary information; Videos 7 , 8 , 9 ) . ( a-a’ , g-g’ , m-m’ ) Arbitrary 10 μm spheres were employed to fill the sprout volume , using the Imaris ‘Spot’ cell tracking tool . ( c-c’ , i-i’ , o-o’ ) Sprout volumes covered by two neighboring spheres are monitored overtime . ( e-e’ , k-k’ , q-q’ ) A single sphere placed at the sprout tip is used to monitor the tip advance and retraction . ( b , h , n ) Quantification by time-lapse microscopy of dVenus signal levels relative to each sphere volume covering the sprout . dVenus intensity sum is represented by arbitrary units ( a . u . ) . For technical reasons absolute intensity sum values are not comparable between experiments . ( d , j , p ) Quantification by time-lapse microscopy of dVenus signal intensity sum in two neighboring sphere volumes . ( f , l ) Quantification by time-lapse microscopy of the sprout tip displacement along the three axes x , y , z is shown together with dVenus intensity sum from ( d , j ) . ( r ) To counteract the effect of a sprout drift along the x axis on x-y-z displacement ( See supplementary information; Video 9 ) only the y-displacement is quantified by time-lapse microscopy and shown together with dVenus intensity sum from ( p ) . ( For details on the tracking technique see Materials and methods ) DOI: http://dx . doi . org/10 . 7554/eLife . 12167 . 01210 . 7554/eLife . 12167 . 013Figure 4—figure supplement 1 . Dll4-dVenus reporter expression is dynamically and differentially regulated in vitro in individual endothelial cells . DOI: http://dx . doi . org/10 . 7554/eLife . 12167 . 01310 . 7554/eLife . 12167 . 014Figure 4—figure supplement 2 . Additional examples of dynamic Dll4-reporter intensity profiles and sprout tip displacement illustrating a Dll4- and Vegfa-level dependent switch from individual cell to synchronous cell signaling and behaviour . DOI: http://dx . doi . org/10 . 7554/eLife . 12167 . 014 WT control EB and Dll4-dVenus EBs showed a highly branched network of long and slender sprouts under normal Vegfa conditions ( Figure 2a , c ) . Dll4 protein and dVenus showed the typical salt-and-pepper distribution correlating with normal sprouting patterns ( Figure 2b , d ) . However , high Vegfa conditions led to conspicuously short and stunted sprouts , with much wider diameter ( Figure 3a ) . Interestingly , this morphological change from elongation to expansion invariably correlated with altered Dll4 and dVenus distribution . Although still most strongly expressed at the tip , Dll4 and dVenus were expressed at very high levels in clusters of neighbouring cells close to the tip ( Figure 3b ) . Thus , as predicted by computational modelling ( Bentley et al . , 2008; 2009 ) ( Figure 1c ) , high Vegfa levels appear to disrupt the typical lateral-inhibition salt-and-pepper pattern and instead lead to more synchronized fluctuations of Dll4 levels in contiguous cells , thereby disrupting branching . Also the 3Dll4-Emerald ES cells responded to high Vegfa with stunted sprouting , reduced branching and increased sprout diameter ( Figure 3c ) . Emerald expression was also strongest in clusters of cells at the front ( Figure 3d ) . Computational modelling further predicted that raising Dll4 levels should also lead to synchronized fluctuations ( Figure 3—figure supplement 1; Video 3 , 4 and 5 ) . Indeed , ES cells carrying multiple copies of the Dll4-dVenus transgene ( 7Dll4 ) showed high but dynamic levels of Dll4 expression , and synchronized in clusters of cells that correlated with stunted sprouting and vessel expansion ( Figure 3 e-i ) . DAPT treatment restored branching and elongation under high Vegfa or Dll4 overexpression conditions ( Figure 3—figure supplement 2 ) , demonstrating that this synchronization and morphogenesis effect is Notch mediated . 10 . 7554/eLife . 12167 . 015Video 3 . sprout simulation 2 Dll4 copy ( WT ) . DOI: http://dx . doi . org/10 . 7554/eLife . 12167 . 01510 . 7554/eLife . 12167 . 016Video 4 . sprout simulation 3 Dll4 copy . DOI: http://dx . doi . org/10 . 7554/eLife . 12167 . 01610 . 7554/eLife . 12167 . 017Video 5 . sprout simulation 7 Dll4 copy . DOI: http://dx . doi . org/10 . 7554/eLife . 12167 . 017 To directly observe Dll4 dynamics , we established time-lapse imaging of EBs ( Figure 4 ) . Signal intensity heatmap illustrations highlight the rise and fall in dVenus expression in individual cells ( Figure 4—figure supplement 1; Video 6 ) . Tracing the sprout over time , we quantified the intensity of Dll4-dVenus in arbitrary spheres filling the volume of the sprouts ( Figure 4—figure supplement 2 ) . Intensity plots of the individual spheres showed dynamic behaviour , with temporal profiles similar to those observed in monolayers in vitro with RNA analysis ( Figure 1 and Figure 4 ) . However , the profiles between the spheres filling the sprouts showed no synchrony between neighbouring cell fluctuations , in line with the salt-and-pepper distribution of Dll4 between neighbouring cells ( Figure 4a–d ) . The asynchronous dynamic Dll4 expression behaviour correlated with a continuous forward movement of the leading tip ( Figure 4e–f , Figure 4—figure supplement 2a , b; Video 7 , 10 and 11 ) . 10 . 7554/eLife . 12167 . 018Video 6 . EB 3Dll4-dVenus from Figure 4 – Related to figure supplement 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 12167 . 01810 . 7554/eLife . 12167 . 019Video 7 . EB 3Dll4-dVenus normal Vegf from Figure 4 . DOI: http://dx . doi . org/10 . 7554/eLife . 12167 . 01910 . 7554/eLife . 12167 . 020Video 8 . EB 7Dll4-dVenus normal Vegf from Figure 4 . DOI: http://dx . doi . org/10 . 7554/eLife . 12167 . 02010 . 7554/eLife . 12167 . 021Video 9 . EB 3Dll4-dVenus high Vegf from Figure 4 . DOI: http://dx . doi . org/10 . 7554/eLife . 12167 . 02110 . 7554/eLife . 12167 . 022Video 10 . EB 3Dll4-dVenus normal Vegf ( Video 1 ) from Figure 4—figure supplement 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 12167 . 02210 . 7554/eLife . 12167 . 023Video 11 . EB 3Dll4-dVenus normal Vegf ( Video 2 ) from Figure 4—figure supplement 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 12167 . 023 When performing the same dynamic imaging of EBs comprised of ES cells overexpressing Dll4 ( 7Dll4 clone ) , we observed very similar dynamic fluctuations in individual spheres ( Figure 4g , h ) . However , collectively , the patterns of Dll4 expression between neighboring spheres appeared more synchronized ( Figure 4i , j ) . Intriguingly , these wider sprouts moved forward led by a cluster of several cells synchronously expressing high Dll4-dVenus levels ( Video 8 , 12 and 13 , Figure 4—figure supplement 2c , d ) . A short time later , the same sprout retracted , concomitant with reduced Dll4-dVenus levels in the cluster . Later again , regaining higher expression , the same sprout moved forward with a bright Dll4-dVenus cluster at the tip ( Figure 4k , l ) . Similarly , high Vegfa levels induced clustering and iterative extension and retraction concomitant with synchronized dll4 reporter fluctuations at the tip ( Figure 4m–r , Video 9 , 14 and 15 , Figure 4—figure supplement 2e , f ) . These results provide the first experimental evidence for a synchronized fluctuation in Dll4 expression and cell behaviour by high Vegfa levels or Dll4 overexpression in an in vitro sprouting assay . 10 . 7554/eLife . 12167 . 024Video 12 . EB 7Dll4-dVenus normal Vegf ( Video 1 ) from Figure 4—figure supplement 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 12167 . 02410 . 7554/eLife . 12167 . 025Video 13 . EB 7Dll4-dVenus normal Vegf ( Video 2 ) from Figure 4—figure supplement 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 12167 . 02510 . 7554/eLife . 12167 . 026Video 14 . EB 3Dll4-dVenus high Vegf ( Video 1 ) from Figure 4—figure supplement 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 12167 . 02610 . 7554/eLife . 12167 . 027Video 15 . EB 3Dll4-dVenus high Vegf ( Video 2 ) from Figure 4—figure supplement 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 12167 . 027 To investigate the possibility of synchronized fluctuations in signalling and behaviour in vivo , we injected mVegfa 165 into the vitreous of early post-natal Dll4 reporter and WT pups ( P4 ) ( Figure 5 ) . Fluorescent ISH allowed the detection of nascent dll4 mRNA ( one or two dots ) in the nucleus as well as mature mRNA in the cytoplasm of actively transcribing ECs ( Figure 5—figure supplement 1 ) . dVenus and dll4 mRNA expression showed a salt-and-pepper distribution in control retinas , but strong clustering of dll4 positive cells in Vegfa injected retinas 20 hr post-injection ( Figure 5a–h ) . Intriguingly , despite the strong Vegfa stimulation , clusters of dll4 negative cells were also present , arguing against a simple linear relationship between Vegf and Dll4 expression . Similar results were obtained for Dll4 protein expression ( Figure 5—figure supplement 2a–f ) . The change in spatial Dll4 expression distribution correlated with a dramatic shift from branching and extension to vessel enlargement , remarkably similar to what we observed in the EB system ( Figure 3 ) . At the sprouting front , most of the sprouts had disappeared and were replaced by a dramatically widened peripheral vessel with occasional protrusions . In the plexus , all vessels increased substantially in diameter while the overall branching frequency was significantly reduced ( Figure 5—figure supplement 3 ) . Radial expansion of the vascular plexus was also significantly reduced , consistent with earlier reports ( Gerhardt et al . , 2003 ) . 10 . 7554/eLife . 12167 . 028Figure 5 . Endothelial cell Dll4 expression synchronization under high Vegfa in vivo . ( a–d ) Dll4 ( red in a , b ) , ERG ( red in c , d ) , anti-GFP ( green ) and isolectin B4 ( blue ) staining in representative overview tile-scan ( a , b ) and high magnification ( c , d ) images of whole-mounted 3Dll4-dVenus P5 retinas not injected ( CNT; a , c ) and injected with mVegfa165 ( Vegf inj; b , d ) . For anti-GFP ( dVenus ) and ERG signal a median filter of 3 and 5 pixel , respectively , was used . ( e–h ) Representative confocal images of whole-mounted WT P5 retina not injected ( e ) and mVegfa165 injected ( f ) , labeled for dll4 mRNA detected by fluorescent ISH ( green ) and isolectin B4 ( red ) . White dashed boxed areas in each panel ( e , f ) are magnified in ( g , h ) images . To facilitate endothelial cell nuclei visualization ( DAPI; blue ) together with dll4 mRNA only one stack is shown in panels g and h . White dashed lines delimit endothelial cells ( Iso B4 ) . Asterisks represent EC negative for dll4 mRNA expression . ( i ) Computational simulation of collagen IV deposition ( blue ) after high Vegfa stimulation ( for details on simulation , see Materials and methods ) . At t1 a normal Vegfa condition with a linear gradient extending above the sprout is simulated . High uniform Vegfa levels are simulated from t2 through t6 . Cells with high Dll4 expression and 'tip cell phenotype' are represented in green; cells with low Dll4 expression and 'stalk cells phenotype' are shown in purple . ( j , k ) Representative confocal images of the collagen IV distribution at the sprouting front of WT P5 retinas not injected ( j ) and injected with mVegfa165 ( k ) . Collagen IV is shown in red; Endomucin in green . Arrows indicate empty collagen IV sleeves . ( l–o ) Quantification of the total number ( l , m ) and length of empty collagen IV sleeves ( n , o ) in WT retinas three ( l , n ) and twenty ( m , o ) hours post-injection . Mean ± S . D values are indicated ( n , o ) . n= number of retinas analyzed ( l , m ) ; n= total number of collagen sleeves observed ( n , o ) . P values calculated using a two-tailed , unpaired t test . ( p ) Quantification of the radial expansion in P4 ( 0H post injection ) and P5 ( 20H post injection ) WT retinas not injected and injected with mVegfa165 . n= number of retinas analyzed . Values represent mean ± S . D . p values calculated using a two-tailed , unpaired t test . DOI: http://dx . doi . org/10 . 7554/eLife . 12167 . 02810 . 7554/eLife . 12167 . 029Figure 5—figure supplement 1 . Detection of mature and nascent dll4 mRNA transcripts using whole mount fluorescent ISH . DOI: http://dx . doi . org/10 . 7554/eLife . 12167 . 02910 . 7554/eLife . 12167 . 030Figure 5—figure supplement 2 . Dll4 protein expression is synchronized under pathological Vegfa concentration . DOI: http://dx . doi . org/10 . 7554/eLife . 12167 . 03010 . 7554/eLife . 12167 . 031Figure 5—figure supplement 3 . High Vegf concentrations during retinal angiogenesis result in aberrant vascular patterning . DOI: http://dx . doi . org/10 . 7554/eLife . 12167 . 031 In these experiments Vegf is acutely elevated , by injection , after a normal salt-and-pepper pattern had previously been established through normal development , This scenario is different to the previous simulations where endothelial cells experience high Vegf from the onset , prompting us to re-evaluate the computational predictions in simulations of Vegfa injections at a late timepoint , once the normal salt-and-pepper pattern had established using our existing Vegf/Notch feedback model . We observed that the impact of elevated vegf is strong enough alone to drive a full shift in patterning dynamics matching the observed phenotypic changes in vascular patterning of the retina plexus . Simulations of collagen matrix deposition of advancing sprouts in this model further predicted that the iterative synchronized sprouting and retraction movement generated by the synchronized fluctations in Notch signaling should lead to accumulation of empty collagen sleeves ahead of the vascular front ( Figure 5i; Video 16 ) . Testing this prediction , we assessed the pattern of collagen IV basement membrane deposition at the vascular sprouting front 3 hr ( P4 retinas ) and 20 hr ( P5 retinas ) after Vegfa injection . Retinas injected with Vegfa showed a dramatic mismatch between the vasculature and collagen IV deposition in the blunted regions of the sprouting front ( Figure 5j , k ) . In these areas , already 3 hr after Vegfa injection , collagen IV sleeves protruded ahead of the blunted vasculature , and further increased in number and length over time ( Figure 5l–o ) . The observed radial expansion delay ( Figure 5p ) is thus likely a result of unproductive and iterative sprouting and retraction . 10 . 7554/eLife . 12167 . 032Video 16 . Collagen IV sleeve simulation . DOI: http://dx . doi . org/10 . 7554/eLife . 12167 . 032 A plausible alternative explanation for the observed enlargement in vessel diameter upon high Vegfa levels is cell proliferation . Indeed , injecting Vegfa into the vitreous leads to widespread endothelial proliferation throughout the plexus ( Figure 6a , c ) . To discriminate between proliferation and collective endothelial cell synchronization effects , we inhibited proliferation by systemic treatment with mitomycin C ( MMC ) 24 hr prior to and again concomitant with intraocular mVegfa 165 injection , and analysed proliferation and retinal patterning 24 hr post Vegfa treatment . EdU incorporation revealed a complete block in proliferation in the MMC treated samples ( Figure 6a–d ) . Blocking proliferation alone without Vegfa treatment lead to reduced vascular density in particular in the peripheral zone that developed over the duration of MMC treatment , in line with the proliferation zone in normal development ( Figure 6b ) . Interestingly however , the diameter increase at expense of branching complexity after Vegfa treatment was equally evident in samples from MMC plus Vegfa treated pups ( Figure 6k , l ) . To our knowledge , this is the first direct evidence for the morphogenic effects of quantitative changes in Vegfa to be independent of proliferation . Although proliferation will likely contribute to vessel size increase under high Vegfa , surprisingly , it is not required . Instead we propose that the increased coupling of cells through elevated Dll4/Notch signaling , drives clustering of cells through synchronization of cellular dynamics that interfere with branching , but promote vessel diameter increase . 10 . 7554/eLife . 12167 . 033Figure 6 . High levels of Vegfa induce vessel expansion even in absence of EC proliferation . Representative image of a tile scan showing IsoB4 ( red ) and EdU staining ( green ) in non treated ( control; a–a’ ) , MMC ( bb’ ) , Vegfa 165 ( cc’ ) and MMC_Vegfa 165 ( d–d’ ) treated retinas . ( e ) Quantification of the radial expansion in control ( a ) , MMC ( b ) , Vegfa 165 ( c ) and MMC_Vegfa 165 ( d ) treated retinas . High magnification images showing IsoB4 ( red ) and EdU staining ( green ) in non-treated ( control; f ) , MMC ( g ) , Vegfa 165 ( h ) and MMC_Vegfa 165 ( i ) treated retinas . Quantification of the number of tip cells ( j ) , branching points ( k ) , number of loops ( l ) , vascular density ( m ) , vessel diameter ( n ) and loop area ( o ) in non treated ( control ) , MMC , Vegfa 165 and MMC_Vegfa 165 treated retinas . Scale bar correspond to 400 μm ( a–d’ ) and 100 μm ( fi ) respectively . Statistical comparison and number of animals analyzed are indicated in the graphs . DOI: http://dx . doi . org/10 . 7554/eLife . 12167 . 033 These observations also suggest that the pathological vascular patterning in disease conditions with elevated Vegfa levels might be a consequence of synchronized Dll4 dynamics and the consequent abnormal iterative sprouting and retraction behaviour or cell rearrangement abrogation . To directly test this prediction , we investigated the expression and distribution of dVenus , dll4 mRNA and protein as well as collagen IV empty sleeves formation in two different pathological systems with aberrant vascular patterning; the oxygen induced retinopathy ( OIR ) model , which reproduces aspects of the pathobiology of human retinopathies ( Smith et al . , 1994 ) and the glioblastoma brain tumour model ( GBM ) . In OIR , temporal vessel regression leads to ischemia of the neural retina , inducing a neovascular response that is characterized by dramatic vessel expansion and ineffective ballooning sprouts ( glomeruloid tufts ) that penetrate the inner limiting membrane ( Smith et al . , 1994 ) . In 3Dll4-dVenus reporter mice , OIR leads to strong clustering of endothelial cells positive or negative for dVenus in the neovascular tufts ( Figure 7a–d ) . Similar to Vegfa injection , OIR also leads to an overall increase in Dll4-dVenus levels ( Figure 7a ) . dll4 mRNA and protein expression in WT retinas confirmed clustering of cells with high and low expression , compatible with the idea of localized synchronization of endothelial Dll4 fluctuations in OIR ( Figure 7e–h; Figure 5—figure supplement 2g–i ) . 10 . 7554/eLife . 12167 . 034Figure 7 . Epiretinal tufts in oxygen induced retinopathy show Dll4 expression sychronization . ( ac ) Representative overview tile-scan of a whole-mounted P15 3Dll4-dVenus OIR retina ( a ) . White dashed boxes highlight two different tuft regions; to facilitate tufts visualization only one optical section is shown in the boxed area . Full line boxes are magnified in b and c . For dVenus and ERG signals ( b , c ) a median filter of 3 and 5 pixel , respectively , was used . dVenus ( anti-GFP ) is shown in green , endothelial nuclei are labeled in red ( ERG ) and endothelial cells in blue ( Iso B4 ) . ( d ) Y–Z confocal section of ( c ) . dVenus ( anti-GFP ) expression is shown in green , endothelial cell are labeled in blue ( Iso B4 ) and endothelial nuclei in red ( ERG ) . Asterisks represent EC negative for dVenus expression . ( e–g ) Representative confocal overview image ( e ) and high magnification images ( f , g ) of two different tufts in a WT P15 OIR retina . dll4 mRNA , detected by fluorescent ISH , is shown in green , endothelial cells ( Iso B4 ) are shown in red and endothelial nuclei in blue ( DAPI ) . White dashed boxes in panel ( e ) highlight the regions of the tufts analyzed in f and g . In f and g , white dashed lines delineate endothelial cells on each panel to help visualization . Asterisks represent EC negative for dll4 mRNA expression . ( h ) X-Z confocal section of ( g ) . Stainings for dll4 mRNA ( green ) , IsoB4 ( red ) and DAPI ( blue ) are shown . Asterisks represent EC negative for dll4 mRNA expression . ( i ) Overview image of collagen IV distribution in a mTmG WT P15 OIR retina , labeled with collagen IV ( red ) . New sprouting front and tuft regions are highlighted in the merged image . ( j , k ) Representative images of collagen IV empty sleeves protruding ahead of the new sprouting front ( j ) and of tufts region ( k ) in the mTmG WT p15 OIR retina . Collagen IV is labeled in red . Arrows indicate empty collagen IV sleeves . DOI: http://dx . doi . org/10 . 7554/eLife . 12167 . 034 Collagen IV deposition poorly matched the vasculature , revealing that substantial extensions had previously formed beyond the sprouting front and tufts in the OIR retinas ( Figure 7i–k ) . These results indicate that the glomeruloid tuft formation in OIR is associated with iterative extension and retraction of endothelial processes , which however fail to establish functional new branches . Angiogenesis is one of the hallmarks of cancer ( Hanahan and Weinberg , 2011 ) . However , tumour vessels show highly abberant and dysfunctional patterns , with diameter variability and irregular branching . In many tumours , including glioblastoma , hypoxia driven Vegfa expression drives the angiogenic response . To investigate the implications of Dll4/Notch signalling and behavior synchronization between endothelial cells for vascular patterning in tumours , we analyzed dll4 mRNA and protein expresion together with collagen IV deposition in a mouse syngeneic glioblastoma model ( GBM ) . C57Bl6 derived CT-2A glioblastoma cell spheres implanted into the brain of mice developed highly vascularized solid tumours over the course of 4 weeks ( Martínez-Murillo and Martínez , 2007 ) . Tumours expressed high levels of Vegfa and Hif1a indicating hypoxia , and developed a highly irregular vasculature ( Figure 8—figure supplement 1 ) . dll4 fluorescent ISH performed on 200 µm vibratome sections of GBM revealed a highly irregular pattern of expression ( Figure 8a ) . Whilst some vessels showed little or no signal at all , other vessels showed high levels of expression in all neighboring endothelial cells ( Figure 8 a–c ) . Vessels close to the hypoxic core generally exhibited the highest expression and strong clustering frequently associated with local vessel diameter increase . Similarly Dll4 protein expression showed clustering of high levels in adjacent cells ( Figure 5—figure supplement 2j , k ) . Furthermore , the aberrant vascular patterning in GBM was associated with extensive empty collagen IV sleeves radiating from the tumour vessels ( Figure 8d–f ) . These results provide the first evidence for iterative sprouting and retraction and , together with the Dll4 expression pattern , indicate that endothelial cell Dll4/Notch signalling and sprouting behaviour becomes locally synchronized between neighboring cells during tumour angiogenesis . 10 . 7554/eLife . 12167 . 035Figure 8 . Synchronized Dll4 expression and sprouting behavior in tumour angiogenesis . ( a ) Tile scan representative overview of dll4 mRNA expression detected using fluorescent ISH ( green ) in CT-2A glioblastoma tumor vessels labeled with endomucin ( red ) . Asterisks and arrows indicate tumor vessels negative and positive for dll4 mRNA expression , respectively . White dashed boxes indicate the tumor region analyzed at high magnification on panel b and c . ( b , c ) High magnification of the positive tumor vessel for dll4 mRNA highlighted in panel ( a ) . dll4 mRNA ISH is shown in green , cell nuclei stained with DAPI in blue and endothelial cells , detected using endomucin , in red . To facilitate endothelial cell nuclei visualization together with dll4 mRNA expression only one stack is shown in the panels where nuclear staining ( DAPI; blue ) is included ( c ) . ( d–f ) Confocal overview image ( d ) and high magnification images ( e , f ) showing collagen IV ( red ) deposition around healthy ( d , e ) and tumor vasculature ( d , f ) in the glioblastoma tumor model ( GBM ) developed in a mTmG cre-reporter mouse brain . Endothelial cells express membrane eGFP by PdgfbiCreERT tamoxifen-induced recombination . Dashed line separates the healthy brain ( hb ) from the tumor region on d . Arrows indicate empty collagen IV sleeves . DOI: http://dx . doi . org/10 . 7554/eLife . 12167 . 03510 . 7554/eLife . 12167 . 036Figure 8—figure supplement 1 . Hypoxic tumor cells induce high Vegfa concentrations leading to vessel expansion in mouse glioblastoma model . DOI: http://dx . doi . org/10 . 7554/eLife . 12167 . 036 Together our results demonstrate that the salt-and-pepper differential pattern of Dll4 expression normally associated with vascular branching switches to synchronized fluctuations under conditions of experimental high Vegfa , in retinopathy and tumour angiogenesis . Dynamic observations in the EB model suggest that nonlinear synchronized fluctations in Dll4/Notch drive vessel expansion and disrupt effective cell rearrangement and migration leading to reduced branching and elongation . Our current results provide evidence that Notch signalling activity , between activated endothelial cells , undergoes a phase transition between two distinct operational modes , depending on the levels of Vegfa . At normal physiological levels , that typically correspond with developmental angiogenesis , Dll4 fluctuates dynamically in individual cells; the lateral-inhibition feedback with the Vegfr system functions to generate differences between neighbouring cells that manifest themselves as a salt-and-pepper distribution of Dll4 high and low cells . As a consequence , the salt-and-pepper distribution of tip and stalk cell phenotypes , although individually transient , support the differential behaviour of neighbouring cells at any given time . Dynamic observations in embryoid body sprouting assays indicate that the differential behaviour is critical for branching and continuous elongation . At higher Vegfa levels however , the dynamic fluctuations between neighbouring cells become synchronized , corresponding with a loss of salt-and-pepper patterning . This synchronization , and thus local clustering of cells with high or low Dll4 levels , is dependent on Notch activity , as dampening or inhibiting Notch activity disrupts synchronization . Observations of dynamic behaviour in EBs stimulated with high Vegfa demonstrate that synchronization of Dll4 levels coincides with synchronized sprouting activity . As a consequence , sprouting and branching is disrupted and the vessel instead expands ( Figure 9 ) . The finding of clustered Dll4 expression by in situ hybridisation , Dll4 reporter and protein staining in the vessel expansions in the OIR model and in dilated vessels after Vegfa injection and in mouse glioblastomas together indicate that synchronization is an important principle of vessel malformation in disease . 10 . 7554/eLife . 12167 . 037Figure 9 . Schematic illustrating the important role of dynamic dll4 fluctuations in driving vessel expansion . Individual cells Dll4 levels fluctuate asynchronously during normal vessel growth . Under high VEGF or high Dll4 these fluctuations become more synchronized leading to homogenous cellular dll4 levels that fluctuate between high and low levels together , driving vessel expansion rather than branching . In contrast , when Dll4 levels are homogeneous , but not fluctuating ( termed 'static dll4' scenarios here for simplicity ) the result is that cells either all remain activated and behave as tip cells ( hypersprouting phenotype ) or are all constantly inhibited , prohibiting vessel extension and driving a sparse branching phenotype . Thus the expansion phenotype is specifically driven by the dynamic fluctuations between homogeneous Dll4 high/low levels , rather than by the homogeneity of the Dll4 levels alone . DOI: http://dx . doi . org/10 . 7554/eLife . 12167 . 037 Dll4 overexpression , despite under the normal regulatory control , in the absence of altered Vegfa levels , also leads to the same synchronization and thus clustering and loss of branching . Together with the observed up-regulation of Dll4 under Vegfa control , this suggests that Vegfa-induced Dll4 up-regulation is the key to synchronization . The observed clustering , stunted sprouting and expansion upon Dll4 overexpression in the absence of an altered Vegfa environment further suggests that synchronization itself is the major driver of the switch from branching to expansion . When injecting Vegfa in vivo or raising the concentration in vitro , not only the concentration of Vegfa , but also its spatial distribution changes . We previously proposed that Vegfa injection disrupts the Vegfa gradient , thereby reducing tip cell migration , while driving stalk cell proliferation ( Gerhardt et al . , 2003; Ruhrberg , 2003; Ruhrberg et al . , 2002 ) . Based on static images , we concluded that this shift in the balance between tip migration and stalk proliferation reduces branching while driving expansion . Our new dynamic data , the transient nature of tip and stalk cell phenotypes , and the new insights from blocking proliferation under this condition and from computational modelling give rise to a different interpretation . Computational modelling previously predicted that reduced Vegfa gradients interfere with tip/stalk specification as the reduced differences in the local environment make it more difficult for the lateral-inhibition feedback loop to establish differences between neighboring endothelial cells ( Bentley et al . , 2008; 2009 ) . A loss in gradient already pushes the system towards synchronization , an effect that is dramatically aggravated by higher Vegfa levels . Our present findings indicate that the underlying mechanism driving expansion and clustering , instead of branching and elongation , in situations of reduced Vegfa gradients and high Vegfa levels lies in Dll4/Notch mediated synchronization of adjacent cells as they fail to establish a dynamic salt and pepper pattern . Intriguingly , overexpression of Dll4 in tumour cells has been shown to lead to reduced vessel branching and dramatically increased tumour vessel diameter , with evidence that tumour cell Dll4 activates endothelial Notch receptors ( Li et al . , 2007 ) . Why and how this leads to tumour vessel expansion has since remained unclear . In the light of our current results it is tempting to speculate that overexpression of Dll4 in tumour cells adjacent to the endothelium will cause synchronization of endothelial Notch dynamics , similar to the situation of overexpressed endothelial Dll4 . Conversely , inhibition of Dll4 by antibody , genetic haploinsufficiency or Notch inhibition by DAPT all lead to a dramatic increase in vessel branching , but reduced vessel diameter in tumour vessels ( Ridgway et al . , 2006; Noguera-Troise et al . , 2006 ) . We propose that this switch in tumour vessel patterning is driven by the loss of synchronization , and thus a switch from collectively synchronized cell behaviour to differential individual behaviour . Given our recent identification of Notch-regulated differential VE-cadherin adhesion ( Bentley et al . , 2014 ) , it is conceivable that synchronization leads to vessel expansion through the loss of adhesion differences . According to this concept , differential adhesion drives intercalation and thus vessel elongation , whereas synchronization disrupts intercalation , thus causing cells to cluster and thereby enlarge vessel diameter . As endothelial cell migration as opposed to endothelial cell apoptosis appears to present the key mechanism in vessel remodelling ( Franco et al . , 2015 ) , our current finding of Vegfa-induced vessel enlargement and loss of branching in the absence of proliferation further supports the idea that synchronization disrupts the normal differential migration/intercalation behaviour of endothelial cells and thereby leads to the observed morphological changes . Our observations in EBs overexpressing Dll4 or treated with high Vegfa levels demonstrate that Notch inhibition is sufficient to break the synchronisation and restore branching morphogenesis . Other developmental morphogenesis systems make use of Notch driven synchronization , a phenomenon first described and best understood in somitogenesis , the formation of mesodermal tissue blocks that form the segmental pattern of vertebrates ( Jiang , 2000; Tajbakhsh and Spörle , 1998 ) . The presomitic mesoderm close to the tip of the tail bud shows dynamic activity of Dll/Notch and Notch target genes of hes/hey family . Intriguingly , the zone closest to the tip shows unsynchronized salt-and-pepper patterning of this activity , whilst temporal waves of synchronized activity emerge just rostral of this zone . Although many genes and components of the Notch , Fgf and Wnt-signaling pathway oscillate in the presomitic mesoderm ( Aulehla et al . , 2008; Wahl et al . , 2007; Goldbeter and Pourquié , 2008 ) and single cells oscillate autonomously through feedback loops in the regulation of the hes family of transcriptional repressors ( Kageyama et al . , 2007; Webb et al . , 2016 ) , it has become clear that Dll/Notch activity between the cells is the driver of synchronization . In the absence of Notch activity or Dll expression cells gradually drift out of synchrony , leading to loss of somite pattern and thus vertebrae defects . Recent work by Oates and colleagues using mathematical modelling and experimentation identified that the levels of Dll expression not only critically influence synchronization , but also affect the periodicity of the synchronized oscillations by changing the strength of coupling between the neighboring cells ( Herrgen et al . , 2010 ) . Whether cell-autonomous hes-driven oscillations occur in endothelial cells remains unclear , but speculations on waves of activity of the Bmp and Notch pathway in angiogenesis have recently been raised ( Beets et al . , 2013 ) . Our computational model does not simulate cell-intrinsic oscillations that are coupled via Dll4/Notch . Yet , the ultimate behaviour of the Vegfa-Dll4-Notch-Vegfr feedback loop shows highly similar elements when Dll4 levels change . Thus , although the precise wiring of the signalling circuit is different , there is a striking analogy in that rising Dll4 levels under the influence of Vegfa stimulation will increase the coupling strength between neighbouring endothelial cells . This will ultimately drive the cells to switch from differential to synchronized behaviour and thus to iterative sprouting and retraction , disrupting branching morphogenesis . Although Vegfa has long been known to have different effects at different concentrations , akin to a morphogen , the underlying mechanism has not been understood . Our current results provide a mechanistic explanation for the dosage effects of Vegfa . Further , the identified switch in behaviour from differential to synchronized by changing Dll4/Notch coupling strength between endothelial cells under the influence of rising Vegfa concentrations identifies a novel , and to date unique , mechanism of action for a morphogen . From an applied perspective , the identified mechanism has wide reaching implications for our understanding of the effects of anti-Vegfa therapy in cancer , and will hopefully stimulate new research into dosing regimes of both Vegfa and Dll4/Notch inhibitors , separate and in combination , and effects on vessel normalization . Mice were maintained at London Research Institute under standard husbandry conditions . All protocols were approved by the UK Home Office ( P80/2391 ) . Glioblastoma studies were performed at the Vesalius Research Center , VIB , KU Leuven where housing and all experimental animal procedures were performed in accordance with Belgian law on animal care and were approved by the Institutional Animal Care and Research Advisory Committee of the K . U . Leuven ( P105/2012 ) . WT animals , the mouse lines Pdgfb-iCre ERT ( Claxton et al . , 2008 ) and R26mTmG ( Muzumdar et al . , 2007 ) bred on a C57Bl/6 background were used in this investigation , together with the Dll4 reporters mouse lines 3Dll4-dVenus and 3Dll4-Emerald generated for this study . Details of the different mice treatments are specified below . A 500 bps region encompassing the last intron and exon 11 ( without the STOP codon ) of mouse Dll4 gene ( 5’ homologous region , 5’HR ) was amplified by PCR from a genomic Dll4 BAC clone ( RP23_46P4 , BACPAC CHORI ) and inserted into the pEnt-Emr/Tet vector , which bears a P2A sequence validated in mouse ( Hsiao et al . , 2008 ) . For dVenus reporter , destabilised version of Venus ( dVenus ) coding sequence ( CDS ) was amplified by PCR and inserted in the same vector , in frame with Dll4 exon 11 and the P2A sequence . For Emerald reporter , Emerald CDS was already present in the original pEnt-Emr/Tet vector in frame with the P2A sequence . The first 500 bps of mouse Dll4 3’ UTR ( 3’ homologous region , 3’HR ) were amplified by PCR from the same Dll4 genomic BAC clone and inserted into the PL45 vector ( kindly provided by National Cancer Institute Fredericks ) , which contains a Neomycin/Kanamycin resistance cassette flanked by two loxP sites and driven by a prokaryotic promoter ( em7 ) and a eucaryotic promoter ( Pgk ) . Dll4 5’HR/P2A/dVenus CDS cassette was released by KpnI and SalI digestion and ligated into PL45/ Dll4 3’HR digested with the same enzymes . The targeting vectors so generated ( GenBank accession numbers: BankIt1637803 pTVDll42AdVenus KF293660 , Dll4-dVenus; BankIt1641166 pTVDll42AEmerald KF293661 , Dll4-Emerald ) were linearized using KpnI restriction enzyme and inserted by homologous recombination ( Yu et al . , 2000 ) between exon 11 and 3’UTR of the Dll4 gene contained in the Dll4 genomic BAC clone , employing SW105 bacteria ( kindly provided by National Cancer Institute Fredericks ) . Chloramphenicol and Neomycin/Kanamycin resistance cassettes were utilised to select clones that have undergone recombination . Recombined Dll4-dVenus and Dll4-Emerald BAC clones were linearized by AscI digestion , purified and electroporated into mouse ES cells by standard protocols ( Joyner , 2000 ) with a Biorad Gene pulser electroporator . Positive ES cells were selected for Neomycin resistance . Random insertion and homologous recombination of one or two Dll4 alleles were screened with a ViiA7 Real-Time PCR System and the TaqMan Copy Number Assay ( Life technologies , Carlsbad , California , see manufacturer for details ) . Primers for Dll4 ( Mm00537881_cn ) , dVenus ( custom made , AI6RNTP ) and Emerald ( Mr00660654_cn ) were employed together with the Mouse TaqMan Copy Number Reference Assay , Tfrc ( Life technologies ) , to detect the copy number of Dll4 gene , dVenus and Emerald in the ES genome . ES clones with single homologous recombination were selected , together with clones bearing single ( 3 Dll4-dVenus and 3 Dll4-Emerald ) or multiple ( 7 Dll4-dVenus ) random integrations . To enable dynamic observation of Dll4 reporter in individual endothelial cells the ES clones selected were used to generate embryoid bodies in 3D matrices . ES clones with single homologous recombination and with single ( 3 Dll4-dVenus and 3 Dll4-Emerald ) random integration were also employed to generate mouse reporter lines; briefly , 10–15 ES cells were injected into blastocyst stage embryos collected from C57BL/6J female mice that had been mated to C57BL/6J male mice . Injected embryos were transferred to pseudopregnant recipient ( day 2 . 5 dpc ) mice according to standard protocols ( Nagy , 2000 ) . Mouse lines carrying double Dll4-dVenus homologous recombination were generated by further mice crossing . Offspring was screened and genotyped by performing the same TaqMan Copy Number Assay ( Life technologies ) on Dll4 gene , Venus and Emerald used to screen ES cells . Intraocular injections of 300ng murine Vegfa 165 ( Product 450–32; Peprotech , Rocky Hill , NJ ) , were performed at postnatal day 4 ( P4 ) under isoflurane anaesthesia . Injections were performed using 10 μl gas-tight Hamilton syringes equipped with 34 gauge needles attached to a micromanipulator . Eyes were collected 3H ( P4 ) or 20H ( P5 ) later to proceed with the different experiments . For a detailed description of the experiments in which Vegf injected retinas have been used , together with the conditions required for eye and retina preparation see the specific section of Materials and methods . WT pups were IP injected with 20 μl/g of a 0 . 5 mg/ml Mitomycin C ( Sigma , St . Louis , Missouri ) solution 24 hr ( postnatal day 3; P3 ) and directly before intraocular injection of murine Vegfa 165 at postnatal day 4 ( P4 ) was performed . At P5 , 17H after Vegfa injection , all pups were IP injected with 5-ethynyl-2´-deoxyuridine ( EdU; Invitrogen , ) . Eyes were collected 3H later EdU injection was performed to proceed with retina dissection and EdU_Isolectin B4 staining . ( see EdU staining section below ) Pathological eye angiogenesis was induced as described in Smith et al . ( 1994 ) . However , the 75% oxygen condition was applied only during P7-9 , before returning to normoxia during P10-15 . Conditions used for eye fixation and retina preparation are specified for each technique in which OIR samples have been used . When R26mTmG Pdgfb-iCre ERT mice were used under OIR protocol , tamoxifen ( Sigma ) was injected intraperitoneally ( IP; 20 μl/gr from 4 mg/ml stock solution ) at postnatal day 13 ( P13 ) ; eyes were then collected at P15 to allow visualization of the retinal vasculature . Tumor implantation was performed on WT mice or Pdgfb-iCre-mTmG mice ( 8–12 weeks ) injected intraperitoneally with 100 μg/g of tamoxifen 10 days prior surgery . Craniotomy was realised by drilling a circle in between lamboid , sagittal and coronal sutures of the skull on Ketamine/Xylasine anesthetised mouse . 250–500 µm diameter CT-2A glioblastoma cells spheroid were injected in the cortex and sealed with a glass coverslip cemented on top of the mouse skull . Human end point of the experiment was reached when the tumor exceed 4 mm diameter or if the animal loosed 15–20% of its original weight . Anesthetised mouse was then intracardially perfused with 2% PFA solution . Mouse brain was harvested and fixed overnight in 4% PFA at 4°C . For in situ hybridization , brain was post-fixed in -20°C cold methanol and store at -80°C prior vibratome sectioning . For immunocytochemistry , brain was washes with PBS and sectioned at the vibratome ( 200 µm thickness sections ) . Embryonic stem cells were cultured and embryoid bodies ( EBs ) were generated as previously described ( Jakobsson et al . , 2006 ) . Briefly , embryonic stem cells were routinely cultured on a layer of irradiated mouse embryonic fibroblasts ( DR4 ) in the presence of leukaemia inhibitory factor ( LIF ) . For experiments , cells were cultured for two passages without feeders , then trypsinized , depleted of LIF and left in suspension as hanging drops during four days . On day four the formed embryoid bodies were transferred to a polymerized collagen I gel ( Jakobsson et al . , 2006 ) with addition of 50 ng/ml ( normal ) , 500 ng/ml ( high ) Vegf ( Product 450–32; Peprotech , Rocky Hill , NJ ) , DMSO or DAPT . Medium with normal , high Vegf , DMSO or DAPT ( all at concentrations of 5 μM; Sigma-Aldrich , LY-374973 ) was changed on day 2 after cell plating and every day thereafter . Sprouted EBs were used to perform live imaging analysis of dVenus expression levels or were fixed and stained for Dll4 and dVenus . To analyse dll4 and dVenus mRNA levels in endothelial cells derived from embryonic stem cells by quantitative real-time PCR , EBs were culture in 2D . Briefly , EBs were generated following the same procedure described for 3D cultured EBs; at day 4 EBs were seeded on gelatine-coated plates and treated with 50 ng/ml ( normal ) Vegfa 165 ( Product 450–32; Peprotech , Rocky Hill , NJ ) supplemented media , allowing the formation of a peripheral vascular plexus in 2D . To monitor Dll4 and dVenus gene expression on bEND5 cells and 2D cultured EBs by quantitative real-time PCR analysis samples were collected directly in RLT lysis buffer ( RNeasy MicroKit , Qiagen , Germany ) and further processed for RNA isolation . Reverse transcription of mRNA was performed using Superscript III reverse transcriptase ( Invitrogen ) following the manufacturer recommended protocol . Quantitative real-time PCR was performed using a ViiA7 Real-Time PCR System and Taqman gene expression probes for Dll4 ( Mm00444619; Applied Biosystems , Foster City , California ) and dVenus ( AI6RNTP; Applied Biosystems ) . GAPDH was used as endogenous control to normalize Dll4 and dVenus gene expression . Eyes were collected at P5 ( 3 hr after EdU injection ) and fixed in 2%PFA at 4°C for 4H , subsequently retinas were dissected in PBS . For EdU detection the Click-iT EdU Alexa Fluor 488 Imaging Kit was used ( C10337; Invitrogen ) . Dissected Retinas were transferred to PBS-0 . 5% Triton at RT for 2H and additionally washed with PBS for 4 times 10 min . After removing the PBS , 100 μl of Click-iT reaction cocktail ( following the manufacture indications from Invitrogen ) was added to each retina and incubated ON at 4°C . The day after , retinas were washed 3 times with PBS-0 . 1% Triton during 15 min each and fixed with 4% PFA at RT before to proceed with the Isolectin staining . When co-staining with isolectin ( Iso B4 ) was required to detect endothelial cells on retinas , samples were equilibrated for 1–2H using PBlec buffer ( PBS pH6 . 8 , 1% Triton X-100 , 0 . 1 mM CaCl2 , 0 . 1 mM MgCl2 , 0 . 1 mM MnCl2 ) and incubated with Iso B4 -488 , -568 , -647 or -594 pre-labelled ( 1:200–1:500 ) O/N at 4°C in a rocking platform . In all IFs DAPI ( Sigma ) was used for nuclei labeling . In general , retinas and glioblastoma samples were mounted on slides using Vectashield mounting medium ( Vector Labs , H-1000 ) except when other specifications are given . Primary antibodies used: Dll4 ( R&D Systems , Abingdon , United Kingdom ) , ERG 1/2/3 ( Santa Cruz antibodies , Dallas , Texas ) , GFP ( Abcam , Cambridge , United Kingdom ) , Collagen IV ( AbD Serotec , United Kingdom ) , Endomucin ( Santa Cruz antibodies ) , Hif-1a ( Upstate , Billerica , Massachusetts ) and Iso B4 -488 , -568 , -647 or -594 pre-labelled ( Invitrogen ) . Secondary antibodies: the adequate Alexa -488 , -555 , -568 or -647 conjugated ( Invitrogen ) . Glioblastoma implanted mouse brains were harvested , tumors were dissected out together with the corresponding contralateral hemisphere region and were separately frozen at -80°C . Samples were lysed in RIPA ( 20 mM Tris pH 7 . 5 , 60 mM NaCl , 1% Triton X-100 , 0 . 5% deoxycholic acid , 0 . 1% sodium dodecyl sulfate , 10% glycerol , 25 mM ß-glycerophosphate , 50 mM sodium fluoride , 2 mM sodium pyrophosphate , 1 mM sodium orthovanadate , and protease inhibitor cocktail , Calbiochem , Billerica , Massachusetts ) , homogenized and sonicated . Proteins were quantitated with BCA kit ( Pierce , Rockford , IL ) . Proteins ( 100 μg ) were analyzed by western blotting using anti-Vegfa ( 0 . 5 µg/ml , R&D Systems ) and anti-ß-actin ( 0 . 2 µg/ml , Sigma ) antibodies . Membranes were incubated with peroxidase-conjugated secondary antibodies ( 1:5000; Pierce ) for 2H at RT , and proteins were visualized with ECL detection reagents ( Pierce ) using ImageQuant LAS-4000 ( GE Healthcare , United Kingdom ) imaging system . Western blot quantifications of 4CT-2A glioblastoma implanted mice samples were performed using BioRad QuantityOne software . Data are expressed in fold change with healthy brain region as a reference . in vitro was kindly provided by Dr . Mailhos ( Mailhos et al . , 2001 ) . Eyes were collected at postnatal day 5 ( P5 ) or 15 ( P15 ) , fixed with 4% PFA for 4 hr and immediately dissected in PBS . Retinas were fixed O/N with fresh 4% PFA at 4°C and transferred to methanol ( METOH ) the day after . Retinas were store at -80°C until dll4 fluorescent ISH was performed . Glioblastoma samples were fixed overnight in 4% PFA at 4C , post-fixed in -20°C cold METOH , vibratome sectioned ( 200 μm ) and used to perform dll4 fluorescent ISH as described below . To develop dll4 fluorescent ISH , samples were rehydrated through washing steps with 75% , 50% and 25% METOH-PBS-0 . 1% Tween-20 at RT , treated with Proteinase K ( Invitrogen ) , fixed with 4%PFA-0 . 1%Glutaraldehyde ( Sigma ) and pre-hybridized at 65°C for 2 hr with Hb4 pre-hybridization buffer [25% Formamide deionized ( Sigma ) , 25% 20X SCC pH 7 , 5 μg/ml yeast RNA ( Sigma ) , 50 μg/ml Heparin ( Sigma ) and 0 . 1%Tween-20] . Probe-hybridization was performed O/N at 65°C using Hb4D5 buffer containing 50 μg/ml Dextran Sulphate powder ( Sigma ) diluted in Hb4 buffer . dll4 anti-sense DIG-probe was used at a concentration of 1μ g/ml to prepare the hybridization mix . The following day , samples were thoroughly washed using 50% Formamide deionized ( Sigma ) in 2XSSCT , 2XSSCT and 0 . 2XSSCT at 65°C , cooled down in PBS-0 . 1%Tween and blocked with PBS-0 . 1%Tween-8% Sheep serum ( Sigma ) for 2 hr at RT . After blocking , samples were incubated O/N at 4°C with anti-DIG POD antibody ( 1:500; Roche , Mannheim , Germany ) . Then samples were washed 4 times 30 min with TBS-0 . 1%Tween and incubated for 20-30min with fluorescein tyramide solution ( TSA; 1:500 ) , diluted on PBS-0 . 1%Tween , at RT . Subsequently , 0 . 001% H2O2 was added to the TSA solution to activate the TSA reaction . After 30–45 min , the TSA reaction was stopped , and the retinas quickly washed with PBS several times . For TSA solution synthesis and products references see: http://wiki . xenbase . org/xenwiki/index . php/Flourescin_Tyramide_Synthesis . Before proceeding with IsoB4 staining , retinas were washed during 2 days at 4°C in slow agitation with PBS-0 . 1%Tween , to reduce ISH unspecific signal , and post-fixed with 4%PFA . For tumor brain samples vasculature was stained using endomucin . The specific protocol for IsoB4 and endomucin staining is listed in the immunofluorescence section . In both samples , additionally DAPI staining was used to visualize endothelial cell nuclei . Retinas and tumor brain samples were mounted on slides using Vectashield mounting medium ( Vector Labs , Burlingame , California H-1000 ) and imaged by confocal microscopy . Confocal laser scanning microscopy was performed using Carl Zeiss ( Germany ) LSM710 , Carl Zeiss LSM780 and Leica ( Germany ) TCS SP8 confocal microscope . Images were processed using Imaris ( Bitplane ) , ImageJ , and Adobe Photoshop software . For time-lapse microscopy embryoid bodies were cultured in collagen I in glass bottom 24-well plates ( MatTek ) using a phenol-red free medium to minimise the autofluorescence background ( GIBCO ) . On day six the plate was transferred to a LSM780 laser-scanning microscope ( Zeiss; equipped with a motorized stage , incubator S-M and POC-R cultivation system ) maintained at 37°C and 5% CO2 with a humidifier . Z-slices were acquired ( 40–60 per field every 30 min using 2% laser capacity ) with a 20x , numerical aperture ( NA ) 1 . 0 , water-immersion and coverslip corrected dipping objective ( Zeiss ) . In order to isolate dVenus and Emerald signals from the autofluorescence background before each experiment dVenus and Emerald spectra were identified and recorded in Lambda mode and the time-lapse acquisitions were performed in Online Fingerprinting mode using the previously recorded spectra . Simulations were developed using the established memAgent-Spring Model ( MSM ) of Dll4-Notch signalling during sprouting angiogenesis . Cells in this model are comprised of computational agents , which represent sections of the membrane ( 'memAgents' ) connected by springs , which confer tension to the cortex beneath the membrane . The cells are initialised in three dimensional space filled with Vegf ( using both a gridded lattice and continuous space mappings ) , memAgents then interact with their local environment and activate their Vegfr-2 and Notch receptors if any Vegf or Dll4 respectively are present leading to dynamic Notch regulated growth of filopodia and cell migration . We reduced the resolution of the established MSM to make the model more coarse and speed up calculation time , allowing the simulation of a larger number of cells as needed in a monolayer simulation ( a lattice site now represented 1 micron cubic volume instead of 0 . 5 microns ) . Each cell was square with sides of 10 microns ( comprising 10x10 memAgents in a 2D square agent spring mesh ) . Each cell was linked by junction springs to its nine square neighbours creating a fully connected checkerboard sheet of simulated endothelial cells . The coarser lattice resulted in the following recalibrated parameters: Filopodia grow one grid site per timestep ( previously this resulted in one timestep representing 15 s to grow one grid site of 0 . 5 microns , to match measured filopodia dynamics in in vivo zebrafish data in Bentley et al . [2008] ) . Here filopodia extend 1 micron per timestep so we must define one timestep in a coarse grid as representing thirty seconds of sprouting . In the original MSM model , the Vegfr-2-Dll4-Notch signalling time delays were set so that the lateral inhibition pathway’s periodicity matched the thirty minute periodicity of the zebrafish segmentation clock ( Giudicelli and Lewis , 2004 ) . Here we are matching mouse cell data and so a longer period of 4 hrs was used . The phase behaviour of the system has been established as independent of the precise periodicity , only the time until selection is affected ( Bentley et al . , 2008 ) . Receptor and ligand maximal levels and the maximum actin level available for filopodia extension , Vmax , Dmax , N , Ftot as defined in ( Bentley et al . , 2008 ) were reduced to a quarter of their original level due to the corresponding reduction in cell surface area . The precise Vegf microenvironment that the cells experience in vitro is not known , but assumed to have no gradient with limited room for cells to extend processes and filopodia in the intercellular space or over neighbouring cells , however short filopodia and cell shape changes do occur and we assume some heterogeneity in Notch signalling is generated . A completely uniform level of Vegf cannot produce heterogeneity in cells Notch signalling without greater room for the stochastic filopodia growth to establish differences and a prolonged period of time ( Bentley et al . , 2008 ) . Thus with constrained room for filopodia to grow ( 10 microns of space only ) abstracted to above the sheet for simplicity ( rather than simulating processes between cells ) a shallow gradient was required to obtain at least transient heterogeneity as the cells compete over time . Vegf was calculated as follows for each grid lattice site g , where Vegf was set to 0 . 9 for normal Vegf ( representing 50 ng ) and 18 for high Vegf ( 1 mg ) , zg is the z axis coordinate position of the grid site and G is the define gradient increase , set to 0 . 1 . VEGFg=VEGF+zgG The values for G and Vegf under normal conditions were calibrated to generate maximise the time spent in a heterogeneous pattern . Though uniform Vegf was also simulated . Using the coarse model as defined above for the monolayer we now initialised ten cells back around a single tube with radius 3 microns , using a gridded lattice of dimensions Xg , Yg , Zg = 90 , 86 , 8 . The vessel is placed as sprouting from midway up the y axis sprouting in the direction as the x axis the increases , periodic boundary conditions on the sprout are switched off so edge effects at either end of the sprout may take effect . To close off one end of the cylinder all memAgents lying on the edge of the exposed cylinder at the tip end were placed at the same x , y position but with their z coordinate = 1 , thus zipped shut adhered to the collagen layer beneath the vessel . All memAgents other than those in filopodia do not move for these simulations to maintain the original vessel shape , without initiation of new branching or complex Vegf environment simulation , lumen formation etc to study the Dll4 dynamics in a single sprout over time in a parsimonious manner . Again little is known about the specific local gradients and concentrations of Vegf that vessel in a dish experience . A simple static Vegf environment was used . It was assumed that a shallow gradient would exist with Vegf increasing towards the tip of the sprout . Sprouts normally migrate along collagen fibres in the embryoid body assay , and astrocytes in the retina , so we assume that there is a fibre/astrocyte beneath the sprout with most of the Vegf adhered to it , creating the strongest gradient below the sprout . We also assume there is a low level Vegf gradient felt all around the sprout given the 3D nature of the environment . The Vegf level in each grid site g ( with coordinate position xg , yg , zg on the integer grid ) is defined as:VEGFg={Vconst+xgVcol , ifzg=1Vconst+xgVcolVfree , otherwise Here Vconst , Vcol , Vfree were set to 2 . 1 , 0 . 008 and 0 . 01 respectively and calibrated to these values to exactly match receptor activation dynamics over time as generated by the validated , less coarse model in Bentley et al . ( 2009 ) . Normally in the model Dll4 is set to rise with activity of the Vegfr-2 receptor by twofold ( the delta parameter from Bentley et al . [2008] ) . Given there are two copies of Dll4 normally this meant that to simulate three and seven copies we simply changed this parameter Delta to 3 and 7 respectively . The original tip cell selection model and parameter set was used as in Bentley et al . ( 2009 ) in order to see the effects of more detailed filopodia and cell shape changes than in the coarse grained model defined above . Cells are initialised around a pre-existing vascular tube , with periodic boundary conditions , two cells per cross section . New tip cells can be selected from it by stimulation with Vegf , which is adhered to a square lattice of collagen extending above the vessel in the y axis ( see Bentley et al . [2009] for more details ) . The following extensions were made to this model here: 1 ) Collagen is deposited by the cell by creating a blue square in every lattice site that the cell body ( excluding filopodia ) touches . Collagen here does not currently change the Vegf signalling but merely shows where the cell body has been located . 2 ) The cell body was allowed to randomly move forward along filopodia tracks with a small probability ( 0 . 001 per filopodia/per timestep ) whereas previously only contact with another tip cell’s filopodia would trigger this migratory behaviour ( see Bentley et al . [2009] for more details ) . This was added to improve realism to the model such that tip cells can migrate slowly before they meet another tip cell when this behaviour would then be enhanced . After the salt and pepper pattern has been established and cells are close to anastamosing ( representing a normal developing retinal vasculature ) an injection of high Vegf ( 100x the normal level ) was simulated which was also assumed to eradicate any gradient of Vegf above the vessel .	Throughout life , blood vessels are constantly remodelled to ensure that oxygen and nutrients reach every part of the body where they are needed . If a tissue is not receiving an adequate blood flow , existing blood vessels may widen or new blood vessels may sprout from their walls . In certain diseases , such as cancer , blood vessels may grow excessively to form disorganized networks , and preventing this growth may help to treat these conditions . However , we do not fully understand how the body controls the size , shape and branching pattern of blood vessels . For a new blood vessel to sprout out of an existing vessel , the tip of the new branch must first develop . The tip forms when the endothelial cells that line the blood vessel are activated by a protein called vascular endothelial growth factor A ( Vegfa ) , which is produced by the surrounding tissue . The activated endothelial cells respond to Vegfa stimulation by producing the protein Dll4 , which talks to neighboring endothelial cells to prevent them from also forming new tips . In a way , this process bears all the signs of a competition between cells , as they fight for which one is allowed to take the lead . The losers of this competition , when forced into subordination by the tips , also serve an important function , as they will help to form and elongate the base of the new sprout . Although it is known that changes in the levels of Vegfa in tissues can cause blood vessel branching to alter dramatically , the mechanisms that enable this to occur are not well understood . Computer simulations of the process predicted that an unexpected synchronization of Dll4 dynamics would be triggered when Vegfa levels increased; however , this remained to be observed in real cells . Ubezio , Blanco et al . have now used fluorescent markers to observe the Dll4 production in lab-grown mouse endothelial cells as they formed new vessel sprouts in response to Vegfa . This revealed that the levels of Dll4 fluctuate widely in individual cells . Time-lapse movies of the cells showed that as a new sprout forms , the levels of Dll4 in neighbouring cells fluctuate in an uncoordinated manner . However , increasing the amount of Vegfa in the cells indeed synchronizes these fluctuations . This causes the new sprout to retract and allows the original blood vessel to widen . Increasing the levels of Dll4 had the same effect . Further experiments confirmed that increasing the amount of Vegfa also reduces blood vessel branching in tumours in mice by synchronizing the fluctuations in the levels of Dll4 in neighbouring endothelial cells . In the future , these results could help refine anti-cancer treatments that work by blocking the activity of Vegfa and Dll4 .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "developmental", "biology", "computational", "and", "systems", "biology" ]	2016	Synchronization of endothelial Dll4-Notch dynamics switch blood vessels from branching to expansion
The conserved Musashi ( Msi ) family of RNA binding proteins are expressed in stem/progenitor and cancer cells , but generally absent from differentiated cells , consistent with a role in cell state regulation . We found that Msi genes are rarely mutated but frequently overexpressed in human cancers and are associated with an epithelial-luminal cell state . Using ribosome profiling and RNA-seq analysis , we found that Msi proteins regulate translation of genes implicated in epithelial cell biology and epithelial-to-mesenchymal transition ( EMT ) , and promote an epithelial splicing pattern . Overexpression of Msi proteins inhibited the translation of Jagged1 , a factor required for EMT , and repressed EMT in cell culture and in mammary gland in vivo . Knockdown of Msis in epithelial cancer cells promoted loss of epithelial identity . Our results show that mammalian Msi proteins contribute to an epithelial gene expression program in neural and mammary cell types . During both normal development and cancer progression , cells undergo state transitions marked by distinct gene expression profiles and changes in morphology , motility , and other properties . The Epithelial-to-Mesenchymal Transition ( EMT ) is one such transition , which is essential in development and is thought to be co-opted by tumor cells undergoing metastasis ( Polyak and Weinberg , 2009 ) . Much work on cell state transitions in both the stem cell and cancer biology fields has focused on the roles that transcription factors play in driving these transitions ( Polyak and Weinberg , 2009; Lee and Young , 2013 ) , such as the induction of EMT by ectopic expression of the transcription factors Snail , Slug , or Twist ( Mani et al . , 2008 ) . Recent work has shown that RNA-binding proteins ( RBPs ) also play important roles in cell state transitions , by driving post-transcriptional gene expression programs specific to a particular cell state . The epithelial specific regulatory protein ( ESRP ) family of RBPs are RNA splicing factors with epithelial tissue-specific expression whose ectopic expression can partially reverse EMT ( Warzecha et al . , 2009; Shapiro et al . , 2011 ) . RBPs have also been implicated in other cell state transitions , such as reprogramming of somatic cells to induced pluripotent stem cells ( iPSCs ) , which have the essential characteristics of embryonic stem cells ( ESCs ) . For example , overexpression of the translational regulator and microRNA processing factor Lin28 along with three transcription factors is sufficient to reprogram somatic cells ( Yu et al . , 2007 ) . The Muscleblind-like ( Mbnl ) family of RBPs promote differentiation by repressing an ESC-specific alternative splicing program , and inhibition of Mbnls promotes cellular reprogramming ( Han et al . , 2013 ) . For ESRP , Lin28 , and Mbnl proteins , the developmental or cell-type-specific expression pattern of the protein provided clues to their functions in the maintenance of epithelial , stem cell , or differentiated cell state . The Musashi ( Msi ) family comprises some of the most highly conserved and tissue-specific RBPs , with Drosophila Msi expressed exclusively in the nervous system ( Nakamura et al . , 1994; Busch and Hertel , 2011 ) . In mammals , the two family members Msi1 and Msi2 are highly expressed in stem cell compartments but are mostly absent from differentiated tissues . Msi1 is a marker of neural stem cells ( NSCs ) ( Sakakibara et al . , 1996 ) and is also expressed in stem cells in the gut ( Kayahara et al . , 2003 ) and epithelial cells in the mammary gland ( Colitti and Farinacci , 2009 ) , while Msi2 is expressed in hematopoietic stem cells ( HSCs ) ( Kharas et al . , 2010 ) . This expression pattern led to the proposal that Msi proteins generally mark the epithelial stem cell state across distinct tissues ( Okano et al . , 2005 ) , with HSCs being an exception . Msi1 is not expressed in the normal adult brain outside a minority of adult NSCs but is induced in glioblastoma ( Muto et al . , 2012 ) . Msi proteins affect cell proliferation in several cancer types . In glioma and medulloblastoma cell lines , knockdown of Msi1 reduced the colony-forming capacity of these cells and reduced their tumorigenic growth in a xenograft assay in mice ( Muto et al . , 2012 ) . Msi expression correlates with HER2 expression in breast cancer cell lines , and knockdown of Msi proteins resulted in decreased proliferation ( Wang et al . , 2010 ) . These observations , together with the cell-type specific expression of Msi proteins in normal development , suggested that Msi proteins might function as regulators of cell state , with potential relevance to cancer . Msi proteins have been proposed to act as translational repressors of mRNAs—and sometimes as activators ( MacNicol et al . , 2011 ) —when bound to mRNA 3′ UTRs , and were speculated to affect pre-mRNA processing in Drosophila ( Nakamura et al . , 1994; Okano et al . , 2002 ) . However , no conclusive genome-wide evidence for either role has been reported for the mammalian Msi family . Here , we aimed to investigate the roles of these proteins in human cancers and to gain a better understanding of their genome-wide effects on the transcriptome using mouse models . To obtain a broad view of the role Msis might play in human cancer , we surveyed the expression and mutation profiles of Msi genes in primary tumors using genomic and RNA sequencing ( RNA-Seq ) data from The Cancer Genome Atlas ( TCGA ) ( Cancer Genome Atlas Network . , 2012 ) . To determine whether Msi genes are generally upregulated in human cancers , we analyzed RNA-Seq data from five cancer types for which matched tumor-control pairs were available . In these matched designs , a pair of RNA samples was obtained in parallel from a single patient's tumor and healthy tissue-matched biopsy , thus minimizing the contribution of individual genetic variation to expression differences . We observed that Msi1 was upregulated in at least 40% of breast , lung , and prostate tumors , while Msi2 was upregulated in at least 50% of breast and prostate tumors ( Figure 1A , top ) . Overall , Msi1 or Msi2 were significantly upregulated in matched tumor-control pairs for 3 of the 5 cancer types , compared to control pairs . Kidney tumors showed the opposite expression pattern , with Msi1 and Msi2 downregulated in a majority of tumors and rarely upregulated , and in thyroid cancer neither Msi1 nor Msi2 showed a strong bias towards up- or down-regulation ( Figure 1A , top ) . In breast tumors , a bimodal distribution of Msi1 expression was observed , with a roughly even split between up- and down-regulation of Msi1 , consistent with the idea that Msi1 upregulation might be specific to a subtype of breast tumors . The bimodality of Msi1 expression was not seen when comparing control pairs , so is not explained by general variability in Msi1 levels ( Figure 1A , bottom , solid vs dotted lines ) . 10 . 7554/eLife . 03915 . 003Figure 1 . Msi genes are frequently overexpressed in breast , lung , and prostate cancer but downregulated in kidney cancer . ( A ) Top: percentage of matched tumor–control pairs with upregulated ( black-fill bars ) or downregulated ( grey-fill bars ) Msi1 or Msi2 in five cancer types with matched RNA-Seq data . Upregulated/downregulated defined as at least two-fold change in expression in tumor relative to matched control . Asterisks indicate one-tailed statistical significance levels relative to control pairs . Bottom: distribution of fold changes for Msi1 and Msi2 in matched tumor–control pairs ( solid red and green lines , respectively ) and in an equal number of control pairs ( dotted red and green lines , respectively . ) Shaded gray density shows the fold change across all genes . ( B ) Percentage of tumors with non-silent mutations in Msi1/Msi2 and a select set of oncogenes and tumor suppressors across nine cancer types . Bold entries indicate genes whose mutation rate is at least two-fold above the cancer type average mutation rate . DOI: http://dx . doi . org/10 . 7554/eLife . 03915 . 00310 . 7554/eLife . 03915 . 004Figure 1—figure supplement 1 . Analysis of Msi1/Msi2 mutation and expression profiles in TCGA datasets . ( A ) Distributions of the percent of tumors with non-silent mutations across cancer types in TCGA DNA sequencing data . Red and green triangles indicate values for Msi1 and Msi2 , respectively . ( B ) Unsupervised hierarchical clustering of breast cancer tumors and matched controls , with overlaid sample labels , clinical markers and PAM50 subtypes . DOI: http://dx . doi . org/10 . 7554/eLife . 03915 . 004 Examining genome sequencing data from matched tumor-control pairs across nine diverse cancer types , we found that Msi1 and Msi2 were not significantly mutated in most of these cancers ( Figure 1B ) . One notable exception was kidney cancer ( KIRC ) , where non-silent mutations in Msi1 were significantly overrepresented , detectable in 9% of tumors ( ranked in the 99th percentile of mutations per gene in this cancer ) ( Figure 1—figure supplement 1A ) . This observation , together with the lower Msi mRNA levels observed in matched kidney tumors ( Figure 1A ) , is consistent with a model in which loss of Msi function is selected for in kidney tumor cells , either as a result of downregulation or mutation . The observation that Msi1/Msi2 was not significantly mutated in most tumors but are overexpressed in several tumor types ( including glioblastoma ) makes their profile more similar to oncogenes like FOS or HER2 , than to tumor suppressors like PTEN and TP53 , which tend to have the opposite pattern ( Verhaak et al . , 2010; Cancer Genome Atlas Network . , 2012 ) ( Figure 1B ) . To determine whether Msi overexpression is specific to a particular cancer cell state , we focused on breast cancer , where tumors with distinct properties can be robustly classified by gene expression ( Parker et al . , 2009; Cancer Genome Atlas Network . , 2012 ) . Unsupervised hierarchical clustering of matched tumor and control samples produced a nearly perfect separation of tumors from control samples , rather than clustering by patient/genome of origin ( Figure 1—figure supplement 1B ) . We overlaid on top of our clustering a classification of samples into Normal , HER2+ , Luminal A , Luminal B , and Basal states using RNA-Seq data to measure expression of the PAM50 gene set ( Parker et al . , 2009 ) . Our clustering using all genes corresponded well to the PAM50 classification ( Cancer Genome Atlas Network . , 2012 ) , separating most Luminal A from Luminal B tumors and showing a general grouping of HER2+ tumors ( Figure 1—figure supplement 1B ) . Using this classification , we found that Msi2 was highly expressed in Luminal tumors ( Figure 2A ) . Msi1 was more variable across tumor subtypes , often showing a bimodal profile , split between up- and down-regulation ( Figure 1A and Figure 2—figure supplement 1B ) . Msi2 expression was highest in Luminal B tumors , which are known to be more aggressive and highly proliferating ( Ki67-high ) than Luminal A types and are thought to share properties with epithelial mammary progenitor cells ( Das et al . , 2013 ) . These observations prompted the hypothesis that Msi proteins might be localized to epithelial cells in breast cancer tumors . The splicing factors Rbfox2 and Mbnl1 were previously identified as regulators of EMT and are upregulated during this transition ( Venables et al . , 2013 ) . Using TCGA expression analysis , we confirmed that Rbfox2 and Mbnl1 are more highly expressed in luminal tumors compared with mesenchymal tumors , as predicted by their role in EMT ( Figure 2—figure supplement 2 ) . 10 . 7554/eLife . 03915 . 005Figure 2 . Msi is associated with the epithelial-luminal state in breast cancer . ( A ) mRNA expression of Msi2 across different breast tumor types in TCGA RNA-Seq . ( B ) Immunofluorescence staining for Ecadherin ( ECAD , red ) and Msi1 ( MSI1 , green ) . Top: luminal human breast tumor with high number of ECAD-positive cells . MSI1 shows primarily cytoplasmic localization ( white arrowheads ) . Inset shows magnified version of ECAD and MSI staining . Bottom: triple negative , basal-like tumor . ECAD-positive cells showed strong cytoplasmic MSI1 stain ( blue arrowheads ) while ECAD-negative cells were MSI1-negative ( red ) . Single confocal stacks shown , 10 μm scale . ( C ) mRNA expression of Msi1 , Msi2 , Ecad , Fn1 , Vim , and Jag1 in breast cancer cell lines by RNA-Seq ( datasets are listed in Supplementary file 1 ) . ( D ) Western blot for MSI1/2 ( MSI1/2 cross react . antibody ) , MSI2 , phosphorylated HER2 ( p-HER2 ) and HER2 in panel of breast cell lines . ‘HMLE + pB’ indicates HMLE cells infected with pB empty vector , ‘HMLE + Twist’ indicates HMLE cells infected with Twist transcription factor to induce EMT . MDAMB231-derived metastatic lines ( 231-Brain , 231-Bone ) and Sum159 are basal , HER2-negative cancer cell lines . BT474 and SKBR3 are HER2-positive , epithelial-luminal cancer cell lines . Epithelial-luminal ( HER2-positive ) lines show increased expression of Msi proteins compared with basal lines , and Twist-induced EMT reduces Msi expression . ( E ) mRNA expression of Msi1 , Msi2 , Ecad , Fn1 , Vim , and Twist1 in GBM tumors classified as mesenchymal ( n = 20 ) or epithelial ( n = 20 ) using an EMT gene signature . DOI: http://dx . doi . org/10 . 7554/eLife . 03915 . 00510 . 7554/eLife . 03915 . 006Figure 2—figure supplement 1 . Expression of Msi1/Msi2 in subtypes of breast cancer cell lines and breast cancer tumors . ( A ) Unsupervised hierarchical clustering of gene expression from RNA-seq of breast cancer cell lines . ( B ) Fold-change in tumor–control pairs of TCGA breast cancer tumors for Msi1 and Msi2 across tumor subtypes . Msi1 shows a variable bimodal distribution of fold changes , while Msi2 is enriched in Luminal B tumors relative to Basal tumors . ( C ) Ratio of luminal to basal cancer cell line fold changes for Msi1 , Msi2 , Jag1 , and Fn1 . DOI: http://dx . doi . org/10 . 7554/eLife . 03915 . 00610 . 7554/eLife . 03915 . 007Figure 2—figure supplement 2 . Expression of Rbfox2 ( Rbm9 ) and Mbnl1 in subtypes of breast cancer tumors from TCGA . Expression values for Rbfox2/Mbnl1 plotted across PAM50 subtypes , after TMM normalization . DOI: http://dx . doi . org/10 . 7554/eLife . 03915 . 007 To examine the expression and distribution of Msi proteins in tumors , we stained a panel of human breast cancer tumors for MSI1 and the epithelial marker E-cadherin ( ECAD ) . MSI1 expression was predominantly cytoplasmic ( Figure 2B , top panel ) . Across luminal tumors , MSI1 was co-expressed with ECAD ( as in Figure 2B , top panel ) . In triple negative/basal-like tumors , a minority of ECAD-positive cells showed strong MSI1 staining , whereas ECAD-negative cells showed little to no expression ( Figure 2B , blue and red arrowheads , respectively ) , supporting an association between Msi and epithelial cell state in tumors . Given the heterogeneity of human tumor samples , it is possible that the increased expression of Msi genes in luminal tumors ( compared with basal ) reflects the generally higher fraction of epithelial cells in these tumors . To explore whether Msi expression is associated with a luminal as opposed to basal state in a more homogenous system , we analyzed RNA-Seq data for luminal and basal breast cancer cell lines generated by multiple independent labs ( RNA-Seq data sets used are listed in Supplementary file 1 ) . Gene expression profiles from the same cell lines generated independently tended to cluster together in unsupervised clustering ( supporting consistency of data across labs ) , and overall the basal cell lines were distinguishable from the luminal lines ( Figure 2—figure supplement 1A ) . Matching the pattern observed in primary tumors , we observed higher Msi1 and Msi2 expression in luminal breast cancer lines than in basal lines ( Figure 2C , left panel ) . Expression of Fibronectin ( Fn1 ) , Vimentin ( Vim ) , and Jagged1 ( Jag1 ) , which are associated with the basal/mesenchymal state ( Yamamoto et al . , 2013 ) , had the opposite pattern , showing strong enrichment in basal over luminal lines ( Figure 2C , right panel ) . The enrichments of these four genes for either the luminal or basal state were unusual when compared to the background distribution of these enrichments across all expressed genes ( Figure 2—figure supplement 1C ) , indicating that these genes are strong indicators of the two states . To further investigate the connection between Msi expression and EMT in breast cancer , we examined Msi expression in a panel of breast cancer-derived cell lines . Consistent with the RNA-Seq data from primary tumors , HER2+ epithelial cell lines expressed higher levels of Msi1 and Msi2 compared with HER2– lines ( Figure 2D , lane 6 and 7 ) . A standard cell culture model of EMT is the immortalized inducible-Twist human mammary epithelial ( HMLE-Twist ) cell line , which undergoes EMT when induced to express the transcription factor Twist ( Mani et al . , 2008 ) . We found that Msi1 was strongly downregulated in HMLE cells following Twist-induced EMT ( Figure 2D ) , consistent with the epithelial-associated expression pattern of Msis in primary tumors ( Figure 2A–C ) . Similarly , Msi protein expression was higher in luminal , HER2+ breast cancer lines ( BT474 , SKBR3 in Figure 2D ) compared with basal HER2– breast cancer lines ( brain and bone metastatic derivatives of MDAMB231 , 231-Brain and 231-Bone , and SUM159 in Figure 2D ) . We next asked whether the epithelial expression signature of Msis is present in other primary tumors . Given the established role of Msi proteins as regulators of Glioblastoma ( GBM ) cell growth and as markers of primary tumors ( Muto et al . , 2012 ) , we examined whether there is a similar subtype expression pattern in GBM tumors from TCGA ( Verhaak et al . , 2010 ) . We used an EMT gene signature to rank GBM tumors from more epithelial to more mesenchymal , based on the similarity of each tumor's gene expression profile to that of cells undergoing EMT in culture ( Feng et al . , 2014 ) . Using this ranking , we found that the top 20 most epithelial tumors expressed higher levels of Msi and epithelial markers like ECAD ( Figure 2E ) . By contrast , the top 20 most mesenchymal tumors expressed lower levels of Msi and higher levels of mesenchymal markers like Fibronectin and Vimentin ( Figure 2E ) . Thus , Msi expression is enriched in epithelial tumors in GBM as well , consistent with the results obtained in breast cancer tumors and cell lines . Taken together , these results show that Msi genes are rarely mutated but frequently overexpressed across human cancers and are strong markers of the epithelial-luminal state . This pattern suggests that Msi proteins may play a role in the maintenance of an epithelial state and/or repression of EMT , in both breast and neural cell types . To better understand the molecular functions of Msi proteins , we turned to a controlled cell culture system . The upregulation of Msi genes in glioblastoma motivated the choice of NSCs as a system to study the molecular roles of Msi proteins , a cell type where both proteins are highly expressed in normal development , and where their target mRNAs are likely to be present . NSCs provide a well-characterized system for homogeneous cell culture ( Kim et al . , 2003 ) , which is not always available for progenitor/stem cell types cultured from other primary tissues like the mammary gland , making NSCs grown in culture amenable to analysis by genome-wide techniques . Furthermore , the conserved expression of Msi genes in the nervous system and their reactivation in human glioblastoma suggests that molecular insights obtained in this system could be informative about the roles of Msi proteins in glioblastoma cells . We cultured cortical NSCs from E12 . 5 embryos obtained from transgenic mice with a Dox-inducible Msi1 or Msi2 allele , and from double conditional knockout mice for Msi1/Msi2 , whose deletion was driven by a Tamoxifen-inducible Cre ( Figure 3A ) . These systems enabled robust overexpression or depletion of Msi proteins ( Figure 3B ) within 48–72 hr of induction . To study the effects of Msi depletion and induction on mRNA processing , expression , and translation , we used ribosome footprint profiling ( Ribo-Seq ) ( Ingolia et al . , 2009 ) and high-throughput sequencing of polyA-selected RNA ( RNA-Seq ) ( Mortazavi et al . , 2008 ) ( Figure 3A ) . 10 . 7554/eLife . 03915 . 008Figure 3 . Genetic system for studying effects of Msi loss/gain of function on gene expression . ( A ) Experimental setup and use of Msi1/2 inducible overexpression and conditional double knockout mice for derivation of neural stem cells , which were then used for ribosome profiling ( Ribo-Seq ) and mRNA sequencing ( RNA-Seq ) . ( B ) Western blot analysis of Musashi overexpression and knockout in neural stem cells . Overexpression and conditional knockout cells were exposed to Dox and 4-OHT for 72 hr , respectively . ( C ) mRNA-Seq expression values ( RPKM ) scatters between Msi1 overexpressing cells and controls ( left ) , Msi2 overexpressing cells and controls right ( 72 hr Dox ) . Msi1/2 each robustly overexpressed with similar magnitude following Dox . ( D ) Comparison of translational efficiency ( TE ) values using Ribo-Seq on Msi1 overexpressing cells on Dox ( 72 hr ) vs controls ( left ) and conditional knockout cells following 4-OHT for 48 hr ( right ) . Colored points indicate select genes with large changes in TE . DOI: http://dx . doi . org/10 . 7554/eLife . 03915 . 00810 . 7554/eLife . 03915 . 009Figure 3—figure supplement 1 . Quality control metrics for Ribo-Seq libraries . ( A ) Quality control metrics for overexpression Ribo-Seq libraries . Left panel: percentage of reads mapped to genome , and the percentages of reads that are unique ( ‘percent_unique’ ) and mapping to rRNA ( ‘percent_ribo’ ) out of those mapped . Right panel: percentage of reads mapping to exons ( ‘percent_exons’ ) , and out of those the percentage of reads in CDS regions ( ‘percent_cds’ ) , 3′ UTRs ( ‘percent_3p_utr’ ) , 5′ UTRs ( ‘percent_5p_utr’ ) . Percentage of reads mapping to introns ( ‘percent_introns’ ) also shown . ( B ) Quality control metrics for knockout Ribo-Seq libraries , same format as ( A ) . DOI: http://dx . doi . org/10 . 7554/eLife . 03915 . 009 When Msi1 or Msi2 were overexpressed , few significant changes in mRNA expression were observed after 48 hr ( Figure 3C ) . This observation suggests that these factors do not directly impact transcription or mRNA stability/decay but leaves open possible effects on other steps in gene expression such as mRNA translation . To determine the genome-wide effects of Msi proteins on translation , we performed Ribo-Seq on Msi1-overexpressing cells and double knockout cells . Reads from these Ribo-Seq libraries showed the expected enrichment in coding exons relative to UTRs and introns , and yielded high scores in various quality control ( QC ) metrics ( Figure 3—figure supplement 1 ) . These QC metrics were highly consistent across libraries , supporting comparative analysis of the resulting data ( Figure 3—figure supplement 1 ) . To examine changes in translation , we computed ‘Translational Efficiency’ ( TE ) values for all protein-coding genes , a measure of ribosome occupancy along messages that is defined as the ratio of the ribosome footprint read density in the ORF to the RNA-seq read density . Examination of TEs across overexpression and knockout samples yielded a handful of genes with very large changes in ribosome occupancy ( Figure 3D , ‘Materials and methods’ ) . Several genes exhibited substantial changes in their translation efficiency in response to overexpression of Msi1 , including six genes with increased TE and three with reduced TE ( Figure 3D ) . Genes with increased translation included the RNA processing factor Prpf3/Prp3p , a U4/U6 snRNP-associated factor , and genes involved in epithelial cell biology such as Kirrel3/NEPH2 . Genes with repressed translation included: Rbm22/Cwc2 , another splicing factor associated with U6 snRNP; Dhx37 , an RNA helicase with possible role in alternative splicing ( Hirata et al . , 2013 ) ; and Jag1 , a ligand of Notch receptors and an important regulator of Notch signaling . No change was detected in translation of previously reported Msi target Numb ( Okano et al . , 2002 ) , though Numb had low coverage of Ribo-Seq reads in NSCs , reducing our statistical power to detect regulation ( ‘Materials and methods’ ) . To explore whether the observed changes are mediated by direct protein binding to RNA targets , we mapped the RNA binding specificity of Msis . To determine sequence-specific RNA binding preferences of Msi proteins , we used ‘RNA Bind-n-Seq’ ( RBNS ) to obtain quantitative and unbiased measurement of the spectrum of RNA motifs bound by recombinant MSI1 protein in vitro ( Lambert et al . , 2014 ) ( Figure 4A ) . For each 6mer , the ‘R value’ was defined as the occurrence frequency in libraries derived from MSI1-bound RNAs divided by the corresponding frequency in the input RNA library , and 6mer ‘enrichment’ was defined as the maximum R value observed across all protein concentrations . The fold enrichment profiles obtained by RBNS for the top five most enriched 6mers and five randomly chosen 6mers are shown in Figure 4B . Enriched 6mers exhibited similar enrichment profiles across concentrations , peaking in fold enrichment at concentrations typically between 16–64 nM ( Figure 4B ) . To summarize the binding preferences of MSI1 from RBNS , we aligned the most enriched 6mers to generate a motif , which emphasizes that MSI1 binds predominantly to UAG-containing sequences , preferentially flanked by Us ( Figure 4C ) . The MSI1 binding site ( G/A ) UAGU from a previous SELEX study was ∼threefold enriched by RBNS , along with highly similar sequences , confirming binding under our assay conditions ( Imai et al . , 2001; Ray et al . , 2013 ) . Closer examination of the RBNS data revealed evidence for longer , higher-affinity motifs containing multiple UAGs with short intervening spacers ( not shown ) . 10 . 7554/eLife . 03915 . 010Figure 4 . Profiling MSI1 binding preferences by RNA Bind-n-Seq . ( A ) Schemaic of Bind-n-Seq experiment for MSI1 protein . Increased concentrations of MSI1-SBP fusion protein incubated with random RNA pool , pulled by straptavidin pull-down , reverse-transcribed and sequenced . ( B ) Fold enrichment of top five enriched 6mers ( red curves ) and five randomly chosen 6mers ( blue curves ) across protein concentrations . ( C ) Binding motif for MSI1 . Position-weight matrix generated by global alignment of top 20 enriched 6mers . ( D ) Two sites in Jag1 3' UTR , region 1 and region 2 , containing a high density of enriched 6mers . Top: PhyloP conservation score for 3' UTR in 20 nt windows ( based on UCSC vertebrates multiple alignment ) . Bottom: number of enriched 6mers from BNS in 20 nt windows of 3' UTR . ( E ) Percent binding of MSI1 protein to region 1 and region 2 ( red curves ) and mutants where UAG sites are disrupted ( blue curves ) , measured by gel-shift ( see Figure 4—figure supplement 1 ) . Kd estimates for region 1 and region 2 are shown ( mean of 2 gel-shifts per sequence ) . ( F ) Western blot analysis of Jag1 regulation by Msi: top left panel , Jag1 expression in Msi1 overexpression cells and controls in cellular fractions ( T—total lysate , C—cytoplasmic and N—nuclear fractions ) . Jag1 is translationally repressed upon induction of Msi1 and detected only in total and cytoplasmic lysates . hnRNP A1 , known to shuttle between the nucleus and the cytoplasm and alpha-Tubulin used as loading controls . ( G ) Increased JAG1 protein levels in double knockout cells . ( H ) Reduced JAG1 protein levels upon Msi2 overexpression . DOI: http://dx . doi . org/10 . 7554/eLife . 03915 . 01010 . 7554/eLife . 03915 . 011Figure 4—figure supplement 1 . Validation by gel-shift of MSI1 binding to Jag1 3' UTR sequences . ( A ) Top: gel-shift MSI1 binding assay for Jag1 3' UTR sequence 1 . Kd estimate shown ( 15 nM ) is average of two gel shifts . Bottom: gel-shift for Jag1 3′ UTR sequence 1 mutant , where UAG sites mutated to UCC . Kd cannot be estimated ( no binding to mutant could be detected . ) ( B ) Top: gel-shift MSI1 binding assay for Jag1 3′ UTR sequence 2 . Kd estimate shown ( 9 nM ) is average of two gel shifts . Bottom: gel-shift for Jag1 3′ UTR sequence 2 mutant , where UAG sites are also mutated to UCC . Kd for mutant sequence was 649 nM . DOI: http://dx . doi . org/10 . 7554/eLife . 03915 . 01110 . 7554/eLife . 03915 . 012Figure 4—figure supplement 2 . Effect of Msi1 gain and loss of function on Jag1 mRNA levels and protein expression . Fold-change in Jag1 expression in Msi1 overexpression and double knockout samples for Ribo-Seq and RNA-Seq experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 03915 . 01210 . 7554/eLife . 03915 . 013Figure 4—figure supplement 3 . Validation of Msi-dependent regulation of Jag1 protein levels using luciferase reporters containing Jag1 3' UTR . Luciferase expression for Jag1 3′ UTR reporter transfected into 293T cells . Mean values shown for three biological replicates ( ±standard deviation ) . For knockdown lines , Jag1 3′ UTR reporter expression was normalized relative to reporter expression in mock transfected 293T cells ( represented by dashed horizontal line . ) Note that Msi2 sh . 4 was effective in knocking down Msi2 , but consistently increased Msi1 mRNA levels , and therefore did not reduce total Msi mRNA levels . This likely explains why Msi2 sh . 4 293T cells did not show increased Jag1 3′ UTR reporter expression . DOI: http://dx . doi . org/10 . 7554/eLife . 03915 . 013 Previous studies suggested that MSI1 binds 3′ UTR regions of mRNAs to regulate translation ( Okano et al . , 2005 ) . We calculated the density of RBNS-enriched 6mers in 3′ UTR regions genome-wide and ranked genes by the density of enriched 6mers in their 3′ UTR ( ‘Materials and methods’ ) . We observed that the 3′ UTR of Jag1—which is translationally repressed by Msi ( Figure 3D ) —contains a moderately high density of RBNS-enriched 6mers , ranking in the 85th percentile of all 3′ UTRs ( Figure 4D ) . To ask whether Msi proteins can directly bind the Jag1 mRNA and test the RBNS motif , we selected two regions of the Jag1 3′ UTR that contained the highest density of RBNS-enriched 6mers for in vitro analysis ( Figure 4B , top ) . A gel-shift assay detected strong binding of RNAs representing both regions by recombinant Msi protein , with estimated Kd values of 15 nM and 9 nM for regions 1 and 2 , respectively ( representative gel shifts are shown in Figure 4—figure supplement 1 ) . Since both sequences contain UAGs ( Figure 4—figure supplement 1 ) , we hypothesized that the UAGs nucleate binding . Mutation of the UAG sites to UCC reduced binding to MSI1 protein by an order of magnitude or more in each case ( Figure 4E ) , supporting a model where MSI1 binding occurs primarily at these sites . Following Msi overexpression , the Ribo-Seq density of the Jag1 coding region was reduced by ∼fivefold , while its mRNA level was little changed , suggesting a predominant effect at the translational level ( Figure 4—figure supplement 2 ) . In double knockout cells , Jag1 mRNA increased ∼1 . 5-fold by RNA-Seq ( Figure 4—figure supplement 2 ) , with a similar increase in Ribo-Seq density , suggesting effects on message stability either in the absence of or as a consequence of translational derepression . Western blot analysis confirmed repression of JAG1 protein by Msi1 overexpression ( Figure 4F ) and derepression in double knockout cells ( Figure 4G ) . The high similarity between MSI1 and MSI2 proteins ( over 70% identity at the amino acid level , with highly similar RNA recognition motifs ) suggests similarity in function , and we confirmed that Msi2 overexpression also repressed JAG1 protein expression by Western analysis ( Figure 4H ) . To directly test the hypothesis that Msi proteins regulate Jag1 translation via UTR binding , we constructed luciferase reporters for the Jag1 3' UTR and transfected these into 293T cells . Knockdown of MSI1 or knockdown of both MSI1 and MSI2 increased luciferase expression in these cells , relative to mock knockdown treatments ( Figure 4—figure supplement 3 ) . This observation also indicates that Msi-dependent regulation of Jag1 translation is conserved from murine to human cells . In sum , our results support a model where Msi proteins directly bind to the Jag1 3′ UTR to mediate post-transcriptional repression of protein levels . Since some of the largest changes in translation observed by Ribo-Seq affected RBPs with functions in RNA splicing , we hypothesized that Msi overexpression might trigger changes in pre-mRNA splicing . Changes in mRNA splicing following Msi overexpression or depletion were assessed by analysis of RNA-seq data using the MISO software ( Katz et al . , 2010 ) . For example , exon 38 in the Myo18a gene , which is predominantly included under control conditions , was modestly repressed following Msi2 overexpression and strongly repressed following Msi1 overexpression ( Figure 5A ) . In total , we observed several hundred alternatively spliced exons that were either repressed or enhanced by overexpression or knockout of Msis ( Figure 5B ) . Msi proteins are predominantly localized in the cytoplasm ( Figure 5—figure supplement 1 ) , even when overexpressed ( Figure 3F ) , suggesting that these changes in pre-mRNA splicing are indirect . For example , these splicing changes may result from changes in the levels of splicing factors whose mRNAs are translationally regulated by Msi proteins . 10 . 7554/eLife . 03915 . 014Figure 5 . Global impact of Msi proteins on alternative splicing . ( A ) Sashimi plot for Myo18a alternative exon 38 with Percent Spliced In ( Ψ ) estimates by MISO ( values with 95% confidence intervals , right panel . ) Exon splicing is repressed by Msi1 overexpression and slightly increased in knockout Msi1/2 cells . ‘+’ indicates samples treated with Dox/Tam for overexpression/knockout cells , respectively . E12 . 5 neural stem cells were used for all samples except Msi1 overexpression for which an additional E13 . 5 NSC time point was sequenced . ( B ) Number of differential events ( MISO Bayes factor ≥10 , ΔΨ ≥ 0 . 12 ) in each alternative RNA processing category ( SE—skipped exons , A5SS—alternative 5′ splice site , A3SS—alternative 3′ splice site , MXE—mutually exclusive exons , RI—retained introns ) for Msi1 overexpression ( ‘Msi1 OE’ ) , Msi2 overexpression ( ‘Msi2 OE’ ) , double knockouts ( ‘Double KO’ ) , and a Dox control pair ( ‘Control’ ) . ( C ) Comparison of ΔΨ in Msi1 overexpression vs control binned by direction ( ‘Spliced in’ or ‘Spliced out’ , x-axis ) to ΔΨ in Msi2 overexpression cells and in double knockout cells ( along with respective Tam and Dox controls , y-axis ) . ( D ) Computational strategy for identifying human orthologs of alternative exon trios regulated in mouse neural stem cells . Orthologous exon trios were identified by synteny using multiple genome alignments . ( E ) Comparison of ΔΨ mouse alternative exons by Msi1 ( comparing overexpression to control , x-axis ) and ΔΨ of their orthologous exon trios in human ( comparing luminal and basal cell lines , y-axis ) . Two pairs of luminal and basal cells compared: BT474 vs MDAMB231 and SKBR3 vs MDAMB231 . ΔΨ value distributions summarized by violin plots with a dot indicating the mean ΔΨ value . DOI: http://dx . doi . org/10 . 7554/eLife . 03915 . 01410 . 7554/eLife . 03915 . 015Figure 5—figure supplement 1 . Subcellular localization of MSI1 protein in murine NSCs . ( A ) Immunofluorescence staining in mouse neural stem cells for MSI1 ( red ) and hnRNP A2/B1 ( green ) . MSI1 shows predominantly cytoplasmic localization , while hnRNP A2/B1 , a splicing factor , is predominantly nuclear . Confocal maximum Z intensity projections shown , 10 μm scale . ( B ) Western blot analysis for MSI1/2 and alpha-Tubulin ( TUB ) in total protein lysate ( T ) , cytoplasmic protein lysate ( C ) and nuclear protein lysate ( N ) in control and Msi2 overexpressing cells . DOI: http://dx . doi . org/10 . 7554/eLife . 03915 . 01510 . 7554/eLife . 03915 . 016Figure 5—figure supplement 2 . Analysis of two conserved Msi-induced splicing changes in breast cancer tumors . ( A ) Distribution of MISO ΔΨ values in matched tumor–control pairs for Erbin ( Erbb2ip ) exon in light blue and Myo18a in dark blue . Right and left shifts from center ( marked by dotted grey line at ΔΨ = 0 ) indicate tumor-enhanced and tumor-repressed splicing patterns , respectively . ( B ) Comparison of RNA fold changes in matched tumor–control pairs for Msi1 and Msi2 in Basal ( left ) and Luminal ( right ) tumors with ΔΨ values for Erbin and Myo18a exons . Points/triangles indicate luminal/basal tumor types determined by PAM50 . DOI: http://dx . doi . org/10 . 7554/eLife . 03915 . 016 To test whether Msi1 and Msi2 affect pre-mRNA splicing in similar ways , we compared the direction of splicing changes following Msi1 or Msi2 overexpression . Exons with increased inclusion following Msi1 overexpression tended to show increased inclusion following Msi2 overexpression as well , while Msi1 OE-induced splicing changes were uncorrelated with Dox-induced changes ( Figure 5C ) . A similar pattern was observed for exons with decreased inclusion ( Figure 5C ) . These observations suggested that Msi1 and Msi2 trigger similar effects on mRNA splicing . Splicing changes observed in the Msi1/Msi2 double knockout cells exposed to 4-OHT were inversely correlated to those observed following Msi overexpression ( Figure 5C ) . This observation further supports that Msi proteins affect splicing at physiological expression levels . No correlation in splicing was observed between Msi1-induced cells and exposure to 4-OHT of double floxed cells lacking the Cre driver ( Figure 5C ) . We next considered whether the splicing changes associated with Msi mis-expression in NSCs might be related to splicing changes observed in human breast cancer cells or with a particular cell state . The natural variation in Msi levels across breast cancer cell lines ( Figure 2C–E ) enabled a comparison of splicing patterns between Msi-high ( luminal ) vs Msi-low ( basal ) cells . To compare mouse and human splicing patterns , we identified human alternative exon trios orthologous to mouse alternative and flanking exon trios using synteny in a multi-genome alignment ( Figure 5D and Supp . ‘Materials and methods’ ) . We first compared changes ( ΔΨ ) in the percent spliced in ( PSI or Ψ ) values of mouse exons between Msi1 overexpressing cells vs controls , to ΔΨ values of orthologous exons between luminal and basal breast cancer cell lines ( Figure 5E ) . The splicing patterns were consistent: the human orthologs of exons up-regulated in Msi1-OE NSCs had higher inclusion in luminal ( Msi-high ) than in basal ( Msi-low ) cell lines , and similarly for down-regulated exons ( Figure 5E ) . Such agreement was observed for several different luminal and basal pairs , but was strongest when comparing HER2+ luminal lines such as BT474 and SKBR3 to basal lines , consistent with the higher Msi levels observed in HER2+ cell lines ( Figure 2D ) . These observations support the proposition that Msi contributes to a luminal splicing program in human breast cancers by triggering changes similar to those induced in mouse NSCs . Two of the most strongly affected alternative exons in murine NSCs , Myo18a exon 38 ( Figure 5A ) and Erbin exon 21 ( Erbb2ip , a direct binding-partner of the breast cancer oncogene HER2/Erbb2 ) were conserved in the human genome and detected in the transcriptomes of all analyzed breast tumors and controls . In primary tumors , these exons showed a striking cancer-associated splicing pattern , with the ERBIN exon enhanced in tumors and the MYO18A exon repressed in tumors ( Figure 5—figure supplement 2A ) . To test whether the regulation of these exons is responsive to Msi levels , we correlated the fold change in Msi expression for each matched tumor–control pair with the ΔΨ value of the ERBIN and MYO18A exons in that pair ( Figure 5—figure supplement 2B ) . We observed high correlation between the extent of Msi overexpression and the change in splicing in luminal tumors , particularly for MSI2 . As in mouse NSCs , increased expression of Msis was associated with increased inclusion of the ERBIN exon and repression of MYO18A exon splicing , suggesting that Msi-dependent regulation of splicing may be conserved not only in breast cancer cell lines but also in primary tumors . To address whether Msi proteins are functionally required for the maintenance of the luminal state , we performed RNAi knockdown of Msi1 and Msi2 in two luminal breast cancer cell lines , BT474 and MCF7-Ras , where Msi proteins are highly expressed ( Figure 2C and Figure 6—figure supplement 1A ) . In the HER2+ luminal cell line BT474 , cells grow in tightly packed epithelial colonies ( Figure 6A ) . We observed a striking morphological change upon knockdown of MSI1 or MSI2 , where cells progressively separated and acquired a basal-like appearance 3–5 days after knockdown ( Figure 6A ) , accompanied by reduced proliferation ( not shown ) . A similar phenotype was observed in MCF7-Ras cells upon knockdown of MSI1 or MSI2 ( Figure 6—figure supplement 1B ) . These results argue that Msi expression is required for the maintenance of the epithelial-luminal state in breast cancer cell lines . 10 . 7554/eLife . 03915 . 017Figure 6 . Msi levels alter EMT processes breast cancer cell lines . ( A ) Knockdown of Msi1/Msi2 in BT474 breast cancer cell line using lentiviruses carrying short hairpins ( shRNAs ) . Brightfield images ( 10x magnification ) shown at 24 , 72 , and 120 hr after Puromycin-selection . ( B ) mRNA expression of epithelial and mesenchymal markers upon knockdown of Msi1/Msi2 in epithelial-luminal breast cancer cell line ( BT474 ) and overexpression of Msi1 in mesenchymal-basal line ( MDAMB231 ) . Values plotted are fold changes normalized to GAPDH . For BT474 knockdown , cells infected with hairpin against luciferase were used as control ( ‘Control sh’ ) . For MDAMB231 overexpression , cells infected with tdTomato were used as controls ( ‘Msi1-tdT’ ) . Msi1 levels were below detection limit in control MDAMB231 cells , therefore Msi1 fold change in MDAMB231 Msi1-overexpression cells ( relative to controls ) was truncated arbitrarily in plot , indicated by ‘^’ . ( C ) Representative transwell assay image for LM2 control and Msi1-OE breast cancer cells . ( D ) Quantification of percent of well covered in transwell assay for LM2 control and Msi1-OE cells ( 4 wells per condition , individual well values plotted as dots . ) . DOI: http://dx . doi . org/10 . 7554/eLife . 03915 . 01710 . 7554/eLife . 03915 . 018Figure 6—figure supplement 1 . Knockdown of Msi1/2 in breast cancer cell lines . ( A ) Western blot for BT474 cells with control ( shLuc ) or Msi1/2 targeting hairpins . ( B ) Morphology of MCF7-Ras cells upon Musashi knockdown . DOI: http://dx . doi . org/10 . 7554/eLife . 03915 . 018 The Notch pathway regulator Jag1 , which we found was translationally repressed by Msi , is known to be required for EMT . Jag1-depleted keratinocytes undergoing TGFβ-induced EMT fail to express mesenchymal markers and retain epithelial morphology ( Zavadil et al . , 2004 ) . Furthermore , knockdown of Jag1 in keratinocytes strongly impairs wound healing ( Chigurupati et al . , 2007 ) , a process that requires cells to acquire mesenchymal properties such as migration and protrusion . Our gene expression analysis also supported the mesenchymal-basal specific expression of Jag1 , which is particularly pronounced in breast cancer ( Figure 2 ) . The epithelial-associated expression pattern of Msi genes and the antagonistic relation between Msi and Jag1 ( Figure 2 ) prompted the hypothesis that Msi activation promotes an epithelial cell identity , effectively blocking EMT . To test the hypothesis that Msi activation may hinder EMT processes by promoting the epithelial state , we assessed the effect of Msi knockdown and overexpression on EMT marker expression . Knockdown of MSI1 or MSI2 in the luminal cell line BT474 generally resulted in a decrease in epithelial marker expression and an increase in mesenchymal marker expression , consistent with Msi loss promoting EMT ( Figure 6B ) . To test whether ectopic expression of Msi in mesenchymal cancer cells can promote an epithelial state , we overexpressed Msi1 in the mesenchymal cell line MDAMB231 , where Msi1 levels are extremely low . Msi1-overexpressing cells had decreased mesenchymal marker expression and increased levels of epithelial marker expression ( Figure 6B ) , consistent with promotion of the epithelial state . We conclude that Msi activation promotes the epithelial state in breast cancer cells . We next asked whether the increase in epithelial markers following Msi overexpression is accompanied by functional changes that reflect the epithelial state . We predicted that ectopic expression of Msi proteins in a mesenchymal cell line would hinder EMT-associated processes such as migration . Msi1 overexpression in the LM2 cell line ( an MDAMB231-derivative ) resulted in sevenfold reduction in migration in a transwell assay ( Figure 6C , D ) . We were unable to observe this phenotype in the mesenchymal cell lines MDAMB231 or SUM159 , where Msi1 overexpression caused no significant change in migration in the same transwell assays ( data not shown ) . In NSCs , overexpression of Msi1 or Msi2 impaired migration as assayed by a scratch assay as well ( data not shown ) , consistent with the phenotype observed in LM2 breast cancer cells . These results show that depending on the cell-type context , Msi activation can decrease the migration capacity of cells , consistent with promotion of an epithelial state and suppression of mesenchymal properties . The association of Msis with the luminal state in breast cancer tumors and their effect on the epithelial-luminal state in breast cancer cell lines prompted us to ask whether Msi proteins play similar roles in the mammary gland in vivo . During maturation , epithelial cells in the mammary gland migrate and form ducts within the mammary fat pad through a process termed mammary ductal branching morphogenesis . The formation of the mammary ductal system is thought to be a kind of EMT ( Chakrabarti et al . , 2012; Foubert et al . , 2010 ) , making mammary gland an attractive system to study the regulation of EMT in vivo . The mammary gland Terminal End Buds ( TEBs ) from which ducts form are organized into discrete layers of cell types , including epithelial luminal and basal cells . The identity of luminal and basal tumors is thought to resemble their mammary gland cell type counterparts . Analysis of RNA-Seq expression analysis of purified mouse mammary luminal ( CD24highCD29+ ) and basal ( CD24+CD29high ) cells generated by dos Santos et al . ( 2013 ) revealed enrichment of Msi1 and Msi2 expression in luminal cells ( not shown ) . As predicted by the mRNA expression profile , we observed higher MSI2 protein levels in the luminal cell layer and far lower levels in the basal ( K14-positive ) cell layer of mouse mammary ducts ( Figure 7A ) . 10 . 7554/eLife . 03915 . 019Figure 7 . Msi2 activation represses EMT and expands mammary luminal cell layer in vivo . ( A ) Immunostaining for MSI2 , K14 , and DAPI in control sections of mammary gland . Scale bar: 50 μm ( B ) qRT-PCR for Msi2 in mammary epithelial cells from control and Msi2 overexpressing mice ( ‘Msi2-OE’ ) . ( C ) Whole mount stain for mammary glands from control and Msi2 overexpressing mice ( left: low magnification , right: high magnification . ) ( D ) Immunostaining for K14 , K8 , and DAPI in mammary gland sections from control and Msi2 overexpressing mice . Scale bar: 100 μm ( E ) qRT-PCR for luminal markers ( K8 , K18 ) , basal markers ( K14 ) , and smooth-muscle Actin ( SMA ) in mammary epithelial cells from control and Msi2 overexpressing mice . ( F ) Staining for E-cadherin ( ECAD ) ( top ) and EMT-marker SLUG ( bottom ) in mammary glands from control and Msi2 overexpressing mice . Luminal cell layer is expanded upon Dox ( arrows ) . Scale bar: 100 μm . ( G ) qRT-PCR for Slug , Gata3 , Twist1 , Twist2 in mammary epithelial cells from control and Msi2 overexpressing mice . Slug expression in basal cell layer is reduced upon Dox ( arrows ) . Scale bar: 50 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 03915 . 01910 . 7554/eLife . 03915 . 020Figure 7—figure supplement 1 . Msi2 overexpression in mouse mammary gland alters mammary duct morphology . ( A ) Msi2 expression in mammary glands co-stained with basal cell marker K14 in control and Msi2 overexpressing mice . ( B ) Quantification of number of branch points in control and Msi2 overexpression mice . Student's t-test was used to compute p-values . ( C ) Lengths of longest mammary ductal branches ( measured from Center of Lymph Node , CLN ) for control and Msi2 overexpression mice . CLN defined as ‘0’: negative length values indicate that longest ductal branch ends prior to start of CLN , positive length values indicate that longest ductal branch grew past center of CLN . Student's t-test was used to compute p-values . ( D ) Co-staining for luminal cell marker K8 and basal cell marker K14 in control ( left ) and Msi2 overexpressing ( right ) mice . DOI: http://dx . doi . org/10 . 7554/eLife . 03915 . 02010 . 7554/eLife . 03915 . 021Figure 7—figure supplement 2 . Msi2 overexpression in mouse mammary gland represses Slug and Jag1 . ( A ) Staining for EMT marker Slug in control and Msi2 overexpressing mice . Scale bar: 50 μm . ( B ) Western blot for JAG1 protein in mammary epithelial cells of control and Msi2 overexpressing mice 7 weeks after induction with Dox . Arrow indicates expected JAG1 band ( 150 kD ) . ( C ) Immunohistochemistry for JAG1 protein in mammary gland from control and Msi2 overexpressing mice 7 weeks after induction with Dox . DOI: http://dx . doi . org/10 . 7554/eLife . 03915 . 021 We next examined the effect of Msi overexpression on epithelial cell state in the mammary gland in order to see whether its in vivo effects on epithelial-luminal state are similar to those observed in culture models . We ectopically expressed Msi2 in the basal cell layer , where it is nearly absent normally ( Figure 7A ) , using a basal cell-specific Dox-inducible driver , K14-rtTA . As expected , mice administered Dox showed significantly higher levels of MSI2 protein in the basal cell layer ( Figure 7—figure supplement 1A ) and overall higher levels of Msi2 mRNA in mammary epithelial cells ( Figure 7B ) . Overexpression of Msi2 altered mammary ductal branching morphology ( Figure 7C ) . Overexpression mice showed both a defective and delayed mammary ductal branching pattern . Msi2 overexpression resulted in fewer mammary duct branch points given , after either 4 or 7 weeks of induction with Dox , with the difference between controls and overexpression mice more pronounced after 7 weeks ( Figure 7—figure supplement 1B ) . The TEBs in glands overexpressing Msi2 were smaller relative to controls , following either 4 or 7 weeks of induction ( Figure 7C , right inset ) . In addition , after 4 weeks of induction , glands from overexpression mice had shorter ductal lengths relative to controls , but ductal lengths returned to lengths similar to wild type after 7 weeks of induction ( Figure 7—figure supplement 1C ) . These results indicate that Msi2 overexpression resulted in a defect in mammary branching morphogenesis ( evidenced by the reduced number of branch points ) , and a delay in this process , as indicated by the slower rate of branch ductal growth . Since branching morphogenesis requires cells to lose their epithelial identity and undergo migration , we hypothesized that the observed defect in branching morphology might result from inability of cells to lose their epithelial identity and/or expansion of an epithelial cell layer . Consistent with this hypothesis , we observed that Msi2 overexpression resulted in expansion of the luminal cell layer ( Figure 7D and Figure 7—figure supplement 1D ) , confirmed by a corresponding increase in expression of luminal cell markers and a decrease in basal markers ( Figure 7E ) . Furthermore , Msi2 overexpression led to an increase in epithelial marker E-cadherin and reduction in Slug , a marker of EMT and mesenchymal cells . Expression of EMT regulators Slug , Twist1 , and Twist2 decreased upon Msi2 overexpression , while expression of the luminal epithelial cell marker Gata3 increased ( Figure 7G and Figure 7—figure supplement 2A ) . Expression of JAG1 protein was also reduced upon Msi2 overexpression , consistent with the results observed in murine NSCs ( Figure 7—figure supplement 2B , C ) . These results support a model in which ectopic Msi expression leads to expansion of epithelial-luminal cells in the mammary gland , effectively blocking EMT processes required for normal branching morphogenesis , and resulting in the defective ductal branching pattern described above . The observed functions of Msi proteins in regulation of mammary epithelial cell state mirror the functions we observed in breast cancer cell lines and murine NSCs , and suggest that Msi proteins play similar roles in a healthy in vivo context as in cancer cells . The specific expression patterns of Msi proteins in stem and epithelial cells have aroused interest in their functional roles . Here , we show that Msi proteins are associated with the epithelial-luminal cell state in several cancer types , notably breast cancer , where Msi genes are highly enriched in luminal tumors and luminal breast cancer cell lines . We showed that in breast cancer cells , knockdown of Msi genes leads to loss of epithelial identity and upregulation of mesenchymal markers , while their ectopic activation promotes the epithelial state and suppresses mesenchymal properties such as cell migration . As in cancer cells , overexpression of Msi2 in healthy mammary gland tissue suppressed EMT and resulted in a defective mammary ductal branching pattern . These observations all support a role for Msi proteins in maintenance of a luminal/epithelial cell state and inhibition of EMT ( Figure 8 ) . The consistency between our observations in mammary epithelial cells and NSCs and between mouse and human suggests that these functions are shared across cell types and evolutionarily conserved . 10 . 7554/eLife . 03915 . 022Figure 8 . Model for Msi roles in regulation of cell state . Model for Msi role in the control of the epithelial state . We show that Msi represses translation of Jag1 , a positive regulator of Notch and EMT . We also show that Msi promotes expression of an epithelial-luminal splicing program , which we hypothesize occurs through translational regulation of splicing factors . In the model , both the direct regulation of Jag1 and indirect regulation of splicing contribute to maintenance of an epithelial-luminal cell state and inhibition of EMT . DOI: http://dx . doi . org/10 . 7554/eLife . 03915 . 022 Our genome-wide data support the hypothesis that Msi proteins are translational regulators . We showed that Msi proteins can translationally repress Jag1 , an important regulator of Notch signaling . However , the role of Notch signaling in cancer remains complex and may vary between cancer types ( Dickson et al . , 2007; Lobry et al . , 2011 ) . The upregulation of Jag1 in the basal state suggests that Notch pathway activity is high in and required for the entry into the mesenchymal state , consistent with previous studies ( Zavadil et al . , 2004; Dickson et al . , 2007 ) . In mammary epithelial cells , Jag1-triggered activation of Notch was shown to reduce E-cadherin expression and increase Slug expression ( Leong et al . , 2007 ) . Furthermore , Jag1 activation in breast cancer cells promotes their metastasis into the bone in vivo by activating Notch in neighboring bone cells ( Sethi et al . , 2011 ) . The dependence of EMT on Notch activation has been observed in normal development as well . During heart development , cardiac valves are generated from endocardium through EMT , and Notch activity was shown to be required for this process ( Timmerman et al . , 2004 ) . Collectively , these studies are consistent with our working model in which Msi represses Jag1 translationally , in turn altering Notch activity required for EMT . The molecular mechanisms by which Msi proteins regulate translation of a subset of mRNAs like Jag1 remains unclear . Our genome-wide data and in vitro binding assays indicate that Msi proteins act by binding UAG-containing motifs at 3' UTRs of messages . A model where Msi proteins repress translation by outcompeting eIF4G for PolyA-binding protein ( PABP ) was proposed ( Kawahara et al . , 2008 ) , but the conditions under which binding to mRNA results in translational repression are unclear , since only a subset of mRNAs are detectably regulated . It is possible that co-factors are required in vivo for Msi to affect translation following binding to the mRNA . It is also possible that other RNA-binding factors outcompete Msi protein for binding , though MSI1 has relative high RNA-binding affinity . The molecular mechanism underlying Musashi-dependent translational control and the nature of any co-factors involved are not known . This study complements recent reports of the involvement of post-transcriptional regulatory factors in cell state maintenance and EMT . For example , the epithelial-specific splicing factors of the ESRP family play important roles in maintenance of epithelial state ( Warzecha et al . , 2009; Reinke et al . , 2012 ) . A recent study presented evidence that the transcription factor Snail can promote the mesenchymal state in part by repressing Esrp1 ( Reinke et al . , 2012 ) , further highlighting the importance of post-transcriptional control in driving cell state transitions like EMT . Like master transcription factors , master post-transcriptional regulatory factors globally alter gene expression—by affecting RNA splicing , stability , localization , or translation—which makes them suitable for controlling cell identity ( Jangi and Sharp , 2014 ) . Our study shows that post-transcriptional regulatory factors like Msi proteins can impact both translation and pre-mRNA splicing , utilizing multiple layers of RNA regulation to reshape the transcriptome for a particular cell state . Many of the impacted splicing events are part of an epithelial splicing program , suggesting that effects of Msis on splicing may reinforce the effects of Jag1 repression on maintenance of epithelial cell state . The predominantly cytoplasmic expression of Msis makes it likely that splicing is affected indirectly , e . g . , through translational regulation of specific splicing factors , though our data do not rule out that a small fraction of Msi protein may be nuclear localized and could directly regulate splicing . We have also observed that other RBPs are also enriched in the epithelial state ( Shapiro et al . , 2011 ) , suggesting that RBPs as a group may play a broad role in maintenance of this state , and might provide attractive targets for therapeutic efforts to manipulate cell state . Msi proteins are co-expressed with various proliferation markers in a wide variety of stem cell niches , including the breast , stomach , intestine , lung , and brain . This observation suggests the hypothesis that Msis may act as general epithelial stem cell/progenitor regulators across tissues . Our findings are consistent with this hypothesis , but further study of Msi in multiple stem cell compartments will be needed to directly test it . The role of Msi in the normal development and transformation of other adult tissues will also be important to understand . For example , our observation that Msi is frequently overexpressed in lung tumors suggests that ectopic expression of Msi proteins in the lung could elucidate their role in lung cancer . Furthermore , the systematic downregulation of Msi1/Msi2 and high frequency of Msi1 mutations in kidney tumors suggests that kidney would be an informative model for studying Msi loss-of-function and its consequences in cancer . Inducible overexpression mice ( tetO-Msi1/Msi2 ) were generated as previously described in Beard et al . ( 2006 ) ; Kharas et al . ( 2010 ) . The generation of Msi2 conditional knockout mice was previously described in Park et al . ( 2014 ) , and the generation of Msi1 conditional knockout mice will be described elsewhere ( Yu et al . , under review ) . Mice of the 129SvJae strain were used , and the K14-rtTA strain was obtained from JAX ( stock number: 007678 ) . Animal care was in accordance with institutional guidelines and approved by the Committee on Animal Care , Department of Comparative Medicine , Massachusetts Institute of Technology , under animal protocol 1013-088-16 . For derivation of embryonic neural stem cells ( NSCs ) , littermate embryos were used whenever possible . Cortical NSCs were derived from embryos following Kim et al . ( 2003 ) . Briefly , cortical tissue was isolated from E12 . 5 embryos ( unless otherwise noted ) under a light dissection microscope inside a sterile fume hood and collected by centrifugation . Cortical tissues were dissociated into single cells by trituration in Magnesium/Calcium-free HBSS buffer ( Gibco , Woburn MA ) followed by 15-min incubation at room temperature . Dissociated tissue was collected by centrifugation , resuspended in N2 medium containing growth factors and Laminin ( Life Technologies , Woburn MA , Catalog Number: 23017015 ) and plated onto Polyornithin/Laminin-coated tissue culture dishes as in Okabe et al . ( 1996 ) . NSCs were grown in N2 medium ( Okabe et al . , 1996 ) containing EGF ( 20 ng/ml ) and bFGF ( 20 ng/ml ) and Laminin ( Life Technologies ) . Cells were grown on Polyornithin/Laminin-coated dishes . EMT was induced by switching cells to N2 medium containing LIF/FBS as described in Ber et al . ( 2012 ) . All breast cancer lines were cultured in DME containing 10% FBS , 1% GlutaMAX ( Gibco ) , and Penn/Strep , except for BT474 , which was cultured in RPMI base medium , and SKBR3 which was cultured with McCoy's 5A supplement . Lentiviruses carrying pLKO vectors with hairpins against Msi1 , Msi2 , or Luciferase ( control ) were used for knockdowns . Hairpins were obtained from Broad Institute shRNA library . Cells were infected in a centrifuge spin-infection step ( 1500 RPM , 37°C , 20 min ) following a 2-hr incubation with polybrene or protamine sulfate , and viral medium was added to the cells overnight . Cells were subjected to 4–6 day Puromycin selection ( 2 μg/ml ) 48 hr after infection . Msi1-OE vector ( Thermo OpenBiosystems ) was used for overexpression assays . Virus was prepared was described above and cell lines infected with virus were selected for 4–6 days with Blasticidin ( 5 μg/ml ) 48 hr after infection . Migration assay was performed using the transwells ( Corning 6 . 5 mm Diameter inserts with 8um pore size , polycarbonate membrane; product #3422 , lot #19614003 ) . 50 , 000 cells were seeded into wells in each condition and allowed to migrate for 9 hr . Cells were stained with Crystal Violet and then percent area covered was calculated using ImageJ . Images were threshold filtered on Hue and Saturation ( Hue: 192-255 'pass'; Saturation: 72-255 'pass' ) and passed to the ‘Analyze Particles’ function with a threshold size of 2000 . For western blotting , cells were lysed on ice and protein lysates were loaded onto 4-12% gradient Bis-Tris Gel ( Life Technologies ) . Primary antibodies and dilutions used in western blotting on murine NSCs: anti-MSI1/2 ( Cell Signaling Technology #2154 , 1:800 ) , anti-MSI2 ( Abcam #57341 , 1:800 ) , anti-Jag1 ( Cell Signaling Technology #2620 , 1:800 ) , anti-HER2 ( Cell Signaling Technologies #2248 , 1:1000 ) , anti-phos-HER2 ( Cell Signaling Technology #2241 , 1:1000 ) , anti-alpha-Tubulin ( Sigma-Aldrich T9026 , 1:5000 ) , anti-HNRNPA1 ( Abcam ab5832 , 1:800 ) . Immunofluorescene was performed on cells grown on glass bottom chambers ( LabTek II , #1 . 5 ) , fixed in 4% PFA . Cells were blocked and permeabilized in 5% FBS , . 1% Triton in PBS ( + ) . Antibodies were applied in 1% FBS in PBS ( + ) . Immunofluorescence antibodies and dilutions: anti-MSI1 ( MBL D270-3 , 1:500 ) , anti-HNRNP A2/B1 ( Santa Cruz , sc-374052 , 1:200 ) . For IHC on murine mammary glands , anti-Jag1 ( Santa Cruz , SC-6011 , 1:100 ) was used . For western on murine mammary glands , anti-Jag1 ( Santa Cruz , SC-6011 , 1:1000 ) and anti-Tubulin ( Sigma-Aldrich , T5168 , 1:4000 ) were used . Paraffin-embedded human breast cancer sections were obtained from Biomax US ( BR1505a ) and stained using standard protocols with antigen retrieval . Antibodies used: anti-ECAD1 ( BD Biosciences , 1:50 ) and anti-MSI1 ( MBL D270-3 , 1:200 ) . Confocal imaging was performed using a Perkin–Elmer microscope using oil-immersion 63× objective , imaged with Velocity software . Single confocal stacks or maximum Z intensity projections were obtained using Fiji ( Bioformats-LOCI plugin ) . RNA-Seq libraries were prepared from polyA-selected RNA using standard Illumina protocol . Ribosome profiling libraries were prepared following Ingolia et al . ( 2009 ) with several modifications . Briefly , cells were collected by centrifugation and immediately flash-frozen . Cells were thawed in lysis buffer ( 20 mM HEPES [pH 7 . 0] , 100 mM KCl , 5 mM MgCl2 , 0 . 5% Na-Deoxycholate , 0 . 5% NP-40 , 1 mM DTT , Roche mini EDTA-free protease inhibitor tablets [1 tablet/10 ml] ) and briefly treated with DNase I and RNAse I . Nuclei and cell debris were removed by centrifugation and lysates were treated with RNase I ( NEB ) for 75 min at room temperature to generate monosome-protected RNA fragments . Monosomes were collected by ultracentrifugation in a sucrose cushion , denatured in 8 M Guanidium HCl , and protected RNA fragments ( footprints ) were extracted with Phenol–Chloroform . Footprints were dephosphorylated by PNK treatment and size-selected ( ∼31–35 nt fragments ) by purification from a 15% TBE-Urea gel . Subtractive hybridization of ribosomal RNA from footprints was performed as in ( Wang et al . , 2012 ) . Footprints were then polyA-tailed , and Illumina sequencing adaptors were added in a reverse transcription step to obtain footprint cDNA , which was then isolated by gel purification . cDNA was then circularized , PCR-amplified , and PCR products isolated by gel purification and submitted for sequencing on Illumina Hi-Seq platform . Source code for the pipelines used to analyze RNA-Seq , ribosome profiling and Bind-n-Seq data is available through the open-source library rnaseqlib ( available at the git repository: http://www . github . com/yarden/rnaseqlib ) . Protocols , raw sequencing data and additional information about genomic datasets are available at http://www . musashi-genes . org . Early work on mammalian Musashi proteins by the Okano group and colleagues suggested that Numb mRNA is translationally repressed by MSI1 ( Okano et al . , 2002 ) . A later study by the same group showed that in the gastric system , Msi1 KO mice had lower , not higher , levels of Numb protein , opposite of the expected change under the translational repression model ( Takahashi et al . , 2013 ) . Recent work in HSCs ( where only Msi2 is expressed ) showed a Numb-independent phenotype for Msi2 and found that Msi2 KO HSCs have unchanged levels of Numb protein ( Park et al . , 2014 ) . Thus , it is unclear if Msi1 or Msi2 directly regulate Numb mRNA translation in all systems and whether such regulation always promotes or represses translation of the mRNA . In our data from NSCs , we were unable to detect a large difference in Numb translational efficiency upon Msi1 overexpression as measured by Ribo-Seq , though a small effect cannot be excluded since coverage of the Numb mRNA in our Ribo-Seq data was low . It is possible that Msi1 affects the translation of certain Numb mRNA isoforms in a context-specific manner , potentially through alternative mRNA processing of the Numb mRNA , as proposed by Takahashi et al . ( 2013 ) . All RNA sequencing data was submitted to GEO ( accession GSE58423 ) . Publicly available TCGA data sets ( Level 2 and Level 3 ) were downloaded from NIH ‘Bulk Download’ website ( RNASeqV2: https://wiki . nci . nih . gov/display/TCGA/RNASeq+Version+2 ) . RNA-Seq analyses were performed using ‘RNASeqV2’ TCGA files . Fold changes for genes were normalized by correction with Lowess-fit of MA-values calculated using raw gene expression estimates . Alternative exon expression was quantified using MISO . Syntenic regions for exons in mouse alternative exon trios ( mm9 ) were computed using Ensembl Compara Database ( Release 66 ) PECAN multiple genomes alignment , using the Pycogent Python framework ( Knight et al . , 2007 ) . Syntenic coordinates in human genome ( hg19 ) were then matched to annotated hg19 exon coordinates given in TCGA data files . A streptavidin binding peptide ( SBP ) tag was added to the pGEX6P-1 vector ( GE ) after the Presceission protease site . Full-length Musashi ( Msi1 ) was cloned downstream of the SBP tag with infusion ( Clontech ) using BamHI and NotI cloning sites . Expression of tagged MSI1 was induced with 0 . 5 mM IPTG at 18° for 4 hr in the Rosetta ( DE3 ) pLysS E . coli strain and subsequently purified on a GST GraviTrap column ( GE ) . MSI1 was eluted from the GST column with PreScission protease ( GE ) in 4 mL of Protease Buffer ( 50 mM Tris pH 7 . 0 , 150 mM NaCl , 1 mM EDTA , 1 mM DTT ) at 4° C overnight ( ∼16 hr ) . Protein purity was assayed SDS-PAGE gel electrophoresis and visualized with SimplyBlue SafeStain ( Invitrogen ) . Input random RNA was generated by T7 in vitro transcription: 1 μg T7 oligo was annealed to 1 μg of RBNS T7 template by heating the mixture at 65° C for 5 min then allowing the reaction to cool at room temperature for 2 min . The random RNA was then in vitro transcribed with HiScribe T7 In vitro transcription kit ( NEB ) according to manufacturer's instructions . The RNA was then gel-purified from a 6% TBE-urea gel . Nine concentrations of purified MSI1 ( 0 nM , 0 . 5 nM , 2 nM , 8 nM , 16 nM , 64 nM , 256 nM , 1 μM , and 2 μM ) were equilibrated in 250 μl of Binding Buffer ( 25 mM Tris pH 7 . 5 , 150 mM KCl , 3 mM MgCl2 , 0 . 01% Tween , 1 mg/ml BSA , 1 mM DTT , 30 μg/ml poly I/C [Sigma] ) for 30 min at room temperature . 40 U of Superasin ( Ambion ) and 1 μM random RNA ( final concentration ) was added to the MSI1 solutions and incubated for 1 hr at room temperature . During this incubation , Streptavidin magnetic beads ( Invitrogen ) were washed three times with 1 ml of wash buffer ( 25 mM Tris pH 7 . 5 , 150 mM KCl , 60 μg/ml BSA , 0 . 5 mM EDTA , 0 . 01% Tween ) and then equilibrated in Binding Buffer until needed . MSI1 and interacting RNA was pulled down by adding the RNA/protein solutions to 1 mg of washed streptavidin magnetic beads and incubated for 1 hr at room temperature . Supernatant ( unbound RNA ) was removed from the beads and the beads washed once with 1 ml of Wash Buffer . The beads were incubated at 70° for 10 min in 100 μl of Elution Buffer ( 10 mM tris pH 7 . 0 , 1 mM EDTA , 1% SDS ) and the supernatant was collected . Bound RNA was extracted from the eluate by phenol/chloroform extraction and ethanol precipitation . Half of the extracted RNA from each condition was reverse transcribed into cDNA using Superscript III ( Invitrogen ) according to manufacturer’s instructions using the RBNS RT primer . To control for any nucleotide biases in the input random library , 0 . 5 pmol of the RBNS input RNA pool was also reverse transcribed and Illumina sequencing library prep followed by 8–10 cycles of PCR using High Fidelity Phusion ( NEB ) . As Msi1 concentration was increased , decreasing input RT reaction was required in the PCR . For instance , the highest MSI1 condition required 30-fold less input RT product than the no MSI1 condition . All libraries were barcoded in the PCR step , pooled together , and sequenced one HiSeq 2000 lane . The Jag1 3' UTR was cloned into the pRL-SV40 vector ( Promega ) downstream of Renilla luciferase using the XbaI and NotI restriction sites creating the Renilla-Jag1-UTR construct . Firefly luciferase expression was used as the internal control and expressed from the PGL3 vector ( Promega ) . Renilla and the Firefly luciferase vectors were co-transfected into 293 cells stably expressing hairpins against Msi1 , Msi2 , or both Msi1 and Msi2 , or into mock transfected 293T cells . Cells were harvested between 30–36 hr after transfection and the Renilla and Firefly luciferase signals measured using the Dual-luciferase Reporter Assay System ( Promega ) according to manufacture's instructions . Mice were given Dox ( Sigma ) via drinking water at 2 g/l . Mice were induced with Dox for 7 weeks unless otherwise indicated . Inguinal mammary glands were spread on glass slides , fixed in Carnoy's fixative ( 6:3:1 , 100% ethanol: chloroform: glacial acetic acid ) for 2 to 4 hr at room temperature , washed in 70% ethanol for 15 min , rinsed through graded alcohol followed by distilled water for 5 min , then stained in carmine alum overnight , washed in 70% , 95% , 100% ethanol for 15 min each , cleared in xylene , and mounted with Permount . Mammary glands were fixed in 4% PFA , paraffin-embedded and 5-μm sections were used for immunofluorescence assay . Paraffin sections were microwave pretreated and incubated with primary antibodies , then incubated with secondary antibodies ( Invitrogen ) and counterstained with DAPI in mounting media . The following antibodies were used: anti-K14 ( Abcam ) , anti-K8 ( Abcam ) , anti-E-cadherin ( CST ) , anti-Msi2 ( Novus Biologicals ) , anti-Hes1 ( Abcam ) , anti-Slug ( CST ) . Mouse mammary epithelial cells were prepared according to the manufacturer's protocol ( StemCell Technologies , Vancouver , Canada ) . Briefly , following removal of the lymph node , mammary glands dissected from 10-week-old virgin female mice were digested in EpiCult-B with 5% fetal bovine serum ( FBS ) , 300 U/ml collagenase , and 100 U/ml hyaluronidase for 8 hr at 37°C . After vortexing and lysis of the red blood cells in NH4Cl , mammary epithelial cells were obtained by sequential dissociation of the fragments by gentle pipetting for 1–2 min in 0 . 25% trypsin , and 2 min in 5 mg/ml dispase plus 0 . 1 mg/ml DNase I ( DNase; Sigma ) . Total RNA was isolated from mammary epithelial cells . Complementary DNA was prepared using the MMLV cDNA synthesis kit ( Promega ) . Quantitative RT-PCR was performed using the SYBR-green detection system ( Roche ) . Primers were as follows: Msi2 forward primer: ACGACTCCCAGCACGACC; Msi2 reverse primer: GCCAGCTCAGTCCACCGATA . K8 forward primer: ATCAAGAAGGATGTGGACGAA; K8 Reverse primer: TTGGCAATGTCCTCGTACTG . K14 forward primer: CAGCCCCTACTTCAAGACCA; K14 Reverse primer: AATCTGCAGGAGGACATTGG . K18 forward primer: TGCCGCCGATGACTTTAGA; K18 Reverse primer: TTGCTGAGGTCCTGAGATTTG . RNA was extracted using Trizol and cDNA was prepared using SuperScript III ( Invitrogen ) . Primers used are listed below ( ‘h’ prefix denotes human gene , ‘F’ denotes forward primer , ‘R’ denotes reverse primer ) : hEcad-F: TGCCCAGAAAATGAAAAAGG hEcad-R: GTGTATGTGGCAATGCGTTC hTwist-F: GGAGTCCGCAGTCTTACGAG hTwist-R: TCTGGAGGACCTGGTAGAGG hEpCAM-F: CTTTAAGGCCAAGCAGTGCA hEpCAM-R: CGCGTTGTGATCTCCTTCTG hCD24-F: GGTTTGACTAGATGATGGATGCC hCD24-R: TCCATTCCACAATCCCATCCT hMsi1-F: GGGACTCAGTTGGCAGACTAC hMsi1-R: CTGGTCCATGAAAGTGACGAA hMsi2-F: ACCTCACCAGATAGCCTTAGAG hMsi2-R: AGCGTTTCGTAGTGGGATCTC hJag1-F: GTCCATGCAGAACGTGAACG hJag1-R: GCGGGACTGATACTCCTTGA	All living things start life as a single cell , but many organisms develop into a collection of different , specialized cells . Most of the cells in an organism can only divide to make more of the same type of cell; however , stem cells are different because they can ‘differentiate’ and develop into several different cell types . A key step in the development of an embryo is called the epithelial-to-mesenchymal transition , in which an epithelial cell—a cell type that normally lines body surfaces and cavities—begins to crawl away from the tissue it is in and starts to differentiate . This transition also allows cancer cells to leave tumors and spread around the body , in a process known as metastasis . In mammals , two proteins called Musashi1 and Musashi2 are abundant in stem cells and brain cancers , but are rarely found in specialized tissues and cells . Katz , Li et al . now find that the Musashi proteins are also often overexpressed in human breast , lung , and prostate tumors . In addition , Musashi proteins are much less abundant in cells that have completed an epithelial-to-mesenchymal transition . When Katz , Li et al . artificially reduced the amounts of Musashi proteins in breast cancer cells , the cells migrated and dispersed , as if becoming mesenchymal cells . Furthermore , many of the genes normally used in epithelial cells were switched off . In comparison , artificially increasing the levels of Musashi proteins halted the movement of mesenchymal cells and led to increased levels of genes used in epithelial cells , as if they were reverting to epithelial cells . Therefore , it appears that the Musashi proteins prevent epithelial cells from developing mesenchymal properties . Katz , Li et al . investigated how Musashi proteins work at the molecular level by studying neural and mammary cells in mice . This revealed that Musashi proteins control the steps that lead to the epithelial-to-mesenchymal transition by binding to the tail end of the RNA molecules that include the instructions to make certain proteins . This affects how often these proteins can be made from the RNA molecules . Katz , Li et al . suggest that Musashi proteins may similarly control the behavior of progenitor and stem cells in many other tissues as well; however , further study is needed to confirm this .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "cancer", "biology" ]	2014	Musashi proteins are post-transcriptional regulators of the epithelial-luminal cell state
During all stages of tumor progression , cancer cells are subjected to inappropriate extracellular matrix environments and must undergo adaptive changes in order to evade growth constraints associated with the loss of matrix attachment . A gain of function screen for genes that enable proliferation independently of matrix anchorage identified a cell adhesion molecule PVRL4 ( poliovirus-receptor-like 4 ) , also known as Nectin-4 . PVRL4 promotes anchorage-independence by driving cell-to-cell attachment and matrix-independent integrin β4/SHP-2/c-Src activation . Solid tumors frequently have copy number gains of the PVRL4 locus and some have focal amplifications . We demonstrate that the transformation of breast cancer cells is dependent on PVRL4 . Furthermore , growth of orthotopically implanted tumors in vivo is inhibited by blocking PVRL4-driven cell-to-cell attachment with monoclonal antibodies , demonstrating a novel strategy for targeted therapy of cancer . As many as 90% of all human cancers originate from epithelial tissues . Epithelia have a distinct ability to form and maintain highly organized monolayers , which is reflected in their role in providing the inner lining of hollow organs . This unique architecture is dictated by the requirement for an epithelial cell to be physically anchored on a basement membrane , an organizing substratum composed of specific extracellular matrix ( ECM ) molecules . Cells physically attach to ECM via integrins , a class of signaling molecules that serve to stimulate the survival and proliferation of cells in a matrix attachment-dependent manner ( Hynes , 2002 ) . Conversely , loss of contact with the proper ECM molecules results in initiation of a cell death program known as anoikis ( Frisch and Screaton , 2001 ) , and other constraints on cellular expansion . Early stages of epithelial cancer progression are universally characterized with genetic changes that confer ability to survive and proliferate in the absence of an appropriate matrix anchorage , which allows cellular expansion in a geometrically unconstrained manner . Though acquired early , the ability to tolerate the loss of anchorage remains critical for the survival of cancer cells throughout the course of disease progression , encompassing stages such as invasion of the underlying stroma , extravasation into blood vessels , survival in the bloodstream , and , eventually , metastatic outgrowth at a distant site with a distinct matrix composition . Along with the loss of the requirement for anchorage , a propensity for self-aggregation is a characteristic of aggressive cancer cells . Thus , tumor-derived subclones with greater metastatic capacity in vivo display increased self-aggregation in vitro; at the same time , subclones selected for increased in vitro aggregation were found to be more metastatic in mice ( Updyke and Nicolson , 1986; Saiki et al . , 1991 ) . Invasion of the underlying stroma is frequently undertaken by large groups of tumor cells , a phenomenon known as collective , or cohort , cell migration ( Friedl and Gilmour , 2009 ) . Clusters of circulating tumor cells ( CTCs ) have been identified from the blood samples of breast , colorectal , prostate , and lung cancer patients as well as from mouse tumor models ( Molnar et al . , 2001; Stott et al . , 2010; Hou et al . , 2011 ) . In particular , one report demonstrated that , on average , 50% of all breast and lung CTCs exist in circulation as aggregates ( Cho et al . , 2012 ) . An increase in clustering behavior is not limited to self-aggregation , as interactions with a variety of other cell types have been shown to be essential for the dissemination of cancer cells and subsequent metastatic colonization . Thus , heterotypic interactions of cancer cells with the endothelial lining within the target organ microvasculature were documented as an initiating event for metastatic lesion formation ( Al-Mehdi et al . , 2000 ) . Moreover , signaling elicited by the physical association of cancer cells with other cell types , such as platelets and macrophages , has been demonstrated to be essential for successful seeding and metastatic outgrowth ( Chen et al . , 2011; Labelle et al . , 2011 ) . Taken together , tumor-specific cell–cell contacts can provide a multifaceted survival advantage throughout the course of pathologic progression . Targeted therapies directed towards blocking such cell–cell contacts may therefore represent a novel cancer treatment approach . In this study , we have performed an unbiased genetic screen to identify genes that when overproduced promote anchorage-independent cell growth in a human mammary epithelial cell ( TL-HMEC ) line . Our screen identified a cell adhesion molecule PVRL4 ( poliovirus receptor-like 4 ) , also known as Nectin-4 , which we demonstrate to be a potent mediator of anchorage-independent colony formation in normal epithelial and cancer cells alike . We demonstrate that PVRL4 promotes the attachment of individual cells to each other via engaging its receptor PVRL1 on a juxtaposed cell . PVRL4-mediated cell-to-cell attachment triggers integrin β4 signaling in a matrix attachment-independent manner , and interfering with this pathway blocks PVRL4-driven anchorage-independence . Our findings point to a model in which signaling via PVRL4-PVRL1-driven cell–cell contacts serves as a surrogate for cell–matrix signaling in conditions of anchorage loss , thus enabling anoikis evasion and subsequent cellular expansion . Finally , we show that blocking PVRL4-driven cell–cell contact assembly with monoclonal antibodies potently inhibits anchorage-independent cellular expansion in vitro as well as the growth of orthotopically implanted tumors in vivo , thus validating the therapeutic utility of this approach . To identify novel genes involved in the transformation of epithelial cells , we carried out a genetic screen for colony formation in the absence of substratum attachment using a TL-HMEC in vitro transformation system . In this model , hTERT-immortalized human mammary epithelial cells are transduced with SV40 Large T antigen ( Zhao et al . , 2003 ) . TL-HMECs cannot survive and proliferate in a semi-solid medium , such as methylcellulose , in the absence of substratum attachment; however , ectopic expression of oncogenic H-RASV12 , myristoylated PI3K catalytic subunit ( Zhao et al . , 2003 ) , or shRNA against PTEN ( Westbrook et al . , 2005 ) enables anchorage-independent cellular expansion and formation of macroscopic colonies . We screened a retrovirus-encoded library of 8000 human open reading frames ( ORFeome 1 . 1 ) ( Rual et al . , 2004 ) ( Figure 1A ) and individually recovered stably integrated ORFs from 732 resulting macroscopic colonies by PCR , followed by determination of the ORF identity by sequencing . Forty ORFs that were identified in two independent screen replicates were individually validated to confirm the anchorage-independent growth phenotype ( Figure 1B , see also Supplementary file 1A ) . Eleven ORFs potently induced anchorage-independent colony formation ( a fivefold increase over background ) , and another eight ORFs produced a moderate phenotype ( a two- to fivefold increase over background ) . According to the recently published copy number variation analysis across a variety of tumor types ( Beroukhim et al . , 2010 ) , eight out of 11 ORFs that potently induced anchorage-independent growth localized to statistically defined peaks of focal amplification in at least one tumor subtype ( p=0 . 009 , Fisher’s exact test ) , providing strong evidence for positive selection of these genes in cancer ( Supplementary file 1B ) . In addition , ORFs corresponding to two genes previously shown to promote anchorage-independent colony formation were identified in our screen . Ectopic expression of SULF2 ( sulfatase 2 ) had been previously found to induce anchorage-independent growth in human bronchial epithelial cells ( Lemjabbar-Alaoui et al . , 2010 ) and NPM1 ( nucleophosmin 1 ) overexpression had been previously found to enable NIH 3T3 fibroblast colony formation in methylcellulose ( Kondo et al . , 1997 ) . 10 . 7554/eLife . 00358 . 003Figure 1 . A genetic screen for drivers of anchorage-independent growth in human mammary epithelial cells . ( A ) A schematic of the screen . TL-HMECs were transduced with the ORFeome library ( 8000 CMV promoter-driven human open reading frames [ORFs] in a retroviral vector ) at a multiplicity of infection ( MOI ) of 0 . 2 and plated into semi-solid medium . Macroscopic colonies were isolated , individually expanded , and the identities of ORF inserts were determined by sequencing . ( B ) ORFs recovered from two independent screen replicates were individually transduced into TL-HMECs and plated into semi-solid medium . Colonies were counted and colony numbers were normalized to an empty vector-transduced sample . Asterisks denote strongly validated ORFs that localize to focal amplification peaks in at least one tumor subtype . Assays were performed in triplicate ( error bars ± SD ) . DOI: http://dx . doi . org/10 . 7554/eLife . 00358 . 003 We focused our attention on the top scoring candidate identified in the screen , PVRL4 ( poliovirus receptor-like 4 ) . Ectopic expression of PVRL4 strongly induced anchorage-independent colony formation ( Figure 1B ) , whereas its endogenous expression was not detectable in TL-HMECs ( Figure 2B ) , consistent with the documented absence of its expression in normal mammary gland ( Fabre-Lafay et al . , 2007 ) . 10 . 7554/eLife . 00358 . 004Figure 2 . PVRL4-induced anchorage-independent colony formation is carried out through its extracellular region . ( A ) and ( B ) A series of PVRL4 deletion constructs were designed and their expression confirmed by Western blot . ( C ) PVRL4 mutants from ( A ) were tested for their ability to induce anchorage-independent colony formation in triplicate ( error bars ± SD ) . ( D ) Cells with full-length PVRL4 or the cytoplasmic region deletion mutant were assayed for viability under conditions of anchorage deprivation by measuring total ATP content in cells cultured on ultra-low attachment plates for 72 hr . Values were normalized to an empty vector-transduced sample . Assays were performed in triplicate ( error bars ± SD ) . ( E ) and ( F ) TL-HMECs expressing empty vector , full-length PVRL4 or cytoplasmic region deletion mutant containing cells were cultured on tissue culture-treated ( adherent ) or ultra-low attachment ( suspension ) dishes for 72 hr . RNA was isolated and mRNA levels for TGM1 ( E ) and KRT6A and IVL ( F ) were measured by RT-qPCR . Transcript levels were normalized to β-actin . qPCR was performed in quadruplicate ( error bars ± SD ) . DOI: http://dx . doi . org/10 . 7554/eLife . 00358 . 004 Recently identified to be a receptor for measles virus , PVRL4 ( Muhlebach et al . , 2011; Noyce et al . , 2011 ) belongs to a family of poliovirus receptor-related transmembrane proteins , also known as nectins , which localize to sites of cell junctions in various cell types ( reviewed in Takai et al . , 2008 ) . Similarly to other nectins , PVRL4 contains an extracellular region with three immunoglobulin-like domains , a single transmembrane domain , and a short cytoplasmic tail ( Reymond et al . , 2001 ) . Extracellular regions of nectins engage in trans-interactions on the interface of two juxtaposed cells . Through a stretch of four C-terminal amino acids , trans-interacting nectins recruit afadin , a scaffold-like mediator of downstream signaling , triggering assembly of the afadin/Rap1/p120ctn complex , which positively regulates membrane retention of E-cadherin , assisting in adherens junction ( AJ ) formation in MDCK cells ( Hoshino et al . , 2005 ) . To identify regions of PVRL4 responsible for driving anchorage-independent growth , we created a series of PVRL4 deletion constructs ( Figure 2A , B , see also Supplementary file 1C ) . Remarkably , deletion of the entire cytoplasmic region of PVRL4 did not affect its ability to promote TL-HMEC colony formation ( Figure 2C ) or cell viability ( Figure 2D ) in the absence of substratum attachment . Mammary epithelial cells are known to undergo an alternative anoikis-related cell death program characterized by robust induction of transcripts associated with terminal squamous differentiation when apoptosis is suppressed ( Mailleux et al . , 2007 ) . TL-HMECs in particular do not initiate classic apoptotic responses in response to the loss of substratum attachment , likely due to the anti-apoptotic action of Large T antigen ( not shown ) , but instead potently upregulate multiple squamous differentiation markers , including TGM1 ( transglutaminase 1 ) , KRT6A ( keratin 6 ) , and IVL ( involucrin ) , and we found that both full-length PVRL4 and its cytoplasmic deletion mutant reduced this effect ( Figure 2E , F ) . Taken together , these data demonstrate that PVRL4 protects TL-HMECs from the differentiation associated with anchorage loss and promotes subsequent cellular expansion . Importantly , this phenotype is facilitated by the extracellular part of PVRL4 rather than by its intracellular region . Consistent with its role in promoting cell–cell contact formation , PVRL4 drives rapid association of TL-HMECs into multicellular clusters in suspension ( Figure 3A , B ) . We sought to rule out the possibility that PVRL4-induced colonies were due to an artifact associated with either incomplete dissociation of cells prior to seeding into methylcellulose or potential de novo association of cells during growth in anchorage-independent conditions . To address this , we mixed equal numbers of dsRed- and GFP-labeled TL-HMECs expressing PVRL4 and ( i ) seeded them into methylcellulose or ( ii ) co-cultured the mixed population on an adherent surface , subsequently seeding them into methylcellulose . Examining the colors of resulting colonies revealed that out of 56 colonies from sample ( i ) , all 56 were single-color colonies , whereas in sample ( ii ) , 39 out of 40 colonies were single-color , and only one colony contained both GFP and dsRed-positive cells ( Figure 3—figure supplement 1 ) . The rarity of double-color colonies demonstrates that PVRL4-induced colonies originate from individual cells . Consistent with this observation , passing cells through a 35 μm cell strainer immediately prior to being seeded in the absence of anchorage did not affect colony numbers ( Figure 3—figure supplement 2 ) , further confirming that PVRL4-induced colonies are clonal in origin and that the observed increase in colony numbers is not a result of pre-existing multicellular clusters having a survival advantage over single cells in the absence of anchorage . 10 . 7554/eLife . 00358 . 005Figure 3 . PVRL4 facilitates cell-to-cell attachment , inhibition of which suppresses anchorage-independence . ( A ) and ( B ) PVRL4 promotes cell clustering of TL-HMECs . Cells were dissociated off the tissue culture surface with trypsin-free cell dissociation buffer and kept in suspension for 1 hr . Small ( 3–5 cells ) and large ( >5 cells ) cell clusters per field of view were counted , n = 3 ( error bars ± SD ) . ( C ) GFP-labeled PVRL4-expressing TL-HMECs were allowed to aggregate with dsRed-labeled cells expressing either a PVRL1-targeting shRNA or a control shRNA . Representative phase-contrast and fluorescent images ( red and green channels superimposed ) are shown . ( D ) PVRL4 was co-expressed with the indicated shRNAs and anchorage-independent colony formation in TL-HMECs was assayed . Values were normalized to an empty vector-transduced sample . Assays were performed in triplicate ( error bars ± SD ) . ( E ) PVRL4-expressing TL-HMECs were assayed for clustering in the presence of the indicated antibodies or isotype controls . Cell clusters were quantified as before . ( F ) Anchorage-independent growth induced by PVRL4 or an shRNA against PTEN was assayed in the presence of PVRL4-targeting antibody or control IgG . Colony numbers were normalized to the control sample . Anchorage-independent colony formation assays were performed in triplicate ( error bars ± SD ) . DOI: http://dx . doi . org/10 . 7554/eLife . 00358 . 00510 . 7554/eLife . 00358 . 006Figure 3—figure supplement 1 . PVRL4-driven anchorage-independent colonies originate from single cells . PVRL4 expressing TL-HMECs were stably transduced with dsRed or GFP and mixed in equal proportions , followed by ( i ) immediate plating into semi-solid medium , or ( ii ) co-culturing on an adherent surface for 2 d , followed by plating into semi-solid medium . Resulting colonies were visualized under a fluorescent microscope and each colony was assessed for the presence of red and green fluorescence . Representative phase-contrast and fluorescent images ( red and green channels ) are shown . DOI: http://dx . doi . org/10 . 7554/eLife . 00358 . 00610 . 7554/eLife . 00358 . 007Figure 3—figure supplement 2 . Potential preformed clusters of TL-HMECs do not contribute to anchorage-independent colony numbers . The colony formation efficiency of TL-HMECs transduced with empty vector or vector expressing PVRL4 was compared between unfiltered cell suspensions and cell suspensions that were filtered through a 35 μm nylon mesh strainer prior to plating into methylcellulose . Anchorage-independent colony formation assays were performed in triplicate ( error bars ± SD ) . DOI: http://dx . doi . org/10 . 7554/eLife . 00358 . 00710 . 7554/eLife . 00358 . 008Figure 3—figure supplement 3 . Depletion efficiency of individual anti-PVRL1 shRNAs . efficiency of shRNA-mediated PVRL1 mRNA depletion was measured by RT-qPCR . PVRL1 transcript abundance was normalized to β-actin . qPCR was performed in quadruplicate ( error bars ± SD ) . DOI: http://dx . doi . org/10 . 7554/eLife . 00358 . 00810 . 7554/eLife . 00358 . 009Figure 3—figure supplement 4 . Anti-PVRL4 antibodies block PVRL4-driven cell–cell clustering . PVRL4-expressing TL-HMECs were allowed to aggregate in the presence of the indicated antibodies or isotype controls . Representative images are shown . DOI: http://dx . doi . org/10 . 7554/eLife . 00358 . 00910 . 7554/eLife . 00358 . 010Figure 3—figure supplement 5 . PVRL4-driven cell–cell clustering is inhibited by antibodies against PVRL1 . PVRL4-expressing TL-HMECs were assayed for clustering in the presence of the indicated antibodies or isotype controls . Cell clusters were quantified as before . DOI: http://dx . doi . org/10 . 7554/eLife . 00358 . 010 The extracellular region of PVRL4 exhibits preferential affinity for trans-interaction with PVRL1 ( Reymond et al . , 2001 ) , a cell adhesion molecule endogenously expressed in TL-HMECs . To test whether PVRL1 was necessary for PVRL4-driven cell-to-cell attachment , we used an shRNA to stably deplete the PVRL1 transcript in a dsRed-labeled population of TL-HMECs and allowed cells to aggregate with PVRL4-expressing GFP-labeled TL-HMECs . PVRL1 depletion resulted in exclusion of dsRed-labeled cells from multicellular clusters , whereas control shRNA-expressing cells were readily incorporated ( Figure 3C ) , suggesting that PVRL4-mediated cell–cell contacts are carried out through interaction with its receptor PVRL1 . We then asked whether PVRL1 was similarly required for PVRL4-driven anchorage-independence . Indeed , stable depletion of the PVRL1 transcript by two independent shRNA constructs ( Figure 3—figure supplement 3 ) abolished anchorage-independent colony formation ( Figure 3D ) , paralleling the effect of PVRL1 depletion on PVRL4-driven cell-to-cell attachment . The above data raise the possibility that PVRL4-PVRL1 trans-interaction on juxtaposed cells is mechanistically involved in driving anchorage-independent growth . Therefore , we sought a physical means by which to disrupt PVRL4-induced cell-to-cell attachment to examine its effects on colony formation . We asked whether antibodies directed towards the extracellular domain of PVRL4 could block PVRL4-induced cluster formation and anchorage-independent growth . Cell clustering was completely abrogated in the presence of three independent PVRL4 antibodies , whereas control IgG or a blocking antibody against E-cadherin ( DECMA-1 ) did not produce such an effect ( Figure 3E , Figure 3—figure supplement 4 ) . Similarly , an antibody targeting the extracellular region of PVRL1 inhibited PVRL4-induced cell clustering ( Figure 3—figure supplement 5 ) . PVRL4-driven colony formation was inhibited to almost a basal level in the presence of a monoclonal antibody targeting the extracellular region of PVRL4 , but no such effect was observed on colonies induced by PTEN shRNA ( Figure 3F ) , demonstrating that the observed inhibition is due to a direct effect of an antibody and not due to its general toxicity . Our data indicate that PVRL4 enables anchorage-independent growth in a manner which is dependent on PVRL4-PVRL1-driven cell-to-cell attachment , suggesting that these intercellular contacts may be providing a survival benefit in the context of altered matrix anchorage . To further probe the functional link between PVRL4-driven cell-to-cell attachment and anchorage-independent growth , we asked whether the PVRL4-PVRL1-mediated cell surface interaction alone is sufficient for this phenotype . To address this question , we created chimeric constructs in which extracellular regions of PVRL1 and PVRL4 were fused to the transmembrane regions of an unrelated transmembrane molecule , CD8 ( Figure 4A ) , and the cytoplasmic regions of both molecules were deleted . We first introduced PVRL4-CD8tm into TL-HMECs , concomitantly depleting endogenous PVRL1 by RNAi ( Figure 4D ) . Membrane localization of the chimeric construct was verified by immunofluorescence ( Figure 4—figure supplement 1 ) . Expression of PVRL4-CD8tm induced both cell clustering and anchorage-independent colony formation , and both phenotypes were suppressed by stable depletion of endogenous PVRL1 ( Figure 4B , C ) . Importantly , clustering was restored when two populations of clustering-incompetent cells , PVRL4-CD8tm expressing , PVRL1-depleted cells ( PVRL4+; PVRL1− ) and control cells ( PVRL4−; PVRL1+ ) , were mixed together , independently verifying that cell-to-cell attachment is mediated by the PVRL4-PVRL1 trans-interacting module ( Figure 4—figure supplement 2 ) . Both cell clustering and colony formation defects induced by PVRL1 shRNA were rescued by the expression of the shRNA-resistant PVRL1-CD8tm construct . Taken together , these data indicate an on-target nature of the PVRL1 depletion phenotype and demonstrate that ability to withstand the absence of anchorage is driven by the PVRL4-PVRL1 cell-surface trans-interaction and not by intracellular or transmembrane regions of either molecule . 10 . 7554/eLife . 00358 . 011Figure 4 . Expression of extracellular regions of PVRL4 and PVRL1 on the cell surface is sufficient for anchorage-independence . ( A ) Schematics of chimeric constructs containing extracellular domains of PVRL4 or an shRNA-resistant version of PVRL1 fused to the transmembrane domain of CD8 ( blue ) . ( B ) and ( C ) TL-HMECs were stably transduced with the indicated combinations of expression constructs and assayed for anchorage-independent growth ( B ) and clustering ( C ) . Colony numbers were normalized to the control sample . Anchorage-independent colony formation assays were performed in triplicate ( error bars ± SD ) . ( D ) Expression levels of endogenous and chimeric proteins were verified by Western blot . DOI: http://dx . doi . org/10 . 7554/eLife . 00358 . 01110 . 7554/eLife . 00358 . 012Figure 4—figure supplement 1 . Plasma membrane localization of PVRL4-CD8tm construct in TL-HMECs . TL-HMEC cells infected with PVRL4-CD8tm or empty vector were fixed with methanol and stained with goat polyclonal anti-PVRL4 antibody followed by anti-goat Alexa Fluor 488 secondary antibody . DOI: http://dx . doi . org/10 . 7554/eLife . 00358 . 01210 . 7554/eLife . 00358 . 013Figure 4—figure supplement 2 . Cell–cell clustering is driven by PVRL4-PVRL1 trans-interactions between individual cells . Clustering assays were performed with TL-HMECs expressing the following transgenes: ( i ) empty vector/control shRNA; ( ii ) PVRL4-CD8tm/control shRNA; ( iii ) PVRL4-CD8tm/anti-PVRL1 shRNA; and ( iv ) a 1:1 mixture of ( iii ) and ( i ) . DOI: http://dx . doi . org/10 . 7554/eLife . 00358 . 01310 . 7554/eLife . 00358 . 014Figure 4—figure supplement 3 . Expression of shRNA constructs against PTEN or constitutively active mutants of RAS and PI3K induces anchorage-independent growth . TL-HMECs transduced with the indicated constructs were assayed for ability to induce anchorage-independent colony formation in triplicate ( error bars ± SD ) . DOI: http://dx . doi . org/10 . 7554/eLife . 00358 . 01410 . 7554/eLife . 00358 . 015Figure 4—figure supplement 4 . Constitutively active mutants of RAS and PI3K induce cell–cell clustering . TL-HMECs were stably transduced with the indicated constructs and assayed for cell–cell clustering and colony formation in the absence of substratum anchorage . Representative clustering assay images are shown . DOI: http://dx . doi . org/10 . 7554/eLife . 00358 . 01510 . 7554/eLife . 00358 . 016Figure 4—figure supplement 5 . PTEN depletion induces cell–cell clustering . TL-HMECs expressing two independent PTEN shRNAs or control shRNA were assayed for cell–cell clustering . Representative images are shown . DOI: http://dx . doi . org/10 . 7554/eLife . 00358 . 01610 . 7554/eLife . 00358 . 017Figure 4—figure supplement 6 . Depletion efficiency of individual anti-PTEN shRNAs . The efficiency of PTEN mRNA depletion was measured by RT-qPCR . PTEN transcript abundance was normalized to β-actin . qPCR was performed in quadruplicate ( error bars ± SD ) . DOI: http://dx . doi . org/10 . 7554/eLife . 00358 . 01710 . 7554/eLife . 00358 . 018Figure 4—figure supplement 7 . Cell–cell clustering induced by depletion of PTEN is heterotypic . GFP-labeled PTEN-depleted TL-HMECs were allowed to aggregate with dsRed-labeled control shRNA-expressing cells . Images were taken and processed as before . Representative images are shown . DOI: http://dx . doi . org/10 . 7554/eLife . 00358 . 018 We have shown that PVRL4-driven anchorage-independent growth is dependent on its involvement in driving cell-to-cell attachment , which raises an interesting possibility that in the context of the altered matrix environment that tumor cells encounter , certain types of intercellular interactions function to enable anchorage-independent survival . Along these lines , we hypothesized that other oncogenic perturbations that drive anchorage-independent growth ( Figure 4—figure supplement 3 ) may also promote cell-to-cell attachment as a part of their survival strategy . Either expression of mutant RasV12 , or myristoylated PI3K catalytic subunit ( Figure 4—figure supplement 4 ) , or shRNA-mediated depletion of PTEN ( Figure 4—figure supplement 5; for depletion efficiencies see Figure 4—figure supplement 6 ) triggered multicellular cluster formation in TL-HMECs . None of the perturbations induced endogenous PVRL4 expression , as measured by FACS ( not shown ) , which suggests that clustering in these cases was mediated through other adhesion mechanisms . Mixing PTEN shRNA-expressing GFP-labeled cells with control shRNA-expressing dsRed-labeled cells triggered the formation of multicellular clusters that contained both GFP- and dsRed-labeled cells ( Figure 4—figure supplement 7 ) , demonstrating that , in a manner similar to PVRL4 , depletion of PTEN enabled attachment of TL-HMECs to cells that do not carry the perturbation as well as those that do . Taken together , these observations suggest that increased cell-to-cell attachment is a phenotype that multiple oncogenic perturbations converge upon . We next sought to gain insight into the mechanism by which PVRL4-PVRL1 cell surface trans-interaction becomes translated into survival advantage in conditions of anchorage loss . Neither of the two molecules possesses catalytic activity; in addition , our data show that transmembrane as well as cytoplasmic regions are dispensable for anchorage-independence . Therefore , we considered the possibility that the assembly of a PVRL4-PVRL1 interacting module on the interface of two neighboring cells may be triggering lateral recruitment and/or activation of cell surface-localized proteins , which , in turn , conveys a prosurvival signal . To identify potential cell surface-localized binding partners of PVRL4 , we created a C-terminally HA/FLAG-tagged PVRL4 construct and first verified that cell–cell clustering as well as anchorage-independent growth phenotypes were not affected by the addition of a C-terminal tag ( not shown ) . We then performed immunoprecipitations with anti-HA agarose beads from lysates of TL-HMECs induced to express either HA/FLAG-tagged PVRL4 or HA/FLAG-tagged GFP and subjected the eluates to tandem mass spectrometry analysis . Since only the extracellular regions of PVRL4 and PVRL1 are required for anchorage-independent colony formation , we searched the list of PVRL4 IP-specific peptides for those that had at least three unique peptides corresponding to cell surface-localized proteins , as classified by Gene Ontology . We identified the transmembrane protein integrin β4 as specifically interacting with HA/FLAG-PVRL4 ( Figure 5A ) . To validate the putative PVRL4–integrin β4 interaction directly in cells under conditions of anchorage deprivation , we performed immunoprecipitations with anti-HA beads from lysates of suspension-incubated TL-HMECs that expressed either HA/FLAG-tagged PVRL4 or HA/FLAG-tagged GFP . Immunoblotting for integrin β4 verified a specific association of integrin β4 with PVRL4 ( Figure 5B ) . We next asked if the integrin β4–PVRL4 association was affected by cell clustering . The same amount of integrin β4 coprecipitated with HA/FLAG-tagged PVRL4 from lysates prepared from multicellular clusters and from single-cell suspension . Similarly , the amount of coprecipitating integrin β4 was not affected by RNAi-mediated depletion of PVRL1 ( Figure 5—figure supplement 1A ) . These data point to a cis mode of association between PVRL4 and integrin β4 on the membrane of the same cell . 10 . 7554/eLife . 00358 . 019Figure 5 . PVRL4-driven cell-to-cell attachment promotes anchorage-independence via integrin β4-associated signaling . ( A ) Cell surface-localized proteins interacting with HA/FLAG-tagged PVRL4 , but not with HA/FLAG-tagged GFP , as determined by mass spectrometry . ( B ) TL-HMECs expressing HA/FLAG tagged PVRL4 or HA/FLAG-tagged GFP were detached from the adherent surface with the enzyme-free cell dissociation buffer and incubated in suspension for 1 hr . Immunoprecipitations were performed with HA beads , followed by Western blot with FLAG and integrin β4 antibody . ( C ) TL-HMECs expressing vector control , PVRL4 , or myr-PI3K were stably transduced with the indicated shRNA constructs and integrin β4 levels were assayed by Western blot . ( D ) TL-HMECs from ( C ) were assayed for anchorage-independent colony formation . Colony numbers were normalized to the vector control sample . Assays were performed in triplicate ( error bars ± SD ) . ( E ) TL-HMECs stably transduced with the indicated constructs were detached from the adherent surface with enzyme-free cell dissociation buffer and incubated in 0 . 5% methylcellulose in suspension for 6 hr or cultured on an adherent surface for 48 hr . Levels of pY416-SFK ( Src family kinases ) , total SFK , and vinculin loading control were measured by Western blot . Band intensity was measured with ImageJ software . ( F ) PVRL4-expressing TL-HMEC cells transduced with control or anti-PVRL1 shRNA were incubated in suspension in the conditions indicated . Levels of pY416-SFK ( Src family kinases ) and tubulin loading control were measured by Western blot . Band intensity was measured with ImageJ software . ( G ) TL-HMECs stably expressing PVRL4 or control vector were assayed for anchorage-independent colony formation in the presence of PP2 or vehicle control . Colony numbers were normalized to the vector sample . Assays were performed in triplicate ( error bars ± SD ) . ( H ) TL-HMECs expressing PVRL4 were stably transduced with the indicated shRNA constructs and assayed for anchorage-independent colony formation . Colony numbers were normalized to the vector control sample . Assays were performed in triplicate ( error bars ± SD ) . ( I ) SHP-2 levels were assayed by Western blot in TL-HMEC lysates from ( H ) . DOI: http://dx . doi . org/10 . 7554/eLife . 00358 . 01910 . 7554/eLife . 00358 . 020Figure 5—figure supplement 1 . Interaction of PVRL4 with integrin β4 . ( A ) Anti-HA beads were used to immunoprecipitate HA/FLAG-PVRL4 from the indicated lysates . Immunoprecipitates were blotted with anti-integrin β4 and anti-FLAG antibodies . ( B ) Immunoprecipitations with anti-integrin β4 antibodies or with a control IgG were performed from TL-HMEC lysates expressing either full-length PVRL4 or its cytoplasmic deletion mutant . Immunoprecipitates and input lysates were blotted with anti-integrin β4 and anti-PVRL4 antibodies . Asterisks denote heavy chains of control and anti-integrin β4 antibodies , which cross-react with an anti-goat secondary antibody . DOI: http://dx . doi . org/10 . 7554/eLife . 00358 . 020 We took further steps to investigate the interaction between integrin β4 and PVRL4 , asking whether deletion of the cytoplasmic region of PVRL4 will disrupt its association with integrin β4 . Immunoprecipitation of integrin β4 from TL-HMECs which were induced to express either a full-length or a cytoplasmic deletion version of PVRL4 revealed that both full-length and mutant versions of PVRL4 coprecipitated with integrin β4 , and not with a control IgG ( Figure 5—figure supplement 1B ) . This finding parallels our earlier demonstration that the cytoplasmic region of PVRL4 is dispensable for the anchorage-independence phenotype , prompting us to investigate a potential mechanistic link between integrin β4 and the PVRL4-driven anchorage-independence phenotype . To directly test the involvement of integrin β4 in PVRL4-driven anchorage-independence , we used two independent shRNA constructs for stable depletion of integrin β4 from TL-HMECs ( Figure 5C ) . Depletion of integrin β4 had no effect on TL-HMEC attachment to the tissue culture vessel or on proliferation in adherent conditions; neither did it affect PVRL4-driven cell clustering ( not shown ) . However , PVRL4-induced anchorage-independent colony formation was markedly reduced in the presence of integrin β4-specific shRNAs ( Figure 5D ) . Importantly , reliance on integrin β4 was specific to colony growth promoted by PVRL4 , as PI3K-driven colony numbers were not affected by its depletion . A unique member of the integrin family , integrin β4 contains an atypically long C-terminal region ( 1017 amino acids ) which does not take part in canonical focal adhesion formation , but instead , creates attachment points for keratin-containing intermediate filaments in adherent cells , forming hemidesmosomes ( Giancotti , 2007 ) . In contrast to other integrins , the C-terminus of integrin β4 does not bind FAK . However , it can activate Src family kinases ( SFKs ) in a FAK-independent manner via recruitment of SHP-2 phosphatase . In substratum-attached cells , ligation of integrins to the matrix maintains a constitutive level of active SFKs , whereas loss of anchorage inactivates SFKs . To test whether cells in PVRL4-induced clusters are able to maintain SFK activity in conditions of anchorage loss , we measured levels of autophosphorylated SFKs in TL-HMECs after 6 hr incubation in suspension ( Figure 5E ) . Levels of activated SFK dropped precipitously in control cells while remaining high in PVRL4-expressing cells but not in PVRL4-expressing , PVRL1-depleted cells . Moreover , the addition of anti-PVRL4 antibody also reduced the level of SFK autophosphorylation in suspension . A similar effect was observed after a short-term incubation of PVRL4-expressing TL-HMECs in suspension ( Figure 5F ) , where levels of SFK activation observed in freshly detached cells ( 0 hr time point ) were maintained in cells that were allowed to attach to each other but not in cells in which clustering was prevented by either anti-PVRL4 antibody or PVRL1 depletion . Having shown that PVRL4-driven cell-to-cell attachment could preserve SFK activation even in the absence of substratum attachment , we hypothesized that SFK activation is a critical driver of the observed ability of PVRL4-expressing cells to withstand the loss of anchorage . In support of this , PVRL4-driven but not PI3K-driven anchorage-independent colony formation was abrogated in the presence of PP2 , a chemical inhibitor of SFK activity ( Figure 5G ) . Activation of SFKs by integrin β4 is carried out via recruitment of SHP-2 phosphatase . Since PVRL4-driven anchorage-independent growth is inhibited both by integrin β4 depletion and by chemical inhibition of SFK activity , we reasoned that depletion of SHP-2 might have a similar effect . Indeed , two independent shRNAs against SHP-2 completely abrogated colony formation induced by PVRL4 ( Figure 5H , I ) . Taken together , these data suggest that PVRL4-driven anchorage-independent growth is facilitated via activation of the integrin β4/SHP-2/Src signaling pathway in a manner that is dependent on cell-to-cell attachment . Having uncovered a role for PVRL4 in enabling epithelial cells to escape growth restriction associated with the lack of proper matrix anchorage , we next sought to test its involvement in driving the tumorigenic properties of breast cancer cells . We first focused on an inflammatory breast tumor cell line SUM190 , which contains a particularly high-level focal amplification ( ∼50 kb ) of the PVRL4 genomic locus ( Figure 6A ) ( Beroukhim et al . , 2010 ) . To ask whether PVRL4 is involved in enabling anchorage-independent growth of these cells , we designed four independent shRNAs targeting PVRL4 and stably expressed them in SUM190 cells , confirming the extent of depletion by RT-qPCR ( Figure 6B ) . Paralleling effects seen in TL-HMECs , spontaneous self-clustering of SUM190 cells was abrogated by anti-PVRL4 antibodies ( Figure 6—figure supplement 1 ) or by RNAi against PVRL4 ( Figure 6—figure supplement 2 ) . Similarly , heterotypic attachment of SUM190 cells with human lung microvascular endothelial ( HMVEC-L ) cells was prevented by anti-PVRL4 antibodies ( Figure 6—figure supplement 3 ) and by RNAi ( Figure 6—figure supplement 4 ) . RNAi against PVRL4 potently reduced both anchorage-independent colony formation and clonogenic growth of SUM190 cells , indicating that PVRL4 is involved in the growth and survival of breast cancer cells . Importantly , observed effects correlated with the degree of RNAi-mediated mRNA depletion ( Figure 6B ) . To verify that the effects of PVRL4 depletion are not limited to SUM190 cells , we expressed anti-PVRL4 shRNAs in two additional breast cancer cell lines , Sk-BR-3 and BT-474 ( Figure 6—figure supplement 5 ) , which both express lower levels of PVRL4 than SUM190 cells ( Fabre-Lafay et al . , 2007 ) , and observed a strong defect in clonogenic growth ( Figure 6—figure supplement 6 ) and reduced proliferation in the absence of anchorage ( Figure 6—figure supplement 7 ) . 10 . 7554/eLife . 00358 . 021Figure 6 . PVRL4 is amplified in breast cancer and is essential for the transformed phenotype of cancer cells . ( A ) A view from the integrated Genome Viewer program showing focal amplification of the PVRL4 locus in SUM190 cells . The degree of amplification is denoted by the intensity of the color . ( B ) PVRL4 mRNA was stably depleted from SUM190 cells by four independent shRNAs . Transcript levels were measured by RT-qPCR and normalized to β-actin . qPCR was performed in quadruplicate ( error bars ± SD ) . PVRL4-depleted and control cells were assayed for clonogenic survival and anchorage-independent colony formation . Assays were performed in triplicate ( error bars ± SD ) . All values were normalized to the uninfected control sample . ECM: extracellular matrix . ( C ) The PVRL4-CD8 chimeric construct was used to rescue the defect in clonogenic survival observed with RNAi-mediated PVRL4 depletion . Assays were performed in triplicate ( error bars ± SD ) . Colony numbers were normalized to the control shRNA sample . ( D ) Expression levels of endogenous and chimeric proteins were verified by Western blot . DOI: http://dx . doi . org/10 . 7554/eLife . 00358 . 02110 . 7554/eLife . 00358 . 022Figure 6—figure supplement 1 . PVRL4 induces clustering of SUM190 cells which is blocked by antibodies against PVRL4 . SUM190 cells were assayed for cell–cell clustering in the presence of the indicated antibodies . Four-cell clusters and clusters with more than four cells from representative images were scored separately . DOI: http://dx . doi . org/10 . 7554/eLife . 00358 . 02210 . 7554/eLife . 00358 . 023Figure 6—figure supplement 2 . PVRL4 induces clustering of SUM190 cells which is blocked by RNAi against PVRL4 . SUM190 cells were assayed for cell–cell clustering in the presence of the indicated shRNAs . Four-cell clusters and clusters with more than four cells from representative images were scored separately . DOI: http://dx . doi . org/10 . 7554/eLife . 00358 . 02310 . 7554/eLife . 00358 . 024Figure 6—figure supplement 3 . PVRL4 induces attachment of SUM190 cells to microvascular endothelial cells which is blocked by antibodies against PVRL4 . SUM190 cells ( GFP ) were assayed for heterotypic clustering with HMVEC-L cells ( dsRed ) in the presence of the indicated antibodies . Clusters with at least three cells incorporating both green and red cells from representative images were counted . DOI: http://dx . doi . org/10 . 7554/eLife . 00358 . 02410 . 7554/eLife . 00358 . 025Figure 6—figure supplement 4 . PVRL4 induces attachment of SUM190 cells to microvascular endothelial cells which is blocked by RNAi against PVRL4 . SUM190 cells ( GFP ) were assayed for heterotypic clustering with HMVEC-L cells ( dsRed ) in the presence of the indicated shRNAs . Clusters with at least three cells incorporating both green and red cells from representative images were counted . DOI: http://dx . doi . org/10 . 7554/eLife . 00358 . 02510 . 7554/eLife . 00358 . 026Figure 6—figure supplement 5 . Stable depletion of PVRL4 transcript in BT-474 and Sk-BR-3 cell lines . PVRL4 transcript levels were measured by RT-qPCR and normalized to β-actin . qPCR was performed in quadruplicate ( error bars ± SD ) . DOI: http://dx . doi . org/10 . 7554/eLife . 00358 . 02610 . 7554/eLife . 00358 . 027Figure 6—figure supplement 6 . PVRL4 depletion affects clonogenic growth of BT-474 and Sk-BR-3 cell lines . The clonogenic potential of the indicated cell lines in the presence of control or PVRL4 shRNA constructs was assessed . Assays were performed in triplicate ( error bars ± SD ) . Colony numbers were normalized to the control shRNA sample . DOI: http://dx . doi . org/10 . 7554/eLife . 00358 . 02710 . 7554/eLife . 00358 . 028Figure 6—figure supplement 7 . PVRL4 depletion affects anchorage-independent growth of BT-474 and Sk-BR-3 cell lines . Cell line growth in methylcellulose-containing media on an ultra-low attachment surface was assessed in the presence of control or PVRL4 shRNA constructs . Representative images are shown . DOI: http://dx . doi . org/10 . 7554/eLife . 00358 . 028 To confirm that the observed phenotype was specific for PVRL4 depletion , we used a PVRL4-CD8tm construct to rescue the colony formation defect in SUM190 cells . Coexpression of PVRL4-CD8tm in the presence of PVRL4-targeting shRNAs alleviated the clonogenic survival defect , confirming its on-target nature ( Figure 6C , D ) . This result also demonstrates that the extracellular region of PVRL4 alone is sufficient for potentiating clonogenic survival of breast cancer cells , consistent with PVRL4-CD8tm ability to enable anchorage-independent colony formation in TL-HMECs . To determine whether colony growth inhibition induced by PVRL4 depletion in vitro was also relevant for in vivo tumor growth , we stably expressed either control shRNA or an anti-PVRL4 shRNA in a panel of breast cancer cell lines and orthotopically implanted them into the mammary fat pads of nude mice . Depletion of PVRL4 by RNAi from SUM190 ( Figure 7A , B ) , SUM185 ( Figure 7C ) , and BT-474 cells ( Figure 7—figure supplement 1 ) inhibited xenograft growth , verifying the importance of PVRL4 to cancer cell growth in vivo as well as in vitro . 10 . 7554/eLife . 00358 . 029Figure 7 . Targeting PVRL4 inhibits tumor growth . ( A ) and ( B ) Female nude mice were injected into their mammary fat pads with SUM190 cells expressing PVRL4-targeted or control shRNA ( n = 10 per group , error bars ± SEM ) . The resulting tumors were excised , scaled ( A ) , and photographed ( B ) . ( C ) SUM185 cells were stably transduced with PVRL4-targeted or control shRNA and injected into the mammary fat pads of female nude mice ( n = 10 per group , error bars ± SEM ) . Tumor volume was measured with calipers at the indicated time points . ( D ) Female nude mice with ∼50 mm3 SUM190-eGFP xenografts were randomized into two cohorts ( n = 7 per group ) and injected with anti-PVRL4 monoclonal antibodies or control IgG on the indicated days . Tumor volume was measured with calipers ( error bars ± SEM ) . ( E ) Levels of PVRL4 protein were measured in tumor lysates from anti-PVRL4 antibody or control-treated mice , 7 days after the last treatment . ( F ) Tumor sections from control IgG ( A–C ) or anti-PVRL4 antibody-treated ( D–F ) mice were stained with hematoxylin/eosin and photographed . Representative images are shown . DOI: http://dx . doi . org/10 . 7554/eLife . 00358 . 02910 . 7554/eLife . 00358 . 030Figure 7—figure supplement 1 . PVRL4 depletion inhibits BT-474 xenograft growth . BT-474 cells were transduced with the indicated shRNA constructs and injected subcutaneously into female nude mice ( N = 10 per group ) . Slow-release estrogen pellets were implanted into mice 72 hr prior to cell line injection . Tumor growth was measured at the indicated time points ( error bars: SEM ) . DOI: http://dx . doi . org/10 . 7554/eLife . 00358 . 03010 . 7554/eLife . 00358 . 031Figure 7—figure supplement 2 . Anti-PVRL4 antibodies disrupt cell–cell contacts in xenografts in vivo . Freshly explanted tumors from control IgG or anti-PVRL4 antibody-treated mice ( N = 3 per group ) were visualized using a two-photon confocal microscope . Representative images are shown . DOI: http://dx . doi . org/10 . 7554/eLife . 00358 . 03110 . 7554/eLife . 00358 . 032Figure 7—figure supplement 3 . Anti-PVRL4 antibodies do not induce ADCC in vitro . Europium-labeled SUM190 cells were incubated with fresh human NK cells in the presence of an isotype control or anti-PVRL4 antibody , and the degree of lysis was measured by the DELFIA europium assay . The maximum signal was determined by a complete lysis of labeled SUM190 cells in DELFIA lysis buffer . As a positive control , hMB humanized mouse lymphoma cells were mixed with effector cells in the presence of ADCC-competent anti-CD52 antibody . ADCC: antibody-dependent cytotoxicity . DOI: http://dx . doi . org/10 . 7554/eLife . 00358 . 03210 . 7554/eLife . 00358 . 033Figure 7—figure supplement 4 . Anti-PVRL4 antibody treatment does not affect the degree of macrophage infiltration into SUM190 xenografts . Paraffin-embedded sections of SUM190 xenografts from mice treated with either control IgG or anti-PVRL4 antibody were stained with anti-mouse F4/80 antibody . Representative images are shown . DOI: http://dx . doi . org/10 . 7554/eLife . 00358 . 03310 . 7554/eLife . 00358 . 034Figure 7—figure supplement 5 . Inhibition of PVRL4 by antibodies or by RNAi does not affect expression of EMT markers . ( A ) Tumor lysates from control antibody- or anti-PVRL4 antibody-treated mice were blotted for E-cadherin and vimentin . ( B ) PVRL4 was stably depleted by two independent shRNAs in SUM190 cells and lysates were blotted for E-cadherin and vimentin . DOI: http://dx . doi . org/10 . 7554/eLife . 00358 . 03410 . 7554/eLife . 00358 . 035Figure 7—figure supplement 6 . Anti-PVRL4 antibodies recognize both human and mouse epitopes . 293T cells were transfected with empty pQCXIN ( green line ) , pQCXIN-human PVRL4 ( blue line ) , or pCMV-SPORT6-mouse PVRL4 ( magenta line ) . Live-cell FACS was performed with mouse anti-human PVRL4 antibody followed by anti-mouse secondary antibody conjugated to Alexa Fluor 488 fluorophore . The FITC-A fluorescent signal for three labeled cell populations is shown . DOI: http://dx . doi . org/10 . 7554/eLife . 00358 . 035 Having previously shown that blocking PVRL4-driven cell-to-cell attachment with a monoclonal antibody suppresses SFK activation and anchorage-independent growth of TL-HMECs in vitro , we sought to determine whether this antibody would similarly inhibit SUM190 xenograft growth in an in vivo setting . To address this , we treated mice bearing ∼50 mm3 SUM190-eGFP xenografts with four consecutive intravenous injections of either anti-PVRL4 monoclonal antibody or isotype control IgG at 15 mg/kg each . After the last treatment , we continued to monitor tumor growth for the next 7 days . Whereas tumor volumes in the control group steadily increased over time , the PVRL4 antibody-treated group displayed a remarkable stalling of tumor growth throughout the course of the injection regimen ( Figure 7D ) . Immunoblotting of tumor lysates revealed that PVRL4 antibody treatment resulted in a precipitous decline of PVRL4 protein levels , suggesting that cells with the highest expression of PVRL4 were eliminated by the treatment ( Figure 7E ) . Dissection of excised xenografts revealed a softer , paste-like composition of PVRL4 antibody-treated tumors compared to the more solid consistency of control-treated samples . This latter observation was corroborated by histological analysis ( Figure 7F ) , which revealed areas of widespread necrosis and , importantly , loss of cell contacts ( Figure 7D–F ) when compared to control IgG-treated tumors ( Figure 7A–C ) . Finally , examination of three-dimensional tumor architecture in control- and anti-PVRL4 antibody-treated tumors by two-photon confocal microscopy revealed reduced cell–cell contacts in anti-PVRL4 antibody-treated samples as compared to controls ( Figure 7—figure supplement 2 ) . The loss of cell contacts and necrosis observed in treated tumors supports our model of PVRL4-driven cell-to-cell attachments conferring ability to sustain growth in conditions of matrix anchorage loss . To test whether an in vivo tumor inhibitory effect of an anti-PVRL4 antibody can be explained by antibody-mediated recruitment of components of innate immunity , we asked whether this antibody was capable of inducing ADCC ( antibody-dependent cytotoxicity ) in vitro . Specifically , we mixed SUM190 cells with fresh human NK cells and measured the relative degree of cell lysis induced by either anti-PVRL4 antibody or control IgG . No increase in cell lysis was observed with anti-PVRL4 antibody over isotype control-incubated cells , demonstrating that anti-PVRL4 antibody was unable to efficiently recruit Fc receptor-containing cells ( Figure 7—figure supplement 3 ) . Moreover , immunohistochemical staining of control and anti-PVRL4 IgG-treated tumor sections for a mouse macrophage-specific marker F4/80 revealed no differences in the extent of macrophage infiltration ( Figure 7—figure supplement 4 ) . Taken together , these data suggest that the observed inhibitory effect is likely to be a consequence of a direct blockade of PVRL4 function . Taken together , these data suggest that targeting PVRL4-driven tumors with a monoclonal antibody directed towards its extracellular region results in dramatic inhibition of growth and disruption of cell–cell contacts , demonstrating the therapeutic efficacy of such approach . One caveat associated with targeting junctional proteins as an anticancer therapy strategy is that it may inadvertently induce an EMT phenotype and instigate metastasis . To address this , we measured levels of E-cadherin ( an epithelial marker ) and vimentin ( a mesenchymal marker ) in SUM190 xenografts treated with anti-PVRL4 antibody in vivo ( Figure 7—figure supplement 5A ) as well as in cultured SUM190 cells in which PVRL4 expression was inhibited by RNAi ( Figure 7—figure supplement 5B ) . In both settings E-cadherin and vimentin protein levels remained unchanged , demonstrating that inhibiting PVRL4 in SUM190 cells does not cause EMT; neither does it select for cells with EMT-like characteristics in vivo . Another safety concern associated with targeting PVRL4 with a monoclonal antibody is that such treatment has the potential to induce damaging effects in tissues normally expressing the antigen . Both human and mouse PVRL4 were strongly recognized by anti-PVRL4 antibody when expressed on the cell surface ( Figure 7—figure supplement 6 ) . Among normal mouse tissues , PVRL4 expression is strongest in cornea and skin epidermis ( Wu et al . , 2009 ) . The skin surface of mice treated with anti-PVRL4 antibodies was not visibly affected by treatment , and mice appeared healthy overall , demonstrating that targeted therapy against PVRL4 does not induce acute side effects in this organism . Taken together with our findings that functionally link PVRL4 to tumorigenesis , as well as with its widespread expression in multiple tumor types , targeting PVRL4-driven cell-to-cell attachment can be viewed as a potential therapeutic strategy directed against a broad spectrum of tumor types . In this study we have performed an unbiased genetic screen to identify genes that facilitate cell growth in the absence of anchorage . The screen identified PVRL4/Nectin-4 as a potent inducer of anchorage-independent growth in a manner that relies upon formation of physical contacts between individual cells . PVRL4 is expressed in a limited set of normal tissues—namely , skin epidermis , hair follicles , placenta , trachea , and lung ( Jelani et al . , 2011 ) —yet it becomes overexpressed in a large fraction of breast ( Fabre-Lafay et al . , 2007 ) , NSCLC ( Takano et al . , 2009 ) , and ovarian ( Derycke et al . , 2010 ) tumors . Analysis of a publicly available dataset of copy number alterations involving PVRL4 in 484 breast tumors indicates a low-level copy number gain in 60 . 3% of samples , and high-level amplification in another 7 . 9% of samples ( Cerami et al . , 2012 ) . Moreover , PVRL4 has emerged as a strong predictor of poor postoperative survival of patients with breast and lung cancer ( Takano et al . , 2009; Athanassiadou et al . , 2011 ) . We show that PVRL4-driven colony formation is carried out exclusively via the extracellular regions of PVRL4 and its interacting partner PVRL1 , while transmembrane and cytoplasmic regions are not required for this phenotype . Cytoplasmic regions of nectins recruit afadin , a large scaffold protein involved in mediating intracellular signaling . Interestingly , and in contrast to PVRL4 , loss of afadin has been shown to be strongly associated with poor outcome in breast cancer patients in a number of studies ( Letessier et al . , 2007; Fournier et al . , 2011 ) . Supporting our findings in TL-HMECs , we found that PVRL4 is required for clonogenic and anchorage-independent growth of breast cancer cells in vitro as well as growth of orthotopically implanted tumors in vivo . In lung cancer cells , siRNA-mediated depletion of PVRL4 was shown to negatively affect adherent growth and motility ( Takano et al . , 2009 ) , highlighting the functional role of PVRL4 across multiple cancer types . Importantly , we were able to reverse the clonogenic defect by expressing a chimeric construct consisting of the extracellular region of PVRL4 fused to the transmembrane domain of CD8 , independently verifying that its function is carried out via an extracellular route . Our findings reveal a potential mechanism of PVRL4-driven anchorage-independent growth . Specifically , we identify integrin β4 as a cis-interacting partner of PVRL4 and demonstrate that anchorage-independent growth driven by PVRL4 requires intact integrin β4/SHP-2/SFK signaling . An extensive array of evidence functionally links integrin β4 to breast tumorigenesis . Similarly to PVRL4 , integrin β4 expression is associated with poor prognosis and a basal-like expression profile ( Diaz et al . , 2005; Lu et al . , 2008 ) , and targeting integrin β4 with RNAi was shown to inhibit anchorage-independent growth , invasion , and xenograft growth in breast cancer cell lines ( Lipscomb et al . , 2003; Bon et al . , 2006 ) . Integrin β4 is thought to exhibit a significant degree of signaling autonomy compared to other integrins , due to its atypically long C-terminus which serves as a scaffold for its downstream effectors . This signaling autonomy was convincingly demonstrated in a study utilizing chimeric constructs where extracellular and transmembrane regions of integrin β4 were substituted with those on the TrkB receptor tyrosine kinase . Dimerization of two such chimeric molecules on the cell surface by adding TrkB ligand , BDNF , was sufficient to drive SFK activation , in a manner dependent on SHP-2 phosphatase ( Merdek et al . , 2007 ) . Trans-interactions of cell adhesion molecules are thought to trigger the formation of tightly packed zippers of interacting cell adhesion molecules at a cell–cell contact interface ( Walmod et al . , 2004 ) . One possibility is that trans-interacting PVRL4 and PVRL1 similarly organize into zippers at sites of cell–cell contact , driving individual integrin β4 molecules into proximity sufficient for their activation . Matrix attachment-independent activation of integrin molecules may then allow sustained SHP-2/SFK signaling sufficient for counteracting the growth-restrictive consequences of anchorage loss . In the context of a multicellular tumor , the PVRL4-PVRL1 associated signaling exemplifies a mechanism by which cell-to-cell attachment serves to mimic attachment to matrix , allowing cells to bypass the growth constraint imposed by the requirement for proper anchorage . Interestingly , we found that multiple oncogenic perturbations facilitate increased intercellular adhesiveness . It remains to be further elucidated whether , and through what mechanisms , the increased ability to form contacts functionally contributes to anchorage-independent growth or other tumorigenesis-associated phenotypes in each of these instances . Increased clustering has been previously associated with greater metastatic capacity ( Updyke and Nicolson , 1986 ) , raising an interesting possibility that increased cell–cell adhesiveness is a generalized mechanism that tumor cells employ as a part of their survival strategy . It is possible that while loss of cell adhesion during EMT facilitates invasiveness , it concomitantly places cells under the stress of anchorage deprivation , compromising their survival . Thus , cancer cells may need to replace the loss of homotypic cell–cell contacts with weaker heterotypic interactions capable of promoting survival in the absence of anchorage . In addition to enabling anchorage-independent growth at the primary tumor site , cell–cell clustering in the lymphatic vasculature could promote cell survival . This is particularly relevant for PVRL4 as 100% of breast tumors from patients with two or more lymph nodes that are positive for cancer cells express PVRL4 ( Athanassiadou et al . , 2011 ) . In addition , we demonstrate that PVRL4 can promote the attachment of cancer cells to other cell types . For example , SUM190 breast cancer cells readily attach to lung endothelial cells ( HMVEC-Ls ) , and inhibiting PVRL4 with RNAi or with blocking antibodies abolishes this interaction . CTC-targeted therapy is a promising new treatment possibility , and it will be of particular interest to investigate the potential role of PVRL4 in mediating these types of interactions in vivo . We directly tested the utility of blocking PVRL4-driven cell–cell contacts with monoclonal antibodies as a potential therapeutic strategy . In vitro , we found that such blocking antibodies suppress PVRL4-driven cellular growth and Src family kinase activation in the absence of anchorage , and in vivo , anti-PVRL4 antibody treatment potently inhibited the growth of orthotopically implanted primary tumors . Importantly , post-antibody therapy tumor samples uniformly exhibit marked downregulation of PVRL4 , confirming that the observed response is target-specific and not a consequence of a non-specific antitumor effect of the antibody . In addition , we found no difference in the degree of host macrophage recruitment or in vitro ADCC induction between control IgG and anti-PVRL4 antibodies . These observations suggest that the observed tumor growth inhibition is likely due to a direct inhibition of cell-to-cell attachment mediated by PVRL4-PVRL1 binding , and not due to an Fc receptor-mediated immune response . It remains to be tested whether a stronger inhibitory effect could be achieved with an antibody that is both ADCC-competent and capable of inhibiting PVRL4-driven cell clustering . Even though treatment with anti-PVRL4 antibodies caused dramatic dissolution of cell–cell contacts , we observed no changes in markers of EMT in tumors from mice treated with the antibody; neither did it produce obvious deleterious effects in treated mice . These observations signify not only the potential efficacy but also the safety of the demonstrated approach . Together , these results suggest that anti-PVRL4 monoclonal antibody therapy aimed at disrupting cell–cell interactions may be a viable strategy for treating cancer . For screen candidate validation , ORFs from isolated colonies were subcloned into pRoles retroviral vector ( provided by W Harper ) . For PVRL4 structure-function analysis , a full-length ORF ( accession number BC010423 , aa 1-510 ) or indicated fragments ( Supplementary file 1C ) were generated by PCR and subcloned into pQCXIN retroviral vector ( Clontech , Mountain View , CA ) . PVRL4-CD8 and PVRL1-CD8 chimeric fusions were created the following way: the extracellular region of PVRL4 ( accession number BC010423 , aa 1-342 ) or PVRL1 ( accession number BC113471 , aa 1-354 ) was amplified by PCR using a reverse primer containing a sequence for a 28 aa-long transmembrane domain of the CD8A gene ( accession number NM_001768 , 543–626 bp ) , followed by a STOP codon . The resulting PCR product was subcloned into the pQCXIN vector . To create an shRNA-resistant PVRL1-CD8 construct , site-directed mutagenesis was performed with the following primers: sh3ResFW 5′-CCAGGCGTCCACAGTCAAGTTGTGCAAGTCAATGACTCCATGTATG-3′ and sh3ResRV 5′-CATACATGGAGTCATTGACTTGCACAACTTGACTGTGGACGCCTGG-3′ using a QuikChange II Site-Directed Mutagenesis Kit ( Agilent , Santa Clara , CA ) . Stable RNAi-mediated depletion was performed with shRNAs expressed in either pMSCV-PM or pGIPZ vector in a miR-30 context that were either picked from the Hannon-Elledge shRNA library ( Open Biosystems , Huntsville , AL ) or designed de novo ( design and cloning protocol described in Paddison et al . , 2004 ) . The 21 nt sense sequences of shRNAs used in this study are listed in Supplementary file 1D . Tandem HA/FLAG-tagged PVRL4 construct was obtained by Gateway recombination of PVRL4 entry clone into C-terminal iTAP vector ( provided by W Harper ) . For stable labeling with fluorescent markers , cells were transduced with pHAGE-dsRed and pMSCV-CMV-GFP viruses . For assaying cell clustering in the presence of oncogenes , TL-HMECs were transduced with pWZL-myr-p110-PI3K-neo or empty pWZL-neo , and pBABE-H-RASV12-puro or empty pBABE-puro . Mouse pSPORT6-PVRL4 was purchased from Open Biosystems . Retro- and lentiviral supernatants were generated by transient transfection of 293T cells following the TransIT transfection protocol ( Mirus Bio , Madison , WI ) and harvested 48 hr later . TL-HMECs expressing hTERT and SV40 Large T antigen ( Zhao et al . , 2003 ) were cultured in MEGM ( Lonza , Allendale , NJ ) . SUM190 and SUM185 cells ( provided by K Polyak ) were cultured in a 1:1 mix of MEGM and F12:DMEM ( Invitrogen , Carlsbad , CA ) , supplemented with 5% FBS ( Invitrogen ) . BT-474 and Sk-BR-3 cells were cultured in RPMI-1640 ( ATCC , Manassas , VA ) with 10% FBS ( Invitrogen ) and 10 μg/ml of bovine insulin ( Sigma , St . Louis , MO ) . 293T cells were cultured in DMEM ( Invitrogen ) , supplemented with 10% FBS . HMVEC-L cells ( Lonza ) were cultured in EGM-2 medium ( Lonza ) . Retroviral infections were performed in the presence of 8 μg/ml of polybrene ( Sigma ) . For SUM190 and SUM185 infections , cells were plated in six-well dishes and centrifuged in the presence of viral supernatant and polybrene for 1 hr at 2000 rpm . Successful viral integrants were selected with puromycin ( 2 μg/ml ) or Geneticin ( 200 μg/ml for TL-HMECs , 750 μg/ml for SUM190 ) . Anchorage-independent colony formation assays were performed as previously described ( Westbrook et al . , 2005 ) with minor modifications . Briefly , cells were suspended in reduced growth factor MEGM ( containing 50% of kit-supplied BPA , insulin , EGF , and hydrocortisone ) with 2% methylcellulose ( Sigma ) and plated on tissue culture dishes precoated with 0 . 6% Noble agar ( Sigma ) in MEM ( Invitrogen ) . For assays performed in 6 cm dishes , 4 . 5 × 104 cells per dish were plated . For assays performed in six-well plates , 1 . 2 × 104 cells per well were plated . Colonies were counted after 3 weeks of growth . For each assay , an average of three replicates ± SD is shown . For anchorage-independent colony formation assays in the presence of antibodies , the following antibodies were used: normal mouse IgG ( MAB004; R&D Systems , Minneapolis , MN ) and mouse anti-human PVRL4 IgG2B ( MAB2659; R&D Systems ) at 4 μg/ml . For anchorage-independent colony formation assays in the presence of PP2 inhibitor , PP2 ( EMD Millipore , Billerica , MA ) was used at 10 μM final concentration . For filtering of the cell suspension , cells were passed through a nylon mesh 35 μm cell strainer ( BD Biosciences , Franklin Lakes , NJ ) . For anoikis assays , cells were seeded on ultra-low attachment dishes ( Corning , Midland , MI ) in reduced growth factor MEGM with 1% methylcellulose . For assays performed in 10 cm dishes , 1 . 4 × 105 cells were plated; for assays performed in six-well dishes , 2 . 0 × 104 cells per well were plated . Total ATP measurements were performed using CellTiter GLO reagent ( Promega , Madison , WI ) according to the manufacturer’s protocol after 72 hr of growth in suspension , and luminescence values were read with a Victor X5 plate reader . For isolation of RNA or protein lysates , cell pellets were harvested after 72 hr of growth in suspension and washed with cold PBS prior to lysis . A genetic screen for ORFs promoting TL-HMEC anchorage-independent colony formation was performed the following way: TL-HMECs were transduced with a retroviral pool of human open reading frames ( ORFeome library V1 . 1 ) at a multiplicity of infection of 0 . 2 and representation of 200 . Two independent infections with a complete library were performed . Cells were plated into semi-solid medium ( methylcellulose ) in the absence of attachment , maintaining the library representation . Resulting colonies were isolated after 3 weeks of growth . A total of 732 colonies were isolated from two screen replicates and individually expanded in 96-well plates . Genomic DNA was isolated from clones and the ORF insert was PCR-amplified and sequenced . A total of 40 candidates ( Supplementary file 1A ) were subsequently subcloned into pRoles and individually tested for their ability to induce colony formation in anchorage-free conditions . For assaying clonogenic potential , 1 . 0 × 103 SUM190 cells were seeded in 6 cm tissue culture-treated dishes . After 3 wk of growth , the resulting colonies were stained with 1% methylene blue and counted . For each assay , an average of three replicates ± SD is shown . Cells were gently detached off the tissue culture surface with enzyme-free cell dissociation buffer ( Invitrogen ) and washed once with complete medium . Then 1 . 0 × 105 cells were allowed to aggregate in 1 ml of complete medium in a 15 ml conical tube at room temperature . The tubes were gently flicked during the process to visually assess the progression of clustering . After 1–1 . 5 hr of incubation , the cell suspension was poured into the wells of 12-well dishes and allowed to attach to the bottom of the dish for 5–10 min . Cells were promptly visualized and photographed under phase contrast and fluorescent filters using an AxioVert inverted microscope . When clustering assays were performed in the presence of antibodies , the following antibodies were used: normal mouse IgG ( MAB004; R&D Systems ) , normal goat IgG ( AB-108-1C; R&D Systems ) , mouse anti-human PVRL4 IgG2A ( MAB26591; R&D Systems ) , mouse anti-human PVRL4 IgG2B ( MAB2659; R&D Systems ) , goat anti-human PVRL4 ( AF2659; R&D Systems ) , goat anti-human PVRL1 ( AF2880; R&D Systems ) , and DECMA-1 ( anti-human E-cadherin , ab11512; Abcam , Eugene , OR ) . All antibodies were used at a concentration of 4 μg/ml . For short-term culture of TL-HMECs in suspension , 4 . 0 × 105 cells were allowed to aggregate in 1 ml of complete medium in a 15 ml conical tube at room temperature , mixed with 5 ml of 0 . 5% methylcellulose in reduced growth factor MEGM and incubated in the wells of a six-well ultra-low attachment dish ( two wells per sample ) for indicated periods of time . Cells were lysed in NP-40 buffer ( 1% NP-40 , 25 mM Tris–HCl , pH 7 . 4 , 150 mM NaCl , 1 mM EDTA , 10% glycerol ) in the presence of protease and phosphatase inhibitor tablets ( Roche , Indianapolis , IN ) . Adherent cells were lysed for 15 min on ice , followed by scraping into Eppendorf tubes and centrifugation at 14 , 000 rpm for 15 min at 4°C . Suspension-cultured cells were washed once with cold PBS , followed by lysis and centrifugation . The protein concentration in supernatants was measured using the BCA assay ( Pierce , Rockford , IL ) and lysates were brought to identical concentrations with lysis buffer . Samples were mixed 1:1 with 2× Laemmli buffer ( 125 mM Tris–HCl , pH 6 . 8 , 4% SDS , 20% glycerol , 0 . 004% bromophenol blue ) and DTT was added to final concentration of 25 mM . Samples were boiled for 8 min and loaded on Tris-glycine 4–20% or 4–12% gradient gels ( Invitrogen ) . Transfer/blotting was performed as described elsewhere . Western blotting was performed with the following antibodies: goat anti-PVRL4 ( AF2659; R&D Systems ) , goat anti-PVRL1 ( AF2880; R&D Systems ) , mouse anti-Vinculin ( V9131; Sigma ) , mouse anti-RAN ( 610340; BD Biosciences ) , mouse anti-ITGB4 ( 611232; BD Biosciences ) , mouse anti-FLAG-HRP ( A8592; Sigma-Aldrich , St . Louis , MO ) , rabbit anti-pY416 SFK ( 2101; Cell Signaling , Danvers , MA ) , mouse anti-SFK ( 2110; Cell Signaling ) , mouse anti-SHP-2 ( 610621; BD Biosciences ) , and mouse anti-α Tubulin ( sc-8035; Santa Cruz , Santa Cruz , CA ) . TL-HMECs expressing HA/FLAG-PVRL4 or HA/FLAG-GFP were gently detached off the tissue culture surface with enzyme-free cell dissociation buffer , washed , and lysed for 30 min in MCLB lysis buffer ( 50 mM Tris , pH 7 . 5 , 150 mM NaCl , 1% NP-40 ) in the presence of protease and phosphatase inhibitors ( Roche ) , followed by centrifugation at 14 , 000 rpm at 4°C . Lysates were precleared with protein A/G beads ( sc-2003; Santa Cruz ) . Pulldowns were performed with anti-HA beads ( A2095; Sigma-Aldrich ) overnight at 4°C . Beads were washed five times with MCLB buffer , followed by two washes with elution buffer ( 50 mM Tris , pH 7 . 5 , 150 mM NaCl , 10% glycerol ) . Elutions were performed with 500 μg/ml of HA peptide ( I2149; Sigma-Aldrich ) in elution buffer . Proteins were precipitated from this mixture with 20% trichloroacetic acid ( TCA ) , and the resulting pellet was washed once with 10% TCA and four times with cold acetone . TCA-precipitated proteins were dissolved in 100 mM ammonium bicarbonate ( pH 8 . 0 ) with 10% acetonitrile and 10 ng/µl trypsin ( Promega ) and incubated at 37°C for 5 hr . They were subsequently desalted , dissolved in 5% formic acid/5% acetonitrile , and loaded onto a reversed phase microcapillary column ( 100 mm I . D . ) packed with 18 cm of Maccel C18AQ resin ( 3 mm , 200 A; The Nest Group , Southborough , MA ) . Peptides were eluted using a gradient of 4–26% acetonitrile in 0 . 125% formic acid over 95 min and detected in a hybrid linear ion trap-orbitrap mass spectrometer ( LTQ-Orbitrap Discovery; ThermoFisher , West Palm Beach , FL ) . Precursors selected for MS/MS fragmentation were corrected for errors in monoisotopic peak assignment , and tandem MS spectra were searched using the Sequest algorithm , with mass tolerance set to 25 ppm and two missed cleavages allowed . False discovery rates were estimated with the target-decoy method ( Elias and Gygi , 2007 ) , and linear discriminant analysis ( LDA ) was utilized to filter peptides to an initial 1% peptide-level false discovery rate ( FDR ) . Peptides were then assembled into proteins and further filtered to a protein-level FDR of 0 . 84% ( Huttlin et al . , 2010 ) , resulting in a final peptide-level FDR of 0 . 35% . Cells were lysed in IP lysis buffer ( 20 mM Tris–HCl , pH 8 . 0 , 150 mM NaCl , 1 mM CaCl2 , 1 mM MgCl2 , 10% glycerol , 1% NP-40 ) in the presence of protease and phosphatase inhibitors and centrifuged at 14 , 000 rpm at 4°C . For the immunoprecipitation of integrin β4 , cells were pretreated with a reversible , cell-impermeable homobifunctional cross-linking agent DTSSP ( 21 , 578; Pierce ) at 1 mM concentration for 45 min on ice . The cross-linking agent was then quenched with 20 mM glycine for 15 min and cells were washed twice with PBS , followed by lysis in IP lysis buffer . Immunoprecipitations were performed with anti-HA beads ( A2095; Sigma-Aldrich ) for 2 hr or with anti-ITGB4 antibody ( 555722; BD Biosciences ) for 2 hr , followed by protein A/G beads ( sc-2003; Santa Cruz ) for 1 hr . Beads were washed four times with lysis buffer and boiled in Laemmli buffer with 25 mM DTT . Total RNA was isolated from cells using the RNAeasy Plus kit ( QIAgen , Germantown , MD ) . cDNA was synthesized from 1 μg of total RNA , using Superscript III Reverse Transcriptase ( Invitrogen ) and Oligo-dT primer ( Invitrogen ) , following the manufacturer’s protocol . Quantitative PCR was performed with the Platinum SYBR Green qPCR Supermix-UDG kit ( Invitrogen ) on an Applied Biosystems 7500 Fast Real Time PCR machine . Gene-specific primers were designed using the Universal Probe Library ( Roche Applied Science , Penzberg , Germany ) . PCR reactions were carried out in triplicates or quadruplicates . For each value , an average of at least three replicates ± SD is shown . Sequences of gene-specific primers used for qPCR are listed in Supplementary file 1E . TL-HMEC cells were cultured on chamber slides ( BD Biosciences ) and fixed with cold methanol following by blocking in blocking buffer containing BSA and cold water fish gelatin ( Sigma-Aldrich ) . Cells were incubated with primary goat anti-PVRL4 antibody ( AF2659; R&D Systems ) , at 1:500 dilution at 4°C overnight , followed by 1-hr incubation with chicken anti-goat Alexa-Fluor 488 secondary antibody ( Invitrogen ) at 1:2500 dilution . Cells were mounted in Vectashield mounting medium with DAPI ( Vector Labs , Burlingame , CA ) and visualized under a fluorescent microscope . 293T cells were gently detached off the adherent surface with enzyme-free cell dissociation buffer ( Invitrogen ) and washed once with serum-free DMEM . Cells were incubated with 10% normal goat serum ( Invitrogen ) for 10 min at room temperature , then with a primary antibody at 1:100 dilution in 5% goat serum for 30 min on ice , followed by a goat anti-mouse secondary antibody coupled to Alexa Fluor 488 ( Invitrogen ) at 1:1000 dilution in PBS for 30 min on ice . The fluorescent signal was measured on an LSR II FACS Analyzer and analyzed with FlowJo software . All mouse experiments were performed with the approval of the Massachusetts Institute of Technology ( MIT ) Committee on Animal Care . For subcutaneous xenograft assays with shRNA-transduced cells , nude mice ( Taconic ) were injected orthotopically into the mammary fat pad with 1 . 0 × 106 SUM190 or SUM185 cells , or subcutaneously with 1 . 0 × 107 BT-474 cells in serum-free medium with 50% Matrigel ( BD Biosciences ) , one injection site per mouse . For the BT-474 xenograft experiment , mice were implanted with a 60 d release pellet containing 0 . 72 mg of 17β-estradiol ( Innovative Research of America , Sarasota , FL ) 72 hr prior to cell injections . Tumor growth was monitored by caliper measurements . Tumors were excised after 4 wk of growth , scaled , and photographed . For the antibody treatment experiment , 15 mg/kg of anti-PVRL4 antibody or normal IgG control ( R&D Systems ) was injected 4 times at days 0 , 1 , 4 , and 7 . Treatment response was assessed by caliper measurements of the tumors in anesthetized mice . Mice were euthanized 15 d after treatment initiation and tumors were harvested and representative parts subjected to paraffin embedding , Western blot analysis , and direct microscopy on an inverted Olympus Multiphoton Laser Scanning Confocal Microscope using a 25× objective . Images were analyzed by the IMARIS software package . PVRL4 antibody-dependent cell-mediated cytotoxicity was assessed by using europium-labeled ( DELFIA cell cytotoxicity assay; PerkinElmer , Waltham , MA ) SUM190 cells as target cells in a 96-well format . Primary human NK cells were isolated from a fresh buffy coat using magnetic purification ( EasySep Human CD56 Positive Selection Kit; STEMCELL Technologies ) . Cells were co-incubated at a 1:1 , 1:10 , and 10:1 effector-to-target cell ratio . Next , 25 µg/ml anti-PVRL4 antibody was added to the respective wells . Cells were co-incubated for 4 hr and cell-free supernatants further subjected to time-lapsed fluorescence quantification using a TECAN Infinite 200 PRO plate reader . To determine the maximal range of europium release , target cells were lysed in DELFIA lysis buffer . Levels of released europium are displayed as percentage values of maximum release determined by lysis . To control for NK cell effector functionality , the ADCC assay was performed using the hMB lymphoma model ( Leskov et al . , 2012 ) as a target cell line in the presence of 25 µg/ml of anti-CD52 antibody ( alemtuzumab ) . Immunohistochemical staining with an anti-mouse F4/80 antibody ( MCA497GA; Serotec , Raleigh , NC ) was performed using an Envision kit ( Dako , Glostrup , Denmark ) . Retrieval was performed by incubating slides in Proteinase K ( Dako ) at a 1:5 dilution for 10 min . The primary antibody was used at a 1:10 , 000 dilution for 60 min at room temperature .	Epithelial tissue is one of the four major types of tissue found in animals , and is the only type of tissue that is able to form and maintain layers of cells that are just one cell thick . These layers provide inner linings to various cavities and hollow organs throughout the body—including the lungs and glandular organs such as mammary glands . A single-cell layer of epithelium is separated from the tissues beneath it by a supporting substance called the extracellular matrix . The individual cells within a single-cell layer are physically attached to the matrix , and when displaced from it , they promptly undergo programmed cell death . This mechanism preserves the single-cell layer pattern throughout the body and prevents epithelial cells from growing in inappropriate locations . It is estimated that up to 90% of cancers in humans originate in epithelial tissue , and the cells within such tumors are known to survive and divide even when they are no longer attached to the extracellular matrix . Understanding how cancerous cells gain this ability may lead to new approaches to stopping tumor cells from dividing and colonizing tissues around the body . To address this problem , Pavlova et al . explored which genes enable epithelial cells from the human mammary gland to grow without being attached to the extracellular matrix . They found that the gene that codes for a protein called poliovirus receptor-like 4 ( PVRL4 ) allows attachment-free cell growth and also makes cells cluster together once detached from the matrix . Normally , the PVRL4 gene is not active in breast epithelial cells , but its activity is detected in many breast , lung , and ovarian tumors . Moreover , cancerous cells tend to cluster together when they are detached from the extracellular matrix . This behavior is particularly evident in the cells that divide aggressively to form tumors that subsequently migrate and colonize other tissues around the body . When Pavlova et al . used genetic techniques to silence PVRL4 in cells from breast tumors , they found that it reduced the formation of clusters by the cancer cells and also reduced their ability to grow in the absence of attachment . Pavlova et al . also showed that interactions between the PVRL4 in one cell and a related protein called PVRL1 in a neighboring cell were responsible for holding the cells together in clusters . Moreover , PVRL4 triggers a form of signaling between the cells called integrin β4 signaling that allows them to survive without being anchored to the extracellular matrix . Finally , Pavlova et al . found that injecting anti-PVRL4 antibodies ( mouse proteins that attach to PVRL4 and prevent the formation of clusters ) slows down the growth of breast tumors in mice . These findings suggest that inhibiting PVRL4 action with antibodies can be used as a new approach to the treatment of breast , lung , and ovarian cancers in humans .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "cell", "biology", "cancer", "biology" ]	2013	A role for PVRL4-driven cell–cell interactions in tumorigenesis
Despite being pervasive , the control of programmed grooming is poorly understood . We addressed this gap by developing a high-throughput platform that allows long-term detection of grooming in Drosophila melanogaster . In our method , a k-nearest neighbors algorithm automatically classifies fly behavior and finds grooming events with over 90% accuracy in diverse genotypes . Our data show that flies spend ~13% of their waking time grooming , driven largely by two major internal programs . One of these programs regulates the timing of grooming and involves the core circadian clock components cycle , clock , and period . The second program regulates the duration of grooming and , while dependent on cycle and clock , appears to be independent of period . This emerging dual control model in which one program controls timing and another controls duration , resembles the two-process regulatory model of sleep . Together , our quantitative approach presents the opportunity for further dissection of mechanisms controlling long-term grooming in Drosophila . Grooming is broadly defined as a class of behaviors directed at the external surface of the body . Most animals spend considerable time grooming ( Mooring et al . , 2004; Sachs , 1988 ) , and this near universality suggests that grooming likely fulfills an essential role for animals ( Spruijt et al . , 1992 ) . Grooming assumes a variety of forms in different species—for instance , birds preen the oily substance produced by the preening gland from their feathers and skin , cats and dogs lick their fur , and flies sweep their body parts with their legs . Although in most cases the primary function of grooming is to maintain a clean body surface , different species-specific forms of grooming have roles in diverse functions such as thermoregulation , communication and social relationships ( Dawkins and Dawkins , 1976; Ferkin et al . , 2001; Geist and Walther , 1974; McKenna , 1978; Patenaude and Bovet , 1984; Schino , 2001; Schino et al . , 1988; Seyfarth , 1977; Spruijt et al . , 1992; Thiessen et al . , 1977; Walther , 1984 ) . Many animal behaviors , such as locomotion , have been shown to be controlled by both external stimuli ( stimulated behavior ) and internal programs ( programmed behavior ) . An example of stimulated locomotor activity is the abrupt evasive response triggered by the sudden appearance of a predator . In contrast , programmed locomotor activities , such as daily foraging for food , are essential to maintain vital functions of the organism ( Bergman et al . , 2000 ) . Similar to locomotion , limited data from mammals suggest that grooming may be controlled by both external stimuli and internal programs ( Hart et al . , 1992; Hawlena et al . , 2008; Mooring and Samuel , 1998 ) . For example , stimulated grooming might be performed when the animal is excessively dirty or itchy , and programmed grooming might be performed as a social ritual . Although grooming is a widely observed behavior , the basic mechanisms regulating grooming are still not well understood . The fruit fly Drosophila melanogaster is an ideal model organism with which to dissect the fundamental mechanisms of grooming and its relationship to other behaviors . The fly is known to be a frequent groomer with a rich repertoire of behaviors and a sophisticated genetic toolkit developed to study them ( Connolly , 1968; Owald et al . , 2015 ) . The study of Drosophila grooming can be traced back to the 1960s ( Connolly , 1968; Szebenyi , 1969 ) , and notable progress has since been made in studying grooming stimulated by the application of dust particles to the insect exterior ( Hampel et al . , 2015; Seeds et al . , 2014 ) . While most grooming studies thus far have focused on stimulated grooming , understanding the mechanisms responsible for programmed grooming will not only identify components distinct to each type of grooming but also inform us about how programmed grooming is prioritized with regard to other programmed behaviors such as locomotion , feeding , and sleep in the same organism . A major hurdle in detecting programmed grooming in Drosophila is the lack of practical methodology . In many cases , fly grooming events are extracted by eye ( King et al . , 2016; Phillis et al . , 1993; Yanagawa et al . , 2014 ) . Consequently , these data report only conspicuous behaviors within relatively short durations of observation . To improve resolution and accuracy , a number of sophisticated video-tracking methods have been recently developed for fly behavior ( Kain et al . , 2013; Mendes et al . , 2013 ) . These designs are not amenable to easy scale-up for tracking multiple individuals simultaneously . Moreover , while several of these methods are sufficient for short-term monitoring ( Branson et al . , 2009; Kabra et al . , 2013 ) , continuous multi-hour measurements and rapid , automated quantification methods are required to dissect long-term , unstimulated fly grooming relative to other daily behaviors like locomotion and sleep . To overcome limitations of currently available methods , we developed a new platform for long-term video-tracking and automated analysis of fly grooming . The layout of our hardware takes advantage of a basic design for housing individual flies that is widely used in locomotion and sleep studies ( Gilestro , 2012; Pfeiffenberger et al . , 2010; Zimmerman et al . , 2008 ) . Here , we incorporate this standardized hardware into studies of grooming . Our algorithm maps fly activity onto a three-dimensional behavioral space and utilizes k-nearest neighbors ( kNN ) method , a machine learning technique , to classify each video frame as grooming , locomotion or rest . Results from multi-day recordings reveal that Drosophila spend approximately 13% of their waking time grooming , and the temporal pattern of grooming behavior is tightly regulated by the fly’s internal circadian pacemaker . These findings suggest that grooming , similar to feeding and rest , likely serves one or more critical functions in Drosophila . Additionally , genetic perturbations reveal that the transcription factors CYCLE and CLOCK are critical parts of an internal program that controls the amount of Drosophila grooming . These grooming data , the easily implementable hardware , and the automated analysis package together permit the construction of high-resolution ethograms of stereotypical fly behavior over the circadian time-scale . We used a custom-designed video set-up to monitor fly behavior . Within the set-up , insects were placed individually in cylindrical glass tubes 6 cm long and 5 mm wide with food and cotton at opposite ends ( Figure 1A ) . Tubes were placed in a chamber where temperature and humidity are monitored and controlled . Flies were illuminated from the sides by white light-emitting diodes ( LED ) to simulate day-night conditions and by infrared LED from below for video imaging . Videos were captured by a digital camera above the chambers ( see Materials and methods ) . A sample raw video clip is shown in Video 1 . Because the tubes ( commonly used with Drosophila Activity Monitors or DAMs ) are commercially available for studying circadian and sleep behavior , this set-up can be easily replicated by other labs . We then developed an automated video image analysis package that classifies fly behavior into grooming , locomotion , or rest . ‘Grooming’ in our algorithm is defined as fly legs rubbing against each other or sweeping over the surface of the body and wings ( Szebenyi , 1969 ) ( Videos 2 and 3 ) , ‘locomotion’ as translation of the whole body , and ‘rest’ as the absence of either grooming or locomotion . Figure 1B shows images of grooming behaviors frequently observed in our videos involving the head , legs and wings . Since we are primarily interested in detecting grooming events rather than performing a detailed classification of all types of behavior ( Branson et al . , 2009 ) , other behaviors involving body centroid movements , such as feeding , were initially classified as locomotion . This three-tier classification allowed our algorithm to efficiently and rapidly interpret grooming events in the recordings without incurring any significant errors in reporting locomotion and rest . To classify behavior , raw videos were processed through four major automated steps: fly identification , feature extraction , classifier training ( optional ) , and subset behavior classification ( Figure 1C ) . First , fly identification was accomplished with the following analysis . Fly shape was extracted from a video frame by computing the difference between the current frame and a reference frame . The reference or background frame was created by comparing eight randomly selected frames and erasing all moving objects from one of them ( see Materials and methods ) . The background frame was updated every 1000 s to account for changes in the fly’s surroundings , such as decrease in the level of food and accumulation of debris within the tube , over the course of multiple hours ( Figure 2—figure supplement 1B ) . A preliminary image of flies in the current frame was determined by comparing the frame to background and setting all pixels greater than a threshold C0 ( Figure 2A ) equal to 10 . Despite the use of C0 , some artifacts in the form of small objects still remained in the extracted image . A C0 = 10 rejects artifacts larger than 20 pixels ( Figure 2B ) . Based on this , to further eliminate remaining small objects , we erased all closed objects with areas less than a second threshold C1 = 25 pixels , retaining only the fly silhouette ( Figure 2—figure supplement 1C , right ) . Thus , each individual fly and its movements were distinguished from background structures . Second , we performed feature extraction to distinguish three specific types of behaviors , which are grooming , locomotion , and rest , performed by the individual fly . The features we used were: ( 1 ) periphery movement ( PM ) , which characterizes movements of the legs , head and wings; ( 2 ) core movement ( CM ) , which quantifies movements of the thorax and abdomen; and ( 3 ) centroid displacement ( CD ) , which quantifies whole body displacement . Extracting these three features allowed us to identify patterns corresponding to different types of behavior . To extract PM and CM , we split each fly’s body into a core and a periphery . Based on the grayscale distributions of the two parts ( Figure 2C ) , we set the median of pixel grayscale values as the criterion to split a fly body into core ( darker ) and periphery ( lighter ) . This criterion made the core and periphery areas roughly equal , giving PM and CM equal weight in the feature space . Slight variations in light condition across the arena can cause differing grayscale distribution for each individual . We therefore calculated the median value separately for each fly . After splitting the fly’s body into two parts , PM and CM were extracted by computing the number of non-overlapping periphery and core pixels , respectively , in two consecutive frames . To extract CD , we calculated the average position of all pixels from the individual fly and defined changes in that quantity between every two consecutive frames as CD . Since the fly moves in essentially one dimension through the narrow tube , we ignored movements perpendicular to the long axis of the tube when calculating centroid movement . In subsequent analysis , fly location was represented by its centroid position . Noise in the apparatus may slightly change the centroid position even when a fly is stationary . Figure 2D shows the distribution of such centroid displacements caused by noise . Based on this distribution , we set 0 . 5 pixel length as the minimum actual CD -- that is , displacements smaller than 0 . 5 pixel were ignored . Application of this threshold eliminated 99 . 99% of such false displacements and accurately identified fly centroid displacement . By extracting these three features ( PM , CM and CD ) , we were able to distinguish between locomotion , rest , and grooming . As shown in Figure 2E , relative metrics of PM and CM were different depending on the type of behavior . Specifically , during locomotion , both parts moved significantly ( Figure 2E , bottom-right ) together with substantial changes in CD . During rest , no significant movement was seen either in the periphery or the core ( Figure 2E , bottom-left ) . During grooming , the periphery moved more than the core ( Figure 2E , top-left , top-right ) . Importantly , since differences in fly size can affect values of PM , CM and CD , we normalized these features to individual fly size before proceeding with further analysis ( see Materials and methods ) . The behavior-dependent changes of these features suggest that PM , CM and CD are appropriate metrics for behavior classification . Third , to produce a rapid , objective and automated quantification of grooming behavior , we performed classifier training to teach the algorithm to automatically recognize these features . We classified fly behavior by applying the k-nearest neighbors ( kNN ) technique to the normalized features ( Bishop , 2007; Dankert et al . , 2009; Kain et al . , 2013 ) . Briefly , kNN works by placing an unlabeled sample into a feature space with pre-labeled samples serving as a training set for the algorithm . The label or class of the unlabeled sample is then decided by the label that is most common among its k-nearest training samples . In our case , the nearest neighbors were searched through a k-d tree algorithm ( Sproull , 1991 ) . To construct the kNN classifier , we prepared a training set by visually labeling fly behavior from 25 , 000 frames ( 9322 frames of grooming , 9930 frames of locomotion and 5748 frames of resting from 20 w1118 flies ) and mapping them onto a three-dimensional feature space where the axes correspond to normalized PM , CM and CD ( Figure 2F , color symbols ) . With these training samples , we applied 10-fold cross-validation ( Bishop , 2007; McLachlan et al . , 2005 ) to the kNN classifier with k ranging from 1 to 50 and settled on k = 10 to achieve balance between computing time and accuracy ( Figure 2—figure supplement 1D ) . Finally , to specifically distinguish between grooming behavior and other types of peripheral movement , we pruned output labels from the kNN classifier ( Figure 3A ) . The algorithm calculates features from every two consecutive frames , resulting in some classifications being confounded by short-term fly activity . For example , features extracted from only two frames often cannot distinguish a fly stretching its body parts from one that is grooming ( Video 4 ) . Based on our observations during creation of the training set , a typical bout of grooming lasts >3 s or for 15 frames at our normal frame rate , longer than an average stretching event , which lasts for ~1 s . Accordingly , we devised a strategy in which a ~ 15 frame-long temporal filter slid one frame at a time to eliminate false grooming labels caused by short , grooming-like behavior . Grooming designations were retained only if at least a minimum number of grooming frames were found within the filter ( Figure 3A ) . To determine the size of the filter and the minimum number of grooming frames within , we assessed the accuracy of our classifier with the ‘minimum number of grooming frames/size of filter’ at 4/5 , 8/10 , 8/15 , 10/15 , 10/20 , 12/15 , 14/15 , and 15/20 . These tests were conducted with a 10 min video ( N = 20 Canton S flies ) . As expected , comparison between 8/15 , 10/15 , 12/15 and 14/15 shows ( Figure 3B ) that for fixed filter sizes , a larger number of grooming frames led to fewer false positive ( higher accuracy ) but more frequent false negative identification of grooming ( lower sensitivity ) . On the other hand , <12 minimum number increased risk of misidentifying other short-term grooming-like behaviors as grooming . Based on these findings , we set the pruning filter to be 12/15 , simultaneously minimizing false positive and false negative errors . Because of this pruning process , if fewer than 12 grooming frames were found within a 15-frame sliding window , then all grooming frames were re-labeled as locomotion once the left edge of the window reached the 15th frame ( Figure 3A ) . Thus , these pruned labels were the final output of our grooming classification algorithm , consisting of fly identification , feature extraction and classifier training . The accuracy of our algorithm was evaluated by comparing the computer-identified grooming with manually labeled grooming identified by visual inspection . We tested a total of 450 min of videos from a different set of w1118 flies ( N = 15 ) than the one used in training the classifier . The comparisons showed that , of the grooming events picked out by our algorithm , 92 . 1% were manually verified as true grooming events ( Figure 3C , top panel ) . Furthermore , among all manually scored grooming events , 95 . 5% were successfully identified by our computational method ( Figure 3C , bottom panel ) . Since size and pigmentation differences between genotypes can potentially affect behavioral classification , we investigated robustness of our w1118-trained classifier with manually-labeled data from Canton S , iso31 , and yw strains ( 10 min videos with N = 20 of each type ) . As shown in Figure 3C , error rates in each tested strain less than 10% . Together , these results suggest that our method identifies grooming with high fidelity in several different Drosophila melanogaster strains . The solitary flies in our experiments also spent portions of their time feeding ( Ja et al . , 2007 ) and sleeping ( Hendricks et al . , 2000; Shaw et al . , 2000 ) , behaviors that our classifier did not initially label but that can nevertheless be identified by our algorithm . Prolonged proximity with food ( >3 s , <body length ) was accepted as a proxy for feeding . Rest periods lasting ≥5 min ( Dubowy and Sehgal , 2017 ) were classified as sleep , following the currently accepted definition of the behavior . Together , these additional classifications led to the identification of five major behaviors in our data: grooming , locomotion , feeding , short rest ( < 5 min of quiescence ) , and sleep ( Figure 4 ) . The first four behaviors are mutually exclusive at the level of single events , together defining the wake state of the fly , and collectively complementary to the sleep state ( Figure 4A ) . We found that a typical iso31+ fly under 12 hr light:12 hr dark ( LD ) conditions spent approximately 6% of its daily time grooming , ~24% time locomoting , ~3% time feeding , ~16% resting , and the remaining ~51% sleeping ( Figure 4B ) . That is , the average iso31+ fly spent ~13% of its awake time grooming . It is worth noting here that such behavioral statistics can vary even between wild-type laboratory strains ( Colomb and Brembs , 2014; Zalucki et al . , 2015 ) . For instance , similar analysis of a Canton-S strain showed that these flies groomed ~19% of their awake time ( Figure 4—figure supplement 1A ) . These analyses demonstrate that our platform for long-term video-tracking and automated analysis can provide a quantitative ethological structure for daily basal fly behavior . Since sleep and wake are complementary states , we expected fractional time spent in sleep to negatively correlate with that of the four wake behaviors our method tracks . Pair-wise comparisons ( Pearson’s correlation coefficient , r , see Materials and methods ) of individual fly sleep with grooming , locomotion , short rest or feeding , showed the expected negative relationships ( Figure 4C and Figure 4—figure supplement 1B ) . Interestingly , the strength of negative correlation with sleep ( Figure 4C ) increased with the average fractional time spent in a wake behavior ( Figure 4B ) . We reasoned that similar analysis among the wake behaviors , in contrast , should show positive correlations . Pair-wise comparisons among grooming , locomotion , short rest and feeding showed the predicted positive correlations , although to varying degrees ( Figure 4D and Figure 4—figure supplement 1C ) . The analyses further revealed that the fraction of time a fly spent in short rest was the best predictor of its grooming time ( r = 0 . 42 in iso31+ and 0 . 26 in Canton-S ) while locomotion ( r = 0 . 26 and −0 . 13 ) and feeding ( r = 0 . 27 and 0 . 06 ) were both less reliable in predicting grooming . The weaker grooming-locomotion and grooming-feeding correlations were unexpected for two reasons . First , daily variations in grooming levels had appeared to closely follow those in locomotion ( Figure 4—figure supplement 2A ) , suggesting the possibility that grooming is a by-product of the more robustly driven locomotor activity . Second , feeding activity has been postulated to act as a trigger for grooming with food debris serving as an external stimulus ( Hampel et al . , 2015; Seeds et al . , 2014 ) . To further dissect the lack of predictive relationship between grooming and locomotion , we first examined temporal parameters that describe grooming and locomotion over short timescales ( Figure 4—figure supplement 2A–E ) . Basal locomotor events during mid-day and night ( Figure 4—figure supplement 2A , rectangles ) were relatively sparse compared to grooming episodes during the same times . This difference in inter-event time interval between grooming and locomotion persisted to different degrees throughout the day-night cycle , such that the average longest pause between two subsequent grooming events was ~88 min while that between two locomotor events was ~132 min ( Figure 4—figure supplement 2C ) . Examination of the duration of individual events showed grooming events on average lasted for ~0 . 23 min compared to ~0 . 44 for locomotor events ( Figure 4—figure supplement 2D ) . These analyses revealed significant differences between the two behaviors over short timescales and do not support locomotor activity as a driver of grooming . To focus on temporal dynamics at longer timescales , we binned multi-day data in 30 min ( Figure 4—figure supplement 2F , G ) and applied least square fit to a previously developed mathematical model that describes long timescale variations in fly activity in terms of exponential functions ( Lazopulo and Syed , 2016 , 2017 ) . The functions were defined by four rate parameters bMR , bMD , bER and bED , where subscripts denote morning rise ( MR ) , morning decay ( MD ) , evening rise ( ER ) and evening decay ( ED ) , and two duration parameters that describe the relative durations of morning ( TM ) and evening ( TE ) peaks in activity ( Figure 4—figure supplement 2H and Figure 4—figure supplement 3 ) . Results from this analysis showed that the rate parameter bMR of grooming was smaller than that of locomotion ( Figure 4—figure supplement 2I ) , indicating a slower increase in night-time grooming activity . Additionally , the evening duration parameter ( TE ) for grooming was greater than that for locomotion ( Figure 4—figure supplement 2J ) , indicating that the evening peak in grooming lasted longer . These differences in long timescale kinetics were again inconsistent with locomotor activity as a driver of grooming . Finally , comparison with large timescale variations in feeding patterns showed that peak time in contacting food was offset by 2–4 hr from nearby peaks in grooming ( Figure 4—figure supplement 2O–P ) . The large temporal offset suggests contact with food is also not likely to drive the majority of grooming events observed in our experiments . Thus , according to our analyses of the kinetics of Drosophila ethograms in our system , neither locomotor activity nor feeding is likely to be a primary driver of basal grooming . To identify major drivers of basal grooming , we noted that multi-day time series of the behaviors showed time-of-day-dependent changes in each behavior ( Figure 4E ) . The appearance of repeating patterns raised the possibility that external light-dark ( LD ) cycles alone or in combination with internal programs could be exerting temporal control over several of these behavioral outputs , including grooming . Indeed , environmental light-dark cycles through influence on the circadian clock are known to drive rhythmic changes in fly sleep and wake durations and within the awake state , feeding , and locomotor activities ( Chatterjee and Rouyer , 2016; Pfeiffenberger et al . , 2010 ) . That these rhythms persisted in the absence of LD cycles is generally considered to be strong support for clock control of these behaviors . We set out to determine whether the circadian clock drives rhythmic modulations in fly daily grooming independent of other circadian-regulated behaviors--that is , to test whether grooming exhibits circadian oscillations simply because individual grooming events are mutually exclusive of other individual wake activities . We recognized that the mutual exclusivity of the behaviors seen at the level of individual events ( Figure 4A ) did not persist at the level of fractional time in each behavior where the long timescale modulations are visible ( Figure 4E ) . This is because fractional time data are binned and the only constraint on these data was that the sum of the time spent in each wake behavior ( grooming , locomotion , feeding and short rest ) and sleep equaled one for each time bin ( Figure 4—figure supplement 1F ) . In this representation , therefore , rhythmicity of one behavior ( i . e . grooming ) did not dictate rhythmic status of another ( i . e . locomotion ) . To test the independence of rhythms , we performed a series of ‘shuffling experiments’ using well-established ( Allada and Chung , 2010; Chatterjee and Rouyer , 2016 ) rhythmicities of wakefulness and locomotion as metrics ( Figure 4F ) . In brief , we took data from Figure 4E in which grooming , locomotion and wakefulness have LD-driven ~24 hr rhythms ( Figure 4F , left and power spectra ) and computationally randomized the grooming time-series such that it lost rhythmicity ( Figure 4F , right ) . To account for the randomized grooming , we also adjusted either locomotion ( Figure 4F , upper panel ) or wakefulness ( Figure 4F , bottom panel ) , in both cases ensuring that wakefulness was between 0 and 1 at all times ( see Materials and methods ) . In either case , we found that rhythmicity in locomotion and wakefulness were intact regardless of the rhythmic status of grooming . The simulation result suggested that circadian control of fly locomotion and wakefulness does not guarantee circadian control of underlying basal grooming , at least as measured from changes in the duration of the behaviors . Therefore , demonstration of robust ~24 hr rhythms in grooming in the absence of any external cues should be strong evidence in favor of circadian control of the behavior . To test whether basal grooming is also under circadian control , we first entrained iso31+ + to 2 days of alternating light-dark cycles and then monitored their behavior over multiple days in constant darkness ( DD ) . In the absence of light cues , locomotor , feeding and sleep showed the familiar clock-driven rhythms in their daily timing ( Figure 5A , B ) . Although short rest appeared to undergo rhythmic changes ( Figure 5A ) , spectral analysis indicated these changes did not result in statistically significant rhythms at the p=0 . 05 level ( Figure 5B ) . Lack of rhythms in short rest is consistent with our earlier reasoning that rhythmic wakefulness and locomotion does not necessarily imply rhythmicity of each behavior in the awake state . Grooming data also showed periodic changes in constant darkness ( Figure 5C ) . Power spectra of individual time-series ( ‘WT’ in Figure 5D and Figure 5—figure supplement 1A ) indicated these periodic changes to be statistically rhythmic by revealing peaks significant at p=0 . 01 in 100% of flies ( 29 out of 29 individuals , Figure 5E ) . The average period of oscillations was 23 . 72 hr , with a standard deviation of 0 . 72 hr ( Figure 5—figure supplement 1B ) . The presence of these robust circadian rhythms in the absence of external cues further support the hypothesis that fly basal grooming is under control of the internal timekeeper . Consistent with our prediction that grooming rhythms in DD do not necessarily follow from rhythms in locomotion or wakefulness , we found that knowing locomotion or wakefulness is rhythmic did not inform about the rhythmic status of grooming ( Figure 5—figure supplement 2 ) . This finding further underscored the importance of the DD studies in establishing rhythmicity in basal grooming . It should be noted here that our simulation results do not demonstrate bidirectional independence of rhythmicity in wakefulness and grooming but , only that rhythmicity of wakefulness does not depend on that of grooming . Demonstration of fully independent rhythms in the two behaviors is beyond the scope of the present study . We next took advantage of several circadian mutants to examine further the control of grooming by the circadian clock . The Drosophila clock is composed of two interlocked genetic feedback loops in which period ( per ) is one of the core components and whose transcription is controlled by the primary transcription factors Clock ( clk ) and Cycle ( cyc ) ( Allada and Chung , 2010 ) . The per gene has several well-characterized mutant alleles , two of which---perS and perL---produce short and long circadian rhythms , respectively , while a third , per0 , results in arrhythmic behavior ( Konopka and Benzer , 1971 ) . Population-averaged grooming of perS and perL showed altered oscillations in LD and DD ( Figure 5C , second and third rows ) , with average DD periods of 19 . 23 ± 0 . 57 hr and 28 . 84 ± 1 . 13 hr , respectively ( Figure 5D , E and Figure 5—figure supplement 1A ) . The periods of oscillation in grooming were well within published values of circadian rhythms of these mutants ( Konopka and Benzer , 1971 ) and in agreement with alterations in locomotor rhythms of the flies ( Figure 5—figure supplement 3A ) . Consistent with these results , grooming in per0 flies was arrhythmic ( Figure 5C , bottom row ) and , when analyzed at the individual fly level , the power spectra unveiled the absence of statistically significant rhythms in 19 out of 20 flies at p=0 . 01 level ( Figure 5D , E and Figure 5—figure supplement 1A ) . Moreover , analysis of grooming patterns in cyc01 ( Rutila et al . , 1998 ) and clkJrk ( Allada et al . , 1998 ) , arrhythmic mutants of cyc and clk , also showed loss of circadian rhythms ( Figure 5E and Figure 5—figure supplement 3B–D ) . Together , these results support the hypothesis that the circadian clock temporally modulates fly grooming . To test whether , in addition to regulating the timing of grooming , the circadian clock also regulates grooming duration , we examined the average duration of grooming in circadian mutants . Despite causing major changes in temporal patterns of grooming , the per0 mutation did not significantly change the average duration of grooming in these flies ( Figure 6A ) . In contrast , cyc01 and clkJrk mutants both exhibited increased daily average grooming relative to their respective genetic controls ( Figure 6B , C ) . While both mutants exhibited increased grooming duration , this change was accompanied by opposing changes in their locomotion: cyc01 flies spent less time and clkJrk flies spent almost twice as much time in locomotion ( Figure 6B , C , pie plots ) . Thus , the increase in cyc01 grooming came almost entirely from loss of locomotor activity while the increase in clkJrk grooming came from loss of sleep . These results support the hypothesis that locomotion and grooming are partly independent behaviors and further suggests that the cyc01 and clkJrk mutations alter the insect’s internal homeostasis in distinct ways , similar to phenotypic differences reported previously in sleep studies involving cyc01 and clkJrk ( Hendricks et al . , 2003; Shaw et al . , 2002 ) . Importantly , together with per0 data , the results raise the possibility of non-circadian roles for cyc and clk in setting the duration of internally driven grooming in Drosophila . cycle and clock have also been implicated in stress response , particularly in regulating level of sleep in response to sleep deprivation and adjusting locomotor output in response to nutrient unavailability ( Hendricks et al . , 2003; Keene et al . , 2010; Shaw et al . , 2002 ) . Because grooming and sleep have both been previously linked to stress , we asked whether reduction in sleep is always accompanied by an increase in grooming as seen in our clkJrk data . To address this question , we examined relationship between grooming and sleep in standard LD cycles in two short-sleeping mutants--fumin and sleepless ( sss ) . Consistent with the original studies ( Koh et al . , 2008; Kume et al . , 2005 ) , our method found both strains to have extremely low levels of sleep ( Figure 6D , E , pie plots ) . But , while loss of sleep in fumin was accompanied by an upregulation in grooming ( Figure 6D ) , loss of sleep in sss was accompanied by a dramatic downregulation in grooming , compared to control flies ( Figure 6E ) . These divergent relationships between sleep and grooming ( e . g . sss vs . fumin ) and between locomotion and grooming ( e . g . clkJrk vs . cyc01 ) became more evident when individuals of different genotypes were compared together ( Figure 6—figure supplement 1F , G ) . To better visualize the effects of disparate mutations , data of each genotype in these plots were normalized to the population-mean of its genetic control . These results suggest that resetting of the level of internally-driven grooming can occur via a number of ways with complex compensatory changes in sleep and locomotor behavior . Accumulated data from our experiments suggest that grooming is an innate fly behavior controlled by two major regulators . One of these regulators controls temporal patterns in grooming and the other controls amount of time spent in grooming . Circadian genes per , cyc and clk are involved in controlling the timing of peaks/troughs in grooming rhythms while cyc and clk are also involved in setting how much time is spent grooming . The apparent absence of per from the second regulatory mechanism is consistent with the possibility that the two control mechanisms operate independently . Nearly all animals tested exhibit daily basal grooming behavior , suggesting that grooming is not only fundamental to health but also reflects a generally healthy state . Consistent with this , loss of grooming is indicative of sickness behavior ( Hart , 1988 ) associated with infection or old age , and , in the case of humans , mental illness . A greater understanding of the molecular mechanisms regulating grooming would provide insight into the principles and neural circuits underlying other complex programmed behaviors , as well as potentially identify biomarkers of pathological disease states . Critical to the dissection of these molecular mechanisms is a system for rapid , automated interpretation of grooming in a genetically tractable model organism . The development of our platform will facilitate high-throughput and unbiased analysis of the genetic regulators and neural circuits that control grooming , as well as those responsible for loss of grooming in the context of disease . Grooming continues to be one of the least understood Drosophila behaviors , possibly due to the technical challenges of detecting grooming events in this small insect . Early work describing fly grooming relied on manual scoring ( Connolly , 1968; Szebenyi , 1969; Tinbergen , 1965 ) , which imposes significant limitations on the length of events that can be detected , fidelity and objectivity of detection , and the level of detail that can be extracted from the data . Despite such limitations , these initial studies made a number of noteworthy observations . Szebenyi delineated all the major modes of fly grooming and suggested that repetitive grooming actions may closely follow a preset sequence ( Szebenyi , 1969 ) . A subsequent study in the blowfly offered a more refined mechanistic picture of insect grooming by proposing that the sequential actions form a hierarchical structure ( Dawkins and Dawkins , 1976 ) . Combining modern computational and genetic tools , an elegant study in Drosophila recently confirmed these previous hypotheses ( Seeds et al . , 2014 ) . That fruit flies may groom spontaneously in the absence of any apparent stimulus has also been previously suggested ( Connolly , 1968; Tinbergen , 1965 ) . Consistent with this , our work provides evidence that fruit flies groom as part of their daily repertoire of internally programmed behaviors and often without any obvious external stimulus . Our analysis revealed that over a period of hours , grooming is temporally structured by the fly circadian clock , with peak activity around dawn and dusk . The study also identifies transcription factors CLOCK and CYCLE as critical molecular components that control the amplitude of programmed Drosophila grooming . Machine-learning is increasingly gaining popularity due to its applicability to virtually any problem involving pattern classification , including in studies aimed at deconstructing stereotyped behavior in the fruit fly ( Branson et al . , 2009; Kabra et al . , 2013; Kain et al . , 2013; Mendes et al . , 2013; Valletta et al . , 2017 ) . Similar to these recent efforts , we constructed a computational pipeline incorporating elements of machine learning to automatically identify grooming events in video recordings of behaving flies . Our approach relies , in particular , on a supervised k-nearest neighbors algorithm to broadly classify behavior into grooming , locomotion and rest ( Figure 2 ) . Application of additional optional filters yielded approximate data on feeding and sleep ( Figure 4 ) . While previous methods offer important details on different modes of grooming ( Berman et al . , 2014; Seeds et al . , 2014 ) , leg movements ( Kain et al . , 2013; Mendes et al . , 2013 ) , and fly-fly interactions ( Branson et al . , 2009; Kabra et al . , 2013 ) from short videos , the methods have limited capability for interpreting multi-day and multi-fly recordings . The method presented here offers less detail on modes of grooming , but can instead readily dissect circadian time-scale recordings into three to five behavioral classes on a typical personal computer . The apparatus used in this method ( Figure 1 ) also offers a number of advantages over current ones . First , most items used in the apparatus , including the ~6 cm tubes in which flies are visualized , are standard in a typical circadian experiment studying fly locomotion or sleep ( Lazopulo et al . , 2015; Pfeiffenberger et al . , 2010 ) using the Drosophila Activity Monitor ( DAM ) . The retention of this basic feature should lower the technical hurdle for the interested investigator who is likely to be one already engaged in locomotion and sleep studies in Drosophila . The use of a shared design to house flies also means that both experimental subjects and certain conclusions drawn from one platform may be readily transferred to the other . Most current grooming methods require specialized equipment for fly stimulation and detection ( Seeds et al . , 2014 ) , elaborate optics ( Kain et al . , 2013 ) , or a specific form of fluorescence microscopy ( Mendes et al . , 2013 ) . Second , our apparatus can simultaneously monitor up to ~20 flies , while the existing approaches , although offering higher-resolution data , monitor only one animal at a time . The scalability and high-throughput nature of our platform should appeal to investigators interested in , for example , large-scale genetic studies to identify mechanisms that differentially affect grooming , locomotion and rest ( King et al . , 2016 ) . Finally , the flies in our apparatus are allowed to move freely over a distance roughly 10 times their body length and still remain in the camera’s field of view while technical constraints in other studies limit visualization to short distances ( Mendes et al . , 2013 ) . The relative freedom of mobility , access to food , and long time-scales of observation offered by our apparatus thus facilitate analysis of basal , internally programmed behavior . These properties make our platform amenable to addressing questions of biological relevance , such as the importance of grooming behavior , its temporal regulation with regards to other fly behaviors , and its dependence on the circadian timekeeping system . First , we found that flies consistently devote a significant fraction of time to grooming behavior during periods of wakefulness ( 13% ) , and surprisingly , that grooming behavior is observed even during periods of reduced locomotor activity ( Figure 4—figure supplement 2A ) . This suggests that the benefits of grooming outweigh the caloric resources expended and the resulting interruption of rest , underscoring the hypothesis that daily grooming is a fundamental behavior of Drosophila . A few recent studies ( Hampel et al . , 2015; Phillis et al . , 1993; Seeds et al . , 2014 ) have shown that fly grooming can be directly induced by peripheral stimuli , and there has been considerable progress toward identifying the behavioral and neural aspects of such stimulus-induced grooming . However , programmed grooming , or grooming in the absence of a macroscopic stimulus , remains relatively understudied in Drosophila . To our knowledge , the existence of programmed grooming , first proposed in the mid 60s , still remains unreported . Data from this study suggest that a significant portion of daily fly grooming is driven by internal programs . Flies in our experiments are active for ~34% of the time within a 24 hr period , during which they mostly engage in grooming , locomotion and feeding . Behavioral analysis showed that , like sleep , locomotion and feeding , fly grooming behavior is modulated by oscillations of the circadian clock ( Figure 5 ) . This finding raised the possibility that the observed grooming was stimulated by rhythms in contact with food or locomotor activity . However , closer examination revealed that kinetics in feeding and locomotion were distinct from those of grooming ( Figure 4—figure supplement 2 ) . Additionally , genetic modifications resulted in contrasting changes in these behaviors ( Figure 6 ) . These results together suggest that the majority of grooming events detected in our experiments are not triggered by external stimuli such as light , food and locomotor movements . Rather , internal regulatory mechanisms , independent of external stimuli , likely drive this programmed behavior . Multi-day recordings of wild-type flies in constant darkness showed 24 hr rhythms in daily grooming patterns ( Figure 5 , Figure 5—figure supplement 1 ) . Furthermore , these rhythms were shifted appropriately in the canonical period mutants perL and perS and abolished in arrhythmic per0 flies ( Figure 5 ) . These data support a regulatory model in which timing of programmed grooming behavior is orchestrated by the circadian clock . Notably , since loss of rhythmicity did not significantly affect the amount of grooming ( Figure 6A ) , our results suggest that the primary role of the clock is to organize the behavior in time without influencing the total time flies dedicate to grooming . Intriguingly , two other circadian mutations , cyc01 and clkJrk , increased the proportion of daily time flies spend grooming ( Figure 6B , C ) , implying that the changes in grooming level may not be due to circadian defects . These data are consistent with the hypothesis that clock-independent but cyc- and clk- dependent pathways regulate the amount of programmed grooming behavior . Finally , why are flies innately programmed to groom ? The present study does not directly address this important question , but given that microscopic pathogens can sporulate on the fly cuticle and eventually infect the insect ( St . Leger et al . , 2011 ) , persistent grooming may serve as a first line of defense against such attack . Thus , the immune system may constitute another internal program , similar to the cyc and clk-controlled mechanisms , that drives fly grooming; if so , we hypothesized that mutants with defective immune response may exhibit altered grooming behavior ( Lemaitre et al . , 1995; Michel et al . , 2001 ) . Consistent with this , we found that grooming was reduced in the immune-deficient imd mutant ( Figure 6—figure supplement 1H ) , although a second immune-deficient strain lacking a member of the Toll pathway ( PGRP-SAseml ) did not show a significant change . Further studies are required to clarify these initial results and elucidate the biological function of programmed grooming in Drosophila . Together , our data provide strong supporting evidence for programmed grooming in Drosophila and suggest that this innate behavior is driven by two possibly distinct sets of regulatory systems . The circadian system temporally segregates time-dependent variations in grooming from those of other essential behavioral outputs like feeding and sleep . Circadian coordination of grooming underscores a previously under-appreciated importance of this behavior in the daily routine of the fruit fly . The second regulatory system adjusts the level of grooming relative to other behaviors . This set of regulation likely confers adaptability on the animal by allowing it to up- or downregulate grooming as necessitated by internal and external conditions . The dual control mechanism of grooming proposed here is highly reminiscent of the two-process framework---circadian and homeostatic---that is widely used in understanding sleep regulation ( Borbély , 1982 ) . Although this work has not demonstrated grooming is under homeostatic control , future studies could be aimed at better characterizing the nature of the non-circadian regulatory system of fly grooming . In summary , we present here a new platform to detect innate grooming behavior simultaneously and for days at a time in multiple individual fruit flies . The apparatus can be assembled easily , and the accompanying analytics are available publicly . Utilizing this platform , we report several mechanisms that are possibly responsible for driving the timing and level of programmed grooming in Drosophila . We also suggest future experiments that through use of this platform can lead to deeper understanding of the underlying biology of grooming and its relation to other essential fly behaviors . Clock mutants perS , per L , and per0 were backcrossed for five-six generations to an iso31 with mini-white insertion strain ( iso31+ ) . cyc01 flies , gifts from William Ja ( The Scripps Research Institute ) , have the Canton S background . ClkJrk flies were backcrossed for five generations to iso31 . sssP1 mutant flies , gifts from Amita Sehgal ( Perelman School of Medicine at the University of Pennsylvania ) , have the iso31 background . fumin mutants , gifts from F . Rob Jackson ( Tufts University School of Medicine ) , have the w1118 background . Flies were bred and raised at 23°C and 40% relative humidity on standard cornmeal and molasses food . All experiments were done with 5–8 days old males at 260C and 70–80% relative humidity in a custom-built behavior tracking chamber ( Figure 1 and Figure 2—figure supplement 1A ) . For each experiment , control strain refers to the genetic background of a mutant . WT flies in Figures 4 and 5 refer to the iso31+ line . Using a desktop computer with Intel Core i7-4770 3 . 4 GHz processer and 4 × 4 G DDR3 1600 MHz RAM , it takes ~7 hr to extract grooming , locomotion and rest data from an 8 hr video of 20 flies recorded in 10 Hz ( in total 288 , 000 frames ) at 1280 pixel ×960 pixel resolution . Videos are analyzed every two frames ( 5 Hz ) , which is sufficient to capture grooming events . All computational analyses were done with custom-written Matlab scripts that will be available at https://github . com/sbadvance/Drosophila-Grooming-Tracking . git ( Qiao , 2017; copy archived at https://github . com/elifesciences-publications/Drosophila-Grooming-Tracking ) . Fly shape extraction . Fly shape was extracted by applying a background subtraction algorithm . The background or reference frame is constructed randomly picking two frames , a template and a contrast , and comparing their pixel grayscale values and erasing all moving objects from the template frame . To remove the fly from the template frame , we replace the pixels belonging to the fly with corresponding pixels from contrast frames , relying on the fact that a fly is always darker than the surrounding objects . The template frame with no fly present then becomes the background frame . Additionally , because a fly’s surroundings , including food debris , change substantially during the course of an experiment ( Figure 2—figure supplement 1B ) , the background frame is regenerated every 1000 s . Lastly , if a fly occupies the same area in the template and contrast frames , the overlapping region cannot be erased on the template . To circumvent this problem , every time a background frame is generated , we randomly choose seven , instead of one , frames as contrast frames and compare all of them with the template . When a fly does not move for more than 1000 s , the fly will not be removed from the background and cannot be detected in other frames during this 1000 s . Thus , when a fly is not detected , we consider the fly to be stationary at the position where it was last detected . To reduce effects of charge coupled device ( CCD ) image noise and fluctuations in the system , we set a minimum change C0 as the threshold to accept grayscale changes from fly movements . We denote the grayscale value of a pixel located at ( x , y ) ( in units of pixel , in our case , x ∈ [1:1280] , y ∈ [1:960] ) in the template as Itemplate ( x , y ) and in the contrast frame Icontrast ( x , y ) . Only ifItemplate ( x , y ) −Icontrast ( x , y ) >C0thenItemplate ( x , y ) =Icontrast ( x , y ) While increasing threshold C0 reduces noise , it can also lead to rejection of real movements of the fly . To optimize C0 , we tested noise levels in our images by analyzing a 3-hr video with dead flies . In the test , 30 pairs of consecutive frames were randomly chosen from the video and the differences between their corresponding grayscale pixel values were calculated . The distribution of the differences , stemming from noise , is shown in Figure 2A . Based on this distribution , we set C0=10 , which excludes 99 . 99% of noise-related changes in grayscale values . Feature normalization . Since PM and CM both represent areas ( number of pixels in area ) , while CD represents distance , we take the square root of PM and CM to make the dimensions of the features homogeneous . In addition , fly size varies between individuals and across experimental settings . To facilitate comparison of data in feature space , we therefore normalize PM , CM and CD of each fly with a scale parameter SP equal to the square root of the area of that fly . Thus , the final form of normalized features areNormalized PM=PM/SPNormalized CM=CM/SPNormalized CD=CD/SP Figures 4 and 5 and Figure 5—figure supplements 1–3: To measure periodicity in locomotion and grooming recordings , we applied the Lomb-Scargle periodogram ( Lazopulo et al . , 2015; Scargle , 1982 ) to time-series that were binned into 30 min periods . Power at indicated p values shown in power spectra were calculated according toPower=-ln⁡ ( 1- ( 1-p ) 1/N ) where p is the p-value and N is the number of frequencies computed in Lomb-Scargle periodogram . To test the effect of binning on rhythmicity , we binned grooming activity of individual flies in 30 min , 5 min , and 1 min bin sizes and ran Lomb-Scargle periodogram analysis on these binned data , as well as raw data without any additional binning . Examples of 5 individual spectra of each bin size are shown in Figure 5—figure supplement 1C . As shown in the figure , the separation between statistical cut-off power ( at certain p value , horizontal lines ) and peak power increases with smaller bin size or equivalently , larger number of data points ( N ) . This is because in Lomb-Scargle periodogram , cut-off power grows as log ( N ) while peak power grows as N . In Figure 4F and Figure 5—figure supplement 2 , randomized grooming was generated by randomly shuffling time in raw grooming data . The corresponding modified locomotion and wake were calculated according to Modified locomotion = original locomotion+original grooming – randomized grooming Modified wake = original wakefulness+original grooming – randomized grooming These manipulations modified either locomotion or wake while keeping the other unchanged . No sample size estimation was performed when the study was being designed . Unless otherwise specified , quantitative experiments with statistical analysis were repeated at least three times independently . Exclusion of data applies to flies which were physically damaged ( for example , broken wings or legs ) , physically confined ( for example , trapped by condensation inside tubes ) , or dead during experiments . For testing statistical significance of differences between groups , we first tested the normality of data by one-sample Kolmogorov-Smirnov test . Two-sample F-test was applied for equal variances test . Samples with equal variances were compared using two-tailed t-test . Satterthwaite's approximation for the effective degrees of freedom was applied for samples with unequal variances . Results were expressed as mean ± s . d . , unless otherwise specified . p<0 . 05 , p<0 . 01 , **p<0 . 001 were considered statistically significant . In Figure 4C , D and Figure 4—figure supplement 1B , C , the Pearson correlation coefficient r for each pair of data was calculated according to the standard definitionrX , Y=E[ ( X−μX ) ( Y−μY ) ]σXσYwhere X and Y are time spent in two behaviors X and Y , rX , Y is the Pearson correlation coefficient between two behaviors , E[ ] is the expectation value , μ and σ are , respectively , mean value and standard deviation of a behavior . The statistical significance of r was estimated through bootstrapping . For each two behaviors , we randomly paired data from n flies ( n = 84 for iso31+ and n = 76 for Canton S ) and calculated a correlation coefficient r . This process was repeated 100 , 000 times and the empirical distribution of the randomly paired r values were used for a two-tailed test ( Figure 4—figure supplement 1D ) . p-values for all Pearson correlation coefficients are presented in Figure 4—figure supplement 1E .	From birds that preen their feathers to dogs that lick their fur , many animals groom themselves . They do so to stay clean , but routine grooming also has a range of other uses , such as social communication or controlling body temperature . Despite its importance , grooming remains poorly understood; it is especially unclear how this behavior is regulated . Fruit flies could be a good model to study grooming because they are often used in laboratories to look into the genetic and brain mechanisms that control behavior . Flies clean themselves by sweeping their legs over their wings and body , but little is known about how the insects groom ‘naturally’ over long periods of time . This is partly because scientists have had to recognize and classify grooming behavior by eye , which is highly time-consuming . Here , Qiao , Li et al . have created a system to automatically detect grooming behavior in fruit flies over time . First , a camera records the movement of an individual insect . A computer then analyzes the images and picks out general features of the fly’s movement that can help work out what the insect is doing . For example , if a fly is moving its limbs , but not the main part of its body , it is probably grooming itself . Qiao , Li et al . then borrowed an algorithm from an area of computer science known as ‘machine learning’ to teach the computer how to classify each fly’s behavior automatically . The new system successfully recognized grooming behavior in over 90% of cases , and it revealed that fruit flies spend about 13% of their waking life grooming . It also showed that grooming seems to be controlled by two potentially independent internal programs . One program is tied to the internal body clock of the fly , and regulates when the insect grooms during the day . The other commands how long the fly cleans itself , and balances the amount of time spent on grooming with other behaviors . Cleaning oneself is not just important for animals to stay disease-free: it also reflects the general health state of an individual . For example , a loss of grooming is associated with sickness , old age , and , in humans , with mental illness . If scientists can understand how grooming is controlled at the brain and molecular levels , this may give an insight into how these mechanisms relate to diseases . The system created by Qiao , Li et al . could help to make such studies possible .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "computational", "and", "systems", "biology", "tools", "and", "resources", "neuroscience" ]	2018	Automated analysis of long-term grooming behavior in Drosophila using a k-nearest neighbors classifier
Maintaining constant CO2 and H+ concentrations in the arterial blood is critical for life . The principal mechanism through which this is achieved in mammals is the respiratory chemoreflex whose circuitry is still elusive . A candidate element of this circuitry is the retrotrapezoid nucleus ( RTN ) , a collection of neurons at the ventral medullary surface that are activated by increased CO2 or low pH and project to the respiratory rhythm generator . Here , we use intersectional genetic strategies to lesion the RTN neurons defined by Atoh1 and Phox2b expression and to block or activate their synaptic output . Photostimulation of these neurons entrains the respiratory rhythm . Conversely , abrogating expression of Atoh1 or Phox2b or glutamatergic transmission in these cells curtails the phrenic nerve response to low pH in embryonic preparations and abolishes the respiratory chemoreflex in behaving animals . Thus , the RTN neurons expressing Atoh1 and Phox2b are a necessary component of the chemoreflex circuitry . In mammals , breathing is the prime homeostatic process that regulates CO2 partial pressure ( PCO2 ) in the blood . The respiratory chemoreflex modulates ventilation in response to deviations in arterial or brain PCO2 , mainly through the detection of changes in pH ( Feldman et al . , 2003; Guyenet et al . , 2010 ) . These changes are sensed by central chemosensors located in the brainstem and by the carotid bodies in the periphery , but most of the CO2 chemosensory drive to breathe is thought to arise centrally ( Smith et al . , 2006; Duffin , 2010 ) . The nature of the central chemosensors and of the circuits that mediate the ventilatory response to CO2 is still controversial ( Guyenet et al . , 2010; Huckstepp and Dale , 2011; Nattie , 2011; Guyenet et al . , 2013 ) . Two sites have been proposed to play a dominant role in the central CO2 response: serotonergic neurons in the medulla ( Ray et al . , 2011; Teran et al . , 2014 ) and the retrotrapezoid nucleus ( RTN ) , a loose collection of neurons located ventral and immediately caudal to the facial nucleus ( Goridis et al . , 2010; Guyenet et al . , 2013 ) . RTN neurons—defined by their location , by the expression of Phox2b and of the vesicular glutamate transporter-2 Vglut2 ( also known as Slc17a6 ) and by the absence of markers for the nearby motor and catecholaminergic neurons—are vigorously activated by increases in CO2 ( Mulkey et al . , 2004; Onimaru et al . , 2012b; Wang et al . , 2013 ) . Neurons with the molecular characteristics of RTN neurons are already present ventral of the facial nucleus in the late embryo . The embryonic RTN neurons , previously termed embryonic parafacial cells , were found to have endogenous bursting properties and to couple with and increase the frequency of the respiratory rhythm generating ( Feldman et al . , 2013 ) pre-Bötzinger complex ( pre-BötC ) ( Dubreuil et al . , 2009; Thoby-Brisson et al . , 2009 ) . Impairment of RTN development or function correlates with a blunted response to hypercapnia or acidification ( Nattie and Li , 2002; Dubreuil et al . , 2009; Marina et al . , 2010; Ramanantsoa et al . , 2011; Takakura et al . , 2013 , 2014 ) , and mouse models of congenital central hypoventilation syndrome ( CCHS ) , which phenocopy the blunted or absent chemoreflex of the patients ( Weese-Mayer et al . , 2010; Rossor et al . , 2014 ) , have an atrophic RTN ( Dubreuil et al . , 2008; Ramanantsoa et al . , 2011 ) . However , the part played by the RTN in the respiratory chemoreflex is still highly debated , and the fraction of the CO2 response that is relayed through RTN neurons is undetermined ( Guyenet , 2014 ) . In a recent study , mutants with a damaged RTN displayed a normal CO2/pH sensitivity at embryonic stages in vitro ( Huang et al . , 2012 ) . To resolve the issue of the requirement for RTN neurons in the respiratory chemoreflex , we selectively impaired RTN development or function using intersectional genetic strategies ( Dymecki et al . , 2010 ) and assessed the functional consequences at pre- and post-natal stages . To target the Phox2b-expressing RTN neurons with utmost specificity , we exploited the fact that the majority of them belongs to the small set of neurons that co-express the Phox2b and Atoh1 transcription factors in the embryo ( Dubreuil et al . , 2009; Rose et al . , 2009b ) . We found that invalidation of Atoh1 in Phox2b+ cells , of Phox2b in Atoh1+ cells or of glutamatergic transmission in Phox2b+/Atoh1+ cells curtailed the response to acidification in embryonic brainstem preparations and eliminated the CO2 response in newborn pups . Optogenetic activation of these cells entrained ongoing respiratory rhythm . Together our findings imply that the RTN neurons that express or have expressed Phox2b and Atoh1 are essential for the activation of breathing by increased CO2 or low pH , and that other contributors to the chemoreflex must act via the RTN or in partnership with it . We first verified the ability of embryonic RTN neurons to entrain the respiratory-like motor output using channelrhodopsin-based optogenetics . The Phox2b+ RTN neurons are glutamatergic ( Bochorishvili et al . , 2012 ) and express the glutamate transporter Vglut2 already at embryonic stages ( Dubreuil et al . , 2009 ) . We could thus use expression of the channelrhodopsin-2-YFP ( ChR2-YFP ) fusion protein driven by the Vglut2 promoter in Vglut2::ChR2-YFP mice ( Hägglund et al . , 2010 ) to stimulate embryonic RTN neurons by light . In the transgenic embryos , cells co-expressing ChR2-YFP and Phox2b were concentrated at the medullary surface ventral to the ChR2-YFP-negative facial neurons , thus well accessible to light delivered from the ventral surface ( Figure 1A–C ) . At embryonic-day 14 . 5 ( E14 . 5 ) , single light pulses ( 473 nm , 70 ms , 1–5 mW/mm2 ) applied to the RTN region in brainstem preparations systematically evoked a burst of action potentials in ChR2-YFP expressing cells ( n = 5 ) that resembled the spontaneous rhythmic bursts ( Figure 1D ) . A latency of 192 ± 12 ms ( n = 51 stimulations in three cells ) was measured from the time of onset of the light stimulus to that of the first action potential of the burst response suggesting the requirement of a still unknown , slow obligatory process for burst initiation in the RTN . 1 day later , when the preBötC is coupled to the RTN oscillator and drives a respiratory-like motor outflow ( Thoby-Brisson et al . , 2009 ) , single light pulses ( 473 nm , 150 ms , 1–5 mW/mm2 ) delivered to the medullary surface triggered motor bursts in the C4 phrenic nerve roots ( hereafter C4 ) . When the light pulses were set to activate the RTN in a rhythmic manner at about twice the frequency of the ongoing endogenous rhythm , the C4 motor bursts could be entrained to the stimuli and followed the light-imposed rhythm ( Figure 1E ) . C4 motor bursts could not be evoked when the preBötC excitability was depressed by the μ-opiate agonist D-Ala2-N-Me-Phe4-Glycol5-enkephalin ( DAMGO , 0 . 3 μM , n = 4 preparations ) ( Mellen et al . , 2003 ) ( Figure 1F ) or its development impaired genetically in Dbx1 null mutants ( Bouvier et al . , 2010 ) ( n = 5 preparations ) ( Figure 1G ) indicating that the motor outputs require an intact preBötC . These data suggest that malfunction of the RTN will result in lack of entrainment of the preBötC and thus of the motor output and in a slowed-down C4 activity . 10 . 7554/eLife . 07051 . 003Figure 1 . Effect of photostimulating Vglut2::Chr2-expressing embryonic retrotrapezoid nucleus ( RTN ) neurons on membrane potential and motor output . ( A ) Cartoon of a ventral view of the hindbrain showing the position of nVII ( dotted line ) , RTN ( red outline ) , the preBötC and the C4 phrenic nerve roots . ( B ) Ventral view corresponding to the boxed area in A ( anterior at top , lateral on the left ) of the RTN ( red outline ) and nVII ( stippled outline ) in a Vglut2::ChR2-YFP embryo , labeled for the indicated markers . ( C ) Sagittal section , labeled for the indicated markers , showing the Phox2b+/YFP+/Islet1 , 2− cells ventral to nVII and optimally accessible to light . ( D ) Above , ventral view of YFP expression in the RTN region of an E14 . 5 Vglut2::ChR2-YFP embryo showing the three recorded and biocytin-filled RTN cells ( left panel ) . A high magnification of the one in the boxed area is provided in the right panels showing biocytin and YFP labeling separately and the overlay . Below , membrane potential trajectory of the RTN cell shown at high magnification , featuring spontaneously rhythmic ( stars ) and light-evoked ( blue bars ) bursts of action potentials . ( E ) Left panel , photostimulation of the RTN region in E15 . 5 brainstem-spinal cord preparations from a Vglut2::ChR2-YFP embryo evokes Calcium Green-1 AM ( ΔF/F ) changes in the RTN and adjacent nVII . Right panel , integrated C4 recording in such a preparation showing spontaneous C4 bursts of activity ( stars ) and bursts evoked by photostimulation ( blue bars ) . Note that the light pulses entrain the ongoing endogenous rhythm . ( F ) In the presence of DAMGO that depresses the excitability of the preBötC , but not of the RTN , photostimulation of the RTN region in E15 . 5 preparations still evokes ΔF/F changes in the RTN but not in nVII ( left panel ) . Right panel , above , integrated C4 activity showing systematic failures of photostimulation to evoke C4 responses . Below , set of 10 superimposed C4 activity traces . ( G ) Same experiment in Vglut2::ChR2;Dbx1LacZ/LacZ mutants ( n = 5 ) in which the preBötC but not the RTN is disrupted . The mutation completely abrogates light-evoked nVII and C4 responses . DOI: http://dx . doi . org/10 . 7554/eLife . 07051 . 003 Atoh1 null mutants ( hereafter Atoh1−/− ) do not establish a proper respiratory rhythm and die at birth of respiratory failure ( Rose et al . , 2009b ) , but the development of the RTN and CO2 chemosensitivity have not been examined in these mutants . In these and subsequent experiments , we used calcium imaging to monitor the rhythmic activity and response to acidification of embryonic brainstem preparations incubated in artificial cerebrospinal fluid ( a-CSF ) containing either Calcium Green-1AM or Fluo-8 AM as calcium indicators . Consistent with our previous observations ( Dubreuil et al . , 2009; Thoby-Brisson et al . , 2009 ) calcium imaging of E14 . 5 Atoh1+/− brainstem preparations shows rhythmically active RTN neurons that increase their bursting frequencies in response to acidification . These cells were absent in the Atoh1−/− mutants and could not be revealed by low pH ( Figure 2A , B ) . At E16 . 5 , the mutants displayed a slowed-down respiratory-like rhythm in the C4 nerve roots ( by an average of 56% , p < 0 . 001 , n = 11 and 8 for control and mutants , respectively ) and a complete lack of response to acidification ( p = 0 . 1 ) ( Figure 2C , D ) . We then monitored breathing parameters by plethysmography in E18 . 5 pups delivered by Caesarean section ( Figure 2E , F ) . In the mutants , respiratory patterns ranged from hardly any breathing movements to slow rhythmic breathing that were completely unresponsive to hypercapnia ( p = 0 . 96 , n = 31 and 8 for control and mutants , respectively ) ( Figure 2G–I and Table 1 ) . Together , the results show that Atoh1 is essential for the formation of a functional RTN and for CO2 chemical drive to breathe before and at birth . 10 . 7554/eLife . 07051 . 004Figure 2 . Absence of a functional RTN and lack of CO2 chemosensitivity in Atoh1−/− ( Atoh1CreERT2/CreERT2 ) mice . ( A ) Left , ventral view centered on nVII ( dotted outline ) showing Fluo-8 AM fluorescence changes ( ΔF/F ) of RTN cells in an E14 . 5 Atoh1+/− brainstem preparation . Right , RTN cell maps ( top ) and average population activity traces ( bottom ) at pH7 . 4 and pH7 . 2 . The vertical black bar in front of the traces represents a 1% ΔF/F change as indicated . ( B ) Corresponding maps and tracings for an Atoh1−/− embryo attesting to the absence of a functional RTN . ( C ) Quantification of C4 burst frequencies in baseline ( white bar ) and low pH ( gray bar ) conditions for Atoh1+/− ( n = 11 ) and Atoh1−/− ( n = 8 ) E16 . 5 preparations . Baseline frequency dropped by 56% in the mutants and was not affected by acidification . ( D ) Change of C4 frequency ( ΔC4 ) induced by a pH challenge in Atoh1+/− and Atoh1−/− preparations . ( E and F ) Representative examples of plethysmographic recordings of three live E18 . 5 Atoh1+/− and Atoh1−/− pups ( #1–#3 ) delivered by Caesarean section and breathing normal or hypercapnic air as indicated . ( G–I ) Mean values of ventilation ( VE ) ( G ) , breath duration ( TTOT ) ( H ) and tidal volume ( VT ) ( I ) of control ( n = 31 ) and mutant ( n = 8 ) pups breathing normal or hypercapnic air , excluding periods of apnea . DOI: http://dx . doi . org/10 . 7554/eLife . 07051 . 00410 . 7554/eLife . 07051 . 005Table 1 . Relative changes ( % ) in breathing parameters in response to hypercapnia ( 8% CO2 ) measured by plethysmography in E18 . 5 or P0 pups of the indicated genotypesDOI: http://dx . doi . org/10 . 7554/eLife . 07051 . 005GenotypesΔVEp-valueΔTTOTp-valueΔVTp-valuencontrol60 . 9 ± 8 . 9<0 . 001−19 . 5 ± 2 . 8<0 . 00127 . 5 ± 5 . 7<0 . 00151Atoh1−/−−3 . 7 ± 8 . 10 . 536 . 7 ± 10 . 40 . 58−1 . 3 ± 4 . 90 . 628P2b::CreBAC1;Atoh1lox/lox−8 . 0 ± 12 . 80 . 3514 . 6 ± 10 . 60 . 630 . 4 ± 10 . 90 . 5513Atoh1Cre;Phox2blox/lox18 . 7 ± 13 . 50 . 91−1 . 9 ± 8 . 10 . 319 . 2 ± 9 . 20 . 4510P2b::FLPo;Atoh1FRTCre;Vglut2lox/lox20 . 3 ± 8 . 60 . 20−3 . 6 ± 6 . 10 . 3116 . 7 ± 9 . 20 . 2512 The malformation of the RTN in the mutants may underpin the lack of CO2 responsiveness . In Atoh1−/− embryos , most RTN precursors , whether defined by expression of the Atoh1 locus and Phox2b or by co-expression of Lbx1 and Phox2b , fail to migrate to the medullary surface and accumulate instead dorso-laterally of the facial nucleus ( Figure 3A–F , I ) , consistent with previous findings ( Rose et al . , 2009b ) . In the mutants , 83 ± 1 . 5% of the Atoh1-expressing periVII cells were found at a dorsal location compared to 32 ± 4 . 1% in the controls ( p < 0 . 001 , n = 3 ) . By contrast , the cells co-expressing the Atoh1 locus and Phox2b that surround the trigeminal motor nucleus are present in the mutants at their normal location ( p = 0 . 33 , n = 3 ) ( Figure 3G , H , J ) . 10 . 7554/eLife . 07051 . 006Figure 3 . Ventral to dorsal shift of periVII and normal location of periV cells in Atoh1−/− embryos . ( A and B ) Combined in situ hybridization ( ISH ) with a 3′ UTR probe of Atoh1 ( blue ) and immunohistochemistry ( IHC ) with anti-Islet1 , 2 antibodies ( brown ) to visualize nVII on sagittal sections through the hindbrain from E15 . 5 Atoh1+/− and Atoh1−/− embryos . In Atoh1−/− embryos , the neurons identified by expression of the Atoh1 3′ UTR , which would have expressed Atoh1 protein normally , are depleted ventral of nVII visualized by Islet1 , 2 labeling , but accumulate dorsally . ( C and D ) Combined ISH with a 3′ UTR probe for Atoh1 and IHC for Phox2b on sagittal ( C and D ) ( anterior at left ) sections through the medulla of E15 . 5 Atoh1+/− and Atoh1−/− embryos as indicated . The arrow points to the dorsally shifted periVII cells that express the Atoh1 3′ UTR and Phox2b . The PeriVII cells that would have normally expressed the Atoh1 protein are depleted ventral of nVII in the mutants and accumulate dorsally . ( E and F ) Immunofluorescence for Phox2b ( green ) and Lbx1 ( red ) on coronal sections ( lateral at left ) through the medulla of E15 . 5 Atoh1+/− and Atoh1−/− embryos . The Lbx1+/Phox2b+ periVII cells ( yellow ) are depleted ventral of nVII but accumulate dorsally ( arrowheads ) . ( G and H ) Combined ISH with a 3′ UTR probe for Atoh1 and IHC for Phox2b on coronal sections ( lateral at left ) through the pons of E15 . 5 Atoh1+/− and Atoh1−/− embryos showing the normal location of periV cells in the mutants . ( I ) Quantification of the dorsal and ventral population of periVII cells identified by Atoh1 3′ UTR expression at E15 . 5 . Their total numbers were not significantly different from the controls in Atoh1−/− embryos , but 83 ± 1 . 5% are located dorsally in the mutants vs 32 ± 4 . 1% in the controls . ( J ) Quantification of periV cells identified by Atoh1 3′ UTR expression at E15 . 5 . DOI: http://dx . doi . org/10 . 7554/eLife . 07051 . 006 In the brainstem , Atoh1 is expressed in the rhombic lip progenitors and its early differentiating progeny and essential for their specification . A second , independent site of Atoh1 expression is post-mitotic , in cells that co-express the Phox2b transcription factor and surround the trigeminal motor nucleus ( hereafter periV neurons ) and the facial nucleus ( hereafter periVII neurons , which include a majority of the RTN neurons ) ( Dubreuil et al . , 2009; Rose et al . , 2009b ) . The lack of CO2 responsiveness of Atoh1−/− embryos could thus be caused by malfunction of the RTN or some other Atoh1-dependent structure . Similarly , the Phox2b mutant backgrounds analyzed so far , in which the RTN was deleted and CO2/pH chemosensitivity abrogated , suffered to various extents from a lack of specificity . In these mutants , potentially all Phox2b-expressing cells ( Dubreuil et al . , 2008 ) or all those derived from dB2 progenitors ( Dubreuil et al . , 2009 ) or from rhombomeres 3 and 5 ( Ramanantsoa et al . , 2011 ) could have caused the respiratory defect . Others have used local viral injection in combination with a general Phox2b promoter to activate or silence the RTN ( Marina et al . , 2010; Abbott et al . , 2011; Basting et al . , 2015 ) . However , one drawback of this approach is that the nearby Phox2b+ catecholaminergic neurons , which also control breathing but whose connectivity and functions differ from that of the RTN , are transduced as well . We therefore sought to design genetic strategies to make the Atoh1 or Phox2b mutations more selective . In the E15 . 5 brainstem , co-expression of Phox2b and Atoh1 is restricted to periVII and periV neurons ( Dubreuil et al . , 2009; Rose et al . , 2009b ) ( and see below ) . They are the only cells in the brain with this transcriptional signature , since Phox2b and Atoh1 are absent from more rostral regions ( Rose et al . , 2009a; Hirsch et al . , 2013 ) . To monitor these cells specifically , we constructed the Atoh1FRTCre line ( Figure 4A ) , which expresses Cre recombinase from the Atoh1 locus conditionally to the action of FLP recombinase , and partnered it with the P2b::FLPo allele that expresses FLP from the Phox2b promoter ( Hirsch et al . , 2013 ) . In mice harboring the P2b::FLPo and Atoh1FRTCre alleles , Cre recombinase will thus be active selectively in cells with a history of both Phox2b and Atoh1 expression ( Figure 4B ) . Indeed , in triple transgenic P2b::FLPo;Atoh1FRTCre;TauGFPnLacZ embryos , brainstem expression of the TauGFPnLacZ reporter was specific for periV and periVII cells ( Figure 4C , D ) , verifying that these cells are the only ones with a history of Phox2b and Atoh1 expression ( hereafter Phox2bon/Atoh1on cells ) . Importantly , the nearby catecholaminergic neurons , despite their expression of Phox2b , lie outside the Phox2bon/Atoh1on intersectional population and are thus excluded by our approach . 10 . 7554/eLife . 07051 . 007Figure 4 . Targeting of Phox2bon/Atoh1on cells by intersectional genetics . ( A ) Schematic of the Atoh1FRTCre allele generated by homologous recombination in ES cells . The targeted Atoh1 locus contains from 5′ to 3′ the Atoh1 CDS up to and including the stop codon , a neomycin resistance cassette with three consecutive SV40 polyadenylation sequences flanked by FRT sites , an IRES sequence , the Cre recombinase CDS ( orange ) followed by a bovine growth hormone polyadenylation sequence and the Atoh1 3′ UTR . FLP recombinase-mediated recombination will remove the neomycin cassette with the polyadenylation signals allowing for expression of Cre from the Atoh1 locus . ( B ) Left , schematic of the P2b::FLPo ( top ) and Atoh1FRTCre ( middle ) alleles and the Cre recombinase-responsive R26tdTomato indicator allele ( bottom ) . Right , top , sagittal brainstem cartoon schematizing the structures with a history of Phox2b ( green ) or Atoh1 ( blue ) ( Rose et al . , 2009a and data not shown ) expression and the intersectional population expressing or having expressed both ( purple ) . Below , the periV and periVII cells that will express tdTomato in P2bFLPo;Atoh1FRTCre;R26tdTomato embryos ( red ) . The cells expressing FLPo recombinase from the Phox2b promoter will activate Cre expression from the Atoh1 locus in the intersectional population , which in turn will activate tdTomato expression ( or nuclear βgalactosidase expression if TauGFPnLacZ is used as indicator allele ) . EGL , external granular layer , ECN , external cuneate nucleus , LRt , lateral reticular nucleus , NAmb . , nucleus ambiguus , NTS , nucleus of the solitary tract , PN , pontine nuclei . ( C and D ) Combined ISH for Atoh1 and IHC for Phox2b ( C ) or nuclear βgalactosidase ( D ) on coronal sections through the hindbrain from E15 . 5 wild-type ( WT ) ( C ) or E18 . 5 P2b::FLPo;Atoh1FRTCre;TauGFPnLacZ ( D ) embryos . The periV and periVII cells are the only cells in the brainstem co-expressing Phox2b and Atoh1 ( C ) and also the only cells with a history of both Phox2b and Atoh1expression ( D ) . The arrows in panels C and D point to the periVII cells . The insets in panel D show close ups of periV ( top panel ) and periVII cells ( bottom panel ) , double-labeled for Atoh1 and βgalactosidase . DOI: http://dx . doi . org/10 . 7554/eLife . 07051 . 007 In the embryo , the Atoh1-expressing cells in the vicinity of the facial nucleus represent a large fraction of RTN neurons as defined by Phox2b and Vglut2 expression and the absence of catecholaminergic and motoneuronal markers ( 85% , Dubreuil et al . , 2009 ) . However , the functional characteristics of these cells have not been examined . In P2b::FLPo;Atoh1FRTCre;R26tdTomato embryos , Phox2bon/Atoh1on cells specifically express the fluorescent reporter tdTomato . The tdTomato+ cells located ventral of the facial nucleus co-expressed Phox2b but not the motoneuronal marker Islet1 , 2 ( Figure 5A ) . They increased in numbers between E14 . 5 and E18 . 5 , probably because of the time required for full recombination by Cre recombinase expressed from the Atoh1 locus , which starts only at E12 . 5 ( Dubreuil et al . , 2009 ) . We then used calcium imaging to show that all tdTomato+ cells in the RTN region analyzed ( 297/297 cells from 7 different preparations ) were rhythmically active and responded to acidification by increasing their bursting frequencies ( Figure 5B–D ) . Importantly , the burst frequencies and responses to acidification of the tdTomato–positive cells in P2b::FLPo;Atoh1FRTCre;R26tdTomato embryos were identical to those of the global population of oscillating RTN neurons ( p = 0 . 5 and 0 . 7 , respectively ) ( Figure 5C ) . In E14 . 5 hindbrain preparations , tdTomato+ cells recorded in the whole-cell configuration ( n = 4 ) featured spontaneous rhythmic burst discharges of action potentials whose frequency increased at pH 7 . 2 ( Figure 5E , F ) . These spontaneous bursts appeared as all-or-none voltage-dependent events that could be initiated by depolarizing current pulses ( data not shown ) and forced to terminate prematurely by negative current pulses ( Figure 5G ) . Together , these data show that the Phox2bon/Atoh1on cells share the functional signature of embryonic RTN neurons: they oscillate spontaneously and accelerate their rhythm in low pH ( Dubreuil et al . , 2009; Thoby-Brisson et al . , 2009 ) and thus are part of the intrinsically rhythmic network of embryonic RTN cells characterized by Thoby-Brisson et al . ( 2009 ) . 10 . 7554/eLife . 07051 . 008Figure 5 . Functional characterization of Phox2bon/Atoh1on cells . ( A ) Ventral view of triple immunofluorescence for Islet1 , 2 , Phox2b and tdTomato over the facial area in Atoh1FRTCre;P2b::FLPo;R26tdTomato embryos showing Phox2bon/Atoh1on cells ( yellow ) at E14 . 5 ( left ) , E16 . 5 ( middle ) and E18 . 5 ( right ) . ( B ) RTN activity map and right , close up of the boxed area showing Phox2bon/Atoh1on cells ( red ) loaded with Calcium Green-1 AM ( Ca-G ) ( yellow ) . ( C ) Histogram showing that the burst frequencies of the global population of oscillating RTN neurons ( white bars , n = 27 preparations ) are not significantly different from that of the tdTomato+ ( thus Phox2bon/Atoh1on ) cells in Atoh1FRTCre;P2b::FLPo;R26tdTomato embryos ( red bars , n = 6 preparations ) at pH7 . 4 ( 7 . 4 ) and pH7 . 2 ( 7 . 2 ) . ( D ) Superimposed traces showing spontaneous rhythmic Ca-G fluorescence changes of 16 individual Phox2bon/Atoh1on cells ( red traces ) and their average ( black trace ) at pH7 . 4 ( top , 7 . 4 ) and pH7 . 2 ( bottom , 7 . 2 ) . Note the frequency increase in low pH . ( E ) Left panel , ventral view of an E14 . 5 brainstem preparation showing Phox2bon/Atoh1on tdTomato+ RTN neurons ( red ) on either side of the midline ( marked by the auto-fluorescence of the holding mesh ) during an electrophysiology experiment . Middle panel , close up of the boxed area showing Phox2bon/Atoh1on cells ( red ) and right , one biocytin-filled Phox2bon/Atoh1on cell during whole-cell patch-clamp recording . ( F ) Top row , trace of the membrane potential trajectory of the biocytin-labeled cell in E showing spontaneous burst discharges of action potentials ( indicated by * ) at pH7 . 4 ( 7 . 4 ) ; bottom row , same trace at pH7 . 2 ( 7 . 2 ) . Note the increased frequency of bursts at low pH . ( G ) Close up of a burst discharge ( top ) that can be curtailed by negative current pulses ( bottom ) applied 10 ms ( red traces ) , 50 ms ( blue traces ) or 100 ms ( green traces ) after onset of the first action potential of the burst . ( H ) Photostimulation of Phox2bon/Atoh1on cells expressing ChR2 at E15 . 5 evokes C4 motor responses . Left panel , ΔF/F changes in the RTN and adjacent nVII ( dotted line ) following photostimulation of the RTN region . Right , at top , integrated activity of the C4 root showing spontaneous activity and bursts of activity evoked by light pulses ( blue bars ) triggered 4 s after a spontaneous burst . Bottom left , close up of a set of 10 superimposed spontaneous ( Spont . ) C4 bouts of activity and right , a corresponding set of light-evoked ( Light-evoked ) C4 responses , synchronized on the light pulse ( blue bar ) , showing their variable latencies . DOI: http://dx . doi . org/10 . 7554/eLife . 07051 . 008 We then tested the capacity of Phox2bon/Atoh1on cells to elicit C4 activity upon optogenetic stimulation as shown above for the global population of oscillating RTN neurons ( see Figure 1E ) . In P2b::FLPo;Atoh1FRTCre;Ai32 embryos , expression of ChR2-YFP depends on Cre recombinase ( Ai32 line , Madisen et al . , 2012 ) and is thus specific for Phox2bon/Atoh1on cells . Single light pulses of 150 ms duration , delivered to the RTN region in E15 . 5 brainstem-spinal cord preparations at a fixed delay of 4 s after a spontaneous C4 burst , triggered a motor burst response in C4 and facial neurons ( Figure 5H ) in around 70% of the cases ( 130 out of 188 stimulations from 4 preparations ) . The parameters of spontaneous and evoked bursts measured on 150 spontaneous and 123 evoked C4 bursts , respectively , were not significantly different ( spontaneous vs evoked: amplitude: 201 ± 152 arbitrary units ( a . u . ) vs 183 ± 117 a . u . , p = 0 . 08; time to peak: 228 ± 17 ms vs 242 ± 14 ms , p = 0 . 53; half-width: 450 ± 21 ms vs 442 ± 17 ms , p = 0 . 76; decay time: 510 ± 25 ms vs 519 ± 22 ms , p = 0 . 79 ) . The latency of the C4 response and its variability ( 666 ± 20 ms for 123 photostimulations ) , which cannot be easily explained by the time required for polysynaptic propagation of activity , probably reflect the delays caused by processes necessary for burst intitiation in the RTN ( 192 ± 12 ms , see above ) and in the preBötC ( Kam et al . , 2013 ) . Therefore , the Phox2bon/Atoh1on subset of RTN neurons is able to entrain phrenic nerve activity and thus inspiration as previously reported in adult rats of neurons in the RTN region defined by Phox2b expression alone ( Abbott et al . , 2009 , 2011 ) . In the latter experiments , activation of breathing could not be unambiguously traced back to RTN neurons classically defined as non-catecholaminergic and non-cholinergic , since as many nearby catecholaminergic as non-catecholaminergic and some cholinergic neurons also expressed ChR2 ( Abbott et al . , 2011 ) . These results are consistent with our data obtained in reduced preparations . Together , they show that the RTN regulates inspiration , not only expiration as postulated by others ( Janczewski and Feldman , 2006 ) . In fact , opinions vary regarding the role of Phox2b-expressing RTN neurons in active expiration . Abbott et al . ( 2011 ) found that activation of Phox2b+ cells in the RTN region also triggered active expiration and Marina et al . ( 2010 ) that silencing these neurons in the in situ perfused brainstem-spinal cord preparation reduced hypercapnia-evoked expiratory activity . By contrast , Pagliardini et al . ( 2011 ) reported that photostimulation of cells expressing ChR2 from a pan-neuronal promoter in the RTN region generated mainly active expiration with little effect on inspiration . These authors attributed the effect seen in their experiments to neurons that did not express Phox2b and are thus not targeted by our approach . Similarly , Tupal et al . ( 2014 ) reported that in E18 . 5 brainstem-spinal cord preparations , a lumbar motor output and thus active expiration persisted after the depletion of Atoh1 in RTN neurons . A conditional genetic strategy that combines the criteria of Phox2b and Atoh1 expression will target periV in addition to periVII neurons . We found that periV neurons are not essential to CO2 chemosensitivity . To show this , we compared phrenic nerve activity in a standard brainstem-spinal cord preparation with one from which the pons and thus periV cells have been removed by sectioning ( Figure 6A ) . At E16 . 5 , in baseline conditions , sectioning accelerated C4 discharges by suppressing the inhibitory influence on the respiratory rhythm generator ( RRG ) exerted by pontine neurons ( Ito et al . , 2000; Hilaire et al . , 2004 ) . However , the response to a pH challenge was fully preserved ( Figure 6B , C ) ( p = 0 . 1 , n = 34 ) . Hence , at least in vitro , periV neurons do not seem to contribute to the CO2/pH response . Conversely , affecting periVII cells , but leaving periV neurons intact , suffices to eliminate the CO2 response: indeed , our previous results showed that conditional inactivation of Phox2b or activation of the toxic P2b27Alacki allele using Egr2Cre as Cre driver resulted in the lack of responsiveness to CO2 or acidification ( Dubreuil et al . , 2009; Ramanantsoa et al . , 2011 ) , despite the fact that Egr2cre directs recombination in periVII ( Dubreuil et al . , 2009; Ramanantsoa et al . , 2011 ) but not in periV cells ( Figure 6D–G ) . 10 . 7554/eLife . 07051 . 009Figure 6 . PeriV cells are not essential for the respiratory chemoreflex . ( A ) Schematic of a standard brainstem-spinal cord preparation showing the level of the section ( red line ) to eliminate the pons . ( B ) Comparison of C4 burst frequencies before and after sectioning in E16 . 5 wild-type preparations in normal and low pH conditions . The section ( sect ) increases baseline ( white bar ) frequency but preserves the response to acidification ( grey bar ) , *p < 0 . 001 . ( C ) The changes of C4 frequency ( ΔC4 ) induced by the pH challenge before ( cont ) and after sectioning ( sect ) are not significantly different from each other . ( D and E ) Combined ISH for Atoh1 and IHC for nuclearly localized βgalactosidase on coronal sections through the E15 . 5 pons , in the indicated genotypes . With Egr2 as Cre driver , expression of the TauGFPnLacZ reporter allele is not activated in periV cells ( D ) , but it is expressed there when Cre is provided by Atoh1cre ( E ) . ( F and G ) Combined ISH for Atoh1 and IHC for Phox2b on coronal sections through the E15 . 5 pons in the indicated genotypes . In the presence of Egr2Cre , Phox2b expression by the Atoh1+ cells is preserved in Phox2blox/lox embryos ( F ) , but not when Cre is provided by Atoh1Cre ( G ) . PeriV cells are not in the Egr2 lineage and can thus not be affected in Egr2Cre;Phox2blox/lox or Egr2Cre;P2b27Alacki mice , in which the respiratory chemoreflex is abrogated . DOI: http://dx . doi . org/10 . 7554/eLife . 07051 . 009 We restricted the Atoh1 null mutation to Phox2bon/Atoh1on cells by partnering a floxed Atoh1 locus with Cre expressed from the Phox2b promoter ( P2b::CreBAC1 ) ( D'Autreaux et al . , 2011 ) . Like Phox2b , Cre was switched on in the dB2 progenitor domain ( Figure 7A ) from which RTN neurons arise ( Dubreuil et al . , 2009; Hirsch et al . , 2013 ) . In P2b::CreBAC1; Atoh1lox/lox embryos , Atoh1 expression should be lost selectively in cells that express or have expressed Phox2b , that is , in Phox2bon/Atoh1on cells . Indeed , Atoh1 expression was abrogated in the ventrolateral medulla at E15 . 5 and as early as E12 . 5 ( Figure 7B , C ) attesting to an efficient recombination . Predictably , Atoh1 expression was preserved outside of the Phox2b lineages , that is , in the rhombic lip and the migrating cells of the anterior extramural stream . In P2b::CreBAC1; Atoh1lox/lox embryos , as in Atoh1−/− embryos , the normally Atoh1-expressing periVII cells that can be detected in these mutants by double Phox2b/Lbx1 labeling ( Pagliardini et al . , 2008; Rose et al . , 2009b ) were quantitatively preserved ( p = 0 . 11 ) . However , their distribution was massively shifted from ventral to dorsal ( 75 ± 2 . 6% dorsally located Phox2b+/Lbx1+ cells in the mutants vs 25 ± 2 . 3% in the controls , p < 0 . 001 , n = 3 and 4 for controls and mutants , respectively ) and the ventral Phox2bon/Atoh1on RTN neurons correspondingly depleted , in agreement with previous results ( Huang et al . , 2012 ) ( Figure 7D–F ) . 10 . 7554/eLife . 07051 . 010Figure 7 . Selective removal of Atoh1 in RTN precursors produces a ventral to dorsal shift of periVII neurons . ( A ) Combined Phox2b ( red ) and YFP ( green ) staining on a transverse section of an E12 . 5 P2b::CreBAC1;R26YFP ( Srinivas et al . , 2001 ) hindbrain showing efficient recombination of the reporter allele in dB2 progenitors . In the dA3 domain , where Phox2b and Cre are switched on postmitotically , newly born cells are still YFP− but express it during their ventral migration . ( B and C ) Combined Atoh1 ISH with a CDS probe of Atoh1 and Phox2b IHC on coronal hindbrain sections of P2b::CreBAC1;Atoh1lox/+ ( B ) or P2b::CreBAC1;Atoh1lox/lox embryos ( C ) at E15 . 5 or E12 . 5 as indicated . Cre recombinase expressed from the Phox2b promoter removes Atoh1 message in the RTN precursors but not in the rhombic lip ( RL ) or the cells of the anterior extramural stream ( AES ) of Atoh1lox/lox embryos . Higher magnifications of the boxed areas are shown below . ( D and E ) Immunofluorescence for Lbx1 ( red ) and Phox2b ( green ) on coronal E15 . 5 hindbrain sections of the indicated genotypes ( dorsal at top , lateral on the left ) . The double-labeled periVII cells are in yellow , the insets show close ups corresponding to the boxed areas . The Lbx1+/Phox2b+ cells accumulate dorso-laterally of nVII in P2b::CreBAC1;Atoh1lox/lox mutants . ( F ) Quantification of the Lbx1+/Phox2b+ periVII cells at E15 . 5 located ventrally ( Vent ) or dorsally ( Dors ) of nVII . In P2b::CreBAC1;Atoh1lox/+ controls , 25 ± 2 . 3% of the periVII neurons are located dorso-laterally of nVII vs 75 ± 2 . 6% in P2b::CreBac1;Atoh1lox/lox mutants . DOI: http://dx . doi . org/10 . 7554/eLife . 07051 . 010 In P2b::CreBAC1;Atoh1lox/lox mutants , we looked for cells displaying the functional hallmarks of the embryonic RTN: rhythmic activity , acid-sensitivity and entrainment of the RRG ( Thoby-Brisson et al . , 2009 ) . In E14 . 5 brainstem preparations , calcium activity in the RTN region of the conditional mutants was similar to that observed in Atoh1−/− embryos: few active cells ( 8 ± 2 cells , n = 5 vs 119 ± 11 in the controls , n = 13 ) , absence of synchronized rhythmic fluorescent changes and unresponsiveness to acidification ( Figure 8A , B ) . At E16 . 5 , the frequency of C4 discharges was reduced as in other mutants with impaired RTN function ( Dubreuil et al . , 2009; Ramanantsoa et al . , 2011 ) ( by an average of 49% , p < 0 . 001 , n = 12 and 15 for mutants and controls , respectively ) ( Figure 8C , D ) . The mutant preparations retained a residual response to a pH challenge , which , however , was much attenuated ( an increase in C4 frequency by 51 ± 10% above the pH 7 . 4 value vs 106 ± 17% in the controls , p < 0 . 001 , n = 12 and 15 for mutants and controls , respectively ) and remained below the baseline value of the controls ( Figure 8C , D ) . 10 . 7554/eLife . 07051 . 011Figure 8 . Functional consequences of selective inactivation of Atoh1 in Phox2bon cells . ( A and B ) RTN activity maps and average population activity traces at E14 . 5 in control and P2b::CreBAC1;Atoh1lox/lox mutants at pH7 . 4 and pH7 . 2 showing absence of a functional RTN in the mutant . ( C ) Quantification of C4 burst frequencies in baseline ( white bars ) and low pH ( gray bars ) conditions in control and mutant E16 . 5 preparations . Baseline frequency dropped by 49% in the P2b::CreBac1;Atoh1lox/lox mutants ( lox/lox ) and increased at pH 7 . 2 by 51 ± 10% above the pH7 . 4 value vs 106 ± 17% in the controls ( cont ) . ( D ) Corresponding change of C4 frequency ( ΔC4 ) induced by a pH challenge in P2b::CreBAC1;Atoh1lox/lox ( lox/lox ) compared to control ( cont ) preparations . ( E ) Left panel , photostimulation of the RTN region at E15 . 5 evokes ΔF/F changes in the RTN and adjacent nVII ( dotted line ) in Vglut2::ChR2;P2b::CreBAC1;Atoh1lox/+ preparations . Right panel , above , integrated C4 activity showing spontaneous bursts and bursts evoked by light pulses ( blue bars ) triggered 4 s after a spontaneous burst . Bottom left , close-up of a set of 10 superimposed spontaneous ( Spont . ) C4 bouts of activity and right , a corresponding set of light-evoked ( Evoked ) C4 responses , synchronized on the light pulse ( blue bar ) that appeared with variable latencies after the stimulus . ( F ) In P2b::CreBAC1;Atoh1lox/lox mutants , ΔF/F responses to light are limited to a few cells in the RTN and nVII ( left panel ) . Right panel , at top , integrated activity of C4 showing the reduced frequency of the C4 rhythm and systematic failures of photostimulation ( blue bars ) to evoke C4 responses; bottom , close-up of spontaneous C4 bouts of activity ( Spont . ) and lack of response to light ( blue bar ) . DOI: http://dx . doi . org/10 . 7554/eLife . 07051 . 011 We then examined the C4 response to photostimulation of the RTN region in E15 . 5 brainstem-spinal cord preparations from mutants harboring the ChR2-YFP allele expressed from the Vglut2 promoter . We introduced the ChR2-YFP allele ( Vglut2::ChR2-YFP ) ( Hägglund et al . , 2010 ) in a P2b::CreBAC1;Atoh1lox/lox background and its P2b::CreBAC1;Atoh1lox/+ control . In E15 . 5 control preparations , stimulation of the RTN region by single light pulses of 150 ms duration , delivered at a fixed delay of 4 s after a spontaneous burst ( n = 147 photostimulations in 6 preparations ) , systematically elicited a synchronized calcium response in the RTN neurons , followed by a motor burst in the facial nucleus and the C4 roots ( Figure 8E ) . Here again , the latency of the C4 response and its variability ( 623 ± 16 ms for 147 photostimulations ) reflect probably the engagement of processes necessary for burst intitiation in the RTN and the preBötC . A comparison of spontaneous and evoked motor burst parameters , measured on 150 spontaneous and 147 evoked C4 bursts , showed that the amplitudes and kinetics of spontaneous and evoked motor bursts were not significantly different ( spontaneous vs evoked amplitude: 310 ± 11 a . u . vs 317 ± 13 a . u . , p = 0 . 33; time to peak: 194 ± 9 ms vs 202 ± 12 ms , p = 0 . 61; half-width: 446 ± 15 ms vs 454 ± 17 ms , p = 0 . 73; decay time: 487 ± 16 ms vs 509 ± 20 ms , p = 0 . 40 ) . In the P2b::CreBAC1;Atoh1lox/lox background , the same light stimulus resulted in the activation of only a few RTN neurons and completely failed to trigger bursts in C4 and the facial nucleus ( n = 317 photostimulations in 6 preparations ) ( Figure 8F ) . Therefore , the residual Phox2b and Lbx1 co-expressing RTN precursors that reach the medullary surface ( see Figure 7F ) are unable to mount a motor response . Reasoning that the dorsally misrouted periVII cells may underlie the residual chemoresponse in E16 . 5 preparations , we compared the functional status of ventral and dorsal cells in E15 . 5 slice preparations ( Figure 9A ) . In control preparations , the full complement of rhythmic cells revealed by their calcium activity was located ventral of the facial nucleus . In the P2b::CreBAC1;Atoh1lox/lox mutants , by contrast , rhythmic cells were massively depleted ventrally , but a population of neurons with synchronized rhythmic activity was now found dorso-laterally of the facial nucleus ( n = 5 ) ( Figure 9B , C ) . The dorsally located mutant periVII cells were thus distinct from wild-type dorsal cells and functionally resembled wild-type ventral cells . The ectopic periVII cells in the P2b::CreBAC1;Atoh1lox/lox mutants accelerated their rhythm in low pH conditions ( Figure 9C , E ) . They had conserved their glutamatergic nature ( Figure 9F ) and could thus mediate the attenuated C4 response to acidification in the brainstem-spinal cord preparations . In Atoh1−/− embryos ( n = 5 ) , the dorso-laterally located rhythmic cells still responded to low pH but were fewer and less synchronized ( Figure 9D , E ) , which may explain the complete absence of a response . 10 . 7554/eLife . 07051 . 012Figure 9 . The dorsally misplaced periVII cells in P2b::CreBAC1;Atoh1lox/lox embryos are rhythmic and respond to a pH challenge . ( A ) Cartoon of the brainstem ( top ) showing medullary slice preparation ( delimited by vertical red lines ) used for calcium imaging of dorsally misplaced mutant cells ( bottom , black box ) . ( B and C ) Above , ΔF/F changes and activity maps and below , superimposed traces showing spontaneous rhythmic fluorescence changes ( traces corresponding to individual periVII cells in black and average trace in red ) at pH7 . 4 ( top , 7 . 4 ) and pH7 . 2 ( bottom , 7 . 2 ) in E15 . 5 transverse slices from control ( B ) or P2b::CreBAC1;Atoh1lox/lox ( C ) preparations . The rhythmic cells are found ventrally in the controls and are massively displaced dorsally in the mutants . ( D ) Same experiment for Atoh1−/− preparations , showing fewer rhythmic cells that are less well synchronized . In panels B–D , the matched colors of the circles and the outlines of nVII and the medullary surface represent individual experiments ( B , n = 9; C , n = 5; D , n = 5 ) . ( E ) Histograms showing that the frequency of the oscillations of periVII cell is not significantly different at pH7 . 4 ( 7 . 4 , white bars ) between controls ( CTL ) and P2b::CreBAC1;Atoh1lox/lox or Atoh1−/− mutants and is increased by acidification to pH7 . 2 ( 7 . 2 , gray bars ) in all cases . ( F ) Combined fluorescent ISH for Vglut2 ( red ) and immunofluorescent staining of Phox2b ( blue ) and Lbx1 ( green ) on coronal sections of an E15 . 5 P2b::CreBAC1;Atoh1lox/lox hindbrain showing Vglut2 expression by the dorsally misplaced Phox2b+/Lbx1+ cells; ( e–e′′′ ) , close ups of the boxed area showing the overlay ( e ) , Phox2b ( e′ ) , Lbx1 ( e′′ ) and Vglut2 ( e′′′ ) expression . DOI: http://dx . doi . org/10 . 7554/eLife . 07051 . 012 Finally , we examined by plethysmography the newborn and adult P2b::CreBAC1;Atoh1lox/lox animals , which survived ( around 60% ) . In newborn mutants at the day of birth ( P0 ) , mean ventilation ( VE ) in normal air was reduced by 16% , ( from 22 . 8 ± 0 . 3 to 19 . 2 ± 0 . 6 μl/s/g , p < 0 . 001 , n = 13 and 31 for mutants and controls , respectively ) . Hypercapnic air increased VE of the controls by 60 ± 13% ( p < 0 . 001 ) but had no effect on the mutants ( p = 0 . 34 ) ( Figure 10A and Table 1 ) , and this was true also at P2 . 5 ( Figure 10B–D ) ( p = 0 . 18 , n = 9 and 17 , for mutants and controls , respectively ) . Adult mutants breathing hypercapnic air responded with a 30% increase in VE , corresponding to an average of 42% recovery of the CO2 response with respect to the adult controls ( p < 0 . 001 ) . 10 . 7554/eLife . 07051 . 013Figure 10 . Defective CO2 chemosensitivity after inactivation of Atoh1 in Phox2bon cells . ( A ) Plethysmographic recording of P0 pups . Shown are the mean values of VE in air or in response to 8% CO2 in P2b::CreBAC1;Atoh1lox/lox mutants ( black circles ) and controls ( empty circles ) . Inset , ventilatory response to hypercapnia expressed as the percentage change relative to baseline for controls ( white bar ) and mutants ( black bar ) . ( B–D ) Mean relative changes of ventilation ( VE ) ( B ) , breath duration ( TTOT ) ( C ) and tidal volume ( VT ) ( D ) measured in P0 ( control n = 31; mutant N = 13 ) , P2 . 5 ( n = 17; N = 9 ) and P60 ( n = 19; N = 11 ) mice breathing normal or hypercapnic air ( 8% CO2 ) . *p < 0 . 001 , n . s . = p > 0 . 05 ( hypercapnic vs room air ) . DOI: http://dx . doi . org/10 . 7554/eLife . 07051 . 013 In conclusion , Atoh1 inactivation in Phox2bon cells disrupts the RTN anatomically and functionally , severely curtails the CO2/pH response in embryonic brainstem preparations and abrogates it at birth , with only a partial recovery at later stages . RTN neuron development depends on Phox2b ( Dubreuil et al . , 2009 ) . In a second strategy to genetically impair the RTN , we targeted the inactivation of Phox2b to Atoh1on cells by partnering a floxed Phox2b locus ( Phox2blox , Coppola et al . , 2010 ) with Cre expressed constitutively from the Atoh1 promoter by the Atoh1Cre allele that was generated by germ-line excision of the STOP cassette in Atoh1FRTCre by FLP recombinase ( see Figure 4A ) . In embryos homozygous for the Phox2blox allele and expressing Cre from the Atoh1 locus , Phox2b will be inactivated selectively in periVII and periV cells . A caveat is that Atoh1 and thus Cre are switched on in the migrating RTN precursors only at E12 . 5 ( Dubreuil et al . , 2009 ) , well after Phox2b , which is already expressed in their progenitors ( Hirsch et al . , 2013 ) . Therefore , early differentiation of RTN neurons is expected to proceed normally in Atoh1cre;Phox2blox/lox embryos and functional defects to appear only during late gestation . At E15 . 5 , most RTN neurons , identified by Atoh1 expression and their location ventral of the facial nucleus , had already lost Phox2b , but their number had dropped by only 34 ± 6% ( p = 0 . 013 , n = 4 ) ( Figure 11A–C ) . In brainstem-spinal cord preparations from E16 . 5 Atoh1Cre;Phox2blox/lox mutants , the baseline frequency of C4 discharges was reduced by an average of 56% ( p < 0 . 001 ) , but the response to hypercapnia was preserved ( p = 0 . 002 , n = 5 and 12 for mutants and controls , respectively ) ( Figure 11D , E ) . 10 . 7554/eLife . 07051 . 014Figure 11 . Loss of RTN neurons and the respiratory chemoreflex after selective inactivation of Phox2b in Atoh1on cells . ( A , B , F , G ) Combined ISH for Atoh1 ( blue ) and IHC for Phox2b ( brown ) on coronal hindbrain sections from E15 . 5 ( A and B ) and E18 . 5 ( F and G ) embryos of the indicated genotypes . The insets show high magnifications corresponding to the boxed areas . ( C and H ) Counts of the Atoh1+ periVII cells in control ( Atoh1Cre;Phox2blox/+ ) and mutant ( Atoh1Cre;Phox2blox/lox ) embryos at E15 . 5 ( C ) and E18 . 5 ( H ) . Total periVII cells represent 66 ± 6% of the controls at E15 . 5 ( n = 4 ) and drop to 38 ± 2% at E18 . 5 ( n = 3 ) , while most Atoh1+ cells have lost Phox2b expression already at E15 . 5 . There was no ventral to dorsal shift of periVII cells in the Phox2b mutants ( 26 ± 2% and 21 ± 0 . 3% dorsally located cells in controls and mutants , respectively ) . ( D ) Quantification of C4 burst frequencies in baseline ( white bar ) and low pH ( gray bar ) conditions in control ( cont ) and Atoh1Cre;Phox2blox/lox ( lox/lox ) E16 . 5 preparations . Baseline frequency dropped by 56% in the Atoh1Cre;Phox2blox/lox mutants . It increased by 127% above the pH7 . 4 value in the mutants vs 135% in the controls . ( E ) Corresponding change of C4 frequency ( ΔC4 ) induced by a pH challenge in Atoh1Cre;Phox2blox/lox ( lox/lox ) compared to control ( cont ) preparations . ( I ) Plethysmographic recording of P0 pups . Shown are the mean values of VE in air or in response to 8% CO2 in Atoh1Cre;Phox2blox/lox mutants ( black circles ) ( n = 10 ) and controls ( empty circles ) ( n = 10 ) . The inset represents the ventilatory response to hypercapnia expressed as the percentage change relative to baseline for controls ( white bar ) and mutants ( black bar ) showing the lack of a significant effect of hypercapnia . DOI: http://dx . doi . org/10 . 7554/eLife . 07051 . 014 At E18 . 5 , however , the number of Atoh1+ RTN neurons in the mutants had dropped by 62 ± 2% ( p = 0 . 001 , n = 3 ) ( Figure 11F–H ) , and the newborn Atoh1Cre;Phox2blox/lox pups were completely unresponsive to hypercapnia ( p = 0 . 91 , n = 10 ) while baseline ventilation was reduced by 24% ( p < 0 . 001 , from 24 . 4 ± 0 . 6 μl/s/g in the control , n = 10 to 18 . 4 ± 0 . 5 μl/s/g in the mutants , n = 10 ) . Thus inactivating Phox2b in periVII and periV cells eliminates the CO2 response at birth . Previous data suggested that RTN neurons are potentially glutamatergic and may thus activate the RRG by releasing glutamate ( Mulkey et al . , 2004; Bochorishvili et al . , 2012 ) . Proper functioning of the RRG requires glutamatergic synaptic transmission since mice with germ-line inactivation of Vglut2 , in which the synaptic release of glutamate is blocked , die at birth from respiratory failure ( Wallen-Mackenzie et al . , 2006 ) . To restrict Vglut2 inactivation to Phox2bon/Atoh1on cells , we partnered the floxed Vglut2 locus ( Vglut2lox ) with the P2b::FLPo;Atoh1FRTCre genotype ( Figure 12A ) . In mice homozygous for Vglut2lox and harboring at the same time the P2b::FLPo and Atoh1FRTCre alleles , Vglut2 inactivation will be targeted selectively to periVII and periV cells . The baseline ventilation of P2b::FLPo;Atoh1FRTCre; Vglut2lox/lox pups was only slightly reduced ( by 9%; from 22 . 0 ± 0 . 7 μl/s/g in the control , n = 9 , to 20 . 2 ± 0 . 6 μl/s/g in the mutants , p < 0 . 001 , n = 12 ) , but their hypercapnic response was abolished ( p = 0 . 2 , n = 12 ) ( Figure 12B and Table 1 ) . Thus the RTN mediation of the CO2 response is glutamatergic and the chemoreflex abrogated by blocking glutamatergic transmission specifically in Phox2bon/Atoh1on cells . 10 . 7554/eLife . 07051 . 015Figure 12 . Disrupting the synaptic release of glutamate in Atoh1on/Phox2bon cells suppresses the CO2 response . ( A ) Schematic of the P2b::FLPo ( above ) and Atoh1FRTCre ( middle ) alleles and the floxed Vglut2 locus ( Wallen-Mackenzie et al . , 2006 ) ( below ) . The Phox2b lineage cells expressing FLPo recombinase from the Phox2b promoter will activate Cre expression from the Atoh1 locus selectively in Phox2bon/Atoh1on cells . Inactivation of the floxed Vglut2 locus will thus be restricted to Phox2bon/Atoh1on cells . ( B ) Plethysmographic recording of P0 Vglut2lox/lox;Atoh1FRTCre;P2b::FLPo pups showing slightly slowed-down baseline ventilation in air and lack of response to 8% CO2 ( black circles ) ( n = 12 ) compared to controls ( empty circles ) ( n = 9 ) . The inset shows the ventilatory response to hypercapnia expressed as the percentage change of VE relative to baseline for mutants ( black bar ) and controls ( white bar ) . DOI: http://dx . doi . org/10 . 7554/eLife . 07051 . 015 In the mammalian CNS , localized networks of interneurons play critical roles in controlling behavior . It has , however , proved difficult to pinpoint the contribution of individual classes of neurons to defined behaviors . One example is the RTN . The RTN neurons have been implicated in breathing regulation by CO2 , but their importance for the respiratory chemoreflex in the intact brain remains controversial ( Huckstepp and Dale , 2011; Nattie , 2011; Guyenet et al . , 2013 ) . Here we reexamined this issue using novel genetic backgrounds inspired by prior evidence that RTN neurons express and require the Phox2b and Atoh1 transcription factors ( Dubreuil et al . , 2009; Rose et al . , 2009b ) . We first established that the Phox2bon/Atoh1on subset of RTN neurons are ( i ) rhythmically active , ( ii ) are activated by low pH , and ( iii ) both necessary and sufficient for optogenetic entrainment of the phrenic nerve . These neurons thus have the functional signatures of RTN neurons ( Abbott et al . , 2009; Thoby-Brisson et al . , 2009; Ramanantsoa et al . , 2011 ) . We then show that an Atoh1 constitutive null mutation and three conditional mutations specifically targeting Phox2bon/Atoh1on neurons and disrupting RTN formation or function abolish the CO2-evoked ventilatory response in neonates . Finally , we provide evidence that periV neurons , the only other group of neurons targeted by the conditional mutations , are not involved in the chemoreflex . We conclude that the neurons in the RTN region that have a history of both Phox2b and Atoh1 expression are obligatory elements of the circuitry for breathing regulation by CO2 in neonates . While our phlethymosgraphic data in adults is in line with those of Huang et al . ( 2012 ) , our in vitro data on embryos contrasts with theirs and we find a lack of chemoreflex at birth . Using a distinct Phox2b::Cre line to delete Atoh1 in Phox2b lineage cells , Huang et al . reported a slowed down basal respiratory rhythm , a hallmark of RTN impairment ( Dubreuil et al . , 2009; Ramanantsoa et al . , 2011 ) , but a fully preserved response to acidification in brainstem-spinal cord preparations at late gestation . They did not test the chemoreflex in behaving pups , but it is only in the latter that we find a complete abrogation of the reflex . A possible explanation is that in vitro the RTN precursors misrouted to the dorsal site of the facial nucleus in the conditional Atoh1 mutants are able to mount a motor response to bath acidification whose magnitude may depend on the precise experimental conditions . In support of this , we found that the ectopic dorsal cells in the mutants respond to low pH by accelerating their firing rate . Our results demonstrate that the blunted CO2 response at later postnatal stages in P2b::CreBAC1;Atoh1lox/lox mice is a recovery from an absent chemoreflex at birth . We found a very similar recovery of the chemoreflex in adult Egr2Cre/+;Phox2b27Alacki/+ mutants ( Ramanantsoa et al . , 2011 ) . The partial recovery of the CO2 response in adults could be due to late compensation by residual RTN neurons , to a strengthening of peripheral chemoreceptor input that bypasses the RTN ( Basting et al . , 2015 ) or to some of the multiple sites proposed to function as CO2 sensors in the brain which would now also bypass the RTN ( Nattie and Li , 2009 ) . The lack of the chemoreflex in many human CCHS patients ( Straus et al . , 2010; Weese-Mayer et al . , 2010 ) suggests that these putative compensatory sites also have a history of Phox2b expression . Our findings are compatible with two explanations for the loss of the chemoreflex: the Phox2b+/Atoh1+ RTN neurons could be the main CO2 sensors in the brain or obligatory relays funneling the chemosensory input from other cells to the RRG—or both . On the one hand , RTN neurons are exquisitely sensitive to small changes in CO2/pH ( Mulkey et al . , 2004; Onimaru et al . , 2008 ) , also when synaptically isolated ( Wang et al . , 2013 ) and , in close apposition to numerous capillaries ( Onimaru et al . , 2012a ) , well positioned to sense PCO2 in the blood . On the other , they receive afferences from many brainstem sites that contain putative chemosensors ( Rosin et al . , 2006 ) and respond with depolarization to activation of nearby acid-sensitive astrocytes ( Gourine et al . , 2010 ) . In addition , RTN neurons receive excitatory connections from the carotid bodies via the nucleus of the solitary tract and increase their firing rates in response to carotid body stimulation by hypoxia ( Takakura et al . , 2006 ) . The carotid bodies are not only the main PO2 sensors , but provide also a substantial fraction of the overall CO2 response ( Forster et al . , 2000; Smith et al . , 2006 ) . Still , we observed a complete loss of the chemoreflex in neonates with impaired RTN function . The implication is that in the perinatal period , the RTN is an obligatory relay for all forms of respiratory chemoreception whether mediated by central or peripheral chemoreceptors . Because of its extensive connections with other chemosensory sites and the observation that its stimulation activates breathing , the RTN has been suggested to be a crucial hub for respiratory chemoreception ( Guyenet et al . , 2013 ) . However , the results of previous loss-of-function experiments to assess the part played by RTN neurons in the chemoreflex circuitry are not entirely conclusive . In previous work , we have monitored the chemoreflex in mice in which subsets of Phox2b expressing neurons including the RTN were genetically lesioned . In these genetic backgrounds , RTN neurons were deleted and the chemoreflex lost , but the effect could not be unambiguously attributed to the RTN ( Dubreuil et al . , 2008 , 2009; Ramanantsoa et al . , 2011 ) . Others have impaired RTN function in behaving adult rats using pharmacological or pharmacogenetic approaches , but similar data are not available for the newborn period . Bilateral lesions of the neurokinin1 receptor-expressing neurons in the RTN region by injection of saporin conjugated to a substance P analogue have produced variable results , ranging from a modest attenuation of the chemoreflex in an earlier study ( Nattie and Li , 2002 ) to a substantial decrease in recent work ( Takakura et al . , 2014 ) . Silencing the neurons in the RTN region by the GABAA agonist muscimol also yielded a large reduction of the hypercapnic response ( Takakura et al . , 2013 ) . These experiments , however , lack specificity and the extent of the lesion or inhibition is difficult to control . Marina et al . ( 2010 ) used a pharmacogenetic approach to silence RTN neurons . Administration of allatostatin to rats which had received local injections of a lentivirus vector expressing the allatostatin receptor from an artificial Phox2b promoter reduced the hypercapnic response by an average of 60% . However , the approach used by Marina et al . targets the neurons in the rostral ventro-lateral medulla defined by Phox2b expression alone . They include C1 adrenergic and A5 noradrenergic neurons ( Stornetta et al . , 2006 ) , whose contribution to the observed effect was not assessed . RTN neurons express the vesicular glutamate transporter Vglut2 and establish synapses with preBötC neurons that resemble classic glutamatergic synapses ( Mulkey et al . , 2004; Bochorishvili et al . , 2012 ) , but this does not prove that they use glutamate and not other transmitters such as galanin ( Bochorishvili et al . , 2012 ) to excite RRG neurons . Removal of Vglut2 and thus glutamatergic synaptic transmission selectively from Phox2bon/Atoh1on neurons affected baseline ventilation only very slightly but eliminated the chemoreflex at birth . Therefore , glutamatergic excitatory drive by Phox2bon/Atoh1on neurons underlies the respiratory chemoreflex while normocapnic ventilation seems to depend less on it . A consensus about the sites and circuits that underlie the chemosensory control of breathing has yet to emerge . Our results establish that the Phox2bon/Atoh1on RTN neurons , that is , the RTN neurons that have expressed Atoh1 in the embryo , lie at the core of the chemoreflex circuitry . They are absolutely necessary for the ventilatory response to hypercapnia at birth and are still the major contributors in the adult . Irregular or instable respiration at birth is characteristic for preterm infants and not uncommon in babies that are born at term ( Gaultier and Gallego , 2005 ) . The key role of the RTN for breathing regulation at birth that we show here suggests that defective development or immaturity of the human equivalent , apart from causing CCHS ( Amiel et al . , 2003 ) , could also underlie more common respiratory problems in the newborn period . The following mouse lines were used in this study:Atoh1FRTCre . To generate the Atoh1FRTCRE allele , we inserted a neomycin resistance cassette with three consecutive SV40 polyadenylation sequences flanked by FRT sites at the 3′ of the Atoh1 CDS , which was followed by an IRES sequence and the Cre recombinase CDS ( Figure 4A ) . The targeting vector was assembled in the K667 plasmid and electroporated into BD10 ( MCI-C57BL/6N Tac ) ES cells . Correctly targeted ES cells were injected into C57BL/6 blastocysts . Construction of the targeting vector , ES cell manipulation and blastocyst injection were done by the Mouse Clinical Institute ( Illkirch , France ) . The Atoh1FRTCre mice were genotyped for the presence of Cre by PCR ( forward primer TGATGGACATGTTCAGGGATC , reverse primer GAAATCAGTGCGTTCGAACGCTAG ) . Atoh1Cre expressing Cre recombinase from the Atoh1 locus constitutively was derived from Atoh1FRTCre by germ-line expression of FLPe ( Rodriguez et al . , 2000 ) . The Atoh1cre mice were genotyped for the presence of Cre and the absence of the neomycin gene ( using GATCTCCTGTCATCTCACCT and ATGGGTCACGACGAGATCCT as PCR primers ) . Atoh1CreERT2 . In the Atoh1CreERT2allele , the Atoh1 CDS is replaced by the CreERT2 sequence thus creating a null allele ( Fujiyama et al . , 2009 ) , preserving transcription of the Atoh1 3′ UTR . P2b::CreBAC1 , a BAC transgenic line expressing Cre recombinase from the Phox2b promoter in a pattern mirroring expression of the endogenous Phox2b gene ( D'Autreaux et al . , 2011 ) . Atoh1lox , harboring a floxed Atoh1 locus generating a null allele upon Cre recombinase action ( Shroyer et al . , 2007 ) . The Atoh1 3′ UTR is not expressed from the recombined Atoh1lox allele . TauGFPnLacZ , previously termed TaumGFP ( Hippenmeyer et al . , 2005 ) and R26tdTomato ( Ai9 , Madisen et al . , 2010 ) , floxed STOP reporter lines expressing GFP and nuclear βgalactosidase or tdTomato , respectively , in response to Cre recombinase action . P2b::FLPo , a BAC transgenic line , expressing FLPo recombinase from the Phox2b promoter in a manner mirroring expression of the endogenous Phox2b gene ( Hirsch et al . , 2013 ) . Ai32 expressing a channelrhodopsin2-YFP fusion protein driven by the CAG promoter in the Rosa locus ( Madisen et al . , 2012 ) in response to Cre recombinase action . Vglut2::ChR2 , a BAC transgenic line expressing a channelrhodopsin2-YFP fusion from the Vglut2 promoter ( Hägglund et al . , 2010 ) . Phox2blox harboring a floxed Phox2b locus generating a null allele upon Cre recombinase action ( Coppola et al . , 2010 ) . Vglut2lox harboring a floxed Vglut2 locus generating a null allele upon Cre recombinase action ( Wallen-Mackenzie et al . , 2006 ) . Egr2Cre expressing Cre from the Egr2 locus ( Voiculescu et al . , 2000 ) . All lines have been crossed for at least four generations with C57BL/6 × DBA/2 F1 mice and were maintained on this background . As there were no significant differences seen between mice that had only one copy of the floxed Atoh1 , Phox2b or Vglut2 alleles or lacked Cre and wild-type mice , these types of littermates were grouped together as controls , except when stated otherwise . All animal studies were done in accordance with the guidelines issued by the European Community and have been approved by the research ethics committees in charge ( Comités d'éthique pour l'expérimentation animale ) and the French Ministry of Research . The methods for immunofluorescence , combined bright field or fluorescent in situ hybridization ( ISH ) and immunohistochemistry ( IHC ) have been described ( Dubreuil et al . , 2000; Hirsch et al . , 2007; Dubreuil et al . , 2008 ) . Riboprobes for Atoh1 or Vglut2 were synthesized using a DIG RNA labelling kit ( Roche , Manheim , Germany ) as specified by the manufacturer . Atoh1 probes were derived either from the Atoh1 3′ UTR to reveal the cells that would have expressed Atoh1 protein normally in Atoh1CreERT2/CreERT2 embryos or from the Atoh1 CDS in all other instances . The Vglut2 signal was revealed by Cy5-tyramide working solution ( Perkin Elmer , Waltham , MA ) . The primary antibodies used were: rabbit anti-Phox2b ( Pattyn et al . , 1997 ) , chicken anti-GFP ( Aves Lab , Tilgard , OR ) , rabbit anti-βgalactosidase ( Cappel , Santa Ana , CA ) , guinea pig anti Lbx1 ( Muller et al . , 2002 ) and mouse anti-Islet1 , 2 ( Developmental Studies Hybridoma Bank , Iowa City , IA ) . They were revealed for fluorescent staining by Alexa 488- ( Invitrogen , Carlsbad , CA ) or by Cy3- or Cy5-labelled ( Jackson Immunoresearch , Suffolk , UK ) secondary antibodies of the appropriate specificity , for bright field observation by biotin-labeled secondary antibodies and Vectastain ABC kit ( Vector Laboratories , Peterborough , UK ) revealed with 3 , 3′-diaminobenzamide . The pictures were captured with either a Hamamatsu ORCA-ER or a Leica DFC420C camera mounted on a Leica DM5500B microscope for observation through fluorescence or bright field optics , respectively . Brightness and contrast were adjusted in Adobe Photoshop uniformly across entire images maintaining signal linearity . PeriVII neurons , defined as Phox2b+/Atoh1+ or Phox2b+/Lbx1+ cells surrounding the facial nucleus , were counted throughout an area starting rostrally with the first section containing facial nucleus neurons and stopping 70 μm caudal to the caudal end of the nucleus . PeriV neurons were defined as Atoh1+ cells surrounding the trigeminal motor nucleus . Cell numbers represent bilateral cell counts in a 1:4 series of sections multiplied by 4 . Statistical analysis of cell counts was done using a two-tailed t-test and XLSTAT software with alpha set at 0 . 05 and the results expressed as mean ± SEM . The methods used for preparing brainstem–spinal cord and transverse slice preparations from embryonic day 14 . 5 ( E14 . 5 ) –16 . 5 mouse embryos and maintaining them in oxygenated a-CSF have been described ( Dubreuil et al . , 2009; Thoby-Brisson et al . , 2009; Bouvier et al . , 2010 ) . Briefly , brainstem-spinal cord and slice preparations were dissected in 4°C a-CSF of the following composition ( in mM ) : 128 NaCl , 8 KCl , 1 . 5 CaCl2 , 1 MgSO4 , 24 NaHCO3 , 0 . 5 Na2HPO4 , 30 glucose , pH 7 . 4 . E15 . 5 transverse slices for optically recording of dorsally located periVII neurons were obtained by cutting 100 μm thick sections from caudal to rostral until a slice contained unambiguously the facial nucleus . Then a 600 μm slice was made and its caudal surface imaged . For calcium imaging , brainstem-spinal cord or slice preparations were incubated at room temperature for 40–45 min in oxygenated a-CSF containing the cell-permeant calcium indicator dyes Calcium Green-1 AM ( 10 μM; Life Technologies , Paisley , UK ) , or Fluo-8 AM ( 10 μM; Teflabs , Austin , TX ) . Preparations were then transferred to a recording chamber ( 30°C ) and let to recover for 30 min prior to start optical recordings using a conventional epifluorescence configuration with a FITC filter cube . Fluorescence images were captured from the slice surface and from the ventral surface of brainstem-spinal cord preparations exposing the RTN region , with a cooled Neo sCMOS camera ( Andor Technology Ltd . , Belfast , UK ) using 10× , 20× and 40× objectives , an exposure time of 100 ms and bin size of 4 × 4 for periods of 180 s using Micro-Manager software ( https://www . micro-manager . org/wiki/ ) . When calcium imaging was combined with photostimulation of Vglut2::ChR2-YFP preparations , the calcium imaging back-illumination was set at a minimal level using neutral density filters in the light-path , which caused no noticeable changes in the rhythmic activity of the RTN . Relative fluorescence changes ( ΔF/F ) in response to photostimulation in Figures 1E–G , 5H , 8E , F show the ΔF/F image calculated by averaging a 1 s sequence acquisition ( 10 frames ) starting at the offset of the light stimulus . Whole-cell patch-clamp recordings were performed under visual control using differential interference contrast and infrared video microscopy , a MultiClamp 700B amplifier , a digitizing interface Digidata 1550 and the software program pClamp10 ( all from Molecular Devices , Sunnyvale , CA ) . Patch electrodes pulled from borosilicate glass tubes ( GC 150 TF; Clark Electromedical Instruments , Hamden , CT ) were filled with a solution containing the following ( in mM ) : 123 K-gluconic acid , 21 KCl , 0 . 5 EGTA , 3 MgCl2 , 10 HEPES , pH 7 . 2 , supplemented with 1 mg/ml biocytine ( Molecular Probes , Eugene , OR ) and had a resistance of 4–6 MΩ . Biocytin-labeled neurons were revealed with Extravidine-FITC ( 1:400; Sigma-Aldrich , Saint-Quentin Fallavier , France ) . Phrenic nerve activity was recorded on E16 . 5 brainstem-spinal cord preparations using suction electrodes positioned on the fourth cervical root ( C4 ) as described ( Thoby-Brisson et al . , 2009; Bouvier et al . , 2010 ) . Raw signals were amplified ( High-gain AC , 7P511 , Grass Technologies , Warwick , RI ) , filtered ( bandwidth 0 . 1–3 kHz ) , integrated ( time constant 50 ms , Neurolog System , Digitimer Ltd , Hertfordshire , UK ) before digital sampling at 6 kHz and analysis using pClamp9 ( Molecular Devices ) . Values are given as mean + SEM . Statistical significance was tested using a difference Student's t-test to compare data sets obtained from different mutants and a paired difference Student's t-test to compare the measurements obtained in two different conditions . In optogenetic experiments , photostimulation was provided by either a 473 nm laser ( Ikecool Corporation , Anaheim , CA ) connected to an optical fiber placed over the RTN region on one side ( Figure 1A ) , to illuminate a 250 × 500 μm elliptic area encompassing the RTN ( laser power set at 8 mW ) or by digital holography ( Lutz et al . , 2008 ) to target the RTN region defined by rhythmic calcium activity maps generated on-line ( laser power density 1–5 mW/mm2 ) . Individual light pulses of 70 ms duration were delivered at a fixed delay of 4 s following a spontaneous C4 motor burst detected by a threshold device , or were triggered randomly in C4 silent preparations . Breathing variables of E18 . 5–P60 animals were measured non-invasively in unanaesthetized , unrestrained animals using whole-body barometric plethysmography ( Chatonnet et al . , 2007 ) . E18 . 5 pups delivered by Caesarean sections or P0 pups were placed under a heating lamp and were gently touched for 10 min until their breathing had stabilized . After a 7 min familiarization period in the plethysmograph chambers , breathing parameters ( breath duration [TTOT] , tidal volume [VT] , and ventilation [VE] calculated as VT/TTOT ) were continuously monitored in apnea-free periods . E18 . 5 to P2 . 5 animals were placed in a 30 ml chamber and their ventilation was recorded continuously for a period of 800 s composed sequentially of a 200 s exposure to normal air followed by a 200 s exposure to hypercapnic air ( 8%CO2/21%O2/71%N2 ) before normal air was resumed for the remaining 400 s . Adult animals were recorded in a larger chamber , first for 100 s in normal air , then the animals were exposed to the hypercapnic mix for 3 min before initiation of a 100 s period of recording prior to returning to normal air breathing . Calibrations were performed at the end of each recording session by injecting 2 . 5–5 μl of air in the chamber with a Hamilton syringe . In plots showing the continuous evolution of respiratory parameters before , during and after the hypercapnic challenge , each circle represents the mean , across all animals of a given genotype , of the parameter values binned in 20 s intervals . To quantify breathing parameter changes induced by hypercapnia , we first calculated for each animal the ratio of the mean parameter value during the last 100 s of the hypercapnic period over that during the last 100 s of the preceding normocapnic period , and then calculated the mean of these ratios in the population of control and mutant animals and expressed them as percentage ( ΔVE graphs and Table 1 ) . Values are given as mean + SEM . Statistical significance was tested using a difference Student's t-test to compare data sets obtained from different mutants and a paired difference Student's t-test to compare the measurements obtained in two different conditions . The time-series acquisitions were analyzed using a custom-made extended fork of the public domain ImageJ software with the open-source code deposited in a sourceforge . net repository ( PhysImage [http://physimage . sourceforge . net/] ) . To determine the changes in calcium activity we first calculated a ΔF/F0 time-series using a plugin of PhysImage that implements a running moving average for baseline subtraction . This tool calculates F0 by taking the fluorescence image ( F ) at each time-step ( t ) as follows: ΔF ( t ) /F0 = ( F ( t ) − F0 ( t ) ) /F0 ( t ) , where F0 ( t ) =∑i=ti+wF ( i ) w and w = 100 and the last w frames of the time-series are dropped . To identify rhythmic RTN cells and establish RTN maps of activity we performed cycle-triggered averages ( CTAs ) using the anatomical location of the RTN to define the region of interest ( ROI ) that represents broad RTN activity . Using the built-in ImageJ Z-project function , we calculated the image representing the standard deviation ( SD ) of the CTA and used an iterative thresholding algorithm to identify potential ROIs representing cells ( Hayes et al . , 2012 ) . A trace representing each ROI's activity was extracted and then the ROIs were validated by calculating the power spectral density of these traces . If the peak power of each trace was not in the range 0 . 1–0 . 4 Hz , the corresponding ROI was discarded . For mutant backgrounds with no discernable RTN activity , the same strategy of detection was used except that the SD image of the full ΔF/F0 time-series was taken into account rather than SD of the image after CTA .	An adult at rest will typically breathe in and out up to 20 times per minute , inhaling oxygen and exhaling carbon dioxide in a process that , for the most part , occurs automatically . While we can choose to override this process and exert voluntary control over our breathing , we cannot suppress it indefinitely . Attempting to do so will ultimately trigger a reflex that forces us to start breathing again . This reflex is mostly a response to the rise of carbon dioxide ( CO2 ) in the blood , which lowers the pH of the blood . This rise in CO2 is toxic and triggers an increase in breathing so that the excess CO2 is exhaled . The majority of the sensors that detect CO2 are in the brainstem , which is at the junction of the brain and the spinal cord . However , the precise location of these sensors is not clear . Ruffault et al . now argue that the sensors are in a region called the ‘retrotrapezoid nucleus’ , and that they can be identified by the presence of two proteins , Atoh1 and Phox2b . In the brains of foetal mice , Ruffault et al . recorded cells in the retrotrapezoid nucleus and found that they fired in a rhythmic pattern , as would be expected for cells that control breathing . Moreover , the firing rate of these cells increased when the pH was lowered . Ruffault et al . then created genetically modified mice with mutations in genes for Atoh1 or Phox2b . The retrotrapezoid nucleus was either absent or abnormal in these mutant mice . Moreover , new-born pups with these mutations were not able to increase their breathing when the level of CO2 in their blood rose . These results shed light on the respiratory distress experienced by patients with a rare disorder called congenital central hypoventilation syndrome ( CCHS ) that is caused by mutations in Phox2b . More commonly , unstable or irregular breathing is seen in human infants that are born prematurely , and sometimes in infants born at full term . In the light of the new findings by Ruffault et al . , it is possible that abnormal development or immaturity of the retrotrapezoid nucleus is the cause .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "neuroscience" ]	2015	The retrotrapezoid nucleus neurons expressing Atoh1 and Phox2b are essential for the respiratory response to CO2
Liquid-liquid phase separation ( LLPS ) has been recognized as one of the key cellular organizing principles and was shown to be responsible for formation of membrane-less organelles such as nucleoli . Although nucleoli were found to behave like liquid droplets , many ramifications of LLPS including nucleolar dynamics and interactions with the surrounding liquid remain to be revealed . Here , we study the motion of human nucleoli in vivo , while monitoring the shape of the nucleolus-nucleoplasm interface . We reveal two types of nucleolar pair dynamics: an unexpected correlated motion prior to coalescence and an independent motion otherwise . This surprising kinetics leads to a nucleolar volume distribution , p⁢ ( V ) ∼V-1 , unaccounted for by any current theory . Moreover , we find that nucleolus-nucleoplasm interface is maintained by ATP-dependent processes and susceptible to changes in chromatin transcription and packing . Our results extend and enrich the LLPS framework by showing the impact of the surrounding nucleoplasm on nucleoli in living cells . The nucleolus is the largest structure present in the cell nucleus of eukaryotic cells . This membraneless organelle is a site of ribosomal biogenesis and plays a key role in cell cycle progression and stress response ( Alberts et al . , 2014; Montanaro et al . , 2008; Boulon et al . , 2010 ) . Nucleoli are composed of RNA and proteins and embedded in the chromatin solution inside the nucleus . They form at specific parts of genome called nucleolar organizer regions ( NORs ) containing rDNA , which is transcribed inside the nucleolus ( McClintock , 1934; Ritossa and Spiegelman , 1965; Wallace and Birnstiel , 1966 ) . At the beginning of the cell cycle a small nucleolus forms at each NOR . These nucleoli later fuse into larger ones , while remaining connected to their NORs in somatic cells ( Amenta , 1961; Sullivan et al . , 2001 ) . The lack of a nucleolar membrane has long been intriguing biologists and physicists alike , questioning the physical nature of the nucleolus . Pioneering studies in frogs found that nucleoli in X . laevis oocytes behave like liquid droplets in vivo , as well as when reconstituted in vitro , and suggested that nucleoli form through liquid-liquid phase separation of the nucleolar components in the nucleoplasm ( Brangwynne et al . , 2011; Berry et al . , 2015; Feric et al . , 2016 ) . The volume distribution of such nucleoli was in agreement with a diffusion-limited aggregation process with a constant influx of particles ( Brangwynne et al . , 2011 ) . In addition , the size of nucleoli in the worm C . elegans embryos was found to be dependent on the concentration of nucleolar components in the nucleoplasm which is consistent with the liquid-like nature of the nucleolus ( Weber and Brangwynne , 2015 ) . The nucleolar subcompartments , that is the granular and the dense fibrillar components , were also suggested to form via liquid-liquid phase separation ( Feric et al . , 2016 ) . Recent studies in the fly D . melanogaster suggest that while the nucleolar assembly follows liquid-liquid phase separation , active protein recruitment is also involved ( Falahati and Wieschaus , 2017 ) . Recently , we have shown that human nucleoli also exhibit liquid-like behavior ( Caragine et al . , 2018 ) . By analyzing the shape fluctuations of nucleolar surface and kinetics of the nucleolar fusion in human cells in vivo , we found nucleolar dynamics to be consistent with that of liquid droplets with very low surface tension γ ~ 10-6 Nm-1 surrounded by highly viscous nucleoplasm of viscosity η ~ 103 Pa s ( Caragine et al . , 2018 ) . Strikingly , it is the nucleoplasm viscosity that sets the time scale for the nucleolar coalescence providing resistance to the already very low surface tension that drives the process ( Caragine et al . , 2018 ) . Correspondingly , nucleolar coalescence in human cells takes hours to complete ( until the newly formed nucleolus rounds up , Figure 1A ) , while the neck connecting two coalescing nucleoli is discernable only for minutes after their initial touch ( Figure 1B ) and its radius r grows in time as r ( t ) ∼t1/2 ( Caragine et al . , 2018 ) . Such long coalescence times have been speculated not to interfere with the rDNA transcription inside the nucleoli ( Caragine et al . , 2018 ) . The nucleoplasm ( chromatin solution ) and its physical properties clearly contribute to the nucleolar physiology . Interestingly , while the nucleolar coalescence can be described by a theory of passive liquid droplets within a highly viscous passive fluid ( Caragine et al . , 2018; Paulsen et al . , 2014 ) , nucleoplasm is an active fluid . Specifically , chromatin dynamics was shown to be active , that is ATP-dependent , and coherent , that is exhibiting correlated displacements , over 3–5 µm in human cells ( Zidovska et al . , 2013 ) . Thus , the measured γ and η are likely effective quantities ( Caragine et al . , 2018 ) . Chromatin is known to localize as a denser heterochromatin at the nucleolar surface ( Padeken and Heun , 2014 ) , yet the nature of physical interactions between the nucleolar surface and the chromatin solution remains to be revealed ( Németh and Längst , 2011; Bickmore and van Steensel , 2013; Towbin et al . , 2013 ) . Disruption and dysfunction of the nucleolus is implicated in a large number of human diseases , such as skeletal and neurodegenerative disorders , cardiovascular disease and cancer ( Hannan et al . , 2013; Núñez Villacís et al . , 2018; Ruggero and Pandolfi , 2003; Derenzini et al . , 2009 ) . Thus , elucidating physical principles governing the nucleolus-nucleoplasm interface might contribute to our understanding of the nucleolus in health and disease . In this work , we investigate the physical interactions between the nucleoli and the surrounding nucleoplasm by studying the structural features and dynamical behavior of the nucleoli . Specifically , to illuminate the kinetics of nucleolar assembly process , we examine the changes in the nucleolar size distribution with progressing cell cycle . In addition , we probe the physical nature of the nucleolar subcompartments , specifically , the granular components and the dense fibrillar components , and their contribution to the nucleolar liquid-like properties . To elucidate the role of nucleoplasm in nucleolar coalescence , we interrogate size , shape , position and alignment , as well as mobility inside the nucleus for both nucleoli that are about to fuse as well as those that do not fuse . To determine the role of active processes in maintaining the liquid-like nucleolus-nucleoplasm interface , we deplete ATP and further evaluate its structure and dynamics . Finally , we probe the contribution of specific cellular processes ( such as cytoskeletal forces , transcriptional activity as well as protein synthesis ) to maintaining the nucleolus-nucleoplasm interface by employing targeted biochemical perturbations . To address the kinetics of the nucleolar assembly process , we have evaluated the number and size of the nucleoli at different times during the cell cycle . After mitosis , human nuclei initially contain 10 nucleoli , which later fuse to form fewer larger ones ( Savino et al . , 2001 ) . Thus , due to the changing nucleolar number , the likelihood of their coalescence is expected to vary with the cell cycle progression . First , we measure the nucleolar size distribution in an unsynchronized cell population , which contains cells at all cell cycle stages ( Figure 2A ) . Then we obtain the specific nucleolar size distributions at different , well-defined times of the cell cycle by synchronizing the cell population and monitoring their nucleolar count and size with progressing cell cycle ( Figure 2B–D ) . Specifically , we carry out our measurements 1 . 5 hr and 3 hr after mitosis as well as at the end of the cell cycle , at the G2/M check point ( Figure 2B–D ) . At every time point , we collect data from the entire volume of the cell nucleus by taking a z-stack with focal planes 0 . 5 μm apart . Figure 2A–D shows micrographs of nuclei with fluorescently labeled chromatin ( H2B-GFP ) and nucleoli ( NPM-DsRed ) for all studied populations , respectively . Moreover , Figure 2 , insets 1–4 , shows an enlarged view of the boxed in nucleus from Figure 2A–D , respectively . While Figure 2 , inset 1 depicts a nucleus from an unsynchronized cell population , Figure 2 , insets 2–3 show the same nucleus with progressing time . Note , the presence of both small and large nucleoli early in the cell cycle ( Figure 2 , inset 2–3 ) , with the large ones becoming more spherical between 1 . 5 hr and 3 hr after mitosis , while only large nucleoli are seen at the end of the cell cycle ( Figure 2 , inset 4 ) . Figure 2E shows the distributions of average nucleolar area of nucleoli in one nucleus , ⟨AN⟩ , as a function of the nucleolar number in the given nucleus , NN , for the unsynchronized and synchronized cell populations at the studied time points . The distributions of nucleolar area , AN , for each time point are shown in Figure 2—figure supplement 1 . For each nucleolus we measure its area in its respective focal plane within the collected z-stack . We find that as the cell cycle progresses , the number of nucleoli per nucleus decreases , while the average nucleolar area in the nucleus increases ( Figure 2E ) . Interestingly , this trend persists beyond 3 hr into the cell cycle suggesting that the fusion of nucleoli is not limited to the first two hours of the cell cycle as previously hypothesized ( Savino et al . , 2001 ) . To gain further mechanical insight into the nucleolar coalescence kinetics during the cell cycle , we have analyzed the nucleolar volume distribution for each time point ( Figure 2F ) . We calculated nucleolar volume assuming a spherical shape , VN=4⁢π⁢r3/3 , where r is the radius of a circle with the area equal to the nucleolar area , and using the least square method we fitted the nucleolar volume distribution P⁢ ( VN ) to a power law f ( VN ) ∼VNα . Our data shows that P⁢ ( VN ) can be described by a power law with α ~ -1 for all cell populations , unsynchronized as well as synchronized at all studied time points . The confidence intervals for the fitting parameter α are listed in Figure 2F with the goodness-of-fit R2 > 0 . 98 for all fits . It is noteworthy , that such distribution is divergent , and so is its first moment , the mean , if integrated over all volumes ( from 0 to ∞ ) . However , the measured p⁢ ( V ) distribution does have finite bounds given by the physical cut-offs for the nucleolar size , the minimum and maximum that it can reach inside a cell nucleus . The human nucleolus behaves like liquid droplet ( Caragine et al . , 2018 ) , yet the nucleolar fluid is complex , containing three distinct subcompartments; fibrillar center ( FC ) , dense fibrillar component ( DFC ) and granular component ( GC ) . They all play a different role in ribosome biogenesis and vary in protein composition: While FC contains polymerase I , DFC and GC are enriched in fibrillarin ( FBL ) and nucleophosmin ( NPM ) , respectively ( Boisvert et al . , 2007 ) . Moreover , they show a hierarchical organization , suggested to form via liquid-liquid phase separation ( Feric et al . , 2016 ) , with FCs nested inside DFCs , which are embedded in GC . To address the contributions of these subcompartments to the overall liquidity of the human nucleolus , we examine their physical properties . Figure 3A shows micrographs of three different nuclei with fluorescently labeled chromatin ( H2B-GFP , green ) , GC ( NPM-DsRed , red ) and DFCs ( FBL-mCerulean , blue ) . We obtain the nuclear and nucleolar contours from H2B-GFP and NPM-DsRed signal , respectively . By analyzing the FBL-mCerulean signal we procure the shape , size and number of DFCs inside a nucleolus . A visual inspection of our data reveals that DFCs appear to be close to spherical . To verify this observation , we measure the DFC eccentricity: First , we measure the length of the semi-major DFC axis a and the semi-minor DFC axis b ( Materials and methods ) . Figure 3B displays the distributions of measured lengths of both a ( red ) and b ( green ) , together with the Gaussian fits f ( aDFC ) ∼e ( aDFC−⟨aDFC⟩ ) 2/2σaDFC2 ( red line ) and f ( bDFC ) ∼e ( bDFC−⟨bDFC⟩ ) 2/2σbDFC2 ( green line ) of their respective distributions . From the Gaussian fits we obtain the following average values: ⟨aD⁢F⁢C⟩= 210 ± 50 nm and ⟨bD⁢F⁢C⟩= 180 ± 40 nm . Next , we evaluate the eccentricity e=a/b for each DFC and find that the DFC shape is indeed close to spherical with the average eccentricity ⟨e⟩= 1 . 22 ± 0 . 17 ( where e = 1 corresponds to a circle ) and average area of ⟨AD⁢F⁢C⟩= 0 . 13 ± 0 . 06 µm2 , where AD⁢F⁢C=π⁢a⁢b . The distributions of e and AD⁢F⁢C are shown in Figure 3—figure supplement 1 . Overall , we identified 1279 DFCs over 114 nucleoli in 63 nuclei , and after the removal of the DFCs that were out of focus , we obtain measurements of a , b , e , and ADFC for 1035 DFCs . Next , we evaluate the nucleolar area , AN , as a function of the DFC number , ND⁢F⁢C , inside the given nucleolus ( Figure 3C ) . Our data reveals that AN grows linearly with ND⁢F⁢C , with a linear fit of AN = ( 0 . 92 ± 0 . 05 ) NDFC . This implies that upon nucleolar coalescence , which leads to larger AN , the new nucleolus contains the cumulative number of DFCs , indicating that DFCs do not fuse themselves . This is further corroborated by the volume distribution of DFCs , p⁢ ( VD⁢F⁢C ) , which has a sharp peak at VDFC = 0 . 03 μm3 ( Figure 3—figure supplement 1 ) , indicating that DFCs are largely monodisperse . Moreover , this finding suggests that every DFC is associated with a GC domain of an area AGC ≈ 0 . 79 µm2 . Since we found DFCs to exhibit a close to spherical shape , we can estimate the volume fraction of DFCs and GC phase in the human nucleolus , and find ΦDFC ≈ 0 . 1 and ΦGC ≈ 0 . 9 , respectively . Our recent study revealed that the timescale of the nucleolar coalescence is set by the high viscosity of the surrounding nucleoplasm ( ηnp ~ 103 Pa s ) ( Caragine et al . , 2018 ) . To elucidate the physical interactions of nucleolar droplets with the chromatin solution , we interrogate their size , shape , position and alignment inside the cell nucleus . Moreover , we compare these characteristics for nucleoli that fuse and the ones that do not fuse during our observation . For non-fusing nucleoli , we record time lapses for 60 min with a time step of 5 min and at every time step we collect a z-stack with focal planes 0 . 5 µm apart . By collecting a z-stack , we can monitor all nucleoli present in the given nucleus and obtain measurements in their respective focal planes . To capture a fusion of nucleoli , we observe a pair of nucleoli for 60–270 min with a time step of 5–15 min , and review at the end of the experiment if the fusion has occurred . In three cases , we were able to track three or four nucleoli simultaneously , the nucleoli closest together were then defined as pairs . In case of three nucleoli only one pair was analyzed , that is two closest nucleoli . In one case , a nucleolar pair fused while the measurement was being set up , and was therefore only analyzed in the post-fusion nucleolar population . Figure 4A shows micrographs of a nucleus with fluorescently labeled chromatin ( H2B-GFP ) at t = 0 and 60 min , where the nucleoli correspond to the voids in the H2B-GFP signal and are highlighted by symbols ( circle and triangle ) . In contrast , Figure 4D shows micrographs of nucleus with fluorescently labeled chromatin ( H2B-GFP ) at t = 0 , 60 and 120 min , with fusion occurring shortly before t = 60 min . The nucleoli correspond to the voids in the H2B-GFP signal and are highlighted by triangle and cross before fusion and by circle during and after fusion . Next , we obtain contours for all nucleoli in their respective z-plane and measure their area , AN , by filling their contour . To evaluate the nucleolar shape we compute its eccentricity , e=a/b , with a and b being the semi-major and semi-minor axes of a fitted ellipse , respectively . For e = 1 , the nucleolus is spherical , while for e > 1 the nucleolus has an elliptical shape . Further , we determine the shortest distance of the nucleolar centroid to the nuclear envelope , De , as well as the angle between the nuclear and nucleolar major axes , α , when fitted by an ellipse , respectively . Figure 4B provides an illustration of the measured parameters a , b , De and α . First , we evaluate these quantities for the non-fusing nucleoli ( Figure 4C ) . We find that AN , e and De do not change appreciably , while α fluctuates significantly during the duration of the experiment . In fact , a constant area might indicate that there is no significant addition or removal of nucleolar material during this time . The eccentricity is rather low , often close to 1 , making α susceptible to small fluctuations . For comparison , Figure 4E shows the same quantities for the nucleoli that fused during the experiment . We aligned the different fusion events in time with fusion occurring at t = 0 min , as marked by the red dashed line . The pairs of fusing nucleoli are identified by symbols of the same color , with two nucleoli before fusion indicated by a triangle and cross , and the new nucleolus after fusion by a circle . Interestingly , AN upon fusion is about same as the summed area of both pre-fusion nucleoli . Since we do not observe any significant increase of AN after fusion , there is likely no significant material influx associated with nucleolar fusion . As expected , e decreases after fusion for all nucleoli , consistent with our prior finding of surface tension driving the fusion ( Caragine et al . , 2018 ) . De does not show any leading trends for the position of the new nucleolus; some remain closer to a position of one of the pre-fusion nucleoli , whereas some move into an intermediate De of the two pre-fusion nucleoli . Interestingly , our α measurement shows that nucleoli after fusion are slowly moving into a parallel ( α= 0° ) or a perpendicular ( α= 90° ) alignment with the major nuclear axis . To further elucidate the nucleolus-nucleoplasm interactions , we investigate the dynamics of both non-fusing and fusing nucleoli . We track the nucleolar motion , by tracking its centroid in time , and obtain a trajectory for every nucleolus . By analyzing nucleolar trajectories , we can extract the nucleolar velocity , v , which we measure relative to the nuclear centroid . Further , we evaluate the radial velocity , vr⁢a⁢d , which we define as the nucleolar velocity along the line connecting centroids of two nucleoli , with origin being the line center . vr⁢a⁢d allows us to assess the motion of one nucleolus towards another , where a negative value of vr⁢a⁢d corresponds to motion towards the other nucleolus . We also compute the angle , αv , between the nucleolar velocity v and the line connecting the two nucleoli , with αv ranging from 0° to 180° , informing if a nucleolus is traveling towards the other . Figure 5A shows examples of trajectories for non-fusing nucleoli , with their temporal evolution color-coded ( blue to red ) . An enlarged view of these trajectories and the areas they cover is depicted in Figure 5—figure supplement 1A–B . A distribution of nucleolar velocities v for non-fusing nucleoli is shown in Figure 5B , with a mean velocity ⟨v⟩ ≈ ( 0 . 49 ± 0 . 30 ) × 10-3 μms-1 . Next , we look at the dynamic behavior of a nucleolar pair , and analyze their velocities with respect to each other . We plot the larger velocity , vm⁢a⁢x , against the smaller velocity , vm⁢i⁢n ( Figure 5C ) . The scatter plot in Figure 5C shows a wide spread and no apparent correlation between vm⁢a⁢x and vm⁢i⁢n ( Pearson correlation coefficient ρ= 0 . 41 ) . Moreover , a distribution of vr⁢a⁢d is centered around 0 , thus not pointing towards any preferred direction of motion ( Figure 5D ) . This observation is further corroborated , when we find vr⁢a⁢d to fluctuate around 0 as a function of time ( Figure 5E ) , as well as αv changing seemingly randomly with time ( Figure 5F ) . By fitting a Gaussian curve to the vr⁢a⁢d distribution in Figure 5D we obtain a variance σrad , nonfusing = 3 × 10-4 μms-1 . Next , we analyze the motion of pairs of fusing nucleoli in the same fashion . Examples of such trajectories , with their temporal evolution color-coded ( blue to red ) , are shown in Figure 5G . An enlarged view of these trajectories and the areas they cover is depicted in Figure 5—figure supplement 1C–D . Figure 5H shows the distribution of the nucleolar velocities , v , for fusing nucleoli , with an average velocity ⟨v⟩≈ ( 0 . 33 ± 0 . 26 ) × 10-3 μms-1 , which is ∼ 30% smaller than for non-fusing nucleoli . This difference is statistically significant as corroborated by p-value of 4 × 10-4 . Strikingly , when we review velocities of a fusing nucleolar pair and plot the larger velocity , vm⁢a⁢x , against the smaller velocity , vm⁢i⁢n ( Figure 5I ) , we find a clear linear correlation between them , with the linear fit of vmax = ( 1 . 74 ± 0 . 20 ) vmin ( Figure 5I , blue line ) and a Pearson correlation coefficient ρ= 0 . 88 . Moreover , we find that ⟨vmax/vmin⟩ ≈ 1 . 8 ± 0 . 6 . The distribution of vr⁢a⁢d for fusing nucleoli ( Figure 5J ) is still centered around 0 , but is clearly narrower than for non-fusing nucleoli ( Figure 5D ) with a variance σrad , fusing = 1 × 10-4 μms-1 obtained by fitting a Gaussian curve to the vr⁢a⁢d distribution in Figure 5J . σr⁢a⁢d , f⁢u⁢s⁢i⁢n⁢g is about three times smaller than σr⁢a⁢d , n⁢o⁢n⁢f⁢u⁢s⁢i⁢n⁢g . When reviewed over time , vr⁢a⁢d exhibits much smaller fluctuations ( Figure 5K ) than in case of non-fusing nucleoli ( Figure 5E ) . Lastly , Figure 5L shows αv of the pre-fusion nucleoli as a function of time , monitoring 40 min prior to fusion , which occurs at t = 0 min . Interestingly , the nucleoli seem not to follow any preferred direction , but instead follow a zig-zag motion while approaching each other to fuse . To investigate a possible role of active ( ATP-dependent ) processes in maintaining the nucleolus-nucleoplasm interface , we have examined nucleoli , specifically , their shape , surface roughness and possible fusion events , upon ATP depletion . The ATP was depleted using 2-deoxyglucose and trifluoromethoxy-carbonylcyanide phenylhydrazone ( see Materials and methods ) . Figure 6A and B show micrographs of nuclei with fluorescently labeled chromatin ( H2B-GFP , green ) and nucleoli ( NPM-DsRed , red ) under physiological conditions ( control ) and upon ATP depletion , respectively . In addition , we review the z-projection of nuclear and nucleolar contours obtained in different planes of the respective z-stack ( Figure 6A–B ) . We find that upon ATP depletion nucleoli do not exhibit spherical , but instead irregular shapes ( Figure 6B ) . In fact , some of the larger irregularly shaped nucleoli might originate from nucleoli fusing at the time of ATP depletion , given the absence of nucleoli in the hour-glass shape , characteristic of nucleolar fusion under physiological conditions ( Figure 1 ) . To characterize the morphological changes of nucleoli upon ATP depletion , we define parameters describing their shape and compare against the control nucleoli . Specifically , after we obtain nuclear and nucleolar contours in their respective focal planes , we compute the following six parameters for every nucleolus: AN/AN⁢u⁢c ( the nucleolar area AN normalized by the nuclear area AN⁢u⁢c ) , e ( the eccentricity e=a/b , where a and b are the semi-major and semi-minor nucleolar axis , respectively ) , α ( the angle between the nuclear and nucleolar major axes ) , de ( the shortest distance from the nucleolar centroid to the nuclear envelope normalized by the nuclear circumference in the focal plane of the nucleus ) , fn⁢e⁢g ( the fraction of the nucleolar contour with negative curvature ) and Nn⁢e⁢g ( the number of independent nucleolar contour regions with negative curvature ) , where the curvature corresponds to the in plane curvature of the nucleolar contour . The comparison of these parameters for nucleoli under physiological conditions and upon ATP depletion is shown in Figure 6C . Table 1 provides a summary of means and standard deviations for measured distributions of AN/AN⁢u⁢c , e , α , de , fn⁢e⁢g and Nn⁢e⁢g . In addition , we evaluated the p-values for all measured physical quantities as well as the relative differences of their means with respect to control ( Table 1 ) . The relative difference ( in % ) of the means was calculated as 100 × [ ( μQ−μP ) /μP] , where μP is the mean of the probability distribution of the measured physical quantity under control conditions and μQ after the perturbation . Furthermore , we computed the skew of the measured distributions , which informs about their asymmetry , and the Kullback-Leibler divergence with respect to control ( Table 1 ) . The Kullback-Leibler divergence is a measure of the difference between two probability distributions . It is defined as ∑P⁢ ( i ) ⁢ln⁡ ( P⁢ ( i ) Q⁢ ( i ) ) , where P⁢ ( i ) and Q⁢ ( i ) are the two distributions . Here , P⁢ ( i ) corresponds to the probability distribution of the measured physical quantity under control conditions , Q⁢ ( i ) after the perturbation . The most striking change that we observe upon the ATP depletion is the irregularity of the nucleolar shape . The dramatic increase in the nucleolar surface roughness upon ATP depletion is nicely captured by the growing amount of the nucleolar contour possessing negative curvature as quantified by fn⁢e⁢g and Nn⁢e⁢g , both showing increase of ∼ 20% . Remarkably , when compared to the control nucleoli , the ATP-depleted nucleoli not only show larger parts of their contour to posses negative curvature , but are also more likely to contain several more independent contour regions of negative curvature making them appear lobulated . In addition , the ATP-depleted nucleoli tend to localize further away from the nuclear envelope , as illustrated by de , than the control nucleoli . Surprisingly , there are no significant changes to the average nucleolar area , eccentricity and orientation as per AN/AN⁢u⁢c , e and α , respectively , upon ATP depletion . To probe contributions of specific cellular processes ( such as cytoskeletal forces , transcriptional activity as well as protein synthesis ) to maintaining the nucleolus-nucleoplasm interface , we employ targeted biochemical perturbations . Specifically , to inhibit cytoskeletal forces we treat the cells with blebbistatin , which is a myosin II inhibitor , latrunculin A , which prevents actin polymerization , and nocodazole , which is a microtubule polymerization blocker . To test the contributions of transcription-related processes we apply α-amanitin , which inhibits the RNA polymerase II activity , and flavopiridol , which blocks the positive transcription elongation factor P-TEFb . In addition , we probed the impact of the local chromatin packing state by applying trichostatin A , which prevents histone deacetylation , and thus leads to chromatin decondensation . Finally , since the nucleolus is the site of ribosome biogenesis and thus directly involved in cellular protein production , we explore the role of protein synthesis in maintaining the nucleolus-nucleoplasm interface . To do so , we use cycloheximide , a protein synthesis inhibitor , and evaluate its effect at two time points , t1= 30 min and t2= 6 . 5 hr , upon drug addition . We anticipate that at short timescales , we can observe a direct impact of protein synthesis inhibition on the nucleolar-nucleoplasm interface , while at longer timescales , we can investigate a possible feedback between protein synthesis inhibition and nucleolar size and shape . Figure 7A shows micrographs of nuclei with fluorescently labeled chromatin ( H2B-GFP , green ) and nucleoli ( NPM-DsRed , red ) under the physiological conditions ( control ) and after treatment with blebbistatin , latrunculin A , nocodazole , α-amanitin , flavopiridol , trichostatin A and cycloheximide at t1 and t2 . We also survey the z-projection of nuclear and nucleolar contours obtained in different planes of the respective z-stack ( Figure 7—figure supplement 1 ) . To examine morphological differences under the studied conditions , we evaluate the same six parameters used earlier: AN/AN⁢u⁢c , e , α , de , fn⁢e⁢g , and Nn⁢e⁢g and visualize their distributions as violin plots in Figure 7B . The red dot indicates the mean , while the solid and dashed red lines correspond to the median and quartiles , respectively . Table 1 provides a summary of means and standard deviations for measured distributions of AN/AN⁢u⁢c , e , α , de , fn⁢e⁢g and Nn⁢e⁢g . In addition , we evaluated the skew of measured distributions and computed p-values for all measured physical quantities , the relative differences of their means with respect to control , as well as the Kullback-Leibler divergence with respect to control ( Table 1 ) . A close inspection of the violin plots ( Figure 7B ) and their corresponding statistical characteristics ( Table 1 ) reveals the following morphological changes upon cytoskeletal , chromatin and protein synthesis perturbations . Interestingly , the cytoskeletal perturbations , which act on the cytoskeleton outside the cell nucleus , did not lead to any major changes in the nucleolar morphology except for the actin polymerization inhibitor latrunculin A , which led to an increase of AN/AN⁢u⁢c compared to the control nucleoli . This increase , however , is due to the rounding up of nuclei upon the latrunculin A treatment ( Khatau et al . , 2009; Burnette et al . , 2014 ) , which leads to a decrease in the nuclear area AN⁢u⁢c , and thus causes the apparent increase of AN/AN⁢u⁢c , while the measured nucleolar area AN remains comparable to the AN of control nucleoli . Similarly , the observed decrease in the distance of nucleoli from the nuclear envelope , de , is likely caused by a decrease of the observed nuclear area AN⁢u⁢c . In contrast , chromatin perturbations such as transcription inhibitors α-amanitin , flavopiridol and histone deacetylase inhibitor trichostatin A led to visible changes in the nucleolar morphology as well as in the roughness of the nucleolus-nucleoplasm interface . Specifically , upon perturbing polymerase II activity using α-amanitin , we find that the nucleolar size AN/AN⁢u⁢c increases by ∼ 15% and nucleoli are on average more elliptical ( e ) . Moreover , the roughness of the nucleolus surface strongly increases by ∼ 40–50% as measured by the amount of negative curvature ( fn⁢e⁢g and Nn⁢e⁢g ) . Similarly , when we perturb the transcription elongation using flavopiridol , we observe ∼ 40–45% increase in the nucleolus surface roughness ( fn⁢e⁢g and Nn⁢e⁢g ) and nucleoli become on average more elliptical ( e ) . Finally , when we block the histone deacetylation using trichostatin A , which leads to chromatin decondensation , we find that the nucleolar size AN/AN⁢u⁢c decreases by ∼ 15% , while their eccentricity ( e ) remains unchanged . Strikingly , the nucleolar surface roughness decreases by ∼ 20–25% ( fn⁢e⁢g and Nn⁢e⁢g ) , in other words upon trichostatin A treatment it becomes smoother than in the control . Finally , the protein synthesis inhibition using cycloheximide left the nucleolar size unchanged , while the nucleoli became on average more elliptical ( e ) at longer times ( t2= 6 . 5 hr ) . Furthermore , protein synthesis inhibition led to a moderate ∼ 15% increase in the nucleolar surface roughness ( fn⁢e⁢g and Nn⁢e⁢g ) at both times ( t1= 30 min and t2= 6 . 5 hr ) . The orientation of the nucleoli within the nucleus and the nucleolar distance from the nuclear envelope is not significantly affected by any of the studied perturbations as illustrated by α and de , respectively . In this work , we study the nucleolus as the archetype of cellular organelles formed by liquid-liquid phase separation ( LLPS ) and monitor its size , shape and dynamics during its lifetime in human cells in vivo . We discover a rich phenomenology that grows the LLPS framework in new and unexpected ways: ( i ) We find that nucleoli exhibit anomalous dynamics and anomalous volume distribution during the cell cycle that defies any current theory and necessitates a new one . ( ii ) We uncover that the nucleolar fluid is a colloidal solution containing solid-like granules , the DFCs . ( iii ) We reveal that the surrounding nucleoplasm plays a key role in the LLPS of nucleoli that might have been previously overlooked and find that active ( ATP-dependent ) processes are involved in maintaining the nucleolus-nucleoplasm interface . Moreover , we identify specific biological processes participating in the nucleolus-nucleoplasm interactions . Our findings show that the nucleolar volume distribution scales as P ( V ) ∼V−1 during the entire cell cycle . The scale-free nature of this distribution suggests that nucleoli of any size can coalesce , moreover , there is no preferred size that nucleoli need to reach before/after they coalesce . It also suggests , that nucleoli of different sizes follow the same coalescence kinetics ( Caragine et al . , 2018 ) . Furthermore , the nucleolar volume distribution remains unchanged during the cell cycle , suggesting that the fusion of nucleoli is not limited to the first two hours of the cell cycle as previously hypothesized ( Savino et al . , 2001 ) , but can occur at any time . Nucleolar coalescence occurs from the early stages in the cell cycle , where it is thought to be a part of the nucleolar assembly process ( Savino et al . , 2001 ) . It is conceivable that at later times , the nucleolar coalescence might serve a different purpose as it is less likely to happen with decreasing nucleolar number . Interestingly , a volume distribution P⁢ ( V ) ∼V-1 was previously found also for liquid-like P-granules in the C . elegans oocyte ( Hubstenberger et al . , 2013 ) . In contrast , the volume distribution of nucleoli in the X . laevis oocyte follows ∼V−1 . 5 , which was shown to be consistent with diffusion-limited aggregation with constant influx of particles ( Brangwynne et al . , 2011 ) . The kinetics of human nucleolar assembly likely differs from that of the frog oocyte due to numerous differences between these two systems . For example , the nucleolar count is much lower in human cells ( ∼ 100 times less than in frog oocyte ) , there is a dense actin network present in the frog oocyte nucleus ( germinal vesicle ) , and human somatic nucleoli are connected to the chromatin fiber , thus , they cannot freely diffuse as it is in the case of nucleoli in the X . laevis oocyte ( Gall et al . , 2004; Brangwynne et al . , 2011; Caragine et al . , 2018; Berry et al . , 2018 ) . Further differences between the two systems include a large difference in the nuclear size ( diameter ~ 1000 μm in frog oocytes , ∼ 10 µm in human cells ) and the nucleolar size with volumes of 10–103 µm3 in frog oocytes and 10-2–102 µm3 in human cells ( Gall et al . , 2004; Brangwynne et al . , 2011; Caragine et al . , 2018 ) . The anomalous volume distribution of human nucleoli might likely be connected to their anomalous dynamics . Remarkably , our data suggest that one can predict if a pair of nucleoli is going to fuse by analyzing their motion . The differences in the dynamical behavior of non-fusing nucleoli and the ones in approach for fusion are stark . While the non-fusing ones appear to move randomly through the nucleoplasm , nucleoli that will fuse in the near future , move slower than non-fusing ones and show a linear correlation in their velocities ( Figure 5I ) . Considering the nucleolar size and the fact that they are physically tethered to the chromatin fiber , their motion unavoidably leads to local spatial reorganization of chromatin . Alternatively , a local chromatin rearrangement could facilitate the nucleolar pre-fusion approach . In fact , in our earlier work we found that the velocities of the growth of the neck connecting two fusing nucleoli ( Figure 1B ) are intriguingly similar to the velocities measured for active chromatin motion ( Caragine et al . , 2018; Zidovska et al . , 2013 ) . Since nucleoli move in an active fluid ( chromatin solution ) , we speculate that active processes might be involved in bringing them together to undergo fusion . To explore this hypothesis , future experiments and theories are needed to probe the nucleolar interactions with chromatin . The complex nature of the nucleolar fluid might also contribute to the anomalous behavior of human nuceoli . Strikingly , we find that dense fibrillar components ( DFCs ) behave as monodisperse solid-like colloidal particles ( granules ) suspended in a liquid phase of granular component ( GC ) . Our data shows that DFCs do not undergo aggregation , but remain of well-defined size and dimensions with a semi-major axis length of 210 ± 50 nm and semi-minor axis length of 180 ± 40 nm , as well as shape with aspect ratio of 1 . 22 ± 0 . 17 even upon nucleolar coalescence , which is consistent with solid-like particles . In contrast , the DFCs in frog oocytes were found to be polydisperse with diameter ∼ 2–5 µm , liquid-like with viscoelastic behavior ( Brangwynne et al . , 2011; Feric et al . , 2016 ) , and with their fusion being observed upon latrunculin A treatment ( Feric et al . , 2016 ) . It is also noteworthy that one frog oocyte DFC can be larger than the entire human nucleolus . Furthermore , our data reveal that human nucleoli obey a volumetric ratio for GC and DFC content , with DFC volume fraction ∼ 0 . 1 , which is significantly lower than in frog oocytes ( ∼ 0 . 25 ) ( Feric et al . , 2016 ) . This suggests it is the rRNA-rich GC phase that provides the human nucleolus with its liquid-like properties . To investigate the nucleolar interactions with the surrounding nucleoplasm , we have tested the impact of active ( ATP-dependent ) processes in general as well as specific biological processes such as cytoskeletal and transcriptional activity , chromatin packing state and protein synthesis . Our data suggest that nucleoli are closely dependent on ATP-dependent processes , losing their spherical shape upon ATP-depletion by exhibiting increased surface roughness ( local deformations ) . In our earlier study ( Caragine et al . , 2018 ) we have shown that the surface roughness can serve as a readout of the nucleolar surface tension . Specifically , local nucleolar surface deformations , which may be driven thermally or by active processes , are opposed by the surface tension . Thus , the larger the surface roughness , the lower its surface tension . Hence , a possible interpretation of the increase in nucleolar surface roughness upon ATP-depletion is a reduction of the surface tension γ of the nucleolus-nucleoplasm interface . These findings are consistent with our earlier study , which found that γ under physiological conditions is an effective quantity , and is therefore , likely dependent on some of the ATP-dependent cellular processes ( Caragine et al . , 2018 ) . Our data show that the roughness ( local deformations ) of the nucleolus-nucleoplasm interface is highly sensitive to the transcriptional activity in the nucleus . Interestingly , the inhibition of transcriptional activity ( such as polymerase II activity or mRNA elongation ) in the nucleus leads to an increase of the relative nucleolar size and the nucleolus becomes more elongated ( less spherical ) with a number of local deformations leading to high surface roughness . However , upon blocking the histone deacetylases , which causes a visible chromatin decondensation ( Tóth et al . , 2004 ) , we find not only a reduction in the relative nucleolar size , but also an increasingly smooth nucleolus-nucleoplasm interface . This suggests that the nucleolus-nucleoplasm interface is closely linked to the chromatin packing state as well as its transcriptional activity . Conversely , the perinucleolar chromatin is mostly heterochromatic , that is largely transcriptionally inactive , yet , its peculiar packing at the nucleolar surface might require active remodeling . Moreover , this is in agreement with our hypothesis that the surface tension γ is maintained by active processes and thus is an effective physical quantity . To elucidate the underlying physics , new theories accounting for the non-equilibrium nature of the nucleolus-nucleoplasm liquid interface need to be developed . In contrast , we find that the cytoskeletal forces exerted on the nucleus from the cytoplasm , do not contribute to the local roughness of the nucleolus-nucleoplasm interface , nor do they impact the nucleolar size and shape . Interestingly , in frog oocytes the disruption of the dense nuclear actin network by latrunculin A facilitates nucleolar fusion , leading to an increase in the nucleolar size ( Feric et al . , 2016 ) . Conversely , there is no filamentous actin network present in human cell nucleus . Lastly , our findings reveal that nucleoli are only moderately sensitive to the protein synthesis inhibition at time scales from 30 min to 6 . 5 hr . We do not observe any change in their size , only a slight increase in the surface roughness . However , it is possible that to observe an effect on nucleoli from the lack of protein synthesis much larger times need to be explored . In summary , we speculate that the interplay of the complex nature of the nucleolar fluid , the reduced mobility of nucleoli due to their chromatin tethering , as well as their interactions with the surrounding nucleoplasm , might impact the nucleolar assembly kinetics and lead to the observed anomalous nucleolar volume distribution ( ∼V-1 ) . In conclusion , nucleoplasm plays a major role in the life of nucleoli , the archetype of the liquid condensate formed by liquid-liquid phase separation in biology . Nucleoplasm , the fluid surrounding the nucleoli , is a complex polymeric solution containing chromatin . Chromatin fiber serves as the template for nucleolar formation and later forms a boundary at the nucleolus-nucleoplasm interface ( McClintock , 1934; Ritossa and Spiegelman , 1965; Wallace and Birnstiel , 1966; Bickmore and van Steensel , 2013; Németh and Längst , 2011; Towbin et al . , 2013 ) . Strikingly , the DNA sequences at which nucleoli form ( NORs ) and the genes located at the nucleolar interface are by no means random ( Németh and Längst , 2011 ) . This likely impacts the 3D chromosomal organization in the nucleolar vicinity . Moreover , considering chromatin’s active nature ( Zidovska et al . , 2013 ) and the fact that nucleoli are tethered to it during their lifetime , we speculate that active positional fluctuations ( or rearrangements ) of chromatin could bring nucleoli together , facilitating fusion . While this hypothesis remains to be tested , it is consistent with our observations that nucleoli , which are in approach to fusion , exhibit different dynamics than non-fusing ones . It is also supported by previous studies of colloidal mixtures , where the presence of particles with an actively driven translational motion leads to phase separation of active and passive ( i . e . , thermally driven ) components ( Stenhammar et al . , 2015 ) . Similar behavior has been found for polymer mixtures containing active and passive polymers ( Smrek and Kremer , 2017 ) . We speculate that , with respect to its translational mobility , the nucleolus could be abstracted as a passive droplet ( or colloid ) immersed in an active polymer ( chromatin ) . In such case , the active positional fluctuations of the polymer could cause demixing of the passive phase and thus effectively bring the passive colloids ( nucleoli ) together . That is , the active entities phase separate from the passive entities , enabling nucleolar coalescence . Such phase separation is distinct from the liquid-liquid phase separation by which the nucleoli are thought to form at the beginning of the cell cycle ( Brangwynne et al . , 2011; Berry et al . , 2015; Feric et al . , 2016 ) . The nucleolus plays a key role in cellular protein synthesis , thus any changes in nucleolar composition , structure or function can lead to cell abnormalities often connected with human diseases . For example , mutations in nucleolar proteins , which interact with RNA polymerase I , regulate rRNA transcription or participate in rRNA processing , are associated with cell cycle arrest and improper nucleolar assembly . This can lead to diseases such as skeletal and neurodegenerative disorders , cardiovascular disease and cancer ( Hannan et al . , 2013; Núñez Villacís et al . , 2018; Ruggero and Pandolfi , 2003; Derenzini et al . , 2009 ) . Moreover , in many diseases such as cancer , Alzheimer’s and Parkinson’s disease , but also in aging , human nucleoli change their shape and size ( Hannan et al . , 2013; Tsai and Pederson , 2014; Núñez Villacís et al . , 2018; Tiku and Antebi , 2018 ) , making the nucleolus a potential valuable diagnostic marker . Hence , a mechanistic understanding of nucleolus , its material properties and physical interactions with the nucleoplasm , might illuminate nucleolus in health and disease , contributing to new paths for diagnosis and therapy . The stable HeLa H2B-GFP cell line was cultured according to ATCC recommendations ( CCL-2 ) . Cells were cultured in a humidified , 5% CO2 ( vol/vol ) atmosphere in Gibco Dulbecco’s modified eagle medium ( DMEM ) supplemented with 10% FBS ( vol/vol ) , 100 units/mL penicillin , 100 μg/mL streptomycin ( Invitrogen ) and 4 . 5 µg/mL Plasmocin Prophylactic ( Invivogen ) . Cells were mycoplasma free , as determined by the Invivogen PlasmoTest ( Invivogen ) . For H2B-GFP imaging , cells were plated onto 35 mm MatTek dishes with glass bottom no . 1 . 5 ( MatTek ) 24 hr before imaging . We performed four independent experiments . For concurrent H2B-GFP and NPM-DsRed ( or NPM-mApple ) imaging , cells were plated onto 35 mm MatTek dishes 48 hr before imaging and transiently transfected with NPM-DsRed ( or NPM-mApple ) 24 hr prior to the experiment . All transfections were carried out using Lipofectamine 2000 ( Invitrogen ) following the manufacturer’s protocol . When indicated , cells were synchronized using 10 µM RO-3306 ( Enzo Life Sciences ) and imaged before the drug was removed after 16 hr , as well as 3 . 5 and 5 hr after the drug removal . The synchronized and unsynchronized populations were evaluated in two distinct experiments . For concurrent imaging of H2B GFP , NPM-DsRed and mCerulean-Fibrillarin-7 , cells were plated onto 35 mm MatTek dishes 48 hr before the experiment and transiently transfected with both NPM-DsRed and m-Cerulean-Fibrillarin-7 24 hr prior to the experiment . We performed six independent experiments , all of which were analyzed qualitatively and one quantitatively . NPM-DsRed and FBL-mCerulean ( mCerulean3-Fibrillarin-7 ) were gifts from Mary Dasso ( Addgene plasmid # 34553 ) ( Yun et al . , 2008 ) and from Michael Davidson ( Addgene plasmid # 55368 ) ( Markwardt et al . , 2011 ) , respectively . NPM-mApple plasmid was created as described earlier ( Caragine et al . , 2018 ) . For experiments involving biochemical perturbations , cells were plated onto 35 mm MatTek dishes 72 hr in advance of the experiment , transiently transfected with NPM-DsRed 48 hr prior to the experiment , and replated onto 35 mm MatTek dishes 24 hr prior to the experiment . We performed three independent experiments for each perturbation and six for the control . All imaging experiments were performed in the Gibco CO2-independent media ( Invitrogen ) supplemented with L-Glutamine ( Invitrogen ) and with MatTek dish containing cells mounted on the microscope stage in a custom-built environmental chamber maintained at 37°C with 5% CO2 supplied throughout the experiment . To deplete ATP , cells were treated with 6 mM 2-deoxyglucose ( DOG ) and 1 µM trifluoromethoxy-carbonylcyanide phenylhydrazone ( FCCP ) dissolved in CO2-independent medium supplemented with L-glutamine 2 hr before imaging . For cytoskeletal perturbations 10 µM latrunculin A , 10 µM blebbistatin or 10 µM nocodazole , respectively , in CO2-independent medium supplemented with L-glutamine were added to cells 30 min before imaging . For chromatin perturbations , 20 µg/mL α-amanitin ( Santa Cruz Biotechnology ) , 5 µg/mL cycloheximide ( Santa Cruz Biotechnology ) , 83 nM flavopiridol ( Santa Cruz Biotechnology ) , or 624 nM trichostatin A ( TSA ) , respectively , in CO2-independent medium supplemented with L-glutamine were added to cells 30 min , 30 min , 2 hr , and 24 hr before imaging , respectively . For cycloheximide , additional timepoint , 6 . 5 hr after drug additon , was evaluated . All chemicals were from Sigma Aldrich unless stated otherwise . Cells were imaged with a Yokogawa CSU-X1 confocal head with an internal motorized high-speed emission filter wheel , Spectral Applied Research Borealis modification for increased light throughput and illumination homogeneity on a Nikon Ti-E inverted microscope equipped with an oil-immersion 100× Plan Apo NA 1 . 4 objective lens , an oil-immersion 40× Plan Fluor NA 1 . 3 objective lens , and the Perfect Focus system . The microscope was mounted on a vibration-isolation air table . The pixel size for the 100× and 40× objective was 0 . 065 µm and 0 . 1625 µm , respectively . H2B-GFP and NPM-DsRed ( or NPM-mApple ) fluorescence was excited with a 488 nm and a 561 nm solid-state laser , respectively . To image H2B-GFP and NPM-DsRed ( or NPM-mApple ) at the same time , we illuminated the sample simultaneously with both excitation wavelengths , 488 nm and 561 nm . The emission was separated by the W-View Gemini Image Splitter ( Hamamatsu ) using a dichroic mirror ( Chroma Technology ) , followed by an ET525/30m and an ET630/75m emission filter ( Chroma Technology ) . The two fluorescent signals were allocated to the two halves of the image sensor , producing two distinct images . The exposure time for each frame was 250 ms . For three color imaging , H2B-GFP , NPM-DsRed , and FBL-mCerulean were excited with 488 nm , 561 nm , and 405 nm solid state lasers , respectively , and fluorescence was collected with a 405/488/561/640 multiband-pass dichroic mirror ( Semrock ) and then an ET525/50m , ET600/50m and ET450/50m emission filter , respectively ( Chroma Technology ) . The exposure time was 250 ms , 250 ms , and 1000 ms for H2B-GFP , NPM-DsRed , and FBL-mCerulean , respectively . Z-stacks were taken with a z axis step size of 500 nm , with the shutter closed in-between steps and an exposure time of 250 ms per plane . Images were obtained with a Hamamatsu ORCA-R2 cooled CCD camera controlled with MetaMorph 7 ( Molecular Devices ) . The streams of 16-bit images were saved as multi-tiff stacks . Images were converted to single-tiff images and analyzed with MatLab ( The MathWorks ) . The nuclear and nucleolar contours were determined from the H2B-GFP and NPM-DsRed signal , respectively , using previously published procedures ( Chu et al . , 2017; Caragine et al . , 2018 ) . The nucleolar velocity was determined as the displacement of the centroid of the filled nucleolar contour relative to the displacement of the centroid of the filled nuclear contour , divided by the elapsed time . For radial velocity calculations , we define the radial distance of each nucleolus as its distance from the midpoint of the linear distance between the centroids of two nucleoli . The radial distances for both nucleoli are measured relative to the midpoint of the linear distance between the nucleoli found in two time points , in order to exclude the movement of the other nucleolus in the calculation of nucleolar velocity . Finally , to calculate the radial velocity we divide the change in the radial distance by the time elapsed . The nucleolar distance from the nuclear envelope was found by finding the minimum distance between the nucleolar centroid and the nuclear contour . The angle between the nucleolus and nucleus was determined by fitting both nucleus and nucleolus with an ellipse and measuring the angle between their respective major axes , for angles greater than 90° , its supplement was taken . The nuclear and nucleolar area were determined as the number of pixels filling its respective contours . Nuclear and nucleolar eccentricity was calculated as the ratio of the semi-major axis length and its semi-minor axis length , when fitted to an ellipse . To obtain an accurate count of DFCs , we developed a feature-finding procedure . A mask created by the nucleolar contour was applied to the FBL-mCerulean image to remove the background signal . Using a local-maxima function , we found a large number of local maxima indicating possible features in the image , most of which correspond to noise . We manually selected DFCs from the local maxima that were found . Next , we manually fit each DFC with a circumscribed and an inscribed circle , measuring the semi-major DFC axis a and the semi-minor DFC axis b , respectively .	The inside of a cell is very organized . Just as bodies contain internal organs , cells contain many different compartments , called ‘organelles’ , each with its own specific role . Most organelles are surrounded by a membrane that keeps their contents separate from the cytoplasm , the water-based liquid inside the rest of the cell . Some organelles , however , are not bounded by a membrane . Instead , they act like tiny drops of oil in water , retaining their structure because they have different physical properties from the fluid around them , a phenomenon called liquid-liquid phase separation . One such organelle is the nucleolus , which sits inside the cell’s nucleus ( a membrane-bound organelle containing all the genetic material of the cell in the form of DNA ) . The nucleolus’s job is to produce ribosomes , the cellular machines that , once transported out of the nucleus , will make proteins . Human cells start with 10 small nucleoli in the nucleus , which fuse together until only one or two larger ones remain . Previous research showed that nucleoli form and persist thanks to liquid-liquid phase separation , and they behave like liquid droplets . Despite this , exactly how nucleoli interact with each other and with the fluid environment in the rest of the nucleus remained unknown . Caragine et al . set out to measure the behavior and interactions of nucleoli in living human cells . Microscopy experiments using human cells grown in the laboratory allowed the positions , size and shape of nucleoli to be tracked over time . This also yielded detailed information about the smoothness of their surface . Mathematical analysis revealed that pairs of nucleoli normally moved independently of each other , unless they were about to fuse , when they invariably slowed down and coordinated their movements . In addition , altering the state of DNA in the surrounding nucleus also affected the nucleoli . For example , when DNA was less densely packed , nucleoli shrank and their surfaces became smoother . These results build on our knowledge of how cells are organized by showing , for the first time , that the environment within the nucleus directly shapes the behavior of nucleoli . In the future , a better understanding of how cells maintain healthy nucleoli may help develop new treatments for human diseases such as cancer , which are characterized by problems with this organelle .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "structural", "biology", "and", "molecular", "biophysics", "physics", "of", "living", "systems" ]	2019	Nucleolar dynamics and interactions with nucleoplasm in living cells
Topoisomerase II ( TOP2 ) relieves topological stress in DNA by introducing double-strand breaks ( DSBs ) via a transient , covalently linked TOP2 DNA-protein intermediate , termed TOP2 cleavage complex ( TOP2cc ) . TOP2ccs are normally rapidly reversible , but can be stabilized by TOP2 poisons , such as the chemotherapeutic agent etoposide ( ETO ) . TOP2 poisons have shown significant variability in their therapeutic effectiveness across different cancers for reasons that remain to be determined . One potential explanation for the differential cellular response to these drugs is in the manner by which cells process TOP2ccs . Cells are thought to remove TOP2ccs primarily by proteolytic degradation followed by DNA DSB repair . Here , we show that proteasome-mediated repair of TOP2cc is highly error-prone . Pre-treating primary splenic mouse B-cells with proteasome inhibitors prevented the proteolytic processing of trapped TOP2ccs , suppressed the DNA damage response ( DDR ) and completely protected cells from ETO-induced genome instability , thereby preserving cellular viability . When degradation of TOP2cc was suppressed , the TOP2 enzyme uncoupled itself from the DNA following ETO washout , in an error-free manner . This suggests a potential mechanism of developing resistance to topoisomerase poisons by ensuring rapid TOP2cc reversal . Topoisomerase II ( TOP2 ) is an evolutionarily conserved enzyme capable of generating reversible double strand breaks ( DSBs ) in DNA , which enables the resolution of problematic DNA topological structures that arise during normal cellular processes , such as transcription , replication , and mitosis . This process is mediated by the topoisomerase II cleavage complex ( TOP2cc ) , a transient DNA-protein complex that is formed when each monomer of the dimeric TOP2 anchors itself to the DNA backbone via a covalent 5’-phosphotyrosyl bond . A second DNA strand can then be passed through the enzyme-bridged DSB , after which the break is re-ligated . Under physiological conditions , topoisomerases reliably execute this reaction without error , as TOP2ccs are highly reversible and short lived ( Pommier et al . , 2016 ) . However , when they become trapped , the abortive TOP2ccs fail to re-ligate the transient DSB intermediate , presenting a severe threat to genome integrity which must be resolved quickly via the DNA repair machinery . The resolution of TOP2ccs is , therefore , crucial for cell survival . TOP2 poisons , such as the anti-cancer agent etoposide ( ETO ) , efficiently trap and stabilize TOP2ccs ( Muslimović et al . , 2009; Nitiss , 2009a; Pommier et al . , 2016 ) . The accumulation of these TOP2 DNA-protein complexes can create significant problems for cells by stalling replication forks and the transcription apparatus , generating torsional and genotoxic stress , which leads to the accumulation of single and double strand DNA breaks ( Cowell and Austin , 2012; Muslimović et al . , 2009 ) . How these abortive TOP2ccs are recognized and targeted by the DNA damage response ( DDR ) remains unclear . It has been suggested that a robust DNA damage response ( DDR ) is elicited only after ETO-trapped TOP2ccs have been converted to ‘clean’ protein-free DSB ends ( Mårtensson et al . , 2003; Sunter et al . , 2010; Zhang et al . , 2006 ) . This suggests that a DSB shielded by a covalently bound protein can evade detection . Several mechanisms exist for removal of trapped TOP2ccs . However , the factors that determine how a cell chooses to deploy them are complex and remain poorly understood . The covalently-linked TOP2 protein is partially or completely degraded by the 26S proteasome , after which the residual DNA-peptide adduct is hydrolyzed by tyrosyl-DNA phosphodiesterase 2 ( TDP2 ) to release the remnants of the entrapped protein ( Ledesma et al . , 2009; Gómez-Herreros et al . , 2013; Lee et al . , 2018; Sunter et al . , 2010; Zagnoli-Vieiral and Caldecott , 2017 ) . Current evidence suggests that transcription increases the rate at which TOP2 is processed by the proteasome ( Canela et al . , 2019 ) . It has also been reported that TDP2 , with the co-operation of secondary cofactors , can resolve unprocessed TOP2ccs in a proteasome-independent manner ( Schellenberg et al . , 2017 ) . An alternate , more error-prone mechanism for TOP2cc removal is mediated by MRE11 endonucleolytic cleavage in the vicinity of a trapped TOP2cc , which releases the entire TOP2 DNA-protein complex that results in the loss of small stretches of DNA ( Hoa et al . , 2016; Lee et al . , 2012 ) . Either of these mechanisms are able to produce protein-free DNA breaks that can then be recognized and repaired by the major cellular DSB repair pathways: non-homologous end joining ( NHEJ ) and homologous recombination ( HR ) . It has long been recognized that trapped TOP2ccs are intrinsically reversible upon ETO washout ( Hsiang and Liu , 1989; Long et al . , 1985 ) . Consistently , we observed that inhibiting the proteasome prior to ETO treatment enhanced the number of rapidly reversible ETO-stabilized TOP2ccs across the genome ( Canela et al . , 2019 ) . Other studies have shown that inhibiting the proteasome not only preserves the reversibility of TOP2ccs but also is able to suppress DDR signaling ( Mao et al . , 2001; Zhang et al . , 2006 ) . However , the implications of suppressing the DDR in response to ETO using proteasome inhibitors , on both the long-term genome integrity and the overall viability of the cell , have not been fully explored . Considering that both proteasome inhibitors and topoisomerase poisons are used , sometimes in tandem , as frontline chemotherapeutic agents to treat a variety of cancers ( Cowell and Austin , 2012; Dittus et al . , 2018; Manasanch and Orlowski , 2017; Thomas et al . , 2017 ) , understanding how TOP2cc reversibility may impact the effectiveness of these drugs is clinically relevant . Here , we utilize high-resolution genome-wide mapping of DSBs by END-seq ( Canela et al . , 2016 ) to examine the impact of proteasome inhibition on the fate of TOP2ccs . We found that once ETO-stabilized TOP2ccs had been proteolytically processed throughout the genome , and a robust DDR had been initiated , the large number of newly generated protein-free DSBs were repaired in a highly error-prone manner , resulting in toxic chromosomal translocations and rearrangements that led to cell death . Notably , the major DSB repair pathways ( NHEJ and HR ) contribute to the mis-repair of ETO-induced DSBs following proteasomal processing . By inhibiting proteasome-mediated degradation of TOP2ccs either through chemical inhibition or by ablation of RNF4-mediated ubiquitination of TOP2ccs , an intact and enzymatically competent TOP2 was able to re-seal the protein-linked DSBs without invoking a significant DDR . As a result , cells were protected from genomic instability , which ultimately led to enhanced cell survival after treatment with topoisomerase poisons . Building on previous observations that proteasome inhibition can specifically attenuate the induction of γ-H2AX by ETO ( Mao et al . , 2001; Zhang et al . , 2006 ) , we sought to first confirm these findings in mouse primary splenic B-cells . Since ETO stabilizes both TOP2A and TOP2B isoforms , resting B-cells were activated with cytokines for 12 hr ( Figure 1A ) . This short-term treatment ensures that B-cells remain in G1 where only the TOP2B isoform is expressed ( Canela et al . , 2017 ) . This treatment regime , therefore , allows us to minimize any differential effect of ETO on each isoform ( Errington et al . , 2004; Willmore et al . , 1998 ) and to bypass essential TOP2-dependent processes , such as DNA replication . In agreement with other studies ( Mao et al . , 2001; Zhang et al . , 2006 ) , we found that treatment with a high ( 50 µM ) dose of ETO for 2 hr elicited a strong γ-H2AX response in B-cells ( Figure 1B ) . Pre-incubating cells with proteasome inhibitors ( 10 µM MG132 , 10 µM Bortezomib ( BTZ ) , 2 µM Epoxomicin ( EPN ) , or 10 µM Ixazomib ( IXA ) ) for 1 hr prior to ETO treatment almost completely abolished the γ-H2AX signal ( Figure 1B ) . The suppressed γ-H2AX signal occurred specifically in response to ETO , as pre-treating cells with the same dose of proteasome inhibitor did not affect the γ-H2AX response to ionizing γ-irradiation ( 5 Gy IR ) ( Figure 1B and C ) . Thus , proteasomal activity is required for triggering a robust DDR specifically in response to ETO treatment . As trapped TOP2ccs are subjected to proteasomal degradation , unprocessed TOP2ccs accumulate on DNA in cells pre-treated with a proteasome inhibitor compared to cells treated with ETO alone ( Lee et al . , 2016 ) . To confirm that proteasome inhibitors were having the same effect in our experimental system , we quantified TOP2ccs using the ICE assay ( Anand et al . , 2018 ) in ETO-treated primary B-cells pre-incubated with or without BTZ . In the presence of ETO , TOP2ccs were readily detected on DNA ( Figure 1D ) . Immediately after ETO was washed out , however , the amount of DNA-bound TOP2 diminished strongly ( Figure 1D ) . The proteasome continuously degraded ETO-trapped TOP2ccs , as evidenced by the markedly enhanced accumulation of TOP2ccs when cells were pre-incubated with BTZ ( Figure 1D ) . The majority of these TOP2ccs also dissociated from DNA immediately following ETO washout ( Figure 1D ) . Thus , inhibiting proteasomal degradation appears to preserve the integrity of TOP2ccs enabling reversal of complexes by completion of the enzymes' catalytic cycle upon drug removal . Alternatively , proteasome activity could recover immediately after the removal of ETO and the proteasome inhibitor and would similarly lead to a loss of TOP2cc signal in the ICE assay . To determine the duration and reversibility of proteasomal inhibition , we directly monitored proteasome activity in eHAP cells transfected with a YFP-Degron reporter that has been demonstrated to be a bona fide proteasomal substrate ( Bence et al . , 2001 ) . As expected , a 2 hr treatment with MG132 or BTZ significantly increased the YFP signal from baseline ( Figure 1E ) . Following washout of MG132 or BTZ , the elevated YFP signal persisted for several hours before it began to decrease and BTZ appeared to be a more potent and persistent inhibitor of the proteasome than MG132 ( Kisselev and Goldberg , 2001 ) . As an additional measure of proteasome activity , we quantified the protein levels of p53 , as it is known to be stabilized upon proteasome inhibition ( An et al . , 2000; Halasi et al . , 2014 ) . Consistent with the YFP-degron results in eHAP cells , we observed that p53 protein remained stabilized in primary B-cells for several hours after proteasome inhibitors were washed out , with BTZ again being more potent than MG132 ( Figure 1F ) . Thus , proteasome activity is not readily recovered even after the removal of proteasome inhibitor , suggesting that the rapid loss of ETO-induced TOP2ccs in MG132 pre-treated cells upon washout most likely reflects the reversal of TOP2ccs by completion of the enzymes' catalytic cycle upon drug removal . Accordingly , we did not observe a delayed γ-H2AX induction at either 2 hr or 6 hr post-ETO and BTZ washout , suggesting that proteasomal activity remain suppressed for at least several hours post-washout ( Figure 1G ) . These data imply that persistent proteasome inhibition allows for TOP2cc reversal and prevents trapped TOP2ccs from being converted into protein-free DSB ends that are capable of eliciting a robust DNA damage response ( DDR ) . Contrary to our observations , previous studies have shown that proteasome inhibitors synergize with topoisomerase poisons like ETO in mediating cell killing ( Aras and Yerlikaya , 2016; Ceruti et al . , 2006; Destanovic et al . , 2018; Dittus et al . , 2018; Lee et al . , 2016; von Metzler et al . , 2009 ) . Interestingly , we found that the addition of BTZ prior to or concurrent with ETO suppressed DDR signaling , but incubating B-cells with BTZ post-ETO treatment did not ( Figure 1G ) . These results showed that the timing of proteasome inhibitor treatment relative to ETO treatment is critical to its effects on TOP2cc metabolism and subsequent DDR signaling . To further analyze the influence of the proteasome on TOP2ccs , we employed genome-wide DSB mapping by END-seq ( Canela et al . , 2019; Canela et al . , 2016 ) . While ETO can generate high levels of both SSBs and DSBs ( Baranello et al . , 2014; Gittens et al . , 2019 ) , END-seq only detects DSBs generated by ETO ( Canela et al . , 2017; Canela et al . , 2016 ) . However , this protocol allows us to capture and distinguish both TOP2ccs and protein-free DSBs generated by ETO ( Canela et al . , 2019 ) . First , we assessed whether proteasome inhibition blocked TOP2 from making incisions in DNA . To this end , we used a cocktail of Exonuclease VII ( ExoVII ) and Exonuclease T ( ExoT ) during sample preparation , allowing us to capture all sites of TOP2 activity ( Canela et al . , 2019 ) . As expected , END-seq analysis revealed that TOP2-mediated DSBs were readily detectable in ETO-treated B-cells ( Figure 2A–C ) . The number and intensity of TOP2-mediated DSBs were reproducible across experiments ( Figure 2—figure supplement 1A–B ) . Consistent with the results shown in Figure 1D , we found that BTZ pre-treatment did not dramatically affect the overall number or location of TOP2-mediated DSBs ( Figure 2A–C; Figure 2—figure supplement 1C–D ) ; TOP2-mediated DSBs were still strongly induced in the BTZ pre-treated sample , despite eliciting only a weak γ-H2AX response ( Figure 1B ) . Thus , while proteasome inhibition does not affect the ability of TOP2 to cleave DNA , it helps maintain the resultant DSBs in a protein-linked form ( persistent TOP2cc ) that conceals them from the DDR . Nevertheless , we did observe a small decrease in the overall intensity of the TOP2-mediated DSB signal ( Figure 2A and C ) upon BTZ pre-treatment . This could be due to reversal of TOP2ccs in the BTZ pre-treated sample by completion of the enzymes' catalytic cycle upon drug removal . Since inhibiting proteasomal degradation did not affect the initial formation of TOP2-mediated DSBs , we wanted to understand if proteasome inhibition affected their fate over time . To directly evaluate the impact of proteasome inhibition on repair of TOP2-mediated DSBs , we used END-seq to assess their longevity by harvesting primary B-cells both immediately following ETO/BTZ washout ( 0 hr washout ) and after a 2 hr recovery in drug-free media at 37°C ( 2 hr washout ) . Notably , BTZ exerted a substantial impact on the rate at which TOP2-mediated DSBs resolved over the course of the 2 hr washout , with very few detectable breaks remaining after recovery relative to cells treated with ETO alone ( Figure 2A and D; Figure 2—figure supplement 1E ) . We plotted the fraction of TOP2-mediated DSBs at each washout timepoint relative to the initial level of breakage following ETO treatment ( Figure 2D ) . Using this analysis , we found that immediately following drug washout , 30% of the initial TOP2-mediated DSBs were still detectable in cells treated with ETO only , while just 15% were still present in the corresponding BTZ pre-treated sample ( Figure 2A and D ) . After a 2 hr recovery , B-cells treated with ETO alone still had a detectable break signal ( 17% ) , whereas in BTZ-pretreated cells , it returned almost to baseline ( 4% , Figure 2A and D ) . This suggested that inhibiting the proteasome enhanced the amount of TOP2ccs which rapidly reverse following ETO washout . As a second form of analysis , we normalized the TOP2-mediated DSB intensity at each post-washout timepoint . We introduced a new parameter , termed TOP2-mediated DSB persistence , which was calculated as the ratio of the peak intensity ( RPKM ) for individual TOP2-mediated DSBs at each timepoint ( either 0 hr washout or 2 hr washout ) relative to its initial intensity in the corresponding pre-washout sample ( Figure 2E; Figure 2—figure supplement 1F ) . Through this analysis , we found that approximately 20% of the initial TOP2-mediated DSBs were classified as persistent immediately following ETO washout ( 0 hr washout ) , and 13% of initial TOP2-mediated DSBs were still persistent 2 hr post-washout ( 2 hr washout ) ( Figure 2E ) . By contrast , BTZ pre-treatment more significantly reduced the fraction of persistent TOP2-mediated DSBs both immediately following washout ( ~13% ) , as well as at 2 hr post-washout ( ~3 . 5% ) ( Figure 2E ) . These results , in conjunction with the ICE assay ( Figure 1D ) , suggested that inhibiting the proteasome enhances TOP2cc reversal following ETO washout . To directly measure the reversibility of TOP2ccs , we took advantage of the fact that ExoT lacks the ability to process protein-linked DSBs ( persistent TOP2ccs that still have an intact 5’-phospotyrosyl bond ) and therefore allows for the detection of only protein-free ( i . e . proteolytically processed ) DSBs ( Canela et al . , 2019 ) . As such , the levels of TOP2ccs can be estimated by the difference between total TOP2-mediated DSBs detected by ExoVII+ExoT and protein-free DSBs detected by ExoT alone ( Canela et al . , 2019 ) . Our analyses revealed that 76% of TOP2-mediated DSBs reversed immediately following washout ( 0 hr washout ) , with the majority of remaining lesions being converted into protein-free DSBs ( Figure 2F ) . The fraction of reversible TOP2ccs was even higher in the BTZ pre-treated cells , as 90% of the initial TOP2-mediated DSBs were reversible immediately following washout ( Figure 2F ) . Thus , suppressing proteasomal degradation of TOP2ccs led to increased TOP2cc self-reversal . Finally , we assessed how the timing of proteasome inhibition impacted the rate at which TOP2-mediated DSBs are resolved . We found that compared to ETO alone , both BTZ pre-treatment and co-treatment significantly reduced the number and persistence of TOP2-mediated DSBs after a 2 hr washout ( Figure 2G and H ) . By contrast , BTZ post-treatment slightly increased the number , as well as persistence , of TOP2-mediated DSBs at the same timepoint ( Figure 2G and H ) . These results confirmed that the timing of proteasome inhibition relative to ETO treatment determined its impact on the resolution of TOP2-mediated DSBs . Since protein-free DSBs persist for a significant period of time following ETO washout , it enabled the study of DNA processing at these sites . We observed clear evidence of 5’ to 3’ DNA end-resection of TOP2-mediated DSBs immediately after drug washout ( Figure 3A ) . After 2 hr post ETO treatment , the extent and intensity of resection had increased ( Figure 3A ) , indicating ongoing end-processing at persistent TOP2-mediated DSBs . In ETO-treated cells pre-incubated with BTZ , while there were still low levels of protein-free DSBs present immediately following drug washout ( 0 hr washout ) , we could barely detect end-resection at individual breaks ( Figure 3A ) . Genome-wide , we were able to identify 1289 resected breaks in ETO-treated cells immediately following washout , while only 496 resected breaks were detected in BTZ pre-treated cells , a 2 . 5-fold decrease ( Figure 3B ) . The end-resection detected in ETO-treated cells genome-wide was reproducible across replicates ( Figure 3—figure supplement 1 ) . Therefore , inhibiting the proteolytic degradation of TOP2ccs prevented DNA repair-associated nucleases from processing TOP2-mediated DSBs . Moreover , the extent of resection , defined as the maximum distance away from the break summit , was significantly shorter in BTZ-pretreated cells compared to cells treated with ETO alone ( Figure 3C; Figure 3—figure supplement 1B ) . Finally , we found that after ETO removal , TOP2-mediated DSBs undergoing nucleolytic processing ( resected breaks ) tended to be more persistent than non-resected breaks ( Figure 3D; Figure 3—figure supplement 1C ) . Thus , the longer a TOP2-mediated DSB persists in the genome , the more likely it will be processed and repaired in a potentially error-prone manner . Enzymatic reversal of TOP2ccs is predicted to be an error-free process . By contrast , the degradation of TOP2ccs followed by DSB repair is potentially more susceptible to error . Nevertheless , repair of ETO-induced DSBs by TDP2-dependent NHEJ might be performed with fidelity ( Gómez-Herreros et al . , 2013 ) since it involves simple re-ligation of compatible 4 bp overhangs . To directly address this question , we exposed primary B-cells in G1 to ETO ( 2 hr , 50 uM ) with or without BTZ pre-treatment , or post-treatment . After washing out the drugs , cells were allowed to recover at 37°C in fresh , drug-free media for 24 hr before they were harvested for mitotic chromosome analysis ( Figure 1A ) . Notably , we found that even a short pulse of ETO induced a substantial number and variety of chromosomal aberrations in WT B-cells ( ~14 aberrations per cell ) ( Figure 4A and B ) . The majority of aberrations that we observed were dicentric chromosomal fusions and chromosome breaks ( Figure 4A and B ) . Post-treatment with MG132 did not mitigate the genotoxic effects of ETO ( Figure 4B ) , consistent with the lack of effect it had on ETO-induced γ-H2AX response ( Figure 1G ) . However , pre-treating B-cells with MG132 or BTZ completely protected cells from accruing chromosomal aberrations induced by ETO , with less than one aberration detected per cell ( Figure 4A and B ) . Therefore , the repair of DSBs that arise from proteasome-mediated degradation of TOP2ccs is error-prone , while preventing TOP2cc processing preserves genome integrity . Mitotic chromosome analysis is limited in its ability to determine the extent of chromosomal fusion events , as pieces of two or more chromosomes could be ligated together and still appear as a normal chromosome when stained with only DAPI . We consistently detected the presence of elongated chromosomes in mitotic spreads prepared from ETO-treated B-cells , suggesting that these might result from complex fusions involving multiple chromosomal fragments . To further characterize these seemingly intact long chromosomes , we performed spectral karyotyping ( SKY ) ( Liyanage et al . , 1996 ) . As shown in Figure 4C and D , SKY analysis revealed that ETO-induced chromosomal fusions were extensive and complex . Indeed , these events resembled chromothripsis , in which fragments originating from three or more chromosomes are fused together ( Zhang et al . , 2013 ) . Strikingly , pre-treating B-cells with MG132 before ETO completely protected them from all types of chromosomal fusions ( Figure 4C and D ) . These results highlight the severe genotoxic consequences of a transient high dose pulse of ETO and demonstrate that such ETO-induced genome instability is proteasome-dependent . Consistent with the impact of the proteasome on genome integrity , we found that pre-treatment with MG132 , but not post-treatment , mitigated the induction of cell cycle arrest by ETO evidenced by the increased cellular incorporation of EdU after the start of ETO treatment ( Figure 4E ) . While proteasome inhibition itself caused a reversible block on S-phase entry , as previously described ( Rastogi and Mishra , 2012 ) , proteasome inhibitor pre-treatment but not post-treatment , significantly diminished the cytotoxicity of ETO ( Figure 4F ) . Taken together , our data indicate that proteasome inhibitor pre-treatment maintains cellular proliferative capacity and long-term viability by blocking ETO-induced genome instability . To determine if DDR signaling per se is responsible for the error-prone repair of TOP2-associated DSBs , we conducted mitotic spread analyses in ATM-/- and H2AX-/- primary B-cells . We found that these DDR signaling deficient mutants accumulated somewhat higher levels of chromosomal aberrations following ETO treatment compared to WT ( Figure 5A ) . Importantly , proteasome inhibitor pre-treatment almost completely abolished ETO-induced damage in WT , as well as in ATM-/- and H2AX-/- B-cells ( Figure 5A ) . Thus , proteasome inhibition but not ATM-γ-H2AX driven DDR signaling protects against ETO-induced genome instability . Both NHEJ and HR are thought to contribute to the repair of ETO-induced DSBs ( Ledesma et al . , 2009; Gómez-Herreros et al . , 2013; Gómez-Herreros et al . , 2017; Hoa et al . , 2016; Pommier et al . , 2016 ) . To determine whether classical NHEJ or HR is primarily responsible for the misrepair of DSBs , we tested primary B-cells deficient in the key end joining factor DNA ligase IV ( Lig4-/- ) or the key HR factors BRCA1 ( BRCA1Δ11 ) and BRCA2 ( BRCA2-/- ) . To this end , primary resting ( G0 ) Lig4-/- , BRCA1Δ11 and BRCA2-/- B-cells were first activated with cytokines for 12 hr and 24 hr to drive G1- or S-phase entry , respectively , at which point they were exposed to ETO . After ETO was washed out , B-cells were given an additional 24 hr to recover before mitotic spread analysis . We found that loss of Lig4 , BRCA1 or BRCA2 failed to mitigate aberrant chromosomal rearrangements in ETO-treated G1 and cycling B-cells , respectively ( Figure 5B and C; Figure 5—figure supplement 1 ) . Furthermore , MG132 pre-treatment completely prevented ETO-induced genome instability in Lig4-/- and BRCA1Δ11 cells ( Figure 5B and C ) . Thus , the protective effects of proteasome inhibition on ETO-induced genome instability is not cell cycle dependent , and neither classical NHEJ nor HR appears to be strictly required to generate ETO-induced chromosomal aberrations . Given the lack of clear dependency on canonical DSB repair pathways , we next explored whether Polθ-mediated alternative end joining could be a major contributor of ETO-induced genome instability . Similar to Lig4- and BRCA1/2-deficiency , however , loss of Polθ ( POLQ-/- ) did not significantly attenuate the formation of complex chromosomal rearrangements in ETO-treated cells ( Figure 5—figure supplement 1 ) . Therefore , multiple DNA repair mechanisms likely contribute to the error-prone repair of ETO-induced DSBs . In addition to preventing the degradation of TOP2ccs , previous studies have shown that proteasome inhibitors are able to suppress ATM-γ-H2AX signaling in response to damage induced by TOP1 poisons like Camptothecin ( CPT ) ( Lin et al . , 2008 ) , which stabilizes TOP1ccs in the same mechanistic way ETO stabilizes TOP2ccs ( Pommier et al . , 2016 ) . We , therefore , assessed if cells pre-treated with proteasome inhibitors were protected from chromosomal aberrations caused by CPT . Our results showed that MG132 pre-treatment largely suppressed CPT-induced chromosomal aberrations ( Figure 5D ) , similar to its effects in ETO-treated cells ( Figure 5A–C ) . Thus , inhibiting the proteolytic degradation of either TOP1ccs or TOP2ccs prevents subsequent DNA mis-repair that leads to genome instability . Based on the results presented above , it is clear that the degradation of ETO-stabilized TOP2ccs leads to genome instability , while chemical suppression of proteasomal TOP2cc degradation prevents the accumulation and mis-repair of genotoxic DSBs . Recent work has described a SUMO-ubiquitin ( Ub ) pathway that recognizes DNA-bound topoisomerase-DNA complexes , wherein the SUMO-targeted E3 ubiquitin ligase ( STUbL ) RNF4 ubiquitinates TOP2 leading to its degradation ( Sun et al . , 2019 ) . We therefore assessed whether interfering with the cell’s ability to sense and recruit the proteasome to trapped TOP2ccs would confer a chemo-protective effect similar to cells treated with BTZ and ETO ( Figure 1B , Figure 4A–F ) . To this end , we treated WT and RNF4-/- MEFs ( Hu et al . , 2010 ) with 10 µM ETO for 1 hr with or without BTZ pre-incubation ( 1 µM , 1 hr ) , and probed for γ-H2AX . ETO produced a weakened γ-H2AX response in RNF4-/- MEFs compared to WT MEFs ( Figure 6A ) , suggesting that RNF4 facilitates proteasomal degradation of TOP2ccs , which in turn generates γ-H2AX . Notably , while BTZ pre-treatment suppressed the ETO-induced γ-H2AX response in WT MEFs ( Figure 6A ) , it did not further reduce the γ-H2AX response in RNF4-/- MEFs ( Figure 6A ) , suggesting that RNF4 is epistatic with the proteasome with regard to TOP2cc degradation . To explore this further , we used END-seq to quantify the number and intensity of TOP2-mediated DSBs in ETO treated RNF4-/- and WT primary B-cells . Our results showed that there was a 6-fold reduction of protein-free DSBs detected by END-seq in RNF4-/- B-cells compared to WT ( Figure 6B , top right ) , while the initial levels of total TOP2-mediated DNA cleavage were similar between the genotypes ( Figure 6B , top left ) . This effect was comparable to what we observed by chemically inhibiting the proteasome with BTZ , which reduced the number of protein-free DSBs by ~5 fold ( Figure 6B , bottom right ) . These results indicate that RNF4 ubiquitination is critical for proteasome-mediated processing of TOP2ccs into genotoxic protein-free DSBs . To assess if the loss of RNF4 conferred resistance to ETO , we determined how RNF4 deletion affected short-term cellular viability and long-term colony formation potential in ETO-treated B-cells and MEFs , respectively . We found that RNF4-/- cells were significantly more viable and retained higher colony formation capacity after a 1 hr pulse of ETO compared to WT counterparts ( Figure 6C and D ) . Consistent with increased viability , RNF4-/- B-cells accumulated 60% less chromosomal aberrations after ETO treatment compared to WT B-cells ( Figure 6E ) . Taken together , these results indicated that impairing the proteasome response to trapped TOP2ccs reduces the cytotoxicity and genotoxicity of ETO . Notably , pre-treating RNF4-/- B-cells with BTZ further reduced aberrant chromosomal repair ( Figure 6E ) , suggesting that RNF4-mediated ubiquitination is likely not the only mechanism by which the proteasome is recruited to TOP2ccs . In this study , we describe a mechanism by which ETO-induced genotoxicity can be abolished by inhibiting proteasome mediated degradation of TOP2 cleavage complexes ( TOP2ccs ) . In the absence of proteolytic degradation , TOP2 remains enzymatically competent and is able to reseal the protein-linked DSB without invoking a DNA damage response ( DDR ) . However , once degradation commences , a TOP2cc is no longer reversible , and the previously hidden DSB become unmasked , triggering a potent DDR signaling response . Following degradation of a TOP2cc , the resultant protein free DSBs appear to engage multiple repair pathways , including error-free NHEJ ( Gómez-Herreros et al . , 2013; Gómez-Herreros et al . , 2017; Figure 7A ) . However , due to the large number of lesions induced by ETO , many DSBs remain unrepaired or mis-repaired which can destabilize the genome . In contrast , by preventing proteasome-mediated unmasking of TOP2-mediated DSBs , error-prone repair is not engaged , thereby facilitating the reversal of TOP2ccs through completion of the enzymes' catalytic cycle upon drug removal , and preserving genome integrity ( Figure 7B ) . Both proteasome inhibitors ( e . g . BTZ ) and topoisomerase poisons ( e . g . ETO , doxorubicin ) are used clinically as therapeutic anti-cancer agents , sometimes in combination ( Cowell and Austin , 2012; Dittus et al . , 2018; Manasanch and Orlowski , 2017; Nitiss , 2009b; Palumbo et al . , 2008; Thomas et al . , 2017 ) . Notably , prior studies have shown that proteasome inhibitors potentiate the cytotoxicity of TOP2 poisons ( Ceruti et al . , 2006; Destanovic et al . , 2018; Dittus et al . , 2018; Lee et al . , 2016; von Metzler et al . , 2009 ) , which at first glance appeared to be at odds with our observation that proteasome inhibition enhances genome stability and survival of ETO-treated cells . We addressed this apparent discrepancy by showing that the timing of proteasome inhibition has a remarkable impact on the outcome of ETO treatment . Inhibiting proteasomal function prior to or during ETO administration significantly improved genome stability by minimizing DSB mis-repair associated with TOP2 proteolysis . However , inhibiting the proteasome after ETO treatment did not effectively protect cells from ETO-induced genome instability , presumably because proteasome-mediated degradation of TOP2ccs is an extremely fast process . Taken together , our study suggests that caution needs to be taken when scheduling therapeutic regimens combining topoisomerase poisons with proteasome inhibitors . Our analysis also revealed that TOP2-mediated DSBs , especially ones that are not promptly repaired , undergo extensive DNA end-resection . We interpreted this as an indicator of active DNA repair occurring at TOP2ccs that have been partially or completely degraded by the proteasome . By contrast , when proteasomal degradation was inhibited by BTZ , we could only detect trace levels of DNA end-resection . This suggests the possibility that MRE11 , which is required for DNA end-resection , may not recognize unprocessed TOP2 . Similar to TOP2 , the meiotic recombination initiator SPO11 forms protein-linked DSBs , which are thought to trigger MRE11 nuclease removal of SPO11 ( Neale et al . , 2005 ) . However , recent studies reveal that mammalian meiotic cells accumulate significant levels of DNA-bound SPO11 , which could , in principal be partially proteolyzed ( Paiano et al . , 2020 ) . While the mechanism by which TOP2 ( and perhaps SPO11 ) shields its associated DSB from early detection by DDR surveillance factors is unclear , our results highlight that γ-H2AX is only a reliable marker for proteolytically degraded TOP2ccs . In contrast , END-seq can be used to distinguish intact TOP2ccs , partially proteolyzed TOP2ccs , as well as fully processed protein-free DSBs . Finally , we found that the SUMO-targeted ubiquitin ligase RNF4 played a critical role in recruiting the proteasome to degrade TOP2ccs , confirming a report by another group ( Sun et al . , 2019 ) . The loss of RNF4 afforded cells significant protection from the genotoxic effects of ETO , further highlighting how impairing proteasome-mediated degradation of TOP2ccs can preserve genome integrity . However , RNF4 deletion did not prevent ETO-induced chromosomal aberrations or blunt the γ-H2AX response as efficiently as chemical proteasome inhibition . This indicates the likely existence of additional ubiquitin ligases that are functionally redundant to RNF4 . Nevertheless , our data suggests that RNF4 could be potentially used as a biomarker to predict the effectiveness of chemotherapeutic regimens that incorporate topoisomerase poisons . In addition , de-regulation of RNF4 may represent a potential mechanism for tumors to acquire resistance to topoisomerase poisons by favoring TOP2cc reversal over degradation . C57BL/6 WT ( NCI mouse repository ) , POLQ-/- ( provided by A . D’Andrea ) , H2AX-/- ( Celeste et al . , 2002 ) , ATM-/- ( provided by A . Wynshaw-Boris ) and conditional Lig4-/- ( Lig4f/f;ERT2-Cre , provided by P . McKinnon ) , BRCA1Δ11BRCA1f/f;CD19Cre , Zong et al . ( 2019 ) , BRCA2-/- ( BRCA2f/f;CD19Cre , provided by S . Sharan ) and RNF4-/- ( RNF4f/f;CD19Cre ) mice between 8 and 18 weeks of age were used to prepare single cell suspensions of primary splenic B-cells . All animal experiments were approved by the NCI Animal Care and Use Committee ( Protocol Numbers: EIB-064–3 and 17–042 ) . Mature resting B-cells were isolated from mouse spleen with anti-CD43 MicroBeads ( Miltenyi Biotech ) . B-cells were activated with LPS ( 25 μg/ml; Sigma ) , IL-4 ( 5 ng/ml; Sigma-Aldrich ) and RP105 ( 0 . 5 μg/ml; BD Biosciences ) for 12 hr as described ( Barlow et al . , 2013; Callén et al . , 2007 ) . Lymphocyte separation media ( Corning ) was used after harvesting the B-cells to separate dead cells from live cells prior to use for END-seq protocol . Cellular Viability was measured using CellTiter-Glo Cell Viability Assay Kit ( Promega ) according to manufactures directions . Luminescence was quantified using a Tecan Infinite M200 Pro plate reader . B-cells were pre-treated with proteasome inhibitor , either 10 μM MG132 ( Sigma-Aldrich ) or 5 μM Bortezomib ( Millipore ) for 1 hr then co-treated for an additional 2 hr with 50 μM Etoposide ( E1383 , Sigma-Aldrich ) . Following the ETO pulse , the cells were either harvested in the presence of the drugs , or washed three times in ice-cold drug free media ( 3 × 5 min spin at 1500 rpm and 4°C to pellet B-cells between washes; 15–20 min total time to complete washout ) , and returned to 37°C to recover in fresh drug free media , until they were harvested at various timepoints post-washout for END-seq , western blot , Immunofluorescence , ICE assay , or for mitotic spread analysis as described in Figure 1A . Asynchronous WT B-cells were treated 24 hr post-activation for 2 hr with 50 μM ETO or 12 . 5 μM Camptothecin ( Sigma-Aldrich ) ±1 hr 10 μM MG132 pre-treatment , then fixed following a 24 hr recovery in drug free media for mitotic spread analysis . Abelson-transformed pre-B cells ( Bredemeyer et al . , 2006 ) were retrovirally transduced with the tetracycline-inducible ER-AsiSI , pTRE3G-HA-ER-AsiSI as previously described ( Callen et al . , 2013 ) . WT and RNF4-/- MEFs ( generated by Gary Lyons ) ( Hu et al . , 2010 ) were cultured in DMEM ( Invitrogen ) + 10% FBS and 1% PenStrep ( Invitrogen ) . MEFs were treated with 5–20 μM ETO for 1 hr ±1 μM BTZ 1 hr pretreatment . For END-seq experiments that used Zinc Finger Nuclease ( ZFN ) spike-in normalization , induction of ZFN was done in G1-arrested Lig4-/- pre B cells ( Bredemeyer et al . , 2006 ) by treating with imatinib for 48 hr and 1 μg/ml Doxycycline during the last 24 hr as described ( Canela et al . , 2016; Bredemeyer et al . , 2006 ) . WT and Lig4-/- mouse pre B cell lines were provided by B . Sleckman . WT and RNF4-/- mouse embryonic fibroblasts ( MEFs ) were provided by S . Bunting . These cell lines are negative for mycoplasma . Western blotting for γ-H2AX and pan-p53 was performed using γ-H2AX antibody ( Millipore ) at 1:5000 dilution and p53 antibody ( Cell Signaling ) at 1:2000 dilution . Antibodies recognizing tubulin ( Sigma-Aldrich , 1:5000 ) , H2AX ( Millipore , 1:5000 ) and vinculin ( Cell Signaling , 1:2000 ) were used to control for protein loading . Image analysis was done using Li-Cor Odyssey CLx to quantify band intensity . Topoisomerase II DNA-protein complexes ( TOP2-DPCs ) were isolated and detected using in vivo complex of enzyme ( ICE ) assay as previously described ( Anand et al . , 2018 ) . Briefly , 5 million cells were lysed in sarkosyl solution ( 1% w/v ) after treatment . Cell lysates were sheared through a 25 g 5/8 needle ( 10 strokes ) to reduce the viscosity of DNA and layered onto CsCl solution ( 150% w/v ) , followed by centrifugation in NVT 65 . 2 rotor ( Beckman coulter ) at 42 , 000 RPM for 20 hr at 25°C . The resulting pellet containing nucleic acids and TOP2-DPCs was obtained and dissolved in TE buffer . The samples were subjected to immunoblotting ( slot blot ) with anti-mouse TOP2β antibody ( Novus ) . 2 μg of DNA is applied per sample . TOP2-DPCs were quantified by densitometric analysis using ImageJ . 5 million cells were harvested 24 hr after drugs were washed out for metaphase analysis as described ( Callén et al . , 2007 ) . Quantitative FISH analysis using a Cy3-labeled ( CCCTAA ) peptide nucleic acid probe ( Applied Biosystems ) was performed as described previously ( Callén et al . , 2007 ) . Telomere length measurements were performed on least 15 metaphases for each cell type . DAPI chromosome and Cy3 telomere images were acquired with a constant exposure time that ensured all captured fluorescent signals were within the linear range using Metafer Software on a Zeiss Axios Imager Z2 Microscope . Image analysis was done in the Metafer ISIS software . SKY was performed and analyzed as previously described , using the hiSKY 7 . 2 . 7 Software ( ASI ) on a Leica DMRXA Microscope ( Liyanage et al . , 1996 ) . A minimum of 35 metaphases were imaged and analyzed . WT eHAP cells were transfected for 48 hr using Xtreme Gene 9 transfection reagent with YFPu Degron reporter ( 0 . 5 ng/μL ) as described ( Bence et al . , 2005; Bence et al . , 2001 ) . eHAP cells were then treated 48 hr post-transfection with 1 μM BTZ and 5 μM MG132 for 2 hr to mimic the total amount of time the proteasome inhibitors are incubated in primary B-cells , after which the cells were washed three times and left to recover at 37°C for up to 18 hr . Cells were fixed at 2 hr , 6 hr , and 18 hr after washout with 4% PFA and YFP expression was measured on a FACS CantoII ( BD biosciences ) . To measure DNA synthesis , primary B-cell cultures were stimulated for 12 hr , 24 hr , 28 hr or 36 hr pulsed with 10 μM of EdU ( 5-ethynyl-2′-deoxyuridine ) for 30 min at 37°C and stained using the Click-IT EdU Alexa Fluor 488 Flow Cytometry Assay Kit according to the manufacturer’s specifications ( Thermo Fisher C10425 ) . Samples were acquired on a FACS CantoII ( BD biosciences ) . Immunofluorescence staining with γ-H2AX antibody ( Millipore , 1:10 , 000 ) was performed in parallel in WT pre B cells ( Bredemeyer et al . , 2006 ) to verify DSB induction . Single cell suspensions of primary B-cells ( 15–20 million ) were untreated or treated with drugs to inhibit the proteasome and/or etoposide as indicated in the cell culture methods section . Primary B-cells were washed twice in cold PBS and resuspended in cold cell suspension buffer ( Bio-Rad CHEF Mammalian Genomic DNA plug kit ) , equilibrated for 5 min at room temperature , mixed with 2% melted CleanCut agarose ( Bio-Rad CHEF Mammalian Genomic DNA plug kit ) prewarmed at 37°C for a final concentration of 0 . 75% , and transferred immediately into plug molds and let them solidify at 4°C for 20 min . Embedded cells were lysed and digested using Proteinase K ( 50°C , 1 hr then 37°C for 7 hr ) . Etoposide and drugs were maintained at the same experimental concentrations in all the steps until digestion with Proteinase K . Plugs were rinsed in TE buffer and treated with RNaseA at 37°C , 1 hr . The next enzymatic reactions were performed with the DNA in agarose plugs to prevent shearing . DNA ends were blunted for 1 hr at 37°C with Exonuclease VII ( ExoVII ) followed by Exonuclease T ( ExoT ) ( NEB ) for 1 hr at 25°C to detect all TOP2-mediated DSBs ( protein-linked and protein-free ) , or only with Exonuclease T ( NEB ) for 1 hr at 25°C to detect only protein-free DSBs . After blunting , A-tailing was performed to attach dA to the free 3’-OH , followed by ligation of ‘‘END-seq hairpin adaptor 1 , ’’ listed in reagents section , using NEB Quick Ligase . Agarose plugs were then melted and dialyzed , and DNA was sonicated to a median shear length of 170 bp using Covaris S220 sonicator for 4 min at 10% duty cycle , peak incident power 175 , 200 cycles per burst , at 4°C . DNA was ethanol-precipitated and dissolved in 70 ml TE buffer . Biotinylated DNA was isolated using MyOne Streptavidin C1 Beads ( Thermo Fisher #650–01 ) , followed by end repair ( dNTPs , T4 polymerase ( NEB ) , Klenow ( NEB ) , T4 PNK ) and dA-tailing ( Klenow exo- ( NEB ) , dATP ) . The second end was ligated to ‘‘END-seq hairpin adaptor 2’’ using NEB Quick Ligase . Hairpins were digested using USER ( NEB ) , and the resulting DNA fragments were PCR amplified using TruSeq barcoded primer p5 , AATGATACGGCGACCACCGAGATCTACACNNNNNNNNACACTCTTTCCCTACACGACGCTCTTCCGATCT and TruSeq barcoded primer p7 , CAAGCAGAAGACGGCATACGAGANNNNNNNGTGACTGGAGTTCAGACGTGTGCTCTTCCGATCT , ( NNNNNNNN represents barcode a * a phosphothiorate bond ) listed in reagents . PCR fragments were isolated by size selection from agarose gel , selecting 200–500 bp fragments followed by DNA purification using QIAquick Gel Extraction Kit . Libraries were quantified using KAPA Library Quantification Kit and sequenced using Illumina NextSeq 500 or 550 . A detailed END-seq protocol can be found in Canela et al . ( 2017 ) , as well as more information about the ExoVII and ExoT processing of TOP2-mediated DSBs in Canela et al . ( 2019 ) . Statistical analysis was performed using R version 3 . 5 . 0 ( http://www . r-project . org ) . The statistical tests are reported in the figure legend and main text .	Molecules of DNA contain the archive of a cell’s genetic information and identity . DNA comprises two strands that twist together into a structure known as a double helix . Physical tension tends to build up in the double helix that can cause it to break apart . To avoid this , cells have an enzyme called Topoisomerase II ( TOP2 ) that relieves the tension by attaching itself to DNA and breaking it in a controlled way before re-sealing the break . Drugs known as TOP2 poisons stop TOP2 from working and trap it on the DNA , which may lead to cells accumulating DNA breaks and eventually dying . Cancer cells are particularly prone to acquiring breaks in their DNA , and TOP2 poisons are therefore often used as part of chemotherapy treatments for cancer . However , it remains unclear why TOP2 poisons are more effective at killing some types of cancer cells than others . It is thought that a molecular machine , known as the proteasome , helps cells repair the damage caused by TOP2 poisons by removing the trapped TOP2 proteins and allowing DNA repair proteins access to the broken DNA underneath . Now , Sciascia et al . have used a genetic approach to study the relationship between the proteasome and DNA repair in mouse cells exposed to TOP2 poisons . The experiments found that when the proteasome removed TOP2 proteins that had become trapped on DNA , the subsequent DNA repair was prone to errors . Pre-treating mouse cells with another drug that inhibited the proteasome protected the cells from the effects of the TOP2 poison . Once the TOP2 poison had left the cells , the previously trapped TOP2 proteins correctly fixed the DNA and detached as they would normally . As a result , cells that had been treated with a proteasome inhibitor were more likely to survive treatment with TOP2 poisons . Since both TOP2 poisons and proteasome inhibitors are clinically approved drugs for treating cancer they can be , and already have been , tested for use together in combination drug therapies . However , these findings suggest that caution should be taken when using these drugs together , because instead of harming the cancer cells , the proteasome inhibitors may protect the cells from the toxic effects of TOP2 poisons .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "genetics", "and", "genomics" ]	2020	Suppressing proteasome mediated processing of topoisomerase II DNA-protein complexes preserves genome integrity
During continuous speech , lip movements provide visual temporal signals that facilitate speech processing . Here , using MEG we directly investigated how these visual signals interact with rhythmic brain activity in participants listening to and seeing the speaker . First , we investigated coherence between oscillatory brain activity and speaker’s lip movements and demonstrated significant entrainment in visual cortex . We then used partial coherence to remove contributions of the coherent auditory speech signal from the lip-brain coherence . Comparing this synchronization between different attention conditions revealed that attending visual speech enhances the coherence between activity in visual cortex and the speaker’s lips . Further , we identified a significant partial coherence between left motor cortex and lip movements and this partial coherence directly predicted comprehension accuracy . Our results emphasize the importance of visually entrained and attention-modulated rhythmic brain activity for the enhancement of audiovisual speech processing . Communication is one of the most fundamental and complex cognitive acts humans engage in . In a dialogue , a large range of dynamic signals are exchanged between interlocutors including body posture , ( emotional ) facial expressions , head and eye movements , gestures and a rich acoustic speech signal . Movements of the lips contain sufficient information to allow trained observers to comprehend speech through visual signals . Even for untrained observers and in the presence of auditory signals , lip movements can be beneficial for speech comprehension when the acoustic signal is degraded ( Peelle and Sommers , 2015; Sumby and Pollack , 1954; van Wassenhove et al . , 2005; Zion-Golumbic and Schroeder , 2012 ) . Dynamic lip movements support disambiguation of syllables and can provide temporal onset cues for upcoming words or syllables ( Chandrasekaran et al . , 2009; Grant and Seitz , 2000; Kim and Davis , 2003; Schroeder et al . , 2008; Schwartz and Savariaux , 2014 ) . However , it has remained unclear how dynamic lip movements during continuous speech are represented in the brain and how these visual representations interact with the encoding of the acoustic speech signal . A potential underlying mechanism for the visual enhancement of hearing could be the synchronization of brain rhythms between interlocutors , which has been implicated in the encoding of acoustic speech ( Giraud and Poeppel , 2012; Hasson et al . , 2012; Pickering and Garrod , 2013 ) . Indeed , continuous speech and the associated lip movements show temporal modulations at the syllabic rate ( 3–8 Hz ) ( Chandrasekaran et al . , 2009 ) . These signals produced in the speaker’s motor system supposedly lead to resonance in the listener’s brain that facilitates speech comprehension ( Giraud and Poeppel , 2012 ) . The hallmark of such a process is the synchronization of brain activity at the frequency of dominant rhythmic components in the communication signal ( Schroeder et al . , 2008 ) . Consistent with this idea , previous studies have demonstrated the frequency-specific synchronization between brain activity and continuous auditory speech signals ( Ahissar et al . , 2001; Ding and Simon , 2012; Gross et al . , 2013b; Luo and Poeppel , 2007; Peelle et al . , 2013 ) at frequencies below 10 Hz . This synchronization was found to be stronger for intelligible than non-intelligible speech and facilitated by top-down signals from left inferior frontal and motor areas ( Ding and Simon , 2014; Kayser et al . , 2015; Park et al . , 2015 ) as well as during attention ( Zion-Golumbic and Schroeder , 2012 ) . The correlated temporal dynamics of the acoustic and lip signals raise the possibility that lip-mediated benefits for hearing rely on similar entrainment mechanisms in the observer as the acoustic component . This is plausible as the auditory speech entrainment and speech intelligibility are enhanced when congruent visual speech is present ( Crosse et al . , 2015; Zion Golumbic et al . , 2013 ) . Still , the neural representations of dynamic lip signals and their dependence on attention and the acoustic speech component remain unclear . Here we directly tested four hypotheses: First , we hypothesized that rhythmic components in visual speech entrain brain activity in the observer . Second , to test whether benefits arising from seeing the speaker’s lip movements are mediated through mechanism other than those implicated in auditory entrainment , we asked whether and which brain areas synchronize to lip movements independently of auditory signals . Third , we hypothesized that the synchronization between visual speech and brain activity is modulated by attention and congruence of visual and auditory signals . Finally , we tested whether any observed synchronization is relevant for speech comprehension . We recorded MEG signals while participants perceived continuous audiovisual speech . To dissociate the synchronization to attended and unattended visual and acoustic signals , we manipulated the congruency of visual and acoustic stimuli in four experimental conditions ( Figure 1A ) . 10 . 7554/eLife . 14521 . 003Figure 1 . Experimental conditions and behavioral results . ( A ) Four experimental conditions . ‘A’ denotes auditory stimulus and ‘V’ denotes visual stimulus . The number refers to the identity of each talk . All congruent condition: Natural audiovisual speech condition where auditory stimuli to both ears and visual stimuli are congruent ( from the same movie; A1 , A1 , V1 ) . All incongruent condition: All three stimuli are from different movies ( A2 , A3 , V4 ) and participants are instructed to attend to auditory information presented to one ear . AV congruent condition: Auditory stimulus presented to one ear matches the visual information ( A5 , A6 , V5 ) . Participants attend to the talk that matches visual information . AV incongruent condition: Auditory stimulus presented to one ear matches the visual information ( A7 , A8 , V8 ) . Participants attend to the talk that does not match the visual information . Attended stimulus is marked as red color for the group attended to the left side ( see Materials and methods for details ) . ( B ) Behavioral accuracy by comprehension questionnaires . Congruent conditions show high accuracy rate compared to incongruent conditions ( %; mean ± s . e . m . ) : All congruent: 85 ± 1 . 66 , All incongruent: 77 . 73 ± 2 . 15 , AV congruent: 83 . 40 ± 1 . 73 , AV incongruent: 75 . 68 ± 2 . 88 ) . Statistics between conditions show significant difference only between congruent and incongruent conditions ( paired t-test , df: 43 , p<0 . 05 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 14521 . 003 The four experimental conditions were designed to modulate congruence and informativeness of visual and auditory stimuli ( Figure 1A ) . In each condition , one visual stimulus was presented and two ( identical or different ) speech streams were presented to the left and the right ears , respectively ( see Materials and methods for details ) . The All congruent condition consisted of three congruent stimuli . The All incongruent condition had three incongruent stimuli . In the AV congruent condition , participants attended an auditory stimulus that had a congruent visual stimulus with an additional incongruent auditory stimulus . In the AV incongruent condition participants attended an auditory stimulus that was incongruent to a congruent audiovisual stimulus pair . Overall , participants showed high comprehension accuracy across conditions ( %; mean ± s . e . m . ) : All congruent: 85 ± 1 . 66 , All incongruent: 77 . 73 ± 2 . 15 , AV congruent: 83 . 40 ± 1 . 73 , AV incongruent: 75 . 68 ± 2 . 88 ) . As expected , accuracy was significantly higher when the visual stimulus was congruent with attended auditory stimulus ( i . e . , All congruent and AV congruent conditions ) compared to when the visual stimulus was incongruent with attended auditory stimulus ( i . e . , All incongruent and AV incongruent conditions ) ( Figure 1B; paired t-test , df: 43 , p<0 . 05; All congruent vs . All incongruent: t = 3 . 09 , p=0 . 003 , All congruent vs . AV congruent: t = 0 . 76 , p=0 . 45 ( n . s . ) , All congruent vs . AV incongruent: t = 2 . 98 , p=0 . 004 , AV congruent vs . All incongruent: t = 2 . 15 , p=0 . 03 , AV congruent vs . AV incongruent: t = 2 . 24 , p=0 . 03 , All incongruent vs . AV incongruent: t = 0 . 65 , p=0 . 52 ( n . s . ) ) . Interestingly , performance for AV congruent condition was not significantly different to performance in All congruent condition despite the interfering auditory input . This is likely caused by attentional efforts to overcome interfering input leading to behavioral compensation . To examine the frequency spectrum of the lip signal , we computed the lip area for each video frame ( Figure 2A and Figure 2—figure supplement 1A , B , C ) . The signal is dominated by low-frequency components from 0 to 7 Hz peaking around 0 to 4 Hz ( Figure 2B; from all lip speech signals used in this study; mean ± s . e . m . ) . Next , we computed coherence between these lip signals and the respective acoustic signals to investigate the relationship between visual and auditory components in audiovisual speech . This was computed for all talks used in the study and averaged . The coherence spectrum reveals a prominent peak in a frequency band corresponding to the syllable rate around 4–8 Hz ( red line; mean ± s . e . m . ) ( Figure 2C ) . These results demonstrate the temporal coupling of auditory and visual speech components . 10 . 7554/eLife . 14521 . 004Figure 2 . Lip signals in continuous speech and its entrainment in the brain . ( A ) Lip signals in the continuous audiovisual speech . Lip contour was extracted for each video frame and corresponding area was computed ( see Figure 2—figure supplement 1A , B , C for details ) . One representative lip speech signal ( for around 14 s speech ) is shown here . Speaker’s face is cropped for this publication only but not in the original stimuli . ( B ) Spectral profile of lip speech signals . The power spectra of lip speech signals used in this study were averaged ( mean ± s . e . m . ) . Signal is dominated by low-frequency components from 0 to 7 Hz that robustly peak around 0 to 4 Hz corresponding to delta and theta band neuronal oscillations in the brain . ( C ) Coupling between lip and sound speech signals by coherence . Coherence between matching ( red line ) and non-matching ( blue line ) lip and sound speech signals were computed and averaged across talks used in the study ( mean ± s . e . m . ) . ( D ) Lip speech entrainment in natural audiovisual speech ( All congruent condition ) . Coherence was computed between lip speech signal and brain activity at each voxel and then statistically compared to surrogate data at 1 Hz ( the dominant frequency in the power spectrum in ( b ) ; p<0 . 05 , FDR corrected ) . ( E ) Sound speech entrainment in natural audiovisual speech ( All congruent condition ) . Using sound speech envelope , the same computation described in ( D ) was performed to investigate sound speech entrainment effect ( p<0 . 05 , FDR corrected ) . ( F ) Lip speech- and sound speech-specific entrainment effects . Lip speech ( D ) and sound speech coherence ( E ) were statistically compared ( p<0 . 05 , FDR corrected ) . DOI: http://dx . doi . org/10 . 7554/eLife . 14521 . 00410 . 7554/eLife . 14521 . 005Figure 2—figure supplement 1 . A schematic figure for the analysis of coupling between lip movements and brain activity . We extracted lip movement signals by automatically computing for each video frame the lip contour using our in-house MATLAB script ( A ) . The contour was converted into the three quantities area , major axis and minor axis ( B ) . We then computed the coherence between these measures ( C ) . Area and minor axis ( representing vertical lip movement ) information have almost identical information approaching 1 . 0 in coherence value ( red line; mean ± s . e . m . ) . However , major axis information representing horizontal lip movement only does not show similar pattern with either area or minor axis ( purple and blue lines; mean ± s . e . m . ) . Since area information represents both horizontal and vertical lip movement , we used this information for further analysis . We then computed coherence between lip area signals and MEG signals at each voxel for each frequency band and for all four experimental conditions ( D; see Materials and methods ) . DOI: http://dx . doi . org/10 . 7554/eLife . 14521 . 00510 . 7554/eLife . 14521 . 006Figure 2—figure supplement 2 . Brain activity entrained by lip movements . We show lip-entrained brain activity by coherence in natural audiovisual speech condition ( All congruent ) at 1 Hz ( dominant frequency component in lip signals ) in Figure 2D . Here we show the same effect for other experimental conditions compared to surrogate data . ( A ) AV congruent: Lip entrainment in the AV congruent condition showed similar pattern to All congruent condition ( Figure 2D ) . Distributed bilateral visual areas , auditory areas , left pre/postcentral gyri and right postcentral , supramarginal gyri were observed ( p<0 . 05 , FDR corrected ) . ( B ) All incongruent: Lip entrainment in this condition revealed increased involvement of bilateral visual cortex . This result confirms that there is clear lip entrainment effect irrespective of sound speech entrainment since this shows robust visual entrainment in the absence of a congruent sound speech input in All incongruent condition ( p<0 . 05 , FDR corrected ) . ( C ) AV incongruent: Compared to the other conditions lip entrainment is evident but reduced . Even though there is a congruent sound speech input as in the AV congruent condition , this does not reveal lip entrainment effect as strong as in the AV congruent condition providing more evidence that lip entrainment is modulated by attention ( p <0 . 05 , FDR corrected ) . DOI: http://dx . doi . org/10 . 7554/eLife . 14521 . 00610 . 7554/eLife . 14521 . 007Figure 2—figure supplement 3 . Brain activity entrained by lip movements and sound envelope . We show entrained brain activity by coherence separately for lip movements ( A ) and sound speech envelope ( B ) at each frequency from 2 to 5 Hz for All congruent condition in addition to 1 Hz shown in Figure 2D and E . DOI: http://dx . doi . org/10 . 7554/eLife . 14521 . 007 First , we tested the hypothesis that lip movements entrain the observer’s brain activity . We addressed this by computing coherence between the lip signal and brain signal at each voxel at frequencies ranging from 1 to 7 Hz ( in 1 Hz steps ) covering the spectral profile of the lip signals ( Figure 2B ) . In addition , as a control , we computed surrogate maps ( from time-shifted lip signals , thereby destroying physiologically meaningful coherence ) as an estimate of spatially and spectrally specific biases of the analysis . We first compared natural audiovisual speech condition ( All congruent ) and surrogate data for the frequency that showed strongest power in the lip signal ( 1 Hz ) . This revealed a significant entrainment effect in visual , auditory , and language areas bilaterally ( p<0 . 05 , false discovery rate ( FDR ) corrected; Figure 2D ) . The areas include early visual ( V1; Calcarine sulcus ) and auditory ( A1; Heschl’s gyrus ) areas as well as inferior frontal gyrus ( IFG; BA 44 ) ( see Figure 2—figure supplement 2 for the other conditions at 1 Hz ) . However , since the speech envelope and lip movements are coherent ( Figure 2C ) , it may be that this lip entrainment is induced by speech entrainment and not by lip movements per se . Thus , we performed the same coherence analysis for the sound speech envelope . In accordance with previous work ( Gross et al . , 2013b ) , we observed an extensive auditory network including Heschl’s gyrus and superior/middle temporal gyri bilaterally and left frontal areas ( p<0 . 05 , FDR corrected; Figure 2E ) ( see Figure 2—figure supplement 3 for different frequencies [2–5 Hz] ) . Statistical comparison of lip movement entrainment ( Figure 2D ) to sound speech entrainment ( Figure 2E ) revealed significantly stronger lip entrainment in bilateral visual areas and stronger sound speech coherence in right superior temporal gyrus ( p<0 . 05 , FDR corrected; Figure 2F ) . This demonstrates significant entrainment of brain activity to the lip movements irrespective of entrainment to the acoustic speech signal . In addition , we found significant lip movement entrainment in visual areas in the absence of a congruent auditory stimulus ( Figure 2—figure supplement 2B ) . These results demonstrate for the first time the entrainment of cortical brain oscillations to lip movements during continuous speech . Next , we compared visual lip entrainment across conditions to test our hypothesis that entrainment changes with attention and the congruence of audiovisual stimuli . We focused our analysis on AV congruent condition where a distracting auditory speech stream is presented to one ear . Compared to the All congruent condition , AV congruent demands additional attention to visual speech because the visual signal is informative to disambiguate the two incongruent auditory streams . We therefore contrasted AV congruent with All congruent condition to capture the effect of visual attention . We also contrasted AV congruent condition with All incongruent condition to capture the effect of congruence . Since lip and sound speech signals are coherent ( Figure 2C ) , it is difficult to disentangle visual and auditory contributions to the lip movement entrainment . To measure lip-specific entrainment effects more directly , we computed partial coherence between lip movement signals and brain activity while removing the contribution of acoustic speech signals . This provides an estimate of entrainment by lip movement signals that cannot be explained by acoustic speech signals and allowed us to test our second hypothesis that lip entrainment is not mediated via acoustic entrainment . However , we repeated the same analysis using coherence instead of partial coherence ( Figure 3—figure supplement 1 ) . First , we identified the frequency band showing the strongest attention effect by averaging voxels across visual cortex ( superior/middle/inferior occipital gyri ) defined from AAL ( Automated Anatomical Labeling ) ROI ( Region-of-Interest ) map . We averaged the partial coherence values within the ROI and statistically compared the AV congruent to surrogate data and to All congruent condition at each frequency from 1 to 7 Hz ( Figure 3A; paired t-test , df: 43 , red dashed line: p<0 . 05 , gray dashed line: p<0 . 05 , corrected ) . This revealed significantly stronger lip movement entrainment at 4 Hz in the left visual cortex for AV congruent compared to both , All congruent condition and surrogate data . In AV congruent condition , lip movements are informative and assist comprehension . This result suggests that coupling of low-frequency brain activity to lip movements is enhanced by visual attention . 10 . 7554/eLife . 14521 . 008Figure 3 . Lip-brain partial coherence . ( A ) Modulation of partial lip-brain coherence by attention and congruence in visual ROI . AV congruent condition was compared to the other conditions ( paired t-test , df: 43 , red dashed line: p<0 . 05 , gray dashed line: p<0 . 05 , corrected ) . ( B , C , D ) Attention-modulated partial coherence at each brain voxel ( AV congruent versus surrogate ( B ) , All incongruent ( C ) , All congruent ( D ) ) . It shows significant involvement of left motor cortex ( precentral gyrus; BA 4/6 ) and left visual areas ( p<0 . 05 , FDR corrected; but in ( D ) , left motor cortex is observed at uncorrected p<0 . 05 ) . Entrainment in the left motor cortex shows a systematic modulation such that statistical contrast with a strong difference in visual attention show stronger entrainment ( AV congruent versus surrogate ( B; t43-value: 3 . 42 ) > All incongruent ( C; t43-value: 3 . 20 ) > All congruent ( D; t43-value: 2 . 24 ) ) . DOI: http://dx . doi . org/10 . 7554/eLife . 14521 . 00810 . 7554/eLife . 14521 . 009Figure 3—figure supplement 1 . Lip-brain coherence . ( A , B ) Modulation of lip entrainment by attention and congruence in ROIs . Contrasting AV congruent with All congruent for the effect of attention and with All incongruent for the effect of congruence are shown in left visual ( superior/middle/inferior occipital gyri ) and right auditory ( Heschl’s gyrus ) areas defined by AAL ROI maps ( paired t-test , df: 43 , red dashed line: p<0 . 05 , gray dashed line: p<0 . 05 , corrected ) at each frequency from 1 to 7 Hz . The strongest difference was observed at 4 Hz for both contrasts in both areas . ( C , D ) Spatial distribution of lip entrainment modulation by attention and congruence at 4 Hz ( p<0 . 05 , FDR corrected ) . Significant effects are observed in left visual areas , right superior and middle temporal gyri including Heschl’s gyrus , posterior superior temporal sulcus ( pSTS ) , left precentral gyrus and right inferior frontal gyrus ( IFG; BA 44/6 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 14521 . 00910 . 7554/eLife . 14521 . 010Figure 3—figure supplement 2 . Partial coherence between lip movements and left motor cortex . We here show a frequency-specific plot for partial coherence between lip movements and left motor cortex . We extracted the maximum voxel in the left motor cortex ( precentral gyrus ) from the contrast for attention effect ( AV congruent vs . All congruent shown in Figure 3D; MNI coordinates = [-44 -16 56] ) . This confirmed lip movements entrain left motor cortex at 4 Hz irrespective of auditory speech signals . DOI: http://dx . doi . org/10 . 7554/eLife . 14521 . 010 Next , we studied attentional lip movement entrainment at 4 Hz across the entire brain . This revealed a significant partial coherence between lip movements and left motor cortex ( precentral gyrus; BA 4/6 ) in addition to the left visual areas ( p<0 . 05 , FDR corrected; Figure 3B , C , D ) . Entrainment in the left motor cortex shows a systematic modulation by attention . Specifically , contrasts with a stronger difference in visual informativeness exhibit stronger lip entrainment ( AV congruent versus surrogate ( Figure 3B; t43-value: 3 . 42 ) > All incongruent ( Figure 3C; t43-value: 3 . 20 ) > All congruent ( Figure 3D; t43-value: 2 . 24 ) ) . Contrasting AV congruent with All congruent conditions ( Figure 3D ) , which both have congruent visual speech , also revealed an effect in the same motor cortex area ( p<0 . 05 , uncorrected; for the frequency-specific plot for left motor cortex , see Figure 3—figure supplement 2 ) . In the addition , the left visual cortex shows significantly stronger lip entrainment for AV congruent compared to surrogate data ( Figure 3B ) , All incongruent condition ( Figure 3C ) and All congruent condition ( Figure 3D ) . This demonstrates that activity in left motor cortex and left visual areas show significant alignment to lip movements independent of sound speech signals and that this alignment is stronger when visual speech is more informative and congruent to the auditory stimulus . To address our fourth hypothesis that lip-entrained brain activity has an impact on speech comprehension , we identified brain regions where lip entrainment correlates with behavioral performance . We performed regression analysis using comprehension accuracy across subjects on the partial coherence map at 4 Hz in each condition . We then contrasted the condition with high visual attention and behavioral performance ( AV congruent ) to the condition with low visual attention and behavioral performance ( AV incongruent; see behavioral results in Figure 1B ) . The regression t-values in the two conditions were transformed to standardized Z-value at each voxel and the two Z-maps were subtracted . This revealed that entrainment in left motor cortex predicts attention-modulated comprehension accuracy ( Figure 4A; Z-difference map at p<0 . 005 ) . 10 . 7554/eLife . 14521 . 011Figure 4 . Behavioral correlates of attentional lip entrainment . ( A ) Lip-entrained brain regions predicted by attention-modulated comprehension accuracy . Regression analysis using comprehension accuracy across participants on the partial coherence map was performed at 4 Hz in each condition . Then Z-difference map was obtained from the regression analysis of conditions showing strongest difference in comprehension accuracy ( AV congruent versus AV incongruent; see behavioral results in Figure 1B ) . This revealed that the left motor cortex entrainment predicts attention-modulated comprehension accuracy ( Z-difference map at p<0 . 005 ) . ( B ) Correlation between partial coherence in left motor cortex and comprehension accuracy . Partial coherence values from the maximum coordinate in the left motor cortex and comprehension accuracy across subjects for the AV congruent condition was positively correlated ( Pearson’s coefficient of Fisher’s Z-transformed data R = 0 . 38 , P = 0 . 01; Spearman rank correlation R = 0 . 32 , P = 0 . 03 ) . ( C ) Difference of attentional lip-entrainment in left motor cortex between good and poor performing group . Group t-statistics between good and poor performing group on the extracted partial coherence values in the left motor cortex for the AV congruent condition was performed . The two groups were divided using median value ( 90%; 23 good versus 21 poor performers ) of comprehension accuracy for the AV congruent condition . Good performers showed higher partial coherence value in left motor cortex than poor performers ( two-sample t-test on Fisher’s Z-transformed data; t42 = 2 . 2 , P = 0 . 03 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 14521 . 011 To confirm this effect , we correlated partial coherence values from the maximum coordinate in the left motor cortex with comprehension accuracy across subjects for the AV congruent condition . This revealed significant positive correlation demonstrating that individuals with higher levels of comprehension show higher partial coherence in the left motor cortex . Since the accuracy has discrete values such as 70 , 80 , 90% ( correct response rate out of 10 questions at the comprehension testing ) , we plotted the correlation as a box plot ( Figure 4B; Pearson’s coefficient of Fisher’s Z-transformed data R = 0 . 38 , P = 0 . 01; Spearman rank correlation R = 0 . 32 , P = 0 . 03 ) . To confirm this result in a separate analysis , we performed t-statistics between good and poor performing group on the extracted partial coherence values in the AV congruent condition . The two groups were divided using median value of performance leading to 23 good performers and 21 poor performers . Good performers showed significantly higher partial coherence value in the left motor cortex than poor performers ( Figure 4C; two-sample t-test on Fisher’s Z-transformed data; t42 = 2 . 2 , P = 0 . 03 ) . Taken together , these results demonstrate that stronger attentional lip entrainment in the left motor cortex supports better speech comprehension . Recent studies provide converging evidence that detailed information about the identity or specific features of sensory stimuli can be decoded from the low-frequency phase of LFP or MEG/EEG signals ( Ng et al . , 2013; Panzeri et al . , 2015; Schyns et al . , 2011 ) . For example , a recent study has demonstrated that the identity of different audiovisual movie stimuli can be decoded from delta-theta phase in occipital MEG sensors ( Luo et al . , 2010 ) . In the auditory domain , this stimulus-specificity is at least partly due to a phase synchronization of rhythmic brain activity and the auditory speech envelope ( Gross et al . , 2013b; Peelle et al . , 2013 ) . Recent studies have shown that congruent visual stimulation facilitates auditory speech entrainment ( Crosse et al . , 2015; Zion Golumbic et al . , 2013 ) . Here , we extend these results by showing low-frequency phase synchronization between speakers’ lip movements and listeners’ brain activity . This visual entrainment is clearly distinct from auditory entrainment for three reasons . First , some areas show stronger entrainment to the visual compared than to the auditory stimulus . Second , removing the auditory contribution using partial coherence still results in significant visual entrainment . Third , we report visual entrainment in the absence of a congruent auditory stimulus . Together , this establishes the existence of a visual entrainment mechanism for audiovisual speech in addition to the well-studied auditory entrainment . In correspondence to the auditory domain , this effect relies on the quasi-rhythmic nature of visual speech that is particularly prominent in the subtle but salient lip movements . Our coherence analysis reveals an extensive network comprising speech processing areas such as bilateral primary sensory areas ( V1 ( Calcarine sulcus; BA 17 ) and A1 [Heschl’s gyrus; BA 41] ) , extended sensory visual ( BA 18/19 ) and auditory areas ( superior/middle temporal gyri; BA 21/22/42 ) , and posterior superior temporal sulcus ( pSTS ) as well as inferior frontal gyrus ( IFG; BA 44/6 ) . Although these areas show synchronization to lip movements , there is large overlap with areas showing synchronization to sound envelope . This is not surprising because both stimulus signals are coherent . But as expected , in occipital areas visual entrainment is significantly stronger compared to auditory entrainment . An interesting asymmetry emerges in auditory areas in the comparison of lip entrainment to speech entrainment ( Figure 2F ) . The right superior temporal cortex is significantly stronger coupled to speech envelope than to lip movements while there is no significant difference in left superior temporal cortex . This seems to suggest that visual speech predominantly entrains left temporal areas ( in addition to visual areas ) . This is consistent with a recent fMRI study demonstrating preferential processing of visual speech in left superior temporal areas ( Blank and von Kriegstein , 2013 ) . In summary , we find that visual areas are entrained by lip movements and left and right auditory areas by lip movements and speech envelope with only the right auditory cortex showing a stronger coupling to speech envelope compared to lip movements . We studied the effect of attention in an ecologically valid scenario with congruent audiovisual speech when a distracting auditory stimulus was present ( AV congruent condition ) or absent ( All congruent condition ) . While we did not explicitly manipulate attention in our paradigm , the distracting auditory stimulus in the AV congruent condition renders the visual speech relevant and informative . Attention to visual speech is known to help to disambiguate the competing auditory inputs ( Sumby and Pollack , 1954; Zion Golumbic et al . , 2013 ) . This condition was compared to the All congruent condition where attention to the visual stimulus is not required for speech comprehension . Interestingly , attention to lip movements leads to significantly increased coupling in left hemispheric visual areas . Previous studies have demonstrated that observers process audiovisual speech preferentially based on information from the left side of a speaker’s face ( observers’ right visual hemifield ) as compared to the right side ( Behne et al . , 2008; Campbell et al . , 1996; Smeele et al . , 1998; Swerts and Krahmer , 2008 ) ( but see Nicholls and Searle , 2006 ) . This has been related to an attentional bias to the observers’ right visual hemifield due to the left hemisphere dominance for speech processing ( Thompson et al . , 2004 ) . Since the right visual hemifield is represented in the left visual cortex , this attentional bias could explain the observed lateralization of entrainment . While visual areas show both coherence and partial coherence to lip movements , temporal areas only emerge from coherence analysis ( Figure 2D and Figure 3—figure supplement 1 ) . This seems to suggest that the temporal alignment of auditory cortex activity to lip movements is not independant of the congruent acoustic speech stimulus and therefore speaks against a direct entrainment of auditory cortex activity by lip movements . However , the informativeness of lip movements still modulates significantly the coherence between lip movements and right temporal brain areas ( Figure 3—figure supplement 1B , C , D ) and indicates an indirect effect of visual attention . Similarly , left posterior superior temporal sulcus ( pSTS ) shows a significant effect of congruence ( Figure 3—figure supplement 1D ) that is only evident in coherence and not in partial coherence map . This is consistent with previous reports of pSTS as a locus of multisensory integration ( Hocking and Price , 2008; Noesselt et al . , 2012 ) with particular relevance for visual speech recognition ( Beauchamp et al . , 2004; Blank and von Kriegstein , 2013; Werner and Noppeney , 2010 ) . Attended audiovisual speech leads to significant entrainment in left inferior precentral gyrus corresponding to the lip representation in motor cortex ( Figure 3—figure supplement 1C , D ) ( Giraud et al . , 2007 ) . In addition , partial coherence analysis where the effect of speech envelope on lip entrainment is removed reveals a more superior left lateralized motor area ( Figure 3B , C , D ) . Partial coherence in this area is modulated by congruence ( Figure 3C ) and attention ( Figure 3D ) and predicts comprehension accuracy ( Figure 4A ) . There is ample evidence for the activation of left motor cortex during audiovisual speech perception ( Evans and Davis , 2015; Meister et al . , 2007; Mottonen et al . , 2013; Watkins et al . , 2003; Wilson et al . , 2004; Ylinen et al . , 2015 ) . Our results indicate that left motor areas are not only activated but also entrained by audiovisual speech . This establishes a precise temporal coupling between audiovisual sensory inputs and sensory and motor areas that could temporally coordinate neuronal computations associated with speech processing . As part of a proposed dorsal auditory pathway , motor areas could provide access to an internal forward model of speech that closely interacts with auditory sensory systems ( Bornkessel-Schlesewsky et al . , 2015; Rauschecker and Scott , 2009 ) . During speech production , an efference copy is sent to auditory areas to allow for efficient monitoring and control . During speech perception , sensory signals could be used to constrain the forward model in simulating the speaker’s motor program and predicting upcoming sensory events ( Arnal and Giraud , 2012; Bornkessel-Schlesewsky et al . , 2015; Lakatos et al . , 2013 ) . Indeed , a recent study has demonstrated direct top-down control of left auditory cortex from left motor cortex during continuous speech processing . The strength of low-frequency top-down signals in the left motor cortex correlates with the coupling of auditory cortex and sound speech envelope ( Park et al . , 2015 ) ( see also Kayser et al . , 2015 ) . This suggests that the motor cortex plays a predictive role in speech perception consistent with recent demonstrations of its contribution to the temporal precision of auditory speech perception ( Morillon et al . , 2014; 2015; Wilson et al . , 2008 ) . Visual speech per se is not critical for speech comprehension , however , it facilitates auditory speech processing as it aids temporal prediction and can prime the auditory system for upcoming concordant auditory input ( Peelle and Sommers , 2015; van Wassenhove et al . , 2005 ) . Overall , left motor cortex seems to be an important area for facilitating audiovisual speech processing through predictive control in auditory active sensing ( Meister et al . , 2007; Morillon et al . , 2015 ) . In summary , our study provides the first direct evidence that lip movements during continuous speech entrain visual and motor areas , and that this entrainment is modulated by attention and congruence and is relevant for speech comprehension . This adds to similar findings in the auditory domain and provides a more comprehensive view of how temporally correlated auditory and visual speech signals are processed in the listener’s brain . Overall , our findings support an emerging model where rhythmic audiovisual signals entrain multisensory brain areas and dynamically interact with an internal forward model accessed via the auditory dorsal stream to form dynamically updated predictions that improve further sensory processing . Through these mechanisms brain oscillations might implement inter-subject synchronization and support the surprising efficiency of inter-human communication . Data were obtained from 46 healthy subjects ( 26 females; age range: 18–30 years; mean age: 20 . 54 ± 2 . 58 years ) and they were all right-handed confirmed by Edinburgh Handedness Inventory ( Oldfield , 1971 ) . All participants provided informed written consent before participating in the experiment and received monetary compensation for their participation . All participants had normal or corrected-to-normal vision and normal hearing . None of the participants had a history of developmental , psychological , or neurological disorders . Only native English-speaking volunteers with British nationality were recruited due to the British accent in the stimulus material . Two subjects were excluded from the analysis ( one subject fell asleep and one had MEG signals with excessive noise ) . This left dataset from 44 participants ( 25 females; age range: 18–30 years; mean age: 20 . 45 ± 2 . 55 years ) . This study was approved by the local ethics committee ( CSE01321; University of Glasgow , College of Science and Engineering ) and conducted in conformity with the Declaration of Helsinki . Neuromagnetic signals were obtained with a 248-magnetometers whole-head MEG ( Magnetoencephalography ) system ( MAGNES 3600 WH , 4-D Neuroimaging ) in a magnetically shielded room using a sampling rate of 1017 Hz . MEG signals were denoised with information from the reference sensors using the denoise_pca function in FieldTrip toolbox ( Oostenveld et al . , 2011 ) . Bad sensors were excluded by visual inspection . Electrooculographic ( EOG ) and electrocardiographic ( ECG ) artifacts were eliminated using independent component analysis ( ICA ) . Participants’ eye fixation and movements were recorded during the experiment using an eye tracker ( EyeLink 1000 , SR Research Ltd . ) to ensure that they fixate on the speaker’s lip . T1-weighted structural magnetic resonance images ( MRI ) were acquired at 3 T Siemens Trio Tim scanner ( Siemens , Erlangen , Germany ) with the following parameters: 1 . 0 x 1 . 0 x 1 . 0 mm3 voxels; 192 sagittal slices; Field of view ( FOV ) : 256 x 256 matrix . Data will be available upon request by contacting corresponding authors . The stimuli used in this study were audiovisual video clips of a professional male speaker talking continuously ( 7–9 min ) . The talks were originally taken from TED talks ( www . ted . com/talks/ ) and edited to be appropriate to the stimuli we used ( e . g . editing words referring to visual materials , the gender of the speaker ) . Eleven video clips were filmed by a professional filming company with high-quality audiovisual device and recorded in 1920 x 1080 pixels at 25 fps ( frame per second ) for video and sampling rate of 48 kHz for audio . In a behavioral study , these videos were rated by 33 participants ( 19 females; aged 18–31 years; mean age: 22 . 27 ± 2 . 64 years ) in terms of arousal , familiarity , valence , complexity , significance ( informativeness ) , agreement ( persuasiveness ) , concreteness , self-relatedness , and level of understanding . Participants were instructed to rate each talk on these 9 items using Likert scale ( Likert , 1932 ) 1 to 5 ( for an example of concreteness , 1: very abstract , 2: abstract , 3: neither abstract nor concrete , 4: concrete , 5: very concrete ) . Talks with excessive mean scores ( below 1 and over 4 ) were excluded and 8 talks were selected for the experiment . There were four experimental conditions: All congruent , All incongruent , AV congruent , AV incongruent ( Figure 1A ) . In each condition , one video recording was presented and two ( identical or different ) auditory recordings were presented to the left and the right ear , respectively . All congruent condition . Natural audiovisual speech condition where auditory stimuli to both ears and visual stimuli are congruent ( from the same movie; e . g . A1 , A1 , V1 – where the first A denotes stimulus presented to the left ear , second A denotes stimulus presented to right ear and V denotes visual stimulus . The number refers to the identity of each talk . ) . All incongruent condition . All three stimuli are from different movies ( e . g . A2 , A3 , V4 ) and participants are instructed to attend to auditory information presented to one ear . AV congruent condition . Auditory stimulus presented to one ear matches the visual information ( e . g . A5 , A6 , V5 ) . Participants attend to the talk that matches visual information . AV incongruent condition . Auditory stimulus presented to one ear matches the visual information ( e . g . A7 , A8 , V8 ) . Participants attend to the talk that does not match the visual information . The color ( yellow or blue ) of a small fixation cross which was overlaid on the speaker’s lip indicates the side of attention ( left or right talk , e . g . “If the color of fixation cross is yellow , please attend to left ear talk . ” ) . The functional meaning of the color of fixation cross was counterbalanced across subjects . In All congruent condition ( natural audiovisual speech ) , participants were instructed to ignore the color and just to attend to both sides . Participants were instructed to fixate on the speaker’s lip all the time in all experimental conditions even if they found it difficult to do so in some conditions ( e . g . incongruent conditions; All incongruent and AV incongruent ) . The fixation cross was used to guide participants’ fixation on the speaker’s lip . Furthermore , the gaze behavior was monitored by an eye tracker . The importance of eye fixation on the speaker’s lip was stressed at the task instruction and they were notified that their eye fixation would be monitored by an eye tracker . There were two groups ( 22 subjects each ) . Participants in one group attended to left ear talk and participants in another group attended to right ear talk in the experiment ( for all four conditions ) . Since the attended side is the same within subjects , in order to avoid presenting all the same color of fixation cross ( e . g . yellow ) within subjects and to prevent them from sensing that they always attend to one side ( left or right ) , the All congruent condition was always presented in the middle ( second or third ) among the four conditions using color of fixation cross indicating opposite side ( e . g . blue ) . As expected there was no significant difference in comprehension accuracy between groups ( two sample t-test , df: 42 , p>0 . 05 ) and for the questions addressed here data was pooled across both groups . In order to prevent talk-specific effects , we used two sets of stimuli consisting of different talks in the combination of audiovisual talks and these two sets were randomized across subjects ( each set 1 and 2 was used for a half of subjects ( 22 subjects ) ) . For example , talks used for AV incongruent condition in the set 1 were used for All incongruent condition in the set 2 . To assess the level of comprehension , we designed questionnaires for each talk . Each questionnaire consists of 10 questions about the talk and tests general comprehension of the talk ( e . g . , “What is the speaker’s job ? ” ) . These questionnaires were also validated from another set of participants ( 16 subjects; 13 females; aged 18–23 years; mean age: 19 . 88 ± 1 . 71 years ) to ensure the same level of difficulty ( accuracy ) across questionnaires for the talks and the length ( word count ) of the questionnaires also matched across all the questionnaires . The attended 4 talks ( in the 4 conditions ) were counterbalanced across conditions in the two sets , thus comprehension of 4 talks were validated . Three participants who showed poor performance ( below 60% accuracy ) were excluded from the analysis . There were no significant differences in the comprehension accuracy between the talks ( mean ± s . e . m . accuracy ( % ) for talk 1: 84 . 62 ± 2 . 91; talk 2: 87 . 69 ± 2 . 31; talk 3: 90 . 77 ± 2 . 39; talk 4: 85 . 38 ± 2 . 43; p>0 . 05 at all pair-wise t-tests between talks ) . In order to recombine audiovisual talks for the four experimental conditions and to add fixation cross , we used Final Cut Pro X ( Apple Inc . , Cupertino , CA ) . The stimuli were controlled with Psychtoolbox ( Brainard , 1997 ) under MATLAB ( MathWorks , Natick , MA ) . Visual stimuli were delivered with a resolution of 1280 x 720 pixels at 25 fps ( mp4 format ) . Auditory stimuli were delivered at 48 kHz sampling rate via a sound pressure transducer through two 5 meter-long plastic tubes terminating in plastic insert earpieces . Each condition ( continuous audiovisual speech ) lasted 7–9 min . After each condition , comprehension questionnaire was performed about the attended talk . Here we measured both accuracy and response time and participants were asked to respond as accurately and quickly as possible . After the experiment , post-experimental questionnaire was administered to obtain participants’ feedback about the experiment . The analysis of MEG data was performed using the FieldTrip toolbox ( Oostenveld et al . , 2011 ) and in-house MATLAB codes according to guidelines ( Gross et al . , 2013a ) . We used in-house Matlab code to extract lip contour of the speaker for each frame of each movie ( Figure 2A ) . From the lip contour we computed area information ( area within lip contour ) , major axis information ( horizontal axis within lip contour ) and minor axis information ( vertical axis within lip contour ) . For our analysis we used area information of lip contour ( see Figure 2—figure supplement 1A , B , C for details ) although use of vertical axis leads to qualitatively similar results . This signal was resampled at 250 Hz to match the sampling rate of the preprocessed MEG signal . We computed the amplitude envelope of sound speech signals ( Chandrasekaran et al . , 2009 ) . We constructed eight frequency bands in the range 100–10 , 000 Hz to be equidistant on the cochlear map ( Smith et al . , 2002 ) . Then sound speech signals were band-pass filtered in these bands using a fourth-order Butterworth filter ( forward and reverse ) . Hilbert transform was applied to obtain amplitude envelopes for each band . These signals were then averaged across bands and resulted in a wideband amplitude envelope . Finally , these signals were downsampled to 250 Hz for further analysis . MR image of each participant was co-registered to the MEG coordinate system using a semi-automatic procedure . Anatomical landmarks such as nasion , bilateral pre-auricular points were manually identified in the individual’s MRI . Based on these three points , both coordinate systems were initially aligned . Subsequently , numerical optimization was achieved by using the ICP algorithm ( Besl and McKay , 1992 ) . Individual head model was created from structural MRI using segmentation routines in FieldTrip and SPM8 . Leadfield computation was based on a single shell volume conductor model ( Nolte , 2003 ) using a 8-mm grid defined on the MNI ( Montreal Neurological Institute ) template brain . For spatial normalization to the standard template , the template grid was transformed into individual head space by linear spatial transformation . Cross-spectral density matrices were computed using Fast Fourier Transform on 1-s segments of data after applying multitaper . Source localization was performed using DICS ( Gross et al . , 2001 ) and beamformer coefficients were computed sequentially for the frequency range from 1–10 Hz . In this study , we used coherence as a frequency-domain measure of dependency to study how rhythmic components in lip movements in continuous speech entrain neuronal oscillations . First , frequency-specific brain activation time-series were computed by applying the beamformer coefficients to the MEG data filtered in the same frequency band ( fourth order Butterworth filter , forward and reverse , center frequency ± 3 Hz ) . The lip speech signals were filtered in the same frequency band . Then , coherence was computed ( Rosenberg et al . , 1989 ) between the lip speech signal and source-localized brain signal for each voxel and each frequency band across 1-s-long data segments overlapping by 0 . 5 s ( Figure 2—figure supplement 1D ) . This computation resulted in a volumetric map describing lip-entrained brain oscillations for each frequency band in each individual . This computation was performed for all experimental conditions: All congruent , All incongruent , AV congruent , AV incongruent . In addition , surrogate maps were created by computing coherence between brain signals and 30 s-shifted lip speech signals for each of the four experimental conditions , thus destroy existing temporal dependencies between the two signals . This surrogate data serves as control data as computations use the same data after controlled manipulation that destroys the effect of interest ( here the temporal shift removes temporal dependencies in coherence measure ) . Surrogate data therefore provide an estimate of coherence that can be expected by chance for each condition . In audiovisual speech , auditory and visual inputs are coherent ( Chandrasekaran et al . , 2009 ) ( also shown in Figure 2C ) . The main purpose of this study was to investigate whether visual signals ( lip movements ) entrain/modulate brain activity and where this entrainment occurs . In order to rule out functional coupling ( coherence ) explained by auditory signals , here we additionally computed partial coherence ( Rosenberg et al . , 1998 ) , i . e . , the coherence partialling out sound speech signals . The analysis process was identical to the process for the coherence above . This partial coherence provides entrained brain activity explained by lip movements that cannot be accounted for by auditory speech signal . As explained above , in natural audiovisual speech , auditory and visual information are robustly correlated ( Chandrasekaran et al . , 2009 ) . Here we computed the coherence between lip speech ( visual ) and sound speech ( auditory ) signals in all experimental conditions except All incongruent ( All incongruent does not have matching lip-sound signals ) . Further , we computed coherence between non-matching lip and sound signals in all conditions except All congruent ( All congruent does not have non-matching lip-sound signals ) . Statistical analysis was performed on the data of all 44 participants using non-parametric randomization statistics in FieldTrip ( Monte Carlo randomization ) . Specifically , individual volumetric maps were smoothed with a 10-mm Gaussian kernel and subjected to dependent samples t-test . We compared each condition with corresponding surrogate data and other experimental conditions . The null distribution was estimated using 500 randomizations and multiple comparison correction was performed using FDR ( False Discovery Rate ) ( Genovese et al . , 2002 ) . Only significant results ( p<0 . 05 , FDR corrected ) are reported . To study the relationship between lip movement entrainment and behavioral performance , we performed regression analysis across subjects using comprehension accuracy from each individual as regressors on the partial coherence map . In this regression analysis , we detected brain regions positively predicted by comprehension accuracy . To maximize sensitivity of this analysis , we compared AV congruent condition to the condition that showed strongest difference in behavioral performance – AV incongruent ( Figure 1B ) . We performed the regression analysis for each condition , then t-values at each brain voxel from the regression analysis were transformed to standard Z-values to be compared between conditions . Then the Z-values were subtracted between the two conditions ( p<0 . 005 ) .	People are able communicate effectively with each other even in very noisy places where it is difficult to actually hear what others are saying . In a face-to-face conversation , people detect and respond to many physical cues – including body posture , facial expressions , head and eye movements and gestures – alongside the sound cues . Lip movements are particularly important and contain enough information to allow trained observers to understand speech even if they cannot hear the speech itself . It is known that brain waves in listeners are synchronized with the rhythms in a speech , especially the syllables . This is thought to establish a channel for communication – similar to tuning a radio to a certain frequency to listen to a certain radio station . Park et al . studied if listeners’ brain waves also align to the speaker’s lip movements during continuous speech and if this is important for understanding the speech . The experiments reveal that a part of the brain that processes visual information – called the visual cortex – produces brain waves that are synchronized to the rhythm of syllables in continuous speech . This synchronization was more precise in a complex situation where lip movements would be more important to understand speech . Park et al . also found that the area of the observer’s brain that controls the lips ( the motor cortex ) also produced brain waves that were synchronized to lip movements . Volunteers whose motor cortex was more synchronized to the lip movements understood speech better . This supports the idea that brain areas that are used for producing speech are also important for understanding speech . Future challenges include understanding how synchronization of brain waves with the rhythms of speech helps us to understand speech , and how the brain waves produced by the visual and motor areas interact .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "neuroscience" ]	2016	Lip movements entrain the observers’ low-frequency brain oscillations to facilitate speech intelligibility
Within land vertebrate species , snakes display extreme variations in their body plan , characterized by the absence of limbs and an elongated morphology . Such a particular interpretation of the basic vertebrate body architecture has often been associated with changes in the function or regulation of Hox genes . Here , we use an interspecies comparative approach to investigate different regulatory aspects at the snake HoxD locus . We report that , unlike in other vertebrates , snake mesoderm-specific enhancers are mostly located within the HoxD cluster itself rather than outside . In addition , despite both the absence of limbs and an altered Hoxd gene regulation in external genitalia , the limb-associated bimodal HoxD chromatin structure is maintained at the snake locus . Finally , we show that snake and mouse orthologous enhancer sequences can display distinct expression specificities . These results show that vertebrate morphological evolution likely involved extensive reorganisation at Hox loci , yet within a generally conserved regulatory framework . Even though vertebrate species can display different morphologies , they all contain a strikingly similar repertoire of transcription factors and signalling molecules . In particular , genes with critical functions during embryonic development are often largely pleiotropic and highly conserved across species ( for references , see Kirschner et al . , 2005; Duboule and Wilkins , 1998 ) . This universality of genetic and genomic principles has changed the evolutionary paradigm from the question of the nature of similarities to that of how distinct traits could evolve using such related developmental pathways ( Carroll , 2008; De Robertis , 2008 ) . Initially , Hox genes , as well as their structural and functional organization into genomic clusters were found well conserved across bilateria ( Duboule and Dolle , 1989; Akam , 1989; Garcia-Fernandez and Holland , 1994; Graham et al . , 1989; McGinnis et al . , 1984 ) . In addition , their mis-expression led to changes in the identity of both insect and vertebrate segments , illustrating their crucial role in the patterning of animal structures , even though the structures they specify are of very different nature in various taxa ( e . g . ( Maeda , 2006; Lewis , 1978; Krumlauf , 1994 ) . Tetrapods generally have four clusters of Hox genes ( HoxA , HoxB , HoxC and HoxD ) , originating from genome duplications early in the vertebrate lineage ( see e . g . Lemons and McGinnis , 2006 ) and located on different chromosomes , unlike fishes or some jawless vertebrates , which have more ( Prince et al . , 1998; Mehta et al . , 2013; Amores et al . , 1998 ) . In addition , all vertebrate Hox clusters described thus far implement a particular type of regulatory process referred to as collinearity , whereby Hox genes are expressed sequentially in both time and space following their topological organization within each genomic cluster ( Gaunt et al . , 1988; Izpisua-Belmonte et al . , 1991 ) . This regulatory property is first observed during axial extension and , subsequently , in some structures such as the limbs ( see ( Deschamps , 2007; Deschamps and van Nes , 2005 ) . In this latter case , and while the detailed underlying mechanism may be distinct from that at work in the major body axis ( Kmita and Duboule , 2003 ) , the general principle remains the same and was likely co-opted in the course of tetrapod evolution ( Spitz et al . , 2001 ) , through the emergence of global enhancers located at remote positions on both sides of the cluster ( Lonfat et al . , 2014 ) . These complex regulations were extensively studied in the mouse , in particular at the HoxD locus , by using various targeted approaches in vivo . The HoxD cluster is surrounded by two gene deserts of approximately 1 Mb ( megabase ) in size , each one containing distinct sets of enhancers capable of activating specific sub-groups of target Hoxd genes depending on their location within the cluster . Each of these two gene deserts can be superimposed to a Topologically Associating Domain ( TAD ) , i . e . a chromatin domain where DNA-DNA interactions in cis are privileged , for example between promoters and enhancers , and determined through chromosome conformation capture technologies ( Dixon et al . , 2012; Nora et al . , 2012 ) . The centromeric gene desert can activate the transcription of the Hoxd9 to Hoxd13 genes , whereas the telomeric gene desert , which is further subdivided into two sub-TADs ( Andrey et al . , 2013 ) controls the expression of Hoxd1 to Hoxd11 ( see [Lonfat and Duboule , 2015] ) . This bimodal regulation allows for the selected expression of Hoxd gene sub-groups in a series of secondary embryonic structures . During limb development , for instance , the telomeric TAD initially controls all genes from Hoxd3 to Hoxd11 in the proximal part of the limb bud , whereas more posterior genes such as Hoxd13 or Hoxd12 are controlled subsequently in the most distal aspect of the incipient limb by enhancers located within the centromeric TAD ( Andrey et al . , 2013 ) . This latter regulatory landscape also controls transcription of the same posterior genes during the outgrowth of external genitalia ( Lonfat et al . , 2014 ) . Since snakes are limbless animals and they display highly specialized and divergent external genitals ( Tschopp et al . , 2014 ) , the existence of such a bimodal type of regulation at the snake HoxD locus was uncertain . Therefore , we set out to investigate Hox gene regulation in snakes . While these animals cannot yet be considered as genuine model systems ( Guerreiro and Duboule , 2014; Milinkovitch and Tzika , 2007 ) , recent advances in their genomic analyses make their study increasingly interesting in an Evo-Devo context ( Castoe et al . , 2013; Gilbert et al . , 2014; Ullate-Agote et al . , 2014; Vonk et al . , 2013 ) . These analyses revealed that snakes , regardless of their extreme morphologies , have a tetrapod-like complement of Hox genes with only a few exceptions ( Vonk et al . , 2013; Di-Poï et al . , 2010 ) . Consequently , the serpentiform body plan may have evolved either along with changes in time and space of Hox gene expression or with a different interpretation of Hox protein functions ( see [Di-Poï et al . , 2010; Woltering et al . , 2009] ) . The analysis of Hox gene expression in the developing corn snake ( Pantherophis guttatus ) revealed a surprisingly well conserved collinear mRNA distribution along the anterior-posterior axis . However , the rather strict correlation between morphological landmarks and the anterior borders of Hox transcript domains , usually seen in mammals and birds , was not always present in snakes ( Burke et al . , 1995; Woltering et al . , 2009 ) ( see also Head and Polly [2015] ) . It was thus concluded that some Hox proteins had likely changed ( part of ) their functionality . In addition , the fact that the most posterior Hox genes were poorly expressed in the extending tailbud was tentatively associated to the unusually large number of segments ( Di-Poï et al . , 2010 ) , together with an increased pace in segmentation ( Gomez et al . , 2008 ) . In this work , we used a combination of experimental approaches to try and elucidate the nature of the differences in Hoxd gene regulation between snakes and mice at comparable stages of their early development . We find that , even though the structural organization of the corn snake HoxD cluster resembles that of tetrapods , the extreme body plan observed in snakes is associated with an extensive regulatory restructuring . In snakes , mesodermal enhancers are mostly located inside the cluster itself , whereas other vertebrates make use of long-range regulations located at remote positions . In addition , we show that despite the loss of limbs , the bimodal chromatin organisation at the Hoxd locus found in tetrapods is conserved in the snake lineage . However , we find that the regulation of snake Hoxd genes during the development of the external genitalia is different from that of other tetrapods , even though the general logic is conserved . In this latter case , the change in enhancer activity from a limb to an external genital specificity seems to have occurred . Altogether , we conclude that Hoxd gene regulation in the snake is in many ways distinct from the situation in mammals . We discuss the possible causative nature of these changes in the evolutionary transformation towards a serpentiform body plan . Because the silencing of transposable elements is often paralleled by the modification of histone H3 at lysine 9 ( H3K9me3 ( Kidwell and Lisch , 1997; Martens et al . , 2005; Friedli and Trono , 2015 ) , heterochromatin-like islands within the snake Hox clusters may be associated with severe modifications in gene regulation ( Di-Poï et al . , 2009; Woltering et al . , 2009 ) . H3K9me3 modifications are normally not found at Hox loci in tetrapods , which like many other genomic loci containing genes of importance for development , are also poor in transposons ( Simons et al . , 2007 ) . Therefore , we performed a ChIP-seq experiment with an antibody against this histone modification on micro-dissected embryonic snake brain ( Figure 1 ) , a tissue that we routinely use as a negative control for Hox gene expression ( Figure 1B ) . No particular H3K9me3 enrichment was scored over the length of the HoxD cluster and the closest located peak was identified in an intron of the Lunapark gene , i . e . at a position unlikely to have any critical impact on Hoxd gene expression ( Figure 1C ) . In tetrapods , the proper collinear regulation of Hox gene transcription was associated with the progressive removal of H3K27me3 coverage ( Soshnikova and Duboule , 2009 ) , a histone modification deposited by the Polycomb complex PRC2 ( Margueron and Reinberg , 2011 ) . We checked if this repressive system would operate similarly during the elongation of the snake body axis by performing an H3K27me3 ChIP-seq , either in the embryonic brain , or in a part of the posterior trunk excluding the post-cloacal region ( Figure 1C ) . In the absence of Hox gene transcription ( brain ) , the entire cluster was decorated with H3K27me3 marks , forming a dense domain of Polycomb repression as seen previously in other species . In contrast , the posterior trunk tissue displayed an H3K27me3 coverage specifically over the 5’ part of the gene cluster , containing the most ‘posterior’ Hoxd genes ( Figure 1C ) . In parallel , whole-mount in situ hybridization ( WISH ) to assess Hoxd gene expression revealed a clear correlation between the domain of active Hoxd genes and the absence of the H3K27me3 mark ( Figure 1B ) . From these experiments , we concluded that both spatial collinearity and the associated dynamics of chromatin structure accompanying progressive gene activation are implemented in snakes as in any other vertebrate species studied thus far . In tetrapods , regulatory elements controlling Hox gene expression are found at various positions . The mouse HoxD cluster for instance contains regulatory elements , which are mainly involved in driving Hox gene expression along the anterior-posterior body axis during gastrulation , whereas remote enhancers located outside of the cluster itself regulate transcription in other organs or structures ( Spitz et al . , 2001; Lonfat and Duboule , 2015 ) . Therefore , to try and identify snake-specific differences in the modes of regulations , we compared the regulatory potential of the snake HoxD cluster with that of other vertebrates by using a BAC transgenic approach in mice , whereby BACs containing HoxD clusters of either human , mouse , chicken , snake and zebrafish were randomly integrated in the mouse genome ( Figure 2A ) . The expression of Hoxd4 was monitored by in situ hybridization with species-specific probes and , under these experimental conditions , all mammalian transgenic BACs showed transcript patterns restricted to the dorsal part of the main embryonic body axis ( Figure 2B ) , resembling the pattern obtained when a single copy Hoxd4/LacZ transgene was used ( Tschopp et al . , 2012 ) . Interestingly , however , this pattern represented only a subset of the full Hoxd4 expression pattern as seen either by WISH on control embryos ( Figure 2B ) , or on previous reporter Hoxd4/lacZ transgenes likely integrated as tandem repeats ( Zhang et al . , 1997 ) . Indeed , expression was scored mostly in the neural tube , yet not in the ventral mesodermal tissues of the upper trunk , i . e . above the level of hindlimbs . To better determine which mesodermal components had their Hox gene expression affected in the isolated human BAC line , we performed a Hoxd4 WISH in a sectioned embryo . We found that , at least at this stage of development , the human Hoxd4 gene was expressed only in the neural tube ( Figure 2—figure supplement 1A ) . 10 . 7554/eLife . 16087 . 006Figure 2 . Location of Hoxd trunk mesodermal enhancers . ( A ) Schematic representation , at the same scale , of the mouse , human , chicken , corn snake and zebrafish BAC clones used to generate the transgenic mouse lines . Exons are represented by black rectangles . ( B ) Lateral view of whole-mount in situ hybridizations of Hoxd4 using E11 . 5 mouse embryos transgenic either for the mouse , the human , the chicken , the zebrafish or the corn snake BAC . ( C ) Schemes illustrating the various deletion stocks ( top ) and whole-mount in situ hybridization of E12 . 5 mouse embryos with the Hoxd4 probe in corresponding deleted mutant embryos ( bottom ) . LoxP sites are indicated as red triangles , the HoxD cluster is represented by a black rectangle and other genes are shown with grey rectangles . vm indicates expression in the ventral mesoderm and white asterisks represent the absence of expression in this tissue . ( D ) ChIP-seq analysis over the mouse and snake HoxD loci of H3K27acetylation using anterior trunk mesodermal tissue of E11 . 5 mouse embryos and 5 . 5 dpo corn snake embryos ( left ) . Green boxes under each ChIP-seq mapping represent peaks called by the MACS algorithm ( Zhang et al . , 2008 ) . On the right , a graphical representation is shown of the percentage of conserved regions between the mouse and corn snake HoxD loci , which are enriched for H3K27ac in each species . DOI: http://dx . doi . org/10 . 7554/eLife . 16087 . 00610 . 7554/eLife . 16087 . 007Figure 2—figure supplement 1 . Detailed analysis of the mesodermal enhancer activity in the 3’ gene desert . ( A ) Whole-mount in situ hybridization of Hoxd4 in E12 . 5 bisected embryos in the control , the Del ( Attp-Sb3 ) line and the Human and Snake BAC lines . ( B ) Expression of Hoxd3 and Hoxd8 in E12 . 5 Del ( Attp-Sb3 ) embryos and respective controls . DOI: http://dx . doi . org/10 . 7554/eLife . 16087 . 00710 . 7554/eLife . 16087 . 008Figure 2—figure supplement 2 . Regulatory potential of the mouse 3’-located , telomeric gene desert in trunk mesoderm . A schematic of the two small ( Del ( Attp-Sb2 ) and Del ( 65-Sb3 ) deletions and of the ( Inv ( Attp-CD44 ) inversion , which takes the telomeric desert away from the HoxD cluster . The expression of Hoxd4 in these mutant lines at E12 . 5 is depicted below . The LoxP sites are represented by red triangles . The black boxes indicate the position of the HoxD cluster and grey boxes represent other nearby located genes . DOI: http://dx . doi . org/10 . 7554/eLife . 16087 . 00810 . 7554/eLife . 16087 . 009Figure 2—figure supplement 3 . Regulatory potential of a mesodermal enhancer sequence ( MSS ) located in the telomeric gene desert . ( A ) ChIP-seq mapping over the HoxD cluster and the flanking 3’-located , telomeric gene desert of H3K27 acetylation in dissected mouse upper trunk tissue . Green boxes under the mapping represent peaks identified by the MACS software . The MSS peak is identified by an asterisk . The HoxD cluster is represented by a black rectangle and the grey box represents the Mtx2 gene ( B ) H3K27ac ChIP-seq mapping and conservation plots of the MSS sequence . The plots show a conservation from 50% to 100% , with a pink colour when conservation is above 75% . ( C ) The MSS enhancer activity is shown as assessed by LacZ reporter assay at E12 . 5 . The pattern of lacZ expression is representative of the one obtained in 4 different embryos . DOI: http://dx . doi . org/10 . 7554/eLife . 16087 . 009 We then investigated the expression of Hoxd4 from either the chicken or the zebrafish BAC transgenic lines and found a similar expression pattern , again mostly limited to the neural tube as well as the dorsal-most part of the somites ( Figure 2B ) . Altogether , these results suggested that enhancers controlling the robust expression of Hoxd genes in various mesodermal derivatives are , for the most part , located outside of the cluster itself . Alternatively , some mesodermal enhancers could be located inside the HoxD cluster , yet they may require additional sequences located at remote positions to properly impact upon the transcription of target genes in physiological conditions . Consequently , we searched for the location ( s ) of such enhancers outside the HoxD cluster by using a set of targeted deletions flanking the locus on either side of it . We first analysed the expression of Hoxd4 in mouse embryos lacking the centromeric TAD , which contains strong enhancers with digit and genital specificities ( Lonfat et al . , 2014 ) . Mutant embryos carrying this HoxDDel ( Atf2-Nsi ) deletion ( Montavon et al . , 2011 ) showed a domain of Hoxd4 expression comparable to control embryos ( Figure 2C; Del ( Atf2-Nsi ) . In contrast , embryos carrying the HoxDDel ( Attp-Sb3 ) deletion of the opposite TAD , located telomeric to the HoxD cluster , which contains various enhancer sequences ( Andrey et al . , 2013; Delpretti et al . , 2012 ) , displayed reduced amounts of mRNA steady state levels in the ventral mesoderm ( Figure 2C; Del ( Attp-Sb3 ) . Consistently , the repositioning of potential telomeric enhancers several megabases far from the target genes , through the HoxDInv ( Attp-CD44 ) inversion , displayed no clear expression in the ventral mesoderm of the thoraco-lumbar region ( Figure 2—figure supplement 2 ) . These results indicate that the telomeric gene desert contains most of the enhancers necessary for Hox gene expression in ventral mesoderm . However , unlike what was observed in the Human BAC line , Hoxd4 expression in the HoxDDel ( Attp-Sb3 ) and HoxDInv ( Attp-CD44 ) mutant lines was also scored in the dorsal-most part of the somites and not exclusively in the neural tube ( Figure 2—figure supplement 1A and 2 ) . To more precisely localize potential mesodermal enhancers within the deleted DNA interval , we used four additional mutant stocks carrying smaller deletions . Both the HoxDDel ( Sb2-Sb3 ) and the HoxDDel ( Sb2-65 ) mutant alleles ( Andrey et al . , 2013 ) resulted in expression patterns for Hoxd4 similar to that obtained with the Del ( Attp-Sb3 ) deletion of the entire gene desert ( Figure 2C; Del ( Sb2-Sb3 ) , Del ( Sb2-65 ) , i . e . lacking any detectable expression in ventral mesoderm . In contrast , such mesodermal expression was scored in the smaller HoxDDel ( 65-Sb3 ) and HoxDDel ( Attp-Sb2 ) deletion alleles ( Figure 2—figure supplement 2; Del ( 65-Sb3 ) , Del ( Attp-Sb2 ) . This set of analyses indicated the presence of mesodermal enhancer ( s ) within a segment of the telomeric gene desert . In addition , the distribution of H3K27ac modifications in the mouse trunk mesodermal tissue , a histone mark associated with putative active enhancers and promoters , was clearly enriched in the telomeric gene desert when compared to the centromeric counterpart ( Figure 2D , top ) with 18 significant peaks telomeric to the cluster versus only 7 located in the centromeric gene desert . We thus concluded that most trunk mesodermal enhancers acting over Hoxd4 and presumably affecting other Hoxd genes , are located in the telomeric gene desert . Because the regulatory sequences located in the telomeric TAD were described to globally drive concomitant expression of several genes located in the central part of the gene cluster rather than individual Hoxd genes ( Delpretti et al . , 2013; Andrey et al . , 2013 ) , we also analysed the expression of both Hoxd3 and Hoxd8 in the absence of the telomeric gene desert . Similar to Hoxd4 , the expression of these two other Hoxd genes was lost in the ventral mesoderm ( Figure 2—figure supplement 1B ) thus suggesting that the telomeric gene desert contains sequences necessary for the expression of multiple Hoxd genes in the ventral mesoderm of the upper trunk . Next , we analysed the transgenic line carrying the snake HoxD cluster and found that , in this case , expression of Hoxd4 in the trunk was not dorsally restricted as observed in all other vertebrate BACs assayed thus far . The expression pattern in the main body axis was in fact reminiscent of the endogenous mouse Hoxd4 expression , with equally strong signals in both the neural tube and mesodermal derivatives ( Figure 2B ) . Therefore , in contrast to other vertebrate species , enhancers located within the snake cluster appear sufficient to drive Hoxd gene expression in the ventral mesoderm . In order to assess if this increase in regulatory potential within the cluster was correlated with a reduction of long-range regulatory elements in the surrounding gene deserts we performed a comparative analysis of H3K27ac profiles between snake and mouse trunk tissues dissected from similar body parts . A global assessment of the profiles suggested that there were relatively less enriched sequences outside of the snake HoxD cluster than outside its mouse counterpart ( Figure 2D , bottom ) . In order to be able to directly compare the ChIP-seq datasets in mouse and snake , we identified 27 DNA regions conserved between the two species and located within the telomeric desert and scored their enrichments with acetylation of H3K27 . While 40% of these conserved sequences were acetylated in the mouse sample , only 22% of them were significantly decorated by this chromatin mark in the snake tissue ( Figure 2D , right ) . Overall , these results indicate that the enhancers required to control snake Hoxd gene expression in the trunk mesoderm are , at least for the most part positioned within the cluster rather than in the telomeric gene desert . These DNA segments acetylated in snakes were found clustered in two regions of the gene desert as demonstrated by peak calling , whereas the mouse acetylated DNA regions span a larger portion of the gene desert ( Figure 2D ) . Of note , one of the acetylated peaks in the mouse was scored over a region conserved in mammals , birds and amphibians , but not in snakes ( Figure 2—figure supplement 3A and B ) . To confirm the enhancer activity of this sequence ( MSS ) , we cloned the mouse version upstream of a LacZ reporter gene . As expected , MSS was able to drive expression in the trunk mesoderm from the forelimb to more posterior parts of the embryo ( Figure 2—figure supplement 3C ) . At the mouse HoxD locus , a bimodal regulatory strategy associated with particular chromatin conformations was reported to control Hox gene expression in a variety of organs and structures . Such global controls involve separate sets of target Hoxd genes , which are thus re-activated after the major body axis is laid down ( Andrey et al . , 2013; Spitz et al . , 2001 ) . Most of these structures , however , are either missing in snakes , such as the limbs or the intestinal cecum or whenever present , they are nevertheless substantially different from their mammalian counterparts . Because mouse Hoxd genes contact such remote enhancer sequences via long-range interactions included within two opposite TADs ( Montavon et al . , 2011 ) , we set out to see whether such a bimodal type of regulatory topology would also exist in snakes , even in the absence of many of the related functionalities . We thus used whole mouse and snake embryos of similar size to characterize the interaction profile of Hoxd genes with their surrounding regulatory landscapes . We used the 4C-seq version of chromosome conformation capture ( Dekker et al . , 2002; de Laat and Dekker , 2012 ) with four different Hoxd genes as viewpoints to assess their potential interaction tropism with either one of the flanking gene deserts . As observed in the mouse , significant interactions between the snake viewpoints and the centromeric TAD were observed , mostly when the Hoxd13 bait was used and , to a lower extent , with Hoxd11 ( Figure 3A , B , compare tracks 1 and 2 ) . In both cases however , substantial contacts were also observed with the opposite , telomeric TAD . These latter interactions increased when the snake Hoxd9 and Hoxd4 baits were used , whereas at the same time , interactions with the centromeric landscape almost disappeared ( Figure 3A , B , tracks 3 and 4 ) . Therefore , as previously reported in the case of mouse tissues and ES cells , genes located at different relative positions within the HoxD gene cluster show distinct interaction tropisms . 5’-located genes such as Hoxd13 interact primarily with the 5’-located gene desert , whereas genes located in a more 3’ part of the cluster , such as Hoxd4 , interact mostly with the 3’-located gene desert ( Figure 3C ) . 10 . 7554/eLife . 16087 . 010Figure 3 . 4C-seq bimodal interaction profiles for Hoxd genes in mouse and snake embryos . ( A ) The four tracks show the interaction profiles established by either the Hoxd13 , Hoxd11 , Hoxd9 or the Hoxd4 viewpoints in E11 . 5 total mouse embryo . While Hoxd13 mostly interacts with the centromeric landscape ( left ) , Hoxd4 contacts preferentially the telomeric landscape . Both Hoxd11 and Hoxd9 show intermediate profiles . The centromeric ( C-DOM ) and telomeric ( T-DOM ) TADs are represented as black boxes on top of the profiles . ( B ) The four tracks show the snake orthologous series of genes used as baits on 2 . 5 dpo corn snake whole embryos . The same general interaction profiles are observed . Brackets indicate the location of the two telomeric sub-TADs:’ a’ and ‘b’ . Under the profiles the HoxD locus is represented . The black rectangle is the HoxD cluster , grey boxes are neighbouring genes and red boxes represent known constitutive contacts in the mouse that are conserved in snakes . ( C ) Graphical representation of the percentage of interactions either in 5’ or in 3’ of the gene clusters , calculated for the different viewpoints for the mouse ( blue ) or the snake ( green ) . DOI: http://dx . doi . org/10 . 7554/eLife . 16087 . 01010 . 7554/eLife . 16087 . 011Figure 3—figure supplement 1 . 4C-seq in the telomeric gene desert of mouse and snake tissue . ( A ) Profiles are from either Hoxd11 or Hoxd9 viewpoints in proximal limb and whole embryo E11 . 5 mouse samples . The track obtained from mouse whole embryo is a zoom in of the profile in Figure 3A , whereas the proximal limb profile is from ( Andrey et al . , 2013 ) . ( B ) 4C-seq profiles using Hoxd11 and Hoxd9 as viewpoints in snake 2 . 5 dpo full embryos represent a zoom in of the track shown in Figure 3B . The brackets represent the two telomeric sub-TADs’ a’ and ‘b’ . The HoxD locus is shown below with the cluster as a black rectangle , grey boxes for neighbouring genes and red boxes for CNS39 , a constitutive contact observed in the mouse and conserved in snakes . ( C and D ) Graphical representation of the percentage of interactions either in the ‘a’ or in the ‘b’ subTADs over the total interactions that map on the telomeric desert , calculated for the different viewpoints for either the mouse ( C ) , or the snake ( D ) . Arrowheads indicate an increase ( mouse proximal limb ) or decrease ( snake tissue ) of interactions in region ‘b’ , as compared to the interactions scored using the same viewpoint in the mouse whole embryo sample . DOI: http://dx . doi . org/10 . 7554/eLife . 16087 . 011 Interestingly , while the general tendency in the bimodal distribution of interactions was thus comparable between mouse and snake full embryos , the pattern of contacts presented important differences between the two species . In the mouse developing proximal limb for instance , Hoxd11 and Hoxd9 preferentially interact with the telomeric sub-TAD referred to as region ‘b’ rather than the sub-TAD ‘a’ ( Figure 3—figure supplement 1A and C ) ( Andrey et al . , 2013 ) . Mouse region ‘b’ thus likely contains important proximal limb regulatory sequences . In whole embryo tissue , we found that the contacts were rather equally distributed between the two regions ‘a’ and ‘b’ ( Figure 3—figure supplement 1A and C ) . In contrast , Hoxd4 preferentially interacted with the sub-TAD ’a’ ( Figure 3A; brackets ) , similar to Hoxd1 in proximal limbs ( Andrey et al . , 2013 ) . In snake embryonic cells , however , Hoxd9 and to a much lesser extent Hoxd11 , displayed an interaction pattern related to that of Hoxd4 with contacts enriched within the sub-TAD ’a’ , suggesting the absence of strong regulatory controls located in region ‘b’ of the snake gene desert ( Figure 3—figure supplement 1B and D ) . In the mouse , strong contacts between Hoxd13 and its flanking regulatory landscape were associated with its function during the development of both digits and external genitals ( Lonfat et al . , 2014; Montavon et al . , 2011 ) . As snakes lack digits , we investigated whether the conservation of this particular chromatin domain was related to the existence of external genital organs . Male snakes display hemipenes ( HP ) , resulting from symmetrical genital buds during development . As it was proposed that the genitals of mammals have a different embryonic origin than those of other amniotes such as squamates ( Tschopp et al . , 2014 ) , the existence of the same global regulation in snakes was unclear . In situ hybridization revealed that 3’-located genes such as Hoxd4 are expressed in the snake HP ( Figure 4A , top ) , in contrast to the mouse where neither Hoxd3 nor Hoxd4 are transcribed in this structure ( Lonfat et al . , 2014 ) . This difference was also scored when Hox gene expression was analysed in the various BAC transgenic lines . While the human , mouse , chicken and zebrafish Hoxd genes were expressed mostly along the main body axis and transcribed neither in the limbs , nor in the external genitals , in agreement with previous results ( see above and ( Lonfat and Duboule , 2015 ) ( Figure 4A , bottom and Figure 4—figure supplement 1 ) , the snake BAC expressed the Hoxd11 to Hoxd4 genes in the developing limbs and genital bud ( Figure 4A and Figure 4—figure supplement 1 ) . This likely reflects a lack of repression of these genes into such structures , rather than the presence of limb and genital enhancers located inside the cluster . We investigated this issue by RT-qPCR in the incipient genitals of both mouse and snake , using as a control a region of the trunk located at the exact same anterior-posterior level . In such conditions , while the mouse Hoxd9 to Hoxd3 genes were expressed at much higher levels in the trunk when compared to the genitals , the steady state levels of snake mRNAs were nearly the same in both tissues ( Figure 4B ) . Therefore , these results suggested that the snake HoxD cluster lacks the sequences necessary to prevent Hox gene expression from the trunk lateral mesoderm , a critical factor to properly develop limbs and genitals in mouse ( see discussion ) . 10 . 7554/eLife . 16087 . 012Figure 4 . Regulation of mouse and corn snake Hoxd genes in developing genitals . ( A ) Endogenous Hoxd4 expression both in a E12 . 5 control mouse embryo and in a 8 . 5 dpo corn snake embryo . Higher magnifications of the cloacal regions are shown on the right , with the positions of the GT and HP delineated in white . Below are in situ hybridization of either control or E11 . 5 embryos transgenic for the human and snake BAC clones using species-specific probes . ( B ) Quantifications of Hoxd13 , Hoxd11 , Hoxd10 , Hoxd9 , Hoxd4 , Hoxd3 and Hoxd1 transcript levels either in mouse E12 . 5 GT ( n=2 ) or in snake 4 . 5 dpo HP ( n=2 ) by RT-qPCR . The log2 ratios were calculated between genital and control trunk tissue expression values . Hoxd13 ( P = 0 . 0378 ) , Hoxd9 ( P = 0 . 0375 ) , Hoxd4 ( P = 0 . 0298 ) and Hoxd3 ( P = 0 . 0342 ) log2 ratios are significantly different between mouse and corn snake while Hoxd11 ( P = 0 . 8303 ) and Hoxd10 ( P = 0 . 8539 ) values are not ( *P < 0 . 05; unpaired two-tailed t-test ) . Bars indicate the average . ( C ) Smoothed 4C-seq mapping using mouse and snake Hoxd13 and Hoxd11 as viewpoints and GT ( mouse ) and HP ( snake ) as samples along with a control sample ( left ) . The BamCompare subtract function was used for each viewpoint to compare sequence coverage in GT/HP versus control tissues . Genes are represented by grey rectangles and previously characterized mouse limb or GT enhancers are represented by red boxes below . The vertical shaded zones in pink represent sequences that displayed increased read coverage in GT versus control tissue , whereas the grey zones point to sequences showing increased contact in mouse but not in snake genitals . The percentages show the relative amount of interactions over this particular landscape , calculated as in Figure 3 . The centromeric TAD C-DOM is represented by a black rectangle above the mouse 4C profile . An asterisk highlights strong contacts of Hoxd13 with Island I in the snake . DOI: http://dx . doi . org/10 . 7554/eLife . 16087 . 01210 . 7554/eLife . 16087 . 013Figure 4—figure supplement 1 . Interspecies comparison of the regulatory potential associated with the HoxD cluster . Hoxd11 , Hoxd10 and Hoxd9 species-specific probes were used for whole mount in situ hybridization of E11 . 5 transgenic mice containing either a human , a mouse , a snake or a zebrafish HoxD cluster integrated randomly in the genome . The snake HoxD cluster seems to trigger a much wider transcriptional response in mesoderm derivatives . DOI: http://dx . doi . org/10 . 7554/eLife . 16087 . 013 To further explore this difference in Hox gene regulation between murine and snake external genitalia , we assessed whether Hoxd gene regulation in the developing genitals also relied upon enhancer sequences located in the regulatory landscape upstream of Hoxd13 . We performed 4C-seq analyses using embryonic mouse GT and snake HP tissues at comparable stages , as well as control trunk tissues . As expected , both the mouse Hoxd11 and Hoxd13 genes showed more interactions with the centromeric gene desert in the GT material than in control trunk material ( Figure 4C ) . In contrast , the snake Hoxd13 and Hoxd11 interaction profiles were not significantly different , when either HP or the control samples were used . We searched the snake gene desert for the presence of the GT1 and GT2 sequences , two DNA segments described in the mouse counterpart to specifically interact with Hoxd13 in the developing GT ( Lonfat et al . , 2014 ) ( Figure 4C , bottom left ) and could identify them ( Figure 4C , bottom right ) . However , even though these sequences are well conserved in the snake , they did not significantly increase their interactions with Hoxd13 during the development of the snake genitals ( Figure 4C , right ) . In fact , the comparative analysis of the snake 4C-seq data revealed that , if anything , only the Prox sequence , a known mouse GT and limb enhancer ( Gonzalez et al . , 2007 ) appeared to gain interactions in the snake HP sample , when compared to control tissue . Therefore , in addition to the fact that snakes are limbless and that their external genitalia may derive from a different embryonic origin ( Tschopp et al . , 2014 ) , our results pointed to distinct modalities in the implementation of global gene regulation at the HoxD locus . Consequently , we searched for the presence within the snake centromeric TAD of the digit- and genital bud-specific enhancers previously identified in mammals ( Lonfat et al . , 2014; Montavon et al . , 2011 ) . Surprisingly , all these murine enhancers conserved within mammals and up to birds showed some level of conservation in snakes ( Figure 5A ) . To try and assess the functional potential of these sequences , we isolated the snake sequences Prox , GT2 and Island I ( Figure 5A , B ) and used them separately in a mouse transgenic enhancer assay . The mouse counterpart of the Prox sequence displayed activity in both the developing digits and GT . The mouse GT2 sequence is a genital only-specific enhancer and the mouse Island I sequence displays limb-only enhancer specificity when placed upstream of a lacZ reporter in a transgenic assay ( Montavon et al . , 2011; Gonzalez et al . , 2007; Lonfat et al . , 2014 ) ( Figure 5C , top ) . 10 . 7554/eLife . 16087 . 014Figure 5 . Enhancer activity of mouse limb and GT enhancers . ( A ) Conservation plot over the 5’ gene desert ( centromeric in the mouse ) using mouse as reference sequence . Peaks represent a conservation higher than 50% . The alignment was made with the mVista program using sequences from mouse , chicken , corn snake and zebrafish . Genes are represented by grey boxes . The various mouse limb and/or GT enhancers conserved from mammals to chicken are represented by red boxes , whereas mouse limb enhancers either poorly or not conserved at all in chicken are in blue . ( B ) Conservation plots of selected mouse limb and GT enhancers using the mouse sequence as a reference . Coloured peaks represent a conservation of above 75% . ( C ) Enhancer activities of the mouse , chicken , lizard and snake Island I ( E12 . 5 ) ( top ) , and mouse and snake enhancer activities of the GT2 ( E14 . 5 ) and Prox ( E12 . 5 ) sequences ( bottom ) , in transgenic mouse foetuses . Magnifications of the genital region are included . DOI: http://dx . doi . org/10 . 7554/eLife . 16087 . 014 The snake Prox element elicited a robust lacZ staining in the developing mouse GT . Interestingly however , and in contrast to the mouse sequence , staining was not detected in the growing limb buds , indicating that the snake Prox sequence had lost its potential to drive transcription in digits ( Figure 5C , bottom ) . Likewise , the snake GT2 sequence was able to drive reporter gene expression in the mouse GT . However , the level of lacZ staining obtained was consistently weaker than with the mouse GT2 sequence ( Figure 5C , bottom ) , perhaps related to the weak ( if any ) contact observed by 4C between the snake Hoxd13 gene and this sequence ( Figure 4C ) . Therefore , in the two cases where a mouse sequence displayed an enhancer potential for the developing GT , the cognate snake sequence appeared to share part or all of this potential ( Figure 5C , bottom ) . We then turned to the snake Island I , a mouse limb-specific enhancer sequence and , noteworthy , the snake version was able to drive reporter gene expression in the mouse GT while unable to elicit any staining in the developing limb buds ( Figure 5C , top ) . Therefore , in this case , the same regulatory sequence conserved between the mouse and the snake was interpreted either as a limb- or as a genital-specific enhancer by the mouse , when introduced as transgenes . To further evaluate this striking change in enhancer potential , we cloned and investigated the regulatory capacity of both the chicken and the green anole lizard Island I sequences . The chicken construct elicited staining in limbs but not in the GT , i . e . in a pattern reminiscent of the mouse rather than the snake Island I ( Figure 5C , top ) . Noteworthy , while the lizard Island I also drove reporter gene expression in limbs , weak staining was scored in the GT , somehow displaying an intermediate enhancer specificity between the snakes and the other amniotes assayed ( Figure 5C , top ) . Proper sequential Hox gene activation in time and space during the elongation of the main body axis is critical for the correct patterning of the axial skeleton ( Deschamps and van Nes , 2005 ) . While the underlying regulatory mechanisms remain to be fully understood ( see e . g . ( Gaunt , 2015 ) , they likely involve control sequences located both inside and outside of the Hox gene clusters ( Tschopp et al . , 2009; Tschopp and Duboule , 2011 ) as well as concurrent epigenetic modifications ( Soshnikova and Duboule , 2009 ) . Transgenic approaches have identified several cis-regulatory elements , which could reproduce , for the most part , a Hox-like expression pattern in the main body axis and which mapped close to their target gene ( s ) ( for examples , see ( Bel-Vialar et al . , 2002; Brend et al . , 2003; Charité et al . , 1995; Kwan et al . , 2001; Oosterveen et al . , 2003; Sharpe et al . , 1998 ) . However , our various transgenic mouse lines containing HoxD clusters from different vertebrate species showed that , in most vertebrates , regulatory sequences located in the gene deserts flanking the HoxD cluster are necessary for proper expression in the embryonic trunk mesoderm . Indeed , when using the expression of Hoxd4 as a read-out of the regulatory potential contained either in the mouse , the human , the chicken or the zebrafish clusters , ventral mesodermal expression was not detected . This result is at odds with previous reports describing the presence of mesodermal enhancers on a short mouse transgene derived from this locus ( Zhang et al . , 1997; Morrison et al . , 1997 ) . Similarly , we found Hoxd3 expression to be absent from the ventral mesoderm in mouse embryos that lack the telomeric gene desert while previous work has reported that a small single-copy construct containing the Hoxd3 locus was able to drive reporter gene expression in this tissue ( Tschopp et al . , 2012 ) . These discrepancies may derive from a qualitative aspect whereby a short transgene may reveal the potentialities absent from a more complex BAC environment , such as in the case of Hoxd3 . They may also reflect quantitative differences , for example when comparing ( close to ) single copy BAC integrations with a large number of head to tail integrations of a shorter transgene , as for Hoxd4 . In any case , a different result was obtained when the snake HoxD BAC was used since , unlike the other vertebrate clusters tested , the snake transgene appeared to contain the regulatory elements necessary for expression in the trunk mesoderm . The H3K27 acetylation profiles in mouse and corn snake trunks supported this result for only few acetylation peaks were scored outside of the snake cluster compared to the mouse . In an evolutionary context , it is possible that in snakes , long-range mesodermal enhancers were progressively complemented by local enhancers to regulate Hoxd gene expression in the developing body axis , perhaps to provide an increased fine-tuned control in the expression balance between single Hoxd genes during embryonic development . In the absence of any genetic approach to study snake development , it is difficult to evaluate the functional relevance of this difference in regulation . It is however worth noting that the snake HoxD cluster BAC was the only transgenic configuration where expression was strong in the limbs and external genitals , whereas BACs derived from animals with limbs did not elicit an expression of Hoxd genes into the transgenic limbs , even though the endogenous genes were strongly expressed there . Also , isolated mouse mesodermal enhancers were often described to activate reporter gene expression in secondary structures such as limbs ( Charité et al . , 1995; Kwan et al . , 2001; Sharpe et al . , 1998; Renucci et al . , 1992 ) . In tetrapods , this apparent paradox may reflect the necessity for a highly specific type of regulation to control both Hoxa and Hoxd genes in specific domains of the growing limb buds ( Andrey et al . , 2013; Berlivet et al . , 2013; Woltering et al . , 2014 ) . Implementing such global limb regulations may require previous regulatory inputs to be terminated . Our results using the corn snake BAC suggest that , in the absence of limbs , this negative control may have been lost in the course of evolution , leading to the maintenance of transcription in all mesoderm derivatives . Whether or not this increased ‘mesoderm potential’ present in the snake HoxD cluster may somehow relate to the presence of ventral mesodermal enhancer remains to be clarified . Hox clusters of jawed vertebrates show high levels of chromatin compaction ( Noordermeer et al . , 2011; Fabre et al . , 2015 ) , a feature associated with the unusual level of gene packaging and organization , which occurred at the roots of the vertebrate lineage ( Duboule , 2007 ) together with the emergence and generalisation of long-range regulations at these loci ( Darbellay and Duboule , 2016 ) . Squamate Hox clusters however seem to slightly deviate from this rule by having accumulated a large number of transposable elements ( Di-Poï et al . , 2009 ) , a situation rarely found around genetic loci of developmental importance ( Simons et al . , 2007 ) due to the potential effects of such sequences to elicit genetic and morphological variations ( Kidwell and Lisch , 1997; Friedli and Trono , 2015 ) . While the presence of such repeated elements within and around the snake HoxD cluster may have been associated with differences in the location of enhancer sequences , the distribution of chromatin modifications did not point to any drastic regulatory re-organization of this gene cluster in snakes . Indeed , the H3K9me3 histone mark , which in some cases is associated with TEs such as LTRs , LINEs and SINEs ( Friedli and Trono , 2015; Mikkelsen et al . , 2007 ) was not found within the HoxD cluster itself . In addition , the analysis of other chromatin marks present in trunk tissue during the sequential activation of the gene cluster displayed distributions similar to those found in the mouse cluster . These global similarities between mouse and snake in the structure of the HoxD cluster were also noticed when interaction profiles were considered . There again , the snake embryonic material displayed the bimodal distribution of contacts on both sides of the gene cluster , as expected either from several studies using specific mouse samples ( Lonfat et al . , 2014; Andrey et al . , 2013 ) , or from the full embryonic material used in this work . In all cases , most Hoxd genes tend to naturally interact within the ‘telomeric’ ( 3’-located ) TAD , whereas Hoxd13 was strongly associated with the ‘centromeric’ ( 5’-located ) TAD . However , a significant difference was scored in the interaction profiles of Hoxd9 , which displayed more contacts further away of region CNS39 in the mouse than in the snake sample . As this regulatory region is particularly active in proximal limbs , the lower frequency of contacts observed in snake embryos was not unexpected . This observation indicates that , while the general bimodal TAD regulatory structure is conserved at the HoxD locus , differences between vertebrate species may exist either in the extent , or in the internal organization of interactions within the TADs . This latter point was of particular interest as the mouse centromeric TAD was initially defined due the presence of many limb and genital-specific enhancers ( Montavon et al . , 2011; Lonfat et al . , 2014 ) , which were not necessarily expected to be conserved in the snake orthologous landscape due to both the absence of limbs and the presumably distinct origin of snake external genitalia following the relative shift of the cloaca over the course of evolution ( Tschopp et al . , 2014 ) . However , tetrapod limb-specific enhancers are often conserved in snakes and a significant overlap between limb and genital cis-regulatory mechanisms was recently reported ( Infante et al . , 2015 ) , in agreement with the similarities between the molecular mechanisms employed to generate the two structures ( Kondo et al . , 1997; Cohn , 2011; Yamada et al . , 2006 ) . Indeed when a mouse limb- and genital-specific Tbx4 enhancer sequence was isolated from snake , it could only recapitulate genital expression , thus having lost the limb-specific regulatory potential ( Infante et al . , 2015 ) . Consistent with this observation , we find that the snake counterpart of the mouse limb and genital enhancer Prox ( Gonzalez et al . , 2007 ) has lost its limb regulatory potential , while keeping a strong capacity to drive expression in the developing genitalia . This suggests that the mouse Prox consists of two regulatory modules , while the snake Prox has kept the genital specificity only . More strikingly however , Island I , which in the mouse is a limb-only specific enhancer ( Montavon et al . , 2011 ) , switched its regulatory capacity in the snake to become a genital-only specific enhancer , accompanied by elevated enhancer-promoter interactions as assayed by 4C ( Figure 4C – asterisk ) . On the other hand , the chicken Island I revealed enhancer specificities almost identical to those of the mouse sequence despite the fact that birds are more closely related to squamates than to mammals , suggesting that Island I had a limb-only activity at the time of divergence between mammals and reptiles/birds . Interestingly , the lizard sequence displayed a weak activity in the genital bud , in addition to the limb , indicating that the co-option to a genital function likely occurred in the squamate clade and thus preceded limb loss in snakes . External genitalia of squamates , unlike that of mammals , were proposed to have the same embryonic origin as limbs ( Tschopp et al . , 2014 ) . Although , this could bias our mouse-based transgenic analysis , the snake Prox sequence could drive reporter gene expression specifically in the mouse GT and not in the limb , making it unlikely that differences in embryonic origin could interfere with this particular analysis . While the generation of the same transgenics in snakes would be necessary to fully rule out this bias , such experiments are not feasible for the moment . Altogether our results show that , while the general bimodal regulatory strategy is conserved , some profound differences in the regulation are scored at the HoxD locus between two species displaying strikingly distinct morphologies . It is as yet unclear if such changes were causative of the extensive morphological changes that snakes experienced over the course of evolution , or whether they are merely consequential . It nevertheless indicates that vertebrates with extreme variations in those systems known to be under the control of Hox genes ( vertebral number and identities , limbs , genitals ) rely upon the same general regulatory architecture and principles at Hox loci . In this view , vertebrate morphological evolution was accompanied by changes in Hox gene regulation , yet such variations were constrained within the general regulatory framework found at these loci . This may reflect selective pressures that impose essential basic properties to vertebrate body plans , while other more subtle morphological specificities , less likely to result in adverse effects , may arise in the course of evolution . Maintenance of , and experiments on animals were approved by the Geneva Canton ethical regulation authority ( authorization 1008/3421/1R to M . C . M . and GE/81/14 to D . D . ) and performed according to Swiss law . A Pantherophis guttattus BAC library containing 55’296 clones was constructed from liver tissue of one single individual ( Amplicon Express ) . Degenerate primers were designed in DNA regions conserved between mammallian and bird species within the DNA interval spanning from the Atf2 gene to the CNS65 region ( mouse chr2:73653618–75292344 in mm9 ) . The amplified DNA fragments ranging from 400 bp to 1 kb ( kilobases ) were cloned into a PGEMTeasy vector and labelled with DIG-High Prime ( Roche , Switzerland ) to screen filters provided by the company . The lengths and positions of positive BACs were evaluated by PCR and BAC end sequencing . 13 BAC clones were selected for sequencing at BGI ( Beijing Genomics Institute ) . Sequencing was performed on 500 bp and 2 kb-large insert libraries using the Illumina HiSeq2000 and assembly was done using SOAP denovo . The resulting sequence is deposited in GenBank under accession number KU866087 . Exons of the corn snake Hox genes were identified using GENSCAN ( http://genes . mit . edu/GENSCAN . html ) ( Burge and Karlin , 1997 ) and sequence comparison against Hox coding sequences of closely related species . Conservation of non-coding elements was assessed by the use of the mVista software ( http://genome . lbl . gov/vista/mvista/submit . shtml ) with default parameters . Information about different vertebrate Hox clusters and sizes of gene deserts were taken from the UCSC genome browser ( http://genome . ucsc . edu ) and the ncbi genome database ( http://www . ncbi . nlm . nih . gov/genome/ ) . Transposable elements were identified using RepeatMasker ( http://www . repeatmasker . org/ ) and the Repbase vertebrate repeat library combined with a de novo corn snake repeat library described in Ullate-Agote et al . ( 2014 ) . Snake forebrain , anterior trunk and posterior trunk samples as well as mouse anterior trunk samples were dissected and fixed in 1% formaldehyde for 10 min . For each ChIP-seq experiment approximately 100 ng of tissue were used and processed according to ( Lee et al . , 2006 ) or the ChIP-IT High Sensitivity ( Active motif ) specifications . H3K27me3 antibody ( Millipore , 17–622 ) , H3K27ac antibody ( Abcam ab4729 ) and H3K9me3 ( Abcam ab8898 ) were used . Sequencing was performed using 100 bp single-end reads in the Illumina HiSeq system according to manufacturer’s instructions . The reads obtained from the sequencing were mapped to ENSEMBL Mouse assembly NCBIM37 ( mm9 ) or to the corn snake scaffold using the HTSstation mapping pipeline ( http://htsstation . epfl . ch ) ( David et al . , 2014 ) . All ChIP-seq mappings were normalized to total input chromatin using the bamCompare software from the deepTools Galaxy web server ( http://deeptools . ie-freiburg . mpg . de ) ( Ramirez et al . , 2014 ) . Peak calling was done using MACS ( Zhang et al . , 2008 ) . The HoxDDel ( AttP-SB3 ) ( aka Del ( AttP-SB3 ) , HoxDDel ( AttP-SB2 ) ( aka Del ( AttP-SB2 ) ) and HoxDDel ( SB2-65 ) ( aka Del ( SB2-65 ) mutant alleles were generated through TAMERE ( Hérault et al . , 1998 ) and have been described elsewhere ( Andrey et al . , 2013 ) . The HoxDDel ( Atf2-Nsi ) ( aka Del ( Atf2-Nsi ) allele was also produced by TAMERE and described in ( Montavon et al . , 2011 ) . The HoxDDel ( 1–13 ) d11lacZ ( aka Del ( 1–13 ) d11lacZ allele , previously referred to as Del9 in Zákány et al . ( 2001 ) was obtained by loxP/Cre mediated recombination in ES cells . The HoxDinv ( AttP-CD44 ) ( aka Inv ( AttP-CD44 ) was generated using STRING ( Spitz et al . , 2005 ) . Telomeric desert deletions: The HoxDDel ( SB2-SB3 ) ( aka Del ( SB2-SB3 ) and Del ( 65-SB3 ) alleles were generated by TAMERE . All telomeric deletions analyzed were trans-heterozygotes over the HoxDDel ( 1–13 ) d11lacZ balancer allele . The mouse ( RP23-400H17 ) BAC has been previously described in ( Spitz et al . , 2001 ) . The human ( CTD-2086D13 ) , chicken ( CH261-92D10 ) , snake ( Eg-32P1 custom library ) and zebrafish ( 77g24 ) BACs were all recombined to introduce a PISceI site in the vector using EL250 cells ( Lee et al . , 2001 ) . The snake and chicken BACs were further shortened to remove all sequences flanking the HoxD cluster . The final genomic coordinates of the chicken BAC used for transgenesis were chr7:17361344–17440245 ( galGal3 ) . The snake BAC , which initially extended 74 kb downstream of Hoxd1 was reduced to contain only 300 bp 3’ of Hoxd1 . Successful recombination was confirmed by PCR , restriction digest and BAC end sequencing . Prior to injection , BACs were isolated using the Nucleobond Maxiprep kit , linearized with PISceI , purified by phenol/chloroform extraction and dialyzed against microinjection buffer . The linearized BACs were then injected into fertilized mouse oocytes . After having obtained transgenic lines , BAC integrity was confirmed by PCR using primers specific for Hox genes of the appropriate species . In situ hybridization was performed as previously described ( Woltering et al . , 2009 ) with a hybridization temperature of 68°C and 1 . 3x SSC concentration in the hybridization mix . Post-hybridization washes were performed using 2x SSC concentration for four times 30 min . All probes designed for in situ hybridization of BAC transgenic embryos were tested for cross-reactivity by conducting in situ hybridization on control mouse embryos . To produce human , chicken , snake and zebrafish riboprobes , DNA fragments were amplified from BAC DNA that comprised either the first exon or the 3’UTR of the Hox genes ( see Table 1 ) . After ligation with the PGEMTEasy vector ( Promega ) , probes were synthetized using DIG RNA labeling mix ( Roche ) and purified with the QIAGEN RNeasy mini kit . Hoxd11 , Hoxd9 and Hoxd4 mouse probes were previously described ( Gérard et al . , 1996; Zappavigna et al . , 1991; Featherstone et al . , 1988 ) . 10 . 7554/eLife . 16087 . 015Table 1 . List of primers used to clone the probes for in situ hybridization . DOI: http://dx . doi . org/10 . 7554/eLife . 16087 . 015HsHoxd13GGTCCAGGTTGGCCACAGACGTCACTCTACTGATTGCAGCHsHoxd11TTGAGAGCTCCAGGAAGCGCTTCAGTTGCATGGGTTCTGGHsHoxd9CCAATTCCAAGAATGAAGGCACATTTACAACTGGTCCTCGHsHoxd4CAACTCAGAGGCGAGTTCACTCAAGTAGCTTGCTATGGCADrHoxd13ATGATGGTTTCCAGATATGCTGGTGACAGCTGCCCAATCADrHoxd11GAGCCGCTGTTCTTTTCTTCGTCCTATCCGCACGCATATGDrHoxd10CCACCTTTGCCTTCTCTGTGTCCAAAATGTCCTTTCCCAACDrHoxd9TTACTTGGGTCAAGTTGTTGGTGAAGGCAGCAAAAATACTPgHoxd13GCGCTTCTGATCATGTTTGCATAGCTAAACATATAGGCACPgHoxd11CCTAGAGGTTAATATGACTCCCCCATTTAGGCTCCTAGGPgHoxd10CCGAGAACTGACTGCTAATCCAGAATTTATTGCATTATACPgHoxd9AGGAGAGTAACACTTTGAGGCCTCTCTGACATGAGTCTTGPgHoxd4CGGATTTGACCACTTTATAGAACAATATCACCAACACATG Prospective enhancer sequences were obtained by either PCR or restriction digest and cloned upstream of a βglobin-lacZ construct into either PGEMTEasy , or SK Bluescript ( - ) . Constructs were injected into mouse oocytes and embryos were harvested at E12 . 5 and E14 . 5 . Beta-galactosidase staining was performed by fixing in 4% PFA for 30 min , washing in PBS/0 . 1% Tween and incubating in staining solution ( 1 mg/mL Xgal ) overnight at 37°C . A minimum of three transgenics with consistent staining was obtained per construct . Mouse embryos transgenic for either the Island I/LacZ , the GT2/LacZ or the Prox/LacZ constructs had been obtained in previous studies ( Lonfat et al . , 2014; Montavon et al . , 2011; Gonzalez et al . , 2007 ) . One E11 . 5 whole mouse embryo , one 2 . 5 dpo ( days post oviposition ) whole snake embryo , ca . 30 mouse E13 . 5 and corn snake 8 . 5 dpo genital buds as well as mouse and snake trunk tissue dissected from comparable anterior-posterior levels were processed as previously described ( Noordermeer et al . , 2011 ) . Mouse libraries were constructed by using NlaIII as the first restriction enzyme and DpnII ( New England Biolabs ) as the second restriction enzyme and the baits and inverse primers used for the Hoxd4 , Hoxd9 , Hoxd11 and Hoxd13 viewpoints were described in Noordermeer et al . ( 2011 ) . Snake libraries were constructed using DpnII as primary enzyme and NlaIII as secondary enzyme . For the Hoxd4 , Hoxd9 , Hoxd11 and Hoxd13 baits , the primers are listed in Table 2 . All libraries were sequenced in the Illumina HiSeq system to generate 100 bp read length . The reads obtained were then demultiplexed , mapped and analysed using the HTSstation pipeline ( http://htsstation . epfl . ch ) ( David et al . , 2014 ) . The global quantification of telomeric versus centromeric signals was calculated as in ( Andrey et al . , 2013 ) . Signals mapping 5’ of the cluster signal were quantified from the Atf2 gene to 14 kb upstream of the Evx2 gene and 3’ signals were quantified starting from 5 kb downstream of Hoxd1 to 46 kb downstream of the CNS65 enhancer sequence . Quantifications of contacts within regions ‘a’ and ‘b’ were calculated using the same coordinates as for the 3’ signal calculation and by excluding the region chr2: 74964245–75004987 ( mm9 ) in mouse and a comparable interval in snake so that the peak of interaction over the CNS39 region would not be accounted for . Comparison of signals between genital tissue and trunk control tissue was done by using bamCompare subtract function from the deepTools Galaxy web server ( http://deeptools . ie-freiburg . mpg . de ) ( Ramirez et al . , 2014 ) . 10 . 7554/eLife . 16087 . 016Table 2 . List of primers used for 4C-seq amplifications with snake tissues . DOI: http://dx . doi . org/10 . 7554/eLife . 16087 . 016PgHoxd13 DpnIIAATGATACGGCGACCACCGAACACTCTTTCCCTACACGACGCTCTTCCGATCTGGAAAAGGTTGTTAATCAGGPgHoxd13 NlaIIICAAGCAGAAGACGGCATACGACTGCCCTTCTTCAAAGAGACPgHoxd11 NlaIIICAAGCAGAAGACGGCATACGAGCCGCAGTTGTCCAAGTTACPgHoxd11 DpnIIAATGATACGGCGACCACCGAACACTCTTTCCCTACACGACGCTCTTCCGATCTTCCTCCTTGAGAGGGAATCCPgHoxd9 NlaIIICAAGCAGAAGACGGCATACGAAAGAATCCCCATCCTAGTCCPgHoxd9 DpnIIAATGATACGGCGACCACCGAACACTCTTTCCCTACACGACGCTCTTCCGATCTTGTAATCGTAATCAGCATAGPgHoxd4 DpnIIAATGATACGGCGACCACCGAACACTCTTTCCCTACACGACGCTCTTCCGATCTCACTTCATCCTTCGGTTCTGPgHoxd4NlaIIICAAGCAGAAGACGGCATACGATAAACAATGAAGTGAAACGG RNA from genitals and control trunk tissue was extracted from two E12 . 5 mouse embryos and from two 4 . 5 dpo corn snake embryos using the microRNeasy kit ( QIAGEN ) . Biological replicate number was dependent on restricted availability of material and reduced variability of expression values between samples . cDNA was generated using the Promega GoScript reverse transcriptase according to manufacturer’s instructions . qPCR was performed using SYBR select master mix ( Applied Biosystems ) using two technical replicates per biological sample . Primers used are listed in Table 3 . The Hmbs gene expression was used for normalisation and log2 ratios were calculated between GT or HP expression values and trunk control tissue expression . 10 . 7554/eLife . 16087 . 017Table 3 . List of snake and mouse primers used for qPCR . DOI: http://dx . doi . org/10 . 7554/eLife . 16087 . 017PgHoxd13ACGAGACCTACATCTCCATGTTGGTGTAAGGCACTCGCTTCPgHoxd11TCCGAAAAGCCAGAGTTCAGATCTGGTACTTGGTGTAAGGPgHoxd10CGTCTCCAGCCCAGAAAGCGGTTGGAGTATCAGACTTGGPgHoxd9AGGAAAAAGAGGAGCAGCAGTGGAGCGAGCATGAATCCAGPgHoxd4GAAAGTCCACGTTAACTCTGGACTTGCTGCCTGGTATAAGPgHoxd3AGGTATCCAGCTCGCTTACCGCGGACTCTTGTCTTCACAGPgHoxd1AAAGTCAAGAGGAACGCACCACTGGAAGACCCACAAGCTGPgHmbsATTGGGACCAGCTCACTTCGCCTCCTTCTCGTCCAGCTTCMmHoxd13GAAATCATCCTTTCCAGGAGATGCGCCGCTTGTCCTTGTTAATGMmHoxd11AAGAGCGGCGGCACAGTGTTGAGCATCCGAGAGAGTTGGMmHoxd10AGGAGCCCACTAAAGTCTCCCAGACTTGATTTCCTCTTTGCMmHoxd9GACCCAAACAACCCTGCAGTTCAGAATCCTGGCCACCTCMmHoxd4TGCACGTGAATTCGGTGAACGTGAGCGATTTCAATCCGACGMmHoxd3AAGCAGAAGAACAGCTGTGCTAGCGGTTGAAGTGGAACTCCMmHoxd1GGCCCTTTCAGACTGTGTCCCATATTCGGACAGTTTGCTTTTCMmHmbsCGGCTTCTGCAGACACCAGCCCTCATCTTTGAGCCGTTTTC Raw and processed data of 4C-seq and ChIP-seq analysis are available in the Gene Expression Omnibus ( GEO ) repository under accession number GSE79048 .	Animals with a backbone can look remarkably different from one another , like fish and birds , for example . Nevertheless , these animals – which are also known as vertebrates – have many genes in common that shape their bodies during development . These genes include a family called the Hox genes , which control how an animal’s body parts develop from its head to its tail and are needed to shape the animal’s limbs . Hox genes are found clustered in groups within a vertebrate’s DNA , and large regions of DNA on either side of a Hox cluster can , in some cases , physically interact with the Hox genes to regulate their expression . So how do the same genes produce different body shapes ? Different vertebrates regulate where and when their Hox genes are switched off and on in different ways . As such , it is likely that differences in gene regulation , rather than in the genes themselves , lead embryos to develop into the distinct shapes seen across the animal kingdom . Snakes – for example – evolved from a lizard-like ancestor into elongated limbless animals as they have adapted to a burrowing lifestyle . However , it was not known if changes in how Hox genes are regulated have played a role in shaping the distinct body plan of snakes . Guerreiro et al . have now compared how Hox genes are regulated in snakes , mice and other vertebrates , focusing on corn snakes and one particular cluster of Hox genes called the HoxD cluster . The comparison revealed that these Hox genes are regulated differently in developing snakes than in other vertebrate embryos . This is particularly the case for tissues that show the most differences when compared with other animals ( such as the torso and genitals ) or that are absent ( such as the limbs ) . Although Hoxd genes are activated at different times and places in snakes than in other vertebrates , snake Hox genes appear to be regulated using the same general mechanisms as mouse Hox genes . Guerreiro et al . suggest that changes to Hoxd gene regulation have contributed to the evolution of the snake’s shape and have most likely influenced the body shapes of other vertebrates as well . However , the findings also suggest that these gene regulatory changes have been constrained by an existing regulatory mechanism that has been maintained throughout evolution . It remains for future work to address whether these changes in Hox gene regulation are a cause or a consequence of the snake’s extreme body shape , or indeed a combination of the two .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "developmental", "biology", "evolutionary", "biology" ]	2016	Reorganisation of Hoxd regulatory landscapes during the evolution of a snake-like body plan
Many mammals forage and burrow in dark constrained spaces . Touch through facial whiskers is important during these activities , but the close quarters makes whisker deployment challenging . The diverse shapes of facial whiskers reflect distinct ecological niches . Rodent whiskers are conical , often with a remarkably linear taper . Here we use theoretical and experimental methods to analyze interactions of mouse whiskers with objects . When pushed into objects , conical whiskers suddenly slip at a critical angle . In contrast , cylindrical whiskers do not slip for biologically plausible movements . Conical whiskers sweep across objects and textures in characteristic sequences of brief sticks and slips , which provide information about the tactile world . In contrast , cylindrical whiskers stick and remain stuck , even when sweeping across fine textures . Thus the conical whisker structure is adaptive for sensor mobility in constrained environments and in feature extraction during active haptic exploration of objects and surfaces . Many mammals use facial whiskers for navigation ( Vincent , 1912; Dehnhardt et al . , 2001 ) , object localization ( Hutson and Masterton , 1986; Knutsen et al . , 2006; Krupa et al . , 2001; Mehta et al . , 2007; O’Connor et al . , 2010a; Pammer et al . , 2013 ) , texture discrimination ( Carvell and Simons , 1990; Wolfe et al . , 2008; Chen et al . , 2013 ) , and object recognition ( Anjum et al . , 2006 ) . The shapes of mammalian whiskers are diverse . Rodent whiskers are conical ( Birdwell et al . , 2007; Williams and Kramer , 2010; Quist et al . , 2011; Pammer et al . , 2013 ) , whereas sea lion whiskers ( Hanke et al . , 2010 ) and human hair are approximately cylindrical . Whiskers of harbor seals have elliptical cross-sections with an undulated structure ( Hanke et al . , 2010 ) . Differences in whisker shapes across different species likely reflect differences in how animals use their whiskers . For example , the undulating microstructure of harbor seal whiskers suppresses vibrations triggered by vortices and enhances the seal’s ability to analyze water movements ( Hanke et al . , 2010 ) . What could be the advantages of the whisker taper seen in rodents ? Rodents sense their surroundings by moving their whiskers over objects with large amplitudes ( up to 50° peak–peak ) in a rhythmic motion ( Knutsen et al . , 2006; O’Connor et al . , 2010a; O’Connor et al . , 2013; Voigts et al . , 2008 ) . Rodents can localize and recognize objects in three dimensions ( Knutsen et al . , 2006; Krupa et al . , 2001; O’Connor et al . , 2010a; Pammer et al . , 2013; Voigts et al . , 2008 ) and also discriminate subtle differences in surface textures ( Carvell and Simons , 1990; Wolfe et al . , 2008 ) ( reviewed in Diamond , 2010 ) . These behaviors are based on collisions between whiskers and objects , which cause time-varying forces at the whisker base and excitation of sensory neurons in the follicles ( Zucker and Welker , 1969; Szwed et al . , 2006 ) . Whisker mechanics thus couples the tactile world to forces at the whisker base ( Solomon and Hartmann , 2006; Birdwell et al . , 2007; Bagdasarian et al . , 2013; Pammer et al . , 2013 ) . Rodent whiskers are thin , approximately linear and homogenous elastic cones ( Solomon and Hartmann , 2006; Birdwell et al . , 2007; Williams and Kramer , 2010; Pammer et al . , 2013 ) . As a result of the linear taper , whisker bending stiffness decreases with distance from the face over five orders of magnitude . Behavioral measurements have shown that mice use distance-dependent whisker mechanics as a ruler to estimate object location along the length of the whisker ( Pammer et al . , 2013 ) . Here we used theoretical and experimental methods to analyze the interactions of whiskers with objects . We uncover additional decisive advantages of conical whiskers compared to cylindrical whiskers for tactile exploration . Conical whiskers sweep across textures with informative micromotions , whereas cylindrical whiskers get stuck . The steep increase in flexibility from base to tip of conical whiskers allow rodents to maneuver their sensors past objects with relative ease . Conical whisker shape is thus critical for tactile exploration in confined spaces . We modeled rodent whiskers as truncated cones with a cylindrical cross section , base radius rbase , tip radius rtip , and length LW ( Ibrahim and Wright , 1975; Boubenec et al . , 2012 ) ( Figure 1A ) . Whiskers have intrinsic curvature ( Quist and Hartmann , 2012 ) and are further deflected by forces that are caused by interactions with objects ( O’Connor et al . , 2010a; Bagdasarian et al . , 2013; Pammer et al . , 2013 ) ( Figure 1B ) . In our model , contacts occurred either in the ‘concave backward’ ( CB ) or ‘concave forward’ ( CF ) directions ( Figure 1C ) ( Quist and Hartmann , 2012 ) . We quantified contact strength using the push angle θp ( Quist and Hartmann , 2012 ) , the angle through which the whisker is rotated into the object ( Figure 1D ) . By convention , contacts for the CB configuration correspond to θp > 0 , and for the CF configuration to θp < 0; θp = 0 defines the angle of initial touch . In all cases whisker movement and bending were limited to the x-y plane . We computed whisker shape by solving the Euler–Bernoulli beam equation in the quasi-static regime ( Euler , 1744; Birdwell et al . , 2007; Solomon and Hartmann , 2006; Pammer et al . , 2013 ) . The beam equation describing whisker shape was converted to a boundary-value problem formulation ( ‘Materials and methods’; Press et al . , 1992 ) , a set of differential equations with defined boundary conditions at the whisker base and at the point of contact with the object . The object was assumed to be a cylindrical pole perpendicular to the plane of motion , as is typically used in object localization experiments ( Knutsen et al . , 2006; Mehta et al . , 2007; O’Connor et al . , 2010a; Pammer et al . , 2013 ) ( Figure 1B ) . The whisker shape at each time was determined by the static solution computed for the time-varying boundary conditions . Using identical methods we also modeled hypothetical cylindrical whiskers . 10 . 7554/eLife . 01350 . 003Figure 1 . Schematic of the whisker in two dimensions . ( A ) The whisker is modeled as a truncated cone of length LW , virtually extended to length L . ( B ) The base of the whisker ( in the follicle , or attached to a galvo , Figures 4–6 ) is at point ( x0 , y0 ) and angle θ0 , measured clockwise . The position of a point along the whisker is ( x ( s ) , y ( s ) ) and its angle with the x-axis is θ ( s ) . The contacted object is a cylindrical pole with radius rpole centered at ( xcen , ycen ) ; the pole and whisker are shown at a magnified scale in the inset on the left . The whisker contacts the object at the point ( xobj , yobj ) at an angle θobj . The object distance , d , is the distance between ( xobj , yobj ) and the whisker base . The pole applies a force F→ on the whisker . ( C ) The concave forward ( CF , left ) and concave backward ( CB , right ) whisker configurations . Thick black lines , whiskers; solid circle , poles; gray arrows , movement directions . ( D ) , Definition of the push angle , θp , which measures the strain on the whisker imposed by the object . The deflected and undeflected whiskers are shown as black and gray lines respectively . The pole is a dark gray circle . The undeflected whisker is translated and rotated in the plane such that it has the same x0 , y0 and θ0 as the deflected whisker . This generates a virtual undeflected whisker ( dashed gray line ) . A virtual pole ( light gray circle ) is generated by shifting the real pole such that it will be tangent to the virtual undeflected whisker . In addition , the distance from the contact point of the virtual unbent whisker and the virtual pole , ( xvirtual , yvirtual ) and the base ( x0 , y0 ) is equal to d , the distance between ( xobj , yobj ) and the base . The angle between the two line segments connecting the base with the real and virtual contact points is θp . DOI: http://dx . doi . org/10 . 7554/eLife . 01350 . 003 The boundary-value problem for whisker shape generally has two solutions , one stable and the other unstable ( Figure 2A ) . During object contact , the whisker shape matches the stable solution since small perturbations from it will decay back to the stable solution ( Strogatz , 1994 ) . The bending of the stable solution is weaker compared to the unstable solution . As the whisker pushes further into the object it becomes increasingly deflected ( Figure 2B ) . At the same time the whisker slides along the object and the arclength from the whisker base to the point of contact , sobj , increases until the whisker detaches from the object . We note two qualitatively different types of detachment . First , under some conditions detachment occurs suddenly before the end of the whisker has reached the object , sobj < LW; we refer to this type of detachment as ‘slip-off’ ( Figure 2B ) . Second , detachment has to occur when the tip reaches the end of the object , sobj = LW; we refer to this type of detachment as ‘pull-off’ ( Figure 2C , D ) . 10 . 7554/eLife . 01350 . 004Figure 2 . Interactions between whiskers and an object . Solutions of the quasi-static model ( Equations 8–14 ) for a conical whisker ( A and B ) and a cylindrical whisker ( C and D ) . The pole is denoted by a gray circle . The resting shape of the whisker is y = Ax2 , where A = 0 . 02 mm−1 ( Quist and Hartmann , 2012 ) , and the whisker touches the pole in the concave backward configuration . ( A ) Two solutions for a conical whisker . For θp = 10° , there are two solutions for whisker shape , one is stable ( solid line ) and one is unstable ( dashed line ) . ( B ) Whisker shape at the saddle-node bifurcation ( θp = 15 . 6° ) . There is only one solution as the stable and unstable solutions coalesce . The object touches the whisker not at the tip . For any larger θp , static solutions cease to exist and the whisker slips off the pole . ( C ) Whisker shape for a cylindrical ‘whisker’ and θp = 10° . Only one solution ( stable ) exists; the unstable solution is not physical because its computed arclength is longer than Lw = 20 mm . ( D ) For θp = 62 . 7° , the tip of the cylindrical ‘whisker’ reaches the object . Beyond this value of θp , the ‘whisker’ is pulled off the object . Parameters for all panels: Lw = 20 mm , rbase= 30 µm , rtip = 1 . 5 µm for the conical whisker and 30 µm for the cylindrical ‘whisker’ , d=15 . 7 mm , E = 3 GPa , x0 = 0 , y0 = 0 . The pole has rpole=0 . 25 mm and its center is located at xcen = 15 . 13 mm , ycen = 4 . 29 mm . DOI: http://dx . doi . org/10 . 7554/eLife . 01350 . 004 Bending of the whisker can be characterized by the angle of the whisker at the point of object contact , θobj ( Figure 1B ) . The whisker first touches the pole at θp = 0 ( Figure 3A–D , open circles ) . As the whisker pushes into the pole ( \|θp\| > 0 ) , θobj changes monotonically ( Figure 3B ) . In the CB configuration , the whisker bends and its shape becomes more ‘concave backwards’ . The force F acting on the whisker increases as more elastic energy is stored in the whisker ( Figure 3C ) ; sobj , also increases ( Figure 3D ) . At a critical θp , the two solutions ( solid lines , dashed lines ) coalesce and disappear at a saddle-node bifurcation ( SNB ) ( Strogatz , 1994 ) ( Figure 3B–D , solid circles ) . No solution exists above this critical θp value , which also corresponds to a critical sobj < LW . The whisker slips suddenly and rapidly past the pole . The saddle-node bifurcation corresponds to slip-off . In the CF configuration , sobj first decreases as \|θp\| increases from 0 , because touch forces straighten the whisker ( Figure 3D ) . With further increases in θp , the whisker bends in the other direction and sobj increases . 10 . 7554/eLife . 01350 . 005Figure 3 . Analysis of conical ( A–E ) and cylindrical whiskers ( F–J ) pushing into a pole . ( A ) Schematic of a conical whisker . Parameters for panels ( B–E ) : Lw = 20 mm , rbase= 30 µm , rtip = 1 . 5 µm , x0 = 0 , y0 = 0 , rpole = 0 . 25 mm , E = 3 GPa . The equation of the undeflected whisker is y = Ax2 where A = 0 . 02 mm−1 ( Quist and Hartmann , 2012 ) . For CB configurations , xcen = 15 . 13 mm , ycen = 4 . 29 mm; for CF configurations , xcen = 14 . 87 mm , ycen = 4 . 71 mm . Positive and negative values of θp correspond to CB and CF configurations respectively . ( B ) θobj as a function of θp . Left , concave forward ( CF ) ; right , concave backward ( CB ) . Solid lines , stable solutions; dashed lines , unstable solutions ( Equations 8–14 ) . Solid circles denote saddle-node bifurcations ( SNB ) . ( C ) Force F as a function of θp . ( D ) Location of object contact along the whisker arc , sobj , as a function of θp . Arrows correspond to Figure 2A ( a ) and Figure 2B ( b , SNB ) . ( E ) The detachment curve in the θp−d plane bounds the parameter regime with a stable solution for a whisker contacting an object . Black lines represent the points when the stable solution coalesces with an unstable solution and disappears via a saddle-node bifurcation ( slip-offs ) . Blue line represents the points where the whisker is pulled off because the tip has reached the object , sobj = LW ( pull-offs ) . ( F ) Schematic of a cylindrical whisker . Parameters as for conical whisker , except that Lw = 20 mm , rbase= rtip = 30 µm . Panels ( G–J ) correspond to panels ( B–E ) . Two object distances are considered in panels ( G–I ) . Arrows in panel i correspond to Figure 2C , D . d = 15 . 7 mm ( blue lines ) corresponds to the pole location used in ( B–D ) . The ends of the blue lines correspond to pull-offs . Additionally , an object distance d = 10 mm is shown ( black lines ) . The black solid circles correspond to slip-offs ( SNBs ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01350 . 005 The regime in which a stable static solution exists for whisker shape can be visualized by plotting a ‘detachment’ curve in the θp−d plane , where d is the distance of the object from the base of the whisker ( Figure 3E ) . The detachment curve is the set of points in the θp−d plane where detachments occur . It encloses an area where stable contacting solutions exist . When the object is close to the face ( small d ) , the whisker contacts the object near its base and a static solution exists for most practical θp values ( peak-to-peak amplitude of whisker movements , 50° [Voigts et al . , 2008; Curtis and Kleinfeld , 2009; O’Connor et al . , 2013] ) . For larger d , the θp regime with a stable solution decreases approximately linearly . Detachments correspond to slip-offs . The range of θp with a stable solution is larger for the CF than the CB configuration . This is consistent with experimental observations ( Quist and Hartmann , 2012 ) and intuition: in the CB configuration the intrinsic curvature aids slip-off . When the object touches near the whisker tip ( large d ) , the saddle-node bifurcation ceases to exist and the whisker is ‘pulled off’ the object . ( Figure 3E , blue line ) . The pull-offs are the result of whisker truncation , and would not occur for a perfect cone . Identical analyses were performed for hypothetical cylindrical whiskers ( Figure 3F–J ) . Although the bifurcation diagrams were superficially similar for conical and cylindrical whiskers ( c . f . Figure 3B–E , G–J ) , cylindrical whiskers exhibit stable solutions at much larger θp . The SNB occurs for θp > 90° ( Figure 3G–I , black lines and solid circles; d = 10 mm ) , which is beyond plausible ranges of whisking since whiskers cannot move into the face . When cylindrical whiskers touch the object close to their end , they are pulled off at moderate θp , because the whisker tip reaches the object ( Figure 3I , blue line ) sobj = LW . Therefore , cylindrical whiskers do not slip-off the pole . For a homogenous cylinder this effect is independent of the cylinder’s bending stiffness and thus its thickness . Our model thus predicts that conical and cylindrical whiskers interact with objects in a fundamentally different way . For a large range of object distances conical whiskers slip past objects , whereas cylindrical whiskers get stuck . This difference is expected to have profound consequences for object-whisker interactions during haptic sensation . We compared our model with measurements made on mouse whiskers ( conical ) and human hair ( cylindrical ) ( Figure 4 ) . Mouse and human hair have similar Young’s Modulus ( Hu et al . , 2010; Quist et al . , 2011 ) . A C2 mouse whisker was mounted on a galvanometer scanner so that its intrinsic curvature was in the plane of whisker movement ( Figure 4A ) . The whisker was then moved slowly ( fgalvo = 0 . 2 Hz , peak-to-peak amplitude 30° ) against a pole . As the whisker rotated into the pole it was deformed until , at a critical θp , it suddenly slipped off the pole ( Figure 4B , Video 1 ) . The red line in Figure 4B shows the whisker immediately before detachment . Whisker slip-offs occurred before the tip of the whisker had reached the point of contact . In contrast , for the cylindrical hair , slip-offs did not occur . Detachments always coincided with the whisker tip reaching the point of contact and were thus pull-offs ( Figure 4C , red line , Video 2 ) . 10 . 7554/eLife . 01350 . 006Figure 4 . Isolated whiskers interacting with cylindrical poles . ( A ) Top-down view of a mouse C2 whisker mounted on a galvanometer scanner . The scanner rotates the whisker into a vertical pole . The distance of the pole from the base of the whisker , d , is varied across experiments . ( B ) Snapshots of the whisker at 32 Hz as it is smoothly rotated ( 0 . 2 Hz , counter clockwise ) into and past the pole . Red line , whisker shape immediately ( <32 ms ) before slip-off . Note that the end of the whisker had not reached the point of object contact . ( C ) Snapshots of a near-cylindrical hair . Red line , hair shape immediately before pull-off . Note that the end of the hair had reached the point of object contact . ( D ) The detachment curve in the θp−d plane . Solid line , theoretical prediction for conical whisker; open circles , experimental measurements for conical whisker . Dashed line , theoretical prediction for cylindrical hair; solid circles , experimental measurements for cylindrical hair . Blue , pull-offs occur because whisker tip has reached the object . Black , slip-offs occurs because of saddle-node bifurcation . Parameters of the conical whisker: Lw = 15 . 25 mm , rbase= 32 . 5 µm , rtip = 2 µm , A = 0 . 02 mm−1 . Parameters of the approximately cylindrical hair: Lw = 15 . 0 mm , rbase= 30 µm , rtip = 26 . 5 µm , A = 0 . 017 mm−1 . Pole radius , rpole = 0 . 25 mm . DOI: http://dx . doi . org/10 . 7554/eLife . 01350 . 00610 . 7554/eLife . 01350 . 007Video 1 . Example video of a conical whisker mounted on the galvo ( Figure 4 ) slipping off a pole . Speed 16fps , 0 . 5x real-time . DOI: http://dx . doi . org/10 . 7554/eLife . 01350 . 00710 . 7554/eLife . 01350 . 008Video 2 . Example video of a cylindrical hair mounted on the galvo ( Figure 4 ) pulling off a pole . Speed 16fps , 0 . 5x real-time . DOI: http://dx . doi . org/10 . 7554/eLife . 01350 . 008 We performed the same type of measurement for multiple object locations along the whisker ( d , Figure 4A ) . The regime of stable interactions between whisker and pole , bounded by the detachment curve , can be visualized in the θp−d plane ( Figure 4D ) . The experimental results were in agreement with the model . For conical whiskers , slip-off occurred before the whisker tip reached the object , and the critical θp decreased rapidly with object distance ( Figure 4D , black circles ) . The observed deviations between the idealized conical model and actual whisker are expected because the whisker is not a perfect cone ( Ibrahim and Wright , 1975 ) and because the whisker’s Young’s modulus may vary slightly along its length ( Quist et al . , 2011 ) . In contrast , the cylindrical hair only pulled off when the whisker tip reached the pole ( Figure 4C ) , with a close fit between experimental and theoretical results ( Figure 4D , blue circles ) . We next tested if slip-offs occur normally during whisker-dependent behavior ( Figure 5 ) . We analyzed data from head-fixed mice trained in a vibrissa-based object location discrimination task ( Pammer et al . , 2013 ) . Mice reported the presence of a pole at a target position ( the ‘Go stimulus’; proximal ) or in a distracter position ( the ‘No Go stimulus’; distal ) ( Figure 5A ) by either licking ( Go response ) or withholding licking ( No Go response ) . In each trial , the pole was presented at a single location . Whiskers were trimmed so that mice performed the task with a single whisker ( C2 ) . For the trials analyzed here the pole distance from the face was randomly chosen from the range d = 7–13 mm ( measured from the follicle; the No Go stimuli ) . We used high-speed ( 500 Hz ) videography and automated whisker tracking to measure the position and shape of the whisker in two mice ( Clack et al . , 2012; Pammer et al . , 2013 ) ( 140 slip events ) . 10 . 7554/eLife . 01350 . 009Figure 5 . Slip-offs during object location discrimination behavior . ( A ) Schematic of a mouse whisking to touch a pole ( experiments from Pammer et al . , 2013 ) . ( B ) Time series ( 250 Hz ) of whisker shape around example protraction slip event . Frame of slip-off is highlighted in red . ( C ) Detachment curves in the θp−d plane for two whiskers . Solid line , theoretical predictions for conical whisker; open circles , experimental measurements for conical whiskers . Dashed line , theoretical predictions for cylindrical hair . Blue , pull-offs . Black , slip-offs . Left , truncated whisker with parameters: Lw = 12 . 5 mm , rbase= 35 µm , rtip = 8 . 5 µm . Right , whisker parameters: Lw = 15 . 3 mm , rbase = 33 . 5 µm , rtip = 2 µm . For both whiskers , intrinsic curvature was y = A ( x−2 . 2 mm ) 2 where A = 0 . 02 mm−1 . DOI: http://dx . doi . org/10 . 7554/eLife . 01350 . 009 In behaving mice , the intrinsic whisker curvature is not parallel to the plane of whisking and imaging ( Towal et al . , 2011 ) . We corrected for the curvature out of the imaging plane using a simple procedure ( ‘Materials and methods’ ) . Furthermore , whiskers exhibit torsion during movement , rotating from a partially concave backward orientation , thru concave down to partially concave forward during protraction ( Knutsen et al . , 2008 ) . We thus define positive and negative θp to denote whisker movement in the protraction and retraction directions respectively , independent of intrinsic curvature . Slip-offs were more likely for more distant object locations ( Figure 5B ) and occurred at larger θp for smaller object distances ( Figure 5C ) . Overall , slip-offs occurred in approximately 15% of behavioral trials . We again compared model and experiment in the θp−d plane . One of the two whiskers was truncated ( Figure 5C , left ) . For objects touching the whisker near the tip , detachments occurred for small θp ( <20° ) , with the whisker tip reaching the pole ( i . e . , pull-offs ) . For smaller object distances slip-off occurred at larger θp ( >20° ) , consistent with a saddle-node bifurcation ( i . e . , slip-offs ) . The second whisker was less truncated ( Figure 5C , right ) . For the object distances tested we observed slip-offs only along the whisker . These results are consistent with our model . The critical θp values for slip-off varied significantly across trials even for identical object distances . This variability is likely caused by differences in whisker movement and whisker elevation across trials . Rodents move their whiskers over objects to explore surfaces . For example , mice can discriminate surface roughness over a few whisking cycles ( Chen et al . , 2013 ) . Texture is likely inferred from the statistics of whisker micromotions produced by the interactions between whiskers and objects ( Diamond , 2010 ) . In particular , as whiskers move over objects whiskers occasionally get stuck , followed by high-velocity slips . The pattern of stick-slip events is highly informative about surface texture ( Arabzadeh et al . , 2005; Wolfe et al . , 2008 ) . We wondered whether whisker shape determines the nature of the stick-slip events underlying texture exploration . We moved a C2 mouse whisker over extra fine ( 600 grit ) sandpaper using a galvanometer scanner ( fgalvo = 0 . 2 Hz; peak-to-peak amplitude , 30° ) while tracking whisker shape in three dimensions using dual view videography ( Figure 6A , Video 3 ) . The tips of mouse whiskers moved along the surface in an irregular manner , during protractions and retractions . Whisker tips were transiently trapped ( Figure 6B , red ) followed by small , high-velocity slips ( Figure 6B , C ) . The pattern of stick-slip events differed for different wall distances ( Figure 6C ) . 10 . 7554/eLife . 01350 . 010Figure 6 . The whisker taper is necessary for slips across textures . ( A ) Dual-perspective imaging of a conical whisker , mounted on a galvo , sweeping across a texture ( 600 grit sandpaper ) . ‘Top’ , Side view; ‘Bottom’ , Top view . ( B ) Conical whisker swept past the texture at four distances: Free air , push distance dp = 0 . 33 mm , 1 . 5 mm and 4 . 5 mm . dp = \|\|x ( Lw ) −x ( 0 ) , y ( Lw ) −y ( 0 ) \|\|−d , where d is the nearest distance from the base of the whisker to the surface . In other words , dp is the distance the surface is moved radially into the whisker beyond just touching . Red traces indicate frames where the whisker tip is stuck , gray traces where the tip is slipping along the surface . Surface texture is schematic and exaggerated . ( C ) Black lines , histograms of tip position over time . Gray lines , trajectories of the whisker tip over the first three whisking periods . Traces are aligned to peak of theta at base . ( D–F ) , as ( A–C ) , but using a cylindrical hair of similar length . Free air , push distance dp = 0 . 33 mm , 2 mm and 3 . 3 mm . Whisker parameters Lw = 16 . 4 mm , rbase= 33 . 5 µm , rtip = 2 µm . Hair parameters as in Figure 4 . DOI: http://dx . doi . org/10 . 7554/eLife . 01350 . 01010 . 7554/eLife . 01350 . 011Video 3 . Composite video of the conical whisker mounted on a galvo slipping across the textured surface in Figure 6B ( 4 . 5 mm push distance ) . Upper video is the side view , lower video is the top view . Speed 32fps , 1x real-time . DOI: http://dx . doi . org/10 . 7554/eLife . 01350 . 011 In contrast , as cylindrical hairs swept across the surface they were trapped during initial protraction and remained trapped for the remainder of the experiment lasting multiple whisking cycles ( Figure 6D–F , Video 4 ) . When the distance between follicle and the site of trapping was shorter than the hair length , the hair buckled out of the plane of movement ( Figure 6E , top ) . The tips of hairs escaped the traps only when the distance to the tip along the path of an undeflected hair exceeded the actual hair arclength ( Figure 6E , F , pane 2 ) . The whisker tip was thus pulled out of the trap ( Figure 3J , blue line ) . These measurements show that the conical whisker shape is critical for the sweeping motions of whisker tips across objects and surfaces , which supports feature extraction via stick-slip events . More generally , conical whiskers can move past walls and objects , which may be critical for positioning of whiskers in confined spaces , such as tunnels , during directed tactile exploration . 10 . 7554/eLife . 01350 . 012Video 4 . Composite video of the cylindrical hair mounted on a galvo getting stuck on the textured surface in Figure 6E ( 3 . 3 mm push distance ) . Upper video is the side view , lower video is the top view . Speed 32fps , 1x real-time . DOI: http://dx . doi . org/10 . 7554/eLife . 01350 . 012 We investigated whether slip-offs convey specific sensory information to cortex . Silicon probes were inserted into the C2 barrel column ( O’Connor et al . , 2013 ) ( Figure 7A ) . We recorded multi-unit activity across cortical layers 2–5 while mice performed an object location discrimination task with the C2 whisker ( O’Connor et al . , 2013 ) . Mice touched the pole multiple times during a trial ( Figure 7B ) . The first touch within a trial caused a large peak in activity with a rapid rise ( Figure 7B , C ) , consistent with previous work ( Simons , 1978; Armstrong-James et al . , 1992; de Kock et al . , 2007; O’Connor et al . , 2010b; O’Connor et al . , 2013 ) . Later touches within a series , during which slip-offs were more commonly seen , produced smaller responses ( Figure 7D ) ( Ahissar et al . , 2001 ) . When slip-off did not occur , the detach-related signals were almost undetectable . In contrast , when slip-off did occur , the detach-related signals were large , comparable to the first touch ( Figure 7F , G ) . 10 . 7554/eLife . 01350 . 013Figure 7 . Neural signals of slip-off in the barrel cortex . ( A ) Silicon probe recording during a pole localization task ( experiments from O’Connor et al . , 2013 ) . ( B ) Spikes and whisker forces , one behavioral trial . ‘Top’ , multi-unit activity . Arrow , slip-off event . ‘Bottom’ , contact induced forces . Solid circles , time points with non-zero contact-mediated forces calculated from changes in whisker curvature . A 2 ms period during which the whisker was slipping off was removed as the quasi-static model is invalid for such highly dynamic events . ( C–F ) Multi-unit spike responses in the barrel cortex ( shank < 300 μm from the center of the C2 barrel ) . ( C ) Activity aligned to the first touch in a trial ( 720 responses; two animals , three sessions , six electrode shanks ) . ( D ) Same as ( C ) , but aligned to the last touch before a behavioral response ( i . e . , lick ) ( 720 responses ) . ( E ) Same as ( C ) , but aligned to the moment of detach on the last touch before a behavioral response in trials without slip-offs ( 392 responses ) . To prevent contamination by touch-onset this analysis was restricted to touches that were longer than 50 ms . ( F ) Aligned to slip-off ( 34 responses ) . ( G ) Change in spike rate triggered by the event ( activity 10–30 ms post event , minus activity −50 to 0 ms pre event ) . Error bars , SEM . Pairwise comparison showed no significant difference in evoked spikes between first touch and slip-off groups ( p=0 . 20 ) , a significant difference between last touch and slip-off ( p=0 . 037 ) and significant differences between all other groups ( every pair , p<10−12 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01350 . 013 We have developed a mathematical framework for whisker deflection in the context of dynamical systems theory ( Figures 1–3 ) to explore the functional consequences of whisker taper . Recent findings have shown that whisker taper is used as a ruler by mice to gauge the distance to objects with a single whisker ( Pammer et al . , 2013 ) . Tapered whiskers have resonance frequencies that are robust to wear and truncation damage to their tips ( Williams and Kramer , 2010 ) . Tapered whiskers also detach from objects at shallower push angles than cylindrical whisker substitutes . Here we go beyond prior observations and uncover fundamental differences in how tapered and untapered hairs interact with objects; tapered whiskers slip-off when the contact point is along the whisker body ( sobj < Lw ) , whereas cylindrical hairs require the tip to be pulled off ( sobj = Lw ) for biologically plausible push angles ( Figure 3E , J , Figure 4 ) . Thus , tapered whiskers have greater freedom of movement past obstacles compared to cylindrical hairs . This mobility is seen in reduced preparations ( Figure 4 ) and also during active behavior , as whisker slip-off occurs in a variety of object location discrimination tasks ( Figures 5 , 7 ) . The intrinsic difference in detachment also produces qualitatively different interaction patterns during palpation of textured surfaces ( Figure 6 ) . Tapered whiskers sweep past with stick-slip micromotions , whereas cylindrical whiskers become immobilized on surface imperfections . Theoretical treatment of whisker mechanics is a necessary foundation for understanding how sensory input shapes neural representations of the tactile world . Previous work has computed whisker deflections based on the quasi-static solution of the Euler-Bernoulli equation ( Birdwell et al . , 2007; Williams and Kramer , 2010; Quist and Hartmann , 2012; Pammer et al . , 2013 ) . Aspects of whisker vibrations have also been treated , including resonant frequencies ( Hartmann et al . , 2003; Neimark et al . , 2003 ) and wave propagation following contact-induced impulses ( Boubenec et al . , 2012 ) . We framed whisker-object interactions in the language of boundary-value problems . This allowed us to carry out bifurcation analysis and distinguish stable from unstable shapes . We demonstrate that for conical whiskers there are only two possible solutions for whisker shape for a given object distance and push angle , one stable , one unstable . We identify how the saddle-node bifurcations separating the two branches of solutions vary as a function of parameters , such as θp ( Figure 3 ) . This is not possible using previously developed numerical approaches ( Birdwell et al . , 2007; Quist and Hartmann , 2012 ) . Slip-offs occur suddenly at a critical push angle , θp , corresponding to the angle where these saddle-node bifurcations occur . Curves of saddle-node bifurcations in a two-parameter plane define the regime in which stable solutions can be obtained ( Figure 3E , J ) . For conical whiskers the critical push angles are within the normal range of whisking ( Figure 4 ) . For cylindrical hairs the critical angles fall outside the range of whisking . Thus our theory predicts that conical and cylindrical whiskers will interact with objects in a fundamentally different manner . Our theory addresses the effects of whisker truncations ( Figures 3–5 ) . Truncations of conical whiskers make whisker behavior more ‘cylindrical-like’ ( Figure 5 ) . The intuition obtained from our analysis led us to distinguish between the dynamics of conical whiskers and cylindrical hairs during sweeping across textures ( Figure 6 ) . We compared theoretical predictions with videos of whiskers and cylindrical hair rotated into a steel pole . In situations where the whisker curvature was contained within a plane , the agreement between theory and experiment was very good ( Figure 4 ) , despite our model ignoring frictional forces . The small remnant differences between theory and experiment are due to deviations of whisker geometry from perfect conical shape ( AH , KS , DG , unpublished ) and possible inhomogeneities in the Young’s modulus ( Quist et al . , 2011 ) ( but also see Carl et al . , 2012 ) . In behaving mice the whisker droops out of the plane of whisking ( Towal et al . , 2011 ) . Rigorous treatment of whisker deflection by an object would thus require a three-dimensional model . We developed a phenomenological model to predict slip-offs even for behaving mice ( ‘Materials and methods’ ) , which produced qualitative agreement with experiments ( Figure 5 ) . Our results suggest several functions for which the conical shape of rodent whiskers is evolutionarily adaptive . Within their natural habitat , many rodents , including house mice ( Berry , 1968 ) and African pouched mice ( Ellison , 1993 ) , live in group nests consisting of chambers connected by long , body-width tunnels . During running , whiskers point forward to touch unanticipated objects . When a new object is encountered rodents foveate their whiskers on the object for fine-scale exploration ( Grant et al . , 2009 ) . Within these dark , radially constrained tunnels , navigation ( Vincent , 1912; Dehnhardt et al . , 2001 ) , localization of objects ( Hutson and Masterton , 1986; Krupa et al . , 2001 ) , social touch ( Wolfe et al . , 2011 ) , and determination of friend or foe ( Anjum et al . , 2006 ) demands freedom of whisker motion . Whiskers have to be moved past the rough walls of the tunnel . Without the flexibility provided by whisker taper , the whiskers could be trapped in a far protracted or retracted orientation , causing a tactile ‘blind-spot’ . Whisker taper is also desirable outside of constrained spaces . A major sensory avenue for the localization and identification of objects and their properties is via directed sweeping of whiskers across object surfaces ( Carvell and Simons , 1990; Ritt et al . , 2008; von Heimendahl et al . , 2007; Wolfe et al . , 2008 ) . During artificial periodic palpation of fine-grained textures , conical whiskers traversed the surface with complex micromotions , whereas cylindrical hair became trapped against the surface ( Figure 6 ) . Although a precise understanding of the interaction between a tapered whisker and textured surface during a stick-slip event has not been treated mathematically , it is likely that forces at the tip build up until they bend the whisker tip sufficiently to free it from traps . In the cylindrical case , the constant bending stiffness of the body and tip render the whisker incapable of transmitting sufficient lateral force to buckle the much stiffer tip and release it from the surface . Beyond mechanical maneuverability , do slips contribute to the neural representation of tactile sensation ? During active whisking , stick-slip micromotions on textured surfaces drive sparse , precisely timed spikes in barrel cortex that provide a sensory cue for surface texture ( Jadhav et al . , 2009 ) . Neural responses in barrel cortex to repeated contacts between whiskers and objects show strong adaptation during active touch ( Ahissar et al . , 2001; Crochet et al . , 2011 ) and object location discrimination ( Figure 7C , D , G ) . Despite occurring when the circuit is adapted to touch , slip-offs produce strong volleys of cortical activity , of comparable magnitude to pre-adapted touch ( Figure 7F ) . Thus , slip-related excitation can overcome cortical touch adaptation and likely contributes to sensation and perception in a variety of tactile behaviors ( Arabzadeh et al . , 2005 ) . We model whiskers as truncated cones with length LW , base radius rbase , and tip radius rtip ( Figure 1A ) . The conical shape is virtually extended to a perfect cone of length L . The whisker is located in the x-y plane . The arclength along the whisker , s , is s = 0 at the base , s = sobj at the point of object contact , s = LW at the tip , and s = L at the virtual tip ( Figure 1B ) . The whisker base is located at point ( x0 , y0 ) , and the positions of a point along the whisker is ( x ( s ) , y ( s ) ) , 0≤s≤LW . The running angle between the whisker and the x-axis is θ ( s ) , and θ ( 0 ) = θ0 . The whisker radius is rw = ( L−s ) rbase/L and the area moment of inertia is I ( s ) =πrw44 . The Young’s modulus is E = 3 GPa ( Birdwell et al . , 2007; Quist et al . , 2011; Pammer et al . , 2013 ) . Similar calculations were carried out for cylindrical hair with rw = rbase . The bending stiffness of the whisker is the product EI ( s ) . In the absence of contact with an object , the intrinsic curvature of the whisker is κi ( s ) . The object is a cylindrical pole oriented perpendicular to the x-y plane with radius rpole , centered at ( xcen , ycen ) . Upon contact , the whisker touches the object at ( xobj , yobj ) with angle θobj ( Figure 1B ) , where ( 1 ) xobj=xcen−rpolesinθobj yobj=ycen+rpolecosθobj The Euclidian distance between the whisker base and the contact point is d . The object applies force F→ on the whisker: ( 2 ) F→= ( Fx , Fy ) = ( −Fsinθobj , Fcosθobj ) At steady state , the shape of the whisker is determined by the solution of the static Euler-Bernoulli equation ( Landau and Lifshitz , 1986; Birdwell et al . , 2007; Williams and Kramer , 2010; Pammer et al . , 2013 ) ( 3 ) dθds=κi ( s ) +Mz ( s ) EI ( s ) where Mz is the component of the bending moment M→=r→×F→ perpendicular to the x-y plane and r→ ( s ) = ( xobj−x ( s ) , yobj−y ( s ) ) , together with the equations ( 4 ) dxds=cosθ ( 5 ) dyds=sinθ Substituting Equations 1 , 2 in Equation 3 , we obtain ( 6 ) dθds=κi ( s ) +FEI ( s ) [ ( xcen−rpolesinθobj−x ) cosθobj+ ( ycen+rpolecosθobj−y ) sinθobj] We seek a solution for Equations 4–6 given the boundary conditions at the base ( x0 , y0 and θ0 ) , and that the whisker contacts the pole at an ( initially unknown ) arclength sobj . Given the shape of an undeflected whisker as a function of the running arclength s , namely ( x , y ) = ( g ( s ) , h ( s ) ) , the intrinsic curvature is ( 7 ) κi= g' ( s ) h'' ( s ) −h' ( s ) g'' ( s ) {[g' ( s ) ]2+[h' ( s ) ]2}where d/ds is denoted by ‘ . The shape of the undeflected whisker is considered to be parabolic , y = Ax2 ( Quist and Hartmann , 2012 ) . To compute the whisker shape Equations 4–7 are transformed to a form of a boundary-value problem ( BVP ) by introducing a variable σ = s/sobj ( 8 ) dxdσ=sobjcosθ ( 9 ) dydσ=sobjsinθ ( 10 ) dθdσ=sobjg' ( sobjσ ) h'' ( sobjσ ) −h' ( sobjσ ) g'' ( sobjσ ) {[g' ( sobjσ ) ]2+[h' ( sobjσ ) ]2}32+sobjFE ( sobjσ ) I ( sobjσ ) [ ( xcen−rpolesinθobj−x ) cosθobj+ ( ycen+rpolecosθobj−y ) sinθobj] ( 11 ) dFdσ=0 ( 12 ) dθobjdσ=0 ( 13 ) dsobjdσ=0 The differential Equations 8–13 are solved on the interval 0 ≤ σ ≤ 1 together with the equations ( 14 ) I ( sobjσ ) =π4 ( L−sobjσL ) 4 The boundary-value conditions for σ = 0 are: x ( 0 ) = x0 , y ( 0 ) = y0 , θ ( 0 ) = θ0 . The conditions for σ = 1 are: x ( 1 ) = xcen − rpole sin θobj , y ( 1 ) = ycen + rpole cos θobj , θ ( 1 ) = θobj . Solutions to Equations 8–14 have physical meaning if sobj < LW . If the whisker tip reaches the contact point ( sobj = LW ) the whisker detaches because it is pulled off the pole . We solved six differential Equations 8–13 together with their boundary conditions to find six unknown variables ( x , y , θ , F , θobj , sobj ) as functions σ on the interval 0 ≤ σ ≤ 1 . The variable θobj is treated as a separate variable from θ , but the boundary condition θ ( 1 ) = θobj guarantees that the solution is self-consistent . The equations were solved numerically using the iterative shooting method ( Press et al . , 1992 ) . We start with guessed initial values for the unknown variables for σ = 0 and integrate the differential equation until σ = 1 . The initial conditions are then varied to reduce the difference between the given boundary conditions and those that are obtained by the most recent integration . The method converges if the initial conditions are sufficiently close to the solution . We begin by solving the boundary-value problem with θobj corresponding to θp = 0 ( i . e . , the whisker is barely touching the pole ) . We then vary θobj slightly , compute the whisker shape , and repeat the process until the desired θobj is reached . We used the boundary-value problem solver software package XPPAUT ( Ermentrout , 2002 ) . The software package AUTO ( Doedel , 1981 ) , which is incorporated into XPPAUT , was used to compute bifurcation diagrams ( Figure 3 ) , by following the solutions of the boundary-value problem as parameters , such as θobj , vary . The static solution of Equations 8–14 is a fixed point of a spatiotemporal dynamical system representing whisker movement . The full dynamical system can be formulated as a partial differential equation only for small θp and straight beams ( Timoshenko , 1961; Boubenec et al . , 2012 ) . Since the full dynamical system for all θp and beams with intrinsic curvature is not known we cannot linearize a dynamic equation . However , the static solution for small θp must be stable . In addition , bifurcation theory implies that if we increase θp the solution will coalesce with an unstable solution and they both disappear , via a saddle-node bifurcation ( Strogatz , 1994 ) . In principle , a branch of stable solutions can lose stability via a Hopf bifurcation before the saddle-node bifurcation . Slip-off will occur at θp values smaller than predicted by our quasi-static theory . The good correspondence between the computed saddle-node bifurcation and the experimentally measured value for θp at slip-off shows that the static solution disappears via a saddle-node bifurcation ( Figure 4 ) . Undeflected whiskers can be modeled as parabolas within a plane ( Towal et al . , 2011 ) . In the work reported here the whisker is contained entirely within a plane perpendicular to the pole . For a whisker with intrinsic curvature , contact occurs in either the ‘concave backward’ ( CB ) or ‘concave forward’ ( CF ) directions ( Figure 1C ) ( Quist and Hartmann , 2012 ) . To quantify contact strength , we use the push angle θp ( Figure 1D ) ( Quist and Hartmann , 2012 ) . Suppose a whisker originates at ( x0 , y0 , θ0 ) and touches a pole at ( xobj , yobj ) . We plot an undeflected whisker with the same ( x0 , y0 , θ0 ) , and find a point along the whisker with the same Euclidian distance d from ( x0 , y0 ) as ( xobj , yobj ) , defined as ( xvirtual , yvirtual ) The angle between the two rays starting at ( x0 , y0 ) towards ( xobj , yobj ) and ( xvirtual , yvirtual ) is defined as θp . By convention , we define the sign of θp to be positive for CB and negative for CF . Exact treatment of whisker deflection in behaving rodents demands a three-dimensional model that is outside the scope of this work . Instead , we developed a phenomenological two-dimensional model ( Figure 5 ) . First we assume that the whisker touches the object in a concave-down configuration . Second , the deflection of the whisker is described by the two-dimensional model ( Equation 8–14 ) when the projection of the whisker on that plane is treated as a two-dimensional whisker . The area moment of inertia ( I ) is computed by estimating the arclength s along the real whisker from the whisker projection and using this value in Equation 14 . This correction in s was on the order of 3% . We measured the length of the isolated whisker . Estimating the whisker base is inaccurate because of the fur on the face . We therefore determined the effective whisker length from the estimated whisker base to the tip based on the video recordings . If the whisker slips off at its tip , we find the maximal projected length during events of slip-off at the tip . If there are no such events , we compute the projected length that yields the theoretically-obtained slip-off at the largest d for which slip-off is obtained . For all cases , this estimated value is less than 1 mm smaller than the length measured directly . For galvo experiments ( Figures 4 , 6 ) , we used plucked , full-grown mouse C2 whiskers . The shapes of these whiskers were measured under a light microscope at high magnification ( Pammer et al . , 2013 ) . The follicle ends of the whiskers were embedded in the barrel of a cut 21 gauge needle filled with Kwik-Cast silicon sealant ( World Precision Instruments , Sarasota , FL ) . Needles were mounted on the top edge of a galvo scan mirror ( 6800HP; Cambridge Technology , Bedford , MA ) . Whiskers were then rotated into a cylindrical object ( steel Wiretrol II plunger; Drummond Scientific , Broomall , PA ) at 0 . 2 Hz , 30° peak-to-peak amplitude ( Figure 4 ) . Dual-perspective imaging confirmed that whiskers remained in the concave forward or concave backward orientation during the interaction with the pole ( data not shown ) . The same whiskers were used for imaging whisker motion across textured surfaces ( Figure 6 ) . The surface was fine sandpaper ( 600 grit ) rigidly mounted on a glass slide and positioned perpendicular to both planes of imaging . A variety of human hair was characterized . Hair from an Asian female closely matched the whisker diameter close to the base and was used as a cylindrical hair . The hair dimensions were: base diameter , 60 μm; tip diameter , 53 μm; length , 15 . 0 mm . High-speed videography was used to measure the position and shape of mouse whiskers during galvo experiments ( Figures 4 , 6 ) ( X-PRI camera , 32 fps , 0 . 6 ms exposure , 8-bit depth , AOS Technologies , Switzerland ) and behavior ( Figures 5 , 7 ) ( 1000 fps , 0 . 2 ms exposure , 8-bit depth , Basler 504 k , Germany ) . Pixel size was 0 . 07 mm ( Figure 5 ) , or 0 . 031 mm ( Figure 7 ) , or 0 . 032 mm ( Figures 4 , 6 ) . Illumination was with a 940 nm infrared LED delivered through a diffuser and condenser lens and projected directly into the camera . A silver mirror ( PFSQ10-03-P01 , Thor Labs , Newton , NJ ) reflected an orthogonal side view projection onto the same camera ( Knutsen et al . , 2008 ) . Videos were split and cropped prior to whisker tracking . Whiskers were tracked with the Janelia Whisker Tracker ( Clack et al . , 2012 ) ( https://openwiki . janelia . org/wiki/display/MyersLab/Whisker+Tracking ) . The whisker medial axis is stored as an array of points ( xi , yi ) , i = 1 , … , N , where N is on the order of several hundreds . To remove artifacts associated with tracking variation at the base when calculating θ0 , the angle of the whisker base was determined by a linear fit of the fifth through tenth points closest to the base . Forces acting on the whisker ( Figure 7B ) were calculated using published methods ( O’Connor et al . , 2010a; Clack et al . , 2012; Pammer et al . , 2013 ) . The behavioral task and apparatus have been described in detail elsewhere ( O’Connor et al . , 2010a; O’Connor et al . , 2013; Pammer et al . , 2013 ) . Briefly , head-fixed mice judged the distance to a metal pole that was presented at a range of positions along the whisker in the radial dimension ( Figure 5 ) or horizontal dimension ( Figure 7 ) . For radial discrimination , a proximal position ( 5 mm radially from follicle ) was defined as the Go position , distal positions ( 7–13 mm ) were defined as No Go positions . For horizontal discrimination , Go and No go positions were separated by 4 . 5 mm along a parallel to the anteroposterior axis of the mouse at a radial distance of 7–11 mm . Mice performed object location discrimination with a single C2 whisker . Within two days of the behavioral experiments we plucked whiskers and measured their shape and material properties using a macroscope and microgram balance ( Mx5; Mettler Toledo , Columbus , OH ) ( Pammer et al . , 2013 ) . Parts of the electrophysiology dataset is a reanalysis of previously acquired data ( see O’Connor et al . , 2013 for detailed methods ) . During a head-fixed pole location discrimination task , a 32 channel , four shank silicon probe ( Buz32 , Neuronexus , Ann Arbor , MI ) was lowered into barrel cortex , with an estimated tip depth of 375–720 μm from the pia . Prior to insertion , probes were painted with DiI . Following recordings , mouse brains were fixed , stained for cytochrome oxidase and tangentially sectioned to determine the location of the shanks within the barrel field . Shanks within 300 μm of the center of C2 were included for analysis for slip-off responses ( two animals , three behavioral sessions , six shanks ) . Following common signal subtraction , bandpass filtering between 300 and 6 , 000 Hz , spike extraction of 4 s . d . threshold crossing , and spike merging , multiunit responses were aligned to whisker behavioral events . Spikes with peaks <307 . 5 μs jitter on the same shank were considered a single spike . Each multiunit was the sum of activity on all eight electrodes on a single shank ( six total multiunit recordings ) . Slip-off events were rare ( 17 total in three sessions ) compared to detach without slip-off . Significance was calculated as unpaired two-tailed t-tests on the difference between the number of spikes in the period 10–30 ms post event and the 50 ms prior to event normalized to the respective period lengths followed by Bonferroni–Holm correction for multiple comparisons .	When foraging in dark , confined spaces , mammals use the information gathered by their whiskers to ‘see’ the world around them . Mammalian whiskers come in a variety of shapes and sizes , most likely reflecting the way in which they are used . Rodent whiskers are conical and precisely tapered , whereas some harbor seals have flattened whiskers with wave-like undulations . Human hair is cylindrical . Rodents sweep their whiskers back and forth over objects and surfaces without moving their head . They use this process , called whisking , to build up a three-dimensional picture of objects . Whisking allows the rodent to estimate where an object is located , how big it is , and what kind of surface texture it has . Information about surface texture can , for example , help the animal to distinguish a stone from a seed . Hires et al . have used theoretical and experimental methods to analyze the interaction of mouse whiskers with objects . The conical shape of a mouse whisker makes the tip thousands of times more flexible than the base . Hires et al . show that this flexibility gradient allows the whiskers to slip past objects close to the face and to move freely across rough surfaces . Cylindrical whiskers , on the other hand , become stuck behind nearby objects and get caught on tiny features in an object’s surface texture . Hires et al . conclude that conical whiskers are advantageous in the tight confines of the tunnels that mice live , forage and socialize in , because they are able to gather a more complete sensory picture of their surroundings . The maneuverability of the whiskers also allows the mouse to move their whiskers forwards or backwards when rough tunnel walls are close by . By contrast , the sticking experienced by cylindrical whiskers would lead to ‘blind spots’ . In addition to providing insights into the ways that mice interact with their environment , this work could also lead to improvements in the design of the canes used by the visually impaired to navigate human environments .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "neuroscience" ]	2013	Tapered whiskers are required for active tactile sensation
How do DNA transposons live in harmony with their hosts ? Bacteria provide the only documented mechanisms for autoregulation , but these are incompatible with eukaryotic cell biology . Here we show that autoregulation of Hsmar1 operates during assembly of the transpososome and arises from the multimeric state of the transposase , mediated by a competition for binding sites . We explore the dynamics of a genomic invasion using a computer model , supported by in vitro and in vivo experiments , and show that amplification accelerates at first but then achieves a constant rate . The rate is proportional to the genome size and inversely proportional to transposase expression and its affinity for the transposon ends . Mariner transposons may therefore resist post-transcriptional silencing . Because regulation is an emergent property of the reaction it is resistant to selfish exploitation . The behavior of distantly related eukaryotic transposons is consistent with the same mechanism , which may therefore be widely applicable . Transposons are discrete sections of mobile DNA that are amplified as they move from one location in the genome to another . They have had a profound influence on our evolutionary history by providing DNA duplications and rearrangements as substrates for natural selection ( e . g . Hua-Van et al . , 2011 ) . Although retrotransposons , which transpose via an RNA intermediate , can persist for long periods in a given genome , DNA transposons appear to rely on regular horizontal transfer ( Lohe et al . , 1995; Hartl et al . , 1997 ) . In the period following horizontal transfer the transpositional activity of the element is under positive selection . However , once there are multiple copies in the genome , selection is relaxed because a freely diffusing pool of transposase acts on all copies of the element . Gradually , with time , as various copies of the element acquire detrimental mutations , the pool of active transposase is poisoned by dominant-negative complementation leading to the extinction of the transposon ( Lohe et al . , 1995; Hartl et al . , 1997; Le Rouzic and Capy , 2005; Le Rouzic et al . , 2007 ) . Whilst this mechanism likely dictates the ultimate fate of a eukaryotic DNA-transposon , it remains unknown how control is exerted in the short- to medium-term after an active transposon first arrives in a genome . This is a serious problem because transposition is inherently exponential and each new copy of the element is a source of further new copies . The fact that long-lived multicellular organisms tolerate transposons suggests the existence of control mechanisms . Host-mediated epigenetic responses may provide some level of protection . However , these are probably never entirely effective , particularly for euchromatic copies of the transposon ( e . g . Kelleher and Barbash , 2013 ) . Indeed , modeling suggests that unregulated transposition results either in the demise of the transposon or the demise of the host ( Le Rouzic and Capy , 2005 ) . Experiments with a non-autonomous Mos1 element in Drosophila provided the first experimental evidence of autoregulation ( Lohe and Hartl , 1996 ) . The phenomenon , termed overproduction inhibition ( OPI ) , was revealed by a reduction in the frequency of excision when multiple copies of the transposase gene are present , or when transposase is over-expressed from a heat-shock promoter . OPI was later shown to affect other elements , both in vivo and in vitro ( Lampe et al . , 1998; Clark et al . , 2009; Claeys Bouuaert and Chalmers , 2010 ) . Our present work focuses on a resurrected copy of the Hsmar1 transposon , which was active in the human genome about 50 million years ago ( Robertson and Zumpano , 1997; Cordaux et al . , 2006; Liu et al . , 2007; Miskey et al . , 2007 ) . It is closely related to Mos1 and is a member of the mariner family , which is probably the most successful group of transposons in nature , as judged by the depth and breadth of its phylogenetic distribution ( e . g . Robertson and Lampe , 1995; Feschotte and Wessler , 2002 ) . The mariner family is a good model system to address autoregulation because of the demands placed by horizontal transfer on the mechanism of transposition and its control . Ideally , the mechanism should be independent of specific host factors and should allow a rate of transposition sufficiently high to protect the founding element from genetic drift . High activity also provides a high probability of integration into a vector , such as a virus or endosymbiont , which may mediate horizontal transfer . However , this must be balanced against a detrimental effect on host fitness ( e . g . Le Rouzic and Capy , 2005 ) . Here we present a mechanism for autoregulation based on a biochemical analysis of Hsmar1 transposition . The model reveals how the kinetics of a genomic invasion are dictated by the multimeric state of the transposase and the order in which the various components are recruited into the developing transpososome . Once a certain number of copies of the transposon are established in the genome , and the transposase concentration is above a critical threshold , the mechanism provides a steady-state rate , perfectly damping the exponential amplification which is a natural consequence of unregulated transposition . The mechanism is an emergent property of the transposition reaction , based on the competition of active transposase multimers for their primary binding sites at the transposon ends . It is therefore robust and resistant to selfish exploitation . The key prediction of the model is that doubling the transposon copy number , which is tantamount to doubling the transposase concentration , halves the rate of transposition . In vivo transposase dose-response curves for Hsmar1 and the distantly related Sleeping Beauty ( SB ) and piggyBac ( PB ) transposons fit this condition , suggesting they are all regulated in the same way . The documented mechanisms for autoregulation in the bacterial transposons Tn10 and Tn5 provided the starting point for our investigation ( Kleckner , 1990; Reznikoff , 2008 ) . The respective transposases belong to the same family of nucleotidyl-transferases as the Hsmar1 transposase , and use the same cut-and-paste mechanism of transposition ( e . g . Hickman et al . , 2010; see Figure 2A below for an illustration of cut-and-paste transposition ) . The cut-and-paste mechanism does not increase the number of transposons directly . Amplification requires that the empty donor site is repaired by homologous recombination from a sister chromosome , or that the transposon is excised behind a replication fork and integrates in front . However , unless stated otherwise we will assume a maximally efficient reaction in which each transposition event converts one copy of the element into two copies . In Tn10 and Tn5 , the first stage of the reaction is when transposase monomers ( T ) bind to the ends of the element ( A and B ) and bring them into a synapsis ( Figure 1A ) . This complex is known as a paired ends complex ( PEC ) or transpososome . We will refer to this mechanism of transpososome assembly as ‘synapsis-by-protein-dimerization’ ( S-PD ) . A kinetic diagram for the model is provided in Figure 1B . The first catalytic steps of the reaction are double strand breaks at the transposon ends , which are irreversible owing to the loss of enthalpy ( subsumed into k3 in the diagram ) . Any regulation must occur before the formation of a productive synapsis because almost all transposons that achieve the first nick go on to complete cleavage and integration ( Chalmers and Kleckner , 1994 , 1996 ) . 10 . 7554/eLife . 00668 . 003Figure 1 . Dynamic models for genome invasions . ( A ) The mechanism of transpososome assembly in Tn10/5 . In the bacterial elements Tn10 and Tn5 synapsis is mediated by the dimerization of monomers bound to either end of the transposon . T , transposase; T2 , transposase dimer; A and B , transposon ends; PEC , paired ends complex . ( B ) The kinetic model embedded in the computer simulation for the S-PD mechanism of synapsis . Abbreviations are as given in part ( A ) ; Tfree and Tnsb , free and non-specifically bound transposase . ( C ) Simulation of the S-PD model . Kinetic parameters: k0 , 9 . 9 × 106 M−1s−1; k−0 , 139 s−1; k1 , 3 . 8 × 108 M−1s−1; k−1 , 1 . 2 × 10−2 s−1; k2 , 12 . 7 s−1; k−2 , 1 × 10−10 s−1; k3 , 1 . 4 × 10−3 s−1 . Sources of the values are given in Table 2 . ( D ) As in part ( C ) but the transposase is 99% cis-acting . 1% of the transposase leaks into the bulk phase and acts on all copies of the element . ( E ) As in part ( C ) but the transposon expresses a trans-acting inhibitor which reduces the transposase concentration . The inhibitor is 1000-fold more active than the transposase , which approximates the situation in Tn10 where the inhibitor is an antisense RNA . ( F ) The cis-acting transposase and the trans-acting inhibitor from parts ( D ) and ( E ) are combined . ( G ) The mechanism of transpososome assembly in mariner . A transposase dimer bound to one transposon end recruits a naked transposon end . Abbreviations as in part ( A ) . ( H ) The assembly-site occlusion model . When the transposase concentration is low , most transposons are occupied by only one transposase dimer , leading to productive synapsis ( bottom left ) . When the transposase concentration is high , a transposon may be occupied by two transposase dimers , causing overproduction inhibition ( OPI , top right ) . SEC2 is a transposon with a transposase dimer bound to one of the ends . ( I ) Kinetic model embedded in the computer simulation for the ASO mechanism . Abbreviations as given in parts ( A ) and ( B ) . ( J ) Simulation of the S-NEC mechanism and comparison with S-PD . Parameters as in part ( C ) . ( K ) The time scale of the S-NEC simulation in part ( J ) is extended . DOI: http://dx . doi . org/10 . 7554/eLife . 00668 . 003 To explore the dynamics of a genomic invasion by a transposon that uses the S-PD mechanism for transpososome assembly we used the kinetic diagram in Figure 1B to develop a computer model ( ‘Materials and methods’ ) . At the start of the simulation a single transposon appears in the genome and initiates an exponential amplification ( Figure 1C ) . Changing the values of the kinetic constants governing the model changes the scales on the axes but does not alter the exponential character of the reaction . Note that in this case the scale on the X axis is unrealistically short because we allow a productive synapsis to form products almost instantly ( the value of k3 in Figure 1B is large ) . We are thus ignoring the time that would be required in vivo for the maturation of the integration products and repair of the donor site . This allows us to focus on the earlier stages of the reaction where regulation takes place . Amplification of the Tn10 and Tn5 transposons , which use the S-PD mechanism , is controlled by a combination of two effects ( Kleckner , 1990; Reznikoff , 2008 ) . Firstly , the transposases do not diffuse freely and tend to act in cis to the encoding element . This slows the amplification considerably because it prevents the effective concentration in transposase from rising with transposon copy number . However , it does not alter the exponential nature of the curve ( Figure 1D ) . Secondly , the transposons express a trans-acting inhibitor , which in the case of Tn10 is an antisense RNA directed against the transposase mRNA . This strategy is also ineffective in dampening the exponential amplification , even if the inhibitor is much more abundant than the transposase ( Figure 1E ) . In contrast , the combination of a cis-acting transposase and a trans-acting inhibitor is very effective ( Figure 1F ) . This combined strategy is not viable in a eukaryotic host because of the physical separation of transcription and translation . Transposons using the S-PD mechanism would experience exponential amplification in a eukaryote . Whilst the Tn10/5 transposases are monomers in solution , the Hsmar1 enzyme behaves as a dimer in gel filtration experiments ( not shown ) . Furthermore , several mariner transposases readily form a complex known as single end complex 2 ( SEC2 ) , which is composed of a transposase dimer bound to a single transposon end ( Lipkow et al . , 2004; Auge-Gouillou et al . , 2005; CCB & RC data not shown ) . We would like to propose that SEC2 gives rise to the transpososome by capturing a naked transposon end ( Figure 1G ) . We will refer to this mechanism as synapsis-by-naked-end-capture ( S-NEC ) . Recruitment of a naked transposon end immediately suggests a mechanism that can account for OPI ( Figure 1H ) . As the concentration of transposase rises , transposition is inhibited by the progressive reduction in the number of unbound ends available for synapsis . We will refer to this as an assembly-site-occlusion model ( ASO ) . A kinetic diagram for the model is provided in Figure 1I , following the same conventions as before , except that T2 represents the transposase dimer . To explore the dynamics of mariner transposition , we simulated the S-NEC mechanism of transpososome assembly in a second computer model ( ‘Materials and methods’ ) . Using an identical set of parameters , S-NEC has an advantage over the S-PD mechanism in terms of the rate at early time points ( Figure 1J ) . However , as the invasion progresses , the S-NEC model is dampened by ASO , which eventually dominates the reaction ( Figure 1K ) . The peak rate of the reaction is when the free-transposase concentration is equal to its equilibrium binding constant . The ASO model provides testable predictions . The simplest is that the rate of transposition is governed by the ratio of transposase to transposon ends , and not by the absolute concentrations . We tested this using the in vitro excision assay ( Figure 2A , but please also consult Figure 2—figure supplement 1 for a guide to interpretation of the in vitro reactions ) . Excision of the transposon leaves behind the vector backbone , which is an end product of the reaction and provides a direct measure of the efficiency . With 10 nM substrate , the reaction peaked at 25 nM transposase and then declined sharply , as judged by the amount of backbone produced . When the concentration of substrate was increased or decreased by 10-fold , the amount of transposase required for peak activity changed in direct proportion . This is consistent with the ASO model , and excludes models in which OPI arises from a concentration-dependent mechanism of transposase aggregation . 10 . 7554/eLife . 00668 . 004Figure 2 . Experimental test of the ASO model for mariner transposition . ( A ) The efficiency of transposition depends on the ratio of transposase to DNA . The cartoon to the left of the gels illustrates the mechanism of cut-and-paste transposition using a supercoiled ( SC ) substrate . First strand nicking generates an open circular product ( Nick . ) . Second strand nicking at one end yields the linear single-end break product ( Lin . ) . A similar set of nicks at the other transposon end yield the plasmid backbone ( BB ) plus the excised transposon . Transposase was titrated into reactions with three different substrate concentrations . Reactions contained 1 . 5 µg of supercoiled substrate plasmid in a volume of 500 µl , 50 µl and 5 µl , which provided a final concentration of 1 nM , 10 nM and 100 nM , respectively . Transposase was diluted so that the addition of one tenth of the respective reaction volumes achieved the indicated concentrations . Reactions were incubated for 4 hr at 37 °C and deproteinated . Photographs of ethidium bromide stained agarose gels are shown . Consumption of the supercoiled substrate and production of the plasmid backbone both provide a measure of the efficiency of the reaction . Inter . , product of intermolecular integration of the transposon into an unreacted substrate . ( B ) Preassembly of the paired-end complex protects from OPI . Four sets of staged reactions were assembled as indicated . Set A was a standard transposition reaction with 6 . 7 nM plasmid substrate and 20 nM of transposase except that the components were pre-incubated for 3 hr before the addition of the catalytic Mg2+ ion . Set A′ was identical except that Ca2+ was present during the 3 hr pre-incubation period , which enhances PEC assembly . Sets B and B′ were identical to A and A′ except that the mixture was supplemented with 50 nM transposase 30 s before the addition of Mg2+ . OPI inhibited the reaction in Set B . In contrast , the stable PEC assembly supported by the presence of Ca2+ protected Set B′ from the inhibitory effects of the excess transposase added just before the catalytic Mg2+ ion . Photographs of ethidium bromide stained agarose gels are shown . See also Figure 2—figure supplement 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 00668 . 00410 . 7554/eLife . 00668 . 005Figure 2—figure supplement 1 . Kinetic analyses of OPI and interpretation of in vitro transposition reactions . The kinetics of the transposition reaction were analyzed in standard reactions containing 6 . 7 nM plasmid substrate and the indicated concentrations of the transposases . Photographs of ethidium bromide stained agarose gels are shown . ( A ) The effects of the ASO mechanism complicates the interpretation of the gels , particularly the kinetic analyses . In the reactions with 20 nM transposase about half of the substrate is converted to the nicked intermediate in the first 2 min . However , it takes an hour or more to consume the remainder of the substrate . This biphasic behavior arises from the ASO mechanism , which operates in any reaction that contains more than one dimer of transposase . Thus , even if the transposase was 100% active and the reaction was performed with the optimum ratio of one dimer to one transposon , only half of the substrate would react initially . At the start of the reaction half of the transposons would be occupied by a dimer and would react , a quarter would be occupied by two dimers and would suffer OPI and the other quarter would be completely unoccupied . The slow phase of consumption corresponds to the redistribution of dimers from transposons which initially suffered OPI to those that were completely unbound . Even under OPI conditions , when both ends of the transposon are occupied by transposase dimers , their occasional dissociation provides a window of opportunity for synapsis . OPI can therefore be overcome by extending the incubation period so that the accumulated windows of opportunity eventually suffice to complete the reaction . Thus , the reaction with 40 nM transposase goes on to reach completion during the 8 hr incubation , despite initially suffering from OPI . With 200 nM transposase , the extended incubation is unable to overcome OPI because the windows of opportunity provided by unoccupied transposon ends are shorter at this concentration . ( B ) The nicked substrate is more sensitive to OPI and is inhibited at only 40 nM transposase . This is because any free transposon ends , made available by the dissociation of a transposase dimer , are re-bound before they can achieve synapsis . In other words , the slow synapsis of the nicked substrate requires a longer window of opportunity than is available at 40 nM transposase . Time points and transposase concentrations as in part ( A ) . ( C ) The RA104 transposase mutant is resistant to OPI because the higher OFF-rate provides more windows of opportunity for synapsis ( see Figure 4C and text for details ) . Time points and transposase concentrations as in part ( A ) . DOI: http://dx . doi . org/10 . 7554/eLife . 00668 . 005 Another prediction is that pre-assembly of the PEC will sequester the transposon ends in productive interactions , and protect the reaction from the subsequent addition of excess transposase . To test this hypothesis we measured the kinetics of four transposition reactions , which we assembled in different stages prior to the addition of the catalytic Mg2+ ions at time zero ( Figure 2B ) . Set A is a standard transposition reaction in which the components were incubated for 3 hr prior to the addition of Mg2+ . Set A′ is identical except that Ca2+ was present during the pre-incubation period . Although Ca2+ does not support DNA cleavage , it supports assembly of the PEC ( Claeys Bouuaert and Chalmers , 2010 ) . In Set A′ this is evident from the faster consumption of the supercoiled substrate . Upon addition of Mg2+ , 50% of the substrate was converted from the supercoiled to the nicked form within 1 min . Set B is identical to Set A , except that extra transposase was added 30 s before the catalytic metal ion at time zero ( Figure 2B ) . As expected , this inhibited the reaction and very little substrate was converted into product . Set B′ is identical to Set B except that Ca2+ was present during the pre-incubation period , before the addition of excess transposase . The reaction kinetics were then identical to those of Set A′ . Pre-assembly of the transpososome is completely effective in the relief of OPI because the competition for free transposon ends is irrelevant at this stage of the reaction . Transposition is thus regulated during assembly of the transpososome , before productive synapsis of the ends . We have assumed that each copy of the transposon contributes a certain amount of transposase to the nucleus and that this is where transposition takes place . We therefore estimated the rate of diffusion in this compartment and calculated its effects on the association rate constants for specific and non-specific DNA ( ‘Materials and methods’ ) . Compared to the in vitro rates , the values were reduced by about twofold , but their effects cancel each other out and the rate of transposition is unchanged as a result . However , the effect of diffusion on the rate of synapsis is much greater because of the drag experienced by a chromosomally-bound protein . By treating the DNA with the worm-like chain model we obtained an estimate for ‘segmental-diffusion’ , which describes the rate of motion on a short scale , such as the distance between the transposon ends ( ‘Materials and methods’ ) . Our calculated value is consistent with the experimentally determined rate of looping derived from lox/Cre recombination in mammalian cells ( Ringrose et al . , 1999 ) . When the calculated diffusion rates are applied in the computer model , a pseudo-steady-state rate of transposition is established when there are about 200 copies of the transposon ( Figure 3A , compare with Figure 1K ) . This steady-state rate persists well beyond the copy numbers typically achieved during a genomic invasion ( Figure 3—figure supplement 1 ) . 10 . 7554/eLife . 00668 . 006Figure 3 . A semi-quantitative description of mariner transposition . ( A ) The ASO model for mariner transposition was simulated after taking account of the slow diffusion which prevails in vivo: k0 , 5 × 106 M−1s−1; k1 , 1 . 9 × 108 M−1s−1; k2 , 4 . 5 × 10−4 s−1 . Other parameters were as in Figure 1K . ( B ) Transposase binds the transposon end rapidly and tightly . Binding reactions ( 20 µl ) contained 40 fmol of 32P-labeled transposon end and 120 fmol of transposase . The reactions were incubated at 37 °C and separated on a native polyacrylamide gel . The lane indicated as time zero contains no transposase . The 30 s time interval was the time required to mix the sample , load the gel and apply the voltage . SEC1 and SEC2 represent a single transposon end bound by a transposase monomer and dimer , respectively . An autoradiogram is shown . ( C ) Binding reactions were set up and analyzed as in part ( B ) . After 5 min incubation , to allow the complexes to form , a 20-fold molar excess of unlabeled transposon end ( 'cold ITR' ) was added . The lane indicated as time zero contains transposase but no cold competitor . The rate of transposase dissociation can be estimated from the amount of free DNA released from the complexes . ( D ) The rate of transposon excision from a nicked substrate provides an estimate of the rate of synapsis . The kinetics of a transposition reaction was analyzed in standard reactions containing 6 . 7 nM of nicked plasmid substrate and 20 nM transposase . A photograph of an ethidium bromide stained agarose gels is shown . ( E ) The ASO model was simulated as in Figure 1K but taking account of the effects of allostery , as described in the main text and ‘Materials and methods’ . Parameters as in Figure 1K except k2 = 9 . 6 × 10−5 s−1; k−1 = 5 . 8 × 104 s−1 . ( F ) As in part ( E ) but the rate of transposition is divided by the transposon copy number . ( G ) As in part ( A ) , but taking account of allostery and slow diffusion in vivo: k0 , 5 . 0 × 106 M−1s−1; k1 , 1 . 9 × 108 M−1s−1; k2 , 3 . 4 × 10−9 s−1 . The effects of changing transposition efficiency are shown ( see text for definition of efficiency ) . ( H ) The relationship between transposition efficiency and the relative rate of transposition is plotted with the maximum rate scaled to 1: y = 1 . 4427 ln ( x ) . See also Figure 3—figure supplements 1 and 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 00668 . 00610 . 7554/eLife . 00668 . 007Figure 3—figure supplement 1 . The time scale of the graph in Figure 3A is extended . The parameters take account of the slow diffusion in vivo , but take no account of allosteric interactions between transposase subunits . This plot therefore represents the in vivo dynamics of a generic homomeric DNA looping reaction in the absence of allostery . DOI: http://dx . doi . org/10 . 7554/eLife . 00668 . 00710 . 7554/eLife . 00668 . 008Figure 3—figure supplement 2 . EMSA analysis of transpososome assembly shows that SEC1 arises from dissociation of the PEC . Transposon ends were encoded on radiolabeled linear DNA fragments . Binding reactions were incubated at 37 °C for 2 hr , separated on a 5% native polyacrylamide gel and recorded by phosphoimaging . ( A ) SEC1 and SEC2 represent a single transposon end bound by a transposase monomer and dimer , respectively ( see main text for details ) . SEC2 comes to dominate the reaction as the transposase concentration rises . There is a significant transition between 4 and 8 nM transposase when SEC1 largely disappears . This corresponds to the point at which the transposon ends become sub-stoichiometric to the transposase dimers . At this point no free transposon ends remain as they are all bound by transposase . According to the S-NEC mechanism ( see Figure 1 for details ) , SEC2 is converted to the transpososome ( =PEC ) by recruitment of a naked transposon end . OPI occurs when the transposon ends are sub-stoichiometric to transposase dimers and there is a shortage of free transposon ends available for recruitment . Note that the various species observed in these binding reactions are identical to those observed in reactions with the related Mos1 and Himar1 transposons , which display the same behavior for example ( Dawson and Finnegan , 2003; Lipkow et al . , 2004 ) . The present data suggests that the PEC in all three of these related systems is unstable in the EMSA and dissociates into two SEC1 complexes soon after the start of electrophoresis . Thus , in agreement with the S-NEC mechanism , SEC1 disappears at the point in the titration when the transposon ends become sub-stoichiometric to transposase dimers . ( B ) A fixed amount of transposase was titrated with an increasing amount of free transposon ends . The appearance of SEC1 coincides exactly with the appearance of the free transposon ends , which are required for PEC assembly in the S-NEC model . As the amount of transposon ends is increased further , the amount of SEC1 increases . This reflects mass action , which drives PEC assembly by favoring the capture of a free transposon end ( see part D below for confirmation ) . This supports the data in part ( A ) in suggesting that SEC1 arises from the dissociation of the PEC . ( C ) Binding reactions were with a single-chain transposase dimer , in which two monomers are concatenated by a linker peptide joining the C-terminus of one to the N-terminus of the other . Concatenation of the subunits stabilizes the PEC , which is now detected in the gel . As the transposase concentration increases , the PEC disappears at the same point as the free DNA and gives way to SEC2 . This behavior is identical to SEC1 in parts ( A ) and ( B ) . The single-chain dimer of transposase is fully proficient for the transposition reaction ( not shown ) , demonstrating that SEC1 is not an obligate intermediate of the reaction . This supports the data in parts ( A ) and ( B ) further confirming that SEC1 arises from the dissociation of the PEC . ( D ) In vitro transposition reactions were performed with a plasmid substrate encoding a single transposon end . Reactions were stopped at the indicated times and deproteinated before analysis on a 1 . 1% agarose TBE gel stained with ethidium bromide . All three sets of transposition reactions contained the same amounts of transposase and supercoiled plasmid substrate . However , the respective reaction volumes were adjusted to 500 µl , 50 µl and 5 µl to achieve the indicated concentrations . Transpososome assembly requires bimolecular synapsis between ends located on different molecules , as illustrated below the gels . This is inefficient owing to the relatively low concentration of one transposon end with respect to another when on separate molecules . When such a transpososome performs cleavage , followed by integration into an unreacted substrate molecule , the product is a linear molecule three times the size of the substrate ( Claeys Bouuaert et al . , 2011 ) . There is very little reaction when the substrate concentration is low . This reflects the inefficiency of second end recruitment . At high substrate concentration , mass action drives the reaction by favoring second-end recruitment . DOI: http://dx . doi . org/10 . 7554/eLife . 00668 . 008 Up until now we have assumed that synapsis of the transposon ends is the product of two sequential , chemically identical , collision events: a transposase dimer collides with and binds first one transposon end and then the other . This system is useful to illustrate the fundamental dynamics of a generic DNA-looping reaction , and how it is governed by the concentration of the looping protein . However , the Hsmar1 transpososome is more complicated because of communication between the transposase subunits . The allostery was first noted in competition experiments using different topological forms of the substrate ( Claeys Bouuaert et al . , 2011 ) . The experiments revealed that transposase binds quickly and tightly to the first transposon end it encounters , but has a much lower affinity for the second transposon end ( Claeys Bouuaert et al . , 2011 ) . To see how this affects the rate of transposition we estimated the association and dissociation rate constants , and the rate of synapsis , in in vitro experiments . To estimate the association rate constants ( k1 in Figure 1I ) we used an electrophoretic mobility shift assay ( EMSA ) . Assembly of SEC2 was complete after 30 s , which was the minimum time required to load the gel and apply the voltage ( Figure 3B ) . Although this did not provide a direct measure of k1 , it was consistent with the rapid site-specific binding of a helix-turn-helix protein and we therefore retained the previous best estimate from the literature ( ‘Materials and methods’ ) . To estimate the dissociation rate constant ( k−1 in Figure 1I ) we challenged SEC2 with a 10-fold molar excess of unlabeled transposon end ( Figure 3C ) . The complex dissociated slowly and about 50% remained after 20 min . We therefore estimate that k−1 is 5 . 8 × 10−4 s−1 . This is some 20-fold lower than the value for most helix-turn-helix proteins ( ‘Materials and methods’ ) . It is possible that the slow dissociation reflects a second aspect of the aforementioned allostery . In this experiment we also observed single-end-complex 1 ( SEC1 ) , which contains a monomer of transposase bound to a single transposon end . It arises from decay of the PEC , which is unstable in the EMSA ( Figure 3—figure supplement 2A–C ) . SEC1 is highly abundant in this experiment because the excess of cold competitor ends drives PEC assembly by mass action ( Figure 3—figure supplement 2D ) . To measure the rate of synapsis ( k2 ) we used a transposon encoded on an open circular plasmid ( Figure 3D ) , which is probably the most relevant substrate because eukaryotic DNA contains very little free supercoiling . The half-time for synapsis was estimated at 2 hr , which corresponds to a pseudo-first order rate constant of 9 . 6 × 10−5 s−1 ( ‘Materials and methods’ ) . When the experimental estimates of k−1 and k2 were applied in the computer model for the generic DNA looping reaction in vitro , the rate of transposition was greatly reduced and the accelerating phase at the beginning of the reaction was largely eliminated ( Figure 3E , compare with Figures 1K and 3A ) . Although the overall rate of transposition remains constant , the rate for any particular copy of the element decays exponentially ( Figure 3F ) . Finally , when we also take account of the slow diffusion in vivo , the rate of transposition is reduced even further ( Figure 3G , solid line ) . Cut-and-paste transposition relies on the host homologous recombination machinery to reinstate a copy of the transposon at the donor site following excision . The highest rate of amplification , which we define as an efficiency of two , is achieved if , furthermore , the transposon excises after passage of a replication fork and inserts into an un-replicated region ahead of the fork . In this situation , one element on one chromosome becomes four elements on two sister chromosomes . The rate of transposition is quite sensitive to the efficiency of the reaction ( Figure 3G , H ) . If the transposon inserts into a replicated region of the chromosome , or if the donor site is not reinstated , or if the excised transposon fails to reintegrate , the efficiency is reduced to 1 . 5 and the rate of transposition is halved . The rate of transposition may also be negative ( Figure 3H ) . The computer model reveals an inverse linear relationship between the association rate constant ( k1 ) and the rate of transposition ( Figure 4A ) . Increasing the affinity of the transposase for the transposon end reduces the rate of transposition , and vice versa . This is because the transposase acts primarily as an inhibitor of its own activity once the steady-state is established . In principle , this behavior could cause the preferential amplification of transposons which had acquired mutations in the transposase binding sites . However , when we reduced the association rate constant by 100-fold it was clear that the advantage was accompanied by a penalty in the form of a lag phase at the start of the invasion ( Figure 4B ) . A low-affinity founding element would therefore be vulnerable to genetic drift during the period before sufficient copies had been produced to sustain the transposase concentration required for efficient amplification . 10 . 7554/eLife . 00668 . 009Figure 4 . Association and dissociation rate constants . ( A ) Simulation as in Figure 3E , with k1 fivefold up or down . We have retained the effects of allostery but here and in subsequent simulations we have ignored the effects of the slow diffusion in vivo . This allows the algorithm to run more smoothly and shortens the scale on the x axis , but does not affect the conclusions , which are based on the differential responses to changing the various parameters . ( B ) Simulation as in part ( A ) ( solid line ) , with k1 100-fold down . ( C ) Binding reactions with the RA104 mutant transposase were set up as in Figure 3B . In the lane with 20 nM transposase the smear between SEC2 and the position of free DNA is probably due to complexes that have dissociated during electrophoresis . Autoradiogram of an EMSA is shown . ( D ) The kinetics of the transposition reaction were analyzed in standard reactions containing 6 . 7 nM of supercoiled plasmid substrate and the indicated concentrations of the transposases . Photographs of ethidium bromide stained agarose gels are shown . With 200 nM wild-type transposase , the windows of opportunity for synapsis , provided by periods when one transposon end is unoccupied , are too short to allow for synapsis . The RA104 transposase mutant is resistant to OPI because the higher dissociating rate provides more windows of opportunity for synapsis . ( E ) Mutant and wild type transposase were assayed in HeLa cell culture with 8 ng or 1000 ng of helper plasmid and 500 ng of neomycin resistant reporter plasmid . DOI: http://dx . doi . org/10 . 7554/eLife . 00668 . 009 To verify this property of the model we mutagenized the transposase to lower its affinity for the transposon end . The RA104 mutation is located within the second helix-turn-helix motif , and abolishes an ionic interaction with the phosphodiester backbone ( Roman et al . , 2007; Richardson et al . , 2009 ) . In the EMSA the RA104 transposase produced very little SEC2 , even when the standard transposase concentration was increased threefold ( Figure 4C , compare with Figure 3B ) . However , in the in vitro plasmid transposition assay , RA104 consumed the supercoiled substrate faster than wild type ( Figure 4D ) . Presumably , the mutant protein redistributes more quickly from those plasmids initially suffering OPI due to the double occupancy of their transposon ends . When the transposase concentration was increased 10-fold the wild type activity was almost abolished , whereas the mutant was almost unaffected ( Figure 4D ) . We next turned to an established eukaryotic cell culture assay . It is based on the co-transfection of a helper plasmid , expressing transposase , and a reporter plasmid , encoding a transposon that confers neomycin resistance when it integrates into a host chromosome ( e . g . [Grabundzija et al . , 2010] ) . The rate of transposition is given by the number of stable transfectants obtained after drug selection . With the optimum amount of helper plasmid RA104 was only 25% as active as wild type . However , with higher levels RA104 was less severely inhibited ( Figure 4E ) . Thus , the RA104 transposase reflects the behavior of the low affinity element in Figure 4B , which is at a disadvantage to wild type when the transposase concentration is low . Once a transposon is established in the genome , ASO provides a constant rate of transposition . With a given set of kinetic parameters for transposase binding and synapsis , the actual rate achieved depends on the amount of transposase expressed by each copy of the transposon ( Figure 5A ) . Reducing the expression level by fivefold increases the steady-state rate by the same factor . However , it takes slightly longer to establish the final rate because of the difficulty locating the transposon ends at the start of the invasion when the transposase concentration is low . The limits of this effect are reached when the expression of transposase dimers is fractionally sub-stoichiometric to transposon ends ( Figure 5B ) . Since ASO requires at least two dimers per transposon , the accelerating phase is never dampened and amplification , although slow at first , is exponential . 10 . 7554/eLife . 00668 . 010Figure 5 . Transposase expression level . ( A ) Simulation as in Figure 3E showing the effects of changing the transposase expression fivefold up or down . Unless noted otherwise , each transposon copy produces 500 transposase dimers , which are considered as being contained within a 500 fl nucleus . ( B ) As in part ( A ) ( 1× line ) but with a transposase expression level of 1 . 9 dimers per transposon . ( C ) Parameters as part ( A ) ( 1× line ) , plotting the relationship between transposase expression and the rate of transposition at the point in the invasion when there are 1000 copies of the transposon present in the genome . ( D ) HeLa cell assay for Hsmar1 transposition . HeLa cells were transfected with 500 ng of neomycin reporter plasmid and the indicated amount of helper plasmid expressing transposase . The rate of transposition is given by the number of neomycin resistant colonies recovered after drug selection . Bars indicate standard error of the mean , n = 3 . R2 is least squares goodness of fit to the line y = ax−1 . ( E ) As in part ( D ) but with isogenic Sleeping Beauty ( SB100X ) transposon reporter and helper plasmids . Data points are a mean of three experiments and were extracted and re-plotted from Figure 2A of ( Grabundzija et al . , 2010 ) . ( F ) As in part ( D ) but with isogenic piggyBac transposon reporter and helper plasmids . n = 3 . See also Figure 5—figure supplement 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 00668 . 01010 . 7554/eLife . 00668 . 011Figure 5—figure supplement 1 . Transposase expression in HeLa cells . HeLa cells were transfected with an increasing quantity of the helper plasmid , which expresses Hsmar1 transposase from a strong viral promoter . The titration was identical to that shown in Figure 5D . Cells were harvested after two days and analyzed by SDS-PAGE and Western blotting with antibodies against β-tubulin and the Hsmar1 transposase . Transposase expression increased throughout the titration . This indicates that transposase expression is not toxic . If transposase expression had been toxic the amount of transposase recovered from the cells would have plateaued or decreased with higher amounts of helper plasmid transfected . DOI: http://dx . doi . org/10 . 7554/eLife . 00668 . 011 We next considered a genome with 1000 copies of the transposon and calculated the rate of transposition over a range of transposase expression levels ( Figure 5C ) . This revealed an inverse-exponential relationship between transposase expression and the rate of transposition . This is noteworthy because it suggests that once the steady-state is established , a post-transcriptional silencing response may potentiate the genomic invasion . It also suggests that titration of the transposase by non-autonomous elements may not contribute to the demise of a transposon as has been previously suggested . To verify the inverse relationship between the transposase concentration and the rate of transposition , we titrated the helper plasmid in the HeLa cell culture assay ( Figure 5D ) . The inhibitory response followed the power law relationship predicted from the ASO mechanism ( y = ax−1 ) . A Western blot was used to confirm that transposase expression increased with increasing amounts of helper plasmid ( Figure 5—figure supplement 1 ) . We also examined the transposase dose-response relationships for two other eukaryotic DNA transposons ( Figure 5E , F ) . Sleeping Beauty ( SB ) is distantly related to Hsmar1 within the mariner/Tc1 superfamily ( Goodier and Davidson , 1994; Ivics et al . , 1997 ) . The lepidopteran transposon piggyBac is very distant from Hsmar1 ( Fraser et al . , 1996 ) . In both cases the transposase dose-response curve matched the predicted power law relationship . Mariner transposons are widely distributed in plants and animals where genome sizes and nuclear volumes vary greatly . In the S-NEC computer model , the rate of transposition is barely sensitive to volume changes less than 100-fold ( Figure 6A ) . This is because changes in the amount of transposase bound to specific and non-specific sites almost exactly cancel each other . The S-PD mechanism of synapsis responds in the opposite direction ( Figure 6B ) . 10 . 7554/eLife . 00668 . 012Figure 6 . Genome size and nuclear volume . ( A ) Simulation as in Figure 3E , changing the nuclear volume by the indicated amounts . ( B ) Simulation of regulated S-PD mechanism as in Figure 1F , changing the nuclear volume by the indicated amounts . ( C ) Simulation as in part ( A ) , changing the genome size by the indicated amounts . ( D ) Simulation as in part ( B ) , changing the genome size by the indicated amounts . DOI: http://dx . doi . org/10 . 7554/eLife . 00668 . 012 Increasing the size of the genome in the S-NEC/ASO model increases the rate of transposition in direct proportion ( Figure 6C ) . This is a result of the inverse relationship between the transposase concentration and the rate of transposition , which was verified in Figure 5D . The effect arises from the facts that the number of non-specific binding sites scales with the genomes size , and that these relieve OPI by absorbing free transposase . Whether or not non-specific sites are composed of DNA or chromatin does not matter . The ASO mechanism thus tailors the rate of transposition according to the genome size . The S-PD model responds in the opposite direction , and smaller genomes receive progressively higher rates of transposition ( Figure 6D ) . The rice transposon Osmar14 has a secondary transposase binding site close to one of the two principal binding sites at the transposon ends ( Yang et al . , 2009 ) . It was suggested that transposase binding at the secondary site may down-regulate transposition . To investigate the behavior of such a system , we introduced an inhibitory binding site into the S-NEC reaction scheme ( Figure 7A ) . The steady-state rate of transposition was reduced in proportion to the binding affinity of the transposase . Note that the behavior of the system would be identical if the transposon instead expressed a transposase variant or a completely different protein , provided that binding interfered with transposition . 10 . 7554/eLife . 00668 . 013Figure 7 . Variations and alternative mechanism of regulation . ( A ) The S-NEC mechanism is simulated as in Figure 3E but with a secondary transposase binding site that inhibits transposition when occupied . 0 , the secondary site has no affinity for the transposase; 1× , the secondary site has the same affinity for transposase as the primary binding sites at the transposon ends; 2× , secondary site with twice the affinity for transposase . ( B ) The dimerization end occlusion model for OPI . ( C ) All possible states of the DEO model are illustrated . A and B are transposon ends , T is an active transposase monomer , T2 is an inhibitory transposase dimer . Solid lines indicate reaction pathways allowed in the model . The active species is shown in red . ( D ) Simulation of the DEO model using in vitro , non-allosteric parameters as in Figure 1C . Inhibitory dimers bind transposon ends with the same affinity as the active monomers . Active monomers dimerize with the same affinity as they bind transposon ends . ( E ) DEO model as in part ( D ) but with in vivo slow diffusion rate as in Figure 3E . Affinity of the monomers in the inhibitory dimer is doubled as indicated . ( F ) DEO model as in part ( E ) showing the effect of doubling in the genome size and nuclear volume . ( G ) DEO model as in part ( E ) showing the effect of changing the association rate constant of the active monomers and the inhibitory dimers by the indicated amount . DOI: http://dx . doi . org/10 . 7554/eLife . 00668 . 013 In the context of the S-PD reaction scheme , a secondary transposase binding site is identical to the trans-acting inhibitor modeled in Figure 1E . Thus , even if the inhibitory binding site is stronger than the principal binding sites , it will not halt the exponential amplification , but merely slow it down . This highlights the weakness of ‘secondary site’ models for regulation in the absence of the dose-dependent inhibition provided by the ASO mechanism . ASO is reminiscent of the dimerization end-occlusion model ( DEO ) first proposed for Tn5 and later in more detail for the Mos1 mariner transposon ( Weinreich et al . , 1994; Townsend and Hartl , 2000 ) . The model postulates an S-PD mechanism of transpososome assembly , accompanied by the concentration-dependent formation of inactive dimers , which bind and occlude the transposon ends ( Figure 7B ) . Tn5 was later shown to be regulated by other factors ( above ) , and the current biochemical analysis excludes the DEO mechanism for mariner transposition . Nevertheless , DEO is a viable model and we wished to explore its properties: firstly to compare them with ASO , and , secondly , to facilitate its recognition should it be present in any natural element that may be characterized in the future . All potential states of the DEO model are illustrated in Figure 7C . However , in constructing a computer model we used a slightly simplified version in which only free monomers produce inactive dimers . In this scheme there is a single potentially active species , shown in red . As previously demonstrated , the model provides the dose-dependent inhibition needed to prevent exponential amplification ( Townsend and Hartl , 2000; Figure 7D , E ) . The eventual steady-state rate is inversely proportional to the stability of the inhibitory dimer ( Figure 7E ) . As in the ASO model , the steady-state rate is also proportional to the genome size ( Figure 7F ) . However , larger genomes experience a longer lag phase at the start of the invasion . This is an inherent disadvantage of the S-PD mechanism of transpososome assembly and reflects the improbability of transposase monomers binding to both ends of the element simultaneously when their concentration is low ( Figure 1J ) . Interestingly , once established , the steady state rate is independent of the association and dissociation rate constants , k1 and k−1 , provided that dimerization does not change either of these values ( Figure 7G and not shown ) . Transposition in somatic cells leaves no trace in the genomic fossil record and does not help the vertical spread of the transposon through the population . It is also detrimental to the host , particularly in organisms where there is a risk of cancer . Furthermore , since somatic cells may spend a long time in the G1 phase of the cell cycle , when double strand breaks are processed by non-homologous end joining ( NHEJ ) , transposition may cause a net decrease in the number of transposon copies in the event of a failure to reintegrate ( Figure 3H ) . The only selective advantage of transposition in soma that we can envisage is that it provides an opportunity for integration into a vector , which may mediate horizontal transfer . Indeed , mariner transposes readily in the soma and is among the most widespread of elements in nature . In contrast , the P element has a limited phylogenetic distribution and its activity is restricted to the germ line ( Engels , 1983 ) . Transposition in the germ line aids the vertical spread of the transposon in the population and leaves a genomic fossil record of some of the events , which we still observe today . Cut-and-paste transposition is favored in the germ line because homologous recombination , which is required to reinstate the donor site , is more prevalent than in the soma throughout all phases of the cell cycle ( Robert et al . , 2008; Tichy et al . , 2010; Serrano et al . , 2011 ) . Synchronization of transposition with S phase of the cell cycle would provide a further advantage due to the general up-regulation of homologous recombination and the chance of integrating ahead of a replication fork . This strategy was first observed in the bacterial IS10 ( Roberts et al . , 1985 ) . P element transposition may also be linked to the cell cycle because it preferentially integrates near origins of replication , which will ensure its early replication ( Spradling et al . , 2011 ) . A counter example is that expression of SB transposase is reported to extend G1 and promote the repair of excision sites by NHEJ in the soma ( Walisko et al . , 2006 ) . It is unknown whether mariner transposition is synchronized with the cell cycle . However , this seems plausible because Hsmar1 transposition is greatly stimulated by negative supercoiling ( Claeys Bouuaert et al . , 2011 ) , which probably exists transiently behind an advancing replication fork . The values of the kinetic parameters determine the magnitude of the steady-state rate of transposition as well as the copy number at which it is achieved . The rate of synapsis is a dominant factor , and is in turn influenced by the multimeric state of the transposase , the rate of diffusion and , in the case of Hsmar1 , by the allostery between the subunits . At the start of an invasion , a preformed transposase multimer overcomes the improbability of two monomers simultaneously binding opposite ends of the same element when their concentration is low . The advantage arises from the continuity of the DNA between the transposon ends , which ensures that the first dimer-bound-site has a high concentration with respect to its partner . The molar concentration of one DNA site with respect to another is abbreviated as jM , and is about 55 nM for the ends of a 1 . 3 kb transposon , such as Hsmar1 . The magnitude of this value dictates the effectiveness of the ASO mechanism . In dilute solutions , where synapsis is fast , OPI is not significant until the free transposase concentration is a significant fraction of jM . This is evident in Figure 1K where 50% inhibition is not achieved until there are 107 copies of the transposon , which corresponds to 40 haploid human-genomes of DNA . However , under the slow diffusion regime that prevails in vivo , OPI becomes significant much earlier , and the steady state is achieved when there are about 200 copies of the transposon ( Figure 3A ) . Allostery further increases the effectiveness of ASO by reducing the affinity of the developing transpososome for the second transposon end . This increases the competition for binding sites to such an extent that the steady-state rate is established when there are only a few copies of the transposon ( Figure 3G ) . The influence of jM on the rate of transposition helps to explain the success of miniaturized transposons , which often greatly outnumber their cognate master elements ( e . g . Yang et al . , 2009 ) . However , these relationships have additional complexities . Osmar14 is an autonomous transposon in the rice genome and provides the cognate transposase for Ost35 , a more numerous miniature element ( Yang et al . , 2009 ) . This relationship is not obvious from a comparison of the respective transposon end sequences , which appear quite divergent . Our finding that OPI is relieved by reducing the affinity between the transposase and the transposon end provides a further explanation for the unexpected efficiency of this cross-mobilization in addition to their short length ( Figure 4A ) . Another factor that may explain why Osmar14 is less numerous than Ost35 is that it has a secondary transposase binding site near its 3′-end , which inhibits transposition and may represent a mechanism of autoregulation ( Yang et al . , 2009 ) . However , as noted already , a repressive binding site such as this does not change the underlying dynamics of ASO ( Figure 7A ) , nor does it dampen exponential amplification in the absence of ASO or a similarly effective mechanism ( Figure 1E ) . Another type of secondary-site model is one in which transposase represses its own expression . This is conceptually similar to the cis-action observed in some bacterial systems , in that it prevents the concentration of transposase rising in proportion to the transposon copy number . The weakness of such models is that they only prevent the transposase concentration from rising and can never cause a reduction ( Townsend and Hartl , 2000 ) . For example , expression of the P element transposase in the soma is repressed in part by binding of a truncated splice variant . However , co-expression of the truncated transposase with the full-length version in the germ line is not sufficient to establish the repressive P-cytotype , which probably requires piwi RNA arising from specific telomeric copies of the element ( Misra and Rio , 1990; Brennecke et al . , 2008; Jensen et al . , 2008 ) . A more general argument against the effectiveness of ‘secondary-site’ models is that they are vulnerable to selfish exploitation by any element that lacks the site . Thus , the burden of Ost35 transposition , which lacks the secondary site , would presumably negate any advantage accruing to Osmar14 as a result of its self restraint . ASO is less vulnerable than the secondary-site models because it relies on a competition between active multimers of the recombinase for their primary binding sites at the transposon ends . Although it can , in principle , be subverted by the resistance of low-affinity elements to OPI , this effect can only operate within limits ( Figure 4B , note the slow amplification when the copy number is low ) . Although not completely immune , the ASO mechanism is robust to this type of selfish exploitation . ASO also provides a homeostatic mechanism in that a doubling of the transposase concentration always halves the rate of transposition per element , irrespective of the values of the kinetic constants governing the reaction . A larger genome will thus always receive a proportionately higher rate of transposition . In the case of mariner , the relative genetic burden is thus independent of the host genome size and is dictated by adaptive features of the transposon itself . These features include the transposase binding affinity , the strength of the promoter in a given host and the strength of the allosteric coupling between subunits . The latter is a powerful determinant able to change the rate of synapsis over many orders of magnitude . Indeed , allostery provides significant adaptive flexibility , which is unavailable in the S-PD reaction mechanism . Finally , by buffering the rate of transposition against changes in the transposase concentration , ASO may also provide a degree of protection from post-transcriptional silencing and heterochromatinization responses . The wild type and RA104 transposases were expressed from pRC880 and pRC1128 , respectively ( Claeys Bouuaert and Chalmers , 2010 ) . Unless stated otherwise , each 50 µl transposition reaction contained 6 . 7 nM of plasmid substrate and 20 nM transposase in 20 mM Tris-HCl pH8 , 100 mM NaCl , 2 mM DTT , 2 . 5 mM MgCl2 and 10% glycerol . The standard substrate was pRC650 , which encodes a mini-transposon with 30 bp Hsmar1 transposon ends . Plasmid pRC919 has a single 30 bp Hsmar1 end cloned into the pBluescript polylinker . Open circular substrates were prepared using the Nb . BsrDI endonuclease , which nicks the plasmids at several sites some distance away from the transposon ends . Transposition reactions were incubated at 37 °C and analyzed by loading 400 ng of the DNA on each lane of a TBE-buffered 1 . 1% agarose gel . After electrophoresis , the gel was stained with ethidium bromide , destained in water , and photographed on a 310 nm transilluminator using a DC290 camera ( Kodak , Rochester , NY ) with a 590 DF bandpass filter . Digital photographs were quantified using the Image Gauge software package ScienceLab 2003 ( Fuji Corporation , Tokyo , Japan ) . In the EMSA 97 bp DNA fragments carrying Hsmar1 transposon ends were prepared by digesting pRC919 with XmaI and labeled at both ends using [α-32P]dCTP and the Klenow enzyme . Unless stated otherwise each 20 μl reaction contained 250 ng of non-specific plasmid DNA as a carrier , 2 nM labeled substrate and 6 nM transposase . Complexes were assembled for the indicated times in a buffer containing 20 mM HEPES pH7 . 5 , 100 mM NaCl , 2 mM DTT , 10% glycerol , 250 μg/ml BSA . Products were separated on 5 or 7% polyacrylamide Tris-acetate-EDTA gels . Transposition assays in HeLa cells were performed as described by ( Miskey et al . , 2007; Grabundzija et al . , 2010 ) using isogenic plasmids , except that the SB transposase gene and transposon ends were replaced by the respective sequences from PB and Hsmar1 . The kinetic models for interactions between transposase and DNA is presented in Figure 1B , I . A and B represent the identical inverted repeat sequences flanking the transposon , which are the specific binding sites of the transposase . The hyphens joining A and B represent the transposon DNA between the inverted repeats . T2 represents transposase , which is always dimeric in this model , T2 , nsb represents non-specifically bound transposase , T2 , free represents freely diffusing transposase , AT and BT represent transposase monomers bound to A or B , respectively , and AT2 and BT2 represent transposase dimers bound to A or B , respectively . The flux through the reaction was simulated in the Macintosh version of MATLAB R2010b using the ordinary differential equation solver ode15s . For the mariner model , we begin by considering an idealized situation in which we make two significant assumptions . Firstly , that since the A and B sites are identical , they have identical binding kinetics . Secondly , that the two DNA-binding sites on a single transposase dimer have identical behavior and do not interact with each other , meaning that there are no allosteric interactions between the transposase subunits . In this idealized situation the first and second transposon end to be bound by the transposase , which ever this may be , are bound with equal affinity and the rate depends only on the relative concentration of the reactants . Later we will also consider a model in which allosteric interactions between the transposase subunits causes a kinetic barrier to recruitment of the second transposon end . Because of these assumptions mentioned above , our model only needs to include the reaction rate constants shown in Figure 1I , which are k0 , k−0 , k1 , k−1 , k2 , k−2 , and k3 . Of these , k0 and k1 are second order rate constants ( units of M−1s−1 ) and the rest are first order rate constants ( units of s−1 ) . The experimental literature provides reasonable estimates for k−0/k0 , k1 and k−1 , given below , but not for k2 , which we go on to calculate . k3 is also relatively unknown , which we discuss later . For the synapsis-by-protein-dimerization model , as exemplified by Tn10 and Tn5 , we make the same assumptions as detailed above , except that the free transposase is monomeric . In our model , protein expression is instantaneous , producing a specified number of transposase dimers per transposon copy . Unless stated otherwise this is 500 dimers in the mariner model and 1000 monomers in the prokaryotic model . Instantaneous expression is necessary to simulate the reaction under the fast diffusion regime which prevails in vitro . Under the in vivo slow-diffusion regime , the reaction is so slow that the dynamics of protein expression and degradation are irrelevant . The Hsmar1 and Mos1 transposases are 37% identical and align throughout their entire length with only a single residue indel . The respective proteins will therefore have very similar three dimensional structures . A crystal structure for the Mos1 transposase revealed that it has an N-terminal DNA binding domain with two helix-turn-helix ( H-T-H ) motifs that are inserted into the major groove of the DNA close to the end of the transposon ( Richardson et al . , 2009 ) . We can therefore expect that the dissociation constant for transposon end binding ( Kd = k−1/k1 ) will be in the mid-picomolar concentration range . The lactose repressor ( LacI ) is one of the most thoroughly characterized H-T-H proteins , with many studies addressing the kinetic parameters of DNA binding . The LacI dimer is similar to a monomer of Hsmar1 transposase in that it binds to operator sequences by a pair of H-T-H motifs , which are located in adjacent major grooves . We therefore elected to use the kinetic parameters for LacI to model transposase binding to the transposon end . Estimated parameters for the behavior of the transposase are perhaps not ideal . However , our conclusions do not depend on the absolute values of the parameters because they have minimal effect beyond changing the scale on the axes of the graphs generated by the simulation . The relationships between the various parameters , which provide the key insights of the work , remain unchanged , as explained further below and in the main text . The values for the lactose repressor’s dissociation constant quoted in the textbooks and the literature range from about 10−9 to 10−14 M . However , it is now generally accepted that the Kd for LacI , and other dimeric H-T-H proteins , binding to their primary sites is in the mid to high picomolar concentration range . Many of the very low values quoted in the literature date from the late 1970s and early 1980s , and were determined at unphysiologically low salt concentrations . Another factor that complicates the lactose repressor literature is the complexity of the multiple binding sites encoded by the operator , which promote DNA looping and the mutual stabilization of dimers bound in cis to each other . Here we have used the kinetic parameters determined by Wells and Matthews and colleagues for LacI binding to a single site encoded on a 40 bp fragment of the lac operator under a physiological salt concentration ( Table 3 of Hsieh et al . ( 1987 ) ) . In the course of our experiments with Hsmar1 transposase we have never recorded significant inhibition of the reaction by the moderate amounts of non-specific DNA used as a target in many of our experiments . This suggests that the affinity of the transposase for non-specific DNA is relatively low . We have therefore elected to adopt the equilibrium constant ( Kd = k−0/k0 ) of LacI for non-specific DNA because it is probably fairly typical for trans-acting DNA binding proteins . We adopted the value provided by Von Hippel and colleagues because it was determined at a similar physiological salt concentration to the parameters noted above for site-specific binding ( Revzin and Von Hippel , 1977 ) . We used this equilibrium constant to assign arbitrary association and dissociation rate constants , k0 and k−0 . It is difficult to define these rate constants precisely because DNA binding proteins have two distinct modes of non-specific interaction . Firstly , binding from the bulk solution is probably rapid with a rate equal to a sizable fraction of a diffusion-limited reaction . This is followed by a one-dimensional diffusion phase , during which the protein visits non-specific sites that are considered to be 1 bp apart . The equilibrium constant , Kd , is thus the product of one- and three-dimensional events , each of which has its own kon and koff components . Fortunately , the absolute values of k−0 and k0 are not important in the simulation because they serve only to define the equilibrium constant for non-specific sites , which in turn specifies the concentration of free transposase available for transposon end binding in a genome of a given size . The validity of this assumption was confirmed by running the simulation using several different pairs of values for k0 and k−0: the output was unchanged provided that k−0/k0 was always equal to the experimentally determined Kd ( not shown ) . The values for the specific and non-specific parameters are tabulated in Table 1 , where they are also converted into the absolute units used in the simulation ( i . e . , numbers of molecules per cubic micrometer ) . The dissociation constant for non-specific DNA provided by Von Hippel and colleagues is expressed in units of per base pair . We have assumed a haploid genome size of 3 × 109 bp , which provides 6 × 109 non-specific binding sites per diploid genome . The volume of the nucleus is assumed to be 500 µm3 . 10 . 7554/eLife . 00668 . 014Table 1 . Kinetic parameters for DNA bindingDOI: http://dx . doi . org/10 . 7554/eLife . 00668 . 014ReactionKd ( M ) k0 , k1 ( M−1s−1 ) k0 , k1 ( µm3 s−1 ) k−0 , k−1 ( s−1 ) t1/2 ( s ) LacI specific binding3 . 1 × 10−113 . 8 × 108 0 . 6331 . 2 × 10−2 58 †LacI non-specific binding1 . 4 × 10−5 ‡9 . 9 × 106 §0 . 0166139 §5 × 10−3 †*From Table 3 of Hsieh et al . ( 1987 ) . †This is the experimental estimate for LacI bound to non-specific DNA dissociating to the bulk solution ( Elf et al . , 2007 ) . During this time it will have visited about 85 non-specific sites by one-dimensional diffusion . However , for the purposes of the simulation it is convenient to subsume the two phases into a single bulk behavior of the system ( see text for details ) . ‡From Revzin and Von Hippel ( 1977 ) . §Estimated from Kd using the relationship Kd = k−1/k1 or Kd = k−0/k0 . We calculated the first-order reaction rate constant k2 as follows . Throughout this discussion , T2 represents freely diffusing transposase , which is shown as T2 , free in the model diagram . Consider the two elementary reactions from our model that are shown at the bottom-left and the top-right regions of Figure 1I , respectively . Ignoring the transposon’s internal DNA sequences for now , these reactions are ( 1 ) T2 + B → BT2 and ( 2 ) AT2 + B → AT2B . Both reactions involve the binding of transposase to DNA site B , so they are chemically identical . This might suggest that they would have the same reaction rates . However , in actuality , their reaction rates differ because the rate of diffusion for the free transposase is different from that of the DNA-bound transposase . Expressing their reaction rate constants as k1 for reaction 1 ( as in the model diagram ) and k1′ for reaction 2 , the respective reaction rates are ( 1 ) d[BT2]dt=k1[T2][B] , ( 2 ) d[AT2B]dt=k1′[AT2][B] . The Collins and Kimball reaction rate theory ( Rice , 1985 ) enables us to expand these reaction rate constants into one contribution that arises from diffusion and a second that arises from the binding activation energy . According to this theory , the expanded reaction rate constants are ( 3 ) 1k1=1k1 , diff+1kact , ( 4 ) 1k1′=1k1 , diff′+1kact , where k1 , diff and k1 , diff′ are diffusion-limited reaction rate constants and kact is an activation-limited reaction rate constant . These equations include the same kact value because we assumed above that the two reactions are chemically identical; under this assumption , they should have the same binding activation energy and thus the same activation-limited reaction rate constants . The necessary diffusion-limited reaction rate constants can be calculated from equations that Berg derived for the diffusion-limited association rates of proteins to DNA ( Berg , 1984 ) . Berg’s equation 16 is for the association rate of a freely diffusing protein to a specific DNA site , while his equation 27 is for the association rate of a DNA-bound protein to another specific DNA site . In our notation , these equations are ( 5 ) k1 , diff=4πDpR+Dsa ( Ra ) 1/3 , ( 6 ) k1 , diff′=1 . 4Dsa ( Ra ) 1/3 . These equations were derived by treating DNA with the worm-like chain model , in which DNA is represented as a long thin semi-flexible filament that bends smoothly over the course of its length . This model was recently shown to be reasonably accurate for modeling DNA dynamics ( Petrov et al . , 2006 ) . The central model parameter is the filament persistence length , a , which is the characteristic length for filament bending . According to the model , DNA binding sites diffuse rapidly within their local regions , while also gradually diffusing away to more distant regions . The rate of this local ‘segmental diffusion’ is quantified with the diffusion coefficient Ds . Ds is the translational diffusion coefficient of a free DNA fragment with length a ( Berg , 1984 ) . Continuing with the above equations , R represents the distance over which a specific interaction occurs between transposase and its DNA binding sites . It arises from the assumption that the two reactants ( the transposase binding domain and a DNA binding site , in this case ) can be treated as hard spheres that react upon collision . Finally , Dp is the diffusion coefficient of free transposase relative to the center of mass of the DNA . Two final equations need to be introduced to enable us to calculate the rate of synapsis , k2 . The reaction rate equation for reaction 2 , given in equation 2 , can be rearranged by grouping k1′ and [B] , ( 7 ) d[AT2B]dt= ( k1′[B] ) [AT2]=k2[AT2] . The latter equality defines k2 , a pseudo-first order reaction rate constant , as k1′[B] . Here , [B] is the concentration of the B DNA binding site in the vicinity of the A binding site . We used the following empirical equation from Ringrose et al . ( 1999 ) to find [B]: ( 8 ) jM= ( 4a104b ) 3/2exp ( −460a26 . 25b2 ) ( 1 . 25⋅105a3 ) . jM is the local concentration of one DNA site in the vicinity of another in M , a is the DNA persistence length in nanometers , and b is the separation between the sites in base pairs . Table 2 lists the values that we used and derived to estimate k2 . In brief , we used equation 3 , including an experimental value for k1 and a calculated value for k1 , diff from equation 5 , to calculate kact . We then substituted kact into equation 4 , along with a calculated value for k1 , diff′ from equation 6 , to estimate k1′ . Finally , we used equations 7 and 8 to estimate k2 , the rate of synapsis , from k1′ . 10 . 7554/eLife . 00668 . 015Table 2 . Estimation of k2 for transposase-DNA interactions in vitroDOI: http://dx . doi . org/10 . 7554/eLife . 00668 . 015QuantitySymbolValueSourceReaction 1 rate constantk13 . 8 × 108 M−1s−1Table 3 of Hsieh et al . ( 1987 ) . DNA persistence lengtha50 nmWell established ( e . g . Petrov et al . , 2006 ) . Reaction radiusR2 . 5 nmThe transposase dimeric binding domain is 111 amino acids and weighs 26 kDa . From Note 3 of Andrews ( 2012 ) , its radius is about 1 . 9 nm . Add to this about 0 . 6 nm to account for the 0 . 34 nm length of 1 DNA basepair ( because transposase specificity is accurate to 1 bp ) , combined with the 1 nm radius of DNA . Segmental diffusion coefficientDs27 µm2s−1Figure 3 of Petrov et al . ( 2006 ) for 147 bp , which is 1 DNA persistence length . Transposase diffusion coefficientDp61 µm2s−1Note 3 of Andrews ( 2012 ) , using a molecular weight of 80 . 7 kDa for transposase . Assumes DNA center of mass is effectively stationary . Diffusion-limited part of k1k1 , diff1 . 5 × 109 M−1s−1Equation 5 . Activation-limited reaction ratekact5 . 2 × 108 M−1s−1Equation 3 . Diffusion-limited part of k1′k1 , diff′4 . 2 × 108 M−1s−1Equation 6 . Reaction 2 rate constantk1′2 . 3 × 108 M−1s−1Equation 4 . Effective local DNA site concentration[B]5 . 5 × 10−8 MEquation 8 , using a 1287 bp transposon length . Transposase-transposon end dissociation rate constantk−15 . 8 × 10−4 s−1Estimated from Figure 3C . Reaction 2 , pseudo-first order rate constant for synapsisk212 . 7 s−1Equation 7 . Reaction 2 , pseudo-first order rate constant for synapsis with an open circular substrate in vitro . k29 . 6 × 10−5 s−1Estimated using the equation t1/2 = ln2/k2 . t1/2 is the time taken to consume one quarter of the substrate in Figure 3D . See Figure 2—figure supplement 1 for an explanation of why the time to consume one quarter of the substrate is used , rather than the time to consume half of the substrate . These theoretical considerations allow us to develop a model of an idealized transposition reaction , in which the monomers within a transposase dimer bind the first and second transposon ends with equal affinity . In this idealized situation , synapsis of the transposon ends is simply a product of the sequential binding of the transposase dimer to sites at opposite ends of the element . We then go on to consider a more realistic model in which allosteric interactions between the subunits reduces the affinity of the developing transpososome for the second transposon end . The magnitude of this effect is provided by experimental estimates of the rate of synapsis ( Figure 3D , Table 2; Claeys Bouuaert and Chalmers , 2010; Claeys Bouuaert et al . , 2011 ) . In Table 2 , the reaction radius , R , is a fairly rough estimate . However , this is not a major concern because reaction rates are relatively insensitive to the reaction radius ( Berg , 1984 ) ( they scale as R1/3 ) . Other values are likely to be more accurate . A result of the calculations shown there is that k1′ is only about a factor of 1 . 6 slower than k1 . This indicates that the slower diffusion of DNA-bound transposase than of free transposase has only modest impact for in vitro experiments . For Tn10/5 there is no experimental data regarding the affinity of the transposon-end-bound monomers when they collide with each other and achieve synapsis ( see kinetic model in Figure 1B ) . For the purposes of the simulation we assigned an association rate constant equal to that for the mariner transposase binding to a transposon end . In the respective prokaryotic and eukaryotic models , the rate of synapsis is therefore determined by the same association rate constant and the rate of diffusion , which is identical in both systems . The outputs of the respective simulations are therefore directly comparable and reveal differences in the underlying kinetic models . Thus far we have considered reactions in dilute solution . To provide a more realistic description of the reaction in vivo we next had to account for macromolecular crowding , which reduces the rate of diffusion . We therefore computed the same numbers that are shown in Table 2 , but for the in vivo situation , which are shown in Table 3 . The only parameters sensitive to the high viscosity in vivo are the association rate constants , which includes the rate of synapsis . Our calculations show that the association rate constant , k1 , suffers a modest twofold penalty in vivo ( Table 3 ) . This relieves OPI and increases the rate of transposition twofold ( not shown ) . However , this is precisely offset by the penalty on non-specific binding , k0 , which achieves the opposite . The results are very different for the rate of synapsis , which suffers an in vivo diffusion penalty of 2 . 8 × 104-fold ( Tables 2 and 3 ) . However , many DNA interactions are known to occur over much longer-ranges than those between transposon ends . It is therefore possible that other factors in vivo mitigate the severity of the diffusion penalty . For example , there is evidence that the factor jM is doubled by the chromatinization of DNA ( Ringrose et al . , 1999 ) . The in vivo values in Table 3 should therefore be seen as initial estimates rather than an exhaustive treatment of the topic . Nevertheless , our calculated rate of synapsis is very close to the rates of lox/cre recombination measured in vivo with similarly separated recombination sites ( Ringrose et al . , 1999 ) . 10 . 7554/eLife . 00668 . 016Table 3 . Estimation of in vivo transposase-DNA interactionsDOI: http://dx . doi . org/10 . 7554/eLife . 00668 . 016QuantitySymbolValueSourceDNA persistence lengtha50 nmSee Table 2 . Reaction radiusR2 . 5 nmSee Table 2 . Segmental diffusion coefficientDs5 × 10−4 µm2s−1From Figure 2 caption of Marshall et al . ( 1997 ) . Transposase diffusion coefficientDp15 µm2s−1¼ of Table 2 value , based on Note 3 of Andrews , ( 2012 ) . Diffusion-limited part of k1k1 , diff2 . 9 × 108 M−1s−1Equation 5 . Activation-limited reaction ratekact5 . 2 × 108 M−1s−1From Table 2 . Diffusion-limited part of k1′k1 , diff′7 . 8 × 103 M−1s−1Equation 6 . Reaction 1 rate constantk11 . 9 × 108 M−1s−1Equation 3 . Reaction 2 rate constantk1′7 . 8 × 103 M−1s−1Equation 4 . Effective local DNA site concentration[B]5 . 5 × 10−8 MSee Table 2 . Reaction 2 pseudo-first order rate constantk24 . 3 × 10−4 s−1Equation 7 . In the simplest model for synapsis the rate of the reverse reaction , k−2 , would be equal to the rate of transposon end unbinding , k−1 ( see Figure 1I ) . However , in practice we consider synapsis to be essentially irreversible for the following reasons . Transposition reaction mixtures in which the catalytic Mg2+ ions are replaced by Ca2+ support transpososome assembly , but none of the chemical steps of the reaction . After the addition of the catalytic metal ion , the first nick at the transposon end is detected immediately ( Claeys Bouuaert and Chalmers , 2010 ) . This shows that nicking is very much faster than synapsis or synapsis unbinding . Synapsis in the presence of the catalytic metal ion is therefore essentially irreversible because of the loss of enthalpy associated with hydrolysis of the phosphodiester bond . In principle , it is possible that there exists some form of unstable synaptic complex that matures into the stable form that immediately precedes the first chemical reaction . If such a complex existed it would be irrelevant to our experiments , in which the rate of synapsis is estimated from the rate of the first nick . When we consider the rate of synapsis we are therefore restricting ourselves to ‘productive synapses’ , which yield the first nick . For the purposes of the simulation we have set the rate constant for synapsis unbinding ( k−2 ) at 10−10 s−1 . Synapsis in the model is therefore essentially irreversible . When transposase dimer binds to the first transposon end it undergoes a conformational change that lowers the affinity of the unoccupied DNA binding domain for the second transposon end compared to the first ( for details see main text and Claeys Bouuaert et al . , 2011 ) . This means that the actual value of k2 will be much lower than that calculated above for the idealized reaction , in which synapsis is the product of sequential , chemically identical , collisions between DNA binding domains and transposon ends . To account for the allosteric interactions between the transposase subunits we therefore require experimental estimates of k2 and k−1 . Estimates for k−1 and k2 are provided by the kinetics of the in vitro reaction . In Figure 3C we see that about 50% of the transposon end present in SEC2 is released after 20 min incubation in the presence of cold competitor DNA . This corresponds to a dissociation rate constant of 5 . 8 × 10−4 s−1 . We have previously shown that first strand nicking depends on , and is much faster than , synapsis ( Claeys Bouuaert et al . , 2011 and ‘Discussion’ above ) . Consumption of the supercoiled substrate therefore provides an estimate of the rate of synapsis . Under optimal reaction conditions , with about one dimer of active transposase per transposon , only one half of the substrate is initially occupied by a single dimer and is therefore able to react . As explained above , the random association of transposase dimers and transposon ends at the start of the reaction means that a quarter would be occupied by two dimers and would suffer OPI and the other quarter would be completely unoccupied . The time required to consume one quarter of the substrate therefore approximates the half-time for synapsis . Since eukaryotic DNA has very little free supercoiling , the most relevant rate of synapsis is perhaps provided by an open circular substrate . In Figure 3D , the time taken to consume one quarter of the open circular substrate is about 2 hr , which corresponds to a pseudo-first order rate constant of 9 . 6 × 10−5 s−1 . Note that the reason we are able to estimate the rate of synapsis is owing to the optimal reaction condition in vitro where k2 is largely independent of k0 , k−0 , k1 and k−1 . This is because there is very little non-specific DNA present and the transposase concentration is such that most dimers are engaged in productive interactions with transposons . The rate limiting steps of the reaction with various substrates has been determined previously ( Claeys Bouuaert and Chalmers , 2010; Claeys Bouuaert et al . , 2011 ) . With relaxed plasmid and short linear substrates , synapsis is the rate limiting step . Supercoiling in the substrate accelerates synapsis because the transposon ends have a high relative concentration in the plectosome and a favorable angular distribution . In an in vitro reaction , where the plasmid has more than twice its natural level of free supercoiling , the rate of synapsis is faster than cleavage of the second strand at the transposon end . Note that second strand cleavage at the second transposon end releases the transposon from the donor site ( illustrated in Figure 2A ) . In a kinetic analysis of staged in vitro reactions with a supercoiled substrate , a small amount of excised transposon is observed at early time points [this is below the region of the gels shown in Figure 2B , but can be seen in Figure 5B of Claeys Bouuaert et al . ( 2011 ) ] . The rate of integration must therefore be faster than the rate of second strand cleavage . The half life of the excised transposon is approximated by the time required to convert a quarter of the supercoiled substrate into product . From Figure 2B and other experiments this appears to be about 8 min which corresponds to a rate constant for integration of 1 . 4 × 10−3 s−1 . Although this value for k3 is appropriate in the simulations using in vitro parameters , it would be unrealistically short in an in vivo situation where the integration complex must be disassembled and the empty donor site restored . However , our reference in vitro simulation ( Figure 3E ) is relatively insensitive to the value of k3 . This is because the extent to which the Hsmar1 specific factors , namely the slow dissociation rate of SEC2 and the slow recruitment of the second end , slow synapsis . Furthermore , once a steady-state rate is established the length of the maturation process does not matter because the rates of pathway entry and exit are equal . Thus , if k3 is reduced by three orders of magnitude to 1 . 4 × 10−6 s−1 ( t1/2 = 5 . 7 days ) , the final rate of transposon amplification does not change significantly . However , it does take slightly longer to be achieved because of the time required for pathway entry and exit rates to balance ( not shown ) . Note that we assume that every transposon that initiates catalysis goes on to complete excision and integration . This is supported by the in vitro reaction systems for Hsmar1 and Tn10 ( Chalmers and Kleckner , 1994 , 1996; Claeys Bouuaert and Chalmers , 2010; Claeys Bouuaert et al . , 2011 ) . The efficiency of the reaction in vivo is unlikely to be so high and this is dealt with in Figure 3G and the section of text entitled ‘The efficiency of transposition in vivo . ’ Many transposons , and mariner in particular , have a wide host range . The promoters driving expression of transposase genes are therefore assumed to rely on general transcription factors . However , there is little published data pertaining to the possible strength of the promoters . Expression of the Hsmar1 promoter has been detected using a luciferase reporter system , but quantification is lacking for the contribution of a single copy of the element ( Miskey et al . , 2007 ) . A recent proteomic study quantified the abundance of 5000 vertebrate proteins in mouse cells ( Schwanhausser et al . , 2011 ) . The copy number per cell for most proteins ranged from about 100 to 1 million: The median was 16 , 000 and the mode was about 5000 . In the simulation we set the transposase expression at 500 dimers per transposon . Although this is at the low end of the range for vertebrate proteins , it is not unusually low and allows for up to 1000 copies of the transposon before protein levels would reach the high end of the range . However , we note , once again , that our conclusions do not depend on the absolute values of the parameters because they have minimal effect beyond changing the scale on axes of the graphs generated by the simulation . The relationships between the various parameters , which provide the key insights of the work , remain unchanged .	Transposons are regions of mobile DNA that can jump from one location in the genome to another . This represents a genetic burden to the host because there is always the risk that the transposon will inactivate a cellular gene . However , a greater problem is that transposition is accompanied by an increase in the number of copies of the transposon . Since each new copy will be a source of further new copies , amplification of transposons is necessarily exponential . The fact that eukaryotic cells are able to tolerate DNA transposons suggests the existence of regulatory mechanisms to defuse the inevitable genomic melt-down . Host-mediated epigenetic modifications and RNA interference will provide some level of protection . However , they are by no means completely effective and a well-adapted genomic parasite , such as a transposon , might be expected to have its own mechanism of regulation . Now , Claeys Bouuaert , Lipkow and colleagues have used a computer model in combination with in vivo and in vitro experiments to search for this mechanism . Their experiments reveal how a DNA transposon is down-regulated by its own transposase . The transposase is the enzyme that catalyzes the ‘jump’ or transposition . It binds to specific sites at either end of the transposon and brings these together to make up a nucleoprotein complex called the transpososome . It is within this complex that the chemical steps of the reaction take place . When the number of transposons increases , so does the concentration of transposase . Claeys Bouuaert et al . show that the binding sites become saturated at a relatively low transposase concentration and that negative regulation arises from the resulting competition . Thus , the rate of transposition decreases as the number of transposons increases . They further use the computer model to explore how the amplification of the transposon is affected by transposon-specific and cellular-specific factors . Claeys Bouuaert , Lipkow and colleagues based their study predominantly on a resurrected copy of the Hsmar1 transposon , which was active in the human genome 50 million years ago . However , they also tested two distantly related eukaryotic transposons and observed that their behavior was similar , which suggests that this could be a general mechanism that controls the activity of jumping genes . They also note that their competition mechanism is conceptually similar to the immunological ‘prozone effect’ . This is a recurrent theme in protein chemistry and demonstrates once again that less is in fact sometimes more .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "evolutionary", "biology", "biochemistry", "and", "chemical", "biology" ]	2013	The autoregulation of a eukaryotic DNA transposon
Autophagy is a major pathway for the clearance of harmful material from the cytoplasm . During autophagy , cytoplasmic material is delivered into the lysosomal system by organelles called autophagosomes . Autophagosomes form in a de novo manner and , in the course of their formation , isolate cargo material from the rest of the cytoplasm . Cargo specificity is conferred by autophagic cargo receptors that selectively link the cargo to the autophagosomal membrane decorated with ATG8 family proteins such as LC3B . Here we show that the human cargo receptor p62/SQSTM-1 employs oligomerization to stabilize its interaction with LC3B and linear ubiquitin when they are clustered on surfaces . Thus , oligomerization enables p62 to simultaneously select for the isolation membrane and the ubiquitinated cargo . We further show in a fully reconstituted system that the interaction of p62 with ubiquitin and LC3B is sufficient to bend the membrane around the cargo . Cellular homeostasis and quality control require degradation of potentially harmful cytoplasmic material . The lysosomal system mediates degradation of large and bulky substances that cannot be degraded by other means , for example , the proteasome . A major pathway for the degradation of cytoplasmic material is macroautophagy ( hereafter autophagy ) ( De Duve and Wattiaux , 1966 ) . During autophagy , double membrane-bound organelles called autophagosomes are formed that , upon fusion with the lysosomal system , deliver cytoplasmic cargo material for degradation ( Kraft and Martens , 2012 ) . Autophagosomes form in a de novo manner . Initially , small-membrane structures called isolation membranes or phagophores are observed , which gradually enclose cargo material as they grow . Upon closure of the isolation membranes , autophagosomes are formed within which the cargo is isolated from the rest of the cytoplasm . Subsequently , the autophagosomes fuse with the endolysosomal system where the inner membrane and the cargo are eventually degraded ( Kraft and Martens , 2012 ) . It has become clear that autophagy can be highly selective with regard to the cargo that is enclosed and degraded ( Rogov et al . , 2014 ) . Among the many cargos are aggregated proteins ( Bjørkøy et al . , 2005; Kirkin et al . , 2009a; Komatsu et al . , 2007; Pankiv et al . , 2007; Szeto et al . , 2006 ) , damaged mitochondria ( Geisler et al . , 2010; Kanki et al . , 2009; Narendra et al . , 2008; Novak et al . , 2010; Okamoto et al . , 2009 ) , intracellular pathogens ( Gutierrez et al . , 2004; Nakagawa et al . , 2004; Thurston et al . , 2009; Yoshikawa et al . , 2009; Zheng et al . , 2009 ) , surplus peroxisomes ( Farré et al . , 2008; Hutchins et al . , 1999; Iwata et al . , 2006 ) , and ferritin ( Dowdle et al . , 2014; Kishi-Itakura et al . , 2014; Mancias et al . , 2014 ) . Consequently , dysfunctional autophagy results in several pathological conditions such as neurodegeneration , cancer , and uncontrolled infection ( Levine and Kroemer , 2008; Mizushima and Komatsu , 2011; Levine et al . , 2008 ) . The selectivity of autophagic processes is conferred by autophagic cargo receptors that bind the cargo and link it to the isolation membrane ( Johansen and Lamark , 2011 ) . The isolation membrane is specifically recognized by the cargo receptors due to its modification with proteins of the ATG8 family ( Kabeya et al . , 2000 ) . Yeast Atg8 and its homologues are ubiquitin-like proteins that become conjugated to the headgroup of the membrane lipid phosphatidylethanolamine ( Ichimura et al . , 2000 ) . This unusual modification renders the soluble ATG8 proteins membrane-bound and serves as an identifier for the isolation membrane ( Ichimura et al . , 2000 ) . The first autophagic cargo receptor identified was the Saccharomyces cerevisiae Atg19 protein ( Leber et al . , 2001; Scott et al . , 2001 ) . Atg19 acts during the transport of the oligomeric prApe1 peptidase and other cargos into the vacuole ( Hutchins and Klionsky , 2001; Scott et al . , 2001; Suzuki et al . , 2011; Yuga et al . , 2011 ) . Within the vacuole , prApe1 becomes activated and fulfills its enzymatic function . Under basal , nutrient-rich conditions , the prApe1 oligomers are constitutively transported into the vacuole by the cytoplasm-to-vacuole transport ( Cvt ) pathway ( Klionsky et al . , 1992 ) , in which small double membrane-bound vesicles , called Cvt vesicles , tightly enclose the prApe1 cargo ( Baba et al . , 1997 ) . The formation of these Cvt vesicles depends on the core autophagic machinery ( Harding et al . , 1995 ) and it is mechanistically analogous to the formation of selective autophagosomes in complex eukaryotes , including mammals ( Lynch-Day and Klionsky , 2010 ) . The Atg19 receptor binds directly and strongly to the prApe1 cargo ( Morales Quinones et al . , 2012; Sawa-Makarska et al . , 2014; Scott et al . , 2001; Shintani et al . , 2002 ) . In addition , it contains multiple Atg8 binding sites ( Noda et al . , 2008; Sawa-Makarska et al . , 2014; Shintani et al . , 2002 ) . These two properties enable Atg19 to bend the membrane tightly around the cargo and thereby to exclude non-cargo material from the Cvt vesicles ( Baba et al . , 1997; Sawa-Makarska et al . , 2014 ) . Mammals have multiple cargo receptors that mediate the autophagic degradation of cytoplasmic material ( Johansen and Lamark , 2011 ) . While some mammalian cargo receptors such as NCOA4 directly recognize their cargo ( Dowdle et al . , 2014; Mancias et al . , 2014 ) , many mammalian cargo receptors including p62/SQSTM-1 , NBR1 , Optineurin , NDP52 , and Tollip recognize the cargo material due to its modification with ubiquitin ( Bjørkøy et al . , 2005; Kirkin et al . , 2009; Kirkin et al . , 2009; Lu et al . , 2014; Rogov et al . , 2014; Thurston et al . , 2009; Wild et al . , 2011 ) . p62 is a multidomain protein and contains , among other domains , an N-terminal PB1 domain , a LIR motif mediating the interaction with ATG8 family proteins and a C-terminal UBA domain that binds ubiquitin ( Figure 2 ) ( Johansen and Lamark , 2011; Vadlamudi et al . , 1996 ) . The affinity of the UBA domain for ubiquitin is very low ( Long et al . , 2008; Long et al . , 2010 ) but can be increased by phosphorylation on serine 403 ( Matsumoto et al . , 2011 ) . The N-terminal PB1 domain mediates interaction with several other proteins as well as homo-oligomerization ( Lamark et al . , 2003 ) . Recently , it was shown by cryo-electron microscopy that in vitro p62 forms large helical structures in a PB1-dependent manner ( Ciuffa et al . , 2015 ) . 10 . 7554/eLife . 08941 . 003Figure 1 . p62 has only one detectable LC3B/GABARAP-interaction motif . ( A ) Schematic representation of the p62 domain architecture . ( B ) Scheme showing the experimental set-up of the assay . GFP-LC3B-6xHis or GFP-6xHis were tethered to giant unilamellar vesicles ( GUVs ) via nickel-lipids incorporated into the membranes . Recombinant wild-type or LIR mutant mCherry-p62 were added and their membrane recruitment was determined . ( C ) Representative images of GUVs incubated with GFP-LC3B-6xHis and mCherry-p62 wild-type or LIR mutant . The mCherry signal is shown in false color ( ImageJ: fire ) . Scale bars , 5 µm . ( D ) Quantification of mCherry-p62 wild-type or LIR mutant membrane recruitment . Averages and SD from three independent experiments are shown . The p-value was determined using a two-tailed unpaired Student’s t-test . ( E ) Quantification of mCherry-p62 wild-type ( black bars ) or LIR mutant ( gray bars ) recruitment to GFP-LC3B-6xHis or GFP-GABARAP-6xHis coated GUVs . Data are normalized to the wild-type p62 binding intensity . The error bars are derived from three independent experiments . The p-value was determined using a two-tailed unpaired Student’s t-test . ( F ) Anti-GFP and anti-p62 western blots of input ( 8% ) and bead ( 50% ) fractions of a GFP-TRAP pull-down of HeLa cell lysates co-expressing GFP-LC3B , GFP-GABARAP , or GFP ( control ) and wild-type or LIR mutant mCherry-p62 . The endogenous p62 was silenced by siRNA . The experiment was conducted twice . ( D ) Total GUVs counted per condition: GFP-LC3B-6xHis + mCherry-p62 wild-type = 163; GFP-LC3B-6xHis + mCherry-p62 LIR mutant = 152; GFP-6xHis + mCherry-p62 wild type = 254 . ( E ) Total GUVs counted per condition: GFP-LC3B-6xHis + mCherry-p62 wild type = 636; GFP-LC3B-6xHis + mCherry-p62 LIR mutant = 642; GFP-GABARAP-6xHis + mCherry-p62 wild type = 336; GFP-GABARAP-6xHis + mCherry-p62 LIR mutant = 300 . DOI: http://dx . doi . org/10 . 7554/eLife . 08941 . 003 Mutations in the PB1 domain that interfere with its ability to oligomerize inhibit the recruitment of p62 to the autophagosome formation site ( Itakura and Mizushima , 2011 ) . Moreover , deletion of the PB1 domain or oligomerization-inhibiting mutations decrease the interaction with both LC3B and ubiquitin in pull-down assays , suggesting that oligomerization may increase the interaction with these binding partners . Here we show in a variety of in vitro and in vivo systems that the oligomerization of p62 generates high-avidity interactions with ubiquitin and LC3B-coated surfaces , which allows p62 to select for cargo material and the isolation membrane . In particular , we show that oligomerization does not increase the affinity of each individual binding site but reduces the off-rate of the oligomeric unit from ubiquitin and LC3B-coated surfaces , respectively . We further show in a reconstituted system that the concurrent interaction of p62 with ubiquitin and LC3B is sufficient to drive the close apposition of the membrane and the cargo . The S . cerevisiae Atg19 cargo receptor contains multiple low-affinity Atg8 binding sites that enable it to selectively and tightly bind to membrane-localized Atg8 ( Sawa-Makarska et al . , 2014 ) . We asked whether this feature is conserved and turned our attention to p62 , which is a major cargo receptor in mammals , including humans . Only a single LIR motif has been identified in p62 ( Ichimura et al . , 2008; Pankiv et al . , 2007 ) , but there was the possibility that low-affinity-binding sites for ATG8 family proteins such as LC3B and GABARAP were not detected in classical pull-down assays since they fail to detect interactions with high off-rates . We , therefore , used a more sensitive assay to find potential p62–LC3B interaction sites that are independent of the known LIR motif . To this end , we attached GFP-labeled LC3B or GABARAP to the membrane of giant unilamellar vesicles ( GUVs ) . Recombinant mCherry-p62 was added to the GFP-LC3B and GFP-GABARAP-coated GUVs and the recruitment of mCherry-p62 was followed by spinning disk microscopy ( Figure 2 ) . mCherry-p62 was robustly recruited to GFP-LC3B and GFP-GABARAP but not to GFP-coated GUVs . Upon simultaneous mutation of D335 , D336 , D337 , and W338 to A in the LIR motif of p62 ( Ichimura et al . , 2008; Pankiv et al . , 2007 ) , the recruitment of the protein to the GFP-LC3B and GFP-GABARAP-coated GUVs was completely abolished ( Figure 2 ) , strongly suggesting that p62 has only one functional LC3B/GABARAP interaction site . We will refer to this mutant as the LIR mutant . We corroborated these results in GFP-TRAP experiments using HeLa cell lysates ( Figure 2 ) , where the interaction of p62 with LC3B and GABARAP totally depended on its LIR motif . The N-terminal PB1 domain of p62 mediates oligomerization ( Ciuffa et al . , 2015; Lamark et al . , 2003 ) . Within the p62 oligomers , LIR motifs are clustered , similar to the occurrence of multiple Atg8 binding sites in the Atg19 cargo receptor ( Sawa-Makarska et al . , 2014 ) . Indeed , the PB1 domain was shown to enhance LC3B binding in pull-down experiments ( Bjørkøy et al . , 2005 ) . To directly test whether the strength of the p62–LC3B interaction correlates with the ability of p62 to oligomerize , we recombinantly expressed and purified several oligomerization mutants of p62 . The attachment of mCherry to the N-terminus of p62 considerably increased the yield of soluble protein . In order to determine the oligomerization state of our mCherry-p62 variants , we conducted size exclusion chromatography ( SEC ) runs coupled to static light scattering ( SLS ) ( Figure 2A and Figure 2—figure supplement 1 ) . This allowed us to determine the molecular mass of the p62 variants independently of their shape . The wild-type and LIR mutant proteins eluted in the exclusion volumes ( V0 ) of the column . SLS showed that the protein in the V0 was composed of oligomeric particles of on average 24 molecules . Deletion of the PB1 domain resulted in a complete shift of the protein from the V0 to lower molecular weight fractions . Interestingly , SLS showed that p62 delta PB1 is a trimer . The structural basis for the trimeric form is currently unknown . 10 . 7554/eLife . 08941 . 004Figure 2 . Oligomerization of p62 stabilizes binding to LC3B-coated surfaces . ( A ) Size exclusion chromatography ( SEC ) and static light scattering ( SLS ) analysis of recombinant wild-type mCherry-p62 , the LIR mutant and the oligomerization mutants ( K7A/D69A , delta PB1 , and NBR1-p62 chimera ) . The left Y-axis indicates the molecular weight of the protein as determined by SLS . The average sizes of the indicated peak areas obtained by SLS are shown in the table . See Figure 2 for gel . ( B ) Coomassie-stained gel showing a p62 sedimentation assay of recombinant mCherry-p62 wild-type , delta PB1 , and K7A/D69A mutants . For each p62 variant input , supernatant and pellet fractions were loaded . ( C ) Quantification of the p62 sedimentation assay shown in ( B ) . Amounts of p62 in the supernatant ( blue ) and pellet ( red ) are represented as fractions of the input . ( D ) Anti-GFP and anti-p62 western blot of input ( 8% ) and bead ( 50% ) fractions of a GFP-TRAP affinity purification of HeLa cell lysates co-expressing GFP-LC3B or GFP ( control ) and the mCherry-p62 variants . The endogenous p62 was silenced by siRNA treatment ( Figure 2—figure supplement 2 ) . A representative blot of four independent replicates is shown . ( E ) Anti-GST and anti-p62 western blot analysis of input ( 8% ) and bead ( 16% ) fractions of a pull-down experiment using GST-LC3B or GST ( control ) as bait and purified mCherry-p62 variants as prey . A representative blot of three independent replicates is shown . Asterisks denote dimeric GST . ( F ) Quantification of steady-state binding intensities of increasing concentration of wild-type , delta PB1 , or the LIR mutant mCherry-p62 on GST-LC3B-coated beads . The average fluorescence intensity on the beads is plotted against the p62 concentration . Averages and SD of three independent experiments are shown . ( G ) Representative images of the experiment shown in ( FSchematic represen ) . The mCherry signal is shown in false color ( ImageJ: fire ) . ( F ) Total beads quantified: wild-type 0 . 2 µM = 187 - 0 . 5 µM = 198 - 1 µM = 180 - 2 µM = 175 - 5 µM = 73; p62 delta PB1 0 . 2 µM = 133 - 0 . 5 µM = 163 - 1 µM = 179 - 2 µM = 176 - 5 µM = 58; p62 LIR mutant 0 . 2 µM = 74 – 0 . 5 µM = 84 – 1 µM = 75 – 2 µM = 85 – 5 µM = 75 . DOI: http://dx . doi . org/10 . 7554/eLife . 08941 . 00410 . 7554/eLife . 08941 . 005Figure 2—figure supplement 1 . ( A ) Coomassie-stained gel showing the peak fractions of wild-type mCherry-p62 and the K7A/D69A mutant after the size exclusion chromatography ( SEC ) /static light scattering ( SLS ) runs . Corresponding peaks in Figure 2A are indicated . ( B ) Analytical SEC profiles of indicated mCherry-p62 variants run on a Superose 6 10/300 ( GE Healthcare ) column . Arrowheads indicate the elution volume of standard globular protein of indicated molar masses used for calibration . V0 was calibrated with 2MDa blue dextran . DOI: http://dx . doi . org/10 . 7554/eLife . 08941 . 00510 . 7554/eLife . 08941 . 006Figure 2—figure supplement 2 . Western blot of samples shown in Figure 2D showing efficient siRNA-mediated silencing of endogenous p62 in mCherry-p62 co-transfected cells . DOI: http://dx . doi . org/10 . 7554/eLife . 08941 . 00610 . 7554/eLife . 08941 . 007Figure 2—figure supplement 3 . Relative fluorescence intensity plot of data shown in Figure 2F . The data were normalized by setting the absolute binding of p62 at 5 µM to 100% in each sample . Absolute intensities of negative controls ( beads coated with GST only ) were set to 0% in each sample . Data points of wild-type mCherry-p62 and delta PB1 were fitted to a mono-exponential curve ( R2 = 0 . 9976 ) and a line ( R2 = 0 . 9942 ) , respectively . DOI: http://dx . doi . org/10 . 7554/eLife . 08941 . 007 Similarly , when we exchanged the PB1 domain of p62 for the non-oligomerizing PB1 domain of NBR1 ( Lamark et al . , 2003 ) the protein became trimeric . We will refer to this mutant as the NBR1-p62 chimera . Interestingly , introduction of the oligomerization-interfering K7A/D69A double mutation ( Lamark et al . , 2003 ) into the PB1 domain of p62 resulted in an intermediate behavior between the two extremes with a small fraction of the protein eluting in the V0 and another peak representing the trimeric species ( Figure 2A and Figure 2—figure supplement 1 ) . To confirm this result , we tested purified wild-type mCherry-p62 , p62 delta PB1 , and the K7A/D69A mutant in a p62 sedimentation assay ( Ciuffa et al . , 2015 ) . Consistent with the SLS results , the wild-type protein nearly completely sedimented into the pellet , while the delta PB1 mutant remained in the supernatant . Interestingly , the K7A/D69A mutant partitioned into both fractions ( Figure 2B , C ) . To analyze whether the ability of p62 to oligomerize correlates with the strength of its interaction with LC3B , we performed GFP-TRAP experiments using cell lysates of HeLa cells co-transfected with siRNA-resistant versions of the mCherry-p62 variants and GFP-LC3B ( Figure 2D ) . The endogenous p62 was silenced by siRNA ( Figure 2—figure supplement 2 ) . Indeed , there was a strong correlation between the ability of p62 to oligomerize and its presence in the bead fraction ( Figure 2D ) . While the wild-type protein showed the most robust interaction with LC3B , the interaction of the K7A/D69A double mutant ( Lamark et al . , 2003 ) was reduced but still readily detectable . The interaction of delta PB1 p62 and the NBR1-p62 chimera became detectable only after long exposure of the blots . Next , we tested the different purified recombinant mCherry-p62 variants in pull-down assays using GST-LC3B as bait ( Figure 2E ) . Similarly to what we observed in the GFP-TRAP experiments , the ability of the different p62 variants to co-pellet with GST-LC3B correlated strongly with their oligomeric state , suggesting that oligomerization of p62 directly affects its binding to LC3B . Pull-down assays favor interactions with off-rates low enough to resist washing . Therefore , one possible interpretation of these results is that the oligomeric wild-type p62 has a lower off-rate from LC3B clustered on a surface than the non-oligomerizing mutants . However , it is also possible that in p62 oligomers some monomers are simply piggybacked without actively contributing to the interaction with LC3B . Finally , a third possibility would be that the PB1 domain allosterically enhances binding to LC3B . To discriminate between these possibilities , we first measured the steady-state binding of wild-type and delta PB1 mCherry-p62 to GST-LC3B-coated beads . To this end , we recruited wild-type and delta PB1 mCherry-p62 at different concentrations to glutathione beads coated with GST-LC3B and imaged them by spinning disk microscopy at equilibrium . The mCherry-p62 LIR mutant was used as a negative control ( Figure 2F , G ) . The fluorescence signal on the beads correlated well with the protein concentration for both wild-type and delta PB1 mCherry-p62 ( Figure 2G ) . However , the titration curve of the wild-type protein showed a steeper slope compared to the delta PB1 protein and approached a plateau above a concentration of 5 µM ( Figure 2F and Figure 2—figure supplement 3 ) . We could , therefore , estimate a half-maximal binding constant of 1 . 5 µM for wild-type mCherry-p62 . It was impossible to estimate the half maximal binding constant for the delta PB1 mutant since at higher protein concentrations the fluorescence of the unbound protein rendered an accurate quantification of the bead-associated signal impossible . The different shapes of the titration curves ( Figure 2—figure supplement 3 ) suggested that the presence of the PB1 domain does not merely confer piggybacking of p62 molecules , but actively increases the overall affinity of p62 toward LC3B . This could either be due to an oligomerization-dependent increase in avidity or an allosteric effect on the intrinsic affinity of the LIR motif for LC3B . To discriminate between these possibilities , we first performed fluorescence recovery after photo-bleaching ( FRAP ) experiments to determine the exchange rate of mCherry-p62 on GST-LC3B coated beads ( Figure 3A , B ) . In fact , if the PB1 domain increases the avidity of p62 towards surface-localized LC3B via oligomerization , this would result in a lower off-rate of the wild-type protein compared to the non-oligomerizing delta PB1 p62 . This would in turn translate into a slower fluorescence recovery for the wild-type protein . Indeed , while p62 delta PB1 readily recovered 5 min after bleaching , wild-type p62 showed almost no recovery within the same time frame ( Figure 3A ) . 10 . 7554/eLife . 08941 . 008Figure 3 . Oligomerization of p62 renders binding to concentrated LC3B irreversible . ( A ) Fluorescence recovery after photo-bleaching ( FRAP ) curve of the indicated mCherry-p62 variants on GST-LC3B-coated beads . Averages and SD of at least three independent curves are shown . ( B ) Representative pictures for the data shown in ( A ) . Scale bar 5 µm ( C , D ) . Quantification of steady-state binding of indicated mCherry-p62 variants to RFP-TRAP beads and of subsequent GFP-LC3B recruitment to these beads . Absolute fluorescence intensities are shown in ( C ) . A plot of GFP/mCherry ratio is shown in ( D ) . Averages and SD of three independent replicates are shown . Indicated p-values were calculated with a two-tailed unpaired Student’s t-test . ( E ) Quantification of decay of GFP-LC3B fluorescence from RFP-TRAP beads coated with indicated mCherry-p62 variants . Averages and SD of two independent replicates are shown . ( F ) Fluorescence recovery ( FRAP ) curves of wild-type mCherry-p62 recruited to glutathione beads coated with decreasing amounts of GST-LC3B . Averages and SD of four independent curves per sample are shown . ( G ) Plot of extrapolated recovery half-times ( t1/2 ) from ( F ) against the respective LC3B concentration on the beads . Data points were fitted to a mono-exponential equation . Robustness of the fit ( R2 ) and the extrapolated half-maximal LC3B concentration ( c1/2 ) are indicated . ( H ) FRAP curves of the indicated p62 variants on GST-LC3B coated beads . Averages and SD of four independent curves are shown . ( I ) Representative pictures for the graph shown in ( H ) . Scale bar 20 µm . ( C , D ) Total beads quantified: wild type = 101 , delta PB1 = 162 . ( E ) Total beads quantified: wild type = 78 , delta PB1 = 71 . ( Figure supplement 1 ) Total beads quantified: wild type = 98 , delta PB1 = 133 . ( Figure supplement 2 ) Total beads quantified: wild type = 45 , delta PB1 = 49 . ( Figure supplement 4 ) Total beads quantified per condition . Wild type: 0% LC3B = 150; 1% LC3B = 141; 2% LC3B = 130; 4% LC3B = 92; 10% LC3B = 119; 50% LC3B = 92; 100% LC3B = 132 . delta PB1: 0% LC3B = 82; 1% LC3B = 123; 2% LC3B = 69; 4% LC3B = 66; 10% LC3B = 100; 50% LC3B = 93; 100% LC3B = 93 . DOI: http://dx . doi . org/10 . 7554/eLife . 08941 . 00810 . 7554/eLife . 08941 . 009Figure 3—figure supplement 1 . ( A ) Quantification of the decay of the indicated mCherry-p62 variants from GST-LC3B-coated beads . Fluorescence intensities at T = 0 are set to 100% . Averages and SD of three independent replicates are shown . ( B ) Representative pictures for the data shown in ( A ) . For better comparison , the brightness was adjusted so that intensities of beads at time 0 are identical . Scale bars , 25 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 08941 . 00910 . 7554/eLife . 08941 . 010Figure 3—figure supplement 2 . p62 association to GST-LC3B-coated beads . ( A ) Quantification of wild-type mCherry-p62 or delta PB1 recruitment to GST-LC3B-coated beads over time . LC3B-coated beads were added to a 1 µM mCherry-p62 solution and the sample was immediately imaged by spinning disk microscopy . Samples were imaged every 20 s for 1 hr . The increase of mCherry-p62 fluorescence intensity on the beads is plotted against time . The inset on the right shows the increase in fluorescence intensity over the first 5 min for mCherry-p62 wild-type and delta PB1 . ( B ) Representative images of the experiment in ( A ) . The fluorescence intensity on a single bead over time for each p62 variant is shown . The brightness was adjusted so that the fluorescence intensities of the beads at 1 s time are identical . DOI: http://dx . doi . org/10 . 7554/eLife . 08941 . 01010 . 7554/eLife . 08941 . 011Figure 3—figure supplement 3 . Fluorescence recovery after photo-bleaching ( FRAP ) curves of mCherry-p62 delta PB1 recruited to beads coated with the indicated GST-LC3B concentrations . Averages and SD of two independent curves per sample are shown . DOI: http://dx . doi . org/10 . 7554/eLife . 08941 . 01110 . 7554/eLife . 08941 . 012Figure 3—figure supplement 4 . Steady-state binding of the indicated mCherry-p62 variants to beads coated with indicated GST-LC3B amounts . mCherry-p62 variants were incubated at a concentration of 2 µM , 100% GST-LC3B is equivalent to 1 . 5 µg GST-LC3B per µL of beads as described in the Methods . GST only was used as negative control ( 0% GST-LC3B ) . Data are normalized to 100% GST-LC3B for each sample . Data points were fitted to single exponential curves with the plateau set to 100% . Concentrations of LC3B giving half-maximal binding ( c1/2 ) are indicated . DOI: http://dx . doi . org/10 . 7554/eLife . 08941 . 012 We confirmed this result by following the dissociation of the two mCherry-p62 variants from GST-LC3B beads ( Figure 3—figure supplement 1 ) . To this end , GST-LC3B-coated beads were incubated with wild-type mCherry-p62 or the delta PB1 mutant , diluted into empty buffer and imaged over time . While delta PB1 p62 started to dissociate immediately after dilution , the wild-type protein remained stably bound to the beads for up to 1 . 5 hr after dilution . We , therefore , concluded that oligomerization decreases the off-rate of p62 from surface-localized LC3B . We also followed the kinetics of association of wild-type and delta PB1 mCherry-p62 to GST-LC3B-coupled beads . Both proteins showed an initially fast association with the beads ( Figure 3—figure supplement 2 , insert ) . However , while no further increase in bead-associated signal was observed for the delta PB1 protein , the wild-type mCherry-p62 further accumulated on the beads over the time course of 1 hr ( Figure 3—figure supplement 2 ) . Next , we asked whether oligomerization would also positively affect binding of p62 to free LC3B . We , therefore , immobilized p62 on RFP-TRAP beads and added free GFP-LC3B . The recruitment of wild-type and delta PB1 p62 to RFP-TRAP beads was equally efficient ( Figure 3C , black bars ) . To our surprise , mCherry-p62 delta PB1 was twice as efficient as the wild-type protein in recruiting free GFP-LC3B ( Figure 3C , gray bars , and Figure 3D ) . We then went on to measure the decay of the GFP-LC3B signal from the beads upon dilution ( Figure 3E ) . Here , GFP-LC3B readily dissociated from beads coated with both the oligomeric and non-oligomeric p62 variants , with no significant difference . Taken together , these results suggest that oligomerization of p62 specifically promotes interaction with surface-localized , clustered LC3B by drastically reducing the off-rate of p62 from LC3B-coated surfaces . In contrast , oligomerization does not affect the intrinsic affinity of LIR motif for LC3B since binding to free LC3B is not enhanced . This hypothesis predicts that the stability of oligomeric p62 on LC3B-coated surfaces should directly correlate with the density of LC3B on the surface . We tested this hypothesis by recruiting wild-type mCherry-p62 to beads coated with decreasing densities of GST-LC3B and measured the fluorescence recovery rates after bleaching ( Figure 3F ) . Strikingly , decreasing the density of LC3B on the beads resulted in faster recovery rates for wild-type p62 ( Figure 3F ) . In contrast , the recovery rate of p62 delta PB1 was not affected when the density of LC3B on the beads was reduced even by a factor of 10 ( Figure 3—figure supplement 3 ) . We then plotted the recovery rates extrapolated from the FRAP curves against the respective LC3B density on the beads ( Figure 3G ) . The data points fitted robustly to an exponential curve , which showed a half-maximum around 14% of LC3B density . This value is in line with the result we obtained when we measured the steady-state binding of p62 to beads coated with different densities of GST-LC3B ( c1/2 = 9 . 5% for the wild-type protein , Figure 3—figure supplement 4 ) . The data above strongly support a model of oligomerization-dependent LIR motif clustering and hence high-avidity interactions with surfaces on which LC3B is clustered . If this is indeed the case , then the same behavior should be displayed by a non-oligomerizing version of p62 containing multiple LIR motifs . We , therefore , generated a mCherry-p62 delta PB1 protein containing 4 LIR motifs ( 4xLIR ) and tested its exchange rate on LC3B-coated beads by FRAP ( Figure 3H , I ) . Strikingly , p62 delta PB1 4xLIR showed a recovery rate approximately four times slower than delta PB1 p62 containing only one LIR motif ( Figure 3H ) . Given the effect of p62 oligomerization on LC3B binding , we asked whether a similar mechanism applied to the interaction with ubiquitin . Indeed , it was previously reported that the deletion of the PB1 domain resulted in reduced interaction with ubiquitin in a pull-down assay ( Kirkin et al . , 2009 ) . We first tested the interaction of mCherry-p62 with GFP-ubiquitin in GFP-TRAP experiments using cell lysates from transfected HeLa cells ( Figure 4A ) in which the endogenous p62 was downregulated by siRNA treatment . The ability of the p62 variants to co-precipitate with GFP-ubiquitin correlated strongly with their ability to oligomerize . While the wild-type protein and the LIR mutant interacted most robustly with ubiquitin , this interaction was markedly reduced for the K7A/D69A mutant . The non-oligomerizing delta PB1 mutant and the NBR1-p62 chimera showed barely detectable interactions with ubiquitin . 10 . 7554/eLife . 08941 . 013Figure 4 . Oligomerization of p62 promotes ubiquitin binding . ( A ) GFP-TRAP experiment using HeLa cell lysates co-expressing GFP ( control ) or GFP-ubiquitin and the indicated mCherry-p62 variants . The endogenous p62 was silenced by siRNA treatment . Eight percent input and 100% of the bead fractions were analyzed by western blotting using anti-GFP and anti-p62 antibodies . ( B ) Scheme of the set-up of the experiment shown in ( C ) and ( D ) Recombinant GFP-ubiquitin was cross-linked to 2 µm latex beads and incubated with purified mCherry-p62 variants at 50 nM final concentration . Beads were observed using a spinning disk microscope under steady-state conditions . ( C ) Representative images of the recruitment of mCherry-p62 variants on GFP-ubiquitin-coated beads . Pictures were taken using the same microscopy settings and shown in false color for the mCherry-p62 signal ( ImageJ: fire ) . Scale bar 1 μm . ( D ) Quantification of mCherry-p62 recruitment to beads coated with GFP-ubiquitin or GFP . Averages and SD of three independent replicates are shown . Indicated p-values were calculated with a two-tailed unpaired Student’s t-test . ( E ) Quantification of steady-state binding of the indicated p62 variants to the indicated ubiquitin chains cross-linked to 2 µm latex beads . Averages and SD of three independent replicates are shown . All data are normalized to wild-type mCherry-p62 binding to linear tetra-ubiquitin . p-Values were calculated using a two-tailed unpaired Student’s t-test . ( F ) Coomassie-stained gels showing p62 sedimentation assays conducted with recombinant wild-type mCherry-p62 in the presence of the indicated tetra-ubiquitin chains . GST was used as a negative control . For each sample , the input , supernatant , and pellet fractions are shown . Quantifications are shown below the gel . The protein amount in the pellets and supernatants are expressed as fractions of the input . ( G ) Quantification of steady-state binding of the indicated p62 variants to beads coated with GST-mono- , di- or –tetra-ubiquitin . GST was used as negative control . Averages and SD of at least three independent experiments are shown . Data are normalized to wild-type mCherry-p62 binding to GST-tetra-ubiquitin . Data points were fitted to mono-exponential curves ( dashed lines ) . ( H ) p62 co-sedimentation assay with increasing concentrations of linear tetra-ubiquitin . Wild-type mCherry-p62 was incubated with linear tetra-ubiquitin chains at the indicated molar ratios before ultracentrifugation . Inputs , supernatants and pellets were analyzed by SDS-PAGE followed by Coomassie staining . Quantification was performed as described for ( F ) . ( I ) Fluorescence recovery after photo-bleaching ( FRAP ) curves of wild-type mCherry-p62 recruited to mono-ubiquitin or tetra-ubiquitin-coated beads . Averages and SD of six independent FRAP recordings are shown . ( J ) FRAP curves of wild-type mCherry-p62 recruited to beads coated with decreasing concentrations of mono-ubiquitin . For each sample , the averages and SD from six independent FRAP recordings are shown . ( K ) Quantification of wild-type and delta PB1 mCherry-p62 decay from GST-di-ubiquitin-coated beads . Averages and SD of three independent replicates are shown . ( L ) Representative images of data shown in ( K ) . For better comparison , brightness was adjusted so that intensities of beads at time 0 is identical . Scale bars , 25 μm . ( D ) Total beads counted per condition: GFP-ub coated beads + mCherry-p62 wild-type = 565; GFP-ub coated beads + mCherry-p62 K7A/D69A = 383; GFP-ub coated beads + mCherry-p62 delta PB1 = 378; GFP-ub coated beads + mCherry-NBR1-p62 chimera = 476; GFP-ub coated beads + mCherry-p62 LIR mutant = 393; GFP-ub coated beads + mCherry-p62 ∆UBA = 347; GFP coated beads + mCherry-p62 wild-type = 187 . ( E ) Total beads quantified per condition: mCherry-p62 WT: M1 4xUB = 427; K48 4xUB = 332; K63 4xUB = 305; mock = 95 . mCherry-p62 delta PB1: M1 4xUB = 266; K48 4xUB = 239; K63 4xUB = 226; mock = 75 . ( G ) Total beads quantified per condition: mCherry-p62 wild-type: GST = 107; GST-mono-ubiquitin = 182; GST-di-ubiquitin = 149; GST-tetra-ubiquitin = 236 . mCherry-p62 delta PB1 ) GST = 113; GST-mono-ubiquitin = 165; GST-di-ubiquitin = 134; GST-tetra-ubiquitin = 241 . ( K ) Total beads quantified: wild-type = 83 , delta PB1 = 65 . DOI: http://dx . doi . org/10 . 7554/eLife . 08941 . 01310 . 7554/eLife . 08941 . 014Figure 4—figure supplement 1 . p62-LC3B co-sedimentation assay . Wild-type mCherry-p62 was incubated with LC3B at the indicated p62:LC3B ratios for 1 hr . After ultracentrifugation , input , supernatant and pellet fractions were analyzed by SDS-PAGE followed by Coomassie staining . DOI: http://dx . doi . org/10 . 7554/eLife . 08941 . 014 Next , we investigated whether the same was true under equilibrium conditions . To this end , we covalently coupled GFP-ubiquitin to 2 µm beads and added wild-type mCherry-p62 at a final concentration of 50 nM ( Figure 4B–D ) . Spinning disk microscopy was then used to determine the association of p62 with the beads . The non-oligomeric delta PB1 mutant and the NBR1-p62 chimera as well as the ΔUBA mutant showed strongly reduced recruitment to the GFP-ubiquitin-coated beads when compared to the wild-type protein ( Figure 4C , D ) . Consistent with the GFP-TRAP experiment ( Figure 4A ) , the K7A/D69A mutant showed only slightly reduced binding to the GFP-ubiquitin-coated beads . In vivo individual ubiquitin molecules are frequently covalently attached to one another forming longer chains . Depending on the residue used for the linkage formation , different chain types can be formed , each of them with a different functional role in the cell ( Husnjak and Dikic , 2012; Komander and Rape , 2012 ) . p62 was shown to bind preferentially K63- over K48-linked chains ( Long et al . , 2008; Matsumoto et al . , 2011; Seibenhener et al . , 2004 ) . We , therefore , asked whether oligomerization of p62 influences the binding specificity for different ubiquitin chains . To this end , we cross-linked linear ( M1 ) - , K48- , or K63-linked tetra-ubiquitin chains to 2 µm beads and measured the binding of wild-type and delta PB1 mCherry-p62 at equilibrium ( Figure 4E , black bars ) . Consistent with previous reports , wild-type mCherry-p62 bound stronger to K63-linked chains than to the K48-linked chains ( Long et al . , 2010; Matsumoto et al . , 2011; Seibenhener et al . , 2004 ) . The strongest binding was detected for linear ubiquitin . When we compared the binding intensities of p62 delta PB1 ( Figure 4E , gray bars ) with the wild-type protein , we made two observations: first , binding to linear ubiquitin was strongly reduced , and second , there was no longer a significant difference in binding to the three chain types . We concluded that oligomerization of p62 determines specificity toward linear and perhaps weakly toward K63-linked ubiquitin chains , while non-oligomerizing p62 delta PB1 binds indifferently to all three chain types . Interestingly , oligomerization does not promote binding to K48-linked ubiquitin chains . It was reported that addition of K63-linked ubiquitin chains partially disrupted p62 oligomers ( Ciuffa et al . , 2015 ) . Employing the p62 pelleting assay ( [Ciuffa et al . , 2015] and Figure 2B , C ) , we tested the effect of linear , K48- and K63-linked tetra-ubiquitin chains on the oligomerization of p62 . All three chain types had a measurable effect on the oligomerization of p62 ( Figure 4F ) . This effect was specific as GST did not disrupt p62 oligomers . Consistent with previous experiments ( Ciuffa et al . , 2015 ) , we did also not detect any effect of LC3B on the oligomerization of p62 [Figure 4—figure supplement 1] . Addition of K48-linked ubiquitin chains had the strongest disruptive effect on p62 oligomerization ( Figure 4F ) . Together with the fact that these chains were not preferentially bound by oligomeric p62 ( Figure 4E ) this suggests that p62 oligomers may be locally disrupted upon binding to the beads cross-linked with K48-linked ubiquitin chains . Since the strongest oligomerization-dependent binding of p62 to ubiquitin was detected for linear chains , we analyzed this interaction further . When beads coupled to mono-ubiquitin , linear di-ubiquitin , and tetra-ubiquitin were tested for p62 binding , it became apparent that both wild-type and delta PB1 p62 bound stronger to longer ubiquitin chains ( Figure 4G ) . Thus , even though linear ubiquitin chains disrupt p62 oligomers to some extent in a concentration-dependent manner ( Figure 4H ) , they are still bound stronger than mono-ubiquitin ( Figure 4G ) . We then went on to study whether lower off-rates contribute to the stronger binding of the wild-type protein to linear tetra-ubiquitin compared to mono-ubiquitin . Indeed , FRAP analysis of wild-type mCherry-p62 bound to mono-ubiquitin and linear tetra-ubiquitin showed that the recovery rate was higher for mono-ubiquitin ( Figure 4I ) . Lowering the density of mono-ubiquitin on the beads also resulted in increased FRAP recovery rates ( Figure 4J ) , similarly to what we observed for LC3B ( Figure 3F ) . These results suggested that oligomerization of p62 results in clustering of the ubiquitin-binding UBA domain and thus avid binding to surface-localized , clustered ubiquitin , analogous to the interaction with surface-localized LC3B ( Figures 2 and 3 ) . To directly test whether oligomerization of p62 confers more avid interaction with ubiquitin on surfaces , we measured the decay of mCherry-p62 wild-type or delta PB1 from ubiquitin-coated beads upon dilution in buffer ( Figure 4K , L ) . Both proteins showed some degree of dissociation from the surface of the beads but the oligomeric wild-type p62 remained more stably bound . In vivo p62 interacts with ubiquitin when it is concentrated on the cargo and with LC3B when it is localized on the isolation membrane . We , therefore , asked what the consequences of the simultaneous interaction of p62 with ubiquitin and LC3B would be in the context of membrane-localized LC3B and ubiquitin localized to a surface . First , we asked whether p62 actually possesses the ability to simultaneously interact with LC3B and ubiquitin . To this end , we conducted experiments using GST-di-ubiquitin-coated beads to indirectly recruit GFP-LC3B via p62 ( Figure 5A–C , Figure 5—figure supplement 1 ) . First , the recruitment of mCherry-p62 variants to GST-di-ubiquitin-coated beads recapitulated our results with GFP-mono-ubiquitin ( compare Figure 5B , black bars , with Figure 4D ) . We next assessed the ability of the p62 variants to recruit GFP-LC3B to the beads . No significant difference between wild-type p62 and the K7A/D69A mutant was observed , while p62 delta PB1 and the NBR1-p62 chimera recruited significantly less LC3B ( Figure 5B , gray bars ) . However , when the GFP-LC3B signal is normalized to the corresponding mCherry-p62 signal , the non-oligomerizing mutants appear to be about twice as efficient as p62 wild-type in recruiting GFP-LC3B ( Figure 5C ) . This mirrors the results we obtained when we directly tethered mCherry-p62 to RFP-TRAP beads ( Figure 3D ) . In summary , oligomerization of p62 generates high-avidity interactions that increase the residence time of the oligomeric particle on LC3B and ubiquitin-coated structures . However , the efficiency of interaction with LC3B for each p62 monomer within the oligomeric structure is reduced . 10 . 7554/eLife . 08941 . 015Figure 5 . Reconstitution of p62–mediated membrane bending . ( A–C ) Indirect recruitment of GFP-LC3B to GST-di-ubiquitin coated beads via mCherry-p62 . ( A ) Scheme of the experiment . GST-di-ubiquitin was pre-recruited to glutathione agarose beads . Beads were co-incubated with mCherry-p62 variants and GFP or GFP-LC3B . Imaging was performed at equilibrium . ( B ) Quantification of mCherry and GFP intensities on the beads ( see Figure 5—figure supplement 1 for representative pictures ) . All values are plotted as percentages of the wild-type mCherry-p62 intensity . Averages and SD of four independent replicates are shown . Indicated p-values were calculated with a two-tailed unpaired Student’s t-test . p-Values above black bars refer to the mCherry-p62 wild-type bar; p-values above gray bars refer to the GFP-LC3B intensity in the wild-type mCherry-p62 sample . ( C ) Plot of GFP/mCherry ratio of data shown in ( B ) . The ratio for wild-type mCherry-p62 was normalized to 1 . All p-values were calculated with a two-tailed unpaired Student’s t-test . ( D ) Quantification and representative pictures of LC3B-positive giant unilamellar vesicle ( GUV ) membranes bending around 2 µm glutathione beads coated with GST-tetra-ubiquitin and incubated with the indicated mCherry-p62 variants . Averages and SD of four independent experiments are shown . The indicated p-value was calculated with a two-tailed unpaired Student’s t-test . n numbers indicate the total number of beads quantified per sample . Scale bars , 2 µm . ( E ) Quantification and representative pictures of LC3B-positive GUV membranes bending around 2 µm latex beads cross-linked with the indicated mCherry-p62 variants . Averages and SD of three independent experiments are shown . n numbers indicate the total number of beads quantified per sample . Scale bars , 2 µm . ( F ) Quantification and representative pictures of LC3B-positive GUV membranes bending around 2 µm latex beads cross-linked with the indicated ubiquitin chains and incubated with wild-type mCherry-p62 . Averages and SD of four independent experiments are shown . The indicated p-value was calculated with a two-tailed unpaired Student’s t-test . n numbers indicate the total amount of beads counted per sample . Scale bars , 2 µm . ( B ) Total beads counted per condition: GST + mCherry-p62 wild type + GFP-LC3B = 101 , GST-2xUB + mCherry-p62 wild type + GFP = 125 , GST-2xUB + mCherry-p62 wild type + GFP-LC3B = 174 , GST-2xUB + mCherry-p62 DM + GFP-LC3B = 172 , GST-2xUB + mCherry-p62 delta PB1 + GFP-LC3B = 154 , GST-2xUB + mCherry-NBR1-p62 chimera + GFP-LC3B = 153 , GST-2xUB + mCherry-p62 LIR mut + GFP-LC3B = 129 . DOI: http://dx . doi . org/10 . 7554/eLife . 08941 . 01510 . 7554/eLife . 08941 . 016Figure 5—figure supplement 1 . Representative pictures of the data shown in Figure 5B . Scale bars , 20 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 08941 . 016 In order to more fully reconstitute the system in vitro , we attached LC3B-6xHis to the surface of GUVs ( Figure 5D ) . To visualize the vesicles , the membrane was labeled by incorporation of Oregon-green phosphatidylethanolamine . GST-linear tetra-ubiquitin was bound to 2 µm glutathione beads . The beads were then incubated with mCherry-p62 variants and added to GUVs . Strikingly , wild-type mCherry-p62 mediated strong bending of the GUV membrane around the beads . Frequently , the beads were completely submerged into the GUVs ( Figure 5D ) . Membrane bending was dependent on the specific interaction with LC3B as the LIR mutant showed strongly reduced membrane bending activity . The non-oligomerizing delta PB1 mutant showed reduced membrane-bending efficiency , likely due to the fact that less p62 delta PB1 was localized to the ubiquitin-coated beads ( Figure 4 ) . To determine whether membrane bending by p62 also specifically required its interaction with ubiquitin , we directly cross-linked mCherry p62 delta PB1 to beads . Cross-linked p62 delta PB1 efficiently mediated membrane bending in a LIR-dependent manner ( Figure 5E ) , showing that the presence of ubiquitin is not essentially required for membrane bending . Next , we tested the ability of p62 to bend the membrane around beads cross-linked to linear , K48- and K63-linked tetra-ubiquitin chains ( Figure 5F ) . We observed membrane bending events for all the three chain types . However , membrane bending was significantly reduced for K48-linked ubiquitin , consistent with the lower affinity of p62 for this chain ( Figure 4E ) . In summary , we conclude that the interaction of p62 with ubiquitin and LC3B is sufficient to drive bending of a LC3-coated membrane around an ubiquitin-positive cargo . Given our in vitro results , we wanted to know whether oligomerization mutants of p62 would localize to LC3B-positive structures in vivo . We , therefore , examined the co-localization of mCherry-p62 with endogenous LC3B in HeLa cells in which the endogenous p62 was silenced by siRNA ( Figure 6A ) . Consistent with earlier results ( Bjørkøy et al . , 2005; Ichimura et al . , 2008; Pankiv et al . , 2007 ) the wild-type and LIR mutant proteins localized in multiple puncta , but only wild-type p62 extensively co-localized with LC3B . The delta PB1 protein and the NBR1-p62 chimera showed no puncta formation and appeared cytosolic . Interestingly , the K7A/D69A protein was largely cytosolic but still displayed some degree of puncta formation and co-localization with LC3B ( Figure 6A ) . Next , we went on to test whether the ability of p62 to oligomerize would affect its accumulation around cargo particles and its ability to recruit LC3B also in cells . To this end , we adapted a previously described assay that is based on the coating of small latex beads with transfection reagent ( [Kobayashi et al . , 2010] and Figure 6B ) . Upon internalization of the beads by the cell , the transfection reagent damages the endosomal membrane , which then becomes a target for selective autophagy ( Kobayashi et al . , 2010; Thurston et al . , 2012 ) . In order to render the beads themselves a direct target for selective autophagy , we coated them with recombinant TagBFP-ubiquitin before coating with transfection reagent ( Figure 6—figure supplement 1A ) . TagBFP-ubiquitin-coated beads were then added to HeLa cells that had the endogenous p62 protein downregulated by RNAi ( Figure 6C and Figure 6—figure supplement 1B ) and that were co-transfected with mCherry-p62 and GFP-LC3B . Extracellular beads were stained using an anti-ubiquitin antibody allowing us to count only the intracellular beads ( Figure 6—figure supplement 1C ) . mCherry-p62 wild–type was robustly recruited to the TagBFP-ubiquitin-coated beads ( Figure 6D , E ) . The ability of the p62 mutants to associate with the beads strongly correlated with their ability to oligomerize ( Figure 6E ) . While the non-oligomeric delta PB1 mutant and the NBR1-p62 chimera showed almost no recruitment above the experimental background ( mCherry ) , the K7A/D69A mutant was still recruited to a considerable degree ( Figure 6E ) . To follow the recruitment of LC3B to the beads , we quantified the number of beads that were positive for both mCherry-p62 and GFP-LC3B ( Figure 6D , F ) . The recruitment of GFP-LC3B to the beads was also strongly dependent on the ability of p62 to oligomerize . Wild-type p62 showed robust recruitment of LC3B to the beads while p62 delta PB1 and the NBR1-p62 chimera showed very low LC3B recruitment . The p62 K7A/D69A mutant displayed an intermediate behavior between the two extremes with regard to LC3B recruitment . Moreover , we noticed that the effect of the oligomerization on the LC3B recruitment to the beads was largely dependent on the reduced recruitment of the p62 oligomerization mutant . This became obvious when we quantified the total recruitment of LC3B to all intracellular beads , regardless of whether they were positive or negative for p62 ( Figure 6G ) . This quantification showed that there is at least one redundant factor that is able to recruit LC3B to the ubiquitin-coated beads . This redundant factor accounts for 68% of the LC3B recruitment activity ( compare wild-type p62 to mCherry ) . Obvious candidates for this factor are other cargo receptors such as NBR1 ( Kirkin et al . , 2009 ) , optineurin ( Wild et al . , 2011 ) , or Tollip ( Lu et al . , 2014 ) . However , within the dynamic range of our assay for p62 recruitment activity , the p62 oligomerization mutants showed a profound loss of LC3B recruitment . 10 . 7554/eLife . 08941 . 017Figure 6 . Oligomerization of p62 promotes recruitment of p62 and LC3B to ubiquitin-coated beads in HeLa cells . ( A ) Anti-LC3B immunofluorescence analysis of HeLa cells transiently transfected with mCherry-p62 variants . Nuclei were stained with DAPI . Insets show magnifications of the indicated squares . Scale bars , 5 µm . ( B–G ) Quantification of mCherry-p62 and GFP-LC3B recruitment around artificial cargo particles in HeLa cells . ( B ) Schematic outline of the experiment . ( C ) Western blot analysis of HeLa cell lysates overexpressing wild–type mCherry-p62 with or without silent mutations in the siRNA targeting region . ( D ) HeLa cell co-expressing siRNA resistant wild-type mCherry-p62 and GFP-LC3B . Endogenous p62 was silenced by siRNA ( see Figure 6—figure supplement 1 ) . The arrows indicate co-localization of mCherry-p62 and GFP-LC3B at a BFP-ubiquitin-coated 2 µm bead . Scale bar: 5 µm . ( E ) Quantification of mCherry-p62 variants localizing to BFP-ubiquitin-coated beads in mCherry-p62 and GFP-LC3B co-expressing cells . ( F ) Quantification of co-localization of mCherry-p62 variants and GFP-LC3B at BFP-ubiquitin-coated beads . ( G ) Quantification of GFP-LC3B localization to BFP-ubiquitin-coated beads . For all data in ( D–G ) , averages and SD of three independent replicates are shown . Indicated p-values were calculated by a two-tailed equal-variance Student’s t-test . All graphs show the averages and SD . ( E–G ) Total beads quantified per condition: wild-type = 113 beads , K7A/D69A = 145 beads , delta PB1 = 117 beads , NBR1-p62 chimera = 120 beads , mCherry = 144 beads . DOI: http://dx . doi . org/10 . 7554/eLife . 08941 . 01710 . 7554/eLife . 08941 . 018Figure 6—figure supplement 1 . ( A ) Representative picture of 2 µm latex beads cross-linked with BFP-ubiquitin . Left: differential interference contrast; right: BFP fluorescence . ( B ) Immunofluorescence of HeLa cells transfected with a scramble siRNA ( left ) or a siRNA against p62 ( right ) . Cells were stained with an antibody against endogenous p62 . Nuclei were counterstained with DAPI . ( C ) Immunofluorescence of HeLa cells incubated with BFP-ubiquitin-coated beads . Non-permeabilized cells were stained with an antibody against ubiquitin , so that only non-internalized beads are labeled . The dashed lines indicate the cell’s contour . All scale bars , 5 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 08941 . 018 In conclusion , the experiments presented in Figure 6B–G show that both the recruitment of p62 to ubiquitin-positive beads and the recruitment of LC3B to these beads by p62 are promoted by oligomerization of p62 . We next extended our analysis to a more physiological target of p62 and infected HeLa cells with Salmonella typhimurium , an intracellular pathogenic bacterium previously shown to be a p62 target ( Zheng et al . , 2009 ) ( Figure 7 and Figure 7—figure supplement 1 ) . 10 . 7554/eLife . 08941 . 019Figure 7 . Oligomerization of p62 is required for efficient recruitment of p62 to Salmonella typhimurium in HeLa cells . ( A ) Representative pictures of HeLa cells co-expressing GFP-LC3B and mCherry-p62 infected with S . typhimurium . The endogenous p62 was silenced by siRNA . Magnifications of the insets are shown on the right . Pictures of whole cells are shown in Figure 7—figure supplement 1 . Scale bars , 5 µm . ( B ) Quantification of mCherry-p62- and/or GFP-LC3B-positive bacteria . Averages and SD of three independent replicates are shown . Indicated p-values were calculated with a two-tailed unpaired Student’s t-test . Values above the black bars refer to the wild-type mCherry-p62 value in the wild-type mCherry-p62 + GFP-LC3B sample; values above the gray bars refer to the GFP-LC3B value in the same sample . ( B ) Total bacteria counted per condition: mCherry-p62 wild-type + GFP-LC3B = 245 , mCherry-p62 K7A/D69A + GFP-LC3B = 337 , mCherry-p62 LIR + GFP-LC3B = 287 , mCherry-p62 delta PB1 + GFP-LC3B = 296 , mCherry-NBR1-p62 chimera + GFP-LC3B = 292 , mCherry + GFP-LC3B = 318 , mCherry-p62 wild-type + GFP = 325DOI: http://dx . doi . org/10 . 7554/eLife . 08941 . 01910 . 7554/eLife . 08941 . 020Figure 7—figure supplement 1 . Full-size pictures of the data shown in Figure 7A . White squares indicate cropped regions shown in the main figure . Scale bars , 5 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 08941 . 02010 . 7554/eLife . 08941 . 021Figure 7—figure supplement 2 . Quantification of co-localization of the indicated mCherry-p62 variants and GFP-LC3B at bacteria . p-Values refer to mCherry-p62 wild-type + GFP-LC3B sample . DOI: http://dx . doi . org/10 . 7554/eLife . 08941 . 021 LC3B was efficiently recruited to the bacteria even in the absence of p62 ( Figure 7B , gray bars ) . This suggests that there might be other mechanisms for the autophagic targeting of intracellular Salmonella . Wild-type p62 was robustly recruited to intracellular Salmonella and showed extensive co-localization with LC3B in vicinity of the bacteria . The LIR mutant also showed robust recruitment to the Salmonella . However , in contrast to the HeLa cells expressing wild-type p62 , the LIR mutant and LC3B were localized into different patches , showing only partial co-localization ( Figure 7A , Figure 7—figure supplement 1 , 2 ) . The K7A/D69A mutant was still recruited to the bacteria while the oligomerization-deficient delta PB1 mutant and the NBR1-p62 chimera were not robustly recruited to the Salmonella ( Figure 7B ) . Furthermore , all of these mutants showed reduced co-localization with LC3B ( Figure 7B , Figure 7—figure supplement 2 ) , similar to what we observed for ubiquitin-positive latex beads ( Figure 6F , G ) . Interestingly , when we quantified the overall amount of bacteria positive for LC3B , regardless of the presence of p62 , we could see a mild dominant negative effect of p62 oligomerization mutants ( Figure 7B , gray bars ) . In this study , we have shown that the human cargo receptor p62 employs oligomerization to generate high-avidity interactions with ubiquitin and LC3B . Thus , oligomerization enables p62 to simultaneously select for concentrated ubiquitin and LC3B ( Figure 8 ) . There are interesting parallels but also deviations compared to the yeast Atg19 cargo receptor . Atg19 binds its prApe1 cargo with very high affinity and selects for membrane-bound Atg8 via a high-avidity interaction mediated by multiple low-affinity Atg8 interaction sites ( Sawa-Makarska et al . , 2014 ) . These properties are advantageous because the prApe1 cargo is a dedicated selective autophagic cargo that needs to be delivered into the vacuole in order to fulfill its function ( Klionsky et al . , 1992 ) . Thus , the high-affinity interaction of Atg19 with the prApe1 cargo ensures its rapid transport into the vacuole . The multiple Atg8 binding sites in Atg19 subsequently mediate the selective interaction with membrane localized , locally concentrated Atg8 , enabling Atg19 to bend the membrane tightly around the cargo and to exclude non-cargo material from its delivery into the vacuole ( Baba et al . , 1997; Sawa-Makarska et al . , 2014 ) . In contrast , the cargo of p62 is not normally destined to be transported into the lysosomal system but fulfills a function in the cell’s cytoplasm . Only when this material becomes dysfunctional or superfluous it becomes marked with ubiquitin and thereby a target for selective autophagy . Aggregated proteins , for example , are a cargo for p62 ( Bjørkøy et al . , 2005 ) . However , when cytosolic proteins unfold and aggregate , they initially become a target for the ubiquitin-proteasome system ( UPS ) . When the UPS is overwhelmed , unfolded , ubiquitinated proteins form aggregates on which ubiquitin is locally concentrated . Only these structures should become a target for p62 and subsequently be degraded by selective autophagy . The low affinity but high avidity interaction of the p62 oligomer with ubiquitin will select for these structures as ubiquitin is locally concentrated on them ( Figure 8 ) . Interestingly , we found that K48-linked ubiquitin chains are less efficiently bound by oligomeric p62 compared to linear and K63-linked chains , possibly because K48-linked ubiquitin chains disrupt the p62 oligomers more effectively . These results hint to the possibility that K48-linked chains are not the preferred target for p62 in vivo , possibly in order to prevent proteins targeted for the proteasome to become premature targets of p62 . Thus aggregated proteins may need further modification by K63-linked or linear ubiquitin chains in order to render them efficient targets for p62 . 10 . 7554/eLife . 08941 . 022Figure 8 . A model for selective autophagy in yeast and mammalian cells . Multiple binding sites in the yeast cargo receptor Atg19 promote selective and exclusive engulfment of cargo material by Atg8-covered membranes ( Sawa-Makarska et al . , 2014 ) . Oligomerization of p62 allows it to simultaneously select for clustered ubiquitin and ATG8-family proteins . DOI: http://dx . doi . org/10 . 7554/eLife . 08941 . 022 Our data show that oligomerization has no direct effect on the individual LC3B–LIR and ubiquitin–UBA interactions . Instead , oligomerization drastically increases the residence time of p62 on LC3B and ubiquitin-coated surfaces . In the case of wild-type p62 and concentrated LC3B , the interaction even becomes irreversible and might represent the end point of a pathway in which the whole structure is eventually degraded in the lysosome . Moreover , we report that non-oligomerizing p62 mutants are more efficient than wild-type p62 in recruiting soluble LC3B . Consistent with a recently published structure of p62 , which showed a helical arrangement of the oligomer ( Ciuffa et al . , 2015 ) , we hypothesize that p62 oligomers adopt a three-dimensional structure that does not allow all LIR motifs to be engaged in LC3B interactions at the same time . Furthermore , we speculate that in vivo oligomerization is required to tightly appose the growing LC3B-positive membranes to the cargo particle , largely due to the fact that oligomerization mediates the concentration of p62 at the cargo . Additional regulation by cargo-localized kinases such as TBK1 ( Matsumoto et al . , 2011; 2015 ) may generate positive feedback loops resulting in more efficient delivery of ubiquitinated cargo into the lysosome . Furthermore , cooperation of p62 with other cargo receptors such as NBR1 fine-tunes the process and contributes to the clustering of aggregated proteins into larger structures ( Kirkin et al . , 2009 ) . Interestingly , we found that oligomerization mutants of p62 showed a dominant negative effect on LC3B recruitment around Salmonella . It is possible that the presence of p62 oligomerization mutants prevents other cargo receptors or the autophagic machinery in general to trigger the formation of an LC3B-positive isolation membrane . Self-association of autophagic cargo receptors in order to generate high-avidity interaction surfaces to select ubiquitinated cargo and ATG8-family protein decorated membranes may be a reoccurring theme . The cargo receptor optineurin forms higher-order oligomers ( Gao et al . , 2014 ) , while the NBR1 cargo receptor dimerizes ( Kirkin et al . , 2009 ) . It would , therefore , be interesting to test whether a similar molecular mechanism is also employed by these receptors . p62/SQSTM1: NP_003891; MAP1LC3B ( LC3B ) : NP_073729; GABARAP: NP_009209 . 1; ubiquitin: NP_001268649; NBR1: AAH09808 6xHis-TEV-mCherry-p62 constructs were generated as follows: human p62 was first cloned into pmCherry-C1 ( Clontech , Mountain View , CA , USA ) and then the mCherry-p62 fusion was subcloned into pET-Duet1 . A Tobacco Etch Virus ( TEV ) protease recognition site was also included to remove the 6xHis-tag . The K7A/D69A , LIR ( DDDW335-338AAAA ) , delta PB1 ( Δ2-102 ) , delta PB1 LIR mutant and ΔUBA ( Δ389-434 ) mutants were generated by PCR-based mutagenesis . The NBR1-p62 chimera was generated as follows: a fragment coding for amino acids 1–85 from human NBR1 was cloned into pET-Duet1 , followed by insertion of a fragment coding for amino acids 103–443 of p62 . The NBR1-p62 chimera was subcloned into pmCherry-C1 and the whole mCherry-NBR1-p62 construct was finally cloned into pET-Duet1 to generate 6xHis-TEV-mCherry-NBR1-p62 . The TEV site was used to remove the 6xHis-tag . p62 delta PB1 4xLIR was generated as follows: pETDuet-6xHis-TEV-mCherry-p62 delta PB1 was used as template for a PCR reaction using a forward primer with an overhang coding for the GSGSSGGDDDWTHLSS amino acid sequence . Upon self-ligation , the resulting construct coded for a 2xLIR version of p62 with the amino acids 332–343 ( SGGDDDWTHLSS [numbers relative to the wild-type protein] ) inserted after the wild-type LIR motif ( after amino acid 343 ) . The primer inserted an additional GSGS spacer between the two LIRs . p62 delta PB1 2xLIR was used as template for another PCR reaction that introduced HindIII and SalI sites between LIR1 and LIR2 . After self-ligation of this PCR product , an oligo coding for two LIR motifs ( GSSGGDDDWTHLSS ) was inserted via the SalI and HindIII sites . The final 6xHis-TEV-mCherry-p62 delta PB1 4xLIR construct coded for a protein with the following sequence inserted between amino acids 343 and 344: GSGSSGGDDDWTHLSSGSSGGDDDWTHLSSGSSGGDDDWTHLSS ( the numbers are relative to the wild-type protein and the additional three core LIR motifs are underlined ) . The proteins were expressed in Escherichia coli Rosetta ( DE3 ) pLysS cells . Bacteria were grown in Luria broth ( LB ) medium until OD600 ≈ 0 . 8–1 , induced with 0 . 1 mM isopropylthiogalactoside ( IPTG ) and grown at 25°C for 5 hr . Harvested cells were resuspended in lysis buffer 50 mM 4- ( 2-hydroxyethyl ) -1-piperazineethanesulfonic acid ( HEPES ) at pH 7 . 5 , 500 mM NaCl , 10 mM imidazole , 2 mM MgCl2 , 2 mM β-mercaptoethanol , complete protease inhibitor ( Roche , Basel , Switzerland ) and DNase I and lysed by a freeze–thaw cycle followed by brief 30 s sonication . Lysates were cleared by ultracentrifugation at 140 , 000 g for 30 min at 4°C ( Beckman , Brea , CA , USA , Ti45 rotor ) . Supernatants were applied to Ni-NTA columns ( GE Healthcare , Buckinghamshire , UK ) and 6xHis-tagged p62 constructs were eluted via a stepwise imidazole gradient ( 50 , 75 , 100 , 150 , 200 , and 300 mM ) . Protein-containing fractions were pooled and subjected to overnight cleavage with TEV protease at 4°C . Cleaved proteins were applied to a Superdex 200 column ( 16/600 , GE Healthcare ) and eluted with a buffer containing 25 mM HEPES pH 7 . 5 , 500 mM NaCl and 1 mM dithiothreitol ( DTT ) . Fractions containing the purified proteins were pooled , concentrated , frozen in liquid nitrogen , and stored at -80°C . GST-LC3B was generated by insertion of the human LC3B coding sequence into pGEX-4T1 . The last five amino acids of LC3B were deleted to mimic Atg4 cleavage . GST-di-ubiquitin and tetra-ubiquitin plasmids were a courtesy of Fumiyo Ikeda , Vienna , Austria . GST-mono-ubiquitin was generated by insertion of the human ubiquitin coding sequence into pGEX-4T-1 vector . GST-tagged proteins were expressed in E . coli Rosetta ( DE3 ) pLysS cells . Cells were grown in LB medium and induced at OD600 ≈ 0 . 8–1 for 4 hr at 37°C with 1 mM IPTG . Harvested cells were resuspended in a buffer containing 50 mM HEPES at pH 7 . 5 , 300 mM NaCl , 2 mM MgCl2 , 2 mM β-mercaptoethanol , complete protease inhibitor ( Roche ) and DNase I and lysed by freeze–thaw followed by sonication . Lysates were cleared by ultracentrifugation ( 140 , 000 g for 30 min at 4°C in a Beckman Ti45 rotor ) and supernatants were applied to glutathione beads ( GE Healthcare ) for 1 hr at 4°C . Beads were washed five times with 50 mM HEPES , 300 mM NaCl , 1 mM DTT . GST-tagged proteins were eluted with 20 mM reduced L-glutathione in 50 mM HEPES pH 7 . 5 , 300 mM NaCl , 1 mM DTT buffer for 1 hr at room temperature . The supernatant was concentrated and applied to a Superdex 75 column ( 16/600 , GE Healthcare ) previously equilibrated with 25 mM HEPES at pH 7 . 5 , 150 mM NaCl , 1 mM DTT . Fractions containing purified proteins were pooled , concentrated , frozen in liquid nitrogen , and stored at -80°C . eGFP-LC3B and eGFP-GABARAP were obtained by insertion of human LC3B and GABARAP cDNAs into pEGFP-C1 . Fusion proteins were subsequently cloned into pETDuet-1 for bacterial expression . The last five amino acids of LC3B coding sequence and the last amino acid of GABARAP were deleted to mimic Atg4 cleavage . A 6xHis-tag was added C-terminally to recruit the protein to membranes preserving their physiological orientation . To generate a monomeric meGFP-ubiquitin construct , mono-ubiquitin was cloned into pmeGFP-C3 vector , which encodes a monomeric enhanced GFP ( Zacharias et al . , 2002 ) , N-terminally of the cloning site . The fusion protein was subsequently subcloned into the pETDuet-1 vector . A TEV site was added with the forward primer to generate 6xHis-TEV-meGFP-ubiquitin . Blue fluorescently tagged ubiquitin was generated inserting mTAG-BFP into pETDuet1 to generate 6xHis-TEV-BFP followed by insertion of ubiquitin . Fluorescently tagged LC3B , ubiquitin and GABARAP were expressed in E . coli Rosetta ( DE3 ) pLysS cells . Cells were induced at an OD600 of 0 . 5 for 16 hr at 18°C with 0 . 1 mM IPTG . Proteins were purified on Ni-NTA columns as described above . Eluted eGFP-LC3B-6xHis and eGFP-GABARAP-6xHis were concentrated and directly applied to Superdex 75 column ( 16/60 , GE Healthcare ) . 6xHis-meGFP-ubiquitin and 6xHis-BFP-ubiquitin were subjected to overnight TEV cleavage prior to SEC . Proteins were eluted in 25 mM HEPES pH 7 . 5 , 150 mM NaCl , 1 mM DTT buffer , concentrated , frozen in liquid nitrogen and stored at -80°C . LC3B-6xHis was generated by insertion of human LC3B into pETDuet 1 . The last five amino acids of the coding sequence were deleted to mimic Atg4 cleavage . A 6xHis-tag was added C-terminally to recruit the protein to membranes preserving the physiological orientation . The protein was expressed overnight at 18°C in E . coli Rosetta ( DE3 ) pLysS cells in the presence of 0 . 1mM IPTG and subsequently purified via His-Trap and SEC using a Superdex 75 column ( 16/60 , GE Healthcare ) . Recombinant human tetra-ubiquitin ( K48 and K63-linked ) were purchased from Boston Biochem , Cambridge , MA , USA . The lyophilized powder was resuspended in SEC buffer ( 25 mM HEPES pH = 7 , 150 mM NaCl , 1mM DTT ) to a final concentration of 100 µM . Linear tetra-ubiquitin was generated from GST-tetra-ubiquitin by overnight thrombin cleavage at 4°C and subsequent purification via SEC ( Superdex 75 16/60 , GE Healthcare ) . For analytical SEC , 200 µg of the mCherry-p62 variants were applied to a Superose 6 column ( 10/300 , GE Healthcare ) or Superdex 200 column ( 10/300 , GE Healthcare ) and eluted with 25 mM HEPES pH 7 . 5 , 500 mM NaCl , 1mM DTT . 25 µL of 0 . 5 mL fractions were run on a 4-–20% SDS-PAGE gel ( Biorad , Hercules , CA , USA ) and stained with Coomassie . SLS analysis was done with a Superdex 200 column ( 10/300 , GE Healthcare ) . Online Multi-Angle Laser Light Scattering detection was performed with a MiniDawn Treos detector ( Wyatt Technology , Santa Barbara , CA , USA ) via a laser emitting at 690 nm and by refractive index measurement using a Shodex RI-101 ( Shodex , Munich , Germany ) . Sedimentation behavior of the different p62 variants was analyzed by ultracentrifugation of 1 µM p62 solutions at 150 , 000 g for 1 hr 30 min at 4°C . Supernatant and pellet fractions were compared to the protein input by SDS-PAGE followed by Coomassie staining . For p62 co-sedimentation assays with M1 , K48 and K63-linked tetra-ubiquitin chains or with LC3B , proteins were incubated at a p62:Ub or p62:LC3B molar ratio of 1:4 ( unless otherwise stated ) , for 1 hr on ice , before ultracentrifugation . The amount of p62 in the supernatant and pellet fractions was measured by gel densitometry using the ImageJ software . meGFP-ubiquitin , BFP-ubiquitin , and 6xHis-eGFP were cross-linked to carboxylated latex beads ( 4% w/v , Invitrogen ) with a diameter of 2 μm . Next , 50 μL of a 100 μM protein solution in 50 mM MES pH 6 . 0 were added to 25 μL beads suspension , diluted 1:1 in 50 mM MES pH 6 , and incubated at room temperature for 15 min . For direct cross-linking of p62 to beads , 50 µL of 50 µM mCherry-p62 delta PB1 and mCherry-p62 delta PB1 LIR were used . 0 . 8 mg of 1-ethyl-3- ( 3-dimethylaminopropyl ) carbodiimide were added to the mix and further incubated for 2 hr at room temperature . The reaction was quenched by the addition of glycine to a final concentration of 100 mM , followed by incubation for 30 min at room temperature . Beads were washed three times with phosphate-buffered saline ( PBS ) and resuspended in 100 µL of buffer containing 1% bovine serum albumin ( BSA ) in 15 mM HEPES pH 7 . 5 and 135 mM NaCl . Cross-linked beads were stored at 4°C . For HeLa cell transfection , beads were incubated with 0 . 25% BSA in PBS at room temperature for 15 min . After two washes with PBS , beads were stored in PBS at 4°C ( ∼10 µg/µL ) . For M1- K48- or K63-linked tetra-ubiquitin chains , 29 . 4 μL of a 100 µM protein solution were cross-linked to 29 . 4 μL of 2% 2 mm latex beads in 50 mM MES pH 6 . 0 . For mock cross-link , the same amount of SEC buffer was used . Beads were finally resuspended in 58 . 8 µL of 1% BSA , 15 mM HEPES at pH 7 . 5 , 135 mM NaCl buffer , and stored at 4°C . Lipids were purchased from Avanti Polar Lipids , Alabaster , AL , USA . GUVs were formed by electroformation at 30°C as previously described ( Romanov et al . , 2012 ) For protein recruitment , a mixture of 95% 1-palmitoyl-2-oleoyl-sn-glycero-3-phosphocholine ( POPC ) and 5% 1 , 2-dioleoyl-sn-glycero-3-[ ( N- ( 5-amino-1-carboxypentyl ) iminodiacetic acid ) succinyl] ( nickel salt ) ( DGS-Ni-NTA ) was used ( molar ratio ) . For beads engulfment experiments with p62 cross-linked to beads or GST-4xUb-coated beads , a mixture of 46% POPC , 46% 1 , 2-dioleoyl-sn-glycero-3-phosphocholine ( DOPC ) , 5% DGS-Ni-NTA , 3% oregon-green phosphatidylethanolamine ( Oregon-green PE ) was used ( molar ratio ) . For cross-linked ubiquitin chains experiment a mixture of 90% DOPC , 5% DGS-NiNTA , 5% Oregon-green DHPE was used . Electroformed GUVs were diluted 1:2 to 1:3 in GUV buffer ( 15 mM HEPES pH 7 . 5 , 135 mM NaCl , 1 mM DTT ) . The DOPC-containing GUVs were not diluted . eGFP-LC3B-6xHis was added at a final concentration of 400 nM . The mixture was incubated for at least 30 min at room temperature . mCherry-p62 variants were added to final concentration of 100 nM . Proteins were incubated for at least 30 min before imaging . For membrane bending experiments with cross-linked p62 on beads , LC3B-6xHis was added to GUVs at a final concentration of 200 nM and incubated for 30 min at room temperature . Beads were spun at 2000 rpm for 30 s to precipitate aggregates , then 10 µL of the supernatant was added to the GUVs and incubated for 30 min before imaging . For experiments with GST-linear tetra-ubiquitin-coated beads , 2 µm glutathione beads ( Sperotech , Lake Forest , IL , USA ) were used . In total , 100 µL of beads slurry were spun at 160 g for 5 min and washed once with GUV buffer 0 . 25% BSA . Then , 75 µg of GST-linear tetra-ubiquitin were recruited to beads for 30 min at 4°C in 150 µL GUV buffer . Beads were then washed once with GUV Buffer 0 . 25% BSA , divided into three aliquots , and each sample was incubated with 50 µL of 2 µM mCherry-p62 solution for 1 hr at 4°C . Beads were washed once and resuspended in 100 µL of GUV buffer . 10 µL of beads suspension were added to GUVs and incubated for 30 min before imaging . For experiments with ubiquitin chains cross-linked to beads , LC3B-6xHis was recruited to GUVs at 100 nM final concentration . mCherry-p62 wild type was recruited to beads at 500 nM final concentration for 1 hr at 4°C . At the end of the incubation , beads were sonicated for 5 s on ice , spun 2 min at 3500 g , half of the supernatant was removed and beads were resuspended in the remaining 25 μL volume . Then , 5 μL were added to GUVs to a total reaction volume of 45 μL . Before each use cross-linked meGFP-ubiquitin , BFP-ubiquitin , and 6xHis-eGFP beads were resuspended by vortexing , an aliquot was diluted 1:100 and sonicated for 15 min in ice water . Diluted beads were incubated with mCherry-p62 variants at a final protein concentration of 50 nM for at least 20 min at room temperature before imaging at a confocal spinning disk microscope ( Visitron , Puchheim , Germany ) . M1- K48- or K63-linked tetra-ubiquitin cross-linked beads were diluted 1:50 in SEC buffer and 2 . 5 µL were added to 22 . 5 µL of a 0 . 1 µM mCherry-p62 solution . Samples were incubated at room temperature for at least 20 min prior to imaging . 20 µl of Sepharose 4B glutathione beads ( GE Healthcare ) were used . For each reactionbeads were equilibrated by three washes with NETN-E buffer ( Pankiv et al . , 2007 ) . GST-tagged proteins were recruited to beads at 1 µM final concentration for 30 min at 4°C . Beads were washed once in NETN-E buffer and mCherry-p62 variants were added at 100 nM in a total volume of 55 µL . Beads were incubated for 1 hr at 4°C on an orbital shaker , washed twice with NETN-E buffer , and resuspended in 20 µL of 2× Laemmli loading buffer . Protein recruitment assays to glutathione beads were performed in SEC buffer ( 25 mM HEPES pH 7 . 5 , 150 mM NaCl , 1 mM DTT ) . Beads were incubated with 1 . 5 μg GST , GST-LC3B , GST-mono- , di- , or tetra-ubiquitin per μL beads for 30 min at 4°C , washed once and incubated with 2 . 5 μL per μL beads of 2 μM mCherry-p62 solution for 1 hr at 4°C . To measure p62 association to GST-LC3B-coated beads , imaging started immediately ( 1 s ) after the addition of mCherry-p62 solution to the beads . For co-recruitment of LC3B and p62 on ubiquitin-coated beads , mCherry-p62 and GFP-LC3B were incubated together with the beads at 2 μM and 1 μM , respectively . For GST-LC3B and GST-mono-ubiquitin titration , the total amount of protein recruited on beads was kept constant , but decreasing amounts of GST-LC3B and GST-mono-ubiquitin were mixed with increasing amounts of GST . For GST-LC3B , beads were incubated with 1 . 5 µg of total GST/GST-LC3B mixture per µL beads and for GST-mono-ubiquitin with 2 µg/µL beads . At the end of incubation 7 . 5 µL beads were diluted into 50 μL of p62 protein solution ( for steady-state imaging and FRAP ) or empty buffer ( for decay assays ) and imaged within a few minutes from dilution using a spinning disk microscope ( Visitron ) . RFP-Trap agarose beads were used ( ChromoTek , Martinsried , Germany ) . 20 µL beads were incubated with 50 µL of 2 µM mCherry-p62 and GFP-LC3B solution for 1 hr at RT . Beads were diluted in the same protein solution for steady-state imaging or in empty buffer for decay assays . The FRAP experiments were performed with GST-LC3B or GST-ubiquitin-coated beads , incubated with mCherry-p62 variants according to the GST pull-down assay protocol . Defined areas on the beads’ surface were photo-bleached using a 405 nm laser at 100% laser power for 50 ms per pixel and a 10 pixel-wide laser beam . Beads were imaged before and after photo-bleaching using a spinning disk microscope ( Visitron ) . Fluorescence recovery was recorded for the indicated time . Recovery half-times were calculated by fitting the curves to a mono-exponential equation with plateau set at 100% . HeLa human epithelial cells ( CCL-2 , ATCC ) were cultured in Dulbecco’s modified Eagle medium ( DMEM ) high glucose , GlutaMAX , pyruvate ( Gibco , Waltham , MA , USA ) supplemented with 10% heat-inactivated fetal bovine serum ( FBS , Sigma , St . Louis , MO , USA ) and 100 units/mL penicillin and 100 µg/mL streptomycin ( Gibco ) at 37°C and 5% CO2 . Cells were used from passages 2 to 20 . 1 × 105 HeLa cells were seeded ( for IF on a glass coverslip ) in a well of a 6-well plate on day 1 . Transfection of siRNA against endogenous SQSTM1/p62 ( sip62 ) or control siRNA ( siControl ) was performed on day 2 , followed by ( co- ) transfection of appropriate DNA constructs on day 4 . On day 5 , the assay was performed . In total , 50 pmol/well of ON-TARGETplus human SQSTM1/p62 siRNA ( J-010230-05 , Dharmacon , Buckinghamshire , UK ) or ON-TARGETplus nontargeting pool ( D-001810-10 , Dharmacon ) together with 2 . 5 µL Lipofectamine RNAiMax ( Invitrogen , Waltham , MA , USA ) was incubated with serum-free medium for 20 min at room temperature and added to HeLa cells in 2 mL culture medium . siRNA-resistant p62 variants in pmCherry-C1 with silent mutations ( forward nucleotide sequence: ORF 970GAgCAaATGGAaTCcGAc987 ) , full-length LC3B , full-length GABARAP , and/or mono-ubiquitin in pEGFP-C1 were used for transfection . 0 . 75 µg DNA for single transfection , 1 . 0 µg total DNA for co-transfection , or 2 µg BFP-ubiquitin cross-linked latex beads ( 2 µm ) were pre-incubated with FuGene6 ( Promega , Madison , WI , USA ) in a 1 µg:3 µL ratio ( DNA or beads:Fugene6 ) in serum-free medium . After 20 min at room temperature this mix was added to cells supplemented with fresh 2 mL ( DNA ) or 1 mL ( beads ) culture medium per well . Samples with beads were centrifuged at 175 g for 5 min at room temperature to settle down beads , followed by three washes with PBS after 1 hr and an additional centrifugation step . After another 3 hr , cells were washed once with PBS and fixed with 3% paraformaldehyde for 20 min at room temperature . siRNA and/or DNA transfected HeLa cells were washed once with PBS and lysed in 100 µL/well lysis buffer containing 20 mM Tris pH 8 . 0 , 10% glycerol , 135 mM NaCl , 0 . 5% NP-40 , and protease inhibitors ( Complete , EDTA-free , Roche ) for 15 min on ice . After scraping the cells off , lysates were centrifuged at 16 , 100 g for 5 min at 4°C to remove cell debris . In total , 150 µL of the lysis buffer without NP-40 or protease inhibitors ( wash buffer ) was added to the supernatant and this input was incubated with a mix of 2 µL GFP-TRAP_A beads ( ChromoTek ) and 8 µL empty Sepharose 4B beads ( Sigma ) , equilibrated in wash buffer , for 1 hr at 4°C . After washing the beads 3× with wash buffer , beads were taken up in Laemmli loading buffer , boiled for 10 min at 95°C , and loaded on a SDS-PAGE . Proteins were detected by western blotting . The mouse monoclonal anti-GST antibody ( clone 2H3-D10 , diluted 1:1000 ) is available from Sigma . The mouse anti-GFP antibody ( clones 7 . 1 and 13 . 1 , diluted 1:1000 to 1:5000 ) was purchased from Roche ( order number 11814460001 ) . The monoclonal anti-LC3B ( clone 2G6 , diluted 1:500 for immunoblotting or 1:100 for immunofluorescence ) is available from NanoTools , Teningen , Germany . The mouse monoclonal anti-p62 antibody ( diluted 1:1000 to 1:5000 for immunoblotting or 1:100 for immunofluorescence ) was purchased from BD Transduction Laboratories , Franklin Lakes , NJ , USA ( order number 610832 ) . The polyclonal BacTrace goat anti-Salmonella CSA-1 antibody ( diluted 1:200 ) was purchased from KPL , Gaithersburg , MD , USA ( order number 01-91-99 ) . The mouse anti-GAPDH ( clone GAPDH-71 . 1 , diluted 1:50 , 000 ) is available from Sigma . The rabbit anti-ubiquitin serum is available from Sigma . Secondary antibodies for immunofluorescence were Alexa Fluor 488 or 546-conjugated goat anti-mouse IgG ( diluted 1:1000 ) from Invitrogen , Alexa Fluor 647-conjugated goat anti-rabbit IgG ( diluted 1:500 ) and CyTM5-conjugated AffiniPure donkey anti-goat IgG ( diluted 1:400 ) from Jackson ImmunoResearch Laboratories , West Grove , PA , USA and Alexa Fluor 405-conjugated donkey anti goat IgG ( diluted 1:200 ) from Abcam , Cambridge , UK . After paraformaldehyde fixation , cells were washed 4× with PBS and permeabilized with 0 . 1% saponin ( AppliChem , Darmstadt , Germany ) in PBS ( washing buffer ) for 10 min at room temperature . After blocking with 5% BSA in washing buffer for 1 hr , cells were incubated with the primary antibody for 1 hr followed by three washes and the secondary antibody for another hour at room temperature . After three washes cells were mounted with Dapi fluoromount-G ( SouthernBioTech , Birmingham , AL , USA ) and observed on confocal LSM 710 or LSM 700 ( Zeiss , Jena , Germany ) microscopes . To distinguish internal from external beads or Salmonella , IF was first performed on non-permeabilized cells ( without saponin ) followed by IF on permeabilized cells and slides were mounted with ProLong Gold Antifade , Invitrogen . For endogenous LC3B detection , cells were fixed in -20°C cold methanol for 5 min at -20°C . S . typhimurium LT2 wild type was cultured in LB medium with additional 300 mM NaCl overnight at 37°C at 200 rpm . The next day the culture was diluted to OD600 = ∼0 . 2 and grown to OD600 = ∼0 . 9 . siRNA and DNA pre-transfected HeLa cells were washed 2× with PBS and transferred into culture medium without antibiotics at least 2 hr prior to infection . HeLa cells of one well of a 6-well plate were counted and the required inoculum of Salmonella was determined for MOI 50 assuming that OD600 = 0 . 9 corresponds to 1 × 109 Salmonella/mL . The inoculum was added to 1 mL culture medium without antibiotics/well and spun down at 300 g for 5 min at room temperature to synchronize infection . After half an hour of incubation at 37°C and 5% CO2 , HeLa cells were washed 3× with PBS and DMEM with 10% FBS and 100 µg/mL gentamicin ( Sigma ) was added ( time point 0 hr post infection ) . After 1 hr , HeLa cells were washed 2× with PBS and fixed with 3% paraformaldehyde in PBS for 20 min at room temperature . Salmonella were stained with 4' , 6-diamidino-2-phenylindole ( DAPI ) or the anti-Salmonella antibody and imaged using confocal LSM 710 ( Zeiss ) or LSM 700 ( Zeiss ) microscopes . For quantification of protein recruitment to GUVs or beads , one line was drawn across each GUV/bead so that contact points between GUVs/beads as well as protein aggregates would be excluded . The average brightness of an empty portion of each picture was considered as the background for that picture ( Bkg ) . For each line drawn , the protein binding intensity was calculated as the result of the difference ( Max - Bkg ) , where Max denotes the maximal brightness across the line . Where the intensity of two fluorescent proteins was measured , the recruitment of each individual protein was calculated as described . Each bead/GUV was quantified in the same position for both proteins . The recruitment ratio of the two proteins was then calculated dividing the recruitment of the prey protein by the recruitment of the bait protein at the same position for each GUV/bead . For fluorescence decay experiments , at least three fields were acquired per each individual sample . Every field was imaged as a Z-stack spanning all the beads contained in it . Time points were taken every 1 . 25 , 2 . 5 , or 5 min over a total time of 90 min . For quantification , Z-stacks corresponding to the shown time points were projected in single pictures as maximal Z projection ( ImageJ ) ; pictures were assembled in time-lapse stacks and the same positions in every slice were quantified as described . Fluorescence intensities of each bead at every time point were related to respective initial intensities at time point 0 as 100% . FRAP curves were quantified measuring maximal fluorescence intensity at the bead’s rim in the bleached region . The pre-bleaching value was set to 100% and first the post-bleaching time point to 0% . Recovery ( r ) at any following time point ( ix ) was calculated as a fraction of pre-bleach minus post-bleach delta , that is , rix = ( ix – i0% ) / ( i100% – i0% ) *100 . For quantification of membrane bending , only contact points between beads and GUVs were considered . Bending was scored when the membrane was seen deflected or interrupted in correspondence of the bead . Recruitment of mCherry-p62 variants and/or eGFP-LC3Bto beads in HeLa cells was determined by considering only internalized beads ( negative for extracellular anti-ubiquitin antibody signal , but positive for BFP-ubiquitin ) in mCherry-p62 and eGFP-LC3B co-expressing cells , with endogenous p62 knockdown . Puncta to ring structures showing the proteins directly at the beads were counted as positive localization to beads . For quantification of protein recruitment at internalized bacteria , values are expressed as % of positive bacteria for the indicated protein over the total internal bacteria detected . Unless differently stated , for all statistical analyses a two-tailed , unpaired Student’s t-tests were performed .	Cells use a process called autophagy to destroy damaged proteins and other harmful materials . During autophagy , the harmful materials become enclosed in a compartment called an autophagosome , which seals it off from the rest of the cell . The autophagosome – which is made of a double-layered membrane – then fuses with another compartment called a lysosome , which contains enzymes capable of breaking down the harmful material . Autophagy helps to keep cells healthy and defects in this process can contribute to cancer , neurodegeneration and other serious diseases in humans . It is important that cells are able to correctly select harmful materials for autophagy . A protein called p62 is able to specifically bind to damaged proteins that have been covered with a molecule called ubiquitin . It also interacts with proteins of the ATG8 family that are found on the surface of the developing autophagosome . In this way , p62 operates as an adaptor that brings damaged proteins into contact with autophagosomes in preparation for autophagy . However , since a single p62 molecule can only interact with one ubiquitin molecule and one ATG8 protein , it is not clear how it is able to specifically select only the proteins that carry many ubiquitin tags . Individual p62 units can interact with each other to form a group called an oligomer . Wurzer , Zaffagnini et al . use a combination of biochemical and cell biological techniques to study these oligomers in human cells . The experiments show that the oligomers are able to bind to many ubiquitin tags on a single structure . This enables a p62 oligomer to form a stronger connection to the damaged protein than a single p62 unit can . At the same time , the oligomer can interact with many ATG8 proteins , which tend to be found in clusters only on the surface of the autophagosome . Previous studies have shown that when an autophagosome starts to form , its membrane expands and curves around the cargo . Wurzer , Zaffagnini et al . observed that the binding of p62 oligomers to ATG8 proteins and damaged structures with ubiquitin tags , drive the bending of the membrane around these structures . Wurzer , Zaffagnini et al . ’s findings reveal how the formation of oligomers allows p62 to specifically select and target damaged proteins that are covered with many ubiquitin tags for destruction , while sparing other materials . Other proteins that are closely related to p62 also help to select cell materials for autophagy , so a future challenge is to find out whether these proteins work in a similar way .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "biochemistry", "and", "chemical", "biology", "cell", "biology" ]	2015	Oligomerization of p62 allows for selection of ubiquitinated cargo and isolation membrane during selective autophagy
The Neurotransmitter:Sodium Symporters ( NSSs ) represent an important class of proteins mediating sodium-dependent uptake of neurotransmitters from the extracellular space . The substrate binding stoichiometry of the bacterial NSS protein , LeuT , and thus the principal transport mechanism , has been heavily debated . Here we used solid state NMR to specifically characterize the bound leucine ligand and probe the number of binding sites in LeuT . We were able to produce high-quality NMR spectra of substrate bound to microcrystalline LeuT samples and identify one set of sodium-dependent substrate-specific chemical shifts . Furthermore , our data show that the binding site mutants F253A and L400S , which probe the major S1 binding site and the proposed S2 binding site , respectively , retain sodium-dependent substrate binding in the S1 site similar to the wild-type protein . We conclude that under our experimental conditions there is only one detectable leucine molecule bound to LeuT . The Neurotransmitter:Sodium Symporters ( NSSs ) are responsible for clearing neurotransmitters , such as dopamine , serotonin , norepinephrine , glycine and GABA from the synaptic cleft . The transporters are thereby crucial for the regulation of synaptic transmission in the CNS and alterations in their function have been linked to several psychiatric and neurological disorders such as depression , bipolar disorders , attention deficit hyperactive disorder ( ADHD ) , epilepsy , and Parkinson’s disease ( Broer , 2013; Kristensen et al . , 2011 ) . The understanding of the molecular mechanisms and structural ( re ) arrangements underlying NSS function has advanced significantly in recent years . The most detailed insight into structure-function relationships of NSSs comes from studies of the amino acid transporter , LeuT , from Aquifex aeolicus ( Kantcheva et al . , 2013; Kazmier et al . , 2014; Malinauskaite et al . , 2014 , 2016; Piscitelli et al . , 2010; Quick et al . , 2012; Shi et al . , 2008; Singh et al . , 2007; Wang et al . , 2012a , 2012b; Yamashita et al . , 2005 ) . Recent structures of the drosophila dopamine transporter ( dDAT ) ( Penmatsa et al . , 2013 ) and the human serotonin transporter ( Coleman et al . , 2016 ) , which are eukaryotic members of the NSS family , confirm that LeuT is a reliable model protein and proves its value in understanding the molecular function of this class of transporters . Functional studies of LeuT have suggested the existence of a secondary substrate binding site ( S2 ) located in the extracellular vestibule of LeuT approximately 10 Å from the primary substrate binding site ( S1 ) ( Khelashvili et al . , 2013; Quick et al . , 2009; Shi et al . , 2008 ) . The S2 site is suggested to be an allosteric trigger , essential for coupling the energy from the electrochemical gradient to the transport of the solute . The binding of leucine to the S2 site has been measured to have the same affinity ( in nM range ) as binding to the S1 but does not , as the S1 bound substrate , directly coordinate sodium ( Quick et al . , 2012 ) . However , attempts to crystallize LeuT with substrate bound to the S2 site have so far been unsuccessful , and therefore the existence of the S2 site is supported primarily by radioligand binding assays and guided MD simulations ( Quick et al . , 2012; Zhao et al . , 2011 ) . Due to the lack of structural evidence , the existence of a high-affinity S2 site has been questioned ( Piscitelli et al . , 2010 ) , supporting the need for employing new techniques for investigating ligand binding in NSS proteins . Here we investigate the leucine binding properties of LeuT by magic angle spinning ( MAS ) NMR , aiming at a characterization of the proposed S2 binding site . Our approach offers several advantages: ( i ) We use microcrystalline samples of LeuT prepared under experimental conditions allowing for conformations capable of ligand binding to both S1 and S2 ( Quick et al . , 2012 ) . ( ii ) NMR offers information on the full structural ensemble which is unlikely not to include conformers ( even lowly populated ) prone to bind leucine in S2 . ( iii ) Leucine binding to S1 and S2 may be distinguished by characteristic chemical shifts that are expected to be different due to different chemical environments , i . e . interacting residues ( Reyes et al . , 2011 ) . Prior to crystallization and NMR experiments we verified the functionality of the produced LeuT wild-type ( WT ) samples . We initially performed [3H]leucine saturation binding experiments and subsequently assessed Na+-dependency of [3H]leucine binding . All experiments were done at a DDM concentration commonly used in in vitro assays ( i . e . , 0 . 05% corresponding to 5 . 7x CMC ) . At this detergent concentration , LeuT was reported to retain binding to both S1 and the putative S2 site ( Quick et al . , 2012 ) . In scintillation proximity binding assays LeuT WT bound [3H]leucine with a dissociation constant ( Kd ) of 12 ± 1 nM in the presence of 200 mM sodium . The EC50 value calculated for the Na+-dependent binding was 47 ± 4 mM ( Figure 1—figure supplement 1A–B ) . These values are in agreement with those previously reported for LeuT ( Shi et al . , 2008; Singh et al . , 2008 ) . As we were primarily interested in a simple readout reporting solely on substrate binding , we purified and kept LeuT in the presence of 1 mM 15N enriched L-leucine to ensure substrate binding and detection in both sites ( Figure 1A ) . With a 15N natural abundance around 0 . 3% , the background from the protein amides and amines is sufficiently low to distinguish even weakly populated states originating from the enriched substrate only . 10 . 7554/eLife . 19314 . 003Figure 1 . Assessment of L-leucine binding to LeuT WT by solid state NMR . ( A ) Cartoon illustration of experimental approach . 15N enriched L-leucine substrate is added to detergent reconstituted LeuT , which is subsequently crystallized using large scale sitting drop vapour diffusion . Rod-shaped microcrystals form within 24 hr and can be readily harvested . ( PDB ID: 3F3E ) ( B ) LeuT WT purified in NaCl ( red ) and LeuT purified in KCl ( black ) . 15N L-Leucine specific peak is indicated by an asterix with a chemical shift of 38 . 2 ppm . Spectra are tentatively intensity normalized to the 15N natural abundance signal from the LeuT backbone amides . Signal-to-noise is calculated to be 21 . ( C ) 23Na-NMR of LeuT WT ( red ) and LeuT WT in KCl ( black ) in presence of leucine . Minor peak at −8 . 9 ppm represents the shape of one or two structural sodium molecules . Despite inequivalent location of the two sodium sites in the LeuT , the coordination mechanism is almost identical which might account for the observation of a single peak in the 23Na-NMR spectrum instead of two distinct peaks . DOI: http://dx . doi . org/10 . 7554/eLife . 19314 . 00310 . 7554/eLife . 19314 . 004Figure 1—figure supplement 1 . Functional characterization of LeuT WT . LeuT WT was purified in KCl and eluted in 0 . 05% DDM ( pH 8 . 0 ) . Scintillation proximity assay-based measurements of ( A ) [3H]leucine saturation binding to 100 ng LeuT in the presence of 200 mM NaCl and ( B ) Na+-dependent [3H]leucine binding ( 100 nM ) by 100 ng LeuT . Ionic strength was compensated with KCl . Data are displayed as means ± s . e . m . , performed in triplicates , n = 3 . DOI: http://dx . doi . org/10 . 7554/eLife . 19314 . 00410 . 7554/eLife . 19314 . 005Figure 1—figure supplement 2 . Microscopy image of LeuT microcrystals . LeuT microcrystals are needle-shaped and have a length of 1–10 μm . The microscopic image was solely used to assess the quality of the microcrystalline material . DOI: http://dx . doi . org/10 . 7554/eLife . 19314 . 00510 . 7554/eLife . 19314 . 006Figure 1—figure supplement 3 . 1D 15N CP/MAS spectrum of frozen and lyophilized LeuT WT samples . ( A ) Protein concentration: 2 mg/ml , leucine concentration: 1 mM . The spectrum is dominated by the signal arising from the free unbound leucine , at ~42 ppm . Spectrum was recorded on a 400 MHz Bruker shielded wide bore magnet equipped with a 3 . 2 mm MAS HCN operating at ~100 K . Spinning rate: 8000 kHz . CP contact time: 1000 us , recycling delay: 3 s . 24 k scans were required to record presented spectrum . ( B ) 15N CP/MAS spectra from dry L-leucine powder ( top panel ) and from lyophilized leuT samples ( bottom panel ) . Free 15N L-Leucine has a distinct chemical shift at 118 ppm . For the LeuT WT NaCl preparation ( red ) several additional peaks appear around the dominating free state peak . Though these peaks vary slightly in intensity when compared to LeuT WT KCl ( dark ) . These peaks do not originate from structural leucine . Spectra were recorded on a 700 MHz Bruker shielded wide bore magnet equipped with a 4 mm MAS HCN probe operating at 298 K . Spinning rate: 12500 KHz , CP contact time: 1500 us , recycling delay: 2 . 5 s . 65 K scans were required to record presented spectra . DOI: http://dx . doi . org/10 . 7554/eLife . 19314 . 00610 . 7554/eLife . 19314 . 007Figure 1—figure supplement 4 . 15N L-leucine spectrum substrate peak for LeuT WT . ( A ) Line broadening for window function 1 Hz , Signal-to-noise in calculated to be 21 , Full width half height ( FWHH ) of the substrate peak is 31 Hz . ( B ) Line broadening for window function 10 Hz . DOI: http://dx . doi . org/10 . 7554/eLife . 19314 . 00710 . 7554/eLife . 19314 . 008Figure 1—figure supplement 5 . In-solution 1D 15N spectra of free 98% 15N L-leucine at different pH . ( A ) pH titration of the L-leucine amine . The 1D 15N spectra are recorded for 1024 scans at 25°C . ( B ) 15N chemical shift of the L-leucine amine as a function of pH . From the sigmoidal curve fit the pI is estimated to be 9 . 72 ± 0 . 06 . DOI: http://dx . doi . org/10 . 7554/eLife . 19314 . 00810 . 7554/eLife . 19314 . 009Figure 1—figure supplement 6 . 1D 13C CP/MAS spectrum of microcrystalline LeuT WT samples . Spectra for samples prepared in NaCl or KCl are colored in red and black , respectively . Resonances originating from labelled leucine are indicated by their chemical shifts and their respective assignments . These results are in good agreement with the 15N 1D spectra , and only one set of chemical shifts from the ligand can be observed . DOI: http://dx . doi . org/10 . 7554/eLife . 19314 . 009 To achieve sufficiently narrow line widths of the NMR signals and to avoid any signal from unbound leucine , we produced microcrystalline samples of LeuT ( Figure 1—figure supplement 2 ) , and performed cross polarization ( CP ) -based NMR experiments at temperatures above the freezing point . In all other preparations tested ( frozen , lyophilized and proteoliposomes ) the signal from the unspecific or unbound leucine completely dominated the spectra ( Figure 1—figure supplement 3A–B ) . Using the microcrystalline preparations , we were able to produce the high quality CP-based 15N detected spectra showing one significant ( above 2σ – Figure 1—figure supplement 4 ) peak at 38 . 2 ppm that could be assigned to the amine of protein bound leucine ( Figure 1B ) . In addition to the sharp signal from leucine , much broader signals between 110 and 130 ppm were also observed , which originate from the 15N natural abundance of the LeuT amides ( Figure 1B ) . To further assess whether the intense signal at 38 . 2 ppm reflects sodium specific leucine binding to LeuT , we performed a parallel experiment substituting Na+ with K+ . Sodium is required for leucine binding ( Zhao et al . , 2011 ) . By the use of 23Na-NMR we confirmed the presence of only a negligible amount of residual NaCl ( at 7 . 1 ppm ) , and that no detectable Na+ was coordinated in the protein ( Figure 1C ) . In the absence of Na+ , the signal at 38 . 2 ppm in the 15N 1D spectrum disappeared as expected for a signal originating from 15N-leucine bound to LeuT ( Figure 1B ) . Worth of note , the amine NH3+ group of free leucine has a chemical shift of approximately 41 ppm at pH 8 ( Figure 1—figure supplement 5A–B ) , demonstrating that the bound substrate resides in a not fully solvent accessible environment . Importantly , we were unable to detect any signal from any additionally bound leucine . Similarly , the 13C CP/MAS spectra from the same samples clearly displayed only one single set of sodium dependent leucine signals ( Figure 1—figure supplement 6 ) . To investigate whether the origin of the substrate peak at 38 . 2 ppm was due to leucine binding either to the S1 or the S2 site , we recorded solid state NMR spectra of two variants with compromised leucine binding , F253A ( S1 ) and L400S ( S2 ) ( Figure 2A , B ) . For these experiments we lowered the final concentration of the added enriched substrate to 5 μM to ensure proper detimental effect by the mutations . This concentration has previously been reported not to provide any detectable [3H]leucine occupancy in the S2 site of LeuT L400S , but saturated S1 binding ( Quick et al . , 2012 ) . For LeuT WT the specific leucine peak was unaffected by lowering the free leucine concentration ( Figure 2—figure supplement 1 ) . The F253A mutant has previously been shown to impair binding to the S1 site ( Billesbølle et al . , 2015; Wang et al . , 2012b ) . Thus , F253A serves as a S1 disturbing mutant at low substrate concentrations . In the F253A 1D 15N spectrum , we observed sodium dependent substrate binding with a chemical shift of 38 . 4 ppm and a slightly lower signal intensity , when compared to the WT spectra ( Figure 2C ) . The shift in F253A was consistent for both high ( 1 mM ) and low ( 5 µM ) leucine concentrations ( Figure 2—figure supplement 3 ) . Most importantly , the chemical shift difference of ~0 . 2 ppm for the observed bound leucine peak , demonstrates that the ligand is affected by the local environment of the S1 binding site . The L400S mutation was previously suggested to abolish S2 leucine binding ( Quick et al . , 2012 ) . The 15N spectrum obtained for leucine bound to L400S ( Figure 2C , blue ) completely resembled the spectrum obtained with WT protein ( Figure 2C , red ) . We were not able to detect any change in intensity or chemical shift . This argues against the possibility that the substrate signal we observe in LeuT WT samples is reporting on a combination of S1 and S2 binding . Also , we reason that it would be highly unlikely for an S1 bound substrate and a putative S2 site-bound substrate to have the same chemical shift as the environment of the putative S2 binding site markedly differs from the S1 binding site . As a major difference , the proposed S2 binding site does not involve direct sodium binding ( Shi et al . , 2008 ) 10 . 7554/eLife . 19314 . 010Figure 2 . Effects of S1 and S2 site mutations on the L-leucine chemical shift . ( A–B ) Cartoon representation displaying the location of F253 in the S1 site and L400 in the proposed S2 site based on PDB: 3USG ( Wang et al . , 2012a ) . ( C ) 15N 1D NMR spectrum of LeuT WT ( red ) , F253A ( green ) and L400S ( blue ) . Inset: Close-up of L-leucine specific peak . Spectra are tentatively intensity normalized to the 15N natural abundance signal from the LeuT amides . DOI: http://dx . doi . org/10 . 7554/eLife . 19314 . 01010 . 7554/eLife . 19314 . 011Figure 2—figure supplement 1 . Comparison of spectra derived from LeuT WT purified and crystallized in either 1 mM ( red ) or 5 uM ( purple ) free leucine . Spectra are tentatively normalized to the LeuT natural abundance amide signal . As expected the substrate peak relates to LeuT concentration and not to the added free leucine concentration . DOI: http://dx . doi . org/10 . 7554/eLife . 19314 . 01110 . 7554/eLife . 19314 . 012Figure 2—figure supplement 2 . Power spectra of LeuT WT ( red ) , F253A ( green ) and L400S ( blue ) . To rule out that wrong phasing would cause the chemical shift difference observed for the substrate peak of the spectra presented in Figure 2C , we present the them also as power spectra ( which is ultimately a squared magnitude spectra ) where phase signs are not preserved . We have intensified the F253A signal to only demonstrate the chemical shift difference . DOI: http://dx . doi . org/10 . 7554/eLife . 19314 . 01210 . 7554/eLife . 19314 . 013Figure 2—figure supplement 3 . Comparison of LeuT WT and LeuT F253A in the presence of 1 mM free substrate . ( A ) Full spectrum displayed with a line broadening of the windows function of 100 Hz . Tentatively intensity normalized using natural abundance signals . ( B and C ) Close up of the substrate peak region showing preserved chemical shift difference between the substrate bound in LeuT WT ( black dashed line ) and LeuT F253A ( blue dashed line ) in two different free substrate concentrations . DOI: http://dx . doi . org/10 . 7554/eLife . 19314 . 01310 . 7554/eLife . 19314 . 014Figure 2—figure supplement 4 . Cartoon representation of S1 bound substrate ( green ) . Sodium ions are depicted in blue . Measured chemical shift values in red ( in ppm ) and measured distances in black ( in Å ) . Image is made from PDB file: 3F3E . DOI: http://dx . doi . org/10 . 7554/eLife . 19314 . 014 In conclusion , although all LeuT samples used in the present study were prepared in DDM at low concentration to exclude previously reported detrimental effects on the S2 binding site , we were only able to identify one single substrate signal at 38 . 2 ppm in the 15N spectra and one set of signals in the 13C spectrum ( Figure 2—figure supplement 2 ) . We note , however , that at our current signal-to-noise ratio ( ~20 ) , we would not be able to detect species populated less than 5% of the structural ensemble . Based on the minor change in chemical shift in the F253A ( S1 ) mutant , and the completely unaltered signal for the L400S ( S2 ) mutant we reason that the observable bound leucine is located at the S1 binding site , thus supporting the idea that LeuT exhibit one single central binding site . We cannot exclude that the detection of S2 binding may only be possible upon the complete transition of the transporter towards a specific ( yet unknown ) conformation or that unfavourable crystal contacts might complicate S2 binding . However , several crystal structures have shown the binding of antidepressants , which overlaps with the putative S2 site , using these exact conditions ( Singh et al . , 2007; Zhou et al . , 2009 ) . As previously proposed for LeuT ( Piscitelli et al . , 2010 ) we speculate that the S2 substrate binding site , if present , is rather a transient site , responsible for optimal functionality of the transporter . Expression of LeuT WT from Aquifex aeolicus was performed according to the protocol described previously ( Billesbølle et al . , 2015 ) . LeuT WT was expressed in E . coli C41 ( DE3 ) transformed with pET16b encoding C-terminally 8xHis-tagged transporter ( expression plasmid was kindly provided by Dr E . Gouaux , Vollum Institute , Portland , Oregon , USA ) . Briefly , isolated bacterial membranes were solubilized in 1% DDM ( Anatrace , USA ) in the presence of 1 mM 98% 15N-L-Leucine ( Cambridge isotopes , Tewksbury , MA ) and the protein was bound to nickel-charged affinity resin ( Life Technologies , Carlsbad , CA ) . Subsequently , protein was eluted in 20 mM Tris-HCl ( pH 8 . 0 ) , 200 mM KCl , 0 . 05% DDM , 1 mM 15N-L-Leucine and 300 mM imidazole ( KCl sample ) or in the same buffer containing NaCl instead of KCl ( NaCl sample ) . LeuT F253A and L400S variants were generated from the leuT gene using a QuikChange kit ( Agilent Technologies , Santa Clara , CA ) and purified similarly to the LeuT WT protein with the difference that 5 µM of 13C-15N-L-Leucine ( Cambridge Isotopes ) was used for co-purification , and the salt content in all buffers consisted of 50 mM NaCl and 150 KCl . The LeuT F253A variant in the presence of 1 mM substrate was prepared similar to the NaCl sample . Subsequently , all LeuT samples were dialyzed for approx . 36 hr at 4°C in the respective elution buffer without imidazole . Functional characterization of the LeuT WT purified in KCl was performed using a scintillation proximity assay ( SPA ) ( Quick and Javitch , 2007 ) . Saturation binding of [3H]leucine ( 50 . 2 Ci/mmol; PerkinElmer , Waltham , MA ) to purified LeuT WT was performed with 100 ng/well ( 1 . 66 pmol ) of protein in buffer composed of 20 mM Tris-HCl ( pH 8 . 0 ) , 200 mM NaCl , 0 . 05% DDM , 20% glycerol in the presence of 1 . 25 mg/ml copper chelate ( His-Tag ) YSi beads ( PerkinElmer ) . Sodium-dependency was measured at fixed [3H]leucine concentration of 100 nM with increasing concentrations of NaCl ( NaCl was substituted with KCl for equal ionic strength ) again using 100 ng/well LeuT WT . [3H]Leucine binding was monitored using MicroBeta liquid scintillation counter ( PerkinElmer ) and data were fitted to a one-site saturation or dose-response function , respectively , using Prism 7 software ( GraphPad , San Diego , CA ) . Microcrystalline samples were produced by large scale sitting drop vapour diffusion method . 1 mL of the protein sample solutions were mixed 1:1 with the crystallization buffer composed of 100 mM NaCl ( or KCl ) , 120 mM MgCl2 , 28% PEG400 , 100 mM MES or HEPES pH 6 . 5 . The crystallization was carried out at 18°C . After approximately 20 hr a white precipitate could be harvested by centrifugation and transferred directly to the rotor . All samples were freshly prepared immediately before use . Microcrystals where visualized using a Leica M125 microscope with a 1 . 0x PlanApo objective . Protein for lyophilisation was depleted of glycerol during dialysis , snap-freezed in liquid nitrogen and added to a freeze drier . The remaining powder could be transferred directly to the rotor . 15N-L-Leucine was dissolved to a final concentration of 1 mM in the following buffer: 20 mM Tris-HCl ( pH 8 . 0 ) , 200 mM KCl , 0 . 05% DDM and added to an Economy WG5 NMR tube . The Experiment was run on a Bruker Avance III 500 MHz operating at a Larmor frequency of 50 . 667 MHz for 15N . The directly detected 15N spectra were recorded using a recovery delay of 2 s and an acquisition time of 500 ms . The total number of 1024 scans were used . Microcrystalline LeuT nitrogen spectra were recorded on Bruker Avance III 800 MHz wide bore ( 89 mm ) spectrometer equipped with a 4 mm MAS HCN efree probe . Spectra were obtained at 275 K ( measured temperature ) , at 12500 Hz magic-angle-spinning . 15N CP/MAS experiments were run for 60 K scans in blocks of 10 K scans and the magnet was fine-tuned between each block . Cross-polarization contact time was set to 1750 us . Initial recovery delay was set to 3 s . Protons were decoupled at 86 kHz during acquisition . 13C CP/MAS experiments were run for 2 K scans . Cross-polarization contact time was set to 2000 us . The initial recovery delay was set to 3 s . Spectra were displayed using a 1 , 10 or 100 Hz line broadening for EM window function in topspin . Sodium MAS NMR spectra were recorded on a Bruker Avance NMR spectrometer operating at a Larmor frequency of 105 . 8 MHz for 23Na using a double resonance probe equipped for 4 mm ( o . d . ) rotors . All spectra were recorded at room temperature employing a central transition selective 90 degree pulse ( 1 . 8 μs ) , a recycle delay of 2 s , an acquisition time of 40 . 9 ms , a spectral width of 75 . 19 kHz and a spin rate of either 9 or 10 kHz . The spectra are referenced to crystalline NaCl at 7 . 1 ppm .	All living cells need amino acids – the building blocks of proteins – in order to survive , yet few cells can make all the amino acids that they need . Instead , transporter proteins in cell membranes must take these molecules from the outside of the cell and release them to the inside . Some cells , including those in the brain , also release amino acids and molecules derived from them into the spaces outside of the cell to send signals to other nearby cells . Again , transporter proteins must move these signaling molecules back inside cells , to stop the signaling and to allow the molecules to be recycled . Importantly , problems with these uptake mechanisms have been linked to disorders such as depression , epilepsy and Parkinson’s disease . One family of transporters involved in the uptake of amino acids are the “Neurotransmitter:Sodium Symporters” . Though these proteins are involved in processes that are fundamental to life , it remains unclear exactly how they work . Specifically , it has been heavily debated whether this family of transporters require one or two amino acid molecules to bind at the same time in order to help transport them across the membrane . Now Erlendsson , Gotfryd et al . have analyzed a bacterial protein in the Neurotransmitter:Sodium Symporter family . This transporter takes up an amino acid called leucine into cells , and is commonly used as a model to understand this family of transporter proteins more generally . Using a technique called solid state nuclear magnetic resonance , Erlendsson , Gotfryd et al . could detect a single molecule of leucine bound to each transporter , but not a second one . This technique could also pinpoint that the leucine was located at the transporter’s central binding site . Leucine was never found at the proposed secondary binding site . Together these findings suggest that only one molecule of leucine binds to the transporter at any one time , and that it binds to the transporter’s central binding site . Erlendsson , Gotfryd et al . have shown now how solid state nuclear magnetic resonance can be used to explore in detail how Neurotransmitter:Sodium Symporters move molecules across cell membranes . The next challenge is to use the same experimental setup to characterize other Neurotransmitter:Sodium Symporters . Doing so could potentially lay the groundwork for designing more specific and improved drugs to treat disorders like depression and Parkinson’s disease .	[ "Abstract", "Introduction", "Results", "and", "discussion", "Materials", "and", "methods" ]	[ "short", "report", "structural", "biology", "and", "molecular", "biophysics", "neuroscience" ]	2017	Direct assessment of substrate binding to the Neurotransmitter:Sodium Symporter LeuT by solid state NMR
Cytokinesis requires activation of the GTPase RhoA . ECT-2 , the exchange factor responsible for RhoA activation , is regulated to ensure spatiotemporal control of contractile ring assembly . Centralspindlin , composed of the Rho family GTPase-activating protein ( RhoGAP ) MgcRacGAP/CYK-4 and the kinesin MKLP1/ZEN-4 , is known to activate ECT-2 , but the underlying mechanism is not understood . We report that ECT-2-mediated RhoA activation depends on the ability of CYK-4 to localize to the plasma membrane , bind RhoA , and promote GTP hydrolysis by RhoA . Defects resulting from loss of CYK-4 RhoGAP activity can be rescued by activating mutations in ECT-2 or depletion of RGA-3/4 , which functions as a conventional RhoGAP for RhoA . Consistent with CYK-4 RhoGAP activity contributing to GEF activation , the catalytic domains of CYK-4 and ECT-2 directly interact . Thus , counterintuitively , CYK-4 RhoGAP activity promotes RhoA activation . We propose that the most active form of the cytokinetic RhoGEF involves complex formation between ECT-2 , centralspindlin and RhoA . Cell division involves the myosin-mediated contraction of an actin-based contractile ring . In metazoans , contractile ring assembly involves activation of the GTPase RhoA . RhoA directly activates formin-mediated actin polymerization and indirectly promotes myosin activation . As the contractile ring must assemble at the correct position and at the correct time , between the segregating chromosomes in anaphase , RhoA activation is subject to multiple regulatory mechanisms ( see Green et al . , 2012 for review ) . The primary direct activator of RhoA during cytokinesis is the RhoGEF ECT-2 . ECT-2 contains N-terminal BRCT domains and a C-terminal RhoGEF domain ( Kim et al . , 2005; Zou et al . , 2014 ) . Presumably as a consequence of being autoinhibited , ECT-2 function depends on activators . One of the important activators of ECT-2 during cytokinesis is the centralspindlin complex , which is a heterotetramer containing a dimeric kinesin , ZEN-4 ( aka MKLP1 , Pavarotti ) and dimeric CYK-4 ( aka MgcRacGAP , Tum/RacGAP50C ) ( Mishima et al . , 2002 ) . Centralspindlin organizes the spindle midzone and directly recruits numerous regulators of cytokinesis , including ECT-2 , to this location ( Burkard et al . , 2009; Wolfe et al . , 2009 ) . ECT-2 and centralspindlin are conserved among—and restricted to—metazoans ( Frédéric et al . , 2013 ) ( unpublished results ) . Though these proteins are conserved , their names are distinct in each organism . For simplicity , Caenorhabditis elegans names will be used throughout this manuscript with the exception that we will refer to RHO-1 with the more common name RhoA . Recruitment of ECT-2 to the spindle midzone involves regulated binding between ECT-2 and CYK-4 . The BRCT domains of ECT-2 bind to CYK-4 phosphorylated by PLK-1 ( Burkard et al . , 2009; Wolfe et al . , 2009 ) . CYK-4 phosphorylation occurs in a cell cycle and microtubule-regulated manner . Furthermore , CDK-1 phosphorylation of ECT-2 inhibits the ECT-2-CYK-4 interaction during metaphase and inactivates a membrane binding motif within ECT-2 ( Yüce et al . , 2005; Su et al . , 2011 ) . The phosphorylation-dependent interaction between ECT-2 and centralspindlin is required for RhoA activation during cytokinesis in human cells ( Burkard et al . , 2007; Wolfe et al . , 2009 ) . Centralspindlin also localizes in trace , but biologically relevant , amounts on the cell membrane . Centralspindlin accumulation to the midzone and the membrane are independently regulated by its oligomerization , which is inhibited by a 14-3-3 protein and promoted by the chromosome passenger complex ( Douglas et al . , 2010; Basant et al . , 2015 ) . The CYK-4 subunit of centralspindlin contains an evolutionarily conserved Rho family GTPase-activating protein ( RhoGAP ) domain ( Jantsch-Plunger et al . , 2000 ) . The function of this domain has been examined in a number of different contexts . In vitro , the GAP domain of CYK-4 efficiently activates the GTPase activity of the Rho-family GTPases , CED-10/Rac1 and CDC-42 . CYK-4 also has GAP activity towards RhoA , but it is far less active towards RhoA as compared to Rac1 and CDC-42 ( Touré et al . , 1998; Jantsch-Plunger et al . , 2000; Bastos et al . , 2012 ) . Despite extensive effort , there is no consensus for the biological role of this GAP activity ( see White and Glotzer , 2012 for review ) . In some cell types , the GAP activity appears dispensable ( Goldstein et al . , 2005; Yamada et al . , 2006 ) , in others it appears to be important to negatively regulate Rac1 ( D'Avino et al . , 2004; Canman et al . , 2008; Bastos et al . , 2012 ) , whereas in yet others it appears to promote RhoA activation ( D'Avino et al . , 2004; Zavortink et al . , 2005; Loria et al . , 2012 ) . System-specific differences may underlie some of these diverse results . However , these studies differ in the mutations used to assess the function of the GAP domain , which is likely to affect the results . In addition , some of the controversy may be due to misinterpretation of indirect effects . The function of the RhoGAP domain has been examined in C . elegans embryos in some detail . These studies have focused largely on a temperature-sensitive , separation-of-function substitution mutation , E448K , that lies in the RhoGAP domain of CYK-4 , cyk-4 ( or749ts ) ( Canman et al . , 2008 ) . The mutant protein can complex with the centralspindlin kinesin , ZEN-4 , and bundle microtubules in the central spindle . However , cytokinesis does not proceed to completion in these embryos . Interestingly , depletion of CED-10/Rac1 or the actin nucleator subunit ARP-2 enables these embryos to complete cytokinesis ( Canman et al . , 2008 ) . These genetic interactions have been interpreted to indicate that the GAP domain of CYK-4 is important to keep CED-10/Rac1 inactive and prevent the accumulation of branched actin in the equatorial region ( Canman et al . , 2008 ) . However , subsequent analysis demonstrated that cyk-4 ( or749ts ) ; ced-10 ( − ) embryos are phenotypically abnormal ( Loria et al . , 2012 ) . In particular , cyk-4; ced-10 mutant embryos have reduced accumulation of RhoA effectors as compared to ced-10 mutants alone ( Loria et al . , 2012 ) . These results suggest that this mutation in CYK-4 affects more than the RhoGAP activity of CYK-4 or that CED-10/Rac1 is not the relevant target of the GAP domain , or both . Analysis of centralspindlin function in C . elegans embryos is impeded by the existence of a second , parallel pathway that promotes RhoA activation in the early embryo . Upon fertilization , embryos exhibit RhoA-dependent contractility that promotes embryo polarization and culminates in the formation of a transient furrow known as the pseudocleavage furrow . This wave of contractility requires a poorly conserved protein known as NOP-1 and is largely independent of centralspindlin ( Tse et al . , 2012 ) . Cytokinetic contractility , on the other hand , involves both NOP-1 and centralspindlin . However , nop-1 is nonessential; loss of function mutants are viable and fertile ( Rose et al . , 1995 ) . Cytokinesis proceeds to completion in the absence of NOP-1 , although furrow initiation is slightly delayed and RhoA accumulates to lower levels at the cleavage furrow ( Tse et al . , 2012 ) . Due to its role in RhoA activation , polarization in NOP-1-deficient embryos is also delayed . Mutational inactivation of NOP-1 permits direct analysis of centralspindlin-dependent furrow formation . Notably , when NOP-1 is inactivated , cyk-4 ( or749ts ) embryos are completely defective in RhoA activation ( Tse et al . , 2012 ) . The CYK-4 GAP domain is adjacent to a C1 domain that mediates membrane localization of centralspindlin ( Lekomtsev et al . , 2012 ) . Given that the cyk-4 ( or749ts ) substitution renders the protein thermosensitive , it is possible that these phenotypes are not the sole consequence of loss of GAP activity; this mutation could affect other functions of the GAP domain , the adjacent C1 domain may also be affected . To clarify these issues , we performed a targeted structure-function analysis of the C1 and GAP domains of CYK-4 . We demonstrate that the cyk-4 ( or749ts ) allele indeed affects its ability to associate with the membrane and show that this activity contributes to RhoA activation . We further show that the active site of the GAP domain contributes to the accumulation of downstream effectors of RhoA and RhoA-dependent contractility . Furthermore , we find that the catalytic domains of CYK-4 and ECT-2 directly interact in vitro . Finally , we show that hypomorphic defects in CYK-4-mediated RhoA dependent contractility can be suppressed by either loss of the RhoGAP activity provided by RGA-3/4 or by either of two activating mutations in ECT-2 . These activating mutations in ect-2 rescue cyk-4 ( or749ts ) and GAP-deficient CYK-4 . Our results indicate that CYK-4 GAP activity is involved in ECT-2-mediated RhoA activation . We sought to conduct a structure-function analysis of CYK-4 to determine the individual contributions of the GAP and C1 domains of CYK-4 . We established a rescue assay based on single copy integrants of GFP-tagged CYK-4 transgenes driven by the cyk-4 promoter inserted at a defined locus in the C . elegans genome using Mos1-mediated integration ( Figure 1—figure supplement 1 ) ( Frøkjaer-Jensen et al . , 2012 ) . The transgenes were expressed at consistent levels ( Figure 1—figure supplement 2 ) . The transgenes were rendered resistant to an RNAi construct that could effectively deplete endogenous CYK-4 by targeting the 3′ UTR and portions of the coding sequence ( Figure 1—figure supplement 2 ) . By combining the appropriate mutant transgene with RNAi to specifically deplete endogenous CYK-4 , we obtained embryos that express CYK-4MUT . In this and all subsequent experiments , when a given variant is assayed , endogenous CYK-4 is depleted by RNAi; these will be referred to as cyk-4mut embryos . We first sought to validate the rescue assay , by assaying cyk-4E448K embryos and found that they closely phenocopy cyk-4 ( or749ts ) embryos in which endogenous CYK-4 has the E448K substitution ( Figure 1—figure supplement 3 ) . This phenocopy indicates functional depletion of endogenous CYK-4 . Additionally , a wild-type transgene was fully functional as it could complement a large deletion in CYK-4 to viability and fertility ( Figure 6—figure supplement 1 ) . Consistent with the C1 domain promoting membrane association , CYK-4∆C1 does not accumulate on ingressing cleavage furrows ( Figure 1A , B ) . As cyk-4 ( or749ts ) is temperature sensitive , we considered the possibility that the thermosensitivity also destabilizes the C1 domain that lies adjacent to the CYK-4 GAP domain ( Figure 1—figure supplement 1 ) . Indeed , at the restrictive temperature , CYK-4E448K exhibits a similar defect in localization as CYK-4∆C1 ( Figure 1A , B ) . CYK-4 also associates with the membrane in the germline and indirect evidence suggests that this localization is compromised in embryos expressing CYK-4E448K ( Zhou et al . , 2013 ) . We generated strains in which both the endogenous cyk-4 and the GFP-tagged transgene contained the E448K mutation . At the permissive temperature , CYK-4E448K membrane recruitment was readily detected in the germline , however this localization was lost as animals were shifted to the restrictive temperature ( Figure 1—figure supplement 4 ) . Similarly , the C1 domain is required for membrane localization in the gonad ( Figure 1—figure supplement 4 ) . Thus , CYK-4E448K impairs membrane localization , and it is likely to impair the function of both its GAP and C1 domains . 10 . 7554/eLife . 08898 . 003Figure 1 . CYK-4-dependent membrane binding promotes furrow ingression . ( A ) CYK-4 accumulates on the plasma membrane . Membrane accumulation is observed on ingressing cleavage furrows ( boxed regions ) . CYK-4 membrane accumulation requires the C1 domain and is compromised by the E448K substitution in the cyk-4 ( or749ts ) allele . ( B ) Membrane accumulation of CYK-4 variants . The accumulation was quantified as a ratio of the accumulation of CYK-4::GFP/mCherry::PH at the furrow tip as depicted in the schematic; the mean intensity ±s . e . m . are plotted . ( N = 8–12 embryos p < 0 . 05 , one way ANOVA followed by Tukey multiple comparison ) . ( C ) Deletion of the C1 domain and the E448K substitution impair cleavage furrow ingression . Kymographs generated from time-lapse recordings of the equatorial region of embryos of the indicated genotypes expressing the membrane marker mCherry::PH . Kymographs begin at anaphase onset . ( D ) Deletion of the C1 domain and the E448K substitution abrogate centralspindlin-dependent furrow ingression . The progression of cytokinesis was assessed in embryos expressing CYK-4 variants in combination with a loss of function mutation in nop-1 . ( E ) Mutation of CED-10/Rac1 slightly increases furrow ingression in CYK-4∆C1 embryos and allows complete , albeit delayed , furrow ingression in CYK-4E448K embryos . The progression of cytokinesis was assessed in embryos expressing CYK-4 variants in combination with a loss of function mutation in ced-10 . ( F–I ) Quantification of furrow ingression rates in embryos of the indicated genotypes . Representative examples are shown in the accompanying kymographs . ( N = 8–12 embryos; error bars , 95% confidence intervals ) . ( J ) The ability of ced-10/Rac1 to rescue cytokinesis in CYK-4E448K embryos requires the NOP-1 pathway for furrow ingression . Images shown reflect the maximal extent of furrow ingression in embryos of the indicated genotypes expressing the membrane marker mCherry::PH . Unless otherwise specified , all scale bars in all figures are 10 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 08898 . 00310 . 7554/eLife . 08898 . 004Figure 1—figure supplement 1 . CYK-4 structure , mutations , and methods of transgene integration . Schematic depiction of the domain organization of CYK-4 , the variants tested , and the method for single copy transgene integration at a defined locus . DOI: http://dx . doi . org/10 . 7554/eLife . 08898 . 00410 . 7554/eLife . 08898 . 005Figure 1—figure supplement 2 . Transgene expression levels . ( A ) Quantification of the expression of CYK-4::GFP transgenes in embryos . Embryos were assayed in the first cell division and the total fluorescence intensity was measured . ( N ≥ 8 embryos; error bars , s . e . m; p < 0 . 001; n . s . , not significant , by one way ANOVA followed by Tukey multiple comparison ) . ( B ) Quantification of the expression of CYK-4::GFP transgenes . Total worm lysates were prepared from adult worms , standardized for total protein concentration and blotted with an anti-CYK-4 antibody . In the case of the RNAi experiment , 120 gravid adults worms were treated with cyk-4 ( RNAi ) or control treated and subjected to western blotting . DOI: http://dx . doi . org/10 . 7554/eLife . 08898 . 00510 . 7554/eLife . 08898 . 006Figure 1—figure supplement 3 . The CYK-4E448K::GFP transgene combined with cyk-4 ( RNAi ) closely phenocopies cyk-4 ( or749ts ) . Quantification of furrow ingression rates in embryos of the indicated genotypes . Results are quantified as described in Figure 1G . DOI: http://dx . doi . org/10 . 7554/eLife . 08898 . 00610 . 7554/eLife . 08898 . 007Figure 1—figure supplement 4 . Assay for localization of CYK-4 variants to the membrane of the gonad . ( A ) CYK-4E448K exhibits temperature sensitive binding to the incomplete membranes of the syncytial hermaphrodite gonad . Animals in which both endogenous and the GFP-tagged transgene contain the E448K substitution were imaged either immediately after removal from a 16°C incubator or after 60 min at 25°C . At the permissive temperature , CYK-4E448K binds to the gonad membrane , but not at the restrictive temperature . ( B ) The localization of CYK-4WT , CYK-4∆C1 , and CYK-4R459A was analyzed in hermaphrodite gonads . mCherry::PH was used to mark the membranes in the syncytial gonad . Whereas both CYK-4WT and CYK-4R459A label the membranes in the syncytial gonad , CYK-4∆C1 does not . DOI: http://dx . doi . org/10 . 7554/eLife . 08898 . 007 We compared the progression of cytokinesis in embryos expressing CYK-4WT , CYK-4∆C1 , and CYK-4E448K . CYK-4∆C1 embryos exhibit a cytokinetic defect that is largely similar to that of cyk-4E448K embryos . In particular , furrow ingression is slow and incomplete ( Figure 1C , F ) . However , subtle differences were detected; the onset of significant furrow ingression is delayed relative to controls in cyk-4E448K embryos but not in cyk-4∆C1 embryos . This suggests that the CYK-4E448K phenotype could reflect a compound defect , rather than a sole defect in the ability to associate with the membrane . Cleavage furrow ingression in C . elegans depends on the combined action of centralspindlin and a non-essential protein , NOP-1 ( Tse et al . , 2012 ) . To determine whether the C1 domain is essential for centralspindlin-dependent furrow ingression , we expressed CYK-4WT , CYK-4∆C1 , and CYK-4E448K in embryos that lack the NOP-1-dependent pathway for furrow ingression . As expected , CYK-4WT supports full furrow ingression in this sensitized background . In stark contrast , neither CYK-4∆C1 nor CYK-4E448K support detectable furrow ingression in the absence of NOP-1 activity ( Figure 1D , G ) . Previous studies demonstrated that loss of function mutations in ced-10/rac-1 or depletion of the protein partially suppress the cytokinesis defect in cyk-4 ( or749ts ) embryos ( Figure 1E , H ) . However , as described above , it is important to examine the extent of suppression in the absence of the parallel , NOP-1-dependent , pathway . We therefore examined whether ced-10/rac-1 loss of function mutations could suppress the cytokinesis defect in cyk-4∆C1 . Although cleavage furrows in cyk-4∆C1; ced-10 ( n1993 ) embryos ingress somewhat more deeply than cyk-4∆C1 embryos , they do not complete cytokinesis ( Figure 1E , H ) . The simplest interpretation of these results is that although CYK-4E448K diminishes membrane association of CYK-4 , it may retain some function at the restrictive temperature such that it facilitates the abscission step in cyk-4 ( or749ts ) ; ced-10 ( n1993 ) embryos , as CYK-4-mediated membrane association is essential for completion of cytokinesis in cultured human cells ( Lekomtsev et al . , 2012 ) . Importantly , when NOP-1 activity is compromised , inactivation of CED-10/Rac1 does not suppress the defect in furrow ingression caused by CYK-4E448K ( Figure 1J , I ) , suggesting that CYK-4 does not act directly on CED-10 . To begin to determine the role of the catalytic activity of CYK-4 during cytokinesis , we first studied the consequence of mutating the highly conserved catalytic arginine that stabilizes the transition state during GTP hydrolysis . Substitution of the catalytic arginine with alanine strongly attenuates GAP activity against Rac in a variety of CYK-4 orthologs and is widely used to inactivate Rho family GAPs ( Rittinger et al . , 1997; Yamada et al . , 2006; Miller and Bement , 2009; Bastos et al . , 2012; Zanin et al . , 2013 ) . As expected , CYK-4R459A GAP domain retains the ability to bind to RhoA•GTP , demonstrating that the protein is well folded in vitro ( Figure 2—figure supplement 1 ) . While CYK-4 GAP exhibits GAP activity towards both RhoA and CED-10/Rac1 , CYK-4R459A GAP lacks detectable GAP activity towards either GTPase ( Figure 2—figure supplement 2 ) . In order to determine if catalytic activity is required for viability , we used a strain heterozygous for a deletion mutant of CYK-4 , cyk-4 ( ok1034 ) . One quarter of the embryos from these heterozygous hermaphrodites contain maternally provided CYK-4 and lack zygotic CYK-4 . These zygotic null embryos fail to hatch and arrest with a variety of terminal phenotypes ( Figure 2A , left ) . Many , but not all , embryos contain muscle tissue and have undergone partial morphogenesis . The embryos also contain enlarged cells , likely due to defects in cytokinesis ( Sugimoto et al . , 2001 ) . We introduced the GAP-defective transgene into this strain . Remarkably , cyk-4 ( ok1034 ) ; cyk-4R459A animals hatch and develop to adulthood . However , these animals are sterile ( Figure 2A , middle ) . Thus , the GAP activity of CYK-4 is not essential for post embryonic development but it has an important role in gonad development , likely due to a requirement for post-embryonic cell proliferation in the germline . As a consequence , it is not possible to use classical genetic tools to obtain embryos in which CYK-4R459A is the sole form of CYK-4 . 10 . 7554/eLife . 08898 . 008Figure 2 . CYK-4 GAP activity is required for cytokinesis and viability . ( A ) cyk-4 ( ok1034 ) null embryos arrest at variable stages during embryogenesis ( left ) . The GAP-deficient CYK-4R459A transgene rescues the embryonic lethality , but animals arrest as sterile adults ( middle ) . GAP-deficient CYK-4R459A is recessive , no phenotypes is seen in the presence of cyk-4 ( + ) ( right ) . Right scale bar 50 µm . ( B ) GAP-deficient CYK-4R459A embryos fail to complete cytokinesis . Kymographs are generated as in Figure 1D , with the exception that the signal from the CYK-4 transgenes is also shown ( green ) , overlaid on mCherry::PH ( red ) . CYK-4R459A accumulates more strongly at furrow tips as compared to CYK-4WT ( arrows ) . ( C ) The kinetics of furrow ingression in CYK-4WT and CYK-4R459A embryos . Results are quantified as described in Figure 1G . ( D ) Membrane accumulation of CYK-4WT and CYK-4R459A . Results are quantified as described in Figure 1B ( p < 0 . 01 , by t-test ) . DOI: http://dx . doi . org/10 . 7554/eLife . 08898 . 00810 . 7554/eLife . 08898 . 009Figure 2—figure supplement 1 . Binding of CYK-4 variants to RhoA . ( A ) Schematic of the domain structure of full length CYK-4 and the recombinant GAP domains used in binding studies . The positions of the relevant mutations are indicated . ( B ) Binding assays between CYK-4 C1GAP domains and GST-tagged RhoA complexed with GTPγS . CYK-4WT GAP domain binds to RhoA•GTPγS . The CYK-4EE GAP domain exhibits reduced binding to RhoA•GTPγS , whereas CYK-4R459A GAP domain exhibits increased binding to RhoA•GTPγS . ( C ) GST-pulldown assays between GST-RhoA•GTPγS and CYK-4 GAP/C domain variants . Wild-type CYK-4 GAP binds to RhoA•GTPγS . CYK-4R459A GAP exhibits increased binding to RhoA•GTPγS , whereas CYK-4EE GAP exhibits reduced binding to RhoA•GTPγS . None of the CYK-4 GAP variants exhibit binding activity to GST alone . ( D ) Proteins used in binding assays in ( C ) . DOI: http://dx . doi . org/10 . 7554/eLife . 08898 . 00910 . 7554/eLife . 08898 . 010Figure 2—figure supplement 2 . GAP activity of CYK-4 and variants . ( A ) Time course of GTP hydrolysis by RhoA in the presence or absence of 200 nM CYK-4 GAP or CYK-4R459A GAP . ( B ) Titration of CYK-4 GAP activity towards CED-10/Rac1 and RhoA . Free phosphate was measured after 30 min . DOI: http://dx . doi . org/10 . 7554/eLife . 08898 . 010 To study the role of the CYK-4 GAP activity during embryogenesis , we used the aforementioned assay ( Figure 1—figure supplement 1 ) . Embryos , expressing only CYK-4R459A , are largely normal during the initial stages of the first cell cycle . They undergo pseudocleavage , mitotic spindle assembly , chromosome segregation , central spindle assembly during anaphase , and CYK-4R459A becomes highly enriched on the spindle midzone ( Figure 2B ) . Cleavage furrow initiation occurs and the furrow ingresses at near wild-type rates to near completion . However , cytokinesis does not complete and the furrow ultimately regresses; this phenotype was fully penetrant ( Figure 2C ) ; these embryos also fail to complete cytokinesis following meiosis II ( data not shown ) . These results suggest that a late step in cytokinesis is most sensitive to loss of CYK-4 GAP activity . This phenotype is distinct from that of cyk-4 ( or749ts ) embryos–cleavage furrows in cyk-4R459A embryos ingress more rapidly and more deeply than cyk-4E448K embryos ( Figure 1C ) . We assessed the ability of CYK-4R459A to associate with membrane during furrow ingression ( Figure 2B , D ) . CYK-4R459A hyper accumulates on the membrane as compared to WT CYK-4; this localization suggests that CYK-4R459A is well folded in vivo . Therefore , the cytokinetic defect in this strain is unlikely to be an indirect consequence of a failure of CYK-4 to localize to the membrane . To extend these results , confirm that CYK-4 must interact with Rho family GTPases during cytokinesis , and eliminate the possibility that the phenotype of CYK-4R459A is due to enhanced binding of CYK-4 to active RhoA , we engineered mutations in CYK-4 that reduce its binding to RhoA and other GTPases ( Rittinger et al . , 1997; Sekimata et al . , 1999 ) . Two conserved , surface exposed , basic residues in the RhoA interface ( K495 , R499 ) ( Figure 3—figure supplement 1 ) were charge reversed to glutamic acid , generating CYK-4EE , and characterized in the transgenic rescue assay . Embryos expressing only CYK-4EE , like those expressing CYK-4R459A and CYK-4∆C1 , exhibit fully penetrant embryonic lethality ( Figure 3—figure supplement 1 ) . CYK-4EE exhibits reduced binding to RhoA in vitro ( Figure 2—figure supplement 1 ) , and it does not exhibit membrane hyperaccumulation in vivo ( Figure 3A , B ) . Interestingly , cyk-4EE embryos exhibit a stronger furrow ingression defect than cyk-4R459A embryos , as furrow ingression is slower and less complete ( Figure 3C , D ) . Importantly , NOP-1 depletion from cyk-4EE embryos largely eliminates furrow ingression ( Figure 3C , D ) . Thus , Rho GTPase binding by CYK-4 is essential for centralspindlin-mediated cytokinetic ingression . 10 . 7554/eLife . 08898 . 011Figure 3 . RhoA binding by CYK-4 is required for cytokinesis . ( A ) CYK-4R459A , but not CYK-4EE , hyperaccumulates on the plasma membrane of the ingressing cleavage furrows ( boxed regions ) . ( B ) Membrane accumulation of CYK-4::GFP variants was quantified as described in Figure 1B . ( N = 8–12 embryos; p < 0 . 01 , by one way ANOVA followed by Tukey multiple comparison ) . ( C ) The Rho family GTPase binding defective variant of CYK-4 , CYK-4EE , , causes cytokinesis defects . Kymograph analysis of the progression of cytokinesis in CYK-4EE embryos in the presence or absence of NOP-1 function . Kymographs were assembled as described in legend to Figure 1D . ( D ) The kinetics of furrow ingression in CYK-4EE embryos . Results are quantified as described in Figure 1G . ( E ) Inactivation of CDC-42 and CED-10/Rac1 , either alone or in combination , does not cause cytokinesis defects in the sensitized nop-1 background . Kymographs were assembled as described in legend to Figure 1D . Note that depletion of CDC-42 results in symmetric cleavage furrow ingression . ( F ) The kinetics of furrow ingression in embryos deficient in nop-1 and/or CDC-42 and/or CED-10/Rac1 function . Results are quantified as described in Figure 1G . DOI: http://dx . doi . org/10 . 7554/eLife . 08898 . 01110 . 7554/eLife . 08898 . 012Figure 3—figure supplement 1 . Location and conservation of mutated residues in the CYK-4 GAP domain . ( A ) Alignment of a region of the GAP domain of CYK-4 from various species and the unrelated RhoGAP P50 . The positions of the catalytic arginine ( Ce 459 ) and the conserved basic residues ( Ce K495/R499 ) are indicated . Accession numbers Q9H0H5 . 1; Q9WVM1 . 1; AAF58324 . 1; CAB04593 . 1; Q07960 . 1 . ( B ) Embryonic lethality of CYK-4 variants at two temperatures . cyk-4R459A::gfp , cyk-4EE::gfp , and cyk-4∆C1::gfp exhibit fully penetrant embryonic lethality at 16°C and 25°C . cyk-4WT::gfp is included as a control . DOI: http://dx . doi . org/10 . 7554/eLife . 08898 . 012 We next sought to determine the Rho family GTPase to which CYK-4 must bind to fulfill its function in vivo . If furrow formation is dependent on CYK-4 binding to either CED-10/Rac1 or CDC-42 to generate a positive regulatory complex , then inactivation of these GTPases would be predicted to cause a phenotype at least as severe as a mutation that weakens the GTPase binding site of CYK-4 . However , mutation of CED-10/Rac1 , or depletion of CDC-42 , does not affect the rate of cleavage furrow ingression ( Jantsch-Plunger et al . , 2000; Loria et al . , 2012 ) , even when combined with mutations in NOP-1 ( Figure 3E , F ) . Indeed , cytokinesis occurs efficiently and proceeds to completion in embryos in which NOP-1 , CED-10/Rac1 , and CDC-42 are simultaneously inactivated ( Figure 3E , F ) . We infer , therefore , that RhoA is the relevant GTPase that CYK-4 binds to promote cleavage furrow formation . Due to its direct role in furrow ingression , it is not possible to test RhoA in the same manner . We next sought to determine how the GAP activity of CYK-4 promotes cytokinesis . Previous studies proposed at least three models for the phenotype seen in cyk-4R459A embryos . First , the GAP domain could function as canonical GAP that acts on RhoA , causing CYK-4 GAP-deficient embryos fail to complete cytokinesis because of a requirement for RhoA inactivation at late cytokinesis . Second , CED-10/Rac1 could be an important target of CYK-4 GAP activity , causing CYK-4 GAP deficient embryos to accumulate ectopic Rac1 activity that interferes with cytokinesis . Third , although it is counterintuitive , CYK-4 GAP activity could somehow promote RhoA activation , and therefore the CYK-4 GAP deficient embryos may fail cytokinesis due to incomplete RhoA activation . We sought to distinguish between these alternatives . The first and third models make opposite predictions for the outcome of experiments in which RhoA levels are perturbed ( Figure 4A ) . If the failure to complete cytokinesis in cyk-4R459A embryos is due to hyperactivation of RhoA , as would be predicted from the canonical model for the function of a RhoA GAP , then a reduction in active RhoA levels might ameliorate the defect . Conversely , if the GAP active site promotes RhoA activation , then the reduction of RhoA activity would be predicted to exacerbate the phenotype of cyk-4R459A . To distinguish between these models , we reduced RhoA levels by mutationally inactivating NOP-1 . As expected , all control embryos ( nop-1 ( it142 ) ; cyk-4WT ) complete cytokinesis ( Figure 4Bi ) . Surprisingly , nop-1 ( it142 ) ; cyk-4R459A mutant embryos exhibit extremely weak furrow ingression; furrows in these embryos ingressed less than ∼10% of egg width ( Figure 4Biii , C ) . This result supports models in which CYK-4 GAP activity is involved in RhoA activation . 10 . 7554/eLife . 08898 . 013Figure 4 . The GAP activity of CYK-4 is required for centralspindlin-dependent furrowing independent of CED-10/Rac1 . ( A ) Schematic depiction of the known regulators of RhoA . ( B ) GAP defective CYK-4 causes cytokinesis defects that are greatly enhanced by loss of NOP-1 function; this defect is not fully suppressed by mutation of ced-10/rac1 . Kymograph analysis of the progression of cytokinesis in CYK-4R459A embryos in the presence or absence of CED-10/Rac1 and/or NOP-1 function . The kinetics of furrow ingression in cyk-4WT::gfp; nop-1 ( it142 ) and nop-1 ( it142 ) ; ced-10 ( n1993 ) embryos are shown for comparison . Kymographs were assembled as described in legend to Figure 1D . ( C ) The kinetics of furrow ingression in CYK-4R459A embryos . Results are quantified as described in Figure 1G . The kinetics of furrow ingression in cyk-4WT::gfp; nop-1 ( it142 ) and nop-1 ( it142 ) ; ced-10 ( n1993 ) embryos from Figures 1F , 3F are shown as dashed lines for comparison . DOI: http://dx . doi . org/10 . 7554/eLife . 08898 . 01310 . 7554/eLife . 08898 . 014Figure 4—figure supplement 1 . Depletion of ARX-2 , but not RAC-2 , suppresses the cytokinesis defect in CYK-4R459A embryos . The kinetics of furrow ingression in CYK-4R459A embryos depleted of RAC-2 or ARX-2 . Results are quantified as described in Figure 1G . The kinetics of furrow ingression in cyk-4R459A::gfp embryos from Figure 4D is shown for comparison . DOI: http://dx . doi . org/10 . 7554/eLife . 08898 . 01410 . 7554/eLife . 08898 . 015Figure 4—figure supplement 2 . A gain of function mutation , mig-2 ( gm103 ) , does not affect cytokinesis in sensitized backgrounds . NOP-1 or ZYG-9 was depleted from mig-2 ( gm103 ) embryos and the progression of cytokinesis was followed by nomarski imaging . Cytokinesis proceeds to completion in the former case , and both anterior ( left ) and posterior ( right ) furrows are formed in the latter case . ZYG-9 was depleted from cyk-4R459A::gfp embryos for comparison demonstrating loss of the posterior furrow . DOI: http://dx . doi . org/10 . 7554/eLife . 08898 . 01510 . 7554/eLife . 08898 . 016Figure 4—figure supplement 3 . Depletion of neither ARX-2 nor RAC-2 modulates the cytokinesis defect in nop-1; ced-10; cyk-4R459A embryos . The kinetics of furrow ingression in cyk-4R459A::gfp; nop-1 ( it142 ) ; ced-10 ( n1993 ) embryos depleted of either ARX-2 or RAC-2 . Results are quantified as described in Figure 1G . The kinetics of furrow ingression in nop-1 ( it142 ) ; cyk-4WT::gfp , cyk-4R459A::gfp; nop-1 ( it142 ) , and cyk-4R459A::gfp; nop-1 ( it142 ) ; ced-10 ( n1993 ) embryos from Figures 1G , 4C are shown for comparison . DOI: http://dx . doi . org/10 . 7554/eLife . 08898 . 016 Previous studies have implicated CED-10/Rac1 as a target of CYK-4 GAP activity , although these studies utilized the cyk-4 ( or749ts ) mutation that impairs membrane localization of CYK-4 ( Figure 1A , B , Figure 1—figure supplement 4A ) . Therefore , we addressed whether loss of function mutations in ced-10 affect cytokinesis in cyk-4R459A embryos . Interestingly , we found that all ced-10; cyk-4R459A embryos fully ingress and 75% complete cytokinesis ( Figure 4Biv , C ) , suggesting significant , albeit incomplete rescue . Depletion of ARX-2 , a component of the Arp2/3 complex , a downsteam effector of Rac GTPases , provides similar rescue as mutation in ced-10 ( Figure 4—figure supplement 1 ) . Two other Rac related proteins , RAC-2 and MIG-2 , could , in principle , be additional targets of the CYK-4 GAP domain . However , depletion of RAC-2 does not rescue completion of cytokinesis in cyk-4R459A embryos ( Figure 4—figure supplement 1 ) . Furthermore , gain of function mutations in mig-2 ( Zipkin et al . , 1997 ) do not cause cytokinesis defects , even in sensitized genetic backgrounds ( Figure 4—figure supplement 2 ) . Mutations in ced-10 also slightly increase the extent of furrow ingression in cyk-4∆C1 embryos ( Figure 1E , H ) . To more stringently test whether the GAP activity of CYK-4 is linked with CED-10/Rac1 inactivation , we assessed the progression of cytokinesis in embryos that lack NOP-1 function . Crucially , nop-1; ced-10 embryos complete cytokinesis ( Figure 4Bvi ) . If CED-10/Rac1 inactivation is the primary function of the CYK-4 GAP domain , then CYK-4 GAP activity would be predicted to be dispensable in nop-1; ced-10 embryos . However , in stark contrast to this prediction , nop-1; ced-10; cyk-4R459A embryos fail to form ingressing cleavage furrows altogether ( Figure 4Bv , C ) . Significant furrow ingression is not restored by depletion of either RAC-2 or ARX-2 in nop-1; ced-10; cyk-4R459A embryos , suggesting that the cytokinesis defect is not due to activation of other Rac-family proteins ( Figure 4—figure supplement 3 ) . These data demonstrate that the catalytic activity of the CYK-4 GAP domain must have a function that is distinct from maintaining CED-10/Rac1 in an inactive state . To further test models in which CYK-4 RhoGAP catalytic activity is important to either promote RhoA activation or to promote RhoA inactivation , we examined the consequence of depletion of the predominant RhoA GAP in the early embryo , RGA-3/4 ( Schmutz et al . , 2007; Schonegg et al . , 2007 ) . As previously shown , depletion of RGA-3/4 causes cortical hypercontractility in otherwise wild-type embryos , during both pseudocleavage and cytokinesis , and results in embryonic lethality ( Figure 5A ) ( Schmutz et al . , 2007; Schonegg et al . , 2007 ) . When RGA-3/4 is depleted from cyk-4R459A embryos , all embryos complete cytokinesis ( Figure 5B , Figure 5—figure supplement 1A ) , further suggesting that the GAP activity of CYK-4 promotes , rather than counteracts , RhoA activation . 10 . 7554/eLife . 08898 . 017Figure 5 . Depletion of RGA-3/4 rescues cytokinesis in CYK-4R459A embryos . ( A ) Schematic depiction of the known regulators of RhoA . ( B ) Representative embryos demonstrating the effect of RGA-3/4 depletion on cytokinesis in CYK-4R459A embryos both in the presence and absence of NOP-1 function . ( C ) Depletion of RGA-3/4 rescues cytokinesis specifically in CYK-4R459A embryos . CYK-4∆C1 , CYK-4EE , CYK-4R459A where expressed in nop-1 ( it142 ) embryos and the extent of furrow closure measured either in the presence ( green ) or the absence ( red ) of RGA-3/4 . ( D ) Accumulation of the RhoA effector NMY-2 ( tagged with RFP ) in embryos of the indicated genotypes . Embryos are shown at ∼50% ( or maximal ) ingression . Note the reduction of cortical myosin accumulation in CYK-4R459A , CYK-4EE , CYK-4∆C1 embryos as compared to CYK-4WT ( i–iv ) . The severity of this reduction is enhanced by inactivation of NOP-1 ( i′–iv′ ) . Depletion of RGA-3/4 restores myosin accumulation in CYK-4R459A and CYK-4EE embryos ( v , vi ) , even in embryos defective in NOP-1 function ( v′ , vi′ ) . ( E ) Quantification of total NMY-2::mRFP accumulation in the furrow region over the course of cytokinesis in embryos of the indicated genotypes . Error bars , s . e . m . ; n . s . ( not significant ) ; #p < 0 . 05; ###p < 0 . 001 refers to significance relative to wild-type in nop-1 ( + ) and nop-1 ( it142 ) , respectively by one way ANOVA followed by Tukey multiple comparison . p < 0 . 001 for the indicated comparison . DOI: http://dx . doi . org/10 . 7554/eLife . 08898 . 01710 . 7554/eLife . 08898 . 018Figure 5—figure supplement 1 . Furrow ingression of RGA-3/4-depleted embryos expressing CYK-4 variants . ( A ) Line graphs depicting the kinetics of furrow ingression in ( A ) [cyk-4 ( RNAi ) , CYK-4∆C1 , CYK-4EE , CYK-4R459A]; nop-1 ( it142 ) ; rga-3/4 ( RNAi ) embryos and ( B ) [cyk-4 ( RNAi ) , CYK-4EE , CYK-4R459A]; rga-3/4 ( RNAi ) . These data are summarized in Figure 5C . DOI: http://dx . doi . org/10 . 7554/eLife . 08898 . 01810 . 7554/eLife . 08898 . 019Figure 5—figure supplement 2 . Comparison between accumulation of NMY-2::RFP and the RhoA biosensor during cytokinesis . ( A ) Correlation between the accumulation of NMY-2::mRFP and GFP::AHPH ( RhoA biosensor [Tse et al . , 2012] ) in embryos of the indicated genotypes . Each spot corresponds to the average integrated intensity of the saturation > signal >1 . 5cytoplasmic background in a pair of images during cytokinesis ( i . e . , each embryo is represented by 9–11 dots ) . The data are fit to a line as shown . ( B ) Gallery of images showing the distribution of NMY-2::mRFP and GFP::AHPH in the furrow region during cytokinesis in sets of embryos of the indicated genotypes . DOI: http://dx . doi . org/10 . 7554/eLife . 08898 . 019 To test this model more stringently , we asked whether cytokinesis also completes when RGA-3/4 is depleted from cyk-4R459A embryos also lacking NOP-1 ( i . e . , nop-1 ( it142 ) ; rga-3/4 ( RNAi ) ; cyk-4R459A embryos ) . Remarkably , although furrows in cyk-4R459A; nop-1 embryos barely ingress , when RGA-3/4 is depleted , furrow ingression is completed in 100% of embryos ( Figure 5B , C , Figure 5—figure supplement 1B ) . This result also rules out the possibility that RGA-3/4 depletion allows completion of cytokinesis because it stabilizes RhoA that was activated in a NOP-1-dependent manner . Depletion of RGA-3/4 did not significantly modify the cytokinetic phenotype of nop-1; cyk-4 ( RNAi ) embryos ( Figure 5C , Figure 5—figure supplement 1B ) , demonstrating CYK-4 dependence to this suppression . Furthermore , complete suppression was specific to cyk-4R459A embryos , depletion of RGA-3/4 induced deeper but still incomplete ingression in nop-1; cyk-4EE and nop-1; cyk-4∆C1 embryos . These strains formed an allelic series in order of decreasing extents of ingression: cyk-4R459A > cyk-4EE > cyk-4∆C1 ∼ cyk-4 ( RNAi ) ( Figure 5C , Figure 5—figure supplement 1B ) . RhoA is a dose dependent regulator of cleavage furrow formation ( Loria et al . , 2012 ) and CYK-4 is involved in RhoA activation by relieving autoinhibition of ECT-2 ( Kim et al . , 2005; Yüce et al . , 2005 ) . We therefore assayed whether CYK-4 GAP domain mutations affect accumulation of RhoA effectors . Because the RhoA biosensor and the CYK-4 transgenes are both GFP-tagged and integrated at the same position of the genome , we assayed the accumulation of RFP-tagged non-muscle myosin , NMY-2 , a key effector of RhoA , as a proxy for RhoA activation . To validate that NMY-2::mRFP is a valid proxy for RhoA activity levels , we compared the accumulation of these two markers to the cleavage furrow during cytokinesis when co-expressed . The recruitment of NMY-2::mRFP and the RhoA biosensor are highly correlated in space , time , and intensity ( Figure 5—figure supplement 2A , B ) . In addition , the correlation between these markers remains strong when either NOP-1 , CYK-4 , or RGA-3/4 are depleted , despite the significant changes in the extent of recruitment caused by these perturbations . Thus NMY-2::mRFP provides a reliable proxy for RhoA activation . We assayed NMY-2::mRFP levels in CYK-4WT , CYK-4∆C1 , CYK-4R459A , and CYK-4EE embryos during anaphase . Mutations in CYK-4 that reduce the rate and extent of cleavage furrow also reduce NMY-2::mRFP accumulation ( Figure 5D , top row , Figure 5E ) . The defect in myosin accumulation caused by mutations in the GAP domain of CYK-4 is far more severe and apparent in NOP-1-depleted embryos ( Figure 5D , bottom row , Figure 5E ) . Conversely , depletion of RGA-3/4 increases myosin accumulation in CYK-4R459A and CYK-4EE embryos , both in the presence and absence of NOP-1 . These data support models in which the catalytic activity of the CYK-4 GAP domain contributes to RhoA activation . To obtain additional insight into the mechanism by which CYK-4 promotes cytokinesis , we took an unbiased genetic suppression approach . We mutagenized cyk-4 ( or749ts ) animals , grew the mutagenized animals at the permissive temperature for two generations to allow potential suppressors to become homozygous and shifted them to 25°C to select for suppressors . We isolated three strong suppressors out of a total of ∼10^5 mutagenized F1 genomes . Suppressor strains were subjected to sequencing of the cyk-4 locus to identify potential intragenic suppressors . One strain contained a substitution mutation in CYK-4 , H485Y , relatively close to the or749ts substitution E448K ( Figure 6A ) . We also isolated two strong extragenic suppressors , xs110 and xs111 , that rescue cyk-4 ( or749ts ) to viability at the restrictive temperature . The suppressed strains complete cytokinesis with high efficiency ( >90% ) and support high viability ( Figure 6—figure supplement 1 ) . 10 . 7554/eLife . 08898 . 020Figure 6 . Mutations in ECT-2 suppress cyk-4 ( or749ts ) . ( A ) Schematic depicting the domain structure of CYK-4 and ECT-2 with the positions of the mutations isolated in the cyk-4 ( or749ts ) suppressor screen . ( B ) ect-2 ( xs110 ) and ect-2 ( xs111 ) suppresses both cyk-4 ( or749ts ) and CYK-4R459A . Images of embryos of the indicated genotypes are shown at the pronuclear stage and during cytokinesis . Both ect-2 ( xs110 ) ( i ) and ect-2 ( xs111 ) ( v ) embryos exhibit hypercontractility ( arrows ) that is suppressed by cyk-4 ( or749ts ) ( ii and vii ) . Depletion of NOP-1 from cyk-4 ( or749ts ) ; ect-2 ( xs110 ) eliminates contractility during pseudocleavage and greatly reduces contractility during cytokinesis ( iii ) . Depletion of NOP-1 from cyk-4 ( or749ts ) ; ect-2 ( xs111 ) eliminates contractility during pseudocleavage but cytokinesis is still observed ( viii ) . ect-2 ( xs110 ) and ect-2 ( xs111 ) both allow cytokinetic completion in CYK-4R459A ( iv , ix ) . Depletion of CYK-4 prevents completion of cytokinesis in ect-2 ( xs110 ) and ect-2 ( xs111 ) ( v , x ) . Phenotypes shown were seen in ( i ) 18/18 embryos; ( ii ) 11/12; ( iii ) 7/16 , 5/16 showed less contractility; ( iv ) 6/6; ( v ) 7/7; ( vi ) 14/14; ( vii ) 18/18; ( viii ) 13/16; ( ix ) 6/6; ( x ) 6/6 . ( C ) ect-2 ( xs110 ) and ect-2 ( xs111 ) suppress cyk-4 ( or749ts ) . The kinetics of furrow ingression in cyk-4 ( or749ts ) embryos and in the suppressed strains . Results are quantified as described in Figure 1G . ( D ) ect-2 ( xs110 ) causes defects in cleavage plane positioning . The position of furrow initiation and the spindle midzone are indicated in yellow and purple , respectively ( see schematic ) . ( E ) Quantification of the mean position of the central spindle ( ±s . e . m ) as a function of egg length in wild-type and ect-2 ( xs110 ) embryos . ( F ) Quantification of the mean position of furrow initiation ( ±s . e . m ) relative to the center of the spindle midzone in wild-type and ect-2 ( xs110 ) embryos . ( ***p < 0 . 001 , by t-test ) . DOI: http://dx . doi . org/10 . 7554/eLife . 08898 . 02010 . 7554/eLife . 08898 . 021Figure 6—figure supplement 1 . Viability and fertility of cyk-4 mutants and suppressors . Mean brood size and hatch rates of strains of the indicated genotypes ( n/t , not tested ) . All strains were tested at 25°C . DOI: http://dx . doi . org/10 . 7554/eLife . 08898 . 02110 . 7554/eLife . 08898 . 022Figure 6—figure supplement 2 . Conservation of ECT-2 GEF domain and inferred position of the ect-2 ( xs110 ) allele . ( A ) Sequence alignment of DHPH domains from ECT-2 orthologs and other GEFs highlighting the position of the G707 . Accession numbers: Q9H8V3 . 4; AAF50508 . 2; O15085 . 1; Q92888 . 2; CAB54311 . 1; Q64096 . 2 . ( B ) Portion of the crystal structure ( 1XCG ) of PDZRhoGEF ( green , DH domain; red , PH domain; blue , RhoA ) . The equivalent to G707 , N1068 is highlighted . DOI: http://dx . doi . org/10 . 7554/eLife . 08898 . 02210 . 7554/eLife . 08898 . 023Figure 6—figure supplement 3 . Accumulation of myosin during cytokinesis in ect-2 ( xs110 ) . ect-2 ( xs110 ) results in hyperaccumulation of NMY-2::GFP during cytokinesis . DOI: http://dx . doi . org/10 . 7554/eLife . 08898 . 02310 . 7554/eLife . 08898 . 024Figure 6—figure supplement 4 . ect-2 ( xs110 ) and ect-2 ( xs111 ) are dominant gain of function mutations . Nomarski imaging of ect-2 ( xs110 ) /+ and ect-2 ( xs111 ) /+ during pseudocleavage and cytokinesis . Hypercontractility ( arrows ) is observed in the presence of a wild-type allele of ect-2 . DOI: http://dx . doi . org/10 . 7554/eLife . 08898 . 024 Candidate extragenic suppressors were genetically mapped to chromosome II near the ect-2 locus . Substitution mutations in the ect-2 locus were identified in both suppressor strains ( Figure 6A ) . ect-2 ( xs110 ) contained a single nucleotide change in the PH domain , resulting in a G707D substitution ( Figure 6A , Figure 6—figure supplement 2A ) . In a related RhoGEF for which there is a co-crystal structure with RhoA ( PDZRhoGEF ) , the residue analogous to G707 lies in an α helix in the PH domain that comes into close proximity to the α3 helix of RhoA ( Figure 6—figure supplement 2B ) ( Chen et al . , 2010 ) . We recorded the progression of cytokinesis in cyk-4 ( or749ts ) ; ect-2 ( xs110 ) embryos and found that the embryos not only complete cytokinesis as expected but also the delay in furrow initiation and the slow furrow ingression phenotypes characteristic of cyk-4 ( or749ts ) embryos were largely corrected ( Figure 6B , Cii ) . Thus , unlike ced-10 ( n1993 ) , ect-2 ( xs110 ) suppresses the primary defect of the cyk-4 ( or749ts ) mutation . To confirm that the ect-2 ( xs110 ) substitution was causative , we used the CRISPR-associated nuclease Cas9 to re-create this mutation ( Zhang and Glotzer , 2014 ) . We injected cyk-4 ( or749ts ) animals with a plasmid that expresses both Cas9 and a sgRNA designed to create a double strand break near E705 and provided an oligonucleotide repair template containing the G707D substitution . The animals were maintained at the permissive temperature for two generations before shifting to the restrictive temperature . We were able to isolate a strain that was viable and fertile . The ect-2 locus was sequenced and de novo generation of the G707D substitution was confirmed . This mutation in ect-2 therefore suppresses all the essential functions affected by the cyk-4 ( or749ts ) allele . We next investigated whether this mutation causes a detectable phenotype when separated from cyk-4 ( or749ts ) . Interestingly , ect-2 ( xs110 ) embryos exhibit hypercontractility during both pseudocleavage and cytokinesis ( Figure 6Bi ) ; this hypercontractility is associated with enhanced cortical accumulation of myosin II ( Figure 6—figure supplement 3 ) . Hypercontractility is also observed in embryos from ect-2 ( xs110 ) /+ hermaphrodites , indicating ect-2 ( xs110 ) is a dominant , gain of function allele ( Figure 6—figure supplement 4 ) . The hypercontractility is reduced in ect-2 ( xs110 ) ; cyk-4 ( or749ts ) embryos ( ii ) , indicating that ECT-2G707D hyperactivity is partially dependent on CYK-4 and that cyk-4 ( or749ts ) and ect-2 ( xs110 ) exhibit mutual suppression . Comparison of ect-2 ( xs110 ) ; cyk-4 ( or749ts ) embryos to ect-2 ( xs110 ) ; cyk-4 ( or749ts ) ; nop-1 ( RNAi ) ( Figure 6ii vs Figure 6iii ) embryos reveals that NOP-1 also contributes to contractility in ECT-2G707D embryos . An unusual phenotype was observed in ect-2 ( xs110 ) embryos . Following anaphase , the cleavage furrow frequently initiates from a site significantly anterior to the midpoint of the anaphase spindle ( Figure 6D–F ) . As the furrow ingresses , it undergoes a dramatic repositioning so that it ultimately bisects the anaphase spindle . Nevertheless , the ect-2 ( xs110 ) strain is viable and fertile despite exhibiting hypercontractility during polarization and cytokinesis ( Figure 6—figure supplement 1 ) . The second suppressor allele , ect-2 ( xs111 ) , also contains a substitution mutation in ECT-2 . This mutation is located in the linker region between the cryptic BRCT0 domain and BRCT1 ( Zou et al . , 2014 ) ( Figure 6A ) . Several criteria indicate that this mutation is also causal . First , SNP mapping placed suppressor activity near the ect-2 locus . Second , the suppressor was analyzed by one step mapping and whole genome sequencing ( Doitsidou et al . , 2010 ) . ect-2 is the only gene in the candidate region that contained a non-silent mutation that has any role in cytokinesis . Third , biochemical data indicate that this mutation relieves ECT-2 autoinhibition ( see below ) . The ect-2 ( xs111 ) gain of function allele exhibited similar overall characteristics as ect-2 ( xs110 ) ( Figure 6B , C ) , although the spindle positioning defect was less severe ( not shown ) . The one remarkable difference was that ect-2 ( xs111 ) ; cyk-4 ( or749ts ) ; nop-1 ( RNAi ) ( Figure 6Bviii ) embryos fully ingressed during cytokinesis , though they do not form pseudocleavage furrows . The ability of these embryos to complete cytokinesis depends upon residual activity from CYK-4E448K , as depletion of CYK-4 by RNAi prevents completion of cytokinesis in ect-2 ( xs111 ) embryos ( Figure 6Bx ) . As complete furrow ingression is not seen in comparable ect-2 ( xs110 ) embryos ( Figure 6Biii ) , ect-2 ( xs111 ) may be more strongly activated than ect-2 ( xs110 ) . This genetic screen demonstrates that only rare mutations suppress cyk-4 ( or749ts ) and that the essential function of CYK-4 that is inactivated by CYK-4E448K is the ability to activate RhoA . Note that while ced-10 ( n1993 ) can partially suppress cytokinesis defects in CYK-4E448K expressing embryos ( cytokinesis remains delayed and slow in the double mutant; and only ∼67% of embryos complete division ) , ced-10 ( n1993 ) does not rescue cyk-4 ( or749ts ) to viability ( Figure 6—figure supplement 1 ) . CYK-4R459A causes a less severe phenotype than CYK-4E448K , therefore we predicted that ect-2 ( xs110 ) and ect-2 ( xs111 ) could also suppress CYK-4R459A . We used CRISPR/Cas9 to introduce the R459A mutation into the endogenous cyk-4 gene and crossed it into both ect-2 hyperactive mutants . We were able to isolate strains in which the sole source of CYK-4 lacks the critical arginine in the active site ( Figure 6Biv , ix ) . The resulting strains exhibited high viability and fertility ( Figure 6—figure supplement 1 ) . This finding provides independent confirmation that the sole essential function of the RhoGAP active site of CYK-4 is to stimulate ECT-2-mediated RhoA activation . These genetic and cell biological results demonstrate that the GAP activity of CYK-4 contributes to RhoA activation . As ECT-2 is required for all RhoA activity during cytokinesis , the CYK-4 GAP domain is likely to serve this role by modulating ECT-2 . Given that the canonical function of a RhoGAP domain is to inhibit RhoA activity , it is surprising that a protein containing a RhoGAP domain enhances RhoA activation . However , CYK-4 and ECT-2 form a protein complex through their regulatory N-termini ( Burkard et al . , 2007; Wolfe et al . , 2009 ) , therefore the C-terminal GAP domain of CYK-4 will be in the vicinity of the ECT-2 RhoGEF domain . We therefore hypothesized that the interactions between CYK-4 and ECT-2 are not limited to their N-termini . To test this , we purified the C-terminal domains of CYK-4 and ECT-2 ( Figure 7A ) and performed binding assays . We found that the catalytic C-termini of CYK-4 and ECT-2 directly interact ( Figure 7B , Figure 7—figure supplement 1 ) ; a similar complex is also found with human orthologs ( data not shown ) . We assayed for activation of the ECT-2 GEF activity by the CYK-4 GAP domain in vitro . However , we have not yet been able to detect stimulation of GEF activity ( data not shown ) . This negative result could be due to missing components , a requirement for the context provided by the full length , oligomerization competent proteins ( Basant et al . , 2015 ) , or the absence of the plasma membrane to which CYK-4 must bind in vivo in order to activate ECT-2 . 10 . 7554/eLife . 08898 . 025Figure 7 . Biochemical basis for suppression by ECT-2 variants . ( A ) Schematic depiction of the domain organization of CYK-4 and ECT-2 and the recombinant fragments used for biochemical analyses . ( B ) GST pulldown between GST-ECT-2 DHPH and CYK-4 C1GAP . ( C ) Pulldown assay between MBP-ECT-2-N ( wild-type and the E129K variant ) with GST-ECT-2 DHPH ( both wild-type and the G707D variant ) . The MBP proteins were present in the soluble fraction and incubated with the GST-DHPH fragments bound to beads . The wild-type N-terminus associates with wild-type and G707D C-termini . However , the ECT-2 NE129K is defective in binding to wild-type C-terminus . ( D ) The G707D substitution activates the exchange activity of ECT-2 . Exchange assays were performed with RhoA•GDP exchanging for mant-GTP at different concentrations of ECT-2 DHPH and ECT-2 DHPHG707D . Results shown are the average ( ±s . e . m ) of three assays . The gel contains 1 µg of each ECT-2 variant and a BSA standard . ( E ) Working model summarizing the proposed mechanism for ECT-2 activation . Note that only the CYK-4 subunit of centralspindlin is shown . In vivo , centralspindlin is predicted to be oligomeric and the entire complex bound to the plasma membrane . DOI: http://dx . doi . org/10 . 7554/eLife . 08898 . 02510 . 7554/eLife . 08898 . 026Figure 7—figure supplement 1 . Biochemical characterization of CYK-4 and ECT-2 variants . ( A ) GST-pulldown assay between GST-ECT-2 DHPH and CYK-4 GAP/C . Wild-type , R459A , and EE variants of CYK-4 GAP exhibit similar binding activity to GST-ECT-2 DHPH but do not bind to GST alone . ( B ) GST-pulldown assay between GST-ECT-2 DHPH ( wild-type and G707D variant ) and MBP- ECT-2-N ( wild-type and E129K variant ) . The wild-type N-terminus associates with wild-type and G707D DHPH , whereas ECT-2 N E129K is defective in binding to wild-type DHPH . Neither MBP-ECT-2-N variant binds to GST alone . ( C ) Proteins used in binding assays . DOI: http://dx . doi . org/10 . 7554/eLife . 08898 . 02610 . 7554/eLife . 08898 . 027Figure 7—figure supplement 2 . Proposed states of the ECT-2/CYK-4 complex . ( A ) Correspondence between the cyk-4 genotypes , the proposed forms of the CYK-4:ECT-2 complexes , and the observed levels of contractility . ( B ) Schematic depicting the mechanism of activation of p115-RhoGEF by its RGS homology domain and G alpha 13 ( based on small angle x-ray scattering analysis ) compared to the proposed working model for ECT-2 activation . DOI: http://dx . doi . org/10 . 7554/eLife . 08898 . 027 As ect-2 ( xs110 ) and ect-2 ( xs111 ) suppress the phenotypes caused by mutations in the CYK-4 GAP domain , we sought to understand the biochemical basis of activation by the proteins they encode , ECT-2G707D and ECT-2E129K , respectively . Given that the E129K mutation lies near the N-terminal BRCT domain ( Figure 7A ) , we hypothesized that it could interfere with ECT-2 autoinhibition . To test this possibility , we assayed for binding between the N- and C- termini of ECT-2 . Wild-type N- and C- termini form a complex that is readily detected in vitro . However , the E129K substitution , but not G707D , significantly reduces binding of the ECT-2 N- and C-termini ( Figure 7C , Figure 7—figure supplement 1 ) , suggesting that this allele functions by relieving autoinhibition . The G707D mutation in ECT-2 is located in the PH portion of the RhoGEF domain ( Figure 7A , Figure 6—figure supplement 2 ) . In principle , this mutation could promote RhoA activation by a number of mechanisms including activation of the GEF domain , relieving autoinhibition of ECT-2 , stabilizing the interaction with CYK-4 and/or stabilizing the interaction of ECT-2 with the plasma membrane . We did not observe a change in the association of ECT-2_N with ECT-2_CG707D , suggesting that the mutation doesn't relieve autoinhibition . However , as the mutated residue maps to a helix that lies near RhoA in a co–crystal structure of a related RhoGEF ( Figure 6—figure supplement 2 ) , we tested whether it activates RhoGEF activity . We assayed ECT-2 GEF activity in an in vitro exchange assay . The ECT-2G707D variant exhibits a modest increase in GEF activity compared to wild-type ECT-2 over a range of concentrations ( Figure 7D ) , perhaps by increasing the affinity of ECT-2 for RhoA . Diverse mechanisms ensure that the cytokinetic contractile ring assembles at the cell equator following chromosome segregation . These regulatory mechanisms converge to promote local accumulation of active RhoA at the cell equator which is an essential prerequisite for contractile ring assembly . Whereas it is widely accepted that the RhoGEF ECT-2 is the primary activator of RhoA and the centralspindlin component CYK-4 contributes to RhoA activation , the mechanism ( s ) by which CYK-4 promotes RhoA activation have been rather unclear . Here , we demonstrate that CYK-4 has multiple functional domains that are required for it to promote RhoA activation . In addition to the previously characterized binding interaction with ECT-2 , we show that both the C1 domain and the catalytic activity of the RhoGAP domain of CYK-4 are also required for full activation of ECT-2 . Furthermore , our results indicate in order for the CYK-4 GAP domain to promote RhoA activation , it has to act catalytically on RhoA•GTP . This implies that RhoA plays a role in promoting its own activation . Four results suggest that CYK-4 GAP activity promotes RhoA activation . First , CYK-4 GAP activity is required for the completion of cytokinesis and embryos lacking this activity exhibit reduced levels of RhoA effectors . Second , when the NOP-1-dependent , parallel pathway for RhoA activation is eliminated , the requirement for CYK-4 GAP activity for furrow formation and effector recruitment is greatly enhanced . Third , we have demonstrated a biochemical interaction between the GAP domain of CYK-4 and the GEF domain of ECT-2 . Fourth , the requirement for CYK-4 GAP activity can be alleviated by three independent perturbations that each increase RhoA activity levels . One of the perturbations that suppresses the CYK-4 GAP-deficient phenotype is depletion of the primary RhoA GAP , RGA-3/4 . Embryos defective in RGA-3/4 alone exhibit hypercontractility and are largely inviable; these phenotypes are consistent with the canonical function of a RhoA GAP . However , loss of CYK-4 GAP activity and loss of RGA-3/4 counterbalance each other during cytokinesis . Furthermore , suppression by RGA-3/4 depletion is potent , it can restore cytokinesis in embryos deficient in both NOP-1 and CYK-4 GAP activity in which furrows otherwise barely ingress . A large , unbiased , genome-wide screen for suppressors of cyk-4 ( or749ts ) corroborates the model that CYK-4 GAP activity promotes RhoA activation . We identified two strong , extragenic , gain of function suppressor mutations in the RhoGEF ECT-2 ( Figure 6 ) . Because these suppressors rescue cyk-4 ( or749ts ) and cyk-4R459A to viability , the essential function of the CYK-4 GAP activity must be to promote RhoA activation . These results raise the fundamental question: by what mechanism does the GAP activity of CYK-4 contribute to RhoA activation ? We propose a working model in which the most active form of the ECT-2 RhoGEF is a complex containing ECT-2 and CYK-4 with a molecule of RhoA•GDP bound to the GAP active site ( Figure 7E ) . The GTPase bound to CYK-4 is likely to be RhoA , rather than CED-10/Rac1 or CDC-42 . If CYK-4 had to bind CED-10/Rac1 or CDC-42 , then depletion of those GTPases should impair cytokinesis as severely as a mutation that attenuates GTPase binding by the CYK-4 GAP domain . However , neither CED-10/Rac1 nor CDC-42 is required for cytokinesis , even in NOP-1-defective embryos ( Figure 3 ) , whereas weakening GTPase binding by the CYK-4 GAP domain strongly impacts cytokinesis . The model has a further implication: to form the most active ECT-2 GEF complex , CYK-4 GAP binds RhoA•GTP . Therefore , RhoA•GTP plays a role in RhoGEF activation , suggesting the presence of a positive feedback loop during cytokinesis . This working model is supported by the finding that full activation of RhoA and cytokinesis requires that the CYK-4 GAP domain both bind a Rho family GTPase ( Figure 3 ) and activate its ability to hydrolyze GTP ( Figures 2 , 4 , 5 ) ; indeed mutations in the CYK-4 GAP domain that diminish GTPase binding exhibit a stronger defect in RhoA activation than mutation of the GAP active site ( Figure 5 ) . Our mutational analysis has trapped ECT-2 in four distinct states that form an allelic series ( Figure 7—figure supplement 2A ) . We propose that the least active form of ECT-2 is not bound to CYK-4 and has little GEF activity . Once CYK-4 is phosphorylated , it can be bound by the ECT-2 N-terminal BRCT domains rendering it weakly activated ( equivalent to CYK-4EE ) . This form may also exhibit some interactions between the GEF domain of ECT-2 and the GAP domain of CYK-4 , as these domains can interact in vitro without RhoA present . If CYK-4 can bind to RhoA•GTP , it induces a higher activity state , as evidenced by the increased activity of CYK-4R459A , which is sufficient for zygotic development . Finally , if CYK-4 can induce GTP hydrolysis by RhoA , this results in the fully active ECT-2/CYK-4/RhoA•GDP or ECT-2/CYK-4/RhoA•GDP + Pi state populated by the wild-type protein . We speculate that this complex results in full relief from autoinhibition within the ECT-2 GEF domain . Not only does this model explain why CYK-4 retains GAP activity towards RhoA , it also explains why its ability to inactivate RhoA is attenuated relative to Rac and Cdc42 . High turnover rates of RhoA•GTP induced by CYK-4 might rapidly consume RhoA•GTP at the site of production , yielding a futile cycle of RhoA activation and inactivation . However , this working model must be tested by structural studies and biochemical reconstitution assays that reflect the in vivo situation . Accurate reconstitutions will need to account for the facts that cytokinetic RhoA activation involves the CYK-4 C1 domain ( Figure 1C ) , the ability of CYK-4 to bind to ZEN-4 , and the ability of ZEN-4 to oligomerize ( Basant et al . , 2015 ) . We tested the model that GAP activity of CYK-4 is important to maintain CED-10/Rac1 in an inactive state . Some of our results do support this model , as the failure to complete the first cytokinesis in embryos lacking CYK-4 GAP activity can be partially restored by a loss of function mutation in ced-10/Rac1 ( Figure 4B ) . We therefore tested whether inactivation of CED-10/Rac1 suppresses loss of CYK-4 GAP activity in NOP-1-deficient embryos . GAP-deficient CYK-4 does not promote significant ingression of the cleavage furrow in embryos lacking NOP-1 , irrespective of the presence or absence of CED-10/Rac1 ( Figure 4Biii , v ) . Finally , mutations in CED-10/Rac1 do not suppress the lethality of a temperature sensitive mutation in cyk-4 ( Figure 6—figure supplement 1 ) . Thus , because the active site of the CYK-4 GAP domain is required in the absence of CED-10/Rac1 , CED-10/Rac1 inactivation cannot be the primary function of the CYK-4 GAP domain . Rather , these results suggest a model in which loss of CED-10/Rac1 function causes a reduction in overall cortical tension which , in turn , allows an increase in the extent of NOP-1-dependent furrow ingression ( Loria et al . , 2012 ) . The experiments presented here demonstrate that CYK-4 GAP activity promotes RhoA activation and that this function is essential in early C . elegans embryos and in the adult germline ( Figures 2A , 4B ) . Our experiments also addressed the function of CYK-4 GAP activity post-embryonically . We find that whereas zygotic cyk-4 null embryos die during embryogenesis , expression of catalytically inactive CYK-4 provides significant rescue , supporting development into viable , albeit sterile , adults ( Figure 2A ) . Thus , while early embryos require the GAP activity of CYK-4 , this requirement is relaxed post-embryonically . The requirement for CYK-4 GAP activity can be experimentally eliminated by hyperactivation of ECT-2 or depletion of the RhoA GAP RGA-3/4 ( Figures 5B , 6B ) . Interestingly , RGA-3/4 is primarily expressed in the germline and in early embryos ( NextDB , cited in Schmutz et al . , 2007 and data not shown ) , thus regulated RGA-3/4 expression could contribute to the tissue specific requirements . These findings allow us to reconcile many previous results on the role of the CYK-4 GAP domain during cytokinesis . Some studies provided evidence that the GAP activity is dispensable ( Goldstein et al . , 2005; Yamada et al . , 2006 ) , whereas others suggested it is required for Rac1 inactivation ( D'Avino et al . , 2004; Canman et al . , 2008; Bastos et al . , 2012 ) , RhoA inactivation ( Miller and Bement , 2009 ) , or RhoA activation ( D'Avino et al . , 2004; Zavortink et al . , 2005; Loria et al . , 2012 ) . The first set of results is consistent with the results presented here , as some cell types may not require the GAP activity of CYK-4 for cytokinesis , as seen in post embryonic cells in C . elegans . As numerous studies have shown that active RhoA can indirectly inhibit Rac1 ( see Guilluy et al . , 2011 for review ) , some of the results that point to a role for CYK-4 GAP activity in attenuating Rac1 levels ( Bastos et al . , 2012 ) may be indirectly caused by a reduction RhoA activation or by indirectly controlling cortical tension . Thus , many previous results can be explained without proposing that the CYK-4 GAP domain performs different functions in different organisms or cell types . We do not rule out the possibility that , in certain contexts , CYK-4 or its orthologs negatively regulate Rac or Cdc42 . Recent evidence indicates that the Xenopus ortholog of CYK-4 concentrates at cell–cell junctions and negatively regulates GTPases at that site ( Breznau et al . , 2015 ) . Further work is required to resolve why CYK-4 acts as a positive regulator of RhoA in C . elegans embryos and a negative regulator in Xenopus embryos ( Miller and Bement , 2009; Breznau et al . , 2015 ) . The signaling mechanisms we have discovered in cytokinesis have analogies in other signaling pathways . Our favored model , in which RhoA promotes its own activation , is reminiscent of the positive feedback in Cdc42 activation during yeast budding ( Howell and Lew , 2012 ) and the activation of the SOS1 RasGEF domain by a molecule of Ras•GTP that serves as an allosteric activator ( Gureasko et al . , 2008 ) . Interestingly , ECT-2 has also been implicated in SOS regulation ( Canevascini et al . , 2005 ) . Likewise , CYK-4 is not the only protein with a GTPase activating domain that plays a role in promoting GEF activity . A similar function , in cis , has been seen in p115 RhoGEF , which is activated by Gα13 ( Chen et al . , 2012 ) . In this case , the binding of a molecule of Gα13 to an allosteric site on the RhoGEF domain of p115 is stabilized by p115's N-terminal RGS domain ( Figure 7—figure supplement 2B ) . RGS domains accelerate GTP hydrolysis by Gα , that is , they are Gα GAPs ( Tesmer et al . , 1997 ) . As RhoGEF activation is not an obvious function for a RhoGAP domain , additional cases may have gone undetected . RhoA activation is controlled by multiple layers of regulation during cytokinesis . In addition to cell-cycle regulated changes in the phosphorylation state of CYK-4 and ECT-2 that control their binding and localization ( Yüce et al . , 2005; Su et al . , 2011; Zou et al . , 2014 ) , full activation of RhoA also involves membrane binding by CYK-4 ( Figure 1 ) which , in turn , requires centralspindlin oligomerization ( Basant et al . , 2015 ) . Like the requirement for CYK-4 GAP activity , the requirement for the C1 domain of CYK-4 is context dependent . Whereas the C1 domain makes a major contribution to furrow ingression in C . elegans embryos , studies in Hela cells demonstrate that the C1 domain contributes to RhoA activation , but it is not essential ( Lekomtsev et al . , 2012 ) . Thus , centralspindlin has several domains that contribute to maximal activation of ECT-2 . However , not all cell types may require maximal activation of ECT-2 either because of physical properties of the cell ( cell size , cortical tension , and tissue organization ) or because of their biochemical properties ( expression of RGA-3/4 orthologs ) . Nevertheless , the evolutionary conservation of all of these functions suggests that they play critical roles during some stage ( s ) of metazoan development . Animals were grown at 20°C on standard nematode growth media ( NGM ) plates seeded with OP50 Escherichia coli . Some strains were provided by the Caenorhabditis Genetics Center . All strains used in this study are listed in Supplementary file 1 . RNAi was administered by feeding nematodes with E . coli expressing the appropriate double-stranded RNA ( dsRNA ) ( Timmons and Fire , 1998 ) . HT115 bacterial cultures were grown in Luria broth with 100 μg/ml ampicillin overnight at 37°C . Cultures ( 250 μl ) were seeded on NGM plates containing 100 μg/ml ampicillin and 1 mM IPTG and incubated at room temperature for 16 hr . RNAi plasmids were obtained from the library produced by Kamath et al . ( 2003 ) . Young L4 hermaphrodites were picked onto the plates for feeding at 25°C at least 24 hr prior to dissection . For RNAi depletion of temperature-sensitive alleles , L4 larvae were fed for 48 hr at 16°C , then shifted to 25°C for at least 1 hr before imaging . For experiments where two genes were simultaneously knocked down by RNAi , bacterial cultures of E . coli expressing the appropriate dsRNA were mixed in a 1:1 ratio seeded onto NGM plates as described above . If stronger depletion of one of the two genes was desired , embryos were first hatched onto feeding plates targeting the gene . L4 worms were transferred to fresh plates with bacteria expressing dsRNA against both genes . Young gravid hermaphrodites were transferred to fresh seeded NGM plates in triplicate . Remove worms from plates after ∼8 hr of egg laying . The eggs laid on plates were scored manually under dissecting microscope . To determine unhatched embryos , embryos remaining on plates were scored 1 day after the parents were removed . The embryonic lethality percentage is calculated as the number of unhatched embryos divided by the total egg production . To generate CYK-4::GFP MosSCI constructs , ∼2 kb sequences upstream of cyk-4 , cyk-4 genomic DNA tagged with C-terminal GFP coding sequences , and pie-1 3′ UTR sequences were generated by overlapping PCR and inserted to pCFJ150 by SLiCE ( Zhang et al . , 2012 ) . cyk-4 genomic sequences between BamHI and AvrII were recoded to generated RNAi resistant alleles . To introduce cyk-4 mutations , sequences covering mutations were generated by overlapping PCR using pCJF150-cyk-4-gfp as template and the appropriate primers ( see primer sequences in Supplementary file 2 ) : MG4199/MG4276 and MG4200/MG4277 for E448K; MG4199/MG4202 and MG4200/MG4201 for R459A; MG4199/MG4489 and MG4200/MG4488 for K459E/R499E ( EE ) ; MG4199/MG4070 and MG4200/MG4071 for ∆C1 . Overlapping PCR products were inserted into pCFJ15-cyk-4-gfp linearized with NaeI by SLiCE . All constructs were sequence verified . Cas9/sgRNA plasmids were derived from pDD162 vector ( Dickinson et al . , 2013 ) . ect-2 sgRNA target sequences were generated by overlapping PCR using pDD162 as PCR template and the appropriate primers ( see primer sequences in Supplementary file 2 ) , MG4735/MG4773 and MG4774/MG4736 . Overlapping PCR products were inserted into pDD162 linearized with SpeI/BsrBI by SLiCE . Transgenic lines expressing single copy CYK-4::GFP or mutant CYK-4::GFP were generated by integrating constructs into the Mos1 element ttTi5605 on chromosome II using the MosSCI method ( Frøkjaer-Jensen et al . , 2008 ) . For oligonucleotide templates ( ODNs ) based CRISPR experiments ( Zhang and Glotzer , 2014; Zhao et al . , 2014 ) , microinjection was performed by injecting DNA mixture into gonad arms of cyk-4 ( or749ts ) young gravid hermaphrodites . Injected cyk-4 ( or749ts ) were maintained at 16°C for 3–4 days then shifted to 25°C until starvation . Viable worms were isolated and subjected to single worm PCR to identify desired mutations . The injection mixture consists of Cas9/sgRNA plasmids and ODNs . The final concentrations of plasmids and ODNs are Cas9/ect-2 sgRNA vector at 50 ng/μl and ect-2 ODN ( MG4801 5′-TTGTATGGTGCCTGATTCATCGTGACGAGCAAGATGGTGACATTGACACAGTCTTCGAAT-3′ ) at 50 ng/μl . To prepare one-cell embryos for imaging , gravid hermaphrodites were dissected into egg salt buffer ( HEPES pH 7 . 4 5 mM , NaCl 118 mM , KCl 40 mM , MgCl2 3 . 4 mM , CaCl2 3 . 4 mM ) on coverslips , mounted onto 2 . 5% agar pads and sealed with vaseline . For Nomarski imaging , embryos were observed with a Zeiss ( Thornwood , NY ) Axioplan II with a 100×/1 . 3 Plan-Neofluar objective . Images were captured with a charge-coupled device ( CCD ) camera ( Imaging Source , Charlotte , NC ) controlled by Gawker ( gawker . sourceforge . net ) . Images were acquired every 5 s and processed with ImageJ ( http://rsbweb . nih . gov/ij ) . For confocal imaging , embryos were imaged with a 63×/1 . 4 oil-immersion lens on ( 1 ) a Zeiss Axiovert 200M equipped with a Yokogawa CSU-10 spinning-disk unit ( McBain , Simi Valley , CA ) and illuminated with 50-mW , 473-nm and 20-mW , 561-nm lasers ( Cobolt , Solna , Sweden ) , or ( 2 ) a Zeiss Axioimager M1 equipped with a Yokogawa CSU-X1 spinning-disk unit ( Solamere , Salt Lake City , UT ) and illuminated with 50-mW , 488-nm and 50-mW , 561-nm lasers ( Coherent , Santa Clara , CA ) . Images were captured on a Cascade 1K EM-CCD camera or a Cascade 512BT ( Photometrics , Tucson , AZ ) controlled by MetaMorph ( Molecular Devices , Sunnyvale , CA ) . Image processing was performed with ImageJ . Time-lapse acquisitions were assembled into movies using Metamorph and ImageJ . To measure furrow ingression kinetics , a single central plane image of GFP::PH or mCherry::PH was acquired at 10 s intervals starting at anaphase as assessed by the CYK-4::GFP or mCherry::HIS-58 signal . The position of the furrow was assessed in each frame by manual tracking of GFP::PH or mCherry::PH signal . The extent of ingression in each frame was calculated as d/w , where w is the total width of the embryo and d is the distance between the furrow tips . To determine whether furrow ingression kinetics were statistically significant different between multiple genotypes , data sets of normalized cortical distance from 100 s to 410 s after anaphase onset were analyzed with a Kruskal–Wallis non-parametric one-way analysis of variance ( ANOVA ) using Dunnett's multiple comparisons test . To quantitate the abundance of NMY-2::mRFP at the equatorial region , a stack of five planes spanning 2 . 5 μm was captured every 10 s . The Z-stacks were projected using a maximum intensity projection algorithm and corrected for photobleaching . Using custom ImageJ macros , the background signal was measured in a remote region of each frame . A region of fixed size in the equatorial region was thresholded with a minimal value of 1 . 25× background and the total thresholded signal in the region was integrated and normalized to the background . The total value of intensity was summed for a defined number of planes after anaphase onset . To quantitate the abundance of CYK-4::GFP at the furrow tip , a stack of five planes spanning 2 . 5 μm was captured every 10 s . The frame in which furrow ingressed to half of the egg width or the deepest was chosen . The Z-stacks were projected using a maximum intensity projection algorithm and corrected for photobleaching . The background was measured as integrated intensity of a square adjacent to the furrow tip . The GFP intensity was measured as integrated intensity of a square with the same area covering the furrow tip and normalized to the background by subtracting background intensity ( see Figure 1 ) . The coding sequences for CYK-4 C1 and RhoGAP domain ( 342-681aa ) , ECT-2 DH/PH domain ( 356-792aa ) , CED-10/Rac1 and RhoA ( C . e . ) were cloned into the GST expression vector pGEX-4T-tobacco etch virus ( TEV ) , and the coding sequences for ECT-2 BRCT domain ( 1-363aa ) were cloned into MBP expression vector pMAL-c2-TEV . GST- and MBP-tagged proteins were expressed in E . coli strain BL21 by adding 0 . 3 mM IPTG at OD600 reached 0 . 5–0 . 7 at 25°C . Cells were grown for another 4 hr at 25°C and collected . Frozen cells were thawed in lysis buffer ( 50 mM Tris pH 7 . 5 , 150 mM NaCl , 5 mM MgCl2 , 10% Glycerol , 1 mM PMSF , 1 μg/ml leupeptin , 1 μg/ml pepstatin A , 0 . 1% Triton X-100 , 1 mM DTT , 0 . 5 mg/ml lysozyme ) and lysed by sonication . The bacterial lysate was centrifuged at 40 , 000×g at 4°C for 30 min . For GST-tagged proteins , glutathione-Sepharose 4B beads ( bioWORLD ) were added to supernatant and incubated at 4°C for 4 hr . The beads were washed 3× with 50 mM Tris pH 7 . 5 , 150 mM NaCl , 5 mM MgCl2 , 1 mM DTT . Protein-bound beads in were either stored in 50 mM Tris pH 7 . 5 , 150 mM NaCl , 5 mM MgCl2 , 1 mM DTT , 50% glycerol at −20°C , or cleaved from beads by incubation with His-tagged TEV protease at 4°C overnight . TEV protease was removed by incubation with TALON beads ( Clontech ) . Cleaved fusion proteins were stored in 10% glycerol at −80°C . For MBP-tagged proteins , amylose resin ( New England Biolabs ) was added to supernatant and incubated at 4°C for 4 hr . Beads were placed in a poly-prep chromatography column ( Bio-Rad ) , and washed with 12 column volumes of 50 mM Tris pH 7 . 5 , 150 mM NaCl , 5 mM MgCl2 , and 1 mM DTT . Fusion proteins were eluted with 50 mM Tris pH 7 . 5 , 150 mM NaCl , 5 mM MgCl2 , 1 mM DTT , 10 mM maltose , and stored in 10% glycerol at −80°C . C . e . GST-RhoA was loaded with GDP ( Self and Hall , 1995 ) . Beads were washed with low magnesium buffer ( 50 mM Tris pH 7 . 5 , 150 mM NaCl , 1 mM EDTA , 1 mM DTT ) and 1 mM GDP was added . Beads were incubated with shaking at room temperature for 15 min , placed on ice , and 20 mM MgCl2 was added and incubated on ice for 5 min . Beads were washed three times with 50 mM Tris pH 7 . 5 , 150 mM NaCl , 5 mM MgCl2 , 1 mM DTT . GDP-loaded RhoA was cleaved from beads with TEV protease , glycerol added to 10% , flash frozen and stored at −80°C . Fluorescence-based kinetic assays were performed in HORIBA FluoroLog-3 Spectrofluorometer , with fluorescence analog of GTP , mant-GTP ( AnaSpec ) . All nucleotide exchange assays were performed in the presence of 1 μM RhoA-GDP , 200 nM mant-GTP , the indicated concentration of ECT-2 DH/PH domain in 20 mM Tris pH 7 . 5 , 50 mM NaCl , 10 mM MgCl2 , 1 mM DTT , 50 μg/ml BSA , 1% glycerol . The relative fluorescence was monitored for 90 s before adding mant-GTP , and for 510 s after adding mant-GTP; measurements were taken every 15 s . The reaction rate , v , is defined as ∆F/∆t , where F = fluorescence , t = time . CYK-4 GAP ( final concentration from 0 to 800 nM ) and RhoA or Rac1 ( final concentration 9 μM ) were mixed in 1× reaction buffer ( 50 mM Tris pH 7 . 5 , 50 mM NaCl , 5 mM MgCl2 , 1 mM DTT , 1% glycerol ) , then GTP was added to 1 mM to start the reaction . After 30 min , inorganic phosphate was assayed using a malachite green-based assay ( Kodama et al . , 1986 ) ; absorbance was measured with a NanoDrop 2000 spectrophotometer ( Thermo Scientific ) . For time course experiments , 200 nM CYK-4 GAP was added into the reaction and , at the indicated time points , aliquots of the reaction were removed to assess free phosphate . For each binding experiment , purified fusion proteins were added to the protein-bound glutathione-sepharose beads and incubated for 1 hr at 4°C . After three washes in cold wash buffer ( 50 mM Tris pH 7 . 5 , 150 mM NaCl , 5 mM MgCl2 , 1 mM DTT ) , proteins were eluted into loading buffer , separated by SDS-PAGE , and detected by coomassie blue staining .	Cell division is a process in which a cell splits to form two daughter cells . In most cases , the cell first duplicates its genetic material and then the two copies are pulled to opposite ends of the cell . A ring of protein filaments—called the contractile ring—then assembles to form a band around the cell at the site of the division . This ring contracts and the force generated separates the cells in a step known as cytokinesis . A protein belonging to the Rho family , called RhoA , is essential for cytokinesis because it controls the formation of the contractile ring . Rho proteins are switched on by the activities of other proteins called guanine nucleotide exchange factors . Another group of proteins known as ‘GTPase activating proteins’ ( or GAPs for short ) generally act to promote the ability of Rho proteins to turn themselves off . In animals and other multicellular organisms , a GAP called CYK-4 largely concentrates on the spindle midzone , but some of the protein also moves to part of the cell membrane near the future site of cell division . It binds to a guanine nucleotide exchange factor called ECT-2 to switch RhoA on , which in turn promotes the formation of the contractile ring . However , it is not clear why a protein that activates RhoA is also able to trigger its inactivation . In this study , Zhang and Glotzer studied cell division in a roundworm called Caenorhabditis elegans . The experiments show that cells that lacked the GAP activity of CYK-4 were unable to complete cytokinesis because RhoA was not fully switched on . This requirement could be bypassed in cells with mutant forms of ECT-2 that were overactive . Therefore , an activity that was thought to inactivate RhoA actually promotes its activation . Further experiments show that the section ( or ‘domain’ ) of CYK-4 that has GAP activity interacts directly with the guanine nucleotide exchange domain of ECT2 . Zhang and Glotzer suggest that this interaction stimulates ECT2 and thereby promotes the activation of RhoA . Further experiments will reveal how CYK-4 stimulates ECT-2 . In addition , it will be important to determine whether other proteins with GAP domains also work in this unconventional way .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "cell", "biology" ]	2015	The RhoGAP activity of CYK-4/MgcRacGAP functions non-canonically by promoting RhoA activation during cytokinesis
Eukarya , Archaea , and some Bacteria encode all or part of the essential mevalonate ( MVA ) metabolic pathway clinically modulated using statins . Curiously , two components of the MVA pathway are often absent from archaeal genomes . The search for these missing elements led to the discovery of isopentenyl phosphate kinase ( IPK ) , one of two activities necessary to furnish the universal five-carbon isoprenoid building block , isopentenyl diphosphate ( IPP ) . Unexpectedly , we now report functional IPKs also exist in Bacteria and Eukarya . Furthermore , amongst a subset of species within the bacterial phylum Chloroflexi , we identified a new enzyme catalyzing the missing decarboxylative step of the putative alternative MVA pathway . These results demonstrate , for the first time , a functioning alternative MVA pathway . Key to this pathway is the catalytic actions of a newly uncovered enzyme , mevalonate phosphate decarboxylase ( MPD ) and IPK . Together , these two discoveries suggest that unforeseen variation in isoprenoid metabolism may be widespread in nature . Isoprenoids constitute a substantial family of primary and secondary metabolites in all three domains of life . These molecules play essential and specialized roles for their hosts including modulation of membrane fluidity ( cholesterol , hopanoids , squalene ) , chemical defense and communication ( mono- , sesqui- and diterpenes ) , photoprotection and energy transfer ( carotenoids ) ( Lu and Li , 2008 ) , and growth regulation ( giberellins ) ( Hedden and Kamiya , 1997 ) . Additionally , isoprenoids such as quinones , chlorophyll , bacteriochlorophyll , and some cellular proteins are tethered to a polyisoprenoid chain to direct localization to membranes or to potentiate interactions with other biomolecules ( Nowicka and Kruk , 2010; Schafer and Rine , 1992 ) . Isopentenyl diphosphate ( IPP , 1 ) and its isomer , dimethylallyl diphosphate ( DMAPP , 2 ) , are the five-carbon building blocks of all higher order isoprenoids . It is currently accepted that IPP is biosynthesized via one of two metabolic pathways: the mevalonate ( MVA or sometimes MEV ) pathway ( Katsuki and Bloch , 1967; Lynen , 1967 ) or the 1-deoxy-D-xylulose 5-phosphate ( DXP ) pathway ( also known as the 2-C-methyl-D-erythritol 4-phosphate , MEP , or Rohmer pathway ) ( Arigoni et al . , 1997; Eisenreich et al . , 1998; Rohmer , 1999 ) . These two metabolic systems utilize non-homologous enzymes that evolved independently to produce the same 5-carbon end product , IPP ( 1 ) . Typically , a given organism uses one pathway or the other ( Lange et al . , 2000 ) . Eukarya appear to encode the classical MVA pathway ( with some exceptions , see [Cassera et al . , 2004] ) ; plants additionally encode the DXP pathway that operates in the chloroplast . Most bacteria employ the DXP pathway with the exception of several phyla that contain full or partial MVA pathway genes ( Bochar et al . , 1999 , Lombard and Moreira , 2011 ) . While the domain Archaea and the bacterial class Chloroflexi do not encode gene homologs for the DXP pathway , nearly all of these species characterized genomically to date encode an incomplete classical MVA pathway . That is , many of these species lack identifiable genes encoding enzymes for one or both of the terminal steps of IPP biosynthesis through mevalonate . This observation is puzzling due to the essentiality of isoprenoid compounds in all organisms . Recent phylogenetic and experimental data suggest that Archaea ( and probably also the Chloroflexi ) encode an alternative MVA pathway , which bifurcates from the classical pathway following the phosphorylation of mevalonate to phosphomevalonate ( MVAP , also known as mevalonate 5-phosphate , 3 ) ( Grochowski et al . , 2006; Matsumi et al . , 2010; Miziorko , 2010 ) ( Figure 1 ) . Following MVAP biosynthesis , the classical MVA pathway uses phosphomevalonate kinase ( PMK ) to phosphorylate MVAP ( 3 ) producing diphosphomevalonate ( MVAPP , also known as mevalonate 5-diphosphate , 5 ) . MVAPP then undergoes a phosphorylation-dependent decarboxylation catalyzed by mevalonate 5-diphosphate decarboxylase ( MDD , also known as MDC or DPM-DC ) producing the key five-carbon isoprenoid building block , IPP ( 1 ) . 10 . 7554/eLife . 00672 . 003Figure 1 . The classical and alternative MVA pathways . Both branches of the MVA pathway begin with acetyl-CoA ( and acetoacetyl-CoA ) and proceed through a series of enzymatic reactions involving 3-hydroxy-3-methylglutary-CoA Synthase ( HMGS ) , 3-hydroxy-3-methylglutary-CoA Reductase ( HMGR , the presumed early rate-limiting step ) , and mevalonate kinase ( MVK ) before branching . At the bifurcation , the canonical MVA pathway , highlighted by light blue arrows , guides MVAP ( 3 ) through an additional phosphorylation reaction followed by a phosphorylation-dependent decarboxylation carried out by phosphomevalonate kinase ( PMK ) and diphosphomevalonate decarboxylase ( MDD ) , respectively . The alternative MVA pathway , highlighted with light brown arrows , hypothetically decarboxylates MVAP ( 3 ) prior to the phosphorylation reaction carried out by IPK but the former step has not been discovered until now ( Grochowski et al . , 2006 ) . The enzymes MVK , PMK , MDD , and IPK all consume ATP during catalysis . All enzymes are shown in green type . Statins serve as inhibitors of HMGR as highlighted . DOI: http://dx . doi . org/10 . 7554/eLife . 00672 . 003 The alternative MVA pathway is posited to reverse these steps; decarboxylation of MVAP ( 3 ) followed by phosphorylation of isopentenyl phosphate ( IP , 4 ) ( Figure 1 ) . The first step would require a currently undetected enzyme activity , MVAP decarboxylase ( MPD ) . The final phosphorylation activity has been characterized in Archaea and is catalyzed by isopentenyl phosphate kinase ( IPK ) , an ATP-dependent kinase that phosphorylates IP ( 4 ) forming IPP ( 1 ) ( Figure 1 ) . Until now , the IPK gene has not been studied among the Chloroflexi bacteria or within eukaryotic organisms presumably encoding all the enzymes of the classical MVA pathway . We performed extensive phylogenetic analyses to reveal a spotty distribution of IPK-bearing organisms in animals , several fungi , all plants , and some bacteria ( including those from the class Chloroflexi ) . Following heterologous expression , purification , and kinetic characterization of several proteins encoding representative IPK-like genes from each domain of life , we show that fully active and highly specific IPKs unexpectedly exist outside of Archaea . One such fully active IPK was characterized from the Chloroflexi bacteria , Roseiflexus castenholzii . Interestingly , by classical genome annotation , the Chloroflexi bacteria , as well as certain other archaeal and bacterial groups , contain only a subset of components of the classical ( PMK- , MDD-utilizing ) and alternative ( MPD- , IPK-utilizing ) MVA pathways , resulting in what appears to be the existence of two incomplete branches of the MVA pathway in one organism ( Figure 1 ) . For example , Chloroflexi bacteria appear to be missing the PMK gene from the classical pathway and the MPD gene from the alternative pathway ( Lombard and Moreira , 2011 ) . This arrangement contrasts with archaeal species , which do not appear to encode either PMK or MDD from the classical MVA pathway . Through biochemical characterization of the enzymes encoded by the remaining MVA pathway genes ( MDD and IPK ) from the Chloroflexi bacterium , R . castenholzii , we unequivocally demonstrate that a functional alternative MVA pathway does exist and is encoded in a completely unexpected manner . In R . castenholzii , the putative MDD of the classical MVA pathway surprisingly is fully functional , and until now , expresses an unknown MPD activity associated with the alternative MVA pathway . Combined with our observation of a fully functional IPK-like gene from this same organism , these results provide the first definitive bioinformatic and experimental evidence for a fully operative alternative MVA pathway in nature . IPK is a member of the amino acid kinase ( AAK ) superfamily . To date , the search for eukaryotic IPK-like genes has been limited since putative eukaryotic IPK homologs are divergent in sequence and can be difficult to distinguish from other AAK superfamily genes ( Lombard and Moreira , 2011 ) . Within IPK , we previously identified a catalytically essential active site histidine residue that is not conserved among AAK family members belonging to the amino acid kinase superfamily ( Dellas and Noel , 2010 ) . Using this structurally and functionally defined residue as an indication of possible IPK activity ( Dellas and Noel , 2010 ) , we next identified putative IPK homologs from all three domains of life . We used PSI-BLAST and profile Hidden Markov Models ( HMMs ) to detect IPK homologs in public protein , expressed sequence tag ( EST ) , and genome databases . IPK homologs bearing the key histidine signature appear in nearly all archaea ( Lombard and Moreira , 2011 , Matsumi et al . , 2010 ) , a cluster of Chloroflexi bacteria ( Lombard and Moreira , 2011 ) , every sequenced green plant genome , and , in an exceptionally sporadic distribution , across most major eukaryotic lineages ( Figure 2 , Table 1 ) . 10 . 7554/eLife . 00672 . 004Figure 2 . Phylogenetic distribution of IPK across the three domains of life . Maximum likelihood tree of selected IPK protein sequences . Eukaryotes are highlighted with blues , selected archaeal clades with grays and a small group of bacteria with purple . The tree is anchored by several bacterial fosfomycin kinases . See Figure 2—source data 1 for an alignment of IPK homologs . See Figure 2—source data 2 for a table of IPK homolog sequences . DOI: http://dx . doi . org/10 . 7554/eLife . 00672 . 00410 . 7554/eLife . 00672 . 005Figure 2—source data 1 . Alignment of IPKs from the three domains of life . DOI: http://dx . doi . org/10 . 7554/eLife . 00672 . 00510 . 7554/eLife . 00672 . 006Figure 2—source data 2 . Additional IPK sequences from an assortment of sequence databases . DOI: http://dx . doi . org/10 . 7554/eLife . 00672 . 00610 . 7554/eLife . 00672 . 007Table 1 . Phylogenetic distribution of IPK in EukaryaDOI: http://dx . doi . org/10 . 7554/eLife . 00672 . 007ClassificationIPK-bearing speciesSpecies with no IPKMammals35 mammalian genomesBirdsGallus gallusTaeniopygia guttataMeleagris gallopavoReptilesAnolis carolensisPhilodryas olfersii ( EST ) AmphibiansNotophthalmus viridescens ( EST ) Xenopus tropicalisFishCallorhinchus milii ( partial IPK from draft genome ) Danio rerioOryzias latipesTakifugu rubripesTetraodon nigrovidrisGasterosteus aculeatusInvertebrate chordatesBranchiostoma floridaeCiona intestinalisSaccoglossus kowalevskiiCiona savignyiEchinoderms ( Deuterostomes ) Paracentrotus lividus ( EST ) ( almost complete assembly ) Strongylocentrotus purpuratus ( almost complete prediction ) ArthropodsHomarus americanus ( EST ) Drosophila sp . ( Cassera et al . , 2004 ) Apis melliferaAedes aegyptiAnopheles gambiaeCulex quinquefasciatusPediculus humanusIxodes scapularisDaphnia pulexOther bilateriansHirudo medicinalis ( EST ) Capitella teletaEisenia fetida ( EST ) Helobdella robustaLottia giganteaSchistosoma mansoniAplysia kurodai ( EST ) Schistosoma japonicumAplysia californica ( partial prediction ) Brugia malayiMeloidogyne incognitaPristionchus pacificusCnidariansAcropora palmate ( EST ) Hydra magnapilliataMontastraea faveolata ( EST ) Nematostella vectensis ( predicted ) Other early metazoansTrichoplax adherensAmphimedon queenslandicaPre-metazoans ( within Holozoa ) Monosiga brevicollisSalpingoeca rosettaCapsaspora owczarzakiFungiSpizellomyces punctatus34 fungal genomesRhizopus oryzaeMucor circinelloidesEpichloe festucae ( EST ) ‡Green plantsall plants , including:Selaginella moellendorffiiPhyscomitrella patensOstreococcus tauriOstreococcus lucimarinusMicromonas pusillaChlamydomonas reinhardtiiVolvox carteriAmoebozoaDictyostelium discoideumPolysphondylium pallidumEntamoeba hislolytica§Entamoeba invadens§Entamoeba dispar§AlveolataPerkinsus marinusPlasmodium sp . ( Lynen , 1967 ) Cryptosporidium sp . ( Hedden and Kamiya , 1997 ) Tetrahymena thermophilaParamecium tetraureliaIchthyophthirius multifilisDiatomsThalassiosira pseudonanaPhaeodactylum tricornutumKinetoplastidaTrypanosoma sp . ( Chew and Bryant , 2007 ) Leishmania sp . ( Chew and Bryant , 2007 ) ExcavatesGiardia lambliaTrichomonas vaginalisOthersMalawimonas jakobiformis ( EST ) Naegleria gruberiAlexandrium tamarenseEST stands for expressed sequence tag . All other IPKs were found in draft or complete genomes . ‡Since IPK was not found in related fungal genomes , this may be a case of horizontal transfer or even EST library contamination . §Greater similarity to IPK from a clade of archaea than to eukaryotes . IPK appears to have been lost independently in many animal lineages ( Figure 2 , Table 1 ) . It is absent from choanoflagellates and sponges , but found in early branching animals such as Trichoplax , cnidarians ( Nematostella , but not Hydra ) , and corals . It is found in bilaterians , including molluscs ( Aplysia sp . , Lottia gigantea ) , annelids ( earthworm and leech ) , and a crustacean ( lobster ) , but not in any insect or nematode . Within deuterostomes , it is found in the sea urchin and sea star , as well as the hemichordate Saccoglossus kowalevskii ( acorn worm ) and the chordate Branchiostoma floridae ( lancelet ) . Within the vertebrates , it is found in a shark ( Callorhinchus milii ) but no teleost fish; in an amphibian ( the newt Notophthalmus viridescens ) but not in frogs; and in a lizard ( Anolis carolinensis ) and a snake ( Philodryas olfersii ) , but not in any bird or mammal . Within the fungi , IPK is present in one of two sequenced chytrids and two other basal fungal genomes , but is otherwise absent . While all identified IPK homologs retain 11 invariant residues ( see Table 2 ) , their sequences are otherwise highly divergent , and their predicted phylogeny agrees only in part with organismal taxonomy . For instance , while all plant and animal IPKs cluster together , sequences from the archaeal class Thermoprotei fall into four distinct clusters on the calculated phylogenetic tree , and in the eukaryotic genus Entamoeba , sequences cluster with the archaeal class Methanomicrobia ( Figure 2 ) . This sporadic grouping pattern suggests horizontal gene transfer and/or parallel evolution is not uncommon and may relate to shared organismal needs in similar ecological niches . 10 . 7554/eLife . 00672 . 008Table 2 . Conserved residues in IPKDOI: http://dx . doi . org/10 . 7554/eLife . 00672 . 008Highly-conserved residuesPartially conserved residuesK6 , G8 , G9 , K15 , G54 , H60 , P140 , G144 , D213 , T215 , G216K221† , G253‡ , T254‡Numbering is in accordance with IPK from M . jannaschii . †Conserved in nearly all species except one possibly due to a gene prediction error . ‡May be invariant , but mis-predicted in several sequences . To confirm the phylogenetic analyses and identification of putative IPK genes , we next overexpressed , purified , and biochemically characterized seven IPK homologs selected from all three domains of life . These include the previously characterized IPK from M . jannaschii ( Grochowski et al . , 2006; Dellas and Noel , 2010 ) and homologs from two other archaeal species , Methanococcus maripaludis and Sulfolobus solfataricus ( one of the few archaeal species that also encodes the complete classical MVA pathway ) , the bacterium Roseiflexus castenholzii ( an organism with an annotated MDD but no obvious PMK ) , and three eukaryotes , Trichoplax adhaerens ( early-branching metazoan ) , Branchiostoma floridae ( chordate ) , and Arabidopsis thaliana ( plant ) . Kinetic experiments were performed using a quantitative lactate dehydrogenase-pyruvate kinase coupled assay demonstrating that all seven IPK homologs catalyze the efficient phosphorylation of IP ( 4 ) to IPP ( 1 ) with high specificity as reported in Table 3 . 10 . 7554/eLife . 00672 . 009Table 3 . Steady-state kinetic constants for enzymes of the MVA pathwayDOI: http://dx . doi . org/10 . 7554/eLife . 00672 . 00910 . 7554/eLife . 00672 . 010Table 3—source data 1 . Steady-state kinetic plots for IPKs listed above each curve . DOI: http://dx . doi . org/10 . 7554/eLife . 00672 . 01010 . 7554/eLife . 00672 . 011Table 3—source data 2 . Steady-state kinetic plots for PMKs listed above each curve . DOI: http://dx . doi . org/10 . 7554/eLife . 00672 . 01110 . 7554/eLife . 00672 . 012Table 3—source data 3 . Steady-state kinetic plots for MDDs listed above each curve . DOI: http://dx . doi . org/10 . 7554/eLife . 00672 . 01210 . 7554/eLife . 00672 . 013Table 3—source data 4 . Stead-state kinetic plots for MPD from R . castenholzii . DOI: http://dx . doi . org/10 . 7554/eLife . 00672 . 013Organismgi numberEnzymeSubstrateKM ( μM ) kcat ( s−1 ) Ki ( μM ) kcat/KM ( s−1μM−1 ) R2 ( † ) M . jannaschii15668214IPKIP4 . 3 ( ± 0 . 6 ) ‡1 . 46 ( ± 0 . 03 ) –§0 . 34 ( ± 0 . 05 ) 0 . 90M . maripaludis132664414IPKIP21 . 4 ( ± 4 . 3 ) 15 . 2 ( ± 1 . 4 ) 877 ( ± 550 ) 0 . 71 ( ± 0 . 16 ) 0 . 99S . solfataricus15897030IPKIP23 . 6 ( ± 4 . 8 ) 0 . 91 ( ± 0 . 05 ) –0 . 04 ( ± 0 . 01 ) 0 . 9215899698PMKMVAP23 . 1 ( ± 3 . 8 ) 1 . 98 ( ± 0 . 10 ) 5500 ( ± 2200 ) 0 . 09 ( ± 0 . 01 ) 0 . 98R . castenholzii156743980IPKIP4 . 3 ( ± 0 . 7 ) 1 . 70 ( ± 0 . 04 ) –0 . 40 ( ± 0 . 06 ) 0 . 96156740939MPD#MVAP152 ( ± 38 ) 1 . 7 ( ± 0 . 1 ) –0 . 011 ( ± 0 . 003 ) 0 . 98156740939MDDMVAPPND#NDNDNDNDB . floridaeNA¶IPKIP13 . 3 ( ± 2 . 0 ) 27 . 2 ( ± 1 . 2 ) 2820 ( ± 1700 ) 2 . 05 ( ± 0 . 32 ) 0 . 98260829481PMKMVAP38 . 0 ( ± 7 . 3 ) 12 . 6 ( ± 0 . 4 ) –0 . 33 ( ± 0 . 06 ) 0 . 93260794527MDDMVAPP15 . 1 ( ± 3 . 7 ) 1 . 36 ( ± 0 . 09 ) –0 . 09 ( ± 0 . 02 ) 0 . 89A . thaliana22329798IPKIP0 . 79 ( ± 0 . 35 ) 1 . 9 ( ± 0 . 2 ) 522 ( ± 381 ) 2 . 4 ( ± 1 . 1 ) 0 . 9515222502PMKMVAP11 . 8 ( ± 2 . 0 ) 20 . 9 ( ± 0 . 7 ) –1 . 77 ( ± 0 . 31 ) 0 . 9715224931MDDMVAPP15 . 7 ( ± 5 . 0 ) 2 . 02 ( ± 0 . 13 ) –0 . 13 ( ± 0 . 04 ) 0 . 89T . adhaerens195996013IPKIP3 . 1 ( ± 1 . 6 ) 2 . 4 ( ± 0 . 2 ) –0 . 77 ( 0 . 40 ) 0 . 79See Table 3—source data 1 , Table 3—source data 2 , Table 3—source data 3 , and Table 3—source data 4 for Michaelis-Menten fitted kinetic curves for each enzyme . †R2 represents goodness of fit for the kinetic curve to the experimental data . R2 = 1 . 0 − ( SSreg/SStot ) , where SSreg = sum of squares , SStot = sum of squares of the distances between each point and a horizontal line passing through the average of all y values . ‡Values in parentheses represent standard error ( or propagation of error ) for each calculated kinetic constant . §Ki constant was not calculated or was not applicable . #Not detected . ¶B . floridae IPK sequence was predicted from scaffold_167 of the genome assembly ( Putnam et al . , 2008 ) using genewise and manual annotation . These IPK-bearing organisms all appear from previous annotations to encode genes associated with the classical MVA pathway . To explore possible MVA pathway bifurcation , we next overexpressed , purified , and biochemically characterized PMK-like and MDD-like proteins from organisms containing biochemically verified IPKs . PMKs from B . floridae and A . thaliana catalyze the efficient phosphorylation of MVAP as annotated ( 3 ) with similar kinetic constants ( Table 3 ) . PMK from S . solfataricus is catalytically active using ATP and MVAP ( 3 ) as substrates , however , its MDD catalyzes the phosphorylation-dependent decarboxylation of MVAPP ( 5 ) to IPP ( 1 ) at a very slow rate . Under steady state conditions , we were unable to accurately determine kinetic parameters for S . solfataricus MDD . Unexpectedly , the MDD homolog from R . castenholzii does not catalyze the decarboxylation of MVAPP ( 5 ) to IPP ( 1 ) at a measurable rate as annotated . Instead , it efficiently catalyzes the phosphorylation-dependent decarboxylation of MVAP ( 3 ) to IP ( 4 ) ( Table 3 ) . This annotated R . castenholzii MDD , in fact , functions in a completely unexpected way as a bona fide MPD assuming the metabolic role of the long sought MPD of the alternative MVA pathway . The determined catalytic constants of R . castenholzii MPD with MVAP suggest it is less catalytically efficient than bona fide MDD enzymes ( Table 3 ) . R . castenholzii , however , is a thermophilic bacterium and assays at its optimal growth temperature ( 50°C and above ) ( Hanada et al . , 2002 ) were not possible due to stability limitations of the coupled assay system . Fluorescence thermal shift assays ( Pantoliano et al . , 2001 ) demonstrate that recombinant R . castenholzii MPD is stable at temperatures exceeding 90°C ( inflection point of the slope of the fluorescence vs temperature curves ( Tm ) = 94°C at pH 5 . 8 ) , indicating that the enzyme maintains structural integrity at temperatures well beyond what it commonly encounters in nature and is most likely maximally active at high temperatures ( Figure 3 ) . 10 . 7554/eLife . 00672 . 014Figure 3 . Fluorescence thermal shift assays of Roseiflexus castenholzii MDD-like MPD . ( A ) Thermograms for R . castenholzii MPD in 100 mM buffer ( pH 3 . 0–3 . 8 citric acid; pH 4 . 0–4 . 8 sodium acetate; pH 5 . 0–5 . 8 sodium citrate; pH 6 . 0–6 . 8 sodium cacodylate; pH 7 . 0–7 . 8 sodium HEPES; pH 8 . 0–8 . 8 Tris-HCl; pH 9 . 0–11 . 0 CAPSO ) colored from red to violet ( acidic to alkaline pH depicted in the inset ) . ( B ) Negative derivatives of the thermograms ( −dF/dT ) color-coded as in ( A ) . ( C ) Tms for each of the curves show in ( A ) and ( B ) plotted as a function of pH . R . castenholzii MPD was unfolded from pH 2 . 2 to 2 . 8 at 20°C ( data now shown ) . DOI: http://dx . doi . org/10 . 7554/eLife . 00672 . 014 To the best of our knowledge , this is the first enzyme discovered to function as a catalytically efficient MPD . Given the annotation bias associated with functional assignment based upon sequence homology alone , this newly discovered MPD activity encoded by an MDD-like sequence suggests that functional expectations of the enzymes comprising the canonical MVA biosynthetic pathway must be reexamined . To confirm that the functional R . castenholzii MPD ( MDD-like protein sequence ) catalyzes the transformation of MVAP ( 3 ) to IP ( 4 ) in a pathway specific manner with a measurable flux , we next reconstituted each pathway in vitro using a five-enzyme reaction mixture of homogenously purified enzymes and substrates . Briefly , MVAP ( 3 ) and ATP were incubated for 30 min with MPD and IPK from R . castenholzii . The reaction mixture was then diluted into a mixture containing all necessary downstream enzymes to biosynthesize a higher order isoprenoid metabolite , namely the GC-detectable sesquiterpene 5-epi-aristolochene ( 5-EA , 7 ) . Similar reactions were performed for the in vitro reconstitution of the classical MVA pathway using PMK and MDD from each respective organism . The in vitro classical MVA pathway reactions for S . solfataricus , A . thaliana , and B . floridae yield 39% , 49% , and 52% conversion of MVAP ( 3 ) to 5-EA ( 7 ) , respectively ( Figure 4 ) . The in vitro classical MVA pathway reaction from R . castenholzii ( using B . floridae PMK to complete the missing step ) yields less then 0 . 2% conversion to 5-EA ( Figure 4 ) . In contrast , the in vitro alternative MVA pathway from R . castenholzii enzymes yields a 2 . 0% conversion of MVAP ( 3 ) to 5-EA ( 7 ) . B . floridae , A . thaliana , and S . solfataricus were tested for a possible alternative MVA pathway using their IPKs and MDDs . These reactions do not yield any detectable 5-EA ( 7 ) , indicating that these MDDs behave as expected from their gene annotations and lack detectable MPD activity . Control reactions lacking ATP , enzyme , or MVAP ( 3 ) do not yield any 5-EA ( 7 ) . 10 . 7554/eLife . 00672 . 015Figure 4 . In vitro reconstitution and GC-MS analysis of the alternative and classical MVA pathways . ( A ) Terminal steps of the MVA pathways shown with enzymes of the alternative ( orange ) and classical ( blue ) pathways depicted as arrows . The putative neofunctionalization of MDD to MPD is highlighted by a grey arrow . ( B ) In vitro assays include either the alternative or the classical enzymes to produce IPP ( 1 ) ( as shown in panel A ) as well as all downstream enzymes , including isopentenyl phosphate isomerase ( IPPI ) to produce DMAPP ( 2 ) , farnesyl diphosphate synthase ( FPPS ) to produce farnesyl diphosphate ( FPP , 6 ) and tobacco 5-epi-aristolochene synthase ( TEAS ) to produce the sesquiterpene product , 5-EA ( 7 ) . Products are separated by GC and detected by MS ionization and fragmentation . Enzymes used are highlighted in turquoise type . ( C ) Results of the in vitro reconstitutions of various enzyme combinations . The y-axis of the graph represents combinations of enzymes shown in panel A and panel B and the % conversion to the expected sesquiterpene end product , 5-EA ( 7 ) shown as grey bars along the x-axis . Abbreviations of organisms are as follows: Bf = Branchiostoma floridae , At = Arabidopsis thaliana , Ss = Sulfolobus solfataricus , and Rc = Roseiflexus castenholzii . Note , RcMPD is colored both orange and blue on the y-axis , depending on whether it is being tested as an MDD ( blue ) or an MPD ( orange ) . DOI: http://dx . doi . org/10 . 7554/eLife . 00672 . 015 The experiments described above indicate that the MDD-like enzyme from R . castenholzii does not accept MVAPP ( 5 ) as a substrate . Instead , R . castenholzii ‘MDD’ functions as an MPD and possesses a newly discovered biochemical activity , namely the utilization of MVAP ( 3 ) as a substrate for the phosphorylation-dependent decarboxylation of MVAP ( 3 ) to IP ( 4 ) . These biochemical results conclusively demonstrate that this MDD-like gene in fact encodes an MPD . Closer examinations of sequences and three-dimensional structures of MDDs highlight catalytic elements that set MPD from R . castenholzii apart from other bacterial MDDs . A structural model of R . castenholzii MPD was computed using Swissmodel ( Arnold et al . , 2006; Kiefer et al . , 2009 ) and the Staphylococcus aureus MDD structure ( PDBID 2HK2 ) as a template ( 26% sequence identity to R . castenholzii MPD ) ( Byres et al . , 2007 ) . Residues surrounding the diphosphate group of 6F-MVAPP ( 5 ) bound to S . epidermidis MDD ( PDBID 3QT7 ( Barta et al . , 2011 ) ) were then contrasted with the corresponding residues in the R . castenholzii MPD model . A comparison of the phosphate-binding residues of the canonical MDD from S . epidermidis to the same positions in R . castenholzii MPD reveals that the R . castenholzii MPD lacks several conserved hydrogen bonding interactions associated with recognition of the terminal phosphate on the diphosphate moiety of MVAPP ( 5 ) ( Figure 5 ) . Second , the R . castenholzii MPD contains additional residues including arginine and lysine side chains that appear poised to bind the single phosphate of MVAP ( 3 ) . Finally , these residues , unique to atypical MDDs such as the R . castenholzii MPD , clash sterically and electronically with the terminal phosphate on the diphosphate moiety of MVAPP ( 5 ) ( Figure 5 , Table 4 ) . 10 . 7554/eLife . 00672 . 016Figure 5 . Structural comparisons of a bona fide MDD and MPD . Interactions between the terminal phosphate of 6F-MVAPP and the surrounding amino acid residues from the crystal structure of S . epidermidis MDD ( PDB ID 3QT7 ( Barta et al . , 2011 ) ) and the 3D model of MPD from R . castenholzii . ( A ) S . epidermidis MDD has multiple interactions with the diphosphate of MVAPP ( 5 ) . Atoms are colored by type with carbon gold . ( B ) The active site model of R . castenholzii MPD lacks many of the key interactions shown by S . epidermidis MDD in panel A . Atoms are colored by type with carbon green . ( C ) Interactions between the monophosphate of modeled 6F-MVAP and the surrounding amino acids in a superposition of the modeled R . castenholzii MPD , backbone atoms and carbon colored green , on the crystal structure of S . epidermidis MDD , backbone atoms and carbon colored gold . In R . castenholzii MPD , two divergent side chains , Arg83 and Lys161 , putatively provide additional electrostatic interactions with the single phosphate group of MVAP ( 3 ) . These amino acid side chains would clash with the second phosphate of MVAPP ( 5 ) . These models suggest that the predicted active site topology of R . castenholzii MPD facilitates substrate recognition of MVAP ( 3 ) through complementary charged and polarized hydrogen bonds and excludes MVAPP ( 5 ) through steric incompatibility with its second phosphate . DOI: http://dx . doi . org/10 . 7554/eLife . 00672 . 01610 . 7554/eLife . 00672 . 017Table 4 . Active site phosphate-binding residues identified in MDDs across Archaea , Bacteria and EukaryaDOI: http://dx . doi . org/10 . 7554/eLife . 00672 . 01710 . 7554/eLife . 00672 . 018Table 4—source data 1 . Amino acid sequence alignments of archaeal MDDs . DOI: http://dx . doi . org/10 . 7554/eLife . 00672 . 01810 . 7554/eLife . 00672 . 019Table 4—source data 2 . Amino acid sequence alignments of bacterial MDDs . DOI: http://dx . doi . org/10 . 7554/eLife . 00672 . 01910 . 7554/eLife . 00672 . 020Table 4—source data 3 . Amino acid sequence alignments of Chloroflexi MDDs . DOI: http://dx . doi . org/10 . 7554/eLife . 00672 . 02010 . 7554/eLife . 00672 . 021Table 4—source data 4 . Amino acid sequence alignments of eukaryotic MDDs . DOI: http://dx . doi . org/10 . 7554/eLife . 00672 . 021MDD-like active site residuesS . epidermidisTyr18Lys21Ile27Ser139Ser141Ser192BacteriaTyrLysX†SerSerSerSulfolobalesTyrLysAsnSerSerSerEukaryaTyrLysIle/AsnSerSerSerMPD-like active site residuesR . castenholziiTyr74Leu77Arg83Ala201Ser203Thr255Haloarchaea/ThermoplasmatalesTyr /PheMet/TyrArgSerSerPolar‡ChloroflexiTyrLeuArg/ThrAlaSerThrSee Table 4—source data 1 , Table 4—source data 2 , Table 4—source data 3 , and Table 4—source data 4 for alignments . †Not conserved among canonical bacterial MDDs . ‡Includes Glu , Asp , Asn , and Ser . Phylogenetic analyses demonstrate that MDDs from the Chloroflexi are distant from those of other bacteria but similar to those of the two archaeal classes that also encode an IPK but no PMK , the Haloarchaea and Thermoplasmata classes . The model described above thus illustrates a logical route through which atypical MDDs may have over time attenuated MVAPP ( 5 ) substrate selectivity and acquired high selectivity for MVAP ( 3 ) . The Chloroflexi are considered to be one of the oldest phyla of photosynthetic organisms ( Mulkidjanian et al . , 2006 , Hohmann-Marriott and Blankenship , 2011 ) . As photoheterotrophs , Chloroflexi use light for energy but cannot fix carbon dioxide as their primary source of carbon ( Bryant and Frigaard , 2006 ) . Of the six bacterial phyla reported to contain MVA pathway-bearing organisms , only the Chloroflexi contain IPK ( Lombard and Moreira , 2011 ) , but lack an obvious PMK . Although it appears from genome annotations that the Chloroflexi encode two incomplete branches of the essential MVA metabolic pathway , we demonstrate in one Chlorflexi bacterium that this MDD-like enzyme in fact acts catalytically as a bona fide MPD . MPD and IPK from this Chloroflexi bacterium , R . castenholzii , complete the assembly of a hitherto unforeseen alternative MVA pathway that catalyzes sequential MVAP ( 3 ) decarboxylation and IP ( 4 ) phosphorylation resulting in the central hub metabolite of all organisms , the five-carbon isoprenoid building block IPP ( 1 ) . The consequences of the neofunctionalization of MDD to MPD are particularly noticeable in the substrate binding regions surrounding the diphosphate moiety of MVAPP ( 5 ) in canonical MDDs . The majority of the residues surrounding the diphosphate of MVAPP ( 5 ) in canonical MDDs are strictly conserved in structure- and sequence-based analyses ( Byres et al . , 2007 ) . On the other hand , archaeal MDDs from Haloarchaea and Thermoplasmata contain MPD-like active site residues associated with MVAP ( 3 ) binding and turnover and their protein sequences cluster with the MPD from Chloroflexi ( Table 4 ) ( Lombard and Moreira , 2011 ) . These MDDs belong to two archaeal classes that encode IPK but are missing PMK , suggesting that they also use this metabolic assembly of the alternative MVA pathway . The correlation between unique active site residues and alteration in substrate preference , MVAP ( 3 ) over MVAPP ( 5 ) , among MPDs from Chloroflexi bacteria , reflects a lineage specific adaptation that allows for the utilization of the alternative MVA pathway; this adaptation may also apply to MPD-like MDDs of the archaeal phyla Haloarchaea and Thermoplasmata but awaits experimental verification currently in progress . Despite key amino acid differences in the active site of the MDD-annotated R . castenholzii MPD , it nevertheless possesses high sequence similarity to canonical MDDs . Moreover , R castenholzii MPD and its closely related homologs annotated in several sequenced Chloroflexi species’ genomes including , Roseiflexus sp . RS-1 , Herpetosiphon aurantiacus , Chloroflexus aggregans , Chloroflexus aurantiacus , and Chloroflexus sp . Y-400-fl ( http://www . genome . jp/kegg-bin/show_pathway ? map00900 ) , are all labeled as MDDs but share the alternative active site features of R . castenholzii MPD . Unexpectedly , this R . castenholzii gene , which is annotated as an ‘MDD’ , catalyzes the phosphorylation-dependent decarboxylation of MVAP ( 3 ) , thus serving as a long sought MPD requiring functional re-annotation in R . castenholzii and most likely the additional Chloroflexi species noted previously . It is possible that this enzyme has only recently emerged and undergone neofunctionaliation to acquire MPD activity from an MDD ancestor within the larger phylum of Chloroflexi heterotrophic photosynthetic bacteria . The emergence of this MPD activity may have also coincided with the loss of the PMK gene from species within the narrower Chloroflexi class and/or acquisition of an IPK gene . Selection for such an alternative metabolic route to the core isoprenoid building block IPP may be due , in part , to a higher demand for metabolic flux through an alternative set of enzymes associated with the Chloroflexi MVA pathway . Indeed , the Chloroflexi class of bacteria appear to possess an expanded repertoire of downstream enzymes of isoprenoid metabolism suggesting a high demand for isoprenoid metabolites ( http://www . genome . jp/kegg-bin/show_pathway ? map00900 ) . While archaeal species contain an active IPK consistent with the presence of the alternative MVA pathway , most lack typical or atypical MDD genes; furthermore , a gene candidate capable of encoding an enzyme able to convert MVAP ( 3 ) to IP ( 4 ) , the first postulated step of the alternative pathway , has not been identified . One candidate gene previously suggested to encode this decarboxylation activity is a gene encoding a dioxygenase-like protein that resides within the MVA operon in most Archaea ( MJ0403 in M . jannaschii ) ( Grochowski et al . , 2006 ) . MJ0403 encodes a protein that is similar in sequence to subunit B of class III extradiol ring-cleavage dioxygenases and is also homologous to the human protein MEMO ( mediator of erbB2-driven cell motility ) . Thus far , in vitro attempts to demonstrate decarboxylase activity for MJ0403 have failed to show any turnover of MVAP ( 3 ) to IP ( 4 ) even though the protein is quite easily produced heterologously . While most archaea lack the MDD gene , certain species from the archaeal order Sulfolobales ( including the archaeon S . solfataricus ) encode a gene annotated as MDD . In our experiments , S . solfataricus PMK and MDD activities were detected in steady-state kinetic experiments and GC-MS assays , respectively ( Table 3 , Figure 4 ) . A recent publication on the characterization of the classical MVA pathway enzymes from crude extracts of S . solfataricus supports the presence of an active classical MVA pathway in the order Sulfolobales , consistent with our in vitro results ( Nishimura et al . , 2013 ) . While our experiments demonstrate that IPK is active in an in vitro assay , their experiments involving cell-free extracts of S . solfataricus indicate that IPK activity is undetectable . Nevertheless , these combined results apply to only a small set of Archaea , and the question remains as to how most archaeal species convert MVAP ( 3 ) to the essential metabolite , IPP ( 1 ) . Eukaryotic genomes examined to date appear to encode a classical MVA pathway with few exceptions ( Cassera et al . , 2004 ) . We did however identify a spotty distribution of eukaryotes that contain IPK , and from those examined thus far , corresponding IPK activity that could be associated with an alternative route to IPP ( 1 ) through the MVA pathway . Since many eukaryotes encode putative enzymes of unknown function , they may also encode a cryptic MPD that would serve in an alternative biosynthetic route to IPP ( 1 ) . All IPKs tested thus far were fully functional , demonstrating that true IPKs persist across most eukaryotic lineages , while some have been lost during rare evolutionary events , probably due to partial redundancy with the MVA pathway . The unusual phylogeny of IPK coupled with its membership in a family of kinases that phosphorylate such a broad range of substrates leave open the possibility that IPK may assume varied physiological roles , including phosphorylation of an IP-like substrate , IP recycling , and/or most notably , access to a pool of IPP through IP-IPP homeostatic control . While the IPK gene is present within a spotty distribution of eukaryotes , the gene appears to be universally retained across the green plant lineage , which suggests that it plays a more universal role within the plant kingdom . Plants are unique in that they are currently known to encode two IPP-synthesizing pathways , the DXP pathway localized to the chloroplast and the MVA pathway localized to the peroxisome and cytoplasm ( Lange et al . , 2000 , Leivar et al . , 2005; Nagegowda et al . , 2005; Hsieh et al . , 2008 ) . These pathways assume distinct metabolic roles within general isoprenoid biosynthesis ( Nes and Venkatramesh , 1999; Eisenreich et al . , 2001 ) . It would not be surprising to identify yet another variation of the MVA pathway in the green plant lineage . In addition to IPP ( 1 ) recycling through IP ( 4 ) , IPK would afford metabolic control of carbon availability from IP ( 4 ) to IPP ( 1 ) . Noting that the diversity of primary and secondary isoprenoid products produced by plants often localize to subcellular compartments , organelles and specialized cell types ( i . e . trichomes ) , an ubiquitous IPK in plants may regulate spatial and temporal control of isoprenoid diphosphate metabolism destined for a myriad of primary and specialized plant metabolites . These results biochemically elucidate the terminal biosynthetic steps in an alternative or unconventional MVA pathway . The experiments described also definitively demonstrate metabolic routes to IPP ( 1 ) through the classical MVA pathway in previously overlooked organisms by including the kinetic characterization of IPKs , MDDs , MPDs , and PMKs taken from all three domains of life . Significantly , the studies presented provide the first experimental support for the existence of IPK catalytic activity in all three domains of life and a fully functional alternative MVA pathway in the photosynthetic heterotrophic bacterial class Chloroflexi . Archaeal IPK genes from M . jannaschii , M . maripaludis C5 , and S . solfataricus P2 , in addition to MDD and PMK genes from S . solfataricus P2 , were cloned from genomic DNA from American Type Cell Cultures ( ATCC , Manassas , VA ) as previously described for M . jannaschii ( Dellas and Noel , 2010 ) into a pET28a ( + ) vector containing a thrombin-cleavable N-terminal 8-His tag . IPK , MDD , and PMK genes from A . thaliana , T . adhaerens , B . floridae , and R . castenholzii were ordered as synthetic genes from Genscript ( Piscataway , NJ , USA ) and sub-cloned using Gateway technology from Invitrogen ( San Diego , CA , USA ) into pHIS9GW , an in-house pET28-based vector modified to contain a thrombin-cleavable 9-His tag . All proteins were expressed according to a previously described procedure with several modifications ( Dellas and Noel , 2010 ) . Generally , each plasmid containing the gene of interest was transformed into BL21 ( DE3 ) cells ( Novagen® , Germany ) , grown at 37°C in 1 L cultures of TB media to an OD600 nm of 1 . 0 , induced with 1 mM IPTG , and grown overnight at 20°C . All proteins were purified similarly and as previously described ( Dellas and Noel , 2010 ) ; however , only the M . jannaschii protein was incubated at 80°C during purification . Additionally , the 9-His tag was removed from R . castenholzii MPD with thrombin for kinetic assays . MPD was then further purified by anion exchange on a Mono Q column ( GE healthcare , Wauwatosa , WI ) with a linear gradient of 0 M–1 M NaCl in 50 mM Tris-HCl , pH 8 . 0 , over 40 column volumes and ultimately by size exclusion chromatography on Superdex 200 column ( GE healthcare ) developed in 50 mM Tris-HCl , pH 8 . 0 , 0 . 5 M NaCl and 1 mM DTT . Kinetic measurements were performed on IPKs from M . maripaludis , S . solfataricus , and B . floridae using a coupled pyruvate kinase–lactate dehydrogenase assay as previously described that employs IP concentrations ranging from 2 µM to 1 mM ( Dellas and Noel , 2010 ) . The substrate IP was purchased from Isoprenoids , LLC ( >95% purity ) ( Tampe , FL ) . Steady-state kinetic curves were fitted using Prism ( GraphPad Software Inc . , San Diego , CA , USA ) to compute KM , kcat , and where appropriate , Ki . Activity measurements were performed for T . adhaerens and A . thaliana using the coupled assay at four different IP concentrations ( 2 µM , 10 µM , 50 µM , and 100 µM ) in triplicate . Kinetic measurements were performed on MDDs from B . floridae and A . thaliana as discussed above with concentrations of ( RS ) -MVAPP ( 95% purity , Sigma , St . Louis , MO ) ranging from 2 µM to 1 mM . Kinetic measurements were performed on MPD from R . castenholzii with concentrations of ( RS ) -MVAP ( Sigma , 95% purity ) ranging from 4 µM to 4 mM and PMKs from B . floridae , A . thaliana , and S . solfataricus with concentrations of ( RS ) -MVAP ranging from 5 µM to 2 . 5 mM . R . castenholzii MPD Tms were calculated from pH 2 . 2 to 11 . 0 in 100 mM buffer ( pH 2 . 2–3 . 8 citric acid; pH 4 . 0–4 . 8 sodium acetate; pH 5 . 0–5 . 8 sodium citrate; pH 6 . 0–6 . 8 sodium cacodylate; pH 7 . 0–7 . 8 sodium HEPES; pH 8 . 0–8 . 8 Tris-HCl; pH 9 . 0–11 . 0 CAPSO ) . Assays were carried out in white 96-well plates in a LightCycler 480 II ( Roche Applied Science , Indianapolis , IN ) . Each well contained a 20 μl total volume , made up with 2 μl of 320 × SyproOrange Dye ( Invitrogen , Carlsbad , CA ) and 18 μl of MPD ( 2 μM ) in 100 mM of the buffers listed above . The plate temperature was ramped from 20 to 99°C with 10 data points acquired per degree . SyproOrange dye ( Invitrogen ) was excited at 483 nm and fluorescence intensity ( F ) detected at 568 nm using the dynamic integration mode ( max integration time , 999 ms ) . Tms were obtained at temperatures ( T ) where the derivative of the thermograms ( −dF/dT ) was minimum . All GC-MS reconstitution assays were carried out in two steps . First , 1 µM of each of two enzymes ( PMK and MDD for the classical MVA pathway or MDD and IPK for the alternative MVA pathway ) was incubated for 30 min with 500 µM ( RS ) -MVAP , 4 mM ATP , and 10 mM MgCl2 buffered with 50 mM Tris-HCl , pH 8 . 0 . Second , 40 µl of this reaction was transferred to a glass vial containing 10 mM MgCl2 buffered with 50 mM Tris-HCl , pH 8 . 0 , containing at least a 150-fold excess of each of the following enzymes: Escherichia coli isopentenyl diphosphate isomerase ( IPPI ) , E . coli farnesyl diphosphate synthase ( FPPS ) , and tobacco 5-epi-aristolochene synthase ( TEAS ) . This reaction mixture was overlaid with ethyl acetate , incubated overnight , and vortexed to extract hydrocarbons from the aqueous layer the next day . Quantitative GC-MS analyses were performed as previously described ( Dellas and Noel , 2010 ) . All values were compared to a control reaction , where FPP was added in place of MVA pathway enzymes at appropriate concentrations to simulate complete turnover of MVAP . Public protein , cDNA , EST and genomic databases were searched for IPK homologs using individual IPK protein sequences , and profile Hidden Markov models built from several individual IPK clades . Genes were predicted from genomic sequences using Genewise ( Birney et al . , 2004 ) and TimeLogic GeneDetective ( Active Motif Inc . , Carlsbad , CA ) programs , with manual editing . Protein sequences were aligned with Muscle ( Edgar , 2004 ) and edited with ClustalX ( Larkin et al . , 2007 ) and JalView ( Waterhouse et al . , 2009 ) . Figure 2 was created using PhyML ( Guindon et al . , 2005 ) using the SPR model and rooted with fosfomycin kinase sequences . Manual editing was used to merge EST sequences and gene predictions , to correct frameshifts , and to fuse one gene split across two contigs . Discrepancies between individual ESTs were resolved to maximize sequence similarity to highly similar homologs . IPK homologs from the Archaea domain were found in all but three of the 74 complete archaeal genomes found in the Integrated Microbial Genomes ( IMG ) database as of 8 Mar 2010 ( Markowitz et al . , 2008 ) . The exceptions are S . acidocaldarius and S . tokodaii , and Nanoarchaeum equitans , a symbiont archaeon with a reduced genome . Within the Bacteria domain , clear IPK homologs were only found in all five sequenced genomes of the class Chloroflexi , but not within other classes of the phylum Chloroflexi . Divergent homologs were found in Streptomyces wedmorensis , Streptomyces fradiae and one strain of Pseudomonas syringae ( all probably fosfomycin kinases ) , and Shewanella denitrificans . The P . syringae gene is found only in a contig from strain PB-5123 , and not several other sequenced strains . The sequence contains a frameshift within the ORF and lacks the H60 residue , both of which may be the result of sequencing errors . In the Eukarya domain , searches for IPK homologs were made using the non-redundant amino acid ( NRAA ) Genbank database ( Benson et al . , 2010 ) , the database of expressed sequence tags ( dbEST ) ( Boguski et al . , 1993 ) , and a wide variety of genome databases , including those at Ensembl ( www . ensembl . org ) ( Birney et al . , 2004 ) , Joint Genome Institute ( JGI , genome . jgi-psf . org/ ) , Baylor College of Medicine ( www . hgsc . bcm . tmc . edu ) , Sanger Institute ( www . genedb . org/ ) and the Broad Institute ( www . broadinstitute . org ) . Searches were carried out with a series of IPK homologs ( blastp against predicted peptides , tblastn against genome ) using a hidden Markov model profile searched against the genome of interest using Gene Detective . Alignments of MDDs were performed using a combined structure and sequence-based approach . Representatives from each of four groups ( eukaryotes , archaea , bacteria , and Chloroflexi ) were modeled using Swissmodel and superimposed to identify aligning active site residues ( Table 4 ) . Each of these four groups were then aligned with other sequences from the same group using the programs Muscle ( Edgar , 2004 ) and Jalview ( Waterhouse et al . , 2009 ) to generate four separate alignments .	Living things make thousands of chemicals that are vital for life , and are also useful as medicines , perfumes , and food additives . The largest family of these natural chemicals is called the isoprenoids , and members of this family are found in all three domains of life: the eukaryotes ( such as plants and animals ) , the Archaea ( an ancient group of single-celled microbes ) , and bacteria . The isoprenoids are made from a smaller building block called isopentenyl diphosphate , IPP for short , that contains five carbon atoms and two phosphate groups . IPP can be produced in two ways . The classical mevalonate pathway is found in most eukaryotes , including humans; statin drugs are used to inhibit this pathway to treat those with high cholesterol and reduce the risk of heart disease . The second pathway does not use the compound mevalonate and is found in many , but not all , bacteria as well as the chloroplasts of plants . Until recently , however , the enzymes needed for the last two steps of the classical mevalonate pathway appeared to be missing in the Archaea and in some bacteria . Researchers subsequently discovered that an enzyme called isopentenyl phosphate kinase , shortened to IPK , was responsible for one of these two missing steps—the addition of IPP’s second phosphate group . The way this enzyme worked also suggested that there was an alternative mevalonate pathway in which the order of the last two steps was reversed . However , the identity of the enzyme responsible for the other step—the removal of a molecule of carbon dioxide to make the starting material needed by IPK—remained mysterious . Now Dellas et al . have discovered the enzyme responsible for this missing step in Green non-sulphur bacteria , confirming the existence of the alternative mevalonate pathway for the first time . Previously it had been thought that this enzyme acted in the classical mevalonate pathway; but in fact this enzyme has evolved a new function and is not involved in the classical pathway at all . Moreover , Dellas et al . show that Green non-sulphur bacteria , and some eukaryotes , also have functional IPK enzymes . This means that IPK has now unexpectedly been observed in all three domains of life , and hints at another target to medically control mevalonate pathways . The discovery of the missing enzyme in the alternative pathway opens the door to the re-examination of many other living things , to find which have the new pathway and to work out why .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "biochemistry", "and", "chemical", "biology", "genetics", "and", "genomics" ]	2013	Discovery of a metabolic alternative to the classical mevalonate pathway
G protein-coupled receptors ( GPCRs ) signal through allostery , and it is increasingly clear that chemically distinct agonists can produce different receptor-based effects . It has been proposed that agonists selectively promote receptors to recruit one cellular interacting partner over another , introducing allosteric ‘bias’ into the signaling system . However , the underlying hypothesis - that different agonists drive GPCRs to engage different cytoplasmic proteins in living cells - remains untested due to the complexity of readouts through which receptor-proximal interactions are typically inferred . We describe a cell-based assay to overcome this challenge , based on GPCR-interacting biosensors that are disconnected from endogenous transduction mechanisms . Focusing on opioid receptors , we directly demonstrate differences between biosensor recruitment produced by chemically distinct opioid ligands in living cells . We then show that selective recruitment applies to GRK2 , a biologically relevant GPCR regulator , through discrete interactions of GRK2 with receptors or with G protein beta-gamma subunits which are differentially promoted by agonists . G protein-coupled receptors ( GPCRs ) comprise nature’s largest family of signaling receptors and an important class of therapeutic drug targets . GPCRs signal by allostery , and were considered for many years to operate as binary switches that bind to cognate transducer and regulator proteins in a single agonist-induced activated state . Over the past decade an expanded view has taken hold , supported by accumulating in vitro evidence that GPCRs are conformationally flexible ( Lohse and Hofmann , 2015; Mahoney and Sunahara , 2016; Nygaard et al . , 2013; Weis and Kobilka , 2018; Wingler et al . , 2019 ) and a confluence of cell biological and in vivo evidence supporting the existence of functionally selective agonist effects ( Smith et al . , 2018; Urban et al . , 2007; Williams et al . , 2013 ) . According to this still-evolving view , agonists have the potential to promote GPCRs to selectively recruit one transducer or regulator protein over another , introducing bias into the signaling cascade at a receptor-proximal level that is either propagated downstream or eliminated during intermediate transduction steps ( Lau et al . , 2011; Tsvetanova et al . , 2017 ) . Opioid receptors provide a representative example . Interest in selective agonist effects at these GPCRs dates back to the initial demonstration that opioid receptors can be activated by diverse peptide and non-peptide agonists ( Kosterlitz and Hughes , 1977 ) . Early experimental evidence for such selectivity among ligands emerged from the observation of an agonist-induced state of opioid receptors in neuroblastoma cells that discriminates between opioid peptides and opiate alkaloids ( Von Zastrow et al . , 1993 ) . This was followed by the demonstration of agonist-selective control of opioid receptor endocytosis , leading to the identification of functional selectivity among agonists defined by differences in relative ability to drive receptor engagement of G protein relative to beta-arrestin-dependent cellular pathways ( Keith et al . , 1998; Keith et al . , 1996; Whistler et al . , 1999; Whistler and von Zastrow , 1998 ) . This concept further evolved to the present view of biased receptor recruitment of G proteins relative to beta-arrestins , with receptor-proximal selectivity calculated by fitting quantitative measures of downstream pathway or protein response to operational models of receptor-effector coupling ( Schmid et al . , 2017 ) . Two key gaps persist in our present understanding . First , selective protein recruitment by GPCRs in intact cells remains largely calculated rather than directly observed . Accordingly , the understanding of receptor-proximal agonist bias is inherently limited by assumptions of the model used to calculate it ( Kenakin , 2018; Klein Herenbrink et al . , 2016 ) . Indeed , and despite intense efforts motivated by interest in the therapeutic impact of biased agonist effects at opioid receptors ( Johnson et al . , 2017; Schmid et al . , 2017; Whistler et al . , 1999 ) , significant challenges remain in reliably assessing selectivity of receptor-proximal protein recruitment based on downstream cell-based readouts ( Conibear and Kelly , 2019 ) . Second , challenges can arise even using cell-based assays that are direct . For example , multiple methods have been developed to detect GPCR interaction with beta-arrestins in intact cells ( Chen et al . , 2012; Kim et al . , 2017 ) . However , this binding involves multiple biochemical steps and , in particular , it typically requires the receptor to undergo prior agonist-induced phosphorylation ( Eichel et al . , 2018; Gurevich et al . , 1995; Thomsen et al . , 2016 ) . This has been clearly established for opioid receptors ( Whistler and von Zastrow , 1998; Zhang et al . , 1998 ) , for which full interaction with beta-arrestin requires the receptor to be phosphorylated at multiple sites in the cytoplasmic tail through a defined sequence of agonist-dependent reactions which are catalyzed by distinct GPCR kinase ( GRK ) isoforms ( Chiu et al . , 2017; Just et al . , 2013; Lau et al . , 2011; Miess et al . , 2018 ) . Accordingly , beta-arrestin recruitment measured in such assays clearly reflects a process that is considerably more complex than allosteric selection by the receptor . Here we describe an alternative approach to address these knowledge gaps . We delineate a cell-based method to simply assess selective protein recruitment by opioid receptors at the receptor-proximal level , taking advantage of two engineered protein folds established to bind agonist-activated GPCRs in intact cells without requiring or engaging other known cellular proteins ( Stoeber et al . , 2018; Wan et al . , 2018 ) . Using these engineered proteins comparatively as orthogonal receptor-interaction biosensors , we directly demonstrate selectivity in receptor-proximal protein recruitment elicited by various opioid agonists in living cells . We then show how the principle of receptor-proximal protein selection applies in a more complex manner to GRK2 , a biologically relevant regulator . Two agonist-activated opioid receptor complexes have been described in structural detail ( Figure 1A ) , one bound to a nucleotide-free G protein heterotrimer and another to an active state-stabilizing nanobody ( Nb ) ( Huang et al . , 2015; Koehl et al . , 2018 ) . The receptor conformation resolved in each complex is similar but not identical , with Nb and G protein interactions involving distinct molecular contacts on cytoplasmic domains of the receptor . Nbs are inherently orthogonal to intracellular biochemistry but heterotrimeric G proteins engage multiple cellular proteins in addition to activated receptors . Thus we focused on mini-G ( mG ) proteins , engineered versions of the Ras-like domain of G protein alpha subunits which bind directly to activated GPCRs but are not known to engage other cellular proteins ( Nehmé et al . , 2017; Wan et al . , 2018 ) . We assessed binding to receptors in intact cells by redistribution of fluorescently labeled Nb or mG fusion proteins from the cytoplasm to the plasma membrane ( Figure 1B ) . For a mG probe we chose mGsi , derived from the Ras-like domain of Gs alpha but with nine residues at the distal C-terminus replaced by the corresponding residues from Gi alpha1 . These C-terminal residues form a major determinant of G protein coupling specificity ( Conklin et al . , 1993 ) by folding into a helical structure ( alpha-5 helix ) that occupies the agonist-activated GPCR core ( Carpenter and Tate , 2017; Koehl et al . , 2018 ) . Because Gs couples poorly to opioid receptors , we reasoned that a sensor derived from mGsi would primarily detect this interaction . For a Nb probe we selected Nb33 , previously used to detect activated mu ( MOR ) and delta ( DOR ) opioid receptors in living cells ( Stoeber et al . , 2018 ) . Nb33 shares receptor contact residues with Nb39 , a close analog that has been resolved at high resolution in complex with activated MOR ( Huang et al . , 2015 ) and in a similar complex with activated kappa opioid receptor ( KOR ) ( Che et al . , 2018 ) . Because cytoplasmic residues contacted by the Nb in these structures are largely distinct from those engaged by the G protein alpha-5 helix , we reasoned that the Nb-derived sensor has the potential to provide different allosteric information . Fluorescent protein fusions of mGsi or Nb33 localized diffusely when expressed in the cytoplasm of HEK293 cells , and recruitment by receptors was monitored using total internal reflection fluorescence microscopy ( TIR-FM ) in cells co-expressing Flag-tagged KOR ( Figure 1C ) . Importantly , HEK293 cells do not express endogenous opioid receptors or other opioid ligand binding sites , thereby providing a null genetic background on which to directly examine protein probe recruitment mediated specifically by the co-expressed receptor . We observed rapid and robust recruitment of mGsi by KOR upon application of the kappa-selective peptide agonist Dynorphin A ( DynA , Dynorphin 1–17 ) . Recruitment of mGsi was reversible because application of the high-affinity competitive KOR antagonist 5’GNTI resulted in rapid redistribution of the biosensor back to the cytoplasm ( Figure 1D ) . In contrast , mGs was not detectably recruited in response to KOR activation by DynA using the same assay ( Figure 1F ) , verifying assay specificity and that mGsi recruitment is driven primarily by the Gi-derived distal C-terminus . Further , we verified that agonist-induced recruitment of mGsi occurred separately from a change in surface expression of KOR , which was monitored in parallel using anti-Flag antibody ( Figure 1D ) . Nb33 was also rapidly recruited in response to KOR activation by DynA using the same experimental protocol , and this recruitment was also reversible upon antagonist application and occurred without a detectable change in surface receptor expression ( Figure 1E ) . Accordingly , both mGsi and Nb33 can be used as biosensors of ligand-dependent recruitment by KOR in living cells using the TIR-FM assay , and both sensors produce a reversible recruitment signal that is sufficiently robust and fast ( t1/2< 30 s ) to enable reliable detection of protein recruitment without possible complications of later receptor trafficking . We next tested two non-peptide KOR full agonists , U69593 ( U69 ) and U50488 ( U50 ) . We generated concentration-response curves by increasing agonist concentration in a stepwise manner and then adding DynA in excess ( 10 μM ) at the end of each series as an internal reference ( Figure 1G and H ) . Both Nb33 and mGsi were robustly recruited in a concentration-dependent manner in response to DynA and both of the non-peptide full agonist drugs ( Figure 1I–K ) , consistent with the previously established pharmacology of these compounds ( DiMattio et al . , 2015 ) , but we also noted that the concentration-response relationship for mGsi recruitment was consistently left-shifted relative to Nb33 . These results demonstrate that both Nb33 and mGsi are robustly recruited by KOR after activation by peptide and non-peptide full agonists in living cells , but with a potency shift indicating that the interactions are not identical . We then applied the same approach to investigate the effect of the alkaloid agonist etorphine ( ET ) on mGsi and Nb33 recruitment by KOR . ET is an opiate alkaloid drug that is structurally distinct from opioid peptides as well as from U50 and U69 . ET efficaciously promotes G protein activation and signaling but has long been recognized to drive KOR internalization and phosphorylation poorly , supporting its classification as a G protein-biased agonist by operational criteria ( Chu et al . , 1997; DiMattio et al . , 2015; Jordan et al . , 2000 ) . ET behaved as a potent but partial agonist in the mGsi recruitment assay , producing a maximum biosensor recruitment response reaching 67% of that produced by DynA ( Figure 2A and D ) . Remarkably , ET produced little or no recruitment of Nb33 despite a robust response to DynA verified in each assay and in the same cells ( Figure 2B and E ) . This lack of Nb33 recruitment was evident even at very high concentrations of ET ( Figure 2B and C ) , in contrast to mGsi that was potently recruited ( Figure 2C–E ) . Further verifying this difference , selective recruitment of mGsi relative to Nb33 was observed when the biosensors were tagged with distinct fluorophores , co-expressed , and imaged in parallel in the same cells ( Figure 2F ) . Again , mGsi was potently recruited in response to ET but Nb33 was not , despite DynA producing strong recruitment of both probes and in the same cells ( Figure 2G ) . These results indicate that mG and Nb probes can distinguish receptor-proximal agonist effects in intact cells . A simple interpretation of these results is that differential probe recruitment reflects a primary allosteric effect at the level of receptor-proximal protein engagement by the agonist-activated opioid receptor . An alternative possibility is that agonists produce differential probe recruitment as a secondary consequence of agonist-selective post-translational modifications of the receptor . In particular , because agonist-induced internalization of KOR requires multi-site phosphorylation on its cytoplasmic tail , and ET is known to stimulate this phosphorylation less strongly than DynA ( Chen et al . , 2016 ) , we considered the possibility that differential biosensor recruitment occurs secondarily to differential phosphorylation . To test this , we measured biosensor recruitment by a mutant KOR lacking all relevant phosphorylation sites in the cytoplasmic tail ( KOR-TPD for ‘total phosphorylation defective’ , Figure 2H ) . The pronounced difference in mGsi relative to Nb33 recruitment was still observed ( Figure 2I and J ) . Independently verifying this , selective probe recruitment by wild type KOR was not detectably perturbed in the presence of Compound101 ( Figure 2K ) , a chemical inhibitor of GRK2/3 activity known to strongly reduce KOR phosphorylation in HEK293 cells ( Chiu et al . , 2017 ) . Together , these results support the hypothesis that selective recruitment of mG relative to Nb probes occurs as a primary consequence of allosteric protein selection at the receptor , rather than a secondary effect of differential phosphorylation . We next asked if our experimental strategy can also detect differential protein recruitment by MOR . Nb33 is already known to be recruited by agonist-activated MORs ( Stoeber et al . , 2018 ) , and we verified that this is also the case for mGsi . DAMGO , a peptide full agonist of MOR , produced rapid and robust recruitment of mGsi that was rapidly reversed by the competitive antagonist naloxone ( Figure 3A and B ) . Similar to what was observed for recruitment of the engineered protein probes by KOR , the concentration-response relationship for recruitment of mGsi by DAMGO was left-shifted relative to Nb33 ( Figure 3C ) . ET ( also an agonist of MOR ) promoted recruitment of both probes by MOR , and to the same maximum degree when compared to the peptide full agonist ( Figure 3C ) . This contrasts with partial recruitment of mGsi and no detectable recruitment of Nb33 by KOR ( Figure 2 ) , indicating that differential recruitment of the engineered protein probes by opioid receptors is both agonist-dependent and receptor subtype-specific . To expand our search , and taking into account the fact that DAMGO and ET are both generally classified as full agonists at MOR , we next examined morphine and PZM21 . Both of these non-peptide drugs are partial agonists with respect to assays of G protein activation or signaling , but each is derived from a different chemical scaffold and differs in degree of bias estimated using a beta-arrestin recruitment assay ( Manglik et al . , 2016 ) . Using the same experimental protocol , and comparing recruitment promoted by the test ligand relative to the peptide full agonist ( DAMGO ) reference , both morphine and PZM21 produced partial recruitment of mGsi as well as Nb33 ( Figure 3D and E ) . Whereas morphine and PZM21 were similar in the degree of mGsi recruitment that they produced at saturating concentration , morphine was found to be significantly more efficacious than PZM21 in recruiting Nb33 . Together , these results reveal a range of selective protein recruitment effects among chemically diverse MOR partial agonists . The experimental strategy used to compare test agonist effects relative to the peptide reference was robust in practice but , in principle , it could underestimate differences relative to the reference peptide if the test agonist dissociates slowly or has an on-rate much faster than the peptide reference . We found evidence for this when evaluating another chemically distinct MOR partial agonist , the semi-synthetic natural product mitragynine pseudoindoxyl ( MP ) ( Váradi et al . , 2016 ) . Using the sequential agonist addition protocol , MP appeared to be similarly efficacious to DAMGO in promoting recruitment of mGsi because no further increase was elicited by subsequent addition of DAMGO while , in contrast , MP failed to produce any detectable recruitment of Nb33 . However , we noted that DAMGO also failed to promote recruitment of Nb33 in cells that were previously exposed to MP ( Figure 3F ) , despite DAMGO promoting a strong Nb33 recruitment response in cells not previously exposed to MP ( Figure 3B ) . Adding a perfusion wash step , in order to remove excess test agonist between applications , avoided this complication . With this modification , MP was verified to indeed promote mGsi recruitment by MOR , but to a significantly reduced maximal degree relative to DAMGO and without promoting detectable recruitment of Nb33 ( Figure 3G and H ) . These results further expand the range of differential protein recruitment effects documented among chemically diverse MOR agonists . To simplify comparison across agonists and receptors , we defined the maximum recruitment response elicited by each agonist compared to the corresponding peptide full agonist reference ( DynA for KOR and DAMGO for MOR ) as a relative ‘intrinsic activity’ for each agonist ( Figure 4A ) . We then plotted these relative values for each biosensor ( Figure 4B ) . Some non-peptide agonists were indistinguishable from the reference peptide by this analysis , recruiting both protein probes to a similar maximal degree ( corresponding to an ‘I . A . ’ value of 1 for both probes ) , but others departed from the diagonal . This is not consistent with the traditional concept of partial agonism based on a unitary agonist-induced receptor ‘on’ state , which would predict the recruitment responses elicited by all agonists to fall along the diagonal . Rather , the present results support the view that opioid receptors are more flexibly activated , enabling them to selectively recruit one interaction probe over another in living cells . They further suggest that the ability to promote selective protein recruitment is widespread among chemically diverse opioid agonists ( Figure 4C ) . While we found the engineered proteins useful as orthogonal probes to unambiguously assess receptor-proximal recruitment in living cells , their disconnection from endogenous cellular machineries and pathways means that they are not directly related to function . Accordingly , we next asked if agonist-selective protein recruitment applies to a physiologically relevant GPCR-interacting protein . We focused on GRK2 because this kinase is known to be important for generating agonist-selective patterns of multi-site phosphorylation in the MOR cytoplasmic tail , which convey biased effects downstream from the receptor by distinguishing engagement of beta-arrestins and regulating receptor entry into the endocytic network ( Just et al . , 2013; Lau et al . , 2011 ) . We were also intrigued by GRK2 because it is recruited by activated GPCRs through multiple interactions , including with the activated GPCR and with beta-gamma subunits that are exposed on the inner membrane leaflet following activation of the G protein heterotrimer ( DebBurman et al . , 1995; Lodowski et al . , 2003; Figure 5A ) . We began by examining a functional GFP-fusion of GRK2 using the same TIR-FM imaging assay used to monitor orthogonal probe recruitment . We focused on comparing the effects of ET relative to DynA on KOR because these agonist-receptor pairs appeared to differ most dramatically based on the orthogonal biosensor recruitment assay ( Figure 4 ) . DynA promoted rapid , concentration-dependent recruitment of GRK2 to the plasma membrane ( Figure 5B ) while ET , despite being highly potent , produced a degree of GRK2 recruitment clearly lower than that produced by DynA ( Figure 5C and D ) . This difference was not a secondary effect of receptor phosphorylation because ET also produced less maximal GRK2 recruitment than DynA using the phosphorylation-defective mutant KOR-TPD in place of KOR ( Figure 5E and F ) . Although ET promoted recruitment of full-length GRK2 less strongly than DynA , these agonists produced similarly strong recruitment of a probe corresponding to the isolated C-terminal PH domain from GRK2 that interacts with G beta-gamma ( Figure 5G ) . This suggested that GRK2 binding to G beta-gamma subunits , enabled by G protein activation triggered by either agonist , is responsible for partial recruitment promoted by ET . We independently verified this conclusion by returning to assay of full length tagged GRK2 , and testing the effect of blocking Gi activation by pre-exposing cells to pertussis toxin ( PTX ) . In this condition , ET failed to produce any detectable recruitment of GRK2 . However , as expected , DynA still produced a significant recruitment response in the same cells ( Figure 5H ) , but to a reduced degree relative to the recruitment response elicited by DynA in cells not previously exposed to pertussis toxin . The above results indicate that ET and DynA share the ability to promote GRK2 recruitment to the plasma membrane via binding G beta-gamma , and that DynA engages an additional mode of binding that is separate from the G protein and not shared with ET . We hypothesized that this interaction occurs with the activated opioid receptor itself . In order to test this , we devised an assay to resolve GRK2 recruitment to the plasma membrane from GRK2 binding directly to the receptor . To do so , we clustered receptors on the cell surface using an antibody cross-linking protocol , forming clusters that appeared in TIRF images as discrete spots of laterally concentrated KOR ( Figure 6A and B , ‘KOR’ panels ) . We then used this characteristic appearance to distinguish GRK2 recruitment to KOR-containing clusters from recruitment to the surrounding plasma membrane separately from KOR clusters . As expected , in the absence of agonist GRK2 was primarily distributed in the cytosol and not detectably associated with KOR ( Figure 6B left , ‘GRK2’ panel ) . Within ~1 min after application of DynA , GRK2 specifically accumulated at the KOR-containing clusters ( Figure 6B right ) . In contrast , application of ET produced a diffuse increase in GRK2 fluorescence at the plasma membrane but no specific accumulation at KOR-containing clusters ( Figure 6C ) . Quantification of the GRK2 intensity in KOR clusters relative to the surrounding plasma membrane verified significant accumulation of GRK2 with receptors promoted by DynA but not ET ( Figure 6D ) , despite both agonists promoting diffuse membrane recruitment ( Figure 5 ) . These results support a model of GRK2 engagement driven by discrete biochemical modes which are differentially regulated by agonists: DynA and ET share the ability to promote GRK2 recruitment to the plasma membrane via receptor-activated G beta-gamma , but DynA is different from ET in its ability to additionally promote GRK2 recruitment by binding directly to KOR ( Figure 6E ) . The ability of agonists to impose selectivity on protein recruitment by GPCRs has been proposed for many years and is a core hypothesis underlying the present concept of biased agonism ( Schmid et al . , 2017; Smith et al . , 2018; Urban et al . , 2007 ) , but testing this hypothesis in an intact cellular environment has remained challenging due to the complexity of cellular transduction and regulatory pathways that GPCRs typically engage ( Kenakin , 2019 ) . The present study describes a direct , reductionist approach to this problem based on the application of engineered proteins that bind activated receptors but are not known to bind other cellular proteins . We show that agonists differ in relative ability to drive recruitment of the engineered probes to opioid receptors in living cells , and then delineate how the principle of agonist-selective recruitment applies to GRK2 as a physiologically relevant regulator . Our results indicate that selective recruitment of one cellular protein over another not only occurs in intact cells , but it is widespread and elicited by diverse agonists . All partial agonists examined were found to promote mGsi recruitment more strongly than Nb33 when present at saturating concentration . Further , concentration-response curves for mGsi relative to Nb33 recruitment by opioid receptors were left-shifted even for peptide full agonists . The allosteric nature of GPCR activation is well established , and has been recognized since early studies of receptor coupling to heterotrimeric G proteins in vitro ( De Lean et al . , 1980; Maguire et al . , 1975; Strachan et al . , 2014; Sunahara and Insel , 2016 ) . The present results are fully consistent with this concept , and expand it by providing clear biochemical evidence for discrete protein-engaged receptor states that can be selectively produced by diverse agonists in the complex environment of intact , living cells . The engineered interaction probes that we focused on here demonstrate such an additional level of allosteric selection most simply , but our results delineating differential recruitment of GRK2 by receptors suggest that the same principle applies in a more complex manner to biologically relevant GPCR-interacting proteins . In its present state of development , our approach is limited by the number of orthogonal probes available for assessing protein recruitment . We focused here on two previously validated GPCR-interacting proteins , selected based on existing biophysical evidence that each recognizes different structural features of the activated receptor . It is possible , and we think likely , that still more specificity exists in receptor-proximal protein recruitment . In future studies it will be interesting to develop or adapt additional structurally diverse protein folds to address this question , and to explore additional agonist diversity using the existing probes . For example , it will be interesting to determine if ligands can be found that promote recruitment of Nb33 preferentially to mGsi . An important next step is to delineate the biophysical basis for the observed selectivity of protein recruitment induced by opioid agonists , with differential effects of DynA and ET on Nb33 recruitment to KOR providing a striking example . The present results clearly indicate that the complexes responsible for agonist-selective protein recruitment must be distinct , but leave unresolved the nature of the distinction . One possibility is that distinct allosteric complexes reflect unique conformational ensembles of the receptor . Although currently available structural data for KOR ( Che et al . , 2018; Wu et al . , 2012 ) preclude direct assessment of such conformational differences , prior structures of MOR in complex with either Nb39 ( Huang et al . , 2015 ) ( a close analog of Nb33 ) or heterotrimeric Gi ( Koehl et al . , 2018 ) offer some insight . Both Nb39 and Gi alpha stabilize an active MOR conformation in the intracellular domain; however , the precise conformation of the MOR intracellular loop 3 ( ICL3 ) differs between the two structures . Thus , differential recruitment of Nb33 and mGsi may reflect agonist-selective stabilization of distinct active receptor conformations . Alternatively , agonists may promote receptors to adopt similar active conformations , and distinctions in the kinetics of sensor binding , sensor concentration , and/or sensor affinity contribute to differential recruitment . Future studies , combining biophysical and cell biological approaches , will be needed to answer this question . We also note that these classes of mechanism are not mutually exclusive , and think it is likely that both contribute to agonist-selective allosteric effects observed in intact cells . It will be particularly interesting to extend the present approach toward examining kinetic aspects of selective protein recruitment by receptors . We found that the orthogonal probes produce a time-invariant recruitment response within ~30 s after agonist application . This enables the approach to be used as an end-point assay scalable to a drug screening platform , and we focused on steady state recruitment values in the present work for simplicity . However , in light of clear and long-standing evidence for kinetic differences in agonist action at GPCRs ( Klein Herenbrink et al . , 2016; Swaminath et al . , 2004 ) , we anticipate that time-dependent analysis of probe recruitment will provide additional insight into selectivity among agonists . In sum , and viewed more broadly , the present results reinforce an emerging understanding that GPCRs operate as allosteric machines with the potential to communicate significantly more information about local chemical environment than the mere presence or absence of a cognate agonist ( Costa-Neto et al . , 2016; Kenakin , 2019 ) . We propose from the present observations that the mGsi probe reports allosteric effects relevant to G protein engagement by opioid receptors , and that the Nb probe reports additional effects relevant to GRK engagement . Our results further support the hypothesis that agonist bias , now generally defined by operational criteria , can be deconvolved into discrete receptor-proximal molecular selection events . The present study makes initial inroads toward decoding this underlying ‘machine language’ of GPCR signaling , and thus toward precisely delineating how much chemical information content receptors actually convey physiologically . HEK293 ( CRL-1573 , ATCC , female , mycoplasma-tested ) were cultured in Dulbecco’s modified Eagle’s medium ( DMEM , GIBCO ) , supplemented with 10% fetal bovine serum ( UCSF Cell Culture Facility ) . Stably transfected HEK293 cells expressing N-terminally FLAG-tagged MOR or KOR were cultured in the presence of 250 μg/ml Geneticin ( Gibco ) . For transient DNA expression , Lipofectamine 2000 ( Invitrogen ) was used according to manufacturer’s instructions . For live cell imaging , cells were plated on poly-L-lysine-coated 35 mm glass-bottomed culture dishes ( MatTek Corporation ) 48 hr before the experiments . Cells were transfected 24 hr prior to imaging . Per 35 mm culture dish , 200 ng DNA was used for mGsi and Nb33 , 300 ng DNA was used for GRK2 constructs and 1 . 2 μg DNA was used for receptor constructs . GRK2-EGFP and GRK2-pmApple were created by amplifying murine GRK2 and GFP or pmApple DNA by PCR and inserting GRK2 and the respective fluorescent protein using In-Fusion cloning into pCAGGS-SE cut with KpnI and EcoRI . Super ecliptic pHluorin ( SEP ) -KOR was generated by PCR amplification of SEP and KOR , and insertion using In-Fusion cloning into pCAGGS-SE cut with KpnI and EcoRI . ssfKOR-TPD was generated by In-Fusion cloning of three PCR fragments that cover ssfKOR and introduce mutations S356A , T357A , T363A , and S369A . Live cell image series measuring protein recruitment to the plasma membrane were performed at 37°C using a Nikon Ti-E microscope equipped for through-the-objective TIR-FM with a temperature- , humidity- and CO2-controlled chamber ( Okolab ) , objective heater , perfect focus system , and an Andor DU897 EMCCD camera . Images were obtained with a 100 × 1 . 49 NA Apo TIRF objective ( Nikon ) with solid-state lasers of 488 , 561 and 647 nm ( Keysight Technologies ) . Before imaging , receptors at the cell surface were labelled with M1 monoclonal FLAG antibody ( 1:1 , 000 ) conjugated to Alexa647 dye for 10 min at 37°C . Cells were then washed and live imaged in HBS imaging solution ( Hepes buffered saline ( HBS ) with 135 mM NaCl , 5 mM KCl , 0 . 4 mM MgCl2 , 1 . 8 mM CaCl2 , 20 mM Hepes , 5 mM d-glucose adjusted to pH 7 . 4 and 300–315 mOsmol/l ) . Agonists or antagonists were either added by bath application at concentrations indicated in the figure legends or by media perfusion . For the latter , an insert was 3D-printed and placed inside the imaging dish where it left a dead volume of about 300 μL . It was used to perfuse HBS imaging solution with agonists or without agonists ( agonist washout ) at concentrations indicated in the figure legends with a flow rate of 1 . 5 ml/min . To cluster receptors in the plasma membrane , cells transfected with SEP–KOR were treated with a polyclonal rabbit anti-GFP antibody ( 1:100 ) for 15 min at 37°C . Cells were then washed and imaged live in HBS imaging solution . To inhibit GRK2/3 , cells were pre-incubated with Compound101 ( 30 μM ) for 15 min at 37°C and Compound101 was present throughout the imaging experiment . To inhibit KOR coupling to Gai/o , cells were treated with PTX ( 100 ng/mL ) for 16 hr and PTX was present throughout the imaging experiment . For probing protein recruitment to the plasma membrane , HEK293 cells co-expressing the cytosolic protein of interest ( mGsi , Nb33 , or GRK2 ) and MOR or KOR were imaged using TIR-FM . Cells were treated with increasing concentrations of agonist ( bath application ) and imaged at a frame rate of 0 . 2/s ( total movie length 6–8 min ) . Protein intensity during time lapse series was measured using ImageJ . If indicated , values were normalized between 0 ( before agonist ) and 1 ( 10 μM reference agonist ) . Regression curves with Hill slope of 1 were fit using Prism 8 . All quantitative image analysis was performed on unprocessed images using MATLAB ( MathWorks , R2014b ) or ImageJ ( 2 . 0 . 0 ) . For quantifying GRK2-mCherry recruitment to the plasma membrane and receptor clusters , we used a custom written MATLAB script . In brief , a polygon was drawn on the TIR-FM image to encompass the cell of interest . Then , a mask of the receptor clusters was generated by thresholding the SEP-KOR signal within the polygon . The average GRK2-mCherry fluorescence was measured within the cluster mask ( KOR clusters ) and outside of the mask ( membrane ) of the polygon , allowing to calculate the ratio . Quantification was performed in cells imaged before ( t = 0 ) and after ( t = 1–2 min ) agonist addition . Quantification of data are presented as mean ± standard error of the mean ( SEM ) or standard deviation of the mean ( SD ) based on at least three biologically independent experiments with the precise number indicated in the figure legends . Statistical analysis was performed using Prism ( 8 . 1 . 1 , GraphPad ) and using paired or unpaired two tailed Student’s t test .	About a third of all drugs work by targeting a group of proteins known as G-protein coupled receptors , or GPCRs for short . These receptors are found on the surface of cells and transmit messages across the cell’s outer barrier . When a signaling molecule , like a hormone , is released in the body , it binds to a GPCR and changes the receptor’s shape . The change in structure affects how the GPCR interacts and binds to other proteins on the inside of the cell , triggering a series of reactions that alter the cell’s activity . Scientists have previously seen that a GPCR can trigger different responses depending on which signaling molecule is binding on the surface of the cell . However , the mechanism for this is unknown . One hypothesis is that different signaling molecules change the GPCR’s preference for binding to different proteins on the inside of the cell . The challenge has been to observe this happening without interfering with the process . Stoeber et al . have now tested this idea by attaching fluorescent tags to proteins that bind to activated GPCRs directly and without binding other signaling proteins . This meant these proteins could be tracked under a microscope as they made their way to bind to the GPCRs . Stoeber et al . focused on one particular GPCR , known as the opioid receptor , and tested the binding of two different opioid signaling molecules , etorphine and Dynorphin A . The experiments revealed that the different opioids did affect which of the engineered proteins would preferentially bind to the opioid receptor . This was followed by a similar experiment , where the engineered proteins were replaced with another protein called GRK2 , which binds to the opioid receptor under normal conditions in the cell . This showed that GRK2 binds much more strongly to the opioid receptor when Dynorphin A is added compared to adding etorphine . These findings show that GPCRs can not only communicate that a signaling molecule is binding but can respond differently to convey what molecule it is more specifically . This could be important in developing drugs , particularly to specifically trigger the desired response and reduce side effects . Stoeber et al . suggest that an important next step for research is to understand how the GPCRs preferentially bind to different proteins .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "biochemistry", "and", "chemical", "biology", "cell", "biology" ]	2020	Agonist-selective recruitment of engineered protein probes and of GRK2 by opioid receptors in living cells
In light microscopy , refractive index mismatches between media and sample cause spherical aberrations that often limit penetration depth and resolution . Optical clearing techniques can alleviate these mismatches , but they are so far limited to fixed samples . We present Iodixanol as a non-toxic medium supplement that allows refractive index matching in live specimens and thus substantially improves image quality in live-imaged primary cell cultures , planarians , zebrafish and human cerebral organoids . Live imaging is a key tool in understanding the organization and function of cells , tissues and organisms , since it allows the visualization of dynamic processes within their native environment . However , in practice , the live-imaging of multi-layered tissues with different cell types often poses major challenges . Refractive index mismatches between tissue and surrounding medium result in spherical aberrations that misalign the optical paths and ultimately distort and attenuate the microscopic image . This effect increases with complexity and thickness of the specimen , making imaging in deep tissue layers difficult and technically demanding ( Richardson and Lichtman , 2015 ) . Microscope optimization constitutes a first approach to optimize deep imaging . 2-photon microscopy greatly improves depth penetration by excitation with low scattering , near-infrared wavelengths ( Helmchen and Denk , 2005 ) . However , 2-photon microscopy cannot alleviate spherical aberration effects ( Richardson and Lichtman , 2015 ) . These can be partially compensated by the recent introduction of adaptive optics microscopes ( Booth , 2014 ) , yet at the cost of reduced image acquisition rates and the need for intense excitation light . A second approach to improving depth penetration is the direct adjustment of the refractive indexes ( RI ) of sample and environment ( Richardson and Lichtman , 2015 ) . Indeed , recently developed optical clearing techniques can render tissues effectively transparent by equilibrating refractive index heterogeneity within biological samples ( Chung et al . , 2013; Hama et al . , 2015 ) . Unfortunately , these protocols remain limited to fixed specimens due to their reliance on harsh mounting conditions and/or toxic chemicals ( Richardson and Lichtman , 2015 ) . Towards the goal of developing an RI matching medium for live-imaging , we searched for compounds that combine high water solubility as prerequisite for dilution into regular culture media , dilution-dependent RI tuning for effectiveness with a wide range of specimens and finally low toxicity as crucial requirement for live-imaging compatibility . The compound Iodixanol , which was originally developed as an intravenous X-ray contrast agent ( Albrechtsson et al . , 1992 ) and widely used in density gradient applications ( Bettinger et al . , 2002 ) , appeared to have many of the desired properties . Commercially available under the brand name OptiPrep ( TM ) , Iodixanol is optically clear and displays a high refractive index of 1 . 429 as a 60% stock solution , likely at least in parts due to its high density . This value is close to the refractive index of popular fixed tissue clearing solutions such as FocusClear ( RI 1 . 47 ) or CLARITY ( RI 1 . 45 ) ( Richardson and Lichtman , 2015 ) , and Iodixanol has in fact been used in such protocols ( Ku et al . , 2016 ) . As first test of its principal suitability , we evaluated the physicochemical properties of Iodixanol solutions . As Iodixanol is highly water-soluble , simple dilution into aqueous solutions can be used to linearly tune the refractive index of the solutions between RI 1 . 333 – RI 1 . 429 ( Figure 1a ) . For water , PBS and culture media of aquatic model organisms , the medium only minimally affected the refractive index at a given Iodixanol concentration ( Figure 1a ) . Further , we found the temperature dependent change in refractive index ( Beysens and Calmettes , 1977 ) of Iodixanol solutions to be minimal within physiologically relevant temperature ranges ( Figure 1b ) , allowing the use of the same medium at multiple temperatures . Organisms are often immobilized in agarose for live imaging . Agarose polymerization was not prevented at any Iodixanol concentration and agarose concentrations in ranges used for specimen immobilization did not significantly affect the refractive index of Iodixanol solutions ( Figure 1c ) , thus making Iodixanol compatible with agarose embedding protocols . A further important requirement especially for fluorescence-based live imaging applications is low autofluorescence . A spectral emission scan of Iodixanol solutions at the commonly used excitation wavelengths of 405 , 488 , 560 and 640 nm failed to reveal significant autoflorescence in comparison with PBS or highly diluted fluorescent beads as negative or positive controls , respectively ( Figure 1d , Figure 1—figure supplement 1 ) . pH buffering capacity is a further important consideration for potential media supplements . pH titration curves demonstrate that Iodixanol solutions have no significant pH buffering capabilities within the physiological relevant pH range of pH 4 – pH 9 , especially in comparison with PBS as classical physiological buffer ( Figure 1e ) . In fact , Iodixanol is only a slightly stronger acid than water ( Figure 1e ) . Finally , many optical clearing agents , such as ScaleA2 , have a high intrinsic osmolality that makes the reagent intrinsically live specimen incompatible ( Ke et al . , 2013 ) . 60% OptiPrep stock solution displays an osmolality of 212 ± 2 mmol/kg , which is below the typical 290–300 mmol/kg of vertebrate cell culture media ( Figure 1f ) . Further , we measured a linear increase of media osmolality across a dilution series with increasing Iodixanol concentrations ( Figure 1f ) . This means that the contribution of Iodixanol to overall media osmolality can be offset by a corresponding decrease in media salt concentration ( e . g . , NaCl ) . 10 . 7554/eLife . 27240 . 003Figure 1 . Physicochemical properties of the refractive index matching agent Iodixanol . ( a ) . Solvent dependency of the refractive index of Iodixanol . ( b ) Temperature dependency of the refractive index of Iodixanol solutions . Water was used as a solvent . ( c ) The refractive index of Iodixanol gels at various agarose concentrations . ( a–c ) Inset diagrams show a magnified region of the respective data set . Measurements were taken at 10% Iodixanol concentration increments as technical triplicates . In all cases a linear regression curve fit was applied to the series and the coefficient of determination is in all cases r2 > 0 . 999 . Standard deviations ( σ < 0 . 01% in all cases ) and data points were omitted for simplicity . See Figure 1—source data 1–3 for raw measurements . ( d ) Autofluorescene emission spectra of Iodixanol compared to PBS measured at indicated excitation wavelengths . Note that the detected signal is by orders of magnitudes lower than that of a positive fluorescent control , even at 405 nm excitation ( Figure 1—figure supplement 1 ) . ( e ) pH titration curve of 60% Iodixanol stock solution and indicated reference solutions . Starting volume = 50 ml . Data obtained from a single experiment . ( f ) Osmolality of Iodixanol solutions in various solvents . Measurements were taken at 10% Iodixanol concentration increments as technical triplicates and a linear regression curve fit was applied to the series . The coefficient of determination is in all cases r2 > 0 . 981 . Standard deviations ( σ <0 . 5% in all cases ) and data points were omitted for simplicity . See Figure 1—source data 4 for raw data . ( g ) 100 nm sub-diffraction sized beads imaged at 488 nm in unsupplemented aqueous 1% agarose ( top panel ) or in Iodixanol supplemented agarose tuned to the refractive index of the silicon immersion oil used for imaging ( bottom panel ) . Lateral pictures ( left ) show a single optical plane while axial pictures ( right ) represent maximum projected y-stacks . Scale bars: lateral 0 . 5 µm , axial 10 µm . The colour scheme encodes relative intensity ( brightest = white ) and all image acquisitions were performed under identical microscope settings ( h ) Point spread functions of sub-diffraction sized beads as shown in ( g ) . Quantified were peak intensity signal distributions from individual optical planes at indicated excitation wavelengths and direction ( n = 20 , error bars represent S . E . M ) . See Supplementary file 1 for quantified resolutions . Abbreviations: PBS: phosphate buffered saline; PW: planarian water; RI: refractive index; ZFW: zebrafish water . DOI: http://dx . doi . org/10 . 7554/eLife . 27240 . 00310 . 7554/eLife . 27240 . 004Figure 1—source data 1 . Raw measurement values for solvent dependency of the refractive index of Iodixanol . DOI: http://dx . doi . org/10 . 7554/eLife . 27240 . 00410 . 7554/eLife . 27240 . 005Figure 1—source data 2 . Raw measurement values for temperature dependency of the refractive index of Iodixanol solutions . Water was used as a solvent . DOI: http://dx . doi . org/10 . 7554/eLife . 27240 . 00510 . 7554/eLife . 27240 . 006Figure 1—source data 3 . Raw measurement values for the refractive index of Iodixanol gels at various agarose concentrations . DOI: http://dx . doi . org/10 . 7554/eLife . 27240 . 00610 . 7554/eLife . 27240 . 007Figure 1—source data 4 . Raw measurement values for the osmolality of Iodixanol solutions in various solvents . DOI: http://dx . doi . org/10 . 7554/eLife . 27240 . 00710 . 7554/eLife . 27240 . 008Figure 1—figure supplement 1 . Autofluorescence measurements of 60% Iodixanol compared to a highly dilute fluorescent bead solution ( 0 . 04% solids ) as positive controls at indicated excitation wavelengths . DOI: http://dx . doi . org/10 . 7554/eLife . 27240 . 008 To assess the optical effects of Iodixanol supplementation on image quality , we quantified the point spread functions of sub-diffraction sized fluorescent beads using a high NA 1 . 35 silicon oil immersion objective . As expected , tuning of the refractive index of the bead solution to that of the used silicon immersion oil ( RI = 1 . 40 ) , greatly improved both the lateral and axial image resolution compared to controls mounted in conventional aqueous media ( RI = 1 . 33; Figure 1g , h; Supplementary file 1 ) . Overall , the physicochemical properties of Iodixanol are therefore ideally suitable for refractive index tuning of live imaging media . However , toxicity is a further crucial concern in live imaging applications . We therefore quantitatively assessed the health of a range of typical specimens under extended Iodixanol exposure . We first measured the growth rates of human HeLa cell cultures exposed to various concentrations of Iodixanol 24 hr after seeding . Our quantitative measurements failed to detect any Iodixanol concentration dependent effects on HeLa cell proliferation or cell death up to three days after plating , even at the highest tested concentration of 30% Iodixanol ( Figure 2a , Figure 2—figure supplement 1 ) . Importantly , a concentration of 30% Iodixanol ( RI = 1 . 380 ) is higher than the optimal Iodixanol concentration required for HeLa cells . In absence of any toxicity indications , we carried out all subsequent toxicity assessments at the optimal Iodixanol concentration for the respective specimens ( please see Materials and methods and Figure 4—figure supplement 2 for a guide on how to determine a specimen’s optimal Iodixanol concentration ) . 10 . 7554/eLife . 27240 . 009Figure 2 . Iodixanol is live specimen compatible . ( a ) Iodixanol does not affect growth and cell death levels in cultured HeLa cells . Left: Representative low-resolution images of the constitutively expressed nuclear marker H2B-mCherry at indicated incubation times and media conditions . Scale bar = 50 µm; Right: Quantification of cell numbers ( number of nuclei ) and dead cells ( DRAQ7 positive nuclei ) at the indicated time points and Iodixanol concentrations . Iodixanol was applied 24 hr post seeding and measurements were normalized to the 24 hr time point in order to compensate fluctuations in plating density . n = 3; See Figure 2—figure supplement 1 for a complete data representation . ( b ) Iodixanol does not affect developmental growth or survival of dechorionated zebrafish embryos . Left: Representative images of developing embryos at the indicated time points ( hpf = hours post fertilization ) and media conditions . N = 5; Scale bars = 100 µm at 2 and 16 hpf , 500 µm at 48 and 72 hpf . Right: Quantification of body length and survival rate at the indicated time points and media conditions . The initial drop in the survival curves is an effect of dechorionation . N = 30; ( c ) Iodixanol does not affect regeneration of the planarian head or body proportions . Left: Representative images of regenerating Dendrocoelum lacteum amputation fragments at the indicated time points ( dpa = days post amputation ) and under the indicated media conditions . Anterior is always up , Scale bar = 500 µm; Right: Quantification of length/width ratio and projected area at the indicated time points and media conditions . Measurements were normalized to the 0 time point in order to compensate initial size differences between tissue pieces . N = 3; ( a–c ) Error bars represent S . E . M . p>0 . 05 in all cases: ( a ) one way ANOVA ( b , c ) paired t-test . DOI: http://dx . doi . org/10 . 7554/eLife . 27240 . 00910 . 7554/eLife . 27240 . 010Figure 2—figure supplement 1 . Representative low-resolution images of HeLa cell cultures exposed to the indicated Iodixanol concentrations at the indicated time points . The top three rows show H2B-mCherry as constitutively expressed nuclear marker , the bottom three rows show staining of the same cell fields with the dead cell marker DRAQ7 . Scale bar = 50 µm; all images are to scale . DOI: http://dx . doi . org/10 . 7554/eLife . 27240 . 010 We next assessed Iodixanol exposure effects on development by exposing de-chorionated zebrafish embryos to the optimal concentration of 20 % w/v Iodixanol . At 72 hr post fertilization , all embryos developing in Iodixanol displayed normal motility , muscle contractions and body pigmentation . Further , we found survival rates and the head to tail length as measure of developmental growth to be indistinguishable from controls , indicating that Iodixanol exposure over three days of development neither overtly affected development nor survival of zebrafish embryos ( Figure 2b ) . To assess potential long-term effects of Iodixanol exposure on dynamic tissue-level processes , we mounted regeneration-competent tissue fragments of planarian flatworms ( Rink , 2013 ) in 50 % w/v Iodixanol . Remarkably , even after 3 weeks of continuous exposure to a high concentration of Iodixanol , the specimens were healthy , had regenerated morphologically normal heads and succeeded in restoring normal body plan proportions as quantified by length to width ratio and projected area in a manner indistinguishable from controls ( Figure 2c ) . Collectively , these results establish that Iodixanol supplementation minimally impacts survival and growth of cell cultures , embryonic development or tissue turn-over and regeneration in intact animals , thus largely alleviating sample toxicity concerns . We therefore assessed the though-after improvements in live image quality obtainable via Iodixanol refractive index tuning . As reference point we used a current state of the art spinning disc confocal microscope with silicone immersion oil objectives . The refractive index of silicone oil , RI = 1 . 406 closely matches typical live specimens and its introduction has afforded a substantial improvement in live imaging quality ( York et al . , 2012 ) . We started our investigations at the smallest functional scale by imaging clusters of cultured primary zebrafish cells . In unsupplemented mounting media , the structure of nuclear chromatin was indiscernible in cells located ‘behind’ the first layer along the z-axis . Tuning the mounting media RI to 1 . 362 reduced the degradation of image resolution for such cells , demonstrating improvements in high resolution imaging of multi-layered cell culture applications ( Figure 3a , Figure 3—figure supplement 1 ) . Organoids , which are currently emerging as an important ex vivo model of organ development and function ( Simian and Bissell , 2017 ) , represent an imaging challenge at a larger functional scale . Human cerebral organoids appear opaque due to the optical density of neuronal tissues ( Figure 3b ) ( Lancaster and Knoblich , 2014 ) . Consequently , conventional single photon microscopy cannot penetrate significantly beyond 20 µm depth ( Figure 3b ) . By mounting organoids ( 67 days aged ) in Iodixanol supplemented culture media ( RI = 1 . 363 ) , we doubled the penetration depth to ~40 µm as a consequence of improved signal to noise ratios at depth ( Figure 3b ) . Iodixanol supplementation thus improves depth penetration in organoid imaging . 10 . 7554/eLife . 27240 . 011Figure 3 . Refractive index tuning with Iodixanol improves live-imaging of tissue culture systems . ( a ) . Effects of Iodixanol supplementation on live imaging of primary zebrafish cell cultures . Top Left: Brightfield image of a representative cluster of primary zebrafish embryonic cells , approximatly 50 µm in diameter . Centre panel: Images of cell clusters stained with the nuclear dye Hoechst 33342 . Left column: 3D-reconstruction of representative multi-layered cell clusters , imaged in control media ( RI = 1 . 333 , top row ) or in refractive index matched media ( RI = 1 . 362 , bottom row ) under identical imaging conditions . The arrowheads indicate representative deep layer nuclei that are further shown as 2D optical XY-section in the right column . Graphs: Intensity profiles along the solid lines indicated in the respective xy-section image . The flatter and lower intensity profile in the control condition ( top ) quantitatively documents a loss of chromatin structure fine detail in deep nuclei , which is preserved by Iodixanol supplementation ( bottom ) . Scale bars = 3D: 10 µm and 2D: 5 µm See Figure 3—figure supplement 1 for orthogonal sections . ( b ) Effects of Iodixanol supplementation on live imaging of human cerebral organoids . Top left: Dark field image of a representative human cerebral organoid approximately 2 mm in diameter . Centre panel: Human cerebral organoids at culture day 67 stained with the nuclear dye Hoechst 33342 . Centre panel: 3D-imaging of organoids , mounted either in standard media ( RI = 1 . 333 , top row ) or in refractive index matched media ( RI = 1 . 363 , bottom row ) under identical imaging conditions . Left column: Maximum projections of representative z-stacks . The white frame indicates the region shown to the right as optical xy-sections at the indicated tissue depth . The solid white line across the deepest section traces the course of the pixel intensity profile shown to the right . The flatter and lower intensity profile in the standard condition ( top ) quantitatively documents the loss of nuclear signal at 40 µm depth , while Iodixanol supplementation ( bottom ) still allows nuclei detection at that depth . Scale bars = 50 µm . The color scheme encodes relative intensity ( brightest = white ) . DOI: http://dx . doi . org/10 . 7554/eLife . 27240 . 01110 . 7554/eLife . 27240 . 012Figure 3—figure supplement 1 . Zebrafish primary cell culture . Top row shows maximum projected images of cell clusters imaged in standard media ( RI = 1 . 333 , left ) or refractive index adjusted media ( RI = 1 . 362 , right ) . The bottom row shows orthogonal optical sections at the positons indicated with the solid line in the top row . Overall , the images document a loss of resolution along the Z-axis ( = increasing distance from the coverslip ) in standard mounting media , which can be prevented by Iodixanol supplementation . DOI: http://dx . doi . org/10 . 7554/eLife . 27240 . 012 Zebrafish embryos are a popular vertebrate development model system because of their optical transparency ( Vascotto et al . , 1997 ) , yet the segmentation and tracking of cells beyond 100 µm in depth is still a challenge in embryos mounted in culture media ( RI = 1 . 333 , Figure 4—figure supplement 1 ) . To quantitatively assess the effect of Iodixanol supplementation on resolution and thus penetration depth , we imaged embryos injected with sub-diffraction sized fluorescent beads . The quantification of lateral point spread functions between controls and embryos mounted in media tuned to a refractive index of RI 1 . 363 revealed an improvement of lateral resolution ( 792 ± 28 nm ) compared to specimens mounted in regular media ( 918 ± 50 nm ) at a distance of 150 µm from the coverslip ( Figure 4a ) . We found that the resolution benefit of refractive index tuning increases with the distance of the object plane to the coverslip ( Figure 4a , Supplementary file 1 ) , as expected from the increasing impact of spherical aberrations with increasing distance to the objective . Overall , RI tuning of the embedding media to RI 1 . 363 allowed segmentation of nuclei up to 300 µm in depth , thus demonstrating a substantial improvement of deep tissue imaging in developing zebrafish embryos ( Figure 4—figure supplement 1 ) . 10 . 7554/eLife . 27240 . 013Figure 4 . Refractive index tuning with Iodixanol improves model organism live-imaging . ( a ) . Effects of Iodixanol supplementation on zebrafish embryo live imaging . Top left: Stereoscopic image of a dome stage zebrafish embryo of approximately 700 µm diameter . Centre panel: Zebrafish embryos expressing RFP-PCNA were injected at the single cell stage with 200 nm fluorescent sub-diffraction sized beads and imaged at dome stage ( 4 hpf ) . Images ( left column: RFP-PCNA , right column: beads ) represent 50 µm thick y-maximum projections of embryos imaged in regular media ( RI = 1 . 333 , top row ) or in Iodixanol supplemented media ( RI = 1 . 363 , bottom row ) . Scale bars = 50 µm . The graphs to the right depict the quantification of point spread functions of individual beads ( N = 20 ) at shallow ( top ) or deep ( bottom ) imaging depth , each for control and Iodixanol mounted specimens as per the indicated color scheme . The position of the analysed planes is indicated in the bead images to the left . The quantification of the width of the point spread function at half-maximal amplitude ( See Supplementary file 1 for numerical results ) reveals a significant increase in resolution in deep sections . ( b ) Effects of Iodixanol supplementation on planarian live imaging . Top left: Dark-field image of a specimen of the planarian flatworm Dendrocoelum lacteum approximately 4 mm in length . Centre panel: Live Dendrocoelum lacteum were stained with the nuclear marker RedDot1 and mounted in control media ( RI = 1 . 333 , top ) or media Iodixanol-tuned to RI = 1 . 412 ( bottom ) . The large images represent z-maximum projections of image stacks in the head region , with the solid line indicating the position of the single-plane orthogonal yz-section shown to the right . The scatter plots of mean nuclear intensity versus depth to the right quantitatively document an improved signal return upon Iodixanol supplementation , especially from deeper tissue layers . Scale bars = 50 µm . The color scheme encodes relative intensity ( brightest = white ) . DOI: http://dx . doi . org/10 . 7554/eLife . 27240 . 01310 . 7554/eLife . 27240 . 014Figure 4—figure supplement 1 . Improved nuclear segmentation in deep tissue layers by Iodixanol supplementation . four hpf zebrafish embryos expressing RFP-PCNA . Top row: Maximum projection ( left ) and optical XY-section at 200 µm depth ( right ) of an embryo in regular media ( RI = 1 . 333 ) highlights a progressive loss of nuclear signal detection and segmentation beyond 100 µm imaging depth , which is quantified in the scatter plot of mean nuclear intensity versus depth ( right ) . Bottom row: RI tuning with Iodixanol to RI = 1 . 363 enables nuclear detection and segmentation at up to 300 µm imaging depth . Panel order and imaging conditions exactly as above . Scale bars = 50 µm . The colour scheme encodes relative intensity ( brightest = white ) . DOI: http://dx . doi . org/10 . 7554/eLife . 27240 . 01410 . 7554/eLife . 27240 . 015Figure 4—figure supplement 2 . Schematic guidelines for determining the optimal Iodixanol concentration for a given specimen . If the refractive index of the specimen is known , the formula ( left ) can be used to tune the refractive index of the mounting medium between RI 1 . 333 and RI 1 . 429 . If the refractive index of the specimen is not known , the simple phase contrast imaging method by Oster et al . ( Oster , 1956 ) can be used for empirically determining the optimal Iodixanol concentration ( right ) . Briefly , this involves examining the specimen ( ideally dissociated cells ) mounted in different Iodixanol concentrations under a phase contrast microscope and scoring for a loss of contrast between sample and surrounding media . The images illustrate this point by example of two different specimens , with the green frame indicating the lowest contrast and thus the best amongst the tested Iodixanol concentrations . If necessary , a second round of titrations around this value can be added to determine the global optimum . DOI: http://dx . doi . org/10 . 7554/eLife . 27240 . 015 As final imaging challenge , we chose planarian flatworms . Although these animals are widely studied as models of whole body regeneration , live imaging of planarian regeneration has so far not been possible . Even unpigmented species like Dendrocoelum lacteum ( Liu et al . , 2013 ) are optically highly opaque , such that live imaging is largely restricted to the outermost cell layer ( the epithelium; Figure 4b ) . By tuning the refractive index of the embedding medium to RI 1 . 412 , we could partially compensate the opaque appearance of the specimen and significantly improve both signal detection and the overall signal to noise ratio in deeper cell layers ( Figure 4b ) . Together with the lack of overt effects on regeneration ( Figure 2c ) , Iodixanol supplementation therefore brings within reach the live-imaging of cell dynamics during planarian regeneration . Overall , our results establish Iodixanol supplementation as a simple , versatile and effective method for refractive index tuning in live imaging applications . We show that the reduction of spherical aberrations between sample and mounting media by refractive index tuning provides substantial improvements in achievable imaging depth in planaria , zebrafish and human organoids , as well as improved spatial resolution in cell culture applications . Refractive index tuning with Iodixanol therefore enables alignment of an important aspect of the optical axis in live specimens that could so far only be compensated in fixed specimens . What Iodixanol supplementation cannot correct for are refractive index differences within the specimen , such as between neighboring cells or between organelles and surrounding cytoplasm . Such effects are likely responsible for the fact that planarians and organoids appear optically opaque despite lacking pigmentation . Even though Iodixanol can therefore not achieve the in-toto RI matching of fixed tissue protocols ( Richardson and Lichtman , 2015 ) , our results nevertheless demonstrate substantial imaging improvements even in the case of opaque specimens . Overall , we expect that refractive index tuning by Iodixanol supplementation represents a broadly useful addition to the tool kit of live-imaging applications , all the way from cells to tissues and organisms . Iodixanol/OptiPrep was purchased as a 60% w/v stock solution from Sigma ( Cat No . D1556 ) . Planarian water contained 1 . 6 mM NaCl , 1 mM CaCl2 , 1 mM MgSO4 , 0 . 1 mM MgCl2 , 0 . 1 mM KCl , 1 . 2 mM NaHCO3 . Zebrafish medium contained 0 . 3x Danieu’s ( 17 . 4 mM NaCl , 228 µM KCl , 122 µM MgSO4*7H2O , 262 µM Ca ( NO3 ) 2 , 1 . 5 mM HEPES ) . Dissociated zebrafish cells were cultured in 1x Dulbecco’s PBS ( DPBS ) with 0 . 8 mM CaCl2 . Organoids were cultured in Differentiation Medium with vitamin A ( 125 ml DMEM/F12 , 125 ml Neuralbasal , 1 . 25 ml N2 supplement , 2 . 5 ml B27 + vitamin A supplement , 62 . 5 µl insulin , 2 . 5 ml Glutamax supplement , 1 . 25 ml NEAA-MEM and 2 . 5 ml penicillin-streptomycin ) according to Lancaster et al ( Lancaster and Knoblich , 2014 ) . Low gelling temperature SeaPlaque agarose ( Lonza , Cat No . 50100 ) was used for sample embedding . For imaging , samples were mounted in 35 mm No . 1 . 5 glass bottom dishes ( MatTek , Cat No P35G-1 . 5–14-C ) . For the determination of optical resolution in vitro 0 . 1 µm TetraSpeck fluorescent beads ( Thermo Fisher Scientific , Cat No . : T7279 ) mounted in 1% SeaPlaque agarose were used . The resolution was determined in vivo with 0 . 2 µm FluoSpheres ( Thermo Fisher Scientific , Cat . : F8807 ) . Refractive indexes were measured at 20°C unless otherwise indicated . Measurements of the refractive index were performed on a Rudolph Research Automatic Refractometer J457 at a wavelength of 589 . 3 nm . Each measurement was performed as a technical triplicate and refractive indexes were measured at 0% , 10% , 20% , 30% , 50% and 60% final Iodixanol content . Osmolality measurements were performed with a Wesco Vapro Osmometer as technical triplicates . For each Iodixanol dilution series the instrument was independently calibrated . The osmolality was measured at 0% , 10% , 20% , 30% , 50% and 60% final Iodixanol content . pH titration was performed with a freshly calibrated digital PHM210 pH meter ( Radiometer Analytical ) . The 1M HCl and 1M NaOH titration were carried out in separate experiments . In both experiments 50 ml of the indicated solution were titrated by subsequently adding 5 , 10 , 15 , 20 , 25 , 50 , 100 , 200 and finally 500 µl of acid or base . Measurements were taken once the pH meter indicated a stable measurement . Further information on Iodixanols physical properties ( such as density and viscosity ) can be found on the product information sheets of the respective commercial vendors ( an extensive description is provided by Alere Technologies: https://goo . gl/I4owRU ) . Which concentration of Iodixanol ( c%Iodixanol ) needs to be used is highly specimen dependent . If the refractive index of the sample is known the refractive index of the media ( RImedia ) should be adjusted accordingly simply by Iodixanol dilution:c%Iodixanol ≈ ( RImedia − 1 . 333 ) 0 . 0016 ( equation based on data from Figure 1a ) . When the refractive index of the sample is unknown an Iodixanol concentration titration should be performed . In this method introduced by Oster et al . , samples are incubated in various concentrations of Iodixanol and observed with phase contrast microscopy ( Oster , 1956 ) . A loss of contrast between sample and media results from a match of refractive indexes and thus experimentally indicates the target Iodixanol concentration ( Figure 4—figure supplement 2 ) HeLa ‘Kyoto’ cells stably expressing H2B-mCherry were described previously ( Neumann et al . , 2010 ) and obtained from the Ellenberg group at the European Molecular Biology Laboratory Heidelberg . HeLa ‘Kyoto’ cells are not included in the Register of Misidentified Cell Lines v 8 . 0 curated by the International Cell Line Authentication Committee ( Capes-Davis et al . , 2010 ) . The cell line was authenticated using Multiplex Cell Authentication by Multiplexion ( Heidelberg , Germany ) as described ( Castro et al . , 2013 ) . The SNP profiles matched known profiles or were unique . Mycoplasma tests with negative results for contamination were performed using the VenorGeM mycoplasma detection kit ( Sigma-Aldrich , Cat No . MP0025 ) . HeLa cells were cultured at 37C and 5% CO2 in High glucose GlutaMAX DMEM media ( Thermo Fischer Scientific , Cat No . : 10566016 ) supplemented with 10% ( v/v ) heat inactivated FBS , 100 µg/ml Penicillin/Streptomycin and 0 . 5 µg/ml Puromycin as a selection agent . For monitoring cell proliferation and death 700 cells were seeded per well into a 384 well plate ( Greiner Bio-One , Cat No . : 781096 ) . 24 hr post seeding media was replaced with 0% , 10% , 20% or 30% Iodixanol supplemented standard culture media additionally supplemented with 1 . 5 µM DRAQ7 ( Cell Signaling Technologies , Cat No . : 7406S ) as a cell death marker . Due to the high density of Iodixanol , plates were incubated upside down between image acquisitions . Imaging was carried out every 24 hr with the plate being in an upright position ( see below ) . Dendrocoelum lacteum were cultured in planarian water at 13°C and were fed weekly with calf liver paste . Prior imaging experiments animals were starved for 2 weeks . To stain planarian nuclei , animals were incubated for 12 hr with 2x RedDot1 ( Biotium , Cat No . : 40060 ) and 1% ( v/v ) DMSO in planarian water . Prior mounting , animals were anesthetized and relaxed for 1 hr by supplementing planarian water with 0 . 0097% w/v Linalool ( Sigma , Cat No . L2602 ) . Animals’ mucus was removed by a 5 min incubation in 0 . 5% w/v pH neutralized N-Acetyl-L-cysteine ( Sigma , Cat No . A7250 ) . Subsequently , animals were mounted in 1 . 5% SeaPlaque agarose dissolved in planarian water supplemented with 0 . 0097% Linalool . RI matched media had a final 50% Iodixanol content . Zebrafish embryos were kept according to standard conditions . Embryos of wild type ( TLAB ) and transgenic ( Tg ( bactin:RFP-pcna ) ) fish , the latter a generous gift of Caren Norden , were dechorionated by pronase treatment and maintained at 28°C in 0 . 3X Danieu’s medium diluted in distilled water and Iodixanol as indicated . Embryos were mounted in hanging drops of liquid mounting medium in ibidi glass bottom dishes ( 35 mm diameter , 0 . 17 mm coverslip ) , and inverted and submerged in liquid medium for imaging . For zebrafish cell culture , embryos were dissociated into individual cells in 55 mM NaCl , 1 . 75 mM KCl , 1 . 25 mM NaHCO3 , 10% glycerol solution by vortexing in low retention micro-centrifuge tubes . The cell suspension was centrifuged ( 400 g , 1 min ) , supernatant aspirated and replaced with 110 mM NaCl , 3 . 5 mM KCl , 2 . 7 mM CaCl2 , 10 mM Tris/Cl ( pH 8 . 5 ) , 10% glycerol solution . After further centrifugation ( 400 g , 1 min ) , supernatant replaced with DPBS with 0 . 8 mM CaCl2 added . This suspension was centrifuged ( 400 g , 1 min ) , the supernatant replaced with ~20 µl liquified agarose-based cell culture medium ( liquified at 70°C and held at 38°C ) and the cell pellet mechanically resuspended with a plastic micropipette tip . Liquid culture medium with suspended cells was transferred with the same micropipette tip onto the coverslip of an ibidi glass bottom dish . The still liquid mounting medium droplet was sandwiched with an additional 18 mm diameter round coverslip . After about 3 min the added coverslip was mechanically held down while applying 1 ml additional mounting medium . The imaging dish was then capped and sealed airtight with parafilm to prevent evaporation . Dissociation , mounting and imaging were carried out at room temperature without cooling or heating . Control zebrafish cell culture medium was DPBS with 0 . 8 mM CaCl2 . RI-matched medium was prepared in several steps , starting with 0 . 7X Dulbecco-PBS , 20% Iodixanol , 0 . 8 mM CaCl2 . Osmolality was then lowered to 5 mOsm/kg of control medium as a reference by addition of distilled water and repeated osmolality measurement . The RI was then lowered to within 0 . 003 of the cytoplasmic RI ( 1 . 3615 , determined by phase contrast microscopy [Oster , 1956] ) by addition of control medium and repeated RI measurement ( refractometer , 25°C ) . Control and RI-matched media were divided into 2 ml aliquots in microcentrifuge tubes , supplemented with 0 . 7% UltraPure ( Thermo Fisher Scientific , Cat No . : 16520050 ) low melting point agarose . Tubes were closed airtight and heated to 70°C for at least 1 hr , and could then be stored at 4°C for a month at minimum . For DNA staining , Hoechst 33342 stock ( 5 mg/ml ) was spiked into mounting medium aliquots at 1:2000 ( v/v ) ratio before mounting . Human cerebral organoids were generated from human iPSC line SC102A-1 ( System Biosciences ) and cultured according to previously published protocols with minor modification ( Lancaster and Knoblich , 2014 and Camp et al . , 2015 ) . The culture media was replaced with an 18% Iodixanol/media v/v solution 24 hr prior imaging to match the refractive index of the tissue . 2 hr prior imaging this solution was supplemented with 5 µg/ml Hoechst 34580 ( Thermo Fisher Scientific , Cat No . H21486 ) to stain nuclei . Organoids were mounted in 1% SeaPlaque agarose for imaging . The experiments performed with live samples did not require ethical approval according to German law . Autofluorescence of Iodixanol and control solutions was measured on a Tecan Spark 20M plate reader . Fluorescence was measured at 405 , 488 , 560 and 640 nm excitation . The emission signals were detected by emission spectra scans starting at 440 , 520 , 592 and 670 nm respectively . Scans were performed in 2 nm intervals . HeLa cell proliferation and death was monitored using a Cell Voyager 7000 spinning disc high throughput confocal system ( Yokogawa Electric Cooperation ) . H2B-mCherry was excited with a 561 nm solid state laser and the emission signal was detected with a 600/37 nm bandpass filter . DRAQ7 was illuminated with a 640 nm solid state laser and emission was detected with a 676/29 nm bandpass filter . Imaging was performed with a 10x UPlSApo NA 0 . 4 air objective . Fluorescent images of all other experiments were acquired on an Andor Revolution WD Borealis confocal spinning disc system . The Olympus IX83 stand was equipped with an Andor iXon Ultra 888 EMCCD for image acquisition . In vitro point spread functions were determined with an Olympus 100x NA 1 . 35 Sil UPlanSApo objective . For planarian , zebrafish embryo and organoid imaging an Olympus 30x UPlan SApo NA 1 . 05 Sil objective was used . Imaging of cultured zebrafish cells was performed with an Olympus 60x UPlan SApo NA 1 . 30 Sil objective . For Hoechst imaging a 405 nm laser diode was used in combination with 452/45 bandpass filter to detect the emission light . Green fluorescence of TetraSpeck beads was excited with a 488 nm laser diode and emission was collected with a 525/50 bandpass filter . RFP was illuminated with a 561 nm laser diode and the emission was detected with a 607/36 bandpass filter . RedDot1 was excited with a 640 nm laser diode and the emission was detected with a 685/40 bandpass filter . All filters were produced by Semrock . In all comparisons between refractive index matched media to control conditions identical illumination ( laser power ) and detection parameters ( exposure time ) were used on identical hardware setups ( objective , immersion silicone oil , filters , camera ) . Regenerating Dendrocoelum lacteum were imaged on a Nikon AZ 100M widefield microscope stand equipped with dark field illumination and a Nikon AZ Plan Fluor 2x NA 0 . 2 lens mounted . Images were acquired with a Nikon Digital Sight DS-Fi1 camera . Zebrafish embryo development was documented using a Leica M165C stereoscope equipped with a Leica 1x Plan apochromat NA 0 . 35 . Phase contrast imaging of zebrafish primary cell clusters was performed on a Zeiss Axioert 200M widefield microscope equipped with a Zeiss 20x Plan-Apochromat NA 0 . 75 objective . Images were recorded with a Diagnostics Instruments Spot RT camera . Phase contrast imaging of HeLa cells was performed on a widefield Zeiss Observer Z1 microscope stand equipped with a Zeiss Axiocam MRm and a 40x LS Plan – NeoFluoar NA 0 . 6 lens . Images were processed and analyzed with Fiji ( Schindelin et al . , 2012 ) . 3D views were rendered with ClearVolume ( Royer et al . , 2015 ) . Dynamic ranges , signal detection thresholds and object detection parameters were identically set when comparisons between refractive index matched and control conditions were made . For better visualization of intensity levels the ‘Fire’ lookup table ( LUT ) was applied . For segmenting nuclei in zebrafish embryos Fiji’s implemented ‘Otsu’ adaptive thresholding method was used on the raw image stacks and particles larger than 100 pixels were considered nuclei . Mean intensities of the thresholded objects was measured and reported for each slice of the Z-stack . For segmenting nuclei in planaria Fiji’s implemented ‘Moments’ adaptive thresholding method was used on the raw image stacks and particles between 100 and 2000 pixels were considered nuclei . Mean intensities of the thresholded objects was measured and reported for each slice of the Z-stack . To count live ( H2B-mCherry ) and dead ( DRAQ7 ) nuclei of HeLa H2B-mCherry cells , images were automatically thresholded with Fiji’s implemented ‘Otsu’ adaptive thresholding method for the respective channel . Thresholded objects larger than 100 pixels were counted as nuclei . Intensities of beads for PSF determination or object intensities for the demonstration of signal to noise ratios in vivo were determined with Fiji’s implemented ‘Plot Profile’ function along a previously defined line . All experimental numerical data were processed and visualized with Graph Pad Prism software . For PSF determination , a Gaussian distribution function was fit to the raw measurements . The optical resolution was defined as the full-width at half maximum intensity of that function . Display figures were created using Adobe Illustrator software .	Light microscopy is a key tool in biomedical research . For perfect images , light needs to be able to pass through the sample , the material ( or “mounting medium” ) that holds the sample in place , and finally the image-detecting equipment in a straight line . However , in practice , light rays often deviate away from this line because they move at different speeds in different materials; how much the speed of light changes is related to a property called the refractive index of the material . This is exactly the effect that causes a stick stuck into water to look bent at the water’s surface . In light microscopy , mismatches in refractive index significantly reduce quality of the images that can be obtained . Live specimens are particularly challenging to image because different specimens have very different refractive indices compared to the mounting medium , which holds specimens in place but must also keep them alive . Although the addition of chemical compounds can theoretically match the refractive index of the mounting medium to that of the specimen , this approach has so far not been practical because such manipulations tend to kill the specimen . An important challenge has therefore been to identify a compound that can adjust , or “tune” , the refractive index of mounting media over a wide range , yet without harming the specimens . Now , Boothe et al . have identified a chemical called Iodixanol as an ideal and easy to use supplement for tuning the refractive index of water-based live imaging media . Adding Iodixanol to the mounting media did not appear to have any toxic effects on cell cultures , developing zebrafish embryos or regenerating planarian flatworms . Importantly , Boothe et al . found that Iodixanol significantly improved the quality of the images collected from all of these different specimens . It is important to stress that Iodixanol does not change the refractive index of the sample or cancel out refractive index differences within the sample – so it cannot render opaque specimens transparent . Nevertheless , Iodixanol supplementation is a simple and affordable technique to improve image quality in any live imaging application without having to resort to more expensive and highly specialized microscopes .	[ "Abstract", "Introduction", "Results", "Discussion", "Material", "and", "methods" ]	[ "developmental", "biology", "cell", "biology", "tools", "and", "resources" ]	2017	A tunable refractive index matching medium for live imaging cells, tissues and model organisms
Differences in longevity between sexes is a mysterious yet general phenomenon across great evolutionary distances . To test the roles of responses to environmental cues and sexual behaviors in longevity regulation , we examined Caenorhabditis male lifespan under solitary , grouped , and mated conditions . We find that neurons and the germline are required for male pheromone-dependent male death . Hermaphrodites with a masculinized nervous system secrete male pheromone and are susceptible to male pheromone killing . Male pheromone-mediated killing is unique to androdioecious Caenorhabditis , and may reduce the number of males in hermaphroditic populations; neither males nor females of gonochoristic species are susceptible to male pheromone killing . By contrast , mating-induced death , which is characterized by germline-dependent shrinking , glycogen loss , and ectopic vitellogenin expression , utilizes distinct molecular pathways and is shared between the sexes and across species . The study of sex- and species-specific regulation of aging reveals deeply conserved mechanisms of longevity and population structure regulation . Males and females differ in many aspects of biology , including longevity . Sex differences in lifespan are common in animals across great evolutionary distances ( Austad and Fischer , 2016 ) . For example , women live longer than men in almost every country ( WHO , 2016 ) . Moreover , interventions in longevity also display sex-specific patterns in mice ( Austad and Fischer , 2016 ) . However , the underlying mechanisms of sex differences in longevity and responses to aging interventions , and the degree of evolutionary conservation of these mechanisms , are still largely unknown . Interactions between the sexes influence an individual’s longevity ( Fowler and Partridge , 1989; Gems and Riddle , 1996; Partridge and Farquhar , 1981; Van Voorhies , 1992 ) . Sex-specific longevity patterns also exist in Caenorhabditis elegans: males live significantly shorter when maintained in groups , whereas the longevity of hermaphrodites is not influenced by population density ( Gems and Riddle , 2000 ) . What causes this sex-specific longevity pattern remains mysterious . Although worm studies have contributed significantly to our general understanding of longevity mechanisms , how C . elegans male lifespan is regulated is poorly understood , because nearly all lifespan experiments are performed using only hermaphrodites . Analysis of Caenorhabditis males’ longevity not only allows us to test whether known longevity mechanisms are conserved between the sexes , but also provides an opportunity to reveal novel longevity mechanisms . The lifespan of Caenorhabditis females is shortened after mating through receipt of male sperm and seminal fluid ( Shi and Murphy , 2014 ) , and separately by exposure to male pheromone ( Maures et al . , 2014 ) . However , previous studies reported contradictory results on the influence of mating on male lifespan ( Gems and Riddle , 1996; Van Voorhies , 1992 ) . Thus , whether and how male lifespan is affected by prolonged exposure and interactions with females , as well as the effect of pheromone on male lifespan , is unknown . In their natural environments , animals must not only find food , but also avoid competitors , identify appropriate partners , and mate; the effects of these behaviors on lifespan are not well understood . Some of these behaviors are mediated by ascaroside-based pheromones ( Ludewig and Schroeder , 2013 ) . Caenorhabditis females secrete pheromones that attract males ( Chasnov et al . , 2007 ) , while C . elegans hermaphrodites modify their pheromone profile according to their sperm status , becoming more attractive to males once they have used up their own sperm ( Garcia et al . , 2007; Kleemann and Basolo , 2007; Morsci et al . , 2011 ) . Male ascaroside pheromones can directly affect the reproductive system of hermaphrodites , aiding recovery from heat stress and delaying the loss of hermaphrodite germline stem cells ( Aprison and Ruvinsky , 2015 , 2016 ) . The diversity of Caenorhabditis species allows us to evaluate male lifespan regulation from an evolutionary perspective . The Caenorhabditis genus consists of both androdioecious ( male and hermaphrodite ) and gonochoristic ( male and female ) species . In androdioecious species such as C . elegans , the population is dominated by hermaphrodites , which reproduce by self-fertilization . Males are usually very rare ( less than 0 . 2% for the standard lab strain N2 ) and are produced through spontaneous X chromosome nondisjunction ( Chasnov and Chow , 2002; Hodgkin , 1983 ) . Under stressful conditions , more oocytes undergo chromosome non-disjunction; thus androdioecious species may periodically experience male population explosions ( Chasnov and Chow , 2002; Hodgkin , 1983 ) . The existence of males in androdioecious species reduces inbreeding and facilitates adaptation to changing environments ( Anderson et al . , 2010 ) . By contrast , gonochoristic species such as C . remanei ( 50% male , 50% female ) require mating for reproduction . How males in androdioecious and gonochoristic species cope with these different mating situations remains poorly understood . Moreover , the utility of killing females by exposure to male pheromone in gonochoristic populations ( Maures et al . , 2014 ) is not obvious . Here we have found that male-specific population density-dependent death in C . elegans is due to the perception of male pheromone as a toxin; that is , while male pheromone itself is not a general poison to all worms , its perception by C . elegans males leads to death and to male-specific reproductive decline . C . elegans hermaphrodites , while still susceptible , are less sensitive to this toxic aspect of male pheromone . Masculinization of the hermaphrodite nervous system not only increases their sensitivity to male pheromone , but also is sufficient to induce male density-dependent death in these hermaphrodites , suggesting that neurons are key for the male pheromone killing mechanism . We found that the germline of the recipient male is also required for male pheromone-mediated death . This phenomenon is also present in two other independently-evolved androdioecious Caenorhabditis species , suggesting a role for male pheromone killing of males in otherwise hermaphroditic species; by contrast , neither males nor females of three gonochoristic Caenorhabditis species succumb to male pheromone killing . Mating-dependent death and germline-dependent shrinking , by contrast , are shared between all sexes and Caenorhabditis species , suggesting deep conservation . Our work highlights the importance of understanding the shared vs . sex- and species-specific mechanisms that regulate lifespan . When C . elegans males are housed together , they live shorter than solitary males , and the death rate increases with the number of males in a dose-dependent manner ( Gems and Riddle , 2000 ) . C . elegans male lifespan is very sensitive to male density: just two males together significantly reduced each individual’s lifespan , and grouping eight males decreases lifespan by more than 35% , whereas the lifespan of C . elegans hermaphrodites is not affected by population density ( Figure 1A , B ) . It was shown previously that C . elegans hermaphrodites can be killed by pheromone secreted by grouped C . elegans males ( Maures et al . , 2014 ) . We wondered whether the population density-dependent lifespan decrease of grouped C . elegans males is also due to male pheromone toxicity . To study the role of pheromone , we tested the survival of grouped daf-22 ( ascaroside pheromone-production deficient [Golden and Riddle , 1985] ) males . Eight grouped daf-22 ( m130 ) males lived almost as long as solitary wild-type males , suggesting that male pheromone kills grouped wild-type males ( Figure 1C; Figure 1—figure supplement 1A ) . 10 . 7554/eLife . 23493 . 003Figure 1 . Male pheromone leads to early death in C . elegans males . ( A ) Lifespans of grouped fog-2 ( q71 ) males . We used fog-2 ( q71 ) mutants , as fog-2 males are equivalent to wild-type ( N2 ) males ( Schedl and Kimble , 1988 ) . Solitary males: 12 . 0 ± 0 . 4 days , n = 40; two males: 10 . 6 ± 0 . 4 days , n = 40 , p=0 . 0397; four males: 9 . 9 ± 0 . 4 days , n = 60 , p=0 . 0012; eight males: 7 . 7 ± 0 . 2 days , n = 80 , p<0 . 0001 . For all the lifespan assays performed in this study , Kaplan-Meier analysis with log-rank ( Mantel-Cox ) test was used to determine statistical significance . All the lifespan results are included in Supplementary file 1 . ( B ) Lifespans of solitary N2 hermaphrodites: 12 . 3 ± 0 . 4 days , n = 40; grouped N2 hermaphrodites: 12 . 0 ± 0 . 3 days , n = 58 , p=0 . 6436 . ( C ) Grouped daf-22 ( m130 ) males have a similar lifespan to solitary wild-type fog-2 males . ( daf-22 ( m130 ) mutants are ascaroside pheromone-production deficient . ) Solitary fog-2 males: 13 . 8 ± 0 . 7 days , n = 35; eight fog-2 males: 9 . 8 ± 0 . 5 days , n = 48 , p<0 . 0001; eight daf-22 ( m130 ) males: 14 . 7 ± 0 . 7 days , n = 48 , p=0 . 4039 . Details about male-conditioned plates lifespan assays are included in Materials and methods and Figure 1—figure supplement 1B . ( D ) Lifespans of solitary C . elegans daf-22 ( m130 ) hermaphrodites on plates conditioned by eight fog-2 males . Solitary daf-22 hermaphrodites on control plates: 14 . 2 ± 0 . 6 days , n = 35; solitary daf-22 hermaphrodites on male-conditioned plates: 14 . 8 ± 0 . 8 days , n = 35 , p=0 . 4356 . ( E ) Lifespans of solitary C . elegans daf-22 ( m130 ) males on plates conditioned by eight fog-2 males . daf-22 ( m130 ) mutants are ascaroside pheromone-production deficient . Therefore , the effect of male pheromone is due to the male pheromone secreted by wild-type males when conditioning the plates . Solitary daf-22 males on control plates: 19 . 7 ± 0 . 5 days , n = 34; solitary daf-22 males on male-conditioned plates: 13 . 1 ± 0 . 4 days , n = 33 , p<0 . 0001 . ( F ) daf-22 ( m130 ) male lifespans on plates conditioned by wild-type fog-2 males . Solitary daf-22 ( m130 ) : 23 . 0 ± 0 . 9 days , n = 30; daf-22 ( m130 ) on plates conditioned by one fog-2 male: 17 . 3 ± 0 . 7 days , n = 29 , p<0 . 0001; daf-22 ( m130 ) on plates conditioned by eight fog-2 male: 16 . 1 ± 0 . 6 days , n = 30 , p<0 . 0001 . ( G ) Male pheromone-induced shorter lifespan of grouped daf-22 ( m130 ) males is inhibited in the presence of 50 µM FUdR . Grouped daf-22 males on NGM: 12 . 7 ± 0 . 3 days , n = 150; grouped daf-22 males on male-conditioned plates ( MCP ) : 11 . 0 ± 0 . 2 days , n = 150 , p<0 . 0001; grouped daf-22 males on NGM with FUdR: 14 . 9 ± 0 . 2 days , n = 150 , grouped daf-22 males on MCP with FUdR: 15 . 3 ± 0 . 2 days , n = 150 p=0 . 2964 . ( H ) Lifespans of grouped glp-1 ( e2141 ) males . Solitary males: 12 . 7 ± 0 . 8 days , n = 44; eight males: 13 . 3 ± 0 . 8 days , n = 56 , p=0 . 699 . DOI: http://dx . doi . org/10 . 7554/eLife . 23493 . 00310 . 7554/eLife . 23493 . 004Figure 1—figure supplement 1 . Male pheromone-mediated toxicity requires germline . ( A ) Grouped daf-22 ( m130 ) males live slightly shorter than the solitary control ( left ) . Solitary daf-22 males: 21 . 7 ± 1 . 2 days , n = 32; eight daf-22 males: 18 . 8 ± 1 . 0 days , n = 38 , p=0 . 0394 . Lifespans are not different between solitary and grouped daf-22 ( m130 ) in the presence of FUdR ( right ) . Solitary daf-22 ( m130 ) : 15 . 3 ± 0 . 3 days , n = 35; eight daf-22 ( m130 ) : 14 . 7 ± 0 . 3 days , n = 48 , p=0 . 2117 . ( B ) Schematic illustration of how lifespan assays on male-conditioned plates were performed . Detailed description is included in Materials and methods . ( C ) SAM plot of daf-22 ( m130 ) males ± MCP ( 30m ) for 6 days in the presence of 50 µM FUdR . Only 10 genes’ expression significantly changed , even at high FDR ( 17% ) . ( D ) SAM plot of glp-1 ( e2141 ) hermaphrodites ± MCP ( 60m ) for 6 days in the presence of 50 µM FUdR . Only three genes had significant changes in expression , even at high FDR ( 14% ) . ( E ) Lifespans of grouped N2 and glp-1 hermaphrodites on plates conditioned by 60 wild-type fog-2 males . Grouped N2 control: 14 . 7 ± 0 . 5 days , n = 107; grouped N2 hermaphrodites on MCP: 10 . 6 ± 0 . 3 days , n = 98 , p<0 . 0001; Grouped glp-1 ( e2141 ) control: 16 . 9 ± 0 . 5 days , n = 96; grouped glp-1 ( e2141 ) on MCP: 15 . 9 ± 0 . 6 days , n = 97 , p=0 . 8933 . Lifespan assay of glp-1 hermaphrodites was performed at 25°C in the first week of adulthood and at room temperature afterwards . ( F ) Germline also affects the production of male pheromone . Lifespans of solitary daf-22 ( m130 ) males control: 15 . 8 ± 0 . 9 days , n = 25; solitary daf-22 ( m130 ) males on plates conditioned by eight wild-type males: 11 . 8 ± 0 . 5 days , n = 25 , p=0 . 0002; solitary daf-22 ( m130 ) males on plates conditioned by eight germline-less glp-1 males: 13 . 4 ± 0 . 6 days , n = 25 , p=0 . 0339 ( compared to the control ) , p=0 . 0472 ( compared to MCP ( 8 wt m ) ) . DOI: http://dx . doi . org/10 . 7554/eLife . 23493 . 004 The fact that adding just one other male significantly decreases male lifespan ( Figure 1A ) suggests that C . elegans males are very sensitive to male pheromone . To compare the sensitivity to male pheromone between the sexes , we decreased the number of males from 30–150 as was previously used in male-conditioned plates ( Maures et al . , 2014 ) to only eight per plate ( Figure 1—figure supplement 1B ) , and examined the lifespans of daf-22 males and hermaphrodites . This low dosage of male pheromone had no effect on hermaphrodites , but significantly reduced C . elegans male lifespan ( Figure 1D , E ) , suggesting that males are more sensitive than hermaphrodites to male pheromone toxicity . In fact , we found that exposure to pheromone secreted by just one male was sufficient to significantly reduce the lifespan of C . elegans males ( Figure 1F ) . To identify the transcriptional effects of male pheromone treatment on males , we performed expression analysis of daf-22 ( pheromone-less ) males exposed to male pheromone for six days of adulthood . To avoid any germline-dependent effects of male interactions , we added the DNA replication inhibitor 5-fluorodeoxyruridine ( FUdR ) ( Figure 1G ) to the plates , which inhibits germline proliferation ( Shi and Murphy , 2014 ) . ( Adult treatment with the DNA replication inhibitor 5-fluorodeoxyruridine ( FUdR ) has little effect on lifespan and meiosis at low dosage ( 50 µM [Luo et al . , 2009] ) , but rapidly blocks germline proliferation in mated hermaphrodites ( Shi and Murphy , 2014 ) ; FUdR and other germline-blocking approaches are commonly used in expression analyses to avoid confounds [Reinke et al . , 2000; Shaw et al . , 2007; Maures et al . , 2014] . To our surprise , this comparison revealed no differences in gene expression ( Figure 1—figure supplement 1C ) , even at a high % False Discovery Rate ( FDR ) , suggesting that blocking male germline proliferation prevents male pheromone’s effects on gene expression . Indeed , we found that there was no lifespan difference when daf-22 ( m130 ) males were subjected to a high level of exogenous wild-type male pheromone ( 30 males per plate for conditioning ) in the presence of FUdR , mimicking the microarray conditions ( Figure 1G ) . Furthermore , no population density-dependent lifespan decrease was observed when germline-deficient glp-1 ( e2141 ) males were grouped ( Figure 1H ) . Similarly , exposing grouped glp-1 hermaphrodites to very high levels of male pheromone ( 60 wild-type males per plate for conditioning ) also failed to shorten lifespan or to induce significant transcriptional changes ( Figure 1—figure supplement 1D , E ) . Therefore , our results suggest that germline activity in the recipient is required for male pheromone-mediated death in both sexes . Interestingly , we also found that the loss of the germline itself also affects the production of male pheromone: males on plates conditioned by germline-less glp-1 males lived longer than those on plates conditioned by wild-type males ( Figure 1—figure supplement 1F ) , suggesting that there is communication between the status of germline and the production of male pheromone . However , males on plates conditioned by germline-less males still live shorter than the controls , suggesting that the germline cannot be the site of pheromone production , but rather may modulate pheromone levels or quality . The worm perceives environmental cues through its nervous system ( Bargmann , 2006 ) . To test whether the nervous system influences worms’ sensitivity to male pheromone , we utilized a strain of C . elegans hermaphrodite in which the neurons have been masculinized ( EG4389: him-5 ( e1490 ) V; lin-15 ( n765ts ) X; oxEx860[P ( rab-3 ) ::fem-3 ( wt ) ::mCherry ( worm ) ::unc-54 , pkd-2::gfp ( S65C ) , lin-15 ( + ) ] , a gift from the Jorgensen Lab [White et al . , 2007] ) . Solitary neuronally-masculinized hermaphrodites died earlier when exposed to a low level of male pheromone ( eight males per plate for conditioning; Figure 2A ) , whereas normal hermaphrodites were insensitive to male pheromone at this concentration ( Figure 1D ) , suggesting that sex-specific neuronal properties are responsible for male and hermaphrodite’s different sensitivities to male pheromone’s toxicity . 10 . 7554/eLife . 23493 . 005Figure 2 . Neuronal masculinization of C . elegans hermaphrodites . ( A ) Neuronal masculinization of C . elegans hermaphrodites increases their sensitivity to male pheromone toxicity . Lifespans of solitary masculinized hermaphrodites: 12 . 3 ± 0 . 3 days , n = 96; solitary masculinized hermaphrodites on plates conditioned by eight males: 9 . 6 ± 0 . 3 days , n = 56 , p<0 . 0001 . ( B ) Supernatant solutions from Day 5 C . elegans hermaphrodites and masculinized hermaphrodites and Day one fog-2 males were used to do the chemotaxis assay . See Materials and methods for detailed description . Chemotaxis Index ( C . I . ) of wild-type hermaphrodites’ supernatant: 0 . 65 ± 0 . 10 , C . I . of masculinized hermaphrodites’ supernatant: 0 . 03 ± 0 . 03; p=0 . 0261 , unpaired t-test . ( C ) Neuronal masculinization is sufficient to induce male-like population-density-dependent lifespan decrease in hermaphrodites . Lifespans of solitary N2 hermaphrodites: 12 . 3 ± 0 . 4 days , n = 40; grouped N2 hermaphrodites ( about 30 worms per 35 mm plate ) : 12 . 0 ± 0 . 3 days , n = 58 , p=0 . 6436 . Solitary masculinized hermaphrodites: 12 . 4 ± 0 . 5 days , n = 40; grouped masculinized hermaphrodites ( 30 per plate ) : 10 . 4 ± 0 . 3 days , n = 60 , p=0 . 0015 . ( D ) FUdR rescue the lifespans of grouped masculinized hermaphrodites . Lifespans of grouped masculinized hermaphrodites: 9 . 8 ± 0 . 2 days , n = 111 , grouped masculinized hermaphrodites ( 30 per plate ) in the presence of 50 µM FUdR: 12 . 2 ± 0 . 3 days , n = 119 , p<0 . 0001 . DOI: http://dx . doi . org/10 . 7554/eLife . 23493 . 005 Neuronal masculinization also changed the composition of pheromone secreted by the hermaphrodites: C . elegans males are normally attracted to pheromones secreted by old ( self sperm-depleted ) hermaphrodites ( Morsci et al . , 2011; Leighton et al . , 2014; Kleemann and Basolo , 2007 ) , but males were not attracted to pheromones secreted by aged neuronally-masculinized hermaphrodites ( Figure 2B ) . More surprisingly , when these masculinized hermaphrodites were grouped , they lived shorter than the solitary controls ( Figure 2C ) , suggesting that neuronal masculinization of the hermaphrodites is sufficient to induce the production of male-like pheromone in these hermaphrodites , and that neurons are key for male pheromone-mediated death . Inhibiting germline proliferation in these grouped masculinized hermaphrodites by FUdR rescued lifespan ( Figure 2D ) , which again supports the conclusion that the germline , whether male or female , is required for male pheromone-mediated killing . We previously found that mating greatly shortens hermaphrodite lifespan ( Shi and Murphy , 2014 ) , but the effect of mating on male lifespan is not yet known . To distinguish mating effects from the toxic effect of male pheromone , we measured the lifespans of solitary males and single males paired with a single hermaphrodite for 6 days from Day 1 to Day 6 of adulthood . ( We used fog-2 ( q71 ) mutants , as fog-2 males are equivalent to wild-type ( N2 ) males , while fog-2 hermaphrodites are self-spermless ( Schedl and Kimble , 1988 ) , enabling identification of successful mating . ) Mating decreased male lifespan ~35% compared with unmated solitary males ( Figure 3A , Supplementary file 1 ) , similar to the decrease in hermaphrodite lifespan caused by mating ( Shi and Murphy , 2014 ) . Males die faster when paired with a hermaphrodite for a longer period: mating with a hermaphrodite for one day did not significantly affect the lifespan of the male , while 2–3 days’ mating shortened male lifespan by 15% , 4–5 days’ mating reduced their lifespan by 25% , and mating for 6 days reduced lifespan more than 35% ( Figure 3B ) . By contrast , the number of hermaphrodites paired with the single male during mating had little effect ( Figure 3C ) . The time at which mating occurs within the reproductive period is also not critical for males’ post-mating lifespan decrease; given the same mating duration , males mated with hermaphrodites for the first three days of adulthood had a similar lifespan decrease as those mated with hermaphrodites during days 6–8 of adulthood ( Figure 3D ) . 10 . 7554/eLife . 23493 . 006Figure 3 . C . elegans males shrink and die early after mating . ( A ) Lifespans of unmated solitary and mated fog-2 ( q71 ) males . Solitary males: 13 . 1 ± 0 . 6 days , n = 50; mated males: 8 . 3 ± 0 . 4 days , n = 34 , p<0 . 0001 . Each male was paired with a fog-2 ( q71 ) hermaphrodite on a single 35 mm plate during Day 1–6 of adulthood . Unless noted , all the hermaphrodites used are fog-2 ( q71 ) . ( B ) Male post-mating lifespan decrease is mating duration-dependent: Unmated solitary males: 10 . 9 ± 0 . 6 days , n = 35; one male and one hermaphrodite mating on Day 1 of adulthood: 11 . 4 ± 0 . 6 days , n = 31 , p=0 . 3697; mating from Day 1–2: 9 . 0 ± 0 . 6 days , n = 30 , p=0 . 0325; mating from Day 1–3: 9 . 1 ± 0 . 6 days , n = 34 , p=0 . 0452; mating from Day 1–4: 7 . 9 ± 0 . 5 days , n = 32 , p=0 . 0002; mating from Day 1–5: 8 . 3 ± 0 . 4 days , n = 34 , p=0 . 0006; mating from Day 1–6: 6 . 8 ± 0 . 3 days , n = 33 , p<0 . 0001 . ( C ) Lifespans of one male paired with different number of hermaphrodites during Day 1–3 of adulthood: solitary unmated males: 13 . 8 ± 0 . 7 days , n = 35; one male with one hermaphrodite: 10 . 8 ± 0 . 6 days , n = 32 , p=0 . 0175; one male with two hermaphrodites: 11 . 6 ± 0 . 9 days , n = 33 , p=0 . 1435; one male with three hermaphrodites: 10 . 6 ± 0 . 8 days , n = 34 , p=0 . 0147 . ( D ) Lifespans of one male paired with three hermaphrodites for 3 days but at different time of adulthood . Solitary unmated males: 13 . 8 ± 0 . 7 days , n = 35; mating during Day 1–3 of adulthood: 10 . 6 ± 0 . 8 days , n = 34 , p=0 . 0147; mating during Day 6–8 of adulthood: 10 . 8 ± 0 . 6 days , n = 37 , p=0 . 0022 . ( E ) Length of unmated and mated fog-2 males: t-test , p<0 . 01 , p<0 . 001 . ( F ) Representative pictures of the same unmated solitary male and male paired with one hermaphrodite from Day 1-Day 6 of adulthood . ( G ) Male pheromone does not induce body shrinking . Length of solitary fog-2 males on plates conditioned by eight wild-type males . DOI: http://dx . doi . org/10 . 7554/eLife . 23493 . 006 As we previously observed in hermaphrodites ( Shi and Murphy , 2014 ) , males shrank after 6 days of mating; by Day 7 , the mated males were 10% smaller than the unmated solitary males control ( Figure 3E , F , Supplementary file 2 ) . No such shrinking was apparent when males were treated with male pheromone ( Figure 3G ) , suggesting that mating and male pheromone act through different pathways . We wondered whether pheromone is required for mating-induced death in males; however , wild-type males still died early post-mating when paired with a daf-22 hermaphrodite for 6 days ( Figure 4A ) . Likewise , daf-22 mutant males lived shorter after 6 days’ mating with fog-2 hermaphrodites ( Figure 4B ) , indicating that the post-mating lifespan decrease in our single-worm pair lifespan assay is due to mating itself rather than to the presence of pheromone from either sex . 10 . 7554/eLife . 23493 . 007Figure 4 . Mating-induced death is germline-dependent . ( A ) Lifespans of fog-2 males mated with daf-22 ( m130 ) hermaphrodites . Unmated solitary fog-2 males: 12 . 1 ± 0 . 6 days , n = 32; mated males: 9 . 0 ± 0 . 4 days , n = 29 , p=0 . 0001 . In the mated group , one fog-2 ( q71 ) male was paired with one daf-22 ( m130 ) hermaphrodite from Day 1- Day 6 of adulthood . ( B ) Lifespans of unmated and mated daf-22 ( m130 ) males . Unmated solitary daf-22 ( m130 ) males: 13 . 8 ± 0 . 6 days , n = 40; mated daf-22 ( m130 ) males: 7 . 4 ± 0 . 4 days , n = 34 , p<0 . 0001 . In the mated group , one daf-22 ( m130 ) male was paired with one fog-2 ( q71 ) hermaphrodite from Day 1- Day 6 of adulthood . ( C ) FUdR can rescue male post-mating early death . Unmated solitary males: 10 . 5 ± 0 . 5 days , n = 35; one male with three hermaphrodites for three days: 6 . 4 ± 0 . 3 days , n = 31 , p<0 . 0001; one male with three hermaphrodites for three days but in the presence of 50 µM FUdR during the three days’ mating: 10 . 2 ± 0 . 4 days , n = 36 , p=0 . 7086 ( compared with unmated solitary group ) . ( D ) Lifespans of unmated and mated glp-1 ( e2141 ) males: unmated solitary glp-1 males: 8 . 0 ± 0 . 4 days , n = 40; mated glp-1 males: 7 . 2 ± 0 . 4 days , n = 40 , p=0 . 3178 . The assay was performed at 26°C , in mated group , one glp-1 male was paired with one fog-2 hermaphrodite from Day 1–6 . ( E ) Length of mated and unmated glp-1 ( e2141 ) males . ( The same population as in Figure 4D . ) . ( F ) Expression heatmap of genes whose expression is significantly changed in mated males based on SAM analysis . ( G ) Ectopic expression of VIT-2::GFP in mated males is germline-dependent . 5 days’ mating , pictures were taken on Day 6 of adulthood . Representative images are shown above the quantification of VIT-2::GFP expression [maximum ± SE ( error bars ) ] , a . u . , arbitrary units . p<0 . 001 , t-test . ( H ) unc-62 RNAi suppresses male post-mating early death . Unmated solitary male on L4440: 12 . 6 ± 0 . 7 days , n = 25; mated males on L4440: 8 . 8 ± 0 . 5 days , n = 33 , p=0 . 0001 . Unmated males on unc-62 RNAi: 11 . 9 ± 0 . 8 days , n = 25; mated males on unc-62 RNAi: 10 . 6 ± 0 . 5 days , n = 34 , p=0 . 1249 ( compared to unmated males on unc-62 RNAi ) . ( I ) pqm-1 ( ok485 ) mated males have similar lifespans as unmated controls . Unmated solitary pqm-1 ( ok485 ) males: 11 . 9 ± 0 . 5 days , n = 25; mated pqm-1 ( ok485 ) males: 11 . 0 ± 0 . 6 days , n = 29 , p=0 . 2782 . In the mated group , one pqm-1 ( ok485 ) male was paired with one fog-2 ( q71 ) hermaphrodite from Day 1- Day 6 of adulthood . DOI: http://dx . doi . org/10 . 7554/eLife . 23493 . 00710 . 7554/eLife . 23493 . 008Figure 4—figure supplement 1 . Mating-induced lifespan decrease is germline-dependent . ( A ) Germline proliferation blocking with FUdR can rescue male post-mating early death . Unmated solitary males: 13 . 8 ± 0 . 7 days , n = 35; one male with one hermaphrodite for six days: 10 . 3 ± 0 . 6 days , n = 31 , p=0 . 0006; solitary male in the presence of 50 µM FUdR: 13 . 9 ± 0 . 4 days , n = 35 , p=0 . 4079 ( compared to unmated solitary group ) . One male mated with one hermaphrodites for 6 days in the presence of FUdR: 13 . 6 ± 0 . 5 days , n = 34 , p=0 . 3992 compared to unmated solitary group . ( B ) Lifespans of unmated and mated glp-1 ( e2141 ) males: unmated solitary glp-1 males: 11 . 1 ± 1 . 0 days , n = 27; mated glp-1 males: 11 . 1 ± 0 . 5 days , n = 43 , p=0 . 9149 . The assay was performed at 25°C; in mated group , one glp-1 male was paired with one fog-2 hermaphrodite from Day 1–6 . DOI: http://dx . doi . org/10 . 7554/eLife . 23493 . 00810 . 7554/eLife . 23493 . 009Figure 4—figure supplement 2 . Microarray analysis of mated males . ( A ) Clustered heat map of whole transcriptome expression comparison of mated vs unmated males . Individual males were paired with one hermaphrodite for 3 . 5 days and collected on Day four for microarrays . ( B ) Enriched GO terms for significantly up- and down-regulated genes in mated males . ( C ) Enriched motifs in promoter region ( 1 kb upstream of TSS ) of significantly up- and down-regulated genes using ( RSAT ) Regulatory Sequence Analysis Tools ( www . rsat . eu ) . ( D ) VIT-2::GFP expression in males increases significantly after mating . Upper: DIC and GFP images; Lower: GFP intensity quantitation , left: max ± SE ( error bars ) ; right: mean ± SE ( error bars ) , a . u . , arbitrary units . p<0 . 01 , t-test . DOI: http://dx . doi . org/10 . 7554/eLife . 23493 . 009 Elevated germline proliferation is one of the major causes of hermaphrodites’ early death after mating ( Shi and Murphy , 2014 ) . This killing mechanism appears to be conserved in males: when treated with 50 µM FUdR during the three-day mating period , males no longer died earlier ( Figure 4C ) . FUdR treatment also eliminated male post-mating lifespan decrease in our 6 day mating regime ( Figure 4—figure supplement 1A ) . The absence of the germline also prevented mating-induced shrinking: germline-less glp-1 ( e2141 ) males experienced neither shrinking nor lifespan decrease after mating ( Figure 4D , E , Figure 4—figure supplement 1B ) . These results suggest that germline-mediated post-mating death and shrinking is conserved between C . elegans sexes . To identify molecular mechanisms that contribute to post-mating death in males , we performed genome-wide transcriptional analyses of mated and unmated males: we paired a single male with a hermaphrodite for 3 . 5 days of mating ( 150 pairs per biological replicate ) , then picked the males individually from the hermaphrodites on Day four for microarray analysis ( Figure 4—figure supplement 2A ) . As a control , we collected the same number of age-matched solitary males . Only 14 genes were significantly up-regulated and 41 were significantly down-regulated ( FDR = 0%; Figure 4F; Supplementary file 3 ) . Genes whose expression decreased in mated males include extracellular proteins ( scl-11 , scl-12 , zig-4 ) and predicted lipase-related hydrolases ( lips-11 , lips-12 , lips-13 ) that may participate in fat metabolism . As we previously found in hermaphrodites ( Shi and Murphy , 2014 ) , mating decreases fat storage in males ( Figure 5B ) . 10 . 7554/eLife . 23493 . 010Figure 5 . Mating-induced and male pheromone-induced death are distinct . ( A ) Transcriptional profiles of mated males and MCP-treated males are different . Heatmap cluster of mated males ( left ) and MCP-treated vs untreated grouped daf-22 males ( right ) ; Pearson correlation = −0 . 27 . The cluster only contains genes with significant changes in mated males by SAM , 0% FDR . ( B ) Mating induces significant fat loss in males . Representative pictures of Oil red O staining are shown above the quantitation . Males lost about 20% of their fat after mating on Day 4 , p<0 . 001 . Error bars represent SE . ( C ) Male pheromone exposure fails to induce fat loss in males . Four days’ MCP treatment . Representative pictures of Oil red O staining are shown above the quantitation . Unconditioned control males are framed by black lines , and MCP-treated males are framed by green lines . ( D ) Glycogen staining of mated and unmated males . Left: mated fog-2 ( wt ) males lost over 30% glycogen after 5 days’ mating; *p<0 . 001 . Right: mated glp-1 males did not lose glycogen after mating . The staining intensity was normalized to unmated males of each genotype . Representative pictures are shown above the quantitation . Unmated males are framed by dashed lines , and mated males are framed by solid lines . ( E ) No glycogen loss after male pheromone exposure . Four days’ MCP treatment . Representative pictures of iodine staining are shown above the quantitation . Unconditioned control males are framed by black lines , and MCP-treated males are framed by green lines . ( In D–E , error bars represent SD . ) . ( F ) Loss of the DAE-dependent transcription factor PQM-1 suppresses male pheromone-induced death . Lifespans of control solitary pqm-1 ( ok485 ) males: 15 . 6 ± 0 . 6 days , n = 25; solitary pqm-1 ( ok485 ) males on plates conditioned by eight males: 14 . 4 ± 0 . 6 days , n = 25 , p=0 . 1627 . ( G ) DAPI staining of Day six males’ germlines ( left ) . Right: categorical quantification of germline morphology: mating causes more obvious change in male germline morphology than male pheromone does . TZ: transition zone; U: U-shaped turn of male germline . See Figure 5—figure supplement 4 for details . DOI: http://dx . doi . org/10 . 7554/eLife . 23493 . 01010 . 7554/eLife . 23493 . 011Figure 5—figure supplement 1 . Transcriptional profiles of mated and male pheromone-induced males are distinct . ( A ) Heatmap of ( 1 ) mated males ( genes with significant changes by SAM , 0% FDR ) and ( 2 ) MCP treated vs untreated grouped daf-22 males . Top: Pearson correlation . ( B ) Heatmap of ( 1 ) mated males and ( 2 ) MCP treated vs untreated grouped daf-22 males ( genes with significant changes by SAM , 1% FDR ) . Top: Pearson correlation . ( C ) Vit genes are not up-regulated in males exposed to male pheromone . DOI: http://dx . doi . org/10 . 7554/eLife . 23493 . 01110 . 7554/eLife . 23493 . 012Figure 5—figure supplement 2 . Glycogen staining of mated vs unmated hermaphrodites and males . Left: representative pictures of iodine staining of worms . Unmated worms are framed by dashed lines , whereas mated worms are framed by solid lines . In the first picture , mated and unmated fog-2 hermaphrodites were mixed together , with red arrows pointing to mated fog-2 hermaphrodites . Worms were mated from Day 1 – Day 5 and were imaged on Day 5 . Right: quantitation of iodine staining . The intensity of mated worms was normalized to unmated control of the same genotype . Mated fog-2 hermaphrodites have only 30% of the glycogen levels of unmated fog-2 hermaphrodites of the same age ( p<0 . 0001 ) . Mated glp-1 hermaphrodites have 99% glycogen compared to unmated glp-1 hermaphrodites control ( p=0 . 6070 ) . Mated fog-2 males have 64% of the glycogen level of unmated fog-2 males of the same age ( p<0 . 0001 ) . Mated glp-1 males have 101% glycogen compared to unmated glp-1 males control ( p=0 . 7107 ) . Error bars represent SD . p<0 . 0001 , t-test . DOI: http://dx . doi . org/10 . 7554/eLife . 23493 . 01210 . 7554/eLife . 23493 . 013Figure 5—figure supplement 3 . Enriched motifs of male pheromone-induced transcriptional changes . Enriched motifs associated with significantly up- and down-regulated genes predicted by RSAT ( Regulatory Sequence Analysis Tools ) in grouped vs . single neuronal masculinized hermaphrodites ( left ) and MCP treated vs untreated grouped daf-22 males . DOI: http://dx . doi . org/10 . 7554/eLife . 23493 . 01310 . 7554/eLife . 23493 . 014Figure 5—figure supplement 4 . Germline of mated and MCP-treated males . ( A ) Representative micrographs of each category: left: male germline with clear transition zone ( yellow arrow indicates transition zone marked by crescent shaped nuclei ) ; middle: male germline with no apparent transition zone; right: male germline with mature sperm appearing before U shaped turn . ( B ) Mating ( one male one hermaphrodites for 4 days ) or treating solitary males with male pheromone ( 4 days on MCP ( 8m ) ) does not change the number of proliferating nuclei in the male germline . The number of nuclei in mitotic proliferating zone ( only counting germlines with clear transition zone ) is similar in all three conditions . DOI: http://dx . doi . org/10 . 7554/eLife . 23493 . 014 Surprisingly , vitellogenins ( vit-4 , vit-3 , vit-5 , vit-6 , vit-2 ) , which encode yolk protein precursors made in the female/hermaphrodite intestine for transport into developing oocytes ( Kimble and Sharrock , 1983 ) and as such are not normally expressed in males ( Figure 4G , left ) , were the top up-regulated genes in mated males , expressed on average 19 times higher in mated males than in solitary unmated males ( Supplementary file 3 ) . Overproduction of vitellogenins has been shown to be deleterious for hermaphrodites: vitellogenins accumulate in the head and body of older hermaphrodites ( Garigan et al . , 2002 ) ; long-lived insulin signaling mutants repress vit gene expression ( Murphy et al . , 2003; Seah et al . , 2016 ) ; overexpression of vitellogenin reduces the lifespan of long-lived mutants ( Seah et al . , 2016 ) ; and knockdown of the vit genes in wild-type hermaphrodites extends lifespan ( Murphy et al . , 2003; Seah et al . , 2016 ) . Mating induced ectopic expression of VIT-2::GFP in the anterior intestine of males , confirming our gene expression data . Such expression induction was germline function-dependent , as FUdR treatment of males blocked VIT-2::GFP expression in mated males ( Figure 4G , Figure 4—figure supplement 2D ) . In addition to the increase in vit gene expression , the binding motif for UNC-62 , a master transcriptional regulator of vit genes in hermaphrodites ( Van Nostrand et al . , 2013 ) , emerged from unbiased motif analysis of the up-regulated gene set ( Figure 4—figure supplement 2C ) . Using RNAi , we found that knocking down unc-62 was sufficient to rescue the lifespan decrease in mated males ( Figure 4H ) . Thus , the mis-expression of vitellogenins upon mating contributes to post-mating death in males . The DAE ( DAF-16 Associated Element ) motif is present in most vit gene promoters , which are also Class 2 DAF-16 genes ( Murphy et al . , 2003 ) . The DAE is bound by PQM-1 , a transcription factor that is involved in the regulation of multiple processes , including development , stress response , metabolism , and longevity ( Tepper et al . , 2013; Dowen et al . , 2016 ) . We found that mated pqm-1 ( ok485 ) deletion males lived as long as unmated controls ( Figure 4I ) , suggesting that PQM-1 is also required for post-mating male death . To determine whether pheromone-dependent killing , which like mating-induced death requires the germline , utilizes the same downstream mechanisms to kill males , we performed genome-wide transcriptional analysis of worms under conditions where they are exposed to high levels of male pheromone and have short lifespans ( Figures 1G and 2C ) : ( 1 ) grouped daf-22 males on plates conditioned by wild-type males vs . grouped daf-22 males on control plates , and ( 2 ) grouped vs . solitary neuronally-masculinized hermaphrodites ( Supplementary file 4 ) . Three main Gene Ontology groups emerged from these comparisons: innate immunity and defense responses , metal ion/cadmium response , and glycoprotein metabolism . Notably , this pattern of gene expression was very different from that of mated males ( Figure 5A , Figure 5—figure supplement 1; Pearson correlation = −0 . 0594 for the whole transcriptome ) . Specifically , upregulation of vitellogenin genes was not a signature of male pheromone-treated animals ( Figure 5—figure supplement 1C ) , also supporting the notion that the two pathways act through distinct molecular mechanisms . While fat is reduced in males after mating ( Figure 5B ) , MCP treatment caused no significant changes in fat metabolism gene expression ( Figure 5—figure supplement 1A ) or Oil Red O staining ( Figure 5C ) . We showed previously that osmotic stress resistance correlates well with shrinking in mated hermaphrodites , whereas fat loss does not account for such shrinking ( Shi and Murphy , 2014 ) . Likewise , we found that mated wild-type worms lost about 30% of their glycogen stores post-mating in a germline-dependent manner ( Figure 5D ) . The mating-induced glycogen storage decrease and subsequent shrinking is conserved between sexes ( Figure 5—figure supplement 2 ) , while glycogen stores are not affected by male pheromone ( Figure 5E ) . Previously identified gene expression patterns of longevity pathways did not appear in our analysis ( Tepper et al . , 2013; Lakhina et al . , 2015 ) , suggesting a novel pathway for lifespan shortening in the presence of male pheromone . However , comparison of the list of genes significantly up-regulated upon MCP treatment of daf-22 males ( Supplementary file 5 ) to previously published arrays of male pheromone-treated hermaphrodites ( Maures et al . , 2014 ) yielded a significant overlap ( p-value=3 . 05E-06 ) , including ins-11; ins-11 mutants are partially resistant to death by male pheromone ( Maures et al . , 2014 ) . These results suggest that some mechanisms of male-pheromone-induced death are shared between the sexes , but may act independently of known longevity pathways . Unbiased motif analysis revealed that DAE ( DAF-16 Associated Element ) motif was also enriched in these male pheromone-treated conditions ( Figure 5—figure supplement 3 ) . We found that pqm-1 ( ok485 ) null mutant males were not short-lived when treated with male pheromone ( eight males per plate for conditioning; Figure 5F ) , suggesting that male pheromone killing , like mating-induced death , is mediated at least in part by PQM-1 . Both male pheromone-induced killing and mating-induced death require the germline; to examine possible mechanisms of germline effects , we performed DAPI staining of males in pheromone-treated and mated conditions . In a fraction of Day four unmated solitary males , the transition zone ( marked by crescent shaped nuclei ) disappeared ( Figure 5G , Figure 5—figure supplement 4 ) , and the meiotic region expanded to the distal arm , evidenced by the presence of sperm before U-shaped turn ( see Figure 5—figure supplement 4 ) , indicating the loss of mitotic proliferating cells in the germline . Shrinking of the transition zone was observed significantly less in mated animals , but no significant differences were apparent in MCP-treated males ( Figure 5G ) . Therefore , our results suggest that mating causes an increase in the number of mitotically proliferating cells , whereas an unknown signal from the germline , rather than an obvious morphological change , may be responsible for the lifespan shortening effects of male pheromone . Previously , we found that C . remanei females , like C . elegans hermaphrodites , shrink and die faster after mating ( Shi and Murphy , 2014 ) , suggesting that these mechanisms are evolutionarily conserved in females . We found that C . remanei males also lived significantly shorter after mating with a female C . remanei for 6 days ( Figure 6A ) . While female death requires successful cross-progeny production ( Shi and Murphy , 2014 ) , C . elegans males died early when mated with a C . remanei female for 6 days ( Figure 6B ) , even though no cross-progeny result from this mating . This result , together with the germline dependence of mating-induced death , suggests the process of mating and up-regulation of male germline activity is sufficient to induce death , regardless of the species of the recipient female . 10 . 7554/eLife . 23493 . 015Figure 6 . Mating-induced death is evolutionarily conserved , whereas male pheromone-induced death is not . ( A ) Mated C . remanei males also live shorter . Unmated solitary C . remanei males: 31 . 4 ± 1 . 7 days , n = 72; mated C . remanei males: 15 . 7 ± 1 . 2 days , n = 28 , p<0 . 0001 . In mated group: one C . remanei male was paired with one C . remanei female from Day 1-Day 6 of adulthood . ( B ) Lifespans of C . elegans males mated with C . elegans hermaphrodites and C . remanei females . Unmated solitary C . elegans males: 10 . 2 ± 0 . 6 days , n = 35; C . elegans males mated with C . elegans hermaphrodites: 7 . 4 ± 0 . 4 days , n = 35 , p=0 . 0001; C . elegans males mated with C . remanei females: 7 . 4 ± 0 . 4 days , n = 35 , p=0 . 0003 . In mated groups , one C . elegans male was paired with either one C . elegans hermaphrodite or one C . remanei female from Day 1–6 of adulthood . ( C ) Lifespans of solitary C . remanei males on plates conditioned by eight C . remanei males . Solitary C . remanei males on control plates: 35 . 8 ± 2 . 0 days , n = 34; solitary C . remanei males on male-conditioned plates: 37 . 8 ± 1 . 2 days , n = 34 , p=0 . 8501 . ( D ) Lifespans of solitary C . remanei females on plates conditioned by eight C . remanei males . Solitary C . remanei females on control plates: 27 . 6 ± 2 . 2 days , n = 24; solitary C . remanei females on male-conditioned plates: 27 . 0 ± 2 . 5 days , n = 30 , p=0 . 8306 . ( E ) Lifespans of grouped C . remanei females on plates conditioned by 30 males . C . remanei females on control plates: 15 . 8 ± 0 . 9 days , n = 60; C . remanei females on plates conditioned by C . remanei males: 19 . 5 ± 1 . 3 days , n = 30 , p=0 . 0636; C . remanei females on plates conditioned by C . elegans fog-2 males: 18 . 5 ± 0 . 9 days , n = 60 , p=0 . 1770 . ( F ) Lifespans of solitary C . remanei males on plates conditioned by eight C . elegans males . Solitary C . remanei males on control plates: 37 . 2 ± 1 . 7 days , n = 40; solitary C . remanei males on C . elegans male-conditioned plates: 36 . 7 ± 1 . 4 days , n = 38 , p=0 . 7774 . ( G ) Lifespans of grouped C . elegans fog-2 hermaphrodites on plates conditioned with 30 males . fog-2 hermaphrodites control: 14 . 4 ± 0 . 8 days , n = 90 . fog-2 hermaphrodites on plates conditioned by fog-2 males: 10 . 9 ± 0 . 6 days , n = 60 , p=0 . 0004; fog-2 hermaphrodites on plates conditioned by C . remanei males: 11 . 9 ± 0 . 5 days , n = 90 , p=0 . 0042 . ( H ) Chemotaxis of C . elegans ( left , blue ) and C . remanei ( right , red ) to supernatants from C . elegans males , C . remanei males , C . elegans N2 hermaphrodites , and C . remanei females . See Materials and methods for detailed description . C . e . males to supernatant of C . r . females: Chemotaxis Index ( CI ) is 0 . 46 ± 0 . 11 ( mean ± SEM , n = 12 [plates] ) ; C . e . males to supernatant of C . e . hermaphrodites: CI = 0 . 14 ± 0 . 13 ( n = 10 ) ; C . e . males to supernatant of C . e . males: CI = 0 . 17 ± 0 . 17 ( n = 12 ) ; C . e . males to supernatant of C . r . males: CI = 0 . 11 ± 0 . 12 ( n = 11 ) ; C . r . males to supernatant of C . r . females: CI = 0 . 85 ± 0 . 04 ( n = 12 ) ; C . r . males to supernatant of C . e . hermaphrodites: CI = 0 . 04 ± 0 . 08 ( n = 12 ) ; C . r . males to supernatant of C . e . males: CI = 0 . 04 ± 0 . 15 ( n = 12 ) ; C . r . males to supernatant of C . r . males: CI = −0 . 09 ± 0 . 16 ( n = 12 ) . ( I ) Sensitivity to male pheromone-induced lifespan reduction . C . elegans males are the most sensitive to male pheromone-induced killing , whereas both C . remanei sexes are immune to this effect . DOI: http://dx . doi . org/10 . 7554/eLife . 23493 . 015 Because C . elegans is an androdioecious ( males and hermaphrodites ) species , we wondered whether male pheromone-mediated killing also occurs in a true 50/50 male/female ( gonochoristic ) species , where one would expect the level of male pheromone to be high under normal conditions . When exposed to low levels of C . remanei male pheromone ( eight males per plate for conditioning ) , neither C . remanei males nor females were short-lived ( Figure 6C , D ) . At a higher concentration ( 30 males per plate for conditioning ) , multiple trials of C . remanei females and males on male-conditioned plates failed to reveal any sensitivity to either C . remanei or C . elegans male pheromone ( Figure 6E , F ) , in contrast to previous reports in which males and females were grouped ( Maures et al . , 2014 ) ; this result suggests that the lifespan shortening in the latter study was caused by mating rather than by male pheromone . Interestingly , we found that male pheromone toxicity can act across species: C . elegans hermaphrodites died early when exposed to C . remanei male pheromone ( Figure 6G ) , suggesting that the difference in sensitivity to male pheromone might stem from the perception of pheromone as a toxin , rather than from toxicity of the pheromone itself . C . remanei can sense pheromone , but uses it to distinguish potential partners and competitors ( Figure 6H ) , rather than to kill males . The differential sensitivity to male-pheromone-induced killing between C . elegans males and hermaphrodites also suggests that the latter might only be experienced as an off-target effect under extremely high male pheromone conditions ( Figure 6I ) . To determine whether the differences in male-pheromone-induced killing are a more general phenomenon , we examined the effect of male pheromone on other androdioecious and gonochoristic species . Like C . remanei males , two other gonochoristic species , C . brenneri and C . nigoni , are also immune to male pheromone killing ( Figure 7A , B ) . By contrast , males from two evolutionarily distant androdioecious species , C . briggsae and C . tropicalis , died significantly earlier when exposed to male pheromone ( Figure 7C , D ) , just as C . elegans males do . These results strongly indicate that male pheromone-dependent killing is shared by males of androdioecious species . 10 . 7554/eLife . 23493 . 016Figure 7 . Gonochoristic species are immune to male pheromone killing; androdioecious species are susceptible . ( A ) Lifespans of solitary C . brenneri males on plates conditioned by 8 C . brenneri males . Solitary C . brenneri males control: 18 . 1 ± 0 . 8 days , n = 33; solitary C . brenneri males on MCP: 17 . 8 ± 1 . 1 days , n = 32 , p=0 . 9915 . ( B ) Lifespans of solitary C . nigoni males on plates conditioned by 8 C . nigoni males . Solitary C . nigoni males control: 15 . 3 ± 0 . 4 days , n = 32; solitary C . nigoni males on MCP: 15 . 2 ± 0 . 6 days , n = 40 , p=0 . 7443 . ( C ) Lifespans of solitary C . briggsae males on plates conditioned by 8 C . briggsae males . Solitary C . briggsae males control: 13 . 7 ± 0 . 8 days , n = 38; solitary C . briggsae males on MCP: 10 . 3 ± 0 . 3 days , n = 54 , p=0 . 0192 . ( D ) Lifespans of solitary C . tropicalis males on plates conditioned by 8 C . tropicalis males . Solitary C . tropicalis males control: 17 . 7 ± 0 . 8 days , n = 40; solitary C . tropicalis males on MCP: 12 . 2 ± 1 . 0 days , n = 60 , p=0 . 0002 . ( E ) Model of the effects of mating and male pheromone on androdioecious and gonochoristic female and males . C . elegans hermaphrodites ( upper left ) ; C . remanei females ( upper right ) ; C . elegans males ( lower left ) ; C . remanei males ( lower right ) . DOI: http://dx . doi . org/10 . 7554/eLife . 23493 . 016 The fact that male pheromone selectively kills males of multiple , independently-evolved androdioecious species suggests a role for male pheromone killing in these populations , but the lifespan effects we observed to this point are largely post-reproductive , arguing against any reproductive selection under these conditions . However , those experiments were designed specifically to probe adult phenotypes rather than development or mating , by only applying pheromone to adult worms . In order to better mimic conditions in which worms would be exposed to male pheromone their entire lives , we placed fog-2 eggs on male pheromone conditioned plates , and measured developmental rates , lifespan , mating rates , and brood size . Other than a slight difference at 42 hr that disappeared by 48 hr , we observed no significant effects of MCP on the developmental rates of either males or hermaphrodites ( Figure 8—figure supplement 1 ) . By contrast , male pheromone conditioning from egg onward caused a severe ( 36% ) shortening of lifespan ( Figure 8A , Figure 8—figure supplement 2A ) . Moreover , male pheromone significantly decreased male fertility ( Figure 8B , Figure 8—figure supplement 2B ) . ( Note that male pheromone-induced male fertility decrease is distinct from defects in male mating that arise with age ( Chatterjee et al . , 2013 ) ; we observed similar male fertility decline with age in control animals , but the males treated with male pheromone from egg onward exhibited an additional fertility decline compared with age-matched control males . ) Finally , male pheromone treatment decreases the number of progeny produced by those animals who do successfully mate ( Figure 8C ) . By contrast , male pheromone treatment did not affect the brood size of self-fertilized hermaphrodites ( Figure 8D ) . These results suggest that exposure to male pheromone during early life specifically reduces male fertility . 10 . 7554/eLife . 23493 . 017Figure 8 . Male pheromone reduces male offspring . ( A ) Male pheromone conditioning from egg onward causes a more severe lifespan shortening . ( Pooled results from two independent assays . See Figure 8—figure supplement 2 for results of separate lifespan assays , which were also significant . ) Solitary C . elegans fog-2 males control: 18 . 5 ± 0 . 6 days , n = 75; solitary C . elegans fog-2 males on MCP from early adulthood onward ( aMCP ) : 13 . 9 ± 0 . 4 days , n = 79 , p<0 . 0001 ( compared to the control ) ; solitary C . elegans fog-2 males on MCP from egg onward ( eMCP ) : 11 . 4 ± 0 . 4 days , n = 80 , p<0 . 0001 ( compared to the control ) , p<0 . 0001 ( compared to aMCP ) . ( B ) Male pheromone decreases fog-2 male fertility on Day 6 . Male treatment started from egg onward . Each male was paired with one virgin fog-2 hermaphrodite at indicated time for 24 hr . On Day 4 of adulthood , the percent of males who were able to fertilize Day 1 virgin fog-2 hermaphrodites: three biological replicates: ctrl: 82 ± 2%; MCP: 69 ± 6% , p=0 . 12 , unpaired t-test . By Day 6 of adulthood , the percent of males who were able to fertilize Day one virgin fog-2 hermaphrodites ( three biological replicates ) had significantly decreased: ctrl: 69 ± 4%; MCP: 29 ± 4% , p=0 . 0019 . ( C ) Male pheromone treatment decreases the number of progeny produced by those animals who do successfully mate . The difference appears by Day 5 . See Figure 8—figure supplement 3 for detailed numbers . p<0 . 05 , unpaired t-test . ( D ) Brood size of self-fertilized N2 hermaphrodites is not affected by male pheromone . Ctrl: 238 . 2 ± 5 , n = 19; MCP-treated: 240 . 5 ± 5 , n = 18 , p=0 . 7571 , unpaired t-test . ( E ) Theoretical calculation of male pheromone’s effect on male population control ( not considering any other contributing factors ) . See Discussion and Figure 8—figure supplement 3 for more details . DOI: http://dx . doi . org/10 . 7554/eLife . 23493 . 01710 . 7554/eLife . 23493 . 018Figure 8—figure supplement 1 . Male pheromone does not affect developmental rates . Eggs from fog-2 worms were bleached onto control ( n = 359 ) or MCP ( n = 257 ) plates . The number of eggs , L1 , L2 , L3 , and L4 worms was counted at 18 , 24 , 42 , and 48 hr post-bleach . We observed a slight deceleration in development at L3 and no difference by L4 , resulting in no significant change in total developmental rate upon treatment with MCP . DOI: http://dx . doi . org/10 . 7554/eLife . 23493 . 01810 . 7554/eLife . 23493 . 019Figure 8—figure supplement 2 . Male pheromone treatment from egg onward severely affects both lifespan and cross-offspring production . ( A ) Male pheromone conditioning from egg onward causes a severe lifespan shortening ( replicate 1 ) . Solitary C . elegans fog-2 males control: 17 . 4 ± 0 . 7 days , n = 34; solitary C . elegans fog-2 males on MCP from early adulthood onward ( aMCP ) : 13 . 9 ± 0 . 7 days , n = 39 , p<0 . 0001 ( compared to the control ) ; solitary C . elegans fog-2 males on MCP from egg onward ( eMCP ) : 11 . 1 ± 0 . 6 days , n = 40 , p<0 . 0001 ( compared to the control ) , p=0 . 0021 ( compared to aMCP ) . ( B ) Male pheromone conditioning from egg onward causes a severe lifespan shortening ( replicate 2 ) . Solitary C . elegans fog-2 males control: 19 . 5 ± 0 . 9 days , n = 41; solitary C . elegans fog-2 males on MCP from early adulthood onward ( aMCP ) : 13 . 4 ± 0 . 5 days , n = 40 , p<0 . 0001 ( compared to the control ) ; solitary C . elegans fog-2 males on MCP from egg onward ( eMCP ) : 11 . 8 ± 0 . 4 days , n = 40 , p<0 . 0001 ( compared to the control ) , p=0 . 024 ( compared to aMCP ) . ( C ) Percentages of fog-2 males who are able to successfully fertilize virgin fog-2 hermaphrodites in the presence and absence of MCP treatment . See Figure 8—figure supplement 3 for details . ( D ) Male pheromone treatment does not selectively kill one sex in offspring . The male ratio is always ~50% ( checked from the progeny produced by fog-2 hermaphrodites successfully fertilized by males in C ) . DOI: http://dx . doi . org/10 . 7554/eLife . 23493 . 01910 . 7554/eLife . 23493 . 020Figure 8—figure supplement 3 . Male pheromone treatment from egg onward reduces offspring and might be a novel mechanism to cull male population in hermaphroditic species . ( A ) Percentage of fog-2 males who were able to successfully fertilize virgin fog-2 hermaphrodites in the presence and absence of MCP treatment . ( B ) Progeny produced by those animals who do successfully mate . ( C ) Example of theoretical calculation of how male pheromone affects cross progeny production in the next generation ( simplified modeling , not considering any other factors ) . In both control and eMCP conditions , the population starts with 60 males . Assuming they have equal chance of mating from Day 1 – Day 6 of adulthood , the data of percent fertile and number of progeny produced in A and B were used at each time points , in the end , the males in eMCP group produce about 76 . 8% of the total progeny produced by males in control group . This number was used when plotting the curve in Figure 8E , for generation X , percent males of control group is 50% , percent males of eMCP group is 50%* ( 76 . 8% ) X . DOI: http://dx . doi . org/10 . 7554/eLife . 23493 . 020 Here we found that male pheromone killing is the major cause of population density-dependent lifespan decrease in C . elegans males , and is only utilized by androdioecious Caenorhabditis males . By contrast , all sexes of Caenorhabditis species succumb to mating-induced death , while both sexes of gonochoristic Caenorhabditis species are immune to male pheromone toxicity . The toxicity of male pheromone may explain the contradictory results from previous publications in which grouped males were used as the control in testing whether mating affects the lifespan of Caenorhabditis males ( Gems and Riddle , 1996; Van Voorhies , 1992 ) . Masculinization of neurons in hermaphrodites not only increases their sensitivity to male pheromone , but also is sufficient to induce the production of male-like toxic pheromone , suggesting that neurons play two major and distinct roles in this type of killing . The germline and the DAE-dependent transcription factor PQM-1 are required for both mating-induced and male pheromone-mediated death , but the downstream expression changes upon mating and pheromone treatment are distinct; only mating induces vitellogenin gene expression in males and causes shrinking . Thus , we have discovered two distinct mechanisms that accelerate aging in Caenorhabditis males . C . elegans males and hermaphrodites share many post-mating changes . As we found previously for mated females and hermaphrodites ( Shi and Murphy , 2014 ) , Caenorhabditis males also experience germline-dependent shrinking , glycogen loss , and death after mating . Germline up-regulation also leads to ectopic expression of vitellogenins , which contributes to the post-mating lifespan decrease in males . Previously , these yolk protein precursors were only reported to be expressed in hermaphrodites , where vitellogenin proteins are taken up by oocytes; vitellogenin production in males does not have an obvious purpose . Mating also induces significant overexpression of vit genes in hermaphrodites ( DePina et al . , 2011 ) , indicating that vitellogenin expression is closely coupled with mating-induced germline up-regulation in both sexes . Such coupling may be strong enough to overcome the normal repression of male vitellogenin expression . Germline-dependent body shrinking , glycogen loss , and ectopic vitellogenin expression contribute to male post-mating death , which is conserved between the sexes . The striking similarity of germline-dependent post-mating changes in Caenorhabditis males and females suggests that this mechanism is largely conserved between sexes , and may represent an unavoidable cost of reproduction as a result of mating . Germline-dependent lifespan shortening appears to be conserved across species over large evolutionary distances , as it occurs in all Caenorhabditis species we tested . Male post-mating death is also conserved beyond the Caenorhabditis genus , as Drosophila males die earlier after mating , as well ( Partridge and Farquhar , 1981 ) . It was previously noted that the lifespan of Korean eunuchs was significantly longer than the lifespan of non-castrated men with similar socio-economic status ( Min et al . , 2012 ) , analogous to the long lifespan of germline-less C . elegans ( Hsin and Kenyon , 1999 ) and Drosophila ( Flatt et al . , 2008 ) , while the significantly ( 35% ) shortened lifespan of Chinese emperors who were noted to be particularly promiscuous might be an example of the opposite effect on the germline ( Shi et al . , 2015 ) , suggesting that some aspects of germline-dependent male post-mating death may be conserved across great evolutionary distances . C . elegans are subject to killing by male pheromone , while C . remanei are not . Our cross-species results suggest that C . remanei male pheromone is perceived as a toxin by C . elegans , but C . remanei are immune to both elegans and remanei pheromone ( Figure 6C–F ) . The preponderance of males in a 50:50 population , as in the case of C . remanei , makes the use of pheromone as a toxin less likely , as it would cause too much off-target death to be useful for sperm competition purposes . The toxic effect of pheromone may not be due to the pheromone itself , but rather to a perception of pheromone as a toxin , with a greater effect in males than in hermaphrodites . Hermaphrodite death at high male pheromone concentration ( Maures et al . , 2014 ) —which might happen rarely in nature – might simply be collateral damage , as hermaphrodites are far less sensitive than males to male pheromone ( Figure 6I ) toxin . The lack of significant changes observed in the developmental rates of either males or hermaphrodites , or on the brood size of hermaphrodites , indicates that the primary effect of male pheromone might be on male reproductive capacity . Caenorhabditis species might utilize pheromones in such different ways due to their different modes of reproduction . Androdioecious species males do not appear to use pheromones efficiently as chemical messengers to facilitate mating , since they are less able to distinguish hermaphrodites’ pheromone from other species’ female or male pheromone; in fact , C . elegans males are slightly attracted to their own male pheromone , in part explaining their clumping ( Chasnov et al . , 2007 ) ( Figure 6H ) , despite the fact that male pheromone is very toxic to C . elegans males . In the androdioecious species such as C . elegans , males are normally rare ( 0 . 2% ) , so the chance that any worm he encounters will be an appropriate mating partner is very high; thus , there may be less selection pressure to evolve pheromones as chemical messengers to identify mates . By contrast , C . remanei uses pheromone to distinguish males from females , an important requirement for mating in 50:50 mixed populations . C . remanei males are slightly repelled by male pheromone ( Chasnov et al . , 2007 ) ( Figure 6H ) , but are extremely attracted to C . remanei female pheromone , while C . remanei , as well as both males and females from other gonochoristic species , are immune to male pheromone toxicity ( Figure 6C , D ) . Thus , gonochoristic species use pheromones primarily as chemical cues to identify mates , rather than to kill males . The fact that male pheromone toxicity is present in three distantly-related and separately evolved hermaphroditic Caenorhabditis species ( Cho et al . , 2004; Kiontke et al . , 2004; Kiontke et al . , 2011 ) suggests an important role for male pheromone killing . Periodic explosions of male populations in androdioecious species ( e . g . , under stressful conditions ) allow outcrossing and ensure genetic diversity ( Anderson et al . , 2010 ) . After this beneficial period , however , males are more costly to maintain , and there may be pressure to return to a primarily hermaphroditic population ( Figure 8E ) . It is notable that because C . elegans males are XO , rather than XY , males may have no selfish drive to maintain their own chromosomes . Using male pheromone as a dose-dependent toxin may be an effective way to cull the male population and ensure that the species returns to the self-reproduction mode when the stressful condition has passed , aiding the return to hermaphroditism . Because a high fraction of males can only be produced by mating ( mating produces 50% males , while male production rates from hermaphroditic selfing is 0 . 2% [Chasnov and Chow , 2002; Hodgkin , 1983] ) , the combination of decreased mating efficacy and decreased progeny production might be expected to specifically affect the number of males produced each generation ( Figure 8—figure supplement 3 ) . Male pheromone alone could effectively drive the population back to a primarily hermaphroditic state after several generations ( Figure 8E ) . Previous experiments showed that the average time for males to disappear in N2 strain is 12–20 days ( i . e . 4–7 generations ) ( Wegewitz et al . , 2008 ) . The discrepancy between our modeling ( ~15 generations ) and the previous experimental result may indicate that multiple factors , including increased hermaphroditic progeny production and decreased mating rates ( Wegewitz et al . , 2008 ) , decreased copulation performance in aging males ( Chatterjee et al . , 2013 ) , and hermaphrodites’ response to males [Garcia et al . , 2007; Kleemann and Basolo , 2007; Morsci et al . , 2011] ) could act in tandem with pheromone-dependent killing of males to cull the male population and thus promote a return to hermaphroditism . Male-specific culling occurs in species such as Drosophila bifasciata , in which Wolbachia infection leads to the killing of male embryos , suggesting that sex ratio can be controlled through male-killing ( Stevens et al . , 2001 ) . Mathematical modeling shows that selection in C . elegans favors low populations of males ( Stewart and Phillips , 2002 ) , and our model provides a mechanism for how this might be achieved . In summary , germline-dependent early death after mating is conserved between sexes and perhaps even across great evolutionarily distances , and is likely due to an unavoidable cost of mating , the result of mated animals ramping up germline proliferation and subsequently exhausting their own resources as quickly as possible to produce the next generation of progeny . The differential use of pheromones as toxins or chemical messengers by males in androdioecious and gonochoristic species , respectively , demonstrates that they adopt different strategies to compete , mate , and maintain optimal sex ratios . N2 ( wild type ) CB4108: fog-2 ( q71 ) V CB4037: glp-1 ( e2141 ) III DR476: daf-22 ( m130 ) II RB711: pqm-1 ( ok485 ) II RT130: pwIs23 [vit-2::GFP] PB4641: Caenorhabditis remanei PB2801: Caenorhabditis brenneri AF16: Caenorhabditis briggsae JU1422: Caenorhabditis nigoni JU1373: Caenorhabditis tropicalis 6699 EG4389: him-5 ( e1490 ) V; lin-15 ( n765ts ) X; oxEx860[P ( rab-3 ) ::fem-3 ( wt ) ::mCherry ( worm ) ::unc-54 , pkd-2::gfp ( S65C ) , lin-15 ( + ) ] ( gift from the Jorgensen Lab ) All lifespan assays were performed at room temperature ( ~20–21°C ) , except for glp-1 male lifespan assays ( performed at 25–26°C ) . 35 mm NGM plates were used for all the experiments in this study . 20 µl of OP50 was dropped onto each plate to make a bacterial lawn of ~10 mm diameter . The next day , one synchronized late L4 male and one late L4 hermaphrodite/female were transferred onto each 35 mm NGM plate . For experiments in Figures 3C , D and 4C , multiple L4 hermaphrodites were transferred together with one male . One late L4 male of the same age and genotype was transferred onto the control plates . Except for Figure 4A , fog-2 ( q71 ) hermaphrodites were used as the C . elegans hermaphrodites in the mating assay , because fog-2 hermaphrodites do not have self sperm , thus allowing us to easily detect successful mating ( i . e . eggs and progeny on the plates ) . In mated groups , we only included males that were able to produce progeny in our analysis . However , for the experiments regarding glp-1 males , mating on FUdR , and the inter-species cross between C . elegans males and C . remanei females , we included all the males in the analysis . Worms were transferred onto new plates every other day . If the hermaphrodites were lost or bagged , new unmated Day one fog-2 hermaphrodites were added as replacements . Males and hermaphrodites/females were kept together for 6 days ( unless noted otherwise in the text ) ; afterwards only males were transferred on to newly seeded plates every 2–3 days . For RNAi experiments in Figure 4H , synchronized eggs were transferred onto NGM plates with RNAi bacteria , late L4 males were transferred and paired with fog-2 L4 hermaphrodites onto NGM plates seeded with OP50 ( to eliminate the possible effect on mating efficiency for different RNAi treatments ) . Two days later , males and hermaphrodites were transferred onto fresh plates seeded with corresponding RNAi bacteria and males were maintained on RNAi bacteria thereafter . 30–50 worms were included in each group of individual worm lifespan assays . The sample size was similar to previous published study of individual hermaphrodite lifespan assays . When lifespan assays were completed , Kaplan-Meier analysis with log-rank ( Mantel-Cox ) method was performed to compare the lifespans of different groups . The summary of all lifespan experiments is included in Supplementary file 1 . 35 mm NGM plates were used for all the experiments in this study . 20 µl of OP50 was dropped onto each plate to make a bacterial lawn of ~10 mm diameter . The next day , eight synchronized late L4 males were transferred onto each plate . ( Two or four males per plate for experiment in Figure 1A . ) One late L4 male of the same age and genotype was transferred onto the control plates . Males were transferred onto fresh plates every two days , when the males were lost or dead , males from other plates were transferred together to make the size of the group stable . Male-conditioned plates for lifespan assays were prepared similar to the previous description ( Maures et al . , 2014 ) . Briefly , 60 µl of OP50 was dropped onto each 35 mm NGM plate to make a bacterial lawn of ~25 mm diameter . Young Day 1 wild-type males ( fog-2 males ) were transferred onto each plate . Two days later , they were removed and worms for lifespan assays were immediately transferred onto these male-conditioned plates ( MCP ) . These male-conditioned plates were prepared throughout the course of the lifespan assays ( Figure 1—figure supplement 1B ) to ensure fresh MCP plates were available . The number of wild-type males used for conditioning is stated in the text and labeled in the figures . In Figure 1—figure supplement 1F , glp-1 mutant males as well as the wild-type males were used for conditioning at 25°C for 2 days . For MCP treatment from egg onward , 30 Day 1 wild-type males were used to condition plates for 2–3 days . Males were removed and about 15 Day 1 mated fog-2 hermaphrodites were picked onto these MCP plates for 4–5 hr , allowing them to lay 60–80 eggs . Two days later , L4 males were individually transferred onto MCP plates ( conditioned by 8 males , as previously described ) for the lifespan assays . Images of live males on 35 mm plates were taken daily for the first week of adulthood with a Nikon SMZ1500 microscope . Image J was used to analyze the body size of the worms . The middle line of each worm was delineated using the segmented line tool and the total length was documented as the body length of the worm . T-test was performed to compare the body size differences between groups of males in the same day . See Supplementary file 2 for summary . FUdR was added to the NGM media to the final concentration of 50 µM . Late L4 males and hermaphrodites were transferred onto NGM+FUdR plates seeded with OP50 . Worms were transferred every two days , and were kept on FUdR plates for different period of time ( 3 days , 6 days or lifetime as indicated by text ) . Worms were stained according to Bio-protocol ( http://www . bio-protocol . org/wenzhang . aspx ? id=77 ) using VECTASHIELD HardSet Mounting Medium with DAPI from Vector Laboratories ( Burlingame , CA ) . Images were taken with a Nikon Ti . The mitotically proliferating germline region was determined by the crescent shape of DAPI-stained nuclei in the transition zone . Z-series of pictures were taken and the numbers of cells in the mitotically proliferating germline region were counted manually . We scored the germline morphology as ‘1’ ( clear transition zone marked by crescent shaped nuclei and sperm after U turn of the germline ) , ‘2’ ( no clear transition zone ) , and ‘3’ ( sperm appear before U turn of the germline ) . Nonparametric comparison between each treatment group was performed using Prism Graphpad . Mann Whitney test was used to determine the statistical significance . Oil Red O staining was adapted from the published protocol for staining of a small number of worms ( Wählby et al . , 2014 ) . About 20 worms per treatment were imaged with Nikon Ti . Oil Red O quantification was also performed as published ( O'Rourke et al . , 2009 ) . In brief , the color images were split into RGB monochromatic images in Image J . The Oil-Red-O staining arbitrary unit ( a . u . ) was determined by mean gray value within the worm region by Image J ( Intensity in the Green channel was used as the signal , adjusted by the intensity in the Red channel as the background ) . T-test analysis was performed to compare the fat staining of different groups of worms . Glycogen staining was performed according to the published protocol ( Frazier and Roth , 2009 ) . Mating of males was set up as previously described . Right before staining , live males of the same group were picked into an M9 droplet with 1M sodium azide on a 3% agarose pad . Immediately after the liquid was dry , the pad was inverted over the opening of a 50g bottle of iodine crystal chips ( Sigma , St . Louis , MO ) for 1 min . After the color stained by iodine vapor on the pad disappeared ( non-specific staining ) , the worms ( about 20 worms per treatment ) were immediately imaged by a Nikon microscope . Due to uncontrollable differences , it is hard to compare the staining performed at different times . Thus , worms from the groups of comparison were mounted onto the same pad ( using a separate M9 droplet if there is no visible difference ) . Image J was used to compare the mean intensity of iodine staining after the background was subtracted . T-test was performed to compare the staining between different groups ( on the same pad ) . 10–20 worms of each group were imaged by Nikon Ti . Image J was used to measure the mean and the maximum GFP intensity of the whole body area . T-test analysis was performed to compare the GFP intensity of different groups of worms . We paired a single male with a fog-2 hermaphrodite for about 3 . 5 days of mating , then picked the males individually on Day four for microarray analysis . As a control , solitary males were collected at the same time . About 150 males ( on 150 individual 35 mm plates ) were collected for each condition and replicate . Three biological replicates were performed . RNA was extracted by the heat-vortexing method . Two-color Agilent microarrays were used for expression analysis; detailed steps and analysis were performed as we previously reported ( Luo et al . , 2010 ) . Synchronized late L4 daf-22 males were picked on to 35 mm plates ( control and MCP ) . 30 males per plate , 150 males in total were used for each biological replicate . Males were transferred on to freshly seeded plates or MCP plates every two days , and collected on Day six for RNA extraction . Four biological replicates were performed . Synchronized late L4 worms were picked onto 35 mm plates . In the ‘grouped’ condition , 30 hermaphrodites were picked onto one plate , and ~120 worms were used for each replicate . Worms were transferred every two days to exclude progeny , and were collected on Day six for RNA extraction . Four biological replicates were performed . Microarray data can be found in PUMAdb ( http://puma . princeton . edu ) . https://puma . princeton . edu/cgi-bin/publication/viewPublication . pl ? pub_no=576 Including mated males microarrays ( three biological replicates ) ; daf-22 grouped males microarrays ( four biological replicates ) ; glp-1 hermaphrodites treated with MCP microarrays ( four biological replicates ) ; EG4389 masculinized hermaphrodites grouped vs single microarrays ( four biological replicates ) . Significant differentially-expressed gene sets were identified using SAM ( Tusher et al . , 2001 ) . Previously reported microarray results exploring the effect of males on hermaphrodites ( Maures et al . , 2014 ) were downloaded from NCBI and compared to our differentially expressed gene lists . Enriched motifs were found using RSAT ( van Helden , 2003 ) . This assay ( Figure 6H ) was modified from a previous assay ( Chasnov et al . , 2007 ) . 10 Day 1 virgin C . remanei or C . elegans hermaphrodites were placed in 100 µl of M9 buffer at room temperature overnight with shaking . 100 males of either C . elegans or C . remanei were placed in 100 µl of M9 . The supernatant solutions were then used for the pheromone chemotaxis assay . 60 mm NGM plates ( no food ) were used for the chemotaxis assay . Two destination spots ( supernatant and M9 control ) were separated by about 45 mm; the distance from the origin spot to either destination spot is 30 mm . Two 1 µl drops of 1M sodium azide were first applied to the destination spots . When dry , a drop of 1 µl M9 or supernatant was separately added onto the destination spots . Then , over 10 young adult ( Day 2 ) males were placed at the origin spot , transferring as little bacteria as possible . After 60 min , the paralyzed male worms were scored based on their location . The chemotaxis index was calculated as: ( #worms at supernatant destination - #worms at control destination ) / ( #total worms - #worms at origin ) . The chemotaxis assay in Figure 2B was also modified from established protocol ( Chasnov et al . , 2007 ) . 10 Day 5 hermaphrodites of either N2 or 6699 EG4389 were put in 100 µl of M9 buffer at room temperature overnight with shaking . Two destination spots were 3 mm apart . The origin spot was in the middle . 20 Day 1 fog-2 males were used in each assay; two replicates were performed . Males from the control plates and MCP plates were individually paired with one virgin Day 1 fog-2 hermaprhodite at various time points on a seeded 35 mm NGM plate for 24 hr . About 20 pairs were set up for each group in each biological replicate . The percent fertile was calculated from the number of plates with eggs/progeny divided by the total number of plates set up for this group . Each mated hermaphrodite was numbered and was transferred individually onto a new seeded NGM plate every day to count the total progeny/male number .	In many animals , different sexes have different life expectancies . This holds true for a roundworm species called Caenorhabditis elegans that has commonly been used to study aging and lifespan . Unlike some related Caenorhabditis roundworm species ( which consist of male and female worms ) , C . elegans worms are predominantly hermaphrodites and reproduce by self-fertilization . C . elegans males are normally rare . However , under stressful conditions the number of males increases to reduce inbreeding and so help the worm population to adapt to the environment . Investigations into the factors that affect the lifespan of C . elegans have mostly studied hermaphrodites . For example , one recent study showed that mating shortens the lifespan of hermaphrodites . Another study showed that pheromones – hormones that change the behavior of other worms – also shorten hermaphrodite lifespan . The male pheromone is produced by males and sensed by both males and hermaphrodites . But does mating and male pheromone affect the lifespan of male roundworms ? Shi et al . have now studied Caenorhabditis worms of different species and sexes to investigate how sexual behaviors and male pheromone regulate the lifespan of male roundworms . The results of the experiments revealed two distinct mechanisms of male death . Firstly , mating caused the males of many different Caenorhabditis species to shrink and die , and also killed females and hermaphrodites . Secondly , the males of hermaphroditic species – and only these males – could also be killed by male pheromone . The results suggest that death from mating may be an unavoidable cost of reproducing that is seen across all sexes and species of roundworm . In contrast , death by male pheromone may be a way of culling the male population in hermaphroditic species , for example , after stressful conditions have caused a sudden increase in the number of male worms . Further work is now needed to investigate the finer details of the mechanisms by which mating and male pheromone cause death . Ultimately , this work in Caenorhabditis could be extended to help us to understand how other animals regulate their lifespan and maintain an optimum ratio of the sexes .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "developmental", "biology", "evolutionary", "biology" ]	2017	Mating and male pheromone kill Caenorhabditis males through distinct mechanisms
Infection with the food-borne liver fluke Opisthorchis viverrini is the principal risk factor ( IARC Working Group on the Evaluation of Carcinogenic Risks to Humans , 2012 ) for cholangiocarcinoma ( CCA ) in the Lower Mekong River Basin countries including Thailand , Lao PDR , Vietnam and Cambodia . We exploited this link to explore the role of the secreted growth factor termed liver fluke granulin ( Ov-GRN-1 ) in pre-malignant lesions by undertaking programmed CRISPR/Cas9 knockout of the Ov-GRN-1 gene from the liver fluke genome . Deep sequencing of amplicon libraries from genomic DNA of gene-edited parasites revealed Cas9-catalyzed mutations within Ov-GRN-1 . Gene editing resulted in rapid depletion of Ov-GRN-1 transcripts and the encoded Ov-GRN-1 protein . Gene-edited parasites colonized the biliary tract of hamsters and developed into adult flukes , but the infection resulted in reduced pathology as evidenced by attenuated biliary hyperplasia and fibrosis . Not only does this report pioneer programmed gene-editing in parasitic flatworms , but also the striking , clinically-relevant pathophysiological phenotype confirms the role for Ov-GRN-1 in virulence morbidity during opisthorchiasis . Liver fluke infection caused by species of Opisthorchis and Clonorchis remains a major public health problem in East Asia and Eastern Europe . O . viverrini is endemic in Thailand and Laos , where ~10 million people are infected with the parasite ( Sripa et al . , 2011 ) . In liver fluke endemic regions , this infection causes hepatobiliary morbidity including cholangitis , choledocholithiasis ( gall stones ) , and periductal fibrosis , and is the principal risk factor for bile duct cancer , cholangiocarcinoma ( CCA ) ( Sripa et al . , 2011; Sripa et al . , 2007; Mairiang et al . , 2012; Tyson and El-Serag , 2011; Shin et al . , 2010a ) . Indeed , there is no stronger link between a human malignancy and a parasitic infection than that between CCA and infection with O . viverrini ( Pagano et al . , 2004 ) . Northeastern Thailand suffers the highest incidence of CCA in the world , often exceeding 80 cases per 100 , 000 population and for which up to 20 , 000 people annually are admitted for surgery . The prognosis for liver fluke infection-induced cancer remains poor ( Sripa et al . , 2011; Khuntikeo et al . , 2015; Khuntikeo et al . , 2016; Luvira et al . , 2016 ) . How and why opisthorchiasis induces cholangiocarcinogenesis is likely multi-factorial , including mechanical irritation of the biliary tract during migration and feeding of the liver fluke , secretion by the parasite of inflammatory molecules , and nitrosamines in fermented foods that are a dietary staple in northeastern provinces of Thailand ( Songserm et al . , 2012 ) . To survive in the hostile host environment , parasitic helminths produce an assortment of excretory/secretory ( ES ) products including proteins with diverse roles at the host–parasite interface . This interaction has long been thought , but not fully understood , to modify cellular homeostasis and contribute to malignant transformation during chronic opisthorchiasis ( Brindley and Loukas , 2017 ) . Feeding activity of the liver fluke inflicts wounds in the biliary tree , resulting in lesions that undergo protracted cycles of repair and re-injury during chronic infection . The liver fluke secretes mediators that accelerate wound resolution in monolayers of cultured cholangiocytes , an outcome that is compromised following silencing of expression of the liver fluke secreted growth factor Ov-GRN-1 using RNA interference ( Papatpremsiri et al . , 2015; Smout et al . , 2015 ) . We hypothesize that proliferation of biliary epithelial cells induced by Ov-GRN-1 is a pivotal factor in maintenance and progression of a tumorigenic microenvironment in the liver during chronic opisthorchiasis . Progress with development of genetic tools for functional genomic studies with platyhelminth parasites has been limited to date ( Hoffmann et al . , 2014 ) . The use of clustered regularly interspaced short palindromic repeats ( CRISPR ) associated with Cas9 , an RNA-guided DNA endonuclease , has revolutionized genome editing in biomedicine , agriculture and biology ( Hsu et al . , 2014; Sander and Joung , 2014 ) . Progress with CRISPR/Cas9 in numerous eukaryotes including the nematodes Caenorhabditis elegans , Strongyloides stercoralis and Strongyloides ratti has been described ( Sander and Joung , 2014; Waaijers and Boxem , 2014; Lok et al . , 2017; Gang et al . , 2017 ) , but this form of gene editing has not been reported for flatworm parasites . Here , we deployed a CRISPR/Cas9-based approach , aiming to knockout ( mutate ) the Ov-GRN-1 gene and assess the virulence of gene-edited flukes in vitro and in vivo in a hamster model of opisthorchiasis . Following transfection of adult flukes with the gene-editing construct targeting Ov-GRN-1 , the activity and efficiency of programmed editing was evaluated by two approaches . First , quantitative PCR ( qPCR ) was employed , which relies on the inefficiency of binding of a primer ( here termed OVR-F ) overlapping the target genomic sequence of the guide RNA ( gRNA ) , that is where mutations are expected to have occurred , compared to the binding efficiency of flanking primers , that is outside the mutated region ( flanking primers termed OUT-F and OUT-R ) ( Figure 1A and B ) . The ratio between the OVR-F and OUT-R products and OUT-F and OUT-R products provided an estimate of the amplification fold-reduction in the sample of CRISPR/Cas9-edited compared to genomic DNA ( gDNA ) from control , wild-type liver flukes at the target sequence of the sgRNA , that is the annealing site for the OVR primer ( Shah et al . , 2015; Yu et al . , 2014 ) . A reduction in relative fold amplification of 2 . 7% was detected in gDNA from the Cas9-treated worms ( Figure 1E , Figure 1—figure supplement 1C ) . Second , to identify , quantify and characterize the mutations that arose in the genome of Ov-GRN-1-edited ( termed ΔOv-GRN-1 ) flukes , we used an amplicon-sequencing approach . A targeted ( amplicon ) sequence library was constructed from gDNA from some of the flukes ( 7 to 21 days after pCas-Ov-GRN-1 transfection ) . A fragment of 173 bp spanning the predicted site of the programmed double stranded break of Ov-GRN-1 was amplified from gDNA primed with oligonucleotides flanking 1496–1668 nt of Ov-GRN1 . Adaptors and barcodes were ligated into the amplicon libraries . Deep sequencing of the amplicon libraries was undertaken using the Illumina MiSeq system . Insertion-deletion ( INDEL ) /mutation profiles in the sequence reads were compared in multiple sequence alignments with the reference template sequence , nt 1 , 496–1 , 668 of wild type Ov-GRN-1 . The CRISPResso computational pipeline was used to quantify gene-editing outcomes and efficiency ( Canver et al . , 2018; Pinello et al . , 2016 ) ; among >2 million reads aligned against the reference sequence , 27 , 640 sequence reads exhibited non-homologous end joining ( NHEJ ) mutations , including 170 reads with insertions ( 0 . 6% ) , 193 reads with deletions ( 0 . 7% ) and 27 , 277 reads with substitutions ( 98 . 7% ) . Overall , 1 . 3% of the sequenced reads exhibited NHEJ mutations ( Figure 1C ) . Regarding the NHEJ-bearing reads , >100 forms exhibited mutations that would disrupt the coding sequencing of Ov-GRN-1 . Four representatives of the INDEL-bearing traces , aligned with the wild type ( WT ) allele are presented in Figure 1—figure supplement 1B . These and related ( below ) sequence reads are available at GenBank Bioproject PRJNA385864 , Biosample SAMN07287348 , SRA study SRP110673 , accessions SRR5764463-5764618 and SRR8187484-SRR8187487 , at https://www . ncbi . nlm . nih . gov/Traces/study/ ? acc=SRP110673 , Bioproject , www . ncbi . nlm . nih . gov/bioproject/PRJNA385864 . Effects of gene editing on transcription and protein expression in adult flukes were investigated . Levels of both Ov-GRN-1 mRNA transcripts as determined by reverse transcription ( RT ) -qPCR and of Ov-GRN-1 protein , as detected by western blot using anti-Ov-GRN-1 serum , fell significantly from days 1 and 2 after transfection , respectively ( p≤0 . 0001; Figure 1D and E , Figure 1—figure supplement 1C ) . Expression levels of two reference genes encoding actin ( Figure 1—figure supplement 1C ) and the Ov-TSP-2 tegument protein ( Figure 1D ) were not influenced by the programmed mutation of Ov-GRN-1 . These findings , revealing diminished RNA and protein following programmed mutation indicated that CRISPR/Cas9 catalyzed programmed gene-editing of Ov-GRN-1 was active in adult flukes in vitro . Thereafter , to investigate whether gene editing of Ov-GRN-1 impacted in vitro indicators of pathogenesis , the capacity of ES products from WT , mock-transfected and gene-edited flukes to drive proliferation and scratch wound repair of the H69 human cholangiocyte cell line was assessed . ES from WT and mock-transfected adult flukes stimulated cell proliferation and wound closure whereas an equivalent amount of ES products from ΔOv-GRN-1 flukes resulted in significantly reduced cell proliferation over the 6-day course of the assay ( p ≤ 0 . 0001; Figure 2A and B , Figure 2—figure supplement 1A and B ) and significantly reduced in vitro wound closure over 36 hr ( p ≤ 0 . 0001; Figure 2C and D , Figure 2—figure supplement 1C and D ) , consistent with the reduction in Ov-GRN-1 protein secreted from the gene-edited liver flukes . Notwithstanding the marked effects observed with gene-edited , adult developmental forms , the metacercaria ( MC ) ( Figure 3A ) is the infective stage of O . viverrini for humans . Accordingly , we investigated gene knockout in MC . Significant differences in Ov-GRN-1 transcript levels were noted between groups of MC ( p ≤ 0 . 01 ) , but the effect was modest , ≤4% , at each time point ( Figure 3—figure supplement 1 ) , suggesting that delivery of the pCas-Ov-GRN-1 by electroporation through the MC cyst wall was ineffective . Exposure to bile acids and gastric enzymes results in excystation of O . viverrini MC in the duodenum of the mammalian host ( Sripa et al . , 2011 ) . Using trypsin , here the process was mimicked in vitro to release the newly excysted juvenile worms ( NEJ ) ( Figure 3B ) , after which these NEJs were subjected to electroporation with the CRISPR/Cas9 plasmid construct , in like fashion to the adult developmental stage of O . viverrini ( above ) . Following this manipulation , marked depletion of Ov-GRN-1 transcripts in NEJ was evident by 24 hr later ( p ≤ 0 . 0001 ) ( Figure 3C ) . In parallel , hamsters were infected with 100 ΔOv-GRN1 NEJs or WT NEJs immediately after electroporation . At necropsy of the hamsters 14 days later , similar numbers of WT and ΔOv-GRN-1 flukes were observed in the bile ducts , and they were similarly motile ( not shown ) . Strikingly , however , acute infection with ΔOv-GRN-1 parasites failed to induce the marked hyperplasia of the biliary epithelia characteristic of chronic opisthorchiasis . Specifically , infection with WT flukes induced markedly disordered , hyperplasic growth of the epithelium adjacent to the parasites; ~500% thickening of the biliary epithelium compared to uninfected controls as measured in two-dimensional image analysis of H and E-stained thin sections ( p ≤ 0 . 0001 ) . By contrast , infection with the ΔOv-GRN-1 flukes provoked significantly less ( p ≤ 0 . 0001 ) biliary hyperplasia than WT flukes ( 145% thickening compared to uninfected controls; p ≤ 0 . 01 ) . Indeed , the bile ducts from hamsters infected with the ΔOv-GRN-1 flukes generally resembled those of the uninfected control hamsters ( Figure 3D–G ) . At 60 days after infection , significant differences in biliary hyperplasia remained between hamsters infected with WT ( 216% ) and ΔOv-GRN-1 ( 162% ) flukes ( p ≤ 0 . 05 ) , although this was less marked than during acute infection at day 14 ( Figure 3G ) . To evaluate disease during chronic infection with ΔOv-GRN-1 liver flukes and associated chronic biliary morbidity , hamsters were infected with ΔOv-GRN-1 and WT NEJ , and adult flukes were recovered and counted from the livers 60 days post-infection . Similar numbers of worms were recovered from both control and gene-edited liver fluke-infected hamsters ( Figure 4A ) . To assess the impact of infection with ΔOv-GRN-1 on markers of chronic opisthorchiasis including biliary fibrosis , liver sections from infected hamsters were stained with Picro-Sirius Red to localize collagen bundles in the biliary tract ( Figure 4B ) . Minimal deposits of collagen were seen in the periductal regions of the biliary tract of the uninfected control hamsters . By contrast , thick bands of collagen surrounded the enlarged bile ducts in the vicinity of the flukes in the hamsters infected with WT parasites . Significantly less collagen ( 28% ) had been deposited in periductal regions of hamsters infected with ΔOv-GRN-1 flukes compared to livers of hamsters infected with WT flukes ( p ≤ 0 . 001 ) ( Figure 4B and C ) . To further assess fibrosis , thin sections of livers were immuno-stained for alpha-smooth muscle actin ( α-SMA or ACTA2 ) , a marker of hepatic fibrosis ( Guido et al . , 1997 ) . Livers of hamsters infected with WT flukes showed densely packed collagen fibrils that stained for ACTA2 in periductal regions proximal to the parasites . In contrast , livers from hamsters infected with ΔOv-GRN-1 flukes displayed an irregular distribution of less dense collagen fibrils with less ACTA2-specific fluorescence ( Figure 4D , Figure 4—figure supplement 1 ) . Measuring Alexa-594 fluorescence quantified the expression levels of ACTA2 . Median levels of ACTA2 ( quantified using Alexa-594-anti-ACTA2 ) in the livers of ΔOv-GRN-1 fluke-infected hamsters were significantly reduced ( 94% ) compared to those of WT fluke-infected hamsters ( p ≤ 0 . 01 ) ( Figure 4E ) . Bile ducts parasitized by the gene-edited worms displayed a broad range of fibrosis from minimal to marked , as established by staining both with Sirius Red and with antibody specific for alpha-smooth muscle actin . This situation may have reflected unevenness in level of programmed mutation of the Ov-GRN-1 gene in cells within and/or among individual liver flukes . To investigate this situation further , we assessed transcription of the Ov-GRN-1 gene from individual adult flukes recovered from hamsters 60 days after infection with gene edited NEJ . This revealed that levels of Ov-GRN-1 mRNA in the ΔOv-GRN-1 group flukes were 81% lower , in aggregate , than the control wild-type flukes ( Figure 5A ) . Thereafter , to evaluate the mutation rate of the gene editing approach , which involved transfection by electroporation of batches of 750 NEJs , adult flukes at necropsy were assigned to one of three groups based on Ov-GRN-1 mRNA expression levels , as follows: ( i ) ≥ 100% relative to WT mean , that is , low ( L ) efficiency of programmed gene editing; group was termed LΔOv-GRN-1; ( ii ) > 10 to<100% relative to WT mean , that is moderate ( M ) level efficiency of programmed gene editing; termed MΔOv-GRN-1; and ( iii ) ≤ 10% relative to WT mean , that is high ( H ) level efficiency of programmed gene editing; termed HΔOv-GRN-1 . Genomic DNAs pooled from 7 to 10 worms of each group were studied to quantify the efficiency of gene editing , using both the NGS CRISPResso and the tri-primer qPCR approaches . The NGS CRISPResso analysis revealed mutation rates of 1 . 3 , 5 . 9 and 17 . 2% in the L , M , and H groups of ΔOv-GRN-1 worms , respectively . The tri-primer qPCR analysis indicated mutation levels of 0 . 7 , 3 . 2 and 4 . 6% in these groups , respectively . Both approaches confirmed that the efficiency of programmed gene editing negatively correlated with levels of the Ov-GRN-1 transcripts ( Figure 5A and B ) . The combined mutation frequency among all three groups by the two approaches was 8 . 1% and 2 . 7% , with the 2 . 7% rate estimated by tri-primer qPCR indicating the same level as the mutation rate of 2 . 7% observed during culture of adult stage ΔOv-GRN-1 flukes for 7 to 21 days in vitro ( Figure 1E , Figure 1—figure supplement 1C ) . In addition , the NGS CRISPResso analysis of the sequence reads of the gene-edited L , M and H groups compared with those from the control WT group ( GenBank accessions SRR8187484-SRR8187487 , 5 to 10 million reads per targeted amplicon library ) provided details of the nature and types of the mutations as insertions , deletions and/or substitutions following NHEJ events that repaired the programmed cleavage of the Ov-GRN-1 locus . The analysis also revealed increasing ratio of substitutions among the mutations among the LΔOv-GRN-1 , MΔOv-GRN-1 and HΔOv-GRN-1 groups ( Figure 5B ) . Lastly , these findings also demonstrated the longevity of the programmed mutation at Ov-GRN-1; mutations were retained in the parasite for at least 60 days during active infection of the mammalian host . This report , and the accompanying article on schistosomes ( Ittiprasert et al . , 2019 ) , pioneer programmed gene editing using CRISPR/Cas9 of trematodes and indeed genome editing for species of the phylum Platyhelminthes . The findings revealed that somatic tissue gene editing disrupted the expression of liver fluke granulin , resulting in a clinically noteworthy phenotype of attenuated hepatobiliary tract morbidity . Scrutiny of the nucleotide sequence reads indicated that the chromosomal break took place as programmed and was repaired subsequently by NHEJ following Cas9-catalyzed mutation ( Albadri et al . , 2017 ) . Accordingly , the findings confirmed that the bacterial Type II Cas9 system is active in O . viverrini , and we suggest that Cas9-mediated programmed gene editing and repair by homology directed repair and NHEJ will be active in other genes of the liver fluke , and in other trematodes and parasitic platyhelminths generally . Although the findings demonstrated programmed gene editing of the Ov-GRN-1 locus , the somatic mutation rate in the adult developmental stage was generally <5% of the genomes recovered from these multicellular parasites . This low mutation rate contrasted with both the marked reduction in Ov-GRN-1 message detected in vitro and the pathophysiological outcomes and reduced virulence of infection of hamsters with gene-edited flukes . The anomaly might be explained by the tissue expression of secreted Ov-GRN-1 . Although it exhibits generalized expression throughout tissues of the adult liver fluke , predominant expression of Ov-GRN-1 has been immunolocalized to the tegumental surface , tegumental cytons and gut ( Smout et al . , 2009 ) . Given that the flukes were transfected in vitro with the gene editing plasmid by square wave electroporation , gene knockout of the target Ov-GRN-1 locus in nuclei of cells in the tegument and gut may have occurred more frequently than in cells deeper within the fluke . If so , this may explain the marked reduction of expression and secretion of Ov-GRN-1 in tandem with a limited rate of mutation estimated in genomic DNA pooled from the gene-edited flukes . The activity in vitro of liver fluke granulin in cell proliferation , wound repair and angiogenesis has been established ( Papatpremsiri et al . , 2015; Smout et al . , 2015; Smout et al . , 2009 ) , which has prompted the development of therapeutic peptides based on the Ov-GRN-1 scaffold for treatment of non-healing wounds ( Bansal et al . , 2017; Dastpeyman et al . , 2018 ) . The novel findings reported here corroborate earlier in vitro reports and extend the findings in a rodent model of human opisthorchiasis . Programmed gene editing confirmed that secreted parasite granulin induces hyperplasia of the biliary epithelium and fibrosis during chronic infection , and that liver fluke granulin directly contributes to morbidity of the hepatobiliary tract during both acute and chronic opisthorchiasis . The impact of Ov-GRN-1 might emulate the action of interleukin IL−33 , an epithelial mitogen for cholangiocytes , in the development of CCA . IL-33 primes type two innate lymphoid cells to induce proliferation of neighboring cholangiocytes by the release of IL-13 ( Brindley and Loukas , 2017; Li et al . , 2014 ) . The pathophysiological bioactivity of granulin warrants deeper investigation of its role in fibrosis , including the influence on hepatic stellate cells , during liver fluke infection and cholangiocarcinogenesis ( Guido et al . , 1997; Yin et al . , 2013; Gouveia et al . , 2017; Rockey et al . , 2015 ) . The rigor of future gene editing investigations might be enhanced with the inclusion of additional controls including parasites transfected with an otherwise functional vector that lacks target-specific gRNA and/or a gRNA with a scaffold but without seed sequence and/or containing a seed sequence without homology in the genome of O . viverrini . These additional controls would address non-target-specific effects of expression of Cas9 including on the genetic fitness of the genome-edited parasites ( Cox et al . , 2015; Kosicki et al . , 2018; Ihry et al . , 2018 ) . Likewise , in addition to estimation of gene-editing performance and efficiency of somatic cell gene-editing in this multicellular helminth parasite using NGS-based ( Shah et al . , 2015; Canver et al . , 2018; Albadri et al . , 2017 ) and quantitative PCR-based approaches ( Shah et al . , 2015; Yu et al . , 2014 ) , droplet digital PCR ( ddPCR ) -based analysis should provide more sensitive detection and quantification of gene-editing manipulations . The ddPCR approach can provide simultaneous assessment of both homology directed repair and NHEJ , the repair pathways that resolve Cas9 catalyzed double-stranded breaks , and also investigate multiple , simultaneous editing conditions at the target locus ( Miyaoka et al . , 2018 ) . With respect to Ov-GRN1 and its tissue site of expression , the anomaly between the marked knockdown of transcript levels and the minority of genomes mutated by the programmed gene editing among the total number of cells in this liver fluke , is amenable to deeper inquiry . Characterizing by immunolocalization the site of expression in the parasite from hamsters infected with gene-edited NEJ and/or the location of the gene editing plasmid after transfection of the liver fluke should be instructive . The causative agent for many cancers remains obscure including non-liver fluke infection-associated CCA . By contrast , the principal risk factor in liver fluke-endemic regions is well established: infection with O . viverrini and related parasites ( IARC Working Group on the Evaluation of Carcinogenic Risks to Humans , 2012; Fedorova et al . , 2017; Shin et al . , 2010b ) . CRISPR/Cas9-based gene editing and the hamster model of human opisthorchiasis utilized here ( Sripa et al . , 2007 ) , including genetic manipulation of the larval infective stage of the parasite , provide a facile , functional genomics system to interrogate this host-parasite relationship and pathophysiology ( Hoffmann et al . , 2014 ) . In a related model , periductal fibrosis induced by the liver fluke infection combined with ingestion of dimethylnitrosamine or similar nitric oxide carcinogen results in epithelial hyperplasia , cholangiocyte proliferation and DNA damage , which culminates in CCA ( Thamavit et al . , 1987; Maksimova et al . , 2017 ) . Investigation utilizing genome edited liver flukes , mutated at loci encoding granulin or other parasite products can now proceed , including interaction of liver fluke granulin with cholangiocyte signaling pathways that are frequently mutated during liver fluke infection-induced CCA ( Jusakul et al . , 2017 ) . Metacercariae ( MC ) of O . viverrini were isolated from the naturally infected cyprinid fish by pepsin digestion as described ( Pinlaor et al . , 2013 ) . In brief , fishes were homogenized using an electric blender , after which the homogenate was incubated for 120 min at 37°C in 0 . 25% porcine pepsin , 1 . 5% HCl , 150mM NaCl . Subsequently , the digest was filtered sequentially through sieves of 1100 , 350 , 250 and 140 µm diameter pore size . After gravity sedimentation of the final filtrate , the aqueous supernatant was discarded , the MC-enriched sediment was washed once in 150 mM NaCl , and the identity of MC as O . viverrini confirmed using a stereomicroscope . Batches of MC were stored in 150 mM NaCl at 4°C . The newly excysted-juvenile flukes ( NEJ ) were liberated from MC by incubation in 0 . 25% trypsin in 1× PBS supplemented with 2× 200 U/ml penicillin , 200 μg/ml streptomycin ( Gibco ) ( 2× Pen/Strep ) for 5 min at 37°C in 5% CO2 atmosphere , after which NEJ were separated from the discarded cyst walls of the MC by mechanical passage through a 27G ( insulin ) needle ( Papatpremsiri et al . , 2015; Papatpremsiri et al . , 2016 ) . Before use , NEJ were transferred into RPMI medium supplemented with 1% glucose , 2 g/l NaHCO3 , 2× Pen/Strep , 1μM E-64 ( Thermo Fisher Scientific ) for 60 min at 37°C in 5% CO2 atmosphere . To obtain the adult developmental stage of the liver fluke , Syrian golden hamsters ( Mesocricetus auratus ) were infected by intragastric tube at 6–8 weeks of age with 50 MC per hamster ( Sripa and Kaewkes , 2002 ) . The hamsters were maintained at the rodent facility of the Faculty of Medicine , Khon Kaen University , Khon Kaen . Sixty days after infection , hamsters were euthanized , and the liver flukes collected as described ( Sripa and Kaewkes , 2002 ) . The Animal Ethics Committee of Khon Kaen University approved the study , approval number ACUC-KKU-61/60 , which adhered to standard guidelines of the National Research Council of Thailand for the Ethics of Animal Experimentation . To edit the gene Ov-GRN-1 that encodes O . viverrini granulin-1 ( 6 , 287 bp , mRNA GenBank FJ436341 . 1 ) ( Smout et al . , 2009; Young et al . , 2014 ) , online tools including CRISPR design , http://crispr . mit . edu/ ( Ran et al . , 2013 ) and ChopChop , http://chopchop . cbu . uib . no/ ( Labun et al . , 2016; Montague et al . , 2014 ) were employed to design a single guide RNA ( sgRNA ) targeting exon 1 of the Ov-GRN-1 gene at nucleotide position 1589–1608 , 5'-GATTCATCTACAAGTGTTGA ( Figure 1A and B ) . The programmed cleavage site was predicted to be located at three residues upstream of a CGG proto-spacer adjacent motif ( PAM ) sequence in exon 1 of Ov-GRN-1 ( Figure 1B , Figure 1—figure supplement 1B ) . A CRISPR/Cas9-encoding vector encoding this sgRNA under the control of the mammalian U6 promoter and encoding Cas9 ( with nuclear localization signal 1 and 2 ) driven by the CMV promoter was assembled ( GeneArt CRISPR Nuclease Vector Kit , Invitrogen ) , and termed pCas-Ov-GRN-1 ( Figure 1—figure supplement 1A ) . Escherichia coli TOP-10 competent cells were transformed with pCas-Ov-GRN-1 after which the plasmid was recovered from cultures of a positive clone ( NucleoBond Xtra Midi , Macherey-Nagel GmbH , Germany ) . The nucleotide sequence of pCas-Ov-GRN-1 was confirmed by Sanger direct cycle sequencing . Pools of 20 mature adult flukes were simultaneously subjected to transfection with 10 µg pCas-Ov-GRN-1 plasmid DNA in ~500 µl RPMI-1640 ( Sigma ) by electroporation; all 20 flukes were included in the same cuvette during electroporation . The electroporation was performed in 4 mm cuvettes ( Bio-Rad ) with a single square wave pulse of 125 volts for 20 ms using a Gene Pulser Xcell ( Bio-Rad ) ( Papatpremsiri et al . , 2015; Piratae et al . , 2012 ) . Flukes were then washed several times with 150 mM NaCl and an additional five times with RPMI-1640 containing 2× Pen/Strep . Flukes were cultured in RPMI-1640 containing 2× Pen/Strep at 37°C in 5% CO2 atmosphere ( Papatpremsiri et al . , 2015; Piratae et al . , 2012 ) . Two control groups were included: wild-type ( WT ) mature flukes and ‘mock’ control flukes which were exposed to identical electroporation conditions with RPMI-1640 and 1× Pen/Strep in the absence of plasmid DNA . The adult flukes were observed and collected after 1 , 2 , 3 , 5 , 7 , 14 and 21 days of culture following pCas-Ov-GRN-1 transfection . RNA and protein were extracted from individual flukes and Ov-GRN-1 mRNA expression was assessed by RT-qPCR and Ov-GRN-1 protein expression was assessed by western blot . Mutations and/or insertions-deletions ( INDELs ) resulting from CRISPR/Cas were estimated by two discrete types of analysis: 1 ) by Illumina-based Next Generation Sequencing ( NGS ) ( Shah et al . , 2015; Albadri et al . , 2017 ) ; 2 ) by CRISPR efficiency estimation ( Shah et al . , 2015; Yu et al . , 2014; Yang et al . , 2017 ) , a method based on the differences in RT-qPCR efficiencies between amplification of the WT and mutant sequence with a primer spanning the targeted mutation site . MC and NEJ ( 750 larvae per cuvette ) were subjected to square wave electroporation in the presence of pCas-Ov-GRN-1 pDNA as described above for adult flukes . The larvae were washed as above and cultured in RPMI complete medium ( 2× Pen/Strep ) at 37°C in 5% CO2 atmosphere . Transcript levels for Ov-GRN-1 on days 1 , 2 , 3 , and 5 after transfection were ascertained by RT-qPCR , as above . RNA was extracted from pooled or individual transfected flukes using the TRIzol reagent ( Invitrogen ) . Concentration of RNA was estimated by absorbance at 260 nm using a NanoVue spectrophotometer . Genomic DNA was extracted from individual adult flukes using the QIAamp DNA Mini Kit ( Qiagen ) . A dual RNA and DNA extraction was used for individual worms at day 60 after infection of hamsters with Ov-GRN-1 gene-edited NEJ , using RNAzol RT and DNAzol ( Molecular Research Center , Inc . ) ( Chan et al . , 2014; Chen et al . , 2010 ) . In brief , each worm was homogenized in RNAzol RT using a motorized pestle , the DNA and protein from the lysate was precipitated using DNAse-RNAse-free water . The aqueous phase ( top ) was transferred into isopropanol to precipitate the RNA . The DNA/protein pellet was resuspended in DNAzol , and DNA extracted as per the manufacturer’s instructions . Expression levels of Ov-GRN-1 in total RNA recovered from individual liver flukes were determined . To assess the performance of the gene editing approach , following necropsy of hamsters and recovery of the liver flukes , the adult worms were assigned to one of three phenotypes based on the levels of Ov-GRN-1 transcript knockdown , low ( L ) , moderate ( M ) or high ( H ) , as follows: L , ≥100% relative to WT mean ( low efficiency of programmed genome editing ) , group termed LΔOv-GRN-1; M , >10 to<100% relative to WT mean , group termed MΔOv-GRN-1; and H , ≤10% relative to WT mean , group termed HΔOv-GRN-1 . Pools of genomic DNAs from flukes , which had been assigned to each of the L , M and H groups of Ov-GRN1 transcript knockdown levels , were quantified for efficiency of CRISPR/Cas9-programmed gene editing in terms of mutation levels by qPCR and Illumina-based deep sequencing ( below ) ( Yang et al . , 2017; Vasquez et al . , 2018 ) . The data for the pooled samples from each group are based on a single Illumina run , that is n = 1 sample for each of the L , M and H genomic DNA pools . Complementary DNA ( cDNA ) was synthesized from parasite total RNA using an iScript cDNA synthesis kit ( Thermo Fisher Scientific ) prior to proceeding with reverse transcription quantitative real-time PCR ( RT-qPCR ) . RT-qPCR was performed with biological triplicate samples using a SYBR Green kit ( Takara Bio USA , Inc . , Mountain View , CA ) in a thermal cycler ( Light Cycler 480 II , Roche Diagnostics GmbH , Mannheim , Germany ) . Each RT-qPCR reaction consisted of 7 . 5 μl SYBR Green Master Mix , 0 . 5 μl ( 10 μM ) each of specific forward and reverse primers for Ov-GRN-1 ( Figure 1B ) ( forward primer , Ov-GRN-1-RT-F: 5'-GGGATCGGTTAGTCTAATCTCC and reverse primer , Ov-GRN1-RT-R: 5'-GATCATGGGGGTTCACTGTC ) , amplifying 359 base pairs ( bp ) of the product ( nt 7365 of O . viverrini granulin-1 mRNA , GenBank FJ436341 . 1 ) , 2 μl of cDNA and distilled water to a final volume of 15 μl . The thermal cycle was a single initiation cycle at 95°C for 3 min followed by 40 cycles of denaturation at 95°C for 30 s , annealing at 55°C for 30s , extension at 72°C for 45s and a final extension at 72°C for 10 min . The endogenous actin gene ( GenBank EL620339 . 1 ) was used as a housekeeping control ( Papatpremsiri et al . , 2015; Piratae et al . , 2012; Chaiyadet et al . , 2017 ) ( forward primer , Ov-actin-F: 5'-AGCCAACCGAGAGAAGATGA and reverse primer Ov-actin-R: 5'-ACCTGACCATCAGGCAGTTC ) . The fold change in Ov-GRN-1 transcripts was calculated by the 2 ( -ΔΔCt ) method using Ov-actin for normalization ( Papatpremsiri et al . , 2015; Piratae et al . , 2012; Schmittgen and Livak , 2008 ) . Means and standard deviations were calculated and means compared by two-way ANOVA using GraphPad Prism software . One milligram of adjuvanted , recombinant Ov-GRN1 protein ( Smout et al . , 2009; Strannegård and Yurchision , 1969 ) was subcutaneously injected into an outbred New Zealand White rabbit . The rabbit was boosted twice with 500 µg of adjuvanted protein , and 2 weeks after the last booster the rabbit was euthanized after which blood was collected by cardiac puncture ( Animal Ethics Committee , Khon Kaen University , approval no ACUC-KKU-61/60; see above ) . Ov-GRN-1 protein levels were determined by western blot using rabbit anti-recombinant Ov-GRN-1 antiserum . The adult flukes from either WT or ΔOv-GRN-1 groups were collected individually at days 1 , 2 , 3 , 5 , 7 , 14 and 21 after electroporation ( three flukes per group ) . Groups of three flukes were homogenized by sonication ( Sonics and Materials ) in 1× PBS with alternating pulses of 5s duration ( with 5s pause between pulses ) for 45 s at 4°C . The homogenate was clarified by centrifugation at 13 , 000 ×g for 30 min at 4°C , after which the supernatant was stored at −20°C . Protein concentration of fluke homogenates was determined by the Bradford assay . Homogenates were subjected to SDS-PAGE ( 15% ) electrophoresis , and the resolved proteins transblotted to nitrocellulose membrane using a Mini Trans-Blot Cell ( Bio-Rad ) . Membrane strips containing 2 μg of total protein were washed with 0 . 5% Tween-20 in 1× PBS ( PBST ) , blocked with 5% skimmed milk in PBST for 60 min and probed with rabbit anti-Ov-GRN-1 serum or pre-immunization serum , diluted 1:50 with 1% skimmed milk in PBST , for 2 hr with gentle agitation . After washing , the strips were probed with horseradish peroxidase ( HRP ) -goat anti-rabbit IgG ( Invitrogen ) , diluted 1:1000 in antibody buffer , for 60 min . The strips were washed , signals detected using enhanced chemiluminescence ( ECL ) substrate ( GE Healthcare Life Sciences ) and imaged using an Image Quant LAS 4000 mini ( GE Healthcare Life Sciences ) . As a control protein also derived from the tegument of O . viverrini flukes , we also assessed the protein expression levels of Ov-TSP-2 by western blot using a specific antibody raised to the recombinant protein ( Chaiyadet et al . , 2017 ) . Relative protein expression levels as established by western blot were measured by densitometry using Image J , https://imagej . nih . gov/ij/download . html . Levels of protein expressed between groups were compared by independent Student’s t-tests . Adult flukes were collected on days 1 , 2 , 3 , 5 , 7 , 14 and 21 after pCas-Ov-GRN-1 transfection . The genome of each fluke was investigated for mutation ( s ) expected to have resulted from the repair by NHEJ events following the sgDNA programmed double stranded break ( DSB ) of the Ov-GRN-1 locus by Cas9 . For analysis of gDNA from individual adult liver flukes recovered from infected hamsters , we performed a qPCR assay to detect and quantify the frequencies of newly induced mutations . The approach employed two pairs of primers for the target locus , with one putative amplicon extending beyond the putative INDEL site and the other overlapping it , as described ( Yu et al . , 2014 ) . The primers were named Ov-GRN-1-OUT-F , Ov-GRN-1-OVR-F , and Ov-GRN-1-reverse ( OUT/OVR-R ) , respectively . The primer pair of Ov-GRN-1-OUT-F ( 5'-TTCGAGATTCGGTCAGCCG ) and OUT/OVR-R ( 5'-TTGGTCGGCCAGTATGTTCG ) amplified the fragment flanking and spanning the DSB ( 1 , 496–2 , 312 nt ) , whereas the primer pair Ov-GRN-1-OVR-F ( 5'-CAAGTGTTGACGGTGATTTCACTT ) and OUT/OVR-R amplified a region overlapping the DSB ( 1599–2312 ) ( Figure 1B ) . Whereas both primer pairs exhibited equivalent amplification efficiencies with the genomic DNA template from WT flukes , the Ov-GRN-1-OVR-F and OUT/OVR-R primer pair was mutation sensitive , whereas the other pair was not . The OUT and OVR amplicons were 817 and 714 bp in size , respectively , using the following PCR conditions: 7 . 5 μl of SYBR Green Master Mix ( TaKaRa Perfect Real-time Kit ) , 0 . 5 μl ( 0 . 4 μM ) of each primer , 10 ng/μl of gDNA and distilled water to 15 μl . The thermal cycles included initiation for one cycle at 95°C , 3 min followed by 40 cycles of denaturation at 95°C , 30s , annealing at 55°C , 30s , extension at 72°C , 45s , and a final extension at 72°C for 10 min . The SYBR green signal was read at each annealing cycle and reported as threshold cycle ( Ct ) . Efficiency of programmed CRISPR/Cas editing was estimated as the ratio of CtOUT:CtOVR from the experimental group compared with CtOUT:CtOVR of the control group , as described ( Yu et al . , 2014 ) . The CtOUT:CtOVR ratio from the control group would equal ‘1’ ( CRISPR efficiency = 0 ) since there was difference in Ct values from the OUT and OVR primers . By contrast , the OVR primer can be anticipated to be inefficient when compared to the OUT primer for the experimental group , and hence the CtOUT:CtOVR likely would be <1 . Here , we calculated percent mutation indirectly by subtraction of the CRISPR/Cas9 efficiency value from ‘1’ , as indicated ( Yu et al . , 2014; Sentmanat et al . , 2018 ) . Efficiency ( F ) =AveragectOUTAveragectOVRCRISPR/Cas9efficiency=FΔOv−gm−1Fcontrol× ( 100 ) Mutationrate=100%−CRISPR/Cas9efficiency Genomic DNAs from flukes recovered from hamsters 60 days after infection with CRISPR/Cas9-treated NEJ , and which had been assigned to the low ( L ) , moderate ( M ) or high ( H ) groups based on knockdown levels of Ov-GRN-1 transcripts , were pooled by group . The L , M and H groups were assessed and scored for efficiency of CRISPR/Cas9-programmed gene editing in terms of mutation levels by qPCR , as described above . Several Illumina NGS libraries were constructed . First , for analysis of programmed editing of adult flukes that were subjected to gene editing manipulation and subsequently cultured in vitro , genomic DNAs were extracted from the Ov-GRN-1 gene-edited adult liver flukes at each of 7 , 14 and 21 days after transfection . A pool of gDNA was prepared from 15 of these flukes , from five worms from each time point . Second , gDNAs were pooled from 7 to 10 worms from each of the L , M , and H groups of ΔOv-GRN-1 worms ( 25 worms in total ) ( Figure 5A and B ) and also a gDNA pool from 25 control non-gene-edited WT worms . Targeted amplicon NGS libraries were constructed from each of these of gDNA pools . In each case , an amplicon of 173 bp in size that spanned the DSB was amplified using Ov-GRN-1 MiSeq-F primer 5'-TTCGAGATTCGGTCAGCCG ( position 1496–1514 nt ) and Ov-GRN-1 MiSeq-R primer 5'-GCACCAACTCGCAACTTACA ( position 1649–1668 nt ) ( Figure 1B ) . These amplicons were purified ( Agencourt AMPure XP beads , Beckman ) and ligated with Gene Read Adaptors Set A ( Qiagen ) and Illumina compatible adaptor ( s ) and barcode ( s ) using QIAseq 1-step Amplicon library kit ( Qiagen ) . The libraries were quantified using the GeneRead Library Quant Kit ( Qiagen ) with Illumina index/barcode specific primers , and concentration of the libraries established using standard libraries provided in the kit . Illumina NGS was performed by GENEWIZ ( South Plainland , NJ ) . Index/adaptor and primer out sequences were trimmed from the reads . Analysis of the sequenced reads using the SnapGene ( GSL Biotech LLC ) and the CRISPResso software ( https://github . com/lucapinello/CRISPResso ) suites was carried out to validate and characterize programmed mutations of the alleles , including assessment of NHEJ-induced INDELS as insertions , deletions and/or substitutions ( Canver et al . , 2018; Pinello et al . , 2016 ) . The sequences of the alleles were compared to the reference sequence represented by the target amplicon of the WT Ov-GRN-1 gene ( GenBank FJ436341 . 1 ) and to the reads from the control worms for the flukes derived from infection of hamsters with gene-edited NEJ . Of these two analysis methods for performance of CRISPR/Cas9 gene-editing , the qPCR approach ( Yu et al . , 2014 ) is quick and inexpensive in comparison to the targeted amplicon NGS approach ( Canver et al . , 2018; Shalem et al . , 2015 ) . However , the latter approach provides more detailed characterization of the events including the types and frequencies of the INDELS , and is more accurate ( Sentmanat et al . , 2018 ) . To evaluate the effect of Ov-GRN-1 gene editing on liver fluke-driven proliferation of human cholangiocytes , motile WT or ΔOv-GRN-1 adult flukes were co-cultured with cells of the human cholangiocyte cell line H69 in 24-well Trans-well plates ( three wells per group ) ( Papatpremsiri et al . , 2015 ) containing a 4 μm pore size membrane separating the upper and lower chambers ( Corning ) . In brief , 15 , 000 H69 cells were seeded into the lower chamber of the plate and cultured with complete medium containing DMEM/F12 supplemented with 1 × antibiotic , 10% fetal bovine serum , 25 μg/ml adenine , 5 μg/ml insulin , 1 μg/ml epinephrine , 8 . 3 μg/ml holo-transferrin , 0 . 62 μg/ml hydrocortisone , 1 . 36 μg/ml T3 , and 10 ng/ml epidermal growth factor ( Ninlawan et al . , 2010 ) for 24 hr , after which the cells were fasted for 4–6 hr in medium supplemented with only one twentieth of the growth factor content of complete medium . Five viable O . viverrini adult flukes that had been transfected ( or not ) with pCas-Ov-GRN-1 pDNA in a total of 500 μl of RPMI ( or medium alone ) were placed into the upper chamber of each well . The number of cells in each well was determined at days 1 , 2 , and 3 using 1 × PrestoBlue cell viability reagent ( Invitrogen ) ( Tynan et al . , 2012 ) added to cells at 37°C in 5% CO2 atmosphere for up to 60 min . Cell number was determined at 570 nm and calculated from a standard curve before transforming into relative growth compared to control groups . Cell proliferation assays were carried out in triplicate . To assess the effect of Ov-GRN-1 knockout on in vitro wound healing , 300 , 000 cholangiocytes in monolayers were grown in 6-well Trans-well plates with a 4 μm pore size . These cells were cultured in complete media for 2 days at 37°C in 5% CO2 atmosphere then transferred to incomplete media overnight . Monolayers in each well were scratched using a sterile 200 μl autopipette tip ( Papatpremsiri et al . , 2015; Smout et al . , 2015; Liang et al . , 2007 ) and washed with 1× PBS twice to remove disconnected cells or debris . Ten transfected adult or control flukes were added to the upper chamber of the Transwell plate containing the wounded cell monolayer in the lower chamber . The rate of wound closure was measured at 0 , 12 , 24 and 36 hr , respectively . Transwell plates were imaged using an inverted microscope ( Nikon ) and images of all groups were captured at all-time points quantitatively using Adobe Photoshop CS6 . The distances between different sides of the cell monolayer scratch were measured by drawing a line in the middle of the scratch on the captured image ( Papatpremsiri et al . , 2015; Smout et al . , 2015; Liang et al . , 2007; Smout et al . , 2011 ) . The analysis of monolayer wound healing was repeated three times . H69 cells ( Smout et al . , 2015; Grubman et al . , 1994 ) were authenticated using STR profiling by PCR by ATCC and were confirmed in our laboratory to be Mycoplasma-free using the Lookout Mycoplasma PCR detection kit ( Sigma-Aldrich ) . Thirty male Syrian golden hamsters , 6–8 weeks of age , were obtained from the Animal Unit , Faculty of Medicine , Khon Kaen University ( approval number ACUC-KKU-61/60 ) . The hamsters were randomly divided into three groups of 10 animals per group: uninfected control , infected with WT flukes , and infected with ΔOv-GRN-1 flukes . Each hamster was infected with 100 active NEJs through intragastric intubation; the uninfected control group was fed normal saline solution instead of NEJ ( Sripa and Kaewkes , 2002 ) . Hamsters ( five animals per cage ) were maintained under conventional conditions and fed a stock diet ( C . P . Ltd . , Thailand ) and water ad libitum until they were euthanized ( Sripa and Kaewkes , 2002 ) . Following euthanasia , five hamsters from each group were necropsied for histopathological assessment of the hepatobiliary tract at days 14 and 60 post-infection ( Sripa and Kaewkes , 2002 ) . The hamsters were euthanized by overdose of anesthesia with diethyl ether . Subsequently , blood was obtained by cardiac puncture and the livers were removed . Fluke numbers were counted from two livers of both WT and ΔOv-GRN-1 groups at day 60 post-infection and compared using an unpaired two-tailed t-test . The left and right lobes of the liver from five hamsters were dissected , cross-sectioned , and each lobe was divided into three parts . The liver fragments were fixed in 10% buffered formalin and stored overnight at 4°C before processing . Formalin-fixed liver was dehydrated through an ethanol series ( 70 , 95 , and 100% ) , cleared in xylene , and embedded in paraffin . Paraffin embedded sections of 4 µm thickness , cut by microtome , were stained with hematoxylin and eosin ( H&E ) or Picro-Sirius Red , or probed with anti-ACTA2 antibodies , and analyzed for pathologic changes ( below ) . H&E staining was used to assess pathological changes . The sections were deparaffinized in 100% xylene , rehydrated through a descending series of alcohol , stained with H&E for 5 min , dehydrated in an ascending series of alcohol , cleared with 100% xylene , mounted in Permount medium on a glass slide , and slides were dried overnight at 37°C and photographed under light microscopy . Images ( 200× ) from H&E-stained sections from five hamsters infected with WT flukes , five hamsters infected with ΔOv-GRN-1 flukes , and five uninfected hamsters were assessed . Thickness ( width ) of the bile duct epithelium from each thin liver section was measured with ImageJ at eight equidistant positions around the bile duct . To compensate for outliers , the median width for each bile duct was used for the analysis . The two-way ANOVA Holm-Sidak multiple comparisons test was used to compare groups at each time point . Two stains were used separately to assess biliary fibrosis . First , sections were stained with Picro-Sirius Red ( Abcam , Cambridge Science Park , UK ) . Sufficient Picro-Sirius Red solution was applied to completely cover the tissue sections on the slide , the stained slide was incubated at ambient temperature for 60 min , rinsed in two changes of acetic acid solution and dehydrated through two changes of absolute ethanol . Slides were cleared with 100% xylene , mounted in Per-mount , dried overnight at 37°C and photographed by light microscopy to document collagen surrounding the bile ducts . ImageJ was used to auto-color balance the images using the macro by Vytas Bindokas at https://digital . bsd . uchicago . edu/docs/imagej_macros/_graybalancetoROI . txt followed by application of the MRI fibrosis tool to quantify percentage area of fibrosis ( red-stain ) at default settings ( red 1: 0 . 148 , green 1: 0 . 772 , blue 1: 0 . 618 , red 2: 0 . 462 , green 2: 0 . 602 , blue 2: 0 . 651 , red 3: 0 . 187 , green 3: 0 . 523 , blue 3: 0 . 831 ) ( Pereira , 2016 ) . Twenty discrete images ( 200× ) stained with Picro-Sirius Red from each hamster ( five animals per treatment group ) were assessed , that is 100 images per group . Given a broad range of values among groups , comparison of the groups was undertaken using the Kruskal-Wallis with Dunn’s multiple comparisons test . Expression level of smooth muscle alpha-actin ( α-SMA; ACTA2 ) also was assessed as a surrogate for fibrosis . Liver sections from hamsters at 60 days post-infection were deparaffinized 3 times with 100% xylene , 5 min each . Sections were rehydrated with an ascending series of ethanol; 100% , three times , 3 min each , 95% , three times , 3 min each , 70% for 3 min , followed by thorough washing in tap water for 5 min , distilled water for 5 min , and 1× PBS for 5 min . Thereafter , slides were incubated in citrate buffer 0 . 1 M , pH 6 . 0 ( citric acid , anhydrous , 0 . 06 M , sodium citrate dihydrate , 0 . 04 M ) at 110°C for 5 min , allowed to cool at room temperature for 20 min , and then washed in 1× PBS , three times , 5 min each . Thereafter , the sections were blocked with 5% bovine serum albumin ( BSA ) for 30 min in a humidified chamber and washed in three times in 1× PBS , 3 min each with occasional shaking . The slides were probed with Alexa Fluor 594-labeled anti-ACTA2 antibody ( Abcam ) diluted 1:200 in 1% BSA in PBST , 18 hr at 4°C in a humidified atmosphere . Lastly , slides were washed as above , mounted in glycerol , diluted 1:4 with 1 × PBS , and examined under bright and fluorescent lights ( Zeiss Axio Observer; AxioVision SE64 Rel . 4 . 9 . 1 software , Jena , Germany ) . Images with a bile duct containing a fluke were selected and ImageJ used to define regions adjacent to the epithelium that excluded potential blood vessels ( ovate structures ) . Three sites , free of bile ducts and blood vessels , were selected at random in order to establish levels of background fluorescence , each comprising 5–10% of the image . The fluorescence intensity of the biliary epithelium was measured and blanked against the mean of the three background readings and reported as mean intensity per cm2 at 300 pixels per inch ( PPI ) . Twenty-five to 30 discrete images of bile ducts per treatment group ( three hamsters ) were assessed . Zero values from the uninfected group were assigned a value of 1 to enable use of a log axis . The groups were compared using one-way ANOVA with Holm-Sidak multiple comparisons test . These biological replicates represented parallel measurements of biologically discrete samples in order to capture any random biological variation . Technical replicates were undertaken as well; these represented repeated measurements of the same sample undertaken as independent measurements of the random noise associated with the investigator , equipment or protocol . Means for experimental groups were compared to control by one or two ways ANOVA and where appropriate , by Student’s t-test ( GraphPad Prism , La Jolla , CA ) . Values for p of ≤0 . 05 were considered to be statistically significant .	In the rural regions alongside the Mekong River in South East Asia , traditional cuisines often use uncooked or under cooked fish , many of which carry a worm known as Opisthorchis viverrini . Once inside the body , this parasite settles in the human liver , causing a tropical disease known as liver fluke infection . Out of the 10 million people affected by O . viverrini , thousands will also develop a type of liver cancer that is triggered by the presence of the worm . In particular , the parasite secretes a protein known as granulin that may encourage certain liver cells to multiply , potentially raising the risk for cancer . A gene editing technique called CRISPR/Cas9 allows scientist to precisely target and then deactivate the genetic information a cell needs to produce a given protein . While the tool has been used in other species before , it was unknown if it could be applied to O . viverrini . Here , Arunsan et al . harnessed CRISPR/Cas9 to deactivate the gene that codes for granulin and create parasites that can only produce very little of the protein . Hamsters infected with the gene-edited worms had fewer symptoms of liver fluke infection compared to those carrying normal O . viverrini . The animals with parasites that cannot produce granulin also had fewer changes to the liver that are associated with cancer . These findings confirm that granulin has a role in promoting liver fluke infection and liver cancer . Alongside this work , Ittiprasert et al . used CRISPR/Cas9 to inactivate a gene in a species of worm that causes a human disease called schistosomiasis . Together , these findings demonstrate for the first time that the gene editing method can be adapted for use in parasitic worms , which are a major public health problem in tropical climates . This tool should help scientists understand how the parasites invade and damage our bodies , and provide new ideas for treatment and disease control .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "microbiology", "and", "infectious", "disease" ]	2019	Programmed knockout mutation of liver fluke granulin attenuates virulence of infection-induced hepatobiliary morbidity
Eukaryotic replication origin licensing , activation and timing are influenced by chromatin but a mechanistic understanding is lacking . Using reconstituted nucleosomal DNA replication assays , we assessed the impact of nucleosomes on replication initiation . To generate distinct nucleosomal landscapes , different chromatin-remodeling enzymes ( CREs ) were used to remodel nucleosomes on origin-DNA templates . Nucleosomal organization influenced two steps of replication initiation: origin licensing and helicase activation . Origin licensing assays showed that local nucleosome positioning enhanced origin specificity and modulated helicase loading by influencing ORC DNA binding . Interestingly , SWI/SNF- and RSC-remodeled nucleosomes were permissive for origin licensing but showed reduced helicase activation . Specific CREs rescued replication of these templates if added prior to helicase activation , indicating a permissive chromatin state must be established during origin licensing to allow efficient origin activation . Our studies show nucleosomes directly modulate origin licensing and activation through distinct mechanisms and provide insights into the regulation of replication initiation by chromatin . The eukaryotic genome is packaged into a condensed form known as chromatin that presents a barrier to DNA-associated processes . Chromatin is primarily composed of nucleosomes , each of which consists of ~147 base pairs of DNA wrapped around a histone octamer . The location and modification state of nucleosomes is dynamic , regulates access to the DNA and partitions the genome into distinct chromatin states ( Clapier and Cairns , 2009 ) . Nucleosome positioning and modifications influence all DNA processes including replication , transcription , repair and recombination . Thus , maintaining appropriate chromatin states across the genome is critical for cellular viability ( Hargreaves and Crabtree , 2011 ) . Although there is a growing wealth of knowledge concerning the impact of nucleosomes on gene expression , significantly less is known about the role of nucleosomes in regulating DNA replication . Proper eukaryotic DNA replication requires the temporal separation of two key events: origin licensing and origin activation ( Li and Araki , 2013; Siddiqui et al . , 2013 ) . During G1 , origin licensing is initiated by origin-recognition complex ( ORC ) binding to replication origin DNA . ORC then recruits Cdc6 and Cdt1 and these proteins load two inactive Mcm2-7 replicative DNA helicases around the origin DNA ( Bell and Labib , 2016 ) . Origin activation is temporally separated from origin licensing and occurs during S phase . S-phase cyclin-dependent kinases and the Dbf4-dependent Cdc7 kinase ( DDK ) drive recruitment of two helicase-activating proteins , Cdc45 and GINS , forming the active replicative helicase , the Cdc45/Mcm2-7/GINS ( CMG ) complex ( Ilves et al . , 2010; Tanaka and Araki , 2013 ) . Recruitment of DNA polymerases and their accessory proteins to the CMG complex forms a bidirectional pair of replisomes . The majority of these events have been reconstituted in vitro using purified proteins and naked DNA ( Yeeles et al . , 2015 , 2017 ) . Chromatin is proposed to influence multiple aspects of replication initiation including origin licensing , origin activation and the time of replication initiation within S phase . Origin DNA is nucleosome-free to allow ORC DNA binding and , once bound , ORC positions origin-proximal nucleosomes ( Berbenetz et al . , 2010; Eaton et al . , 2010 ) . Repositioning of origin-proximal nucleosomes reduces helicase loading and origin function ( Lipford and Bell , 2001; Simpson , 1990 ) , and loaded helicases appear to interact with these nucleosomes ( Belsky et al . , 2015 ) . Local chromatin states have also been implicated in the activation of eukaryotic origins , each of which is predisposed to initiate earlier or later in S phase ( Rhind and Gilbert , 2013 ) . How chromatin modulates these events and whether specific chromatin regulators impact replication initiation events is unclear . Chromatin-remodeling enzymes ( CREs ) play a major role in determining the chromatin landscape across the genome ( Struhl and Segal , 2013 ) . CREs are multi-protein complexes that use the energy of ATP binding and hydrolysis to assemble , move , slide or alter the composition of nucleosomes ( Clapier and Cairns , 2009; Papamichos-Chronakis and Peterson , 2012 ) . Four subfamilies of CREs are conserved from yeast to humans: ISWI , SWI/SNF , INO80 , and CHD . Members of the ISWI and CHD subfamilies typically function in nucleosome assembly , and they can create regularly spaced nucleosomal arrays by ATP-dependent sliding of nucleosomes ( Hamiche et al . , 1999; Längst et al . , 1999 ) . Similarly , members of the SWI/SNF subfamily mobilize nucleosomes in cis , but these enzymes can also evict nucleosomal histones or eject entire nucleosomes ( Clapier et al . , 2016 ) . Consequently , these CREs typically promote enhanced accessibility of nucleosomal DNA . Finally , members of the INO80 subfamily conduct the post-replicative removal of a particular histone within a nucleosome , and sequential replacement with either a canonical or a variant histone , a process termed nucleosome editing ( Mizuguchi et al . , 2004; Papamichos-Chronakis et al . , 2011 ) . Notably , some members of the INO80 subfamily can also catalyze nucleosome sliding ( e . g . yeast INO80-C ) ( Shen et al . , 2003 ) . Although different CREs can exert a differential impact on nucleosomes , the current view is that each of these enzymes use ATP-dependent DNA translocation as a central mechanism for their activities . Various CREs have been implicated in the regulation of DNA replication ( MacAlpine and Almouzni , 2013 ) . For instance , ISW1-containing remodeling complexes interact with replisome proteins ( Poot et al . , 2005 ) and Chd1 negatively regulate replication initiation ( Biswas et al . , 2008 ) . Similarly , SWI/SNF stimulates replication initiation at specific yeast origins ( Flanagan and Peterson , 1999 ) and is associated with a subset of human origins ( Euskirchen et al . , 2011 ) . Although elimination of different CREs influences DNA replication , whether these effects are direct or indirect and the specific events of replication that are impacted remain elusive . CREs impact multiple processes including transcription , histone modification , and nucleosome assembly ( Clapier and Cairns , 2009 ) leaving open the possibility of indirect effects . In addition , cells express multiple members of each CRE family and overlapping functions of these enzymes could mask the effects of single CRE deletions ( Tsukiyama et al . , 1999 ) . Although the simultaneous deletion of multiple CREs could overcome this issue , in many cases these are lethal events ( Monahan et al . , 2008; Tsukiyama et al . , 1999 ) . Here we describe origin-dependent in vitro replication assays using nucleosomal DNA templates . To address how different nucleosomal states impact DNA replication , we investigated nucleosomal templates that were remodeled by different CREs . Consistent with in vivo studies , these templates showed distinct replication capacities . Most of the nucleosomal DNA templates permitted origin licensing , but ISW2- and Chd1-remodeled templates reduced the efficiency of this event by positioning nucleosomes over the origin DNA , decreasing ORC DNA binding and helicase loading . Although permissive for origin licensing , SWI/SNF- and RSC-remodeled templates showed reduced CMG formation and origin activation . Addition of specific CREs improved replication initiation from these templates but only if the CRE was added prior to CMG formation . Our findings show that local nucleosome status differentially modulates two steps during replication initiation and that specific CREs establish permissive and restrictive states for replication initiation . To investigate the impact of chromatin on replication initiation , we first reconstituted origin licensing using nucleosomal DNA templates . To this end , we used purified ISW1a , Nap1 and budding yeast histone octamers ( Figure 1—figure supplement 1A and B ) to assemble nucleosomes on a 3 . 8 kb linear fragment of Saccharomyces cerevisiae DNA that included the ARS1 replication origin ( Mizuguchi et al . , 2012 ) . We optimized the ratio of DNA to histone octamers to assemble regularly-spaced nucleosome arrays ( Figure 1A and Figure 1—figure supplement 2A ) . After nucleosomes were remodeled , ISW1a , Nap1 and free histones were removed from the template ( Figure 1—figure supplement 2B ) to provide a defined nucleosomal DNA state by preventing additional nucleosome assembly and remodeling . 10 . 7554/eLife . 22512 . 003Figure 1 . Mcm2-7 helicase loading onto nucleosomal DNA templates . ( A ) Nucleosomes were remodeled with bead-coupled ARS1-containing linear DNA , ISW1a , yeast histone octamers and Nap1 . Nucleosome assembly was assessed after partial MNase digestion . ( B ) Outline of the helicase-loading assay using nucleosomal DNA . ( C ) . Comparison of helicase loading on naked DNA and on ISW1a-remodeled nucleosomal DNA . DNA templates were washed with high-salt ( H ) or low-salt ( L ) buffer after loading . Template-associated Mcm2-7 , ORC and H2B was detected by immunoblot . ( D ) Helicase loading onto either wild-type ( WT ) or A-B2- ( mut ) ( Heller et al . , 2011 ) ARS1-containing DNA . As indicated , nucleosomal DNA was remodeled with ISW1a . Assays were performed in either 125 mM ( to allow increased origin non-specific helicase loading ) or 300 mM ( origin specific helicase loading ) potassium glutamate . After a high salt wash , DNA-associated Mcm2-7 was detected by immunoblot . DOI: http://dx . doi . org/10 . 7554/eLife . 22512 . 00310 . 7554/eLife . 22512 . 004Figure 1—figure supplement 1 . Purified proteins used in the in vitro nucleosome . assembly reactions . ( A ) Purified yeast histone octamers and Nap1 were separated by . SDS-AGE and visualized by Coomassie staining . ( B ) Purified CREs were separated . by SDS-PAGE and visualized by Coomassie staining . DOI: http://dx . doi . org/10 . 7554/eLife . 22512 . 00410 . 7554/eLife . 22512 . 005Figure 1—figure supplement 2 . Preparation of in vitro nucleosome templates . ( A ) Nucleosomes were assembled with increasing amounts of histone octamers and fixed amount of DNA with ISW1a . Nucleosome assembly was as in Figure 1A . ( B ) Removal of CRE and unassociated proteins from in- vitro assembled nucleosomes . Amount of CRE and Nap1 associated with nucleosomal DNA before and after washing was detected by anti-CBP ( Ioc3-TAP , Ioc2-TAP , Ino80-TAP , Chd1-TAP , Swi2-TAP and Rsc2-TAP ) , anti-FLAG ( Isw2-FLAG ) , anti-H2B and anti-6xHis ( Nap1 ) immunoblots . DOI: http://dx . doi . org/10 . 7554/eLife . 22512 . 00510 . 7554/eLife . 22512 . 006Figure 1—figure supplement 3 . The ATPase activities of in vitro purified chromatin remodeling enzymes . The ATPase activities of chromatin remodeling enzymes were measured in the presence of 0 . 1 mg/ml plasmid DNA . The fractions of hydrolyzed ATP were normalized with 1 nM remodeling enzymes . DOI: http://dx . doi . org/10 . 7554/eLife . 22512 . 006 Using purified ORC , Cdc6 , Cdt1 and Mcm2-7 ( Kang et al . , 2014 ) , we compared the ability of nucleosomal and naked DNA templates to participate in origin licensing as measured by loading of the Mcm2-7 helicase ( Figure 1B ) . At the end of the reaction , DNA-beads were washed with a low- ( L ) or high-salt ( H ) containing buffer . The low-salt wash retains all DNA-associated proteins whereas the high-salt wash releases ORC , Cdc6 , Cdt1 and incompletely-loaded Mcm2-7 but retains loaded Mcm2-7 complexes associated with successful origin licensing ( Donovan et al . , 1997; Randell et al . , 2006 ) . The amount of ORC DNA binding , helicase association ( low-salt wash [L] ) and helicase loading ( high-salt wash [H] ) were comparable between nucleosomal and naked DNA templates ( Figure 1C ) . Thus , ISW1a-remodeled nucleosomes are permissive for origin licensing . To address the effect of nucleosomes on origin selection , wild-type ( WT ) and mutant ARS1-containing DNA was assembled into nucleosomes and helicase loading was performed under lower-salt conditions that allow Mcm2-7 loading at non-origin sequences ( compare upper and lower panels of ( Figure 1D ) . Under these conditions , nucleosome assembly reduced non-specific Mcm2-7 loading onto mutant ARS1-containing DNA without altering helicase loading onto WT DNA ( Figure 1D , top panel ) . Thus , nucleosomes reduced origin licensing at non-origin DNA sequences , consistent with previous in vivo studies implicating local nucleosomes in origin selection ( Berbenetz et al . , 2010; Eaton et al . , 2010 ) . To address how different local nucleosome landscapes influence replication initiation , we generated ARS1 origin DNA templates with distinct nucleosome patterns . To this end , we assembled nucleosomes onto origin DNA in the presence of seven different purified CREs: ISW1a , ISW1b , ISW2 , INO80-C , Chd1 , SWI/SNF and RSC ( Figure 1—figure supplement 1B ) . The amount of CRE added was normalized according to their relative ATPase activity ( Figure 1—figure supplement 3 ) , ( Smith and Peterson , 2005 ) . After nucleosome assembly , the CRE , Nap1 and non-nucleosomal histones were removed ( Figure 1—figure supplement 2B ) to ensure that the nucleosomes deposited during assembly are not remodeled or moved during subsequent replication-initiation assays . First , we examined nucleosome assembly by partial MNase-digestion . ISW1a , ISW1b , INO80-C , ISW2 and Chd1 each resulted in regularly-spaced nucleosomes on the origin DNA , albeit with different spacings ( Figure 2A ) . In contrast , SWI/SNF- and RSC-remodeled nucleosomes did not show evidence of uniformly-spaced nucleosomes , consistent with previous observations ( Flaus and Owen-Hughes , 2003; Kassabov et al . , 2003 ) . It was possible that SWI/SNF and RSC treatment reduced or eliminated nucleosome assembly . To test this hypothesis , we compared the amount of DNA-associated H2B and H3 ( Figure 2B and Figure 2—figure supplement 1A ) and the amount of mono-nucleosomal DNA produced after extensive MNase treatment ( Figure 2—figure supplement 1B ) . These studies showed that the presence of different CREs did not dramatically change the extent of nucleosome formation . For simplicity , we refer to the different nucleosomal DNA templates by the CRE present during their assembly ( e . g . SWI/SNF template ) . 10 . 7554/eLife . 22512 . 007Figure 2 . Comparison of helicase loading onto nucleosomal DNA templates remodeled with different CREs . ( A ) Comparison of nucleosome assembly with different CREs . Nucleosomes were remodeled with the indicated CRE and assayed by partial MNase digestion . ( B ) Helicase loading onto nucleosomes remodeled with different CREs . After helicase loading , DNA was washed either with high-salt ( H ) or low-salt ( L ) buffer . Mcm2-7 and H2B DNA association was detected by immunoblot . ( C ) Comparison of origin-proximal nucleosome positioning established by different CREs . The positions of nucleosome dyads remodeled with the indicated CRE were analyzed by high-throughput MNase-Seq . Nucleosome dyad density ( Y-axis ) and the corresponding position of the dyad ( X-axis ) are plotted . Zero on the X-axis indicates the first nucleotide of the ARS1 consensus sequence ( ACS ) . The elements of ARS1 ( Marahrens and Stillman , 1992 ) are indicated above . ( D ) ORC association with nucleosomal DNA remodeled with different CREs . Template association of ORC was detected by immunoblot . ( E ) Addition of ORC during nucleosome assembly restores helicase loading on ISW2 and Chd1 templates . Nucleosomes were assembled onto ARS1 DNA with the indicated CRE in the presence or absence of ORC . Helicase loading was performed and analyzed as described in ( B ) . DOI: http://dx . doi . org/10 . 7554/eLife . 22512 . 00710 . 7554/eLife . 22512 . 008Figure 2—figure supplement 1 . Nucleosome assembly with different CREs and their ability to load Mcm2-7 helicase . ( A ) Histone H2B and H3 associated with nucleosomes assembled with the indicated CREs was detected by anti-H2B and H3 immunoblot . ( B ) Mono-nucleosomes produced from nucleosomal templates assembled with different CREs Similar amounts of ISW1a- , INO80-C , Chd1- , RSC- and SWI/SNF assembled templates were digested extensively with MNase and purified mononucleosomal DNA was analyzed by agarose gel electrophoresis . ( C ) Quantification of relative Mcm2-7 loading for nucleosomal templates assembled with the indicated CRE . The amount of Mcm2-7 quantified using online software ImajeJ and statistical analysis was performed using Prism software . For each assay three ( n = 3 ) biological replicates were quantified . Mean value for ISW1a ( High salt wash ) reactions was calculated and set as the 100% . All the other values were calculated as a percentage of that mean value . Error bars indicate standard deviation ( SD ) . DOI: http://dx . doi . org/10 . 7554/eLife . 22512 . 00810 . 7554/eLife . 22512 . 009Figure 2—figure supplement 1—source data 1 . Raw values used in the quantification of Figure 2B , left panel ( n = 3 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 22512 . 00910 . 7554/eLife . 22512 . 010Figure 2—figure supplement 1—source data 2 . Raw values used in the quantification of Figure 2B , right panel ( n = 3 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 22512 . 01010 . 7554/eLife . 22512 . 011Figure 2—figure supplement 2 . ORC1 BAH domain and Abf1 is dispensable for helicase loading of nucleosomal templates . ( A ) Purified ORC and ORCΔBAH were separated by SDS-PAGE and visualized by Coomassie staining . ( B ) Comparison of WT . and ΔBAH ORC mediated helicase loading on ISW1a-remodeled DNA templates . DNA templates were washed with high-salt ( H ) or low-salt ( L ) buffer after loading . Template associated Mcm2-7 and ORC was detected by immunoblot . ( C ) Helicase loading onto ISW1a , ISW2 and INO80-C templates were carried out with WT or ΔBAHORC . DNA templates were washed with high-salt ( H ) after loading . Template-associated Mcm2-7 and H2B was detected by immunoblot . ( D ) Purified Abf1 was separated by SDS-PAGE and visualized by Coomassie staining . ( E ) Comparison of Mcm2-7 loading for naked DNA and ISW1aremodeled nucleosomal templates . After helicase loading , DNA was washed with high salt . Mcm2-7 and H2B DNA association was detected by immunoblot . ( F ) Addition of Abf1 during nucleosome assembly do not restore helicase loading defects of ISW2 templates . Nucleosomes were remodeled onto ARS1 DNA with the indicated CRE in the presence/absence of ORC or Abf1 and ability load helicase was determined . Template associated Mcm2-7 , Abf1 ( anti-Flag ) and H2B was detected by immunoblot after high salt wash . DOI: http://dx . doi . org/10 . 7554/eLife . 22512 . 011 We examined each of the different nucleosomal templates for origin licensing . SWI/SNF and RSC templates showed levels of Mcm2-7 loading similar to ISW1a templates ( Figure 2B , Figure 2—figure supplement 1C and Figure 2—figure supplement 1—source data 1 and 2 ) . ISW1b and INO80-C templates showed modest reductions in loaded Mcm2-7 and Chd1 and ISW2 templates showed progressively less loading . Thus , the CRE present during nucleosomal assembly impacted the extent of origin licensing . To investigate the cause of the differential origin licensing , we determined the position of origin-proximal nucleosomes for the ISW1a , ISW1b , INO80-C , Chd1 and ISW2 templates using MNase-seq ( Cole et al . , 2012; Eaton et al . , 2010 ) . The ISW1a template showed a nucleosome-free region ( NFR ) overlapping ARS1 with well-defined flanking nucleosomes ( Figure 2C ) . In contrast , ISW2 template showed the appearance of a positioned nucleosome overlapping the origin ( centered at −54 bp relative to ACS , Figure 2C ) . In addition , the flanking nucleosome on the opposite side of the origin was shifted towards the ACS ( from +222 to +168 ) in the ISW2 templates . These data support a model in which encroachment of origin-proximal nucleosomes onto origin DNA directly inhibits origin licensing . To determine whether the reduced origin licensing of the ISW2 and Chd1 templates was caused by decreased ORC DNA binding , we examined ORC association with these nucleosomal templates ( Figure 2D ) . The extent of ORC binding to the ISW2 and Chd1 templates correlated with the amount of Mcm2-7 loading ( Figure 2B and D ) . We asked if addition of ORC during nucleosome-assembly reactions restored Mcm2-7 loading . Importantly , when ORC bound DNA prior to Chd1- or ISW2-directed nucleosome assembly , loaded Mcm2-7 levels were restored to levels similar to ISW1a templates ( Figure 2E ) . Together , these data indicate that nucleosome positioning over the origin reduces origin licensing by inhibiting ORC DNA binding and that ORC is not sufficient to move nucleosomes in the absence of a CRE . To further investigate the role of ORC in the establishment of Mcm2-7-loading-competent chromatin states , we evaluated the role of the Orc1 bromo-adjacent homology ( BAH ) domain . BAH domains bind to nucleosomes ( Yang and Xu , 2013 ) and elimination of the Orc1 BAH domain reduces initiation from a subset of replication origins in yeast ( Müller et al . , 2010 ) . We purified ORC lacking the Orc1 BAH domain ( ORC∆BAH , Figure 2—figure supplement 2A ) and performed helicase-loading assays using ISW1a , ISW2 and INO80-C templates ( Figure 2—figure supplement 2B–C ) . Consistent with the limited effect of deletion of the Orc1 BAH domain on ARS1 function in vivo ( Müller et al . , 2010 ) , ORC and ORC∆BAH showed comparable levels of helicase loading onto all the nucleosomal templates . We also examined whether the presence of the ARS1-binding protein , Abf1 , influenced helicase loading in the presence of nucleosomes . Previous studies showed that Abf1 and ORC position nucleosomes on either side of ARS1 ( Lipford and Bell , 2001 ) and that elimination of the Abf1 binding sites reduced ARS1 function ( Marahrens and Stillman , 1992 ) . Addition of purified Abf1 ( Figure 2—figure supplement 2D ) to either naked DNA or ISW1a templates did not improve helicase loading ( Figure 2—figure supplement 2E ) . We also asked whether addition of Abf1 to the ISW2 nucleosome assembly would rescue the helicase-loading defects of ISW2 templates , as we observed for ORC ( Figure 2E ) . In contrast to ORC , Abf1 did not improve helicase loading on the ISW2 template ( Figure 2—figure supplement 2F ) . Next , we examined the effect of nucleosomes on replication-initiation events after origin licensing had occurred . To this end , we performed replication assays ( Gros et al . , 2014; Heller et al . , 2011; On et al . , 2014 ) by sequentially adding DDK and an S-phase extract to helicases loaded onto DNA templates with or without nucleosomes ( Figure 3A ) . ISW1a templates showed comparable levels of replication products to that of naked DNA ( Figure 3B ) . Nucleotide incorporation was Cdc6- ( Figure 3B ) , DDK- ( Figure 3C , and Figure 3—source data 1 ) and origin-sequence-dependent ( Figure 3—figure supplement 1 ) indicating that the DNA synthesis observed was due to replication initiation and elongation ( rather than DNA repair ) . 10 . 7554/eLife . 22512 . 012Figure 3 . Replication initiation on nucleosome templates . ( A ) Outline of nucleosomal DNA replication initiation assay using purified proteins and yeast S-phase cell extract . ( B ) ISW1a templates do not interfere with replication . Naked DNA or ISW1a templates were assayed in the presence or absence of Cdc6 . Radiolabeled replication products were analyzed by alkaline agarose electrophoresis and autoradiography ( top ) . Template-associated H2B was detected by immunoblot ( lower ) . ( C ) Comparison of replication using ISW1a , ISW1b , ISW2 , INO80-C and Chd1 templates in the presence and absence of DDK . Products of the extract-based replication assays were analyzed as in ( B , top ) . H2B levels for each template are shown ( middle ) . Quantification of replication products was performed as in Figure 2B . Error bars show the SD ( n = 3 , lower ) . ( D ) Comparison of replication of ISW1a , SWI/SNF and RSC templates in the presence or absence of DDK . Analysis of replication products , template-associated H2B and quantification ( n = 3 ) as in ( C ) . DOI: http://dx . doi . org/10 . 7554/eLife . 22512 . 01210 . 7554/eLife . 22512 . 013Figure 3—source data 1 . Raw values used in the quantification of Figure 3C ( n = 3 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 22512 . 01310 . 7554/eLife . 22512 . 014Figure 3—source data 2 . Raw values used in the quantification of Figure 3D ( n = 3 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 22512 . 01410 . 7554/eLife . 22512 . 015Figure 3—figure supplement 1 . In vitro nucleosomal DNA template replication initiation is origin specific . WT and mutant ARS ( A-B2- ) containing ISW1a templates were assayed in the presence or absence of Cdc6 . Radiolabeled replication products were analyzed by alkaline agarose electrophoresis and autoradiography ( top ) . Template-associated Mcm2-7 was detected by immunoblot ( lower ) . Products of the extract-based replication assay were analyzed as in Figure 3B . DOI: http://dx . doi . org/10 . 7554/eLife . 22512 . 015 To compare replication initiation from nucleosome templates remodeled with different CREs , we carried out the same replication assay with each template . For ISW1a , ISW1b , ISW2 , INO80-C and Chd1 , the level of replication products closely matched the amount of helicase loading with the same templates ( compare Figure 3C and Figure 2—figure supplement 1C ) . Thus , origin activation and replisome assembly were not further reduced by these nucleosomal templates . In contrast , the RSC and SWI/SNF templates showed a disconnect between the extent of origin licensing and the levels of replication initiation . SWI/SNF , RSC and ISW1a templates showed comparable levels of Mcm2-7 loading ( Figure 2B and Figure 2—figure supplement 1C ) , but the amount of replication products from SWI/SNF and RSC templates was reduced ~5 fold relative to ISW1a templates ( Figure 3D and Figure 3—source data 2 ) . Thus , nucleosomal DNA templates remodeled by SWI/SNF and RSC inhibit one or more events downstream of origin licensing . The presence of multiple CREs in the S-phase extract led us to adapt a fully-reconstituted replication-initiation assay ( Lõoke et al . , 2017; Yeeles et al . , 2015 ) to investigate the cause of the reduced replication of the SWI/SNF and RSC templates ( Figure 4A ) . Compared to the S-phase-extract-based assay , ISW1a templates showed reduced replication using the fully-reconstituted assay ( Figure 4—figure supplement 1A ) , most likely due to a lack of CREs and histone chaperones present in the S-phase-extract-based assay ( Devbhandari et al . , 2017; Kurat et al . , 2017 ) . Nevertheless , replication of the SWI/SNF and RSC templates was similarly reduced relative to their ISW1a-remodeled counterpart using the reconstituted assay ( Figure 4B , Figure 4—figure supplement 1B and Figure 4—figure supplement 1—source data 1 ) . Importantly , the reduced replication observed for the RSC or SWI/SNF templates was not simply because of a lack of uniformly-spaced nucleosomes . When we assembled nucleosomes in the absence of any CRE , the resulting nucleosomes were similarly non-uniformly spaced ( Figure 4—figure supplement 2A ) but the levels of replication from these templates were comparable to ISW1a templates ( Figure 4—figure supplement 2B ) . Thus , the reduced replication capacity of the RSC and SWI/SNF templates requires the activity of the corresponding CRE . 10 . 7554/eLife . 22512 . 016Figure 4 . SWI/SNF and RSC templates show reduced CMG formation . ( A ) Outline of fully-reconstituted nucleosomal DNA replication initiation assay . The proteins added at each step are indicated . ( B ) Comparison of reconstituted nucleosomal DNA replication using ISW1a , SWI/SNF and RSC templates in the presence or absence of DDK . Analysis of replication products and H2B as in Figure 3B . ( C ) Comparison of Mcm2-7 phosphorylation by DDK on ISW1a , SWI/SNF and RSC templates . Phosphorylation of Mcm6 is indicated by reduced electrophoretic mobility and was analyzed by immunoblot ( top ) . Template associated H2B is shown ( lower ) . ( D ) Comparison of replication of ISW1a , RSC and SWI/SNF templates . Reactions were performed with or without DDK and replication products of the reconstituted replication reactions were analyzed as in Figure 3B ( top ) . Template association of Mcm2-7 , Cdc45 , GINS and H2B was measured after a high-salt wash at the end of reconstituted replication assay by immunoblot ( lower panels ) . ( E ) Comparison of CMG formation and activation using ISW1a , SWI/SNF and RSC templates . To prevent replication initiation , the only nucleotide present was ATP and Pol α was left out of the assay . Template association of Mcm2-7 , Cdc45 , GINS , Rfa1 and H2B were measured by immunoblot . DOI: http://dx . doi . org/10 . 7554/eLife . 22512 . 01610 . 7554/eLife . 22512 . 017Figure 4—figure supplement 1 . Reconstituted replication assay . ( A ) Comparison of naked and nucleosomal DNA templates in the reconstituted replication assay . Comparison of replication of naked DNA and ISW1a templates with or without DDK using a fully reconstituted replication assay . Replication products ( top panel ) and Mcm2-7 , Cdc45 , GINS and H2B template association ( lower panels ) were assayed as in Figure 4D . ( B ) Comparison of nucleosomal DNA assembled with different CRE in the reconstituted replication assay . Quantification of the ISW1a , SWI/SNF and RSC templates replication products in the presence of DDK using reconstituted replication assay . Reconstituted replication assays performed as described in Figure 4D . Quantified as in Figure 3B and the material and method section . Error bars indicate standard deviation of three biological replicates ( n = 3 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 22512 . 01710 . 7554/eLife . 22512 . 018Figure 4—figure supplement 1—source data 1 . Raw values used in the quantification of Figure 4B ( n = 3 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 22512 . 01810 . 7554/eLife . 22512 . 019Figure 4—figure supplement 2 . Nucleosomal template assembled without CRE are able to replicate . ( A ) Nucleosomes were assembled with either ISW1a , SWI/SNF , RSC . or no CRE , with bead-coupled ARS1-containing linear DNA , yeast histone octamers and Nap1 . Nucleosome assembly was assessed after partial MNase digestion as in Figure 2A . ( B ) Comparison of ISW1a , SWI/SNF , RSC or no CRE nucleosomal DNA templates in the reconstituted replication assay with or without DDK . H2B template association ( lower panels ) were assayed as in Figure 4D . DOI: http://dx . doi . org/10 . 7554/eLife . 22512 . 019 To identify the replication event ( s ) that was reduced by SWI/SNF- and RSC-remodeled nucleosomes , we monitored different events of origin activation . First , we examined DDK phosphorylation of Mcm2-7 ( detected by retardation of Mcm6 electrophoresis , Francis et al . , 2009 ) . This modification was either unchanged ( SWI/SNF ) or improved ( RSC ) relative to ISW1a templates ( Figure 4C ) , indicating Mcm2-7 phosphorylation by DDK was not reduced . Next , we assessed CMG formation by examining Cdc45 and GINS template association after replication initiation and elongation ( Figure 4D ) . Both SWI/SNF and RSC templates showed reduced Cdc45 and GINS template association compared to ISW1a templates . For these initial experiments , we measured template association at the end of the replication reaction . Thus , the decreases in Cdc45 and GINS template association could be due to inefficient CMG formation during initiation or increased CMG dissociation during elongation . To distinguish between these possibilities , we repeated the replication-initiation assays in the presence of ATP but without other rNTPs or dNTPs ( Figure 4E ) . Under these conditions , the CMG can form and partially unwind DNA but replication cannot initiate ( Yeeles et al . , 2015 ) . As in the previous assays , we observed reduced Cdc45 and GINS association with SWI/SNF and RSC templates compared to the ISW1a templates . Consistent with reduced active helicases and DNA unwinding , the amount of Rfa1 ( a subunit of the eukaryotic single-stranded DNA binding protein RPA ) association with RSC and SWI/SNF templates was also reduced ( Figure 4E ) . Thus , the observed reduction in DNA replication products observed for the SWI/SNF and RSC templates in the complete assays was due to reduced CMG formation and helicase activation . Our previous replication assays were performed in the absence of CREs to address how different chromatin states impact replication initiation . In vivo , however , these enzymes could be present at origin-proximal chromatin during initiation . To address whether the continuous presence of a CRE during replication initiation altered our findings , we asked if the addition of ISW1a , RSC or SWI/SNF during replication-initiation assays improved replication initiation from the RSC and SWI/SNF templates . Adding ISW1a during the helicase-loading step ( DL ) of the assay ( Figure 5A ) restored CMG formation and increased DNA replication of the RSC templates ( Figure 5B , compare lanes 4 and 5 ) . In contrast , adding RSC during the helicase-loading step did not alter either the amount of replication or CMG formation ( Figure 5B , lane 6 ) . Similar experiments with SWI/SNF templates showed that the addition of ISW1a ( but not SWI/SNF ) improved replication of SWI/SNF templates ( Figure 5—figure supplement 1A ) . In contrast to the ability of ISW1a to improve replication from the RSC and SWI/SNF templates , addition of RSC or SWI/SNF to ISW1a templates during helicase loading did not reduce replication levels ( Figure 5—figure supplement 1B ) . Interestingly , consistent with its ability to reduce helicase loading , addition of Chd1 to ISW1a during helicase loading did reduce replication . Together these data indicate that the defects that we observe in CMG formation and replication for the RSC and SWI/SNF templates are not simply due to the lack of a CRE during the replication assay . Instead , our findings suggest that specific CREs create nucleosomal states that facilitate CMG formation and replication initiation . 10 . 7554/eLife . 22512 . 020Figure 5 . Rescue of SWI/SNF and RSC template replication initiation . ( A ) Schematic of ISW1a addition at various steps during the replication assay . ( B ) Addition of ISW1a at the helicase-loading step rescues replication initiation from RSC templates . Reconstituted replication assays were performed on ISW1a and RSC templates with or without DDK . ISW1a or RSC was added to the templates during helicase loading and not deliberately removed ( DL ) or upon addition of the helicase activation and elongation proteins ( I/E ) as indicated . The lane that show I/E is from the same gel as the rest of the panel . Replication products ( top panel ) and Mcm2-7 , Cdc45 , GINS and H2B template association ( lower panels ) were assayed as in Figure 4D . ( C ) Specific CREs improve RSC-template replication . Reconstituted replication assays were performed with RSC templates with or without DDK . ISW1a , ISW1b , INO80-C or ISW2 was added to RSC templates after helicase loading ( AL ) . Replication products ( top ) and Mcm2-7 , Cdc45 , GINS and H2B template association ( lower panels ) were assayed as in Figure 4D . ( D ) Addition of ISW1a after nucleosome assembly facilitates replication and CMG formation of SWI/SNF templates . Reconstituted replication assays were performed with ISW1a or SWI/SNF templates with or without DDK . ISW1a was added to the templates either during helicase loading ( DL ) , after helicase loading ( AL ) or upon addition of the helicase activation and elongation proteins ( I/E ) as indicated . The lane that show I/E is from the same gel as the rest of the panel . Replication products ( top ) and Mcm2-7 , Cdc45 , GINS and H2B template association ( lower panels ) were assayed as in Figure 4D . ( E ) ISW1a addition after helicase loading ( AL ) to RSC templates , but removed before helicase activation improves replication of and CMG complex formation on RSC templates . Reconstituted replication reactions were performed with the indicated templates with or without DDK . ISW1a was added to the RSC templates upon completion of helicase loading ( AL ) . Replication products ( top ) and Mcm2-7 , Cdc45 , GINS and H2B template-association ( lower panels ) were assayed as in Figure 4D . DOI: http://dx . doi . org/10 . 7554/eLife . 22512 . 02010 . 7554/eLife . 22512 . 021Figure 5—figure supplement 1 . ISW1a rescues RSC and SWI/SNF templates prior to the initiation step . ( A ) Addition of ISW1a to SWI/SNF templates , at helicase-loading step rescues SWI/SNF initiation defects . Reconstituted replication assays were performed on . ISW1a and SWI/SNF templates with or without DDK . ISW1a or SWI/SNF was added to the SWI/SNF templates during helicase loading ( DL ) . Replication products were assayed as in Figure 3B . ( B ) Addition ISW1 , Chd1 , SWI/SNF or RSC to ISW1a templates , post at helicase-loading step . Reconstituted replication assays were performed on ISW1a templates with or without DDK . Replication products were assayed as in Figure 3B . DOI: http://dx . doi . org/10 . 7554/eLife . 22512 . 02110 . 7554/eLife . 22512 . 022Figure 5—figure supplement 2 . Origin proximal nucleosome positioning is not directly responsible for CMG formation defects in RSC and SWI/SNF templates . Analysis of origin-proximal nucleosome positioning of ISW1a , SWI/SNF and RSC templates . Comparison of origin-proximal nucleosomes positioning by high-throughput MNase-Seq of ISW1a- , SWI/SNF- and RSC-templates similar to Figure 2C . RSC and SWI/SNF templates were remodeled with ISW1a ( ISW1a-treated RSC or SWI/SNF ) after they were assembled for indicated reactions . Zero on the X-axis indicates the first nucleotide of the ARS consensus sequence ( ACS ) . DOI: http://dx . doi . org/10 . 7554/eLife . 22512 . 02210 . 7554/eLife . 22512 . 023Figure 5—figure supplement 3 . Mcm2 histone-binding motif is dispensable for nucleosomal DNA replication . ( A ) Comparison of WT and mutant Mcm2 helicase loading on ISW1a-remodeled DNA templates . DNA templates were washed with high-salt ( H ) or low-salt ( L ) buffer after loading . Template-associated Mcm2-7 and ORC was detected by immunoblot . ( B ) WT and mutant Mcm2 helicase loading on ISW1aremodeled DNA templates were assayed for replication initiation as described for Figure 4B . DOI: http://dx . doi . org/10 . 7554/eLife . 22512 . 02310 . 7554/eLife . 22512 . 024Figure 5—figure supplement 4 . ISW1a rescues RSC and SWI/SNF templates after DDK step . ISW1a addition to RSC and SWI/SNF templates after DDK phosphorylation step , partially restores RSC and SWI/SNF templates replication and CMG formation defects . Reconstituted replication assays were performed with the indicated templates in the presence and absence of DDK . ISW1a was added to the SWI/SNF and RSC templates upon completion of DDK-phosphorylation step and washed off before adding initiation factors . Replication products ( top panel ) and Mcm2-7 , Cdc45 , GINS and H2B template-association ( lower panels ) were assayed as in Figure 4D . DOI: http://dx . doi . org/10 . 7554/eLife . 22512 . 024 Because we observed a connection between origin-proximal nucleosome positioning and origin-licensing capacity ( Figure 2C ) , we asked if the reduced origin activation of SWI/SNF and RSC templates corresponded to a particular positioning of origin-proximal nucleosomes . Analysis of local nucleosome positioning by MNase-seq did not reveal a nucleosomal pattern that distinguished the RSC and SWI/SNF templates from ISW1a templates in origin proximal region ( Figure 5—figure supplement 2 ) . Consistent with the robust helicase loading observed for all three templates , they each exhibited a nucleosome-free region over the origin . The nucleosome pattern near ARS1 was similar between the ISW1a and SWI/SNF templates . In addition , treatment of SWI/SNF templates with ISW1a did not cause major changes in nucleosome positioning . The pattern of RSC-remodeled nucleosomes was substantially different from SWI/SNF and ISW1a templates on the right side of ARS1 . ISW1a addition to RSC templates enhanced the positioning of one nucleosome centered at ~400 bp on the right side of the ARS1 ACS . Thus , unlike the situation for origin licensing ( Figure 2 ) , there was no apparent correlation between flanking nucleosome positions and the reduced origin activation observed for SWI/SNF and RSC templates . Previous studies have reported a histone-binding motif in Mcm2 ( Foltman et al . , 2013; Huang et al . , 2015 ) , raising the possibility that nucleosome-Mcm2-7 interactions may facilitate replication initiation in a nucleosomal context . To test this possibility , we purified a mutant version of Mcm2-7 that lacks the Mcm2 histone-binding motif . Incorporation of this mutation did not alter helicase loading or DNA synthesis with or without nucleosomes ( Figure 5—figure supplement 3A–B ) , suggesting that this interaction is not critical for replication initiation under the conditions of these assays . This is consistent with the lack of an obvious replication phenotype for this mutation ( Foltman et al . , 2013; Huang et al . , 2015 ) . Given the redundant functions of chromatin remodelers in vivo , we asked if the ability to improve the replication of RSC and SWI/SNF templates was unique to ISW1a or if other CREs could perform the same function . As discussed above , addition of RSC or SWI/SNF to the corresponding nucleosomal templates during helicase loading did not improve CMG formation or replication ( Figure 5B and Figure 5—figure supplement 1 ) . Similarly , ISW2 addition to RSC templates resulted in only limited rescue of both CMG formation and replication ( Figure 5C ) . In contrast , addition of ISW1b or INO80-C to RSC templates after helicase loading improved replication initiation and CMG formation to similar levels as ISW1a addition ( Figure 5C ) . Thus , the ability to restore full replication competence to the RSC or SWI/SNF templates is limited to a subset of CREs , consistent with previous studies indicating that these complexes have distinct functionalities ( Clapier and Cairns , 2009 ) . We also asked when ISW1a needed to be present during a specific replication event to improve replication of SWI/SNF and RSC templates ( Figure 5A ) . When added only during the helicase-loading step ( DL ) or after helicase-loading but removed before DDK treatment ( AL ) , ISW1a significantly improved replication of the SWI/SNF ( Figure 5D ) and RSC templates ( Figure 5B and E ) . Addition of ISW1a to the SWI/SNF and RSC templates after the DDK-step ( but before CMG formation and replication initiation ) also improved replication and CMG formation ( Figure 5—figure supplement 4 ) . In contrast , addition of ISW1a only during the initiation/elongation ( I/E ) step of the replication reaction did not improve CMG formation or replication of RSC ( Figure 5B ) or SWI/SNF templates ( Figure 5D ) . Thus , ISW1a can only improve the replication competence of RSC and SWI/SNF templates if it acts prior to the events of origin activation . Previous studies have shown that replication origins are included within nucleosome-free regions ( NFRs ) and this characteristic is important for origin activity ( Berbenetz et al . , 2010; Eaton et al . , 2010; Lipford and Bell , 2001; MacAlpine et al . , 2010; Simpson , 1990; Xu et al . , 2012 ) . Consistent with nucleosomes impacting origin selection , we found that assembly of DNA into nucleosomes reduced origin-independent initiation ( Figure 1D ) . Given the redundancy of CREs in vivo , which CREs are capable of establishing NFRs at replication origins is unknown . Our findings demonstrate that CRE-dependent differences in local nucleosomes impact origin licensing . Only a subset of CREs positioned nucleosomes in a manner that allowed efficient origin licensing ( Figures 2B , 6A and B ) . ISW2 templates showed the most inefficient Mcm2-7 loading compared to other templates and this reduction correlated with the encroachment of origin-proximal nucleosomes over origin DNA in a manner that inhibited ORC DNA binding ( Figures 2C and 6C ) . This finding is consistent with studies showing that ISW2 slides nucleosomes towards the promoter-proximal NFR to suppress transcription at cryptic transcription-start sites ( Whitehouse et al . , 2007 ) . Our findings suggest that ISW2 and perhaps Chd1 play a similar role in regulating origin usage . Interestingly , once ORC is bound to origin DNA , ISW2 is unable to displace ORC with a nucleosome ( Figure 2D ) . Similarly , once ISW2 establishes nucleosome positioning at the origin , ORC is unable to bind ( Figure 2E ) , suggesting ORC cannot displace interfering nucleosomes . These findings suggest that both the relative timing of ORC binding and histone deposition and the CRE present at this time will influence the use of a given site as an origin . 10 . 7554/eLife . 22512 . 025Figure 6 . Nucleosomes remodeled by different CREs influence replication initiation differently . Nucleosomes affect multiple steps of replication initiation using distinct mechanisms . Schematic of ATP-dependent nucleosome assembly with different CREs and their affect on replication initiation . Opacity of the nucleosome represents nucleosome density at each location . ( A ) Replication permissive nucleosomes are remodeled by ISW1a , ISW1b and INO80-C . These templates are competent for both origin licensing and origin activation . Nucleosome positioning is comparable in these templates . ( B ) SWI/SNF and RSC templates are origin-licensing competent but are inefficient for subsequent origin activation . We propose that the SWI/SNF and RSC templates have alternate/destabilized nucleosome structures indicated by their different color and that these nucleosomes are not conducive to origin activation . Although both reduce origin activation , SWI/SNF and RSC templates do not share similar nucleosome positioning . ( C ) ISW2 ( and Chd1 ) templates have nucleosomes over the replication origin that reduce ORC DNA binding and , therefore , origin licensing . DOI: http://dx . doi . org/10 . 7554/eLife . 22512 . 025 Neither the BAH domain of Orc1 nor the ARS1-binding protein Abf1 contributed to helicase loading in our experiments . The lack of a role for the ORC BAH domain is expected given the modest effect of deletion of the BAH domain on ARS1 replication initiation in vivo ( Müller et al . , 2010 ) . Given the observation that other BAH domains recognize specific modified forms of nucleosomes ( Yang and Xu , 2013 ) , it is also possible that we did not observe a role for the Orc1 BAH domain due to the unmodified status of the histones used in these experiments . Abf1 binding positions nucleosomes on one side of ARS1 in vivo , however , ORC is able to perform this function in the absence of Abf1 at many origins ( Eaton et al . , 2010 ) . One notable difference from the in vivo situation compared to our in vitro studies is that in vivo the TRP1 gene transcribes into ARS1 . Thus , it is possible that Abf1 binding is important to position nucleosomes in the presence of this invasive transcription but not in the absence ( such as in our experiments ) . Our findings indicate that local nucleosomes also impact efficient CMG formation and , therefore , origin activation . In particular , SWI/SNF and RSC templates reduced this event . Our finding that the reduced replication of these templates is rescued by addition of other CREs ( Figure 5 ) makes it clear that these effects are nucleosome-dependent . It remains to be determined how the RSC and SWI/SNF templates modulate this event . Given that these templates allow efficient helicase loading , simple steric inhibition by encroaching nucleosomes is unlikely to explain these effects . This conclusion is reinforced by the presence of a large nucleosome-free region overlapping the origin for both templates ( Figure 5—figure supplement 2 ) . Indeed , despite their different capacities for replication initiation and CMG formation , ISW1a and SWI/SNF templates had a similar pattern of surrounding nucleosomes and the SWI/SNF nucleosomes pattern was not changed by the same ISW1a addition that restored robust CMG formation and replication initiation to these templates ( Figure 5D and Figure 5—figure supplement 2 ) . Another possibility is that the lack of uniformly-spaced nucleosome arrays in the RSC and SWI/SNF templates reduces CMG formation . Of the seven CREs we tested , only SWI/SNF and RSC did not establish uniformly-spaced nucleosomes ( Figure 2A ) . On the other hand , we observed a similar lack of uniformly positioned nucleosomes when we did not add any CRE to the nucleosome assembly reactions , and these templates showed much higher levels of replication initiation than the RSC and SWI/SNF templates ( Figure 4—figure supplement 2 ) . Although it is possible that very subtle changes in nucleosome positioning are responsible , a more likely explanation is that the structure of the nucleosomes remodeled by RSC and SWI/SNF is different ( Figure 6A and B ) . Previous studies have suggested that SWI/SNF family remodelers establish nucleosome structures that are different from canonical nucleosomes ( Lorch et al . , 1998; Schnitzler et al . , 1998; Ulyanova and Schnitzler , 2005 ) . Such altered nucleosomes could have distinct abilities to interact with the replication machinery . SWI/SNF and RSC are also known to remove H2A/H2B dimers from nucleosomes ( Clapier et al . , 2016 ) . Although we do not observe a dramatic reduction in the relative amounts of H3 and H2A for these templates ( Figure 2—figure supplement 1A ) it is possible that a subset of nucleosomes assembled in the presence of RSC or SWI/SNF are lacking the full complement of H2A/H2B . One argument against this possibility is the ability of ISW1a addition to readily restore full replication initiation and CMG formation to these templates ( Figure 5 ) . Given that free histones are removed after initial nucleosome assembly , it is not clear how ISW1a addition could restore full nucleosomes to the RSC or SWI/SNF templates . One interesting possibility to explain the different capacities of the templates to facilitate CMG formation is raised by recent studies suggesting that loaded Mcm2-7 interacts with adjacent nucleosomes ( Belsky et al . , 2015 ) . It is possible that different positions/conformations of local nucleosomes ( see below ) impacts the ability of nucleosomes to interact with Mcm2-7 double hexamers . These interactions could directly or indirectly modulate access of helicase-activating proteins to loaded Mcm2-7 double hexamers . Addition of some CREs to RSC or SWI/SNF templates improved their replication but only if added before the helicase-activation proteins ( Figure 5 ) , suggesting that a positive interaction between loaded Mcm2-7 and nucleosomes must be established prior to CMG formation . Interestingly , we found that RSC and SWI/SNF could not reduce replication from an ISW1a template when added during helicase loading . Perhaps a subset of CREs produce nucleosomes that can interact with the replication machinery positively and once these interactions are established they prevent other CREs from inducing alternative conformations . Although previous studies have identified a histone-binding motif in Mcm2 ( Foltman et al . , 2013; Huang et al . , 2015 ) , incorporation of this mutation into the Mcm2-7 complex did not alter helicase loading or DNA synthesis with or without nucleosomes ( Figure 5—figure supplement 3A–B ) . It is possible , however , that this mutation is not sufficient to eliminate Mcm2-7 interactions with adjacent nucleosomes . Our studies show that the different CREs are not equivalent in their ability to establish replication-competent nucleosomes . These differences were observed both with regard to the initial deposition of nucleosomes on DNA ( e . g . RSC templates inhibiting CMG formation , Figure 5 ) and when CREs were added after deposition ( e . g . ISW1a addition rescuing the reduced CMG formation of RSC templates , Figure 5 ) . The specificity of the different CREs in our assays suggests that the presence of different CREs at origins will impact origin usage . Localization of specific CREs to origin DNA through interactions with the replication machinery ( Euskirchen et al . , 2011; Papamichos-Chronakis and Peterson , 2008 ) or adjacent promoters/transcriptional machinery ( Yen et al . , 2012 ) could impact either origin licensing or activation . Our studies indicate that presence of a specific CRE at an origin during a particular cell cycle stage would have different consequences for DNA replication . A CRE present during S phase would impact the initial assembly of nucleosomes and more likely impact subsequent origin licensing/helicase loading . A CRE that is present during G1 is less likely to impact origin licensing and more likely to modulate subsequent CMG formation . Thus , simple deletion of a CRE or monitoring of CRE association with origin DNA in an asynchronous cell culture is unlikely to reveal their full impact on the events of DNA replication . Interestingly , the RSC and SWI/SNF templates show hallmarks of late-initiating origins . Like late-initiating origins ( Belsky et al . , 2015; Santocanale et al . , 1999; Wyrick et al . , 2001 ) , these templates showed efficient origin licensing but reduced/delayed replication initiation . In addition , replication timing is established in late M or early G1 phase ( Dimitrova and Gilbert , 1999; Raghuraman et al . , 1997 ) and replication timing can only be reprogrammed prior to S phase ( Peace et al . , 2016 ) . Similarly , SWI/SNF and RSC templates can be remodeled to replicate efficiently if certain CREs are added prior to shifting the templates into helicase-activation conditions ( Figure 5B ) , which is the biochemical equivalent of the G1-S phase transition . These similarities suggest that local nucleosome states influence replication timing . Although our studies investigated DNA replication in the context of S . cerevisiae DNA-sequence-defined origins of replication , they are relevant to DNA replication in all eukaryotic organisms . Although most organisms do not use sequence-defined origins of replication , origin-proximal nucleosome-free regions are a common characteristic of origins in many organisms ( Fragkos et al . , 2015 ) . Thus , our findings regarding the impact of local nucleosomes on origin licensing and selection are relevant to these origins as well . Indeed , the absence of specific sequences directing initiation of replication suggests that local chromatin states will have an even more important role in most organisms . Importantly , once helicases are loaded , specific origin sequences have little or no impact on subsequent origin activation ( Gros et al . , 2014 , 2015 ) . The assays described here lay the groundwork for future studies of the impact of nucleosome structure and histone modification on DNA replication , and can be extended to query DNA replication-dependent nucleosome assembly events and epigenetic inheritance mechanisms . Yeast strains used in these studies are derivatives of W303 and are described in Supplementary file 1 . Epitope tagging was performed by PCR-based homologous recombination as previously described ( Longtine et al . , 1998 ) . Plasmids used in this study are described in Supplementary file 2 and were created by conventional molecular-cloning methods . DNA templates were isolated from the pARS1-Nco-Nco plasmid ( Kang et al . , 2014 ) . The plasmid was digested with BamHI , filled in with biotinylated-dATP , dGTP , dCTP and dTTP using Klenow enzyme ( NEB ) to facilitate attachment to beads . After spin column purification ( Plasmid Miniprep Kit from Qiagen ) , the biotin labeled DNA was cut with Nsi I and Sac II followed by a second spin column purification . This creates a large 3 . 8 kb BamHI to Nsi I DNA fragment that is biotinylated at one end and that is entirely derived from native S . cerevisiae sequences surrounding the ARS1 origin of replication . A small biotinylated DNA ( released by Sac II ) is removed by the spin column and the remaining bacterial/vector DNA is not biotinylated and will not bind to streptavidin beads . The 3 . 8 kb biotinylated-DNA was immobilized on streptavidin-coated paramagnetic beads ( Dynabeads M-280 , ThermoFisher ) according to manufacturer instructions and the non-biotinylated DNA fragment was washed away . Yeast histones and hNap1 were purified using previously established methods ( Vary et al . , 2004 ) . Mcm2-7/Cdt1 , ORC , Cdc6 and DDK were purified as previously described ( Heller et al . , 2011; Kang et al . , 2014 ) . S-CDK , Sld3/Sld7 , Sld2 , Cdc45 , Dpb11 , GINS , Pol ε , Pol α/primase , RPA and Mcm10 were purified as described previously ( Lõoke et al . , 2017 ) . Nucleosomes were assembled as previously described ( Mizuguchi et al . , 2012 ) . Nucleosome formation was optimized using a 3 . 8 kb S . cerevisiae DNA fragment ( see below ) , S . cerevisiae histone octamers ( using an DNA:octamer ratio of 1:1 . 3 by mass ) and varying human Nap1 ( hNap1 ) concentration . After determining an optimal histone octamer:hNap1 ratio , nucleosome assembly was further optimized by varying ISW1a concentration . Finally , after optimizing the Nap1 and ISW1a concentrations , the nucleosome assembly reaction was optimized for the DNA:octamer ratio . Consequently , nucleosomes were assembled with ~137 nM yeast histone octamers , 267 nM hNap1 , 10 nM CRE ( ISW1a or ISW1b or ISW2 or INO80-C or Chd1 or SWI/SNF or RSC ) and 120 fmol Dyna bead-immobilized 3 . 8 Kbp ARS1 DNA in a 20 µl reaction . Initially , hNap1 , histone octamers were incubated in ExB 5/50 buffer ( 10 mM HEPES-KOH [pH7 . 6] , 0 . 5 mM EGTA , 5 mM magnesium chloride ( MgCl2 ) , 50 mM potassium chloride ( KCl ) , 10% glycerol , 0 . 1 mg/ml BSA ) in for for 45 min followed by CRE addition and continued incubation for another 15 min . , bringing the total reaction volume to 12 . 5 µl . 7 . 5 µl of ATP regeneration mix ( 5 mM ATP , 30 mM creatine phosphate and 1 μg/ml creatine kinase in 1x ExB 5/50 buffer ) was added to the histone octamers , hNap1 and CRE reaction mix and immediately added to the bead-immobilized DNA and incubated at 30°C at 1400 rpm for 4 . 5 hr in a Thermomixer ( Eppendorf , Hauppauge , NY ) . Nucleosomal-DNA beads were stored at 4°C and used for assays within 12–24 hr . Nucleosome assembly was analyzed by digesting 120 fmol nucleosomal DNA with limiting ( 0 . 04 U ) MNase at 25°C at 1300 rpm for 15 min in a Thermomixer . The resulting DNA fragments were purified using spin columns ( EZ Nucleosomal DNA prep Kit from Zymo Research ) and separated on a 1 . 5% agarose gel and stained with ethidium bromide . Purification protocols for chromatin-remodeling enzymes , histones and hNap1 are described in Supplementary file 2 . Helicase loading was performed as previously described ( Kang et al . , 2014 ) for naked DNA templates with the following modifications . Mcm2-7/Cdt1 , ORC , and Cdc6 were purified as previously described ( Kang et al . , 2014 ) . The bead-coupled nucleosomal DNA was magnetically separated from unassociated or loosely bound proteins and the supernatant was removed . Nucleosomal DNA was washed twice with 20 µl buffer A-0 . 35 ( 25 mM HEPES-KOH [pH7 . 6] , 0 . 5 mM EGTA , 0 . 1 mM EDTA , 5 mM MgCl2 , 10% glycerol , 0 . 02% NP40 , 0 . 1 mg/ml BSA and 0 . 35 M KCl ) and once with 20 µl buffer A-0 . 3 KGlut ( 0 . 3 M potassium glutamate [KGlut] instead of 0 . 35 M KCl in buffer A-0 . 35 ) . Helicase loading was initiated by the addition of 120 fmol ORC , 180 fmol Cdc6 , and 360 fmol Mcm2–7/Cdt1 in a 20 μl reaction containing 60 fmol of bead-coupled 3 . 8 kb ARS1 DNA ( with or without nucleosomes ) in helicase-loading buffer ( 25 mM HEPES-KOH [pH 7 . 6] , 12 . 5 mM magnesium acetate ( MgAc ) , 300 mM KGlut , 20 μM creatine phosphate , 0 . 02% NP40 , 10% glycerol , 3 mM ATP , 1 mM dithiothreitol ( DTT ) , and 2 μg creatine kinase ) . The reaction were briefly vortexed or mixed by pipetting ( if necessary ) to remove any bead clumping . The reactions were incubated at 25°C at 1250 rpm for 25 min in a Thermomixer . Beads were washed three times with 150 µl Buffer H ( 25 mM HEPES-KOH [pH 7 . 6] , 1 mM EDTA , 1 mM EGTA , 5 mM MgAc , 10% glycerol , and 0 . 02% NP40 ) containing 0 . 3 M KGlut for low salt wash experiments . Experiments with high-salt washes substituted buffer H with 0 . 5 M NaCl for the second of the three wash steps . DNA-bound proteins were eluted from the beads using 2x sample buffer ( 120 mM Tris [pH 6 . 8] , 4% SDS and 20% glycerol ) . Eluted proteins were separated by SDS-PAGE and analyzed by immunoblotting . Replication products were measured by incorporation of 32P-dCTP into newly synthesized DNA . Incorporated 32P-dCTP was detected after denaturing gel electrophoresis using a phosphor-imager . For relative replication product quantification , nucleosomal templates assembled with the indicated CRE were quantified with ImageJ software . For each assay , three ( n = 3 ) biological replicates were quantified . The mean value for ISW1a ( reactions with DDK ) was calculated and set as 100% . All the other values were calculated as a percentage of the mean value of the ISW1a experiment ( always performed as part of the same experiment and separated on the same gel ) . Statistical analysis was performed using Prism software . Error bars indicate standard deviation ( SD ) . Quantification of relative Mcm2-7 loading ( immunoblots for Mcm2-7 ) for nucleosomal templates assembled with the indicated CRE was determined with ImageJ software . For each assay three ( n = 3 ) biological replicates were quantified . The mean value for ISW1a ( reaction with high-salt wash ) was calculated and set as 100% . All the other values were calculated as a percentage of the mean value of the ISW1a experiment ( always performed as part of the same experiment and separated on the same gel ) . Statistical analysis was performed using Prism software . Error bars indicate standard deviation ( SD ) . Nucleosomes were remodeled on 120 fmol of DNA as previously described ( nucleosome assembly and analysis section ) with ISW1a , SWI/SNF or RSC . Nucleosomal DNA was digested with 1 U of MNase for 30 min shaking at 25°C at 1350 rpm in Thermomixer . The released DNA was purified using spin columns ( EZ Nucleosomal DNA prep Kit from Zymo Research ) . The purified DNA was separated on a 1 . 5% agarose gel electrophoresis and stained with ethidium bromide . Nucleosomal DNA was washed with buffer A-0 . 35 ( 2X ) and digested with 1 U of MNase in 20 µl MNase-digestion buffer ( 12 . 5 mM HEPES-KOH [pH7 . 6] , 0 . 5 mM EDTA , 5 mM MgCl2 , 2 mM calcium chloride , 5% glycerol and 1 mM DTT ) for 30 min at 25°C shaking at 1350 rpm in a Thermomixer . DNA was purified using spin columns . The purified DNA was separated on a 1 . 5% agarose gel electrophoresis and stained with ethidium bromide . DNA fragments in the size range that includes mono-nucleosomes ( <160 bp ) were extracted from the gel and purified using spin columns ( Freeze 'N Squeeze DNA Gel Extraction Spin Columns , Bio-Rad ) . DNA samples were end-repaired and adaptor-ligated using the SPRI-works Fragment Library System I ( Beckman Coulter Genomics ) and indexed during amplification . Libraries were quantified using the Fragment Analyzer ( Advanced Analytical ) and qPCR before being loaded for paired-end sequencing using an Illumina HiSeq 2000 ( MIT BioMicroCenter ) . Paired-end sequencing reads were aligned to the plasmid sequence using Bowtie 1 . 1 . 2 ( Langmead et al . , 2009 ) with the following parameters: -n 2 -l 20 - -best - -strata . To obtain nucleosome occupancy profiles , the midpoint positions from all 125–175 bp read fragments were extracted . The nucleosome signal was smoothed by constructing a 20 bp Gaussian kernel around each midpoint position , and smoothed kernels were aggregated together to form a nucleosome signal track ( Boyle et al . , 2008 ) . Signal was then normalized to read depth for each sample . Consensus nucleosome positions were determined by finding peaks ( above a threshold of 2 ) in the nucleosome-density signal ( Flores and Orozco , 2011 ) .	Each human cell contains more than two meters of DNA . To fit this length into a cell , remodeling enzymes compact the DNA by helping it to bind to specific proteins . This compaction has the side effect of making the DNA harder to access . DNA replication is one process that requires access to the DNA . Replication occurs each time a cell divides , so that each newly formed cell receives a full set of genetic material . DNA replication starts simultaneously at hundreds of sites across the DNA . At each of these sites , cells assemble a protein called a replicative helicase . Helicases play a important role in many steps of DNA replication , but their most fundamental role is to separate the two DNA strands that make up the double helix; these strands then act as templates during replication . A helicase is initially inactive when loaded at a replication start site . Additional proteins then bind to the helicase to activate it . Studies have shown that DNA compaction influences DNA replication , but it was not known exactly how compacted DNA affects helicase loading and activation . To investigate the effects of compacted DNA during replication in more detail , Azmi et al . created different types of compacted DNA molecules using various remodeling enzymes . Some of the compacted DNAs directly prevented the binding of a protein that is required to load the helicase to the replication start site . In addition , the compaction reduced the number of sites on the DNA where replication could begin . Other types of compacted DNA allowed the helicase to be loaded normally , but inhibited the subsequent activation of the helicase . However , treating these DNA types with particular remodeling enzymes restored helicase activation to normal levels . Overall , the findings presented by Azmi et al . suggest that cells can control helicase loading and activation independently by compacting DNA in different ways . Such control is important to ensure that each time a cell divides , it fully replicates its entire DNA .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "chromosomes", "and", "gene", "expression", "biochemistry", "and", "chemical", "biology" ]	2017	Nucleosomes influence multiple steps during replication initiation
The generation of diverse neuronal subtypes involves specification of neural progenitors and , subsequently , postmitotic neuronal differentiation , a relatively poorly understood process . Here , we describe a mechanism whereby the neurotrophic factor NGF and the transcription factor Runx1 coordinate postmitotic differentiation of nonpeptidergic nociceptors , a major nociceptor subtype . We show that the integrity of a Runx1/CBFβ holocomplex is crucial for NGF-dependent nonpeptidergic nociceptor maturation . NGF signals through the ERK/MAPK pathway to promote expression of Cbfb but not Runx1 prior to maturation of nonpeptidergic nociceptors . In contrast , transcriptional initiation of Runx1 in nonpeptidergic nociceptor precursors is dependent on the homeodomain transcription factor Islet1 , which is largely dispensable for Cbfb expression . Thus , an NGF/TrkA-MAPK-CBFβ pathway converges with Islet1-Runx1 signaling to promote Runx1/CBFβ holocomplex formation and nonpeptidergic nociceptor maturation . Convergence of extrinsic and intrinsic signals to control heterodimeric transcription factor complex formation provides a robust mechanism for postmitotic neuronal subtype specification . Neuronal cell fate specification is a multistep process that can be broadly divided into two stages , early specification of neural progenitor cell identity and postmitotic differentiation of neuronal subtypes . A recurring theme in the early phase of neural progenitor specification of the vertebrate nervous system is that the interplay between intrinsic determinants and extrinsic signals governs neural cell fate decisions ( Fishell and Heintz , 2013 ) . Indeed , cell fate choice of progenitor cells is determined by a combination of intrinsic genetic programs , which define the differentiation potential or the competence state of progenitors , and extrinsic cues , which influence the relative proportions of each cell type generated within the confines of a given competent state ( Livesey and Cepko , 2001 ) . Our understanding of the mechanisms of postmitotic differentiation of neuronal subtypes , on the other hand , is rather limited , and thus the relative contributions of extrinsic and intrinsic signals , and their mechanisms of interplay during late phases of cell fate determination , remain poorly understood . Primary sensory neurons of the dorsal root ganglion ( DRG ) are pseudo-unipolar neurons that extend one axonal branch to the spinal cord and the other to peripheral targets such as the skin . The existence of a large number of functionally specialized DRG sensory neuron subtypes , each endowed with unique morphological and physiological properties enables the somatosensory system to encode diverse sensory information , including nociception , temperature , light touch , limb movement , and body position . DRG neurons can be classified into three major classes: nociceptors , which preferentially respond to noxious stimuli , pruritogenic stimuli , and temperature; mechanoreceptors , which respond to innocuous tactile sensations; and proprioceptors , which detect position and movement of the trunk and limbs . Each functional class is further subdivided into subtypes exhibiting distinct physiological , morphological , and molecular properties ( Lallemend and Ernfors , 2012; Usoskin et al . , 2015 ) . Since the same Neurogenin1-dependent progenitor population gives rise to the three principal DRG neuron types within a relatively short time window ( Frank and Sanes , 1991; Ma et al . , 1999 ) , specification of the diverse DRG neuron subtypes mainly takes place in postmitotic neurons . Therefore , DRG sensory neurons are particularly well suited for mechanistic studies of postmitotic differentiation of neuronal subtypes . The differentiation of nociceptor precursors into nonpeptidergic and peptidergic nociceptors is a well-studied example of postmitotic sensory neuron subtype specification . These two nociceptor subtypes are molecularly , morphologically and functionally distinct . Peptidergic nociceptors are traditionally defined by expression of neuropeptides , such as CGRP and Substance P , while the majority of nonpeptidergic nociceptors have binding sites for the lectin IB4 ( Mulderry et al . , 1988; Silverman and Kruger , 1990 ) . More recently , a single cell transcriptome analysis led to the identification of four molecularly defined nonpeptidergic neuron subtypes ( Usoskin et al . , 2015 ) . Peptidergic and nonpeptidergic nociceptor subtypes are further distinguished by their central and peripheral axonal projection patterns ( Molliver et al . , 1995; Zylka et al . , 2005 ) . Functionally , nonpeptidergic and peptidergic nociceptors are preferentially required for mechanical and thermal nociception , respectively ( Cavanaugh et al . , 2009; McCoy et al . , 2013; Mishra and Hoon , 2010; Vulchanova et al . , 2001 ) . Despite these differences , nonpeptidergic and peptidergic nociceptors derive from a common nociceptor precursor that expresses tropomyosin-receptor kinase A ( TrkA ) , the receptor for nerve growth factor ( NGF ) , and both nociceptor subtypes require NGF–TrkA signaling for survival and development beginning at ~E12 . 5 ( Crowley et al . , 1994; Luo et al . , 2007; Patel et al . , 2000; Silos-Santiago et al . , 1995; Smeyne et al . , 1994 ) . A key developmental event that defines the segregation of nociceptor subtypes is the nonpeptidergic nociceptor-specific switch from NGF responsiveness to glial-derived neurotrophic factor ( GDNF ) family ligand ( GFL ) responsiveness . This occurs by virtue of the upregulation of Ret and GDNF family receptor alpha receptors ( GFRα ) , receptor components for GFL signaling , in nonpeptidergic nociceptors starting from E15 . 5 , as well as postnatal extinction of TrkA expression ( Bennett et al . , 1996; Bennett et al . , 1998; Molliver and Snider , 1997; Molliver et al . , 1997 ) . Interestingly , both NGF–TrkA and GFL–GFRα/Ret signaling are essential for postmitotic differentiation of nonpeptidergic nociceptors . Indeed , NGF–TrkA signaling is required for the acquisition of virtually all nonpeptidergic nociceptor-specific features , including expression of Ret and Gfras , whereas GFL–GFRα/Ret signaling , in turn , plays a critical role in postnatal development of nonpeptidergic nociceptors , including expression of a subset of genes characteristic of mature nonpeptidergic nociceptors ( Luo et al . , 2007; Patel et al . , 2000 ) . The mechanism by which NGF–TrkA signaling governs differentiation of nonpeptidergic nociceptors is unclear . In addition to the extrinsic signals NGF and GFLs , intrinsic determinants of nonpeptidergic nociceptor fate have begun to be defined ( Chen et al . , 2006; Lopes et al . , 2012 ) . Most prominent is Runx1 , a Runx family transcription factor whose expression is primarily restricted to nonpeptidergic nociceptors . Runx1 functions as a master regulator of the nonpeptidergic nociceptor lineage ( Chen et al . , 2006; Yoshikawa et al . , 2007 ) . In mammals , Runx family proteins , which include Runx1 , Runx2 and Runx3 , are key regulators of hematopoietic , osteogenic , and immune cell lineages ( Banerjee et al . , 1997; de Bruijn and Speck , 2004; Ducy et al . , 1997 ) . In those systems , Runx proteins function as heterodimers by forming complexes with CBFβ , a non-DNA-binding cofactor that enhances the DNA binding affinity and protein stability of Runx proteins ( Adya et al . , 2000 ) . In the peripheral nervous system , genetic ablation of Runx1 leads to a near complete loss of nonpeptidergic nociceptor-specific features and a concomitant expansion of neurons exhibiting a peptidergic nociceptor-like phenotype ( Chen et al . , 2006; Kramer et al . , 2006; Yoshikawa et al . , 2007 ) . Interestingly , the dramatic consequences of NGF deficiency on nonpeptidergic-specific gene expression resemble those seen upon Runx1 ablation ( Chen et al . , 2006; Luo et al . , 2007 ) . This observation suggests a potential interaction between the extrinsic signal NGF and the intrinsic factor Runx1 during nonpeptidergic nociceptor development and thus an opportunity to define mechanisms of interplay between extrinsic and intrinsic factors during postmitotic neuronal differentiation . Here , we show that Runx1 controls nonpeptidergic nociceptor development primarily by acting downstream of NGF–TrkA signaling . NGF/TrkA regulates Runx1 function at least in part by promoting expression of CBFβ , which we show is an essential component of the Runx1/CBFβ complex during nonpeptidergic nociceptor development . Mechanistically , NGF–TrkA signaling controls Runx1/CBFβ complex formation through the ERK/MAPK signaling pathway , which is both necessary and sufficient for Cbfb expression . On the other hand , Islet1 , a LIM-homeodomain transcription factor , controls transcriptional initiation of Runx1 but not Cbfb . These findings identify a novel NGF/TrkA–MAPK–CBFβ axis critical for the differentiation of nonpeptidergic nociceptors and reveal a mechanism by which convergence of extrinsic and intrinsic signals promotes formation of a heterodimeric transcription factor complex that instructs postmitotic neuronal subtype specification . The differentiation of nonpeptidergic nociceptors requires the target-derived neurotrophic growth factor NGF as well as the transcription factor Runx1 ( Chen et al . , 2006; Luo et al . , 2007 ) . The phenotypic similarity exhibited by mice lacking NGF and those lacking Runx1 raises the intriguing possibility that NGF and Runx1 function in a common signaling pathway to control nonpeptidergic nociceptor maturation . To address this , we assessed the extent to which initiation of the nonpeptidergic phenotype is co-dependent on NGF and Runx1 during early stages of nociceptor development . We first sought to extend our understanding of Runx1-dependent nonpeptidergic nociceptor-specific genes using unbiased gene expression profiling of DRGs from Wnt1-Cre; Runx1f/f ( Runx1 CKO ) mice , previously characterized neural crest-specific Runx1 conditional mutants and their littermate controls at E16 . 5 , the onset of nonpeptidergic-specific gene expression ( Supplementary file 1 ) ( Chen et al . , 2006 ) . Among the many genes with downregulated expression in Runx1CKO DRGs compared to controls , Ptprt , Myo1a , and Kif21b were shown to exhibit strong Runx1 dependence by in situ hybridization analysis ( Figure 1O–T ) . Further co-localization studies with Runx1 confirmed that their expression is primarily restricted to nonpeptidergic nociceptors ( data not shown ) . We next compared the patterns of expression of these as well as additional , canonical nonpeptidergic nociceptor-specific genes between Runx1 CKO mice and mice lacking NGF at the same developmental stages by in situ hybridization . In order to study survival-independent functions of NGF , nociceptors are kept alive in the absence of NGF by co-deletion of Ngf and the proapoptotic gene Bax ( Ngf-/-; Bax-/- mice ) ( Patel et al . , 2000 ) . In accordance with previous observations , expression of the nonpeptidergic nociceptor markers Mrgprd and Gfra2 was almost completely abolished in presumptive nonpeptidergic nociceptors of both Ngf-/-; Bax-/- and Runx1 CKO mice at P0 ( Figure 1A–D , K–N ) ( Chen et al . , 2006; Luo et al . , 2007 ) . Moreover , newly identified nonpeptidergic-specific genes , including Ptprt , Myo1a , and Kif21b , exhibited dramatically reduced expression in neurons that would normally become nonpeptidergic nociceptors in Ngf-/-; Bax-/- mice compared to littermate controls at P0 ( Figure 1E–J ) , in a manner similar to that observed in Runx1 CKO mice ( Figure 1O–T ) . Furthermore , impairment in expression of Mrgprd , Ptprt , Myo1a , and Kif21b was observed in Ngf-/-; Bax-/- and Runx1 CKO mice as early as E16 . 5 , which is when their expression is normally initiated ( Figure 1—figure supplement 1 ) . As expected , expression of Gfra2 , Myo1a , and Kif21b in non-nociceptors was unaffected in Ngf-/-; Bax-/- and Runx1 CKO mice , thus explaining the residual expression observed in these mutants ( Figure 1D , N and Figure 1—figure supplement 1F , H , N , P ) . These results together demonstrate a general requirement of both NGF and Runx1 for initiation of expression of a large subset of nonpeptidergic neuron genes . 10 . 7554/eLife . 10874 . 003Figure 1 . The majority of nonpeptidergic nociceptor-specific genes depend on both NGF and Runx1 for expression . ( A–J ) Expression of Mrgprd ( Control , 12 . 7% ± 2 . 4%; Ngf-/-Bax-/- , 0% ) , Gfra2 ( Control , 28 . 7% ± 2 . 5%; Ngf-/-Bax-/- , 14 . 3% ± 1 . 1% ) , Ptprt ( Control , 19 . 1% ± 0 . 3%; Ngf-/-Bax-/- , 10 . 3% ± 2 . 1% ) , Myo1a ( Control , 26 . 4% ± 0 . 9%; Ngf-/-Bax-/- , 5 . 8% ± 1 . 5% ) and Kif21b ( Control , 27 . 2% ± 2 . 7%; Ngf-/-Bax-/- , 14 . 05% ± 1 . 8% ) in control and Ngf-/-Bax-/- DRGs at P0 , assessed by in situ hybridization . ( K–T ) Expression of Mrgprd ( Control , 26 . 3% ± 1 . 0%; Runx1 CKO , 0 . 3% ± 0 . 2% ) , Gfra2 ( Control , 40 . 0% ± 3 . 7%; Runx1 CKO , 9 . 6% ± 0 . 7% ) , Ptprt ( Control , 36 . 0% ± 3 . 3%; Runx1 CKO , 10 . 4% ± 2 . 3% ) , Myo1a ( Control , 31 . 1% ± 2 . 3%; Runx1 CKO , 5 . 1% ± 1 . 8% ) and Kif21b ( Control , 24 . 8% ± 1 . 7%; Runx1 CKO , 10 . 9% ± 0 . 8% ) in control and Runx1 CKO DRGs at P0 , assessed by in situ hybridization . Note that Ngf-/-Bax-/- and Runx1 CKO mutants display similar deficits in expression of these representative nonpeptidergic-nociceptor specific genes . For Gfra2 , Myo1a , Kif21b , the apparent NGF- and Runx1-independent expression in large diameter neurons can be attributed to their expression in non-nociceptive DRG neurons . Shown are the means ± SEM for the percentage of DRG neurons expressing indicated genes based on counts from a total of at least 9 sections from three independent animals per genotype . DRG neurons were identified and counted based on combined NeuN immunostaining , which was not shown . Ngf +/-; Bax-/- or Ngf +/+ ;Bax-/- and Runx1f/f mice were used as control animals for analysis of Ngf-/-;Bax-/- and Runx1 CKO mutants , respectively . Scale bar , 50 μm . See also Figure 1—figure supplement 1 . DOI:http://dx . doi . org/10 . 7554/eLife . 10874 . 00310 . 7554/eLife . 10874 . 004Figure 1—figure supplement 1 . The majority of nonpeptidergic nociceptor-specific genes depend on both NGF and Runx1 for initiation of expression . ( A–P ) Expression of Mrgprd , Ptprt , Myo1a and Kif21b in control and Ngf-/-Bax-/- DRGs ( A-H ) , or in control and Runx1 CKO DRGs ( I–P ) at E16 . 5 , assessed by in situ hybridization analysis . Note that expression of each of these nonpeptidergic-specific genes is impaired in both Ngf-/-Bax-/-and Runx1 CKO DRGs as early as E16 . 5 , a time when expression is first detected , indicating a requirement of both NGF and Runx1for initiation of expression . Shown are results representative of at least two independent animals per genotype . Scale bar , 50 μm . DOI:http://dx . doi . org/10 . 7554/eLife . 10874 . 004 While the large majority of nonpeptidergic-specific genes require both NGF and Runx1 for initiation of expression , expression of the nonpeptidergic marker Ret displays differential dependence on NGF and Runx1 . Although Ret expression is severely impaired in Ngf-/-; Bax-/- DRGs at P0 , as previously described ( Figure 2A–C ) ( Luo et al . , 2007 ) , its expression at this time point is only mildly affected in Runx1 CKO DRGs compared to controls , as shown by both in situ hybridization and real-time PCR ( Figure 2F , G , J ) . A time-course analysis of Ret expression in control and Runx1 CKO DRGs using both in situ hybridization and real-time PCR further confirmed the perinatal onset of Runx1 dependence for Ret expression ( Figure 2D–J ) . Therefore , while NGF is required for initiation of Ret expression , Runx1 is only required at a later stage , for maintenance of Ret expression . We conclude that NGF and Runx1 are both required for initiation of expression of a large cohort of nonpeptidergic nociceptor-specific genes , with Ret being a notable exception . The finding that initiation of expression of most nonpeptidergic nociceptor-specific genes analyzed so far is dependent on both NGF and Runx1 suggests a model in which Runx1 is a downstream mediator of NGF signaling during maturation of nonpeptidergic nociceptors . Alternatively , Runx1 may indirectly control nonpeptidergic nociceptor maturation by facilitating NGF signaling . In fact , the level of NGF signaling , as assessed by the fluorescent intensity of phospho-Trk ( pTrk ) immunostaining , is lower in Runx1 CKO DRG neurons relative to controls at P0 ( Figure 3—figure supplement 1A–F ) . This ability of Runx1 to sustain normal levels of TrkA activity is not simply due to an effect on TrkA expression , as TrkA levels in control and Runx1 CKO DRGs are indistinguishable , as determined by immunohistochemistry ( Figure 3—figure supplement 1G , H ) . However , since Ret expression , which is strongly dependent on NGF signaling in nonpeptidergic nociceptors , appears normal at E16 . 5 ( Figure 2D , E ) , it is unlikely that diminished NGF signaling observed at P0 accounts for the profound deficits in initiation of expression of nonpeptidergic-specific genes in E16 . 5 Runx1 CKO mice ( Figure 1—figure supplement 1 ) . Nevertheless , to directly test the possibility that Runx1 controls maturation of nonpeptidergic nociceptors solely by modulating NGF signaling , we asked whether exogenous NGF can rescue nonpeptidergic nociceptor gene expression deficits in Runx1 CKO DRGs . This was addressed in two ways . First , the consequence of excess NGF was evaluated in vivo by in situ hybridization and real-time PCR analysis following intraperitoneal injections of NGF into neonatal mice . With the notable exception of Ret , whose expression was partially restored following NGF injection into Runx1 CKO mice ( Figure 3J–L and Figure 3—figure supplement 2D ) , we found a general and complete lack of effect of NGF administration on expression of nonpeptidergic-specific genes in Runx1 CKO DRGs ( Figure 3A–I and Figure 3—figure supplement 2A–C ) . In a second series of experiments , dissociated DRG cultures from P0 control and Runx1 CKO animals were incubated in the presence or absence of NGF , and the effects of NGF on nonpeptidergic-specific gene expression were assessed by real-time PCR . Importantly , NGF treatment led to a robust increase in expression of the nonpeptidergic-specific genes , Mrgprd , Gfra2 , and Ptprt , in wildtype neurons but not in Runx1-deficient neurons ( Figure 2M–O ) . In contrast , Ret expression was NGF-dependent in both control and Runx1-deficient neurons , albeit to a lesser extent in the absence of Runx1 ( Figure 3P ) . Taken together , these findings are most consistent with a model in which Runx1 primarily acts downstream of NGF to control the initiation of expression of nonpeptidergic-specific genes . 10 . 7554/eLife . 10874 . 005Figure 2 . Ret is an unusual nonpeptidergic nociceptor-specific gene whose expression is differentially dependent on NGF and Runx1 . ( A and B ) Greatly diminished Ret expression in Ngf-/-Bax-/- DRGs compared to controls at P0 , assessed by in situ hybridization . The NGF-independent Ret+ neurons are mechanoreceptors ( Aβ RA-LTMRs ) . ( C ) Real-time PCR analysis of Ret expression in control and Ngf-/-Bax-/- DRGs at P0 confirms its strong NGF dependence in small diameter neurons . Statistical analysis was done using an unpaired t test , N = 4 , **p ≤ 0 . 001 . ( D–I ) Ret expression in control and Runx1 CKO DRGs at E16 . 5 ( D and E ) , P0 ( F and G ) and P10 ( H and I ) , assessed by in situ hybridization . Note that while Ret expression is almost completely eliminated in Ngf-/-Bax-/- DRGs at P0 , its expression at this same time point in Runx1 CKO DRGs is only modestly affected , indicating different temporal requirements of NGF and Runx1 for Ret expression . Shown are representative results of at least two independent animals per genotype at each time point . ( J ) Real-time PCR analysis of Ret expression in control and Runx1 CKO DRGs at P0 and P14 confirms the progressive nature of the Ret expression deficit in Runx1 CKO DRGs . Statistical analyses were done using unpaired t tests , N = 3 for each time point , p ≤ 0 . 05 , **p ≤ 0 . 001 . Ngf +/-; Bax-/- or Ngf +/+; Bax-/- and Runx1f/f mice were used as control animals for analysis of Ngf-/-; Bax-/- and Runx1 CKO mutants , respectively . Scale bar , 50 μm . DOI:http://dx . doi . org/10 . 7554/eLife . 10874 . 00510 . 7554/eLife . 10874 . 006Figure 3 . Runx1 functions downstream of NGF to mediate expression of the majority of nonpeptidergic nociceptor-specific genes , whereas it controls Ret expression at least in part by enhancing NGF signaling . ( A–L ) In situ hybridization analysis of expression of Mrgprd ( A–C ) , Gfra2 ( D–F ) , Ptprt ( G–I ) and Ret ( J–L ) in DRGs of P2 control animals that received BSA injections , Runx1CKO animals that received BSA injections or Runx1 CKO animals that received NGF injections . Note that exogenous NGF administration fails to activate expression of nonpeptidergic-specific genes in Runx1 CKO animals , with the notable exception of Ret , suggesting that Runx1 is a downstream mediator of NGF signaling that is required for expression of the majority of nonpeptidergic-specific genes . The ability of exogenous NGF to upregulate Ret expression in Runx1 CKO animals is most consistent with an indirect role for Runx1 in regulating Ret expression , through enabling NGF signaling . Shown are results representative of at least three independent injection experiments . See also Figure 3—figure supplements 1 , 2 . ( M–P ) Real-time PCR analysis of expression of Mrgprd ( M ) , Gfra2 ( N ) , Ptprt ( O ) and Ret ( P ) in dissociated DRG neurons from P0 control and Runx1 CKO animals cultured in the presence or absence of NGF . Note that , with the exception of Ret , NGF-dependent expression of these nonpeptidergic-specific genes is completely abolished in the absence of Runx1 , further supporting Runx1 as a downstream mediator of NGF in regulating expression of most nonpeptidergic-specific genes . Statistical analyses were done using two-way ANOVA with a Bonferroni post-test , N = 5 for M and P , N = 7 for the rest . p ≤ 0 . 05 , p ≤ 0 . 01 , p ≤ 0 . 001 , ns non-significant . Runx1f/f mice were used as control animals for analysis of Runx1 CKO mutants . Scale bar , 50 μm . DOI:http://dx . doi . org/10 . 7554/eLife . 10874 . 00610 . 7554/eLife . 10874 . 007Figure 3—figure supplement 1 . Runx1 potentiates TrkA activity without regulating TrkA expression . ( A–D ) Double staining of pTrk-SHC and Neurofilament heavy chain ( NFH ) ( A and B ) or pTrk-PLCγ and NFH ( C and D ) in control and Runx1 CKO DRGs at P0 shows greatly diminished pTrk immunoreactivity in NFH-negative neurons in Runx1 CKO DRGs compared to controls , suggesting Runx1-dependence of NGF signaling at this time point . NFH was used to exclude myelinated TrkB , TrkC-expressing DRG neurons from analysis . ( E and F ) Quantification of the NGF signaling deficit based on average fluorescence intensity of pTrk-SHC or pTrk-PLCγ immunoreactivity per cell within the neuronal population that is negative for NFH . An unpaired t test was performed on data from three independent pairs of control and mutant animals , p ≤ 0 . 01 , *p ≤ 0 . 001 . ( G and H ) TrkA immunostaining in control and Runx1 CKO DRGs at P0 shows comparable expression in both genotypes . Shown are results representative of three independent animals per genotype . Scale bar , 50 μm . DOI:http://dx . doi . org/10 . 7554/eLife . 10874 . 00710 . 7554/eLife . 10874 . 008Figure 3—figure supplement 2 . Runx1 controls expression of the majority of nonpeptidergic-specific genes independent of its stimulatory effect on NGF signaling . ( A–D ) Real-time PCR analysis of expression of Mrgprd ( A ) , Gfra2 ( B ) , Ptprt ( C ) and Ret ( D ) in DRGs of P2 control animals that received BSA injections , Runx1CKO animals that received BSA injections or Runx1 CKO animals that received NGF injections . A one-way ANOVA with a Tukey's multiple comparison test was performed on data from three independent experiments , p ≤ 0 . 05 , ns non-significant . DOI:http://dx . doi . org/10 . 7554/eLife . 10874 . 008 To further test this model , we explored the possibility that NGF promotes Runx1 activity during nonpeptidergic nociceptor maturation . To address the mechanism by which Runx1 activity may be modulated by NGF , it was critical to first ask how Runx1 activity is normally controlled in DRG neurons . Outside of the nervous system , Runx family transcription factors function as heterodimers , forming complexes with a common non-DNA-binding cofactor CBFβ . CBFβ is indispensable for Runx1 activity in cells of the hematopoietic lineage because of its role in enhancing the DNA-binding activity and stability of Runx1 proteins ( Adya et al . , 2000 ) . Therefore , we asked whether CBFβ plays a similar role in DRG neurons to promote Runx1 function . To determine whether CBFβ is expressed in DRG neurons at the appropriate time to form a complex with Runx1 , its expression patterns at both the mRNA and protein levels during nociceptor development were assessed . In situ hybridization analysis revealed that Cbfb is expressed at varying levels , in the majority , if not all , DRG neurons over the entire time course of our analysis ( Figure 4—figure supplement 1B–D ) . Due to a lack of CBFβ antibodies that work reliably for immunohistochemistry , we generated a CbfbFlag knockin mouse line , in which N-terminally Flag-tagged CBFβ protein ( Flag-CBFβ ) is produced from the endogenous Cbfb locus ( Figure 4—figure supplement 1A ) . In CbfbFlag mice , the Flag antibody allows for specific detection of endogenous Flag-CBFβ , which appears to be expressed in nearly all DRG neurons , at varying levels ( Figure 4A , B ) . Double labeling of Flag and Runx1 showed overlap between CBFβ and Runx1 throughout nociceptor development suggesting a potential interaction between these two proteins ( Figure 4C–E and Figure 4—figure supplement 1E–G ) . It is noteworthy that CBFβ is broadly expressed and found in more than just Runx1+ neurons , suggesting additional Runx1-independent CBFβ functions in the DRG ( Figure 4C–E ) . To test for a direct physical interaction between Runx1 and CBFβ in DRG neurons , co-immunoprecipitation experiments were done using a Flag antibody and extracts from DRGs of P0 CbfbFlag/Flag animals and wildtype littermate controls . Indeed , Runx1 is enriched in Flag immunoprecipitates from CbfbFlag/Flag animals but not wildtype animals ( Figure 4F ) . Thus , Runx1 and CBFβ are co-expressed and form a complex in developing DRG neurons . 10 . 7554/eLife . 10874 . 009Figure 4 . Runx1 and CBFβ are co-expressed and form a complex in DRG neurons . ( A and B ) Double immunostaining of Flag and Runx1 in wildtype and CbfbFlag/+ DRGs at P0 confirms the specificity of the Flag antibody . Note that Flag immunoreactivity in wildtype DRGs is nearly undetectable . ( C–E ) Double immunostaining of Flag and Runx1 in CbfbFlag/+ DRGs at P0 shows extensive colocalization between Flag-CBFβ and Runx1 ( arrows ) . Note that CBFβ is expressed in many more DRG neurons than those that are Runx1+ . See more examples in Figure 4—figure supplement 1E–G . ( F ) Co-immunoprecipitation experiments using a Flag antibody for immunoprecipitation from DRGs lysates from P0 wildtype and CbfbFlag/Flag animals . Subsequent detection with Runx1 and CBFβ antibodies shows that Runx1 associates with Flag-CBFβ from DRGs of CbfbFlag/Flag animals , but not wildtype controls , indicating the formation of a Runx1/CBFβ complex in the DRG . Scale bar , 50 μm . DOI:http://dx . doi . org/10 . 7554/eLife . 10874 . 00910 . 7554/eLife . 10874 . 010Figure 4—figure supplement 1 . Generation of the CbfbFlag allele and more detailed characterization of the temporal and spatial patterns of Cbfb expression . ( A ) Schematic showing the targeting strategy used for generation of CbfbFlagknockin mice . Following germ-line transmission , the Neo selection cassette was removed by crossing the carrier to a mouse expressing FlpE recombinase in the germ line . A Bstz171 restriction site was introduced immediately downstream of the Flag sequence to facilitate Southern screening of embryonic stem ( ES ) cells . Flag epitope sequences , LoxP and FRT sites are shown as red filled triangles , open and filled triangles respectively . ( B–D ) In situ hybridization analysis of Cbfb expression in wildtype DRGs at E13 . 5 , E16 . 5 and P14 reveals a widespread pattern of expression throughout development . Note that Cbfb is expressed at varying levels in DRG neurons . ( E–G ) Double immunostaining of Flag ( green ) and Runx1 ( red ) in CbfbFlag/ + DRGs at E13 . 5 , E16 . 5 and P14 shows a pattern of CBFβ protein similar to its mRNA distribution . DOI:http://dx . doi . org/10 . 7554/eLife . 10874 . 010 To assess the function of CBFβ in DRG development and , in particular , the role of CBFβ in Runx1-dependent nonpeptidergic nociceptor maturation , which has not previously been feasible because of the early lethality of Cbfb null embryos , we next generated a conditional Cbfb allele ( Cbfbf ) by flanking the putative promoter sequence and the first two exons of this gene with LoxP sites ( Figure 5—figure supplement 1A ) . Effective gene ablation in the DRG was achieved by crossing mice harboring the Cbfbf allele to a Wnt1-Cre mouse line , which drives recombination specifically in the dorsal neural tube and neural crest derivatives ( Danielian et al . , 1998 ) ( Figure 5—figure supplement 1B , C ) . Analysis of Wnt1-Cre; Cbfbf/f ( Cbfb CKO ) animals revealed a wide range of nonpeptidergic nociceptor phenotypes that are virtually identical to those seen in Runx1 CKO mice . Specifically , those genes found to be Runx1-dependent similarly require CBFβ for normal expression , as shown by in situ hybridization and real-time PCR analysis at P0 ( Figure 5A–J and Figure 5—figure supplement 1D , E ) . As has been shown for Runx1 CKO animals , expression of CGRP , a marker of peptidergic nociceptors , was de-repressed in Cbfb CKO animals ( data not shown ) . Furthermore , like Runx1 CKO mice , Cbfb CKO animals at P0 exhibited a marked reduction of sensory innervation of the epidermis ( Figure 5K–P ) . This defect was primarily due to a deficiency in the peripheral projections of nonpeptidergic nociceptors , considering the sparseness of epidermal innervation by peptidergic nociceptors at P0 as well as their unperturbed innervation pattern in the absence of CBFβ ( data not shown ) . We also found that the subepidermal nerve plexus is unaffected in Cbfb CKO mice , indicating a developmental defect in the final stage of peripheral target innervation . Therefore , both molecular and morphological features of nonpeptidergic nociceptors critically depend on both Runx1 and CBFβ during embryonic development . We next asked whether CBFβ is required at postnatal times for development of nonpeptidergic nociceptors . Since Wnt1Cre; Cbfb CKO animals die perinatally due to craniofacial defects , a Runx1CreER knockin allele ( Samokhvalov et al . , 2007 ) combined with postnatal tamoxifen administration was used to ablate Cbfb postnatally . A similar strategy was used to generate a Runx1 inducible knockout mouse model for direct comparison . Both Runx1CreER; Cbfbf/f and Runx1CreER/f mice treated with postnatal tamoxifen were viable and indistinguishable from control littermates . However , following postnatal deletion of either Runx1 or Cbfb , we observed , during the third postnatal week , few or no tyrosine hydroxylase ( TH ) + C-low threshold mechanoreceptors ( C-LTMRs ) , a nonpeptidergic neuronal subtype ( Li et al . , 2011; Seal et al . , 2009 ) . Specifically , key features of C-LTMRs , including TH expression and longitudinal lanceolate endings in the periphery , as marked by expression of a Cre-dependent GFP reporter ( Hippenmeyer et al . , 2005 ) , are nearly completely absent in mice lacking Runx1 or CBFβ at postnatal time points ( Figure 5—figure supplement 1F–M ) . Similar phenotypes were previously reported in a different Runx1 conditional mutant ( Lou et al . , 2013 ) . Therefore , both Runx1 and CBFβ are required during early development for initiation of the nonpeptidergic nociceptor fate and at postnatal times for maturation of at least one specific nonpeptidergic neuronal subtype , the C-LTMR . The phenocopy of Runx1 and Cbfb mutants may be partly attributed to a dramatic defect in Runx1 protein expression in Cbfb CKO DRGs , as shown by immunostaining and western blot analysis at P0 ( Figure 5Q–S ) . This Runx1 protein deficit was evident as early as E13 ( data not shown ) . Runx1 mRNA expression , on the other hand , remained unchanged , if not increased , in Cbfb CKO DRGs compared to controls , as determined by both in situ hybridization and real-time PCR ( Figure 5T–V ) . These findings demonstrate a key role for CBFβ in the post-transcriptional regulation of Runx1 expression , most likely at the level of protein stability . Together , these findings indicate that CBFβ and Runx1 are both essential for induction of the nonpeptidergic nociceptor fate and postnatal nonpeptidergic neuron subtype maturation , and that the CBFβ requirement may reflect its role in controlling the level and activity of Runx1 proteins . 10 . 7554/eLife . 10874 . 011Figure 5 . CBFβ is required for acquisition of molecular and morphological features of nonpeptidergic nociceptors . ( A–J ) Expression of Mrgprd ( Control , 26 . 9% ± 2 . 8%; Cbfb CKO , 0% ) , Gfra2 ( Control , 38 . 8% ± 2 . 8%; Cbfb CKO , 11 . 7% ± 1 . 9% ) , Ptprt , ( Control , 31 . 9% ± 3 . 2%; Cbfb CKO , 7 . 1% ± 2 . 8% ) , Myo1a ( Control , 26 . 9% ± 3 . 2%; Cbfb CKO , 5 . 6% ± 0 . 6% ) and Kif21b ( Control , 20 . 2% ± 0 . 1%; Cbfb CKO , 2 . 4% ± 0 . 5% ) in control and Cbfb CKO DRGs at P0 by in situ hybridization analysis . The gene expression deficits in Cbfb CKO animals phenocopy those observed in Runx1 CKO animals except for Kif21b expression . The discrepancy likely reflects Kif21b expression in proprioceptors where it presumably depends on Runx3 and CBFβ for expression . Shown are the means ± SEMs for the percentage of neurons expressing indicated genes based on counts from a total of at least 9 sections from three independent animals per genotype . DRG neurons were identified and counted based on combined NeuN immunostaining , which was not shown . See also Figure 5—figure supplement 1D , E . ( K–N ) GFP immunostaining of P0 hairy skin to visualize sensory innervation of the epidermis in control and Runx1 CKO animals ( K and L ) or control and Cbfb CKO animals ( M and N ) that also carry the TaumGFP allele . The TaumGFP allele was introduced to label all Cre-expressing neurons including all sensory neurons . Note that there is a dramatic reduction in fiber density specifically in the epidermis in both Runx1 CKO and Cbfb CKO animals relative to their littermate controls . The yellow dotted line denotes the epidermal-dermal junction which was drawn based on TOPRO3 counterstain ( blue ) . ( O and P ) Quantification of sensory innervation of the epidermis in control and Runx1 CKO animals ( O ) or control and Cbfb CKO animals ( P ) reveals a remarkably similar reduction in the innervation density in both mutants at P0 . The innervation density is defined as the fraction of area occupied by GFP+ fibers in the epidermis . An unpaired t test was performed on data from three independent animals per genotype . *p ≤ 0 . 001 . ( Q and R ) Runx1 immunostaining of control and Cbfb CKO DRGs at P0 shows almost complete loss of Runx1 proteins in the absence of CBFβ . Shown are representative images from at least three independent experiments . ( S ) Immunoblot analysis of expression of Runx1 and Cbfb in control and Cbfb CKO DRGs at P0 shows dramatic loss of Runx1 proteins as a result of CBFβ depletion . βIII-Tubulin serves as a loading control . Shown are results from three independent experiments . ( T and U ) In situ hybridization analysis of Runx1 expression in control and Cbfb CKO DRGs at P0 shows comparable levels of Runx1 transcripts in control and mutant animals . ( V ) Real-time PCR analysis of Runx1 expression in control and Cbfb CKO DRGs at P0 shows increased Runx1 mRNA expression in Cbfb CKO DRGs compared to control , which likely reflects an increased ratio of nociceptors to proprioceptors ( data not shown ) . An unpaired t test was performed on data from four independent pairs of control and mutant animals , p ≤ 0 . 01 . Cbfbf/f mice were used as control animals for analysis of Cbfb CKO mutants . Scale bar , 50 μm . DOI:http://dx . doi . org/10 . 7554/eLife . 10874 . 01110 . 7554/eLife . 10874 . 012Figure 5—figure supplement 1 . Generation of the Cbfbf allele and demonstration of a postnatal requirement for both CBFβ and Runx1 in C-LTMR development . ( A ) Schematic showing the targeting strategy for generation of the Cbfbf conditional allele . Following germ-line transmission , the Neo selection cassette was removed by crossing the carrier to a mouse expressing FlpE recombinase in the germ line . A Bstz171 restriction site was introduced immediately downstream of the 3’ loxP site to facilitate southern screening of ES cells . LoxP and FRT sites are shown as open and filled triangles respectively . ( B and C ) Efficient gene ablation in the DRG shown by in situ hybridization analysis of Cbfb expression in control and Cbfb CKO DRGs at P0 . ( D and E ) Real-time PCR analysis of expression of Gfra2 , Mrgprd , Ptprt and Myo1a in control and Cbfb CKO DRGs ( D ) or control and Runx1 CKO DRGs ( E ) at P0 further demonstrates the similarity in nonpeptidergic-specific gene expression deficits between Cbfb CKO and Runx1 CKO mutants . Unpaired t tests were performed on data from four independent pairs of control and Cbfb CKO animals , or three independent pairs of control and Runx1 CKO animals . p ≤ 0 . 01 , p ≤ 0 . 001 . ( F-I ) Expression of TH and GFP in DRGs of P21 Runx1CreER/ + ; TaumGFP/ + and Runx1CreER/f; TaumGFP/ + animals ( F and G ) ( Control , 23 . 2% ± 1 . 6%; Runx1 mutant , 3 . 8% ± 2 . 7% ) or Runx1CreER/ + ;Cbfbf/ + ; TaumGFP/ + and Runx1CreER/ + ; Cbfbf/f; TaumGFP/ + animals ( H and I ) ( Control , 21 . 4% ± 1 . 9%; Cbfb mutant , 8 . 9% ± 2 . 0% ) that received an intraperitoneal injection of tamoxifen at P2 . The TaumGFP allele was introduced to indicate the neurons with active Cre expression . Note that there is a substantial reduction in the number of GFP/TH double positive neurons due to a selective loss of TH expression in the GFP + population in both Runx1 and Cbfb mutant DRGs . Shown are means ± SEM for the percentage of GFP + neurons that express TH based on counts from a total of at least 9 sections from three independent animals per genotype . ( J–M ) Double staining of CGRP and GFP in back hairy skin of P21 Runx1CreER/ + ; TaumGFP/ + and Runx1CreER/f; TaumGFP/ + animals ( J and K ) or Runx1CreER/ + ;Cbfbf/ + ; TaumGFP/ + and Runx1CreER/ + ; Cbfbf/f; TaumGFP/ + animals ( L and M ) that received IP injections of tamoxifen at P2 . Note that in both mutant animals , there is a marked decrease in the number of GFP + longitudinal lanceolate endings characteristic of C-LTMRs , which is accompanied by an increased frequency of GFP + endings that assume a peptidergic neuron-like morphology . Shown are representative images from more than 3 independent experiments . Scale bar , 50 μm . DOI:http://dx . doi . org/10 . 7554/eLife . 10874 . 012 How does NGF control Runx1/CBFβ activity ? Since Runx1 and CBFβ are both necessary for nonpeptidergic neuron maturation , NGF could , in principle , control Runx1/CBFβ function by promoting the activity of Runx1 , CBFβ or both . Considering the profound requirement of NGF for nonpeptidergic nociceptor gene expression , one possibility is that NGF promotes expression of Runx1 , Cbfb , or both , thereby enabling Runx1/CBFβ complex formation , stabilization and function , and thus initiation of the nonpeptidergic nociceptor fate . We therefore examined expression of both Cbfb and Runx1 in DRGs of control and Ngf-/-; Bax-/- animals before and during the acquisition of the nonpeptidergic nociceptor fate . Interestingly , Cbfb expression was found to be sensitive to loss of NGF prior to specification of nonpeptidergic nociceptors . In fact , at E14 . 5 , a time prior to expression of nonpeptidergic nociceptor-specific genes , Cbfb mRNA expression was significantly reduced in Ngf-/-; Bax-/- DRGs relative to controls , specifically in small-diameter nociceptor precursors ( Figure 6A , B , E ) . This Cbfb expression deficit in Ngf-/-; Bax-/- small diameter DRG neurons becomes much more pronounced at later time stages , such as E16 . 5 , the onset of observable deficits in nonpeptidergic-specific gene expression ( Figure 6C–E ) . Similarly , Ntrk1-/-/Bax-/- DRGsdisplayed reduced levels of Cbfb expression at P0 ( Figure 7H , I ) . The deficit of Cbfb mRNA expression was confirmed by real-time PCR analysis ( Figure 6—figure supplement 1A ) . Considering the inability of this assay to distinguish between Cbfb expressed in nociceptors and that expressed in NGF-independent DRG neurons , such as proprioceptors , and non-neuronal cells of the ganglion , it is noteworthy that the real-time PCR measurements are an underestimation of the NGF dependence of Cbfb expression in developing nociceptors . We further addressed NGF-dependence of Cbfb expression at the protein level in two ways . First , Flag immunostaining on dissociated DRG cultures from P0 CbfbFlag/+ animals showed that NGF is essential for CBFβ protein expression in vitro ( Figure 6F–H ) . Second , when DRGs from P0 control and Ngf-/-; Bax-/- animals that were also heterozygous for the CbfbFlag allele were acutely dissociated , the level of Flag immunoreactivity was significantly lower in Ngf-/-; Bax-/- neurons compared to controls , suggesting NGF-dependence of CBFβ protein expression in vivo ( Figure 6—figure supplement 1B–D ) . Cbfb is also sensitive to the dose of NGF as exogenous NGF administered to wildtype neonates via intraperitoneal injection further potentiated Cbfb expression ( Figure 6—figure supplement 1E ) . Together , these findings identify NGF as a key regulator of Cbfb expression prior to the initiation of NGF-dependent nonpeptidergic-specific gene expression . On the other hand , Runx1 mRNA levels are normal in Ngf-/-; Bax-/- DRGs at E14 . 5 , as determined by both in situ hybridization and real-time PCR analysis ( Figure 6I , J , M ) , consistent with our previous observation ( Luo et al . , 2007 ) . It is only at later times , beginning at E16 . 5 , that Runx1 expression becomes affected , suggesting a late requirement of NGF for maintenance of Runx1 expression ( Figure 6K–M ) . Consistent with the late NGF dependence of Runx1 mRNA expression , Runx1 protein expression is unaffected in Ngf-/-; Bax-/- DRGs at E14 . 5 ( Figure 6N , O , R , S ) . At E16 . 5 , the level of Runx1 protein is significantly diminished without any change in the number of Runx1+ neurons ( Figure 6P–S ) . In view of the discrepancy between the dramatic nonpeptidergic phenotypes at E16 . 5 and the relatively modest deficit in Runx1 expression at this time , upregulation of Cbfb expression by NGF is likely to be an important mechanism by which NGF enables Runx1 function during nonpeptidergic nociceptor development . 10 . 7554/eLife . 10874 . 013Figure 6 . NGF regulates the Runx1/CBFβ complex through differential control of Cbfb and Runx1 expression . ( A and B ) In situ hybridization analysis of Cbfb expression in control and Ngf-/-Bax-/- DRGs at E14 . 5 shows a significant reduction in the level of transcripts in small diameter neurons that correspond to prospective nociceptors in Ngf-/-Bax-/- DRGs compared to controls . The insets focus on nociceptor-rich regions . Scale bar for the insets , 10μm . Note that Cbfb in situ hybridization was combined with Runx3 immunostaining to exclude the Runx3+ Cbfb population from the analysis . ( C and D ) In situ hybridization analysis of Cbfb expression in control and Ngf-/-Bax-/- DRGs at E16 . 5 shows more pronounced Cbfb mRNA deficit in Ngf-/-Bax-/- DRGs . ( E ) Quantification of Cbfb expression deficits in nociceptors in Ngf-/-Bax-/- DRGs based on experiments described in ( A–D ) . Intensity of in situ signal in areas devoid of Runx3+ neurons was measured . An unpaired t test was performed using data collected from 3 independent experiments for each time point . p ≤ 0 . 05 . See also Figure 6—figure supplement 1A . ( F and G ) Double staining of Flag and βIII-Tubulin in dissociated DRG neurons from P0 CbfbFlag/+ animals that were cultured without or with NGF . Note that NGF application robustly stimulates CBFβ protein expression as indicated by increased Flag immunoreactivity . ( H ) Quantification of the effect of NGF treatment on CBFβ protein levels based on experiments described in ( F and G ) . CBFβ protein abundance was quantified using the average fluorescence intensity of Flag immunoreactivity per cell . An unpaired t test was performed using data collected from four independent experiments . **p< 0 . 0001 . See also Figure 6—figure supplement 1B–D . ( I and J ) In situ hybridization analysis of Runx1 expression in control and Ngf-/-Bax-/- DRGs at E14 . 5 shows comparable levels of Runx1 transcripts in control and mutant DRGs . Means ± SEM for relative intensity of in situ signals after normalization to the level in control DRGs is as follows: Control , 1 . 00 ± 0 . 16; Ngf-/-Bax-/- , 0 . 73 ± 0 . 12 . p = 0 . 2079 , based on an unpaired t test . ( K and L ) In situ hybridization analysis of Runx1 expression in control and Ngf-/-Bax-/-DRGs at E16 . 5 shows a reduction in the level of signal per cell in Ngf-/-Bax-/- DRGs compared to controls . Control , 1 . 00 ± 0 . 07; Ngf-/-Bax-/- , 0 . 49 ± 0 . 06 . p = 0 . 0003 , based on an unpaired t test . ( M ) Real-time PCR analysis of Runx1 expression in control and Ngf-/-Bax-/- DRGs at E14 . 5 and P0 reveals a late requirement of NGF for Runx1 expression . An unpaired t test was performed on data collected from three independent animals per genotype at each time point . p ≤ 0 . 05 , ns non-significant . ( N–Q ) Runx1 immunostaining in control and Ngf-/-Bax-/- DRGs at E14 . 5 ( N and O ) and E16 . 5 ( P and Q ) shows that the Runx1 protein deficit becomes evident in Ngf-/-Bax-/- DRGs at E16 . 5 , which coincides with the onset of nonpeptidergic nociceptor deficits in Ngf-/-Bax-/- DRGs . ( R and S ) Quantification of Runx1 protein expression in control and Ngf-/-Bax-/- DRGs at E14 . 5 and E16 . 5 based on the percentage of Runx1+ neurons or the fluorescence intensity of Runx1 immunoreactivity . Note that loss of NGF specifically affects the level of Runx1 expression per cell without altering the number of Runx1+ neurons . An unpaired t test was performed on data collected from three independent animals per genotype . p ≤ 0 . 05 , ns non-significant . Ngf +/-; Bax-/- or Ngf +/+; Bax-/- mice were used as control animals for analysis of Ngf-/-; Bax-/- mutants . Scale bar , 50 μm . DOI:http://dx . doi . org/10 . 7554/eLife . 10874 . 01310 . 7554/eLife . 10874 . 014Figure 6—figure supplement 1 . Cbfb expression is NGF-dependent in vivo . ( A ) Real-time PCR analysis of Cbfb expression in control and Ngf-/-Bax-/- DRGs at E14 . 5 and P0 reveals early onset of NGF dependence for Cbfb expression . An unpaired t test was performed using data collected from three independent animals per genotype at each time point . p ≤ 0 . 05 . ( B and C ) Flag immunostaining of acutely dissociated DRG neurons from P0 control and Ngf-/-Bax-/- animals that are also heterozygous for CbfbFlag . ( D ) Quantification of NGF dependence of CBFβ protein expression based on experiments described in ( A and B ) . CBFβ protein abundance was quantified within the βIII-Tubulin + neuronal population based on the average fluorescence intensity of Flag immunoreactivity per cell . An unpaired t test was performed using data collected from three independent experiments , p ≤ 0 . 05 . ( E ) Real-time PCR analysis of expression of Cbfb in DRGs of P2 wildtype animals that received BSA injections or NGF injections . An unpaired t test was performed on data from three independent experiments , p ≤ 0 . 05 . Scale bar , 50 μm . DOI:http://dx . doi . org/10 . 7554/eLife . 10874 . 014 To better understand the mechanism by which NGF controls Cbfb expression in nociceptor precursors , we next sought to identify NGF–TrkA signaling cascades that control Cbfb expression . The canonical ERK1/2 signaling cascade represents a likely candidate because this is a major NGF–TrkA effector pathway , and animals deficient in components of this pathway exhibit nonpeptidergic phenotypes , including reduced Ret expression , impaired innervation of the epidermis , and impaired adult CBFβ protein expression ( Newbern et al . , 2011; Zhong et al . , 2007 ) . To directly assess the role of MAPK signaling in NGF-dependent Cbfb expression during development , both in vitro and in vivo approaches were used . Through immunostaining and immunoblot analysis , we found that pharmacological inactivation of Mek1/2 , direct activators of ERK1/2 , greatly attenuated the ability of NGF to promote CBFβ protein expression in vitro ( Figure 7A–E and Figure 7—figure supplement 1A–C ) . A Nestin-Cre-based conditional knockout mouse model , which targets all four Mapk3 , Mapk1 , Map2k1and Map2k2 alleles , was next used to determine the in vivo role of MAPK signaling for Cbfb expression . Using this in vivo model system , we found that in 4 out of 5 P0 Nes-Cre; Map2k1f/f;Map2k2-/-; Mapk3-/-; Mapk1f/f mutants ( Quadruple ) , Cbfb mRNA expression was severely disrupted , demonstrating a strong dependence of Cbfb expression on MAPK signaling in vivo ( Figure 7F , G ) . The phenotypic variation in this analysis likely reflects incomplete excision of all four floxed alleles . We next asked whether MAPK signaling is sufficient to promote Cbfb expression in vivo , in the absence of NGF or activation of other NGF-dependent signaling pathways . For this analysis , a constitutively active form of B-Raf ( V600E ) was expressed exclusively in the nervous system to drive MAPK signaling in animals that were null for both Ntrk1 and Bax ( O'Donovan et al . , 2014 ) . Remarkably , while Cbfb expression in Ntrk1-/-; Bax-/- DRGs was severely impaired , expression of B-RafV600E in Ntrk1-/-; Bax-/- mice to promote constitutive MAPK signaling restored Cbfb expression to near normal levels ( Figure 7H–J ) . Real-time PCR analysis was carried out as an independent measure of Cbfb expression to further confirm the necessity and sufficiency of MAPK signaling for Cbfb expression in vivo ( Figure 7K ) . Together , these findings indicate that the ERK/MAPK signaling cascade is both necessary and sufficient to mediate NGF/TrkA-dependent expression of Cbfb . 10 . 7554/eLife . 10874 . 015Figure 7 . NGF promotes Cbfb expression through the ERK/MAPK signaling pathway . ( A–D ) Double staining of Flag ( green ) and βIII-Tubulin ( blue ) in DMSO or U0126-treated dissociated DRG neurons from P0 CbfbFlag/+ animals that were cultured without or with NGF . U0126 is a selective inhibitor of MEK1/2 , the direct activators of ERK1/2 . Note that CBFβ protein expression , as defined by Flag immunoreactivity is greatly diminished in U0126-treated neurons as compared to vehicle-treated neurons , all grown in the presence of NGF . ( E ) Quantification of the effect of U0126 treatment on CBFβ protein levels based on experiments as described in ( A–D ) . CBFβ protein abundance was quantified as the average fluorescence intensity of Flag immunoreactivity per cell . Statistical analysis was done using a two-way ANOVA with a Bonferroni post-test , based on data from four independent experiments . **p ≤ 0 . 001 , ns non-significant . See also Figure 7—figure supplement 1 . ( F and G ) In situ hybridization analysis of Cbfb expression in control and quadruple DRGs at P0 reveals a severe deficit in Cbfb mRNA expression in DRGs when MAPK signaling is disrupted in the nervous system . A similar phenotype of varied severity was observed in 4 out of 5 mutant animals . ( H–J ) In situ hybridization analysis of Cbfb expression in Ntrk1 +/-; Bax-/- , Ntrk1-/-; Bax-/- and Ntrk1-/-; Bax-/-; Nes-Cre; V600E DRGs at E18 . 5 shows that constitutive activation of MAPK signaling leads to a dramatic increase in Cbfb expression in TrkA-deficient animals . Shown are representative images from two independent experiments . ( K ) Real-time PCR analysis of Cbfb expression in the same set of loss-of-function and gain-of-function mouse models as described in ( F and G ) and ( H–J ) at P0 further demonstrates the necessity and sufficiency of MAPK signaling for NGF-dependent Cbfb expression in vivo . Shown are averages from two independent experiments after normalization to littermate control . Scale bar , 50 μm . DOI:http://dx . doi . org/10 . 7554/eLife . 10874 . 01510 . 7554/eLife . 10874 . 016Figure 7—figure supplement 1 . In vitro evidence for the necessity of MAPK signaling for NGF-dependent CBFβ expression . ( A ) Immunoblot analysis of expression of Cbfb in DMSO or U0126-treated dissociated DRG neurons from P0 CbfbFlag/ + animals that were cultured in the presence or absence of NGF . Histone 3 serves as a protein loading control . ( B and C ) Quantification of the effect of U0126 on NGF dependence of Flag-CBFβ or CBFβ expression based on ( A ) . Inactivation of MAPK signaling by U0126 attenuates NGF-dependent CBFβ expression . The levels of Flag-CBFβ or CBFβ protein were determined by densitometry , and normalized to that of Histone 3 . The relative protein abundance of Flag-CBFβ or CBFβ after normalization to the level in DMSO-treated cultures that were grown without NGF was subject to a two-way ANOVA test with a Bonferroni post-test , N = 3 . p ≤ 0 . 05 , **p ≤ 0 . 01 , ns non-significant . DOI:http://dx . doi . org/10 . 7554/eLife . 10874 . 016 While Cbfb expression at early stages of nonpeptidergic nociceptor development is dependent on NGF , the apparent NGF-independence of Runx1 expression during early development prompted us to ask whether initiation of Cbfb and Runx1 expression are differentially controlled by intrinsic cues . We tested the involvement of Islet1 , a LIM-homeodomain transcription factor , because in a neural crest derivative-specific Islet1 mutant ( Isl1 CKO ) , a Runx1 protein deficit was noted as early as E12 . 5 , the time when Runx1 proteins are first detected in lumbar DRGs ( Dykes et al . , 2011; Sun et al . , 2008 ) . To define the level at which Runx1 expression is regulated by Islet1 , Runx1 expression was evaluated by in situ hybridization analysis in DRGs of E12 . 5 control and Isl1 CKO animals . Consistent with a central role for Islet in initiating Runx1 expression at the transcriptional level , Runx1 transcripts were virtually undetectable in Isl1 CKO DRGs ( Figure 8A , B ) . In contrast , Cbfb expression was only minimally affected by the same genetic perturbation ( Figure 8C , D ) . The distinct dependence of Runx1 and Cbfb expression on Islet1 was confirmed by a microarray analysis of control and Isl1 CKO DRGs at E12 . 5 ( Figure 8E ) . Thus , expression of Runx1 and CBFβ , obligatory components of a transcription factor holocomplex , are under differential control of Islet1 and NGF; Runx1 and Cbfb at early stages of nonpeptidergic nociceptor differentiation are preferentially regulated by the intrinsic cue Islet1 and the extrinsic cue NGF , respectively . Thus , a convergence of intrinsic and extrinsic signaling events in nonpeptidergic nociceptor progenitors enables formation of the Runx1/CBFβ transcription factor complex , a key event required for nonpeptidergic nociceptor differentiation . Nonpeptidergic nociceptor specification requires proper expression of both Runx1 and Cbfb , whose transcriptional initiation is differentially dependent on Islet1 and NGF signaling . The Runx1/CBFβ holocomplex therefore serves as a coincidence detector for extrinsic and intrinsic signals that promote specification of nonpeptidergic nociceptors . The requirement for NGF , an extrinsic cue that is critical for survival of all nociceptors , ensures that only those nociceptor precursors that gain access to a sufficient amount of NGF and survive the period of naturally occurring cell death will undergo nonpeptidergic nociceptor maturation . Consistent with this , Cbfb expression exhibits NGF dependence at E14 . 5 , immediately following the period of naturally occurring cell death . On the other hand , the dependence of Runx1 expression on Islet1 , a transcription factor required for terminating expression of genes whose expression is restricted to the early stage of sensory neurogenesis , likely contributes to the orderly transition from pan-sensory neurogenesis to specification of nonpeptidergic nociceptors by ensuring that Runx1 expression and hence Runx1/CBFβ-dependent nonpeptidergic nociceptor development are initiated after sensory neurogenesis ( Sun et al . , 2008 ) . Thus the NGF and Islet1 signals coordinate temporal control of the Runx1/CBFβ complex for timely initiation of nonpeptidergic nociceptor differentiation . It is important to note that the extrinsic and intrinsic signals described here are unlikely to be sufficient to instruct Runx1/CBFβ complex formation and the nonpeptidergic nociceptor lineage choice , as neither NGF nor Islet1 functions exclusively in nonpeptidergic nociceptors . Indeed , NGF–TrkA signaling is required for survival , target innervation and normal phenotypic development of both peptidergic and nonpeptidergic nociceptors ( Harrington and Ginty , 2013 ) , and Islet1 is expressed broadly in developing DRG neurons and is required for the generation of virtually all nociceptors ( Sun et al . , 2008 ) . Therefore , one or more additional , unidentified signals must govern the divergence of the two main nociceptive lineages . We found that MAPK signaling is both necessary and sufficient for NGF-dependent expression of Cbfb in developing nociceptors . The generality of MAPK signaling suggests that it plays a similar role in controlling Cbfb expression in cells outside of the nervous system and thus MAPK–CBFβ signaling may have a general role in cell type specification . Moreover , the identification of MAPK as an upstream activator of Cbfb expression may provide an explanation for the observation that Cbfb , unlike Runx genes , is widely expressed ( Wang et al . , 1993 ) . The identity of nuclear targets of the MAPK signaling pathway that directly activate Cbfb expression remains unclear . To this end , we found , through bioinformatic analysis of the putative Cbfb promoter , cAMP-response element ( CRE ) consensus motifs within an evolutionally conserved 458 bp enhancer-like element ( data not shown ) , raising the intriguing possibility that CREB family members , which are well-studied downstream mediators of MAPK signaling in many cell types ( Shaywitz and Greenberg , 1999 ) , are direct transcriptional activators of Cbfb . In view of the role of CREB family members in NGF-dependent growth and survival of neurons ( Liu et al . , 1999; Riccio et al . , 1999 ) , the possibility of CREB-mediated activation of Cbfb expression would further suggest that common effectors of growth factor signaling pathways support distinct biological outcomes , in this case , survival , axon growth and maturation of nociceptors , presumably by controlling expression of distinct target genes . The formation of the Runx1/CBFβ complex downstream of NGF and Islet1 signaling during nonpeptidergic nociceptor maturation illustrates a novel mechanism of interplay between extrinsic and intrinsic factors in controlling postmitotic specification of neuronal subtypes . It is important to note the distinction between this relatively late developmental process and the specification of progenitor domains , which takes place prior to cell cycle exit . While it is well established in vertebrate systems that specification of progenitor domains is coordinately regulated by the intrinsically defined competence state of a progenitor and spatially and temporally controlled extrinsic signals ( Briscoe and Novitch , 2008; Livesey and Cepko , 2001; Molyneaux et al . , 2007 ) , relatively little is known about the contributions of extrinsic cues , intrinsic factors , and their modes of interaction , during postmitotic specification of neuronal subtypes . The gene regulatory mechanism described here leads us to propose a simple model in which convergence of extrinsic and intrinsic signals onto a single heterodimeric transcription factor complex controls lineage-specific differentiation programs and postmitotic specification of neuronal subtypes ( Figure 8F ) . It is conceivable that this simple model is generally applicable to a broad range of neuronal subtypes that rely on heterodimeric transcription factor complexes for their specification . 10 . 7554/eLife . 10874 . 017Figure 8 . Islet1 is required for initiation of Runx1 , but not Cbfb expression . ( A–D ) In situ hybridization analysis of expression of Runx1 ( A and B ) and Cbfb ( C and D ) in control and Isl1 CKO DRGs at E12 . 5 shows that Islet1 deficiency abolishes expression of Runx1 but not Cbfb at an early age . Shown are representative images from two independent experiments . ( E ) Microarray analysis of E12 . 5 control and Isl1 CKO DRGs further confirms the differential dependence of expression of Runx1 and Cbfb on Islet1 . Shown are average expression levels from two independent experiments that are normalized to the control levels for each gene . Expression levels have been normalized using globe scaling . Isl1f/f mice were used as control animals for analysis of Isl1 CKO mutants . ( F ) Schematics illustrating a molecular mechanism underlying specification of nonpeptidergic nociceptors and its general implication in the context of subtype specification . The extrinsic cue NGF and the intrinsic cue Islet1 coordinately regulate the Runx1/CBFβ complex , a nonpeptidergic nociceptor transcription factor complex , by preferentially targeting Cbfb and Runx1 for transcriptional regulation , respectively . Furthermore , the Runx1/CBFβ complex , through an unknown mechanism , enhances the level of NGF-TrkA signaling , resulting in a positive feedback loop between NGF-TrkA signaling and Runx1/CBFβ complex . This gene regulatory mechanism not only underscores the importance of interplay between extrinsic and intrinsic factors during multilineage differentiation , but also illustrates how such interplay can control cell-fate decisions through the convergence of extrinsic and intrinsic signals at the level of a heterodimeric , lineage-specific transcription factor complex . Scale bar , 50 μm . DOI:http://dx . doi . org/10 . 7554/eLife . 10874 . 017 For generation of mice harboring the Cbfbf allele , a 2 kb sequence containing a 1 kb sequence immediately upstream of the transcription start site as well as exon 1 and exon 2 of the Cbfb locus was flanked by two loxP sites using recombineering technology . For generation of mice harboring the CbfbFlag allele , the sequence encoding a single Flag epitope was introduced immediately upstream of the translational start site of the Cbfb gene . Mice were generated using targeted ES cells and standard blastocyst injection techniques . Cbfbf/f mice were mated to either a mouse strain expressing Cre recombinase under control of the Wnt1 promoter ( Danielian et al . , 1998 ) to generate Wnt1-Cre; Cbfbf/f mice or to a mouse strain in which the Cre recombinase coding sequence was inserted into the Runx1 locus ( Samokhvalov et al . , 2007 ) to generate Runx1CreER/+; Cbfbf/f mice . Wnt1-Cre; Runx1f/f and Ngf−/−; Bax−/− mice were generated as described ( Chen et al . , 2006; Patel et al . , 2000 ) . The mouse lines used to generate Nes-Cre; Map2k1f/f; Map2k2-/-; Mapk3-/-; Mapk1f/f and Ntrk1-/-; Bax-/-; Nes-Cre; V600E mice were described previously ( Mercer et al . , 2005; Moqrich et al . , 2004; Newbern et al . , 2011 ) . The TaumGFP allele , a neuronal specific Cre-dependent GFP reporter , was previously described ( Hippenmeyer et al . , 2005 ) . See supplemental information for details on generation of the Cbfbf and CbfbFlag alleles . Digoxigenin ( DIG ) -labeled cRNA probes were used for in situ hybridization . Target sequences for Ptprt , Myo1a , Kif21b probes were amplified using gene specific PCR primers from either cDNA prepared from P0 mouse DRGs or genomic DNA from wildtype ES cells to generate corresponding plasmids for in situ hybridization . In situ hybridization probes for Mrgprd , Gfra2 , Ret and Runx1 were described previously ( Luo et al . , 2007 ) . The in situ hybridization probe for Cbfb was generated from an available cDNA clone ( GenBank: BC026749 . 1 ) . In situ hybridizations were carried out on 14 μm fresh frozen DRG sections as described previously ( Luo et al . , 2007 ) . For combined in situ hybridization and immunostaining , regular BCIP/NBT-based in situ hybridization was performed prior to the standard immunostaining procedure . Bright field and fluorescent images were taken under the same setting . Bright field images were later pseudocolored and merged with fluorescent images . Immunohistochemistry on DRG and skin sections was performed as described previously ( Li et al . , 2011; Luo et al . , 2007 ) . The primary antibodies used were: rabbit anti-Runx1 ( a gift from Dr . Thomas Jessell , Columbia University , 1:10 , 000 ) , rabbit anti-Runx3 ( a gift from Dr . Thomas Jessell , Columbia University , 1:4000 ) , guinea pig anti-Flag ( see supplemental information for antibody generation details , 1:500 ) , rabbit anti-CGRP ( Immunostar , 24112 , 1:1000 ) , chicken anti-GFP ( Aves Labs , GFP-1020 , 1:500 ) , chicken anti-NF200 ( Aves Labs , NFH , 1:500 ) , rabbit anti-Tyrosine Hydroxylase ( Millipore , AB152 , 1:1000 ) , rabbit anti-pTrk-SHC ( Cell Signaling Technology , 4168 , 1:500 ) , rabbit anti-pTrk-PLCγ ( Cell Signaling Technology , 4619 , 1:500 ) , rabbit anti-TrkA ( Millipore , AB1577 , 1:1000 ) , mouse anti-NeuN ( Milllipore , MAB377MI , 1:500 ) , and rabbit anti-βIII-Tubulin ( Covance , PRB-435P , 1:1000 ) . Dissociated DRG cultures from neonatal mice were prepared using a method that was adapted from a previously described protocol for sympathetic neuronal cultures ( Deckwerth et al . , 1996 ) . Briefly , neurons were obtained by sequential steps of enzymatic digestion and mechanical dissociation of DRGs harvested from P0 animals . In general , these neurons were plated on Poly-D-lysine and laminin coated coverslips at a density of ~50 , 000 neurons per well and cultured for 2 days in growth media ( 10% FBS , 1 U/ml penicillin/streptomycin , 5 μm Ara-C ( Sigma ) , 50 μg/ml a pan-caspase inhibitor Boc-aspartyl ( OMe ) -fluoromethylketone ( BAF ) ( MP Biomedicals ) ) supplemented with NGF ( 100 ng/ml; either purified from mouse salivary glands or purchased from Millipore ) , or a neutralizing NGF antibody ( Sigma ) at a dilution of 1:2000 . For experiments that used U0126 , cultures were treated with U0126 ( 50 μm in DMSO; Calbiochem ) or DMSO the morning after plating . For experiments that addressed in vivo dependence of CBFβ protein expression on NGF , neurons were plated for 1 hour in growth media without supplements before being processed for Flag immunostaining . RNA was extracted from freshly isolated DRGs using the RNeasy micro kit ( Qiagen ) according to the manufacturer’s instructions . First strand cDNA was synthesized using an oligo dT primer and the SuperScript III system ( Invitrogen ) . Real-time PCR was performed using the QuantiTect SYBR Green PCR kit ( Qiagen ) and a 7300 Real-Time PCR System ( Applied Biosystems ) . The amount of individual transcripts was normalized to that of PGP9 . 5 , a pan-neuronal marker , unless the comparison was between control and Ngf-/-; Bax-/- , for which GAPDH served as the internal control . Detailed primer sequences for real-time PCR can be found in supplemental information . Acutely dissected DRGs were lysed in ice-old FA-M2 Lysis Buffer ( 50 mM Tris HCl , 150 mM NaCl , 1 mM EDTA , 1% Triton X-100 , pH 7 . 5 ) supplemented with a protease inhibitor cocktail ( Sigma;1:100 dilution ) by sonication . Clarified lysates were either subjected to Flag immunoprecipitation for co-immunoprecipitation experiments or processed directly for SDS-PAGE . Immunoprecipitation of Flag-CBFβ and its associated proteins was done using anti-Flag M2 affinity gel ( Sigma ) according to the manufacturer’s instructions . Immunoblotting was performed using antibodies against Runx1 ( Abcam , 1:5000 ) , CBFβ ( 1:1000 , Santa Cruz ) , Histone 3 ( 1:1000 , Cell Signaling Technology ) and βIII-Tubulin ( 1:1000 , Covance ) , as described ( Kuruvilla et al . , 2000 ) . Statistical differences for mean values between two groups and among multiple groups were analyzed using GraphPad Prism 5 software . The type of test used for statistical analysis is indicated in the figure legend . The criterion for statistical significance was set at p ≤ 0 . 05 . For the Cbfbf allele , a ~2 kb sequence ( chromosome 8: 105169674- 105171592 ) corresponding to a 1 kb sequence immediately upstream of the transcriptional start site as well as exon 1 and exon 2 of the Cbfb locus was flanked by two loxP sites . Recombineering technology was used to generate the targeting vector ( Copeland et al . , 2001; Liu et al . , 2003 ) . Briefly , a 129/SvJ BAC clone containing the targeted region of the Cbfb gene was obtained from Geneservice . An 11 . 5 kb region ( chromosome 8: 105168174-105179387 ) with homology arms that were 1 . 5 kb and 8 kb long was inserted into a PBS-DTA plasmid , the backbone for the final targeting vector , via the first recombineering step . The 3’ loxP site and the FRT-Neo-FRT-5’ loxP cassette were then introduced sequentially during subsequent recombineering steps . A Bstz171 restriction site was engineered 3’ to the 3’ loxP site to facilitate southern screening of ES cells . The targeting construct was linearized with KpnI and then electroporated into mouse 129S6SvEvTac ES cells . ES clones were screened by PCR and correctly targeted ES clones were confirmed by southern blot hybridization using both 5’ and internal probes following Bstz171 digestion ( WT 9 . 8 kb and Mutant 6 . 8 kb , data not shown ) . Chimeric Cbfbf mice were produced by injection of positive ES cells into C57Bl/6 blastocysts . Mice carrying the Cbfbf allele were subsequently generated by mating chimeric mice to germ-line FlpE mice to remove the Neo cassette ( Rodriguez et al . , 2000 ) . Cbfbf mice were genotyped using a 2-primer PCR reaction with the following primers: 5’-GCGCGCCAGTCACTTGTT-3’ and 5’-AAACCATCCCACGAACCGAACCAT-3’ . The sizes of PCR products from wildtype and mutant alleles are 219 bp and 324 bp , respectively . For the CbfbFlag allele , the targeting vector , which was nearly identical to that of the Cbfbf vector , was generated using a combination of recombineering and standard subcloning strategies . The same targeted genomic region was engineered to include the FRT-Neo-FRT-loxP cassette at the position identical to that in the Cbfbf allele , using recombineering technology . The sequence encoding one Flag epitope was introduced into the vector immediately upstream of the translational start site of the Cbfb gene by replacing a 1 . 1 kb NotI/AvrII fragment containing the translational start site with the fragment carrying the insertion using standard cloning techniques . A Bstz171 restriction site was inserted immediately downstream of the Flag sequence for the purpose of southern screening . Subsequent steps for generation of CbfbFlag mice were the same as those described above for Cbfbf mice . CbfbFlag mice were genotyped using a 2-primer PCR reaction with the following primers: 5’-TGAGAGCTGTCTATGGCAAAC-3’ and 5’-TCAGTTCAAGGATGGCAGGTA-3’ . The sizes of PCR products from wildtype and mutant alleles are 232 bp and 336 bp , respectively . Primers used for real-time PCR analysis are provided Table 1 . 10 . 7554/eLife . 10874 . 018Table 1 . Primers used for real-time PCR analysisDOI:http://dx . doi . org/10 . 7554/eLife . 10874 . 018CbfbF-TCGAGAACGAGGAGTTCTTCAGGAR-AGGCGTTCTGGAAGCGTGTCTRunx1F-GCAGGCAACGATGAAAACTACTR-GCAACTTGTGGCGGATTTGTAMrgprdF-TGCTGCTGGAAACACTTCTAGGGAR-GCTGCTGTCAAGAGTGGAGTTCATGfra2F-TCGTACAGACCACTTGTGCCR-ATCAAACCCAATCATGCCAGPtprtF-ACCTGCTTCAACACATCACCCAGAR-TTCATCTTCCTTGGCTGTGTCCCAMyo1aF-ACAGGTGCTTCAACACAGCCAATCR-GCCCTTAAACAGTTCACTGGCACARetF-TCAACCTTCTGAAGACAGGCCACAR-ATGTCAGCAAACACTGGCCTCTTGPGP9 . 5F-CAGACCATCGGAAACTCCTGR-CACTTGGCTCTATCTTCGGGGAPDHF-ATGCCTGCTTCACCACCTTCTTR-ATGTGTCCGTCGTGGATCTGA Mouse pups of the desired genotype were subjected to a single intraperitoneal injection of either NGF ( 2 μg reconstituted in PBS with 1% BSA ) or equal volume of 1% BSA in PBS at both P0 and P1 . Animals were sacrificed at P2 and vertebral columns or DRGs were dissected and processed for analysis . Tamoxifen ( Toronto Research Chemicals ) was dissolved in ethanol ( 20 mg/ml ) . 50 μl ( 1 mg ) of tamoxifen in ethanol was mixed with 50 μl of sunflower seed oil ( Sigma ) , vortexed for 20 min and centrifuged under vacuum for 45 min to remove the ethanol . The tamoxifen solution was delivered at P2 via an intraperitoneal injection into animals harboring the Runx1CreER allele . A total of 6 RNA samples ( ∼1 μg each ) were prepared using Trizol and the RNeasy micro kit from DRGs of three pairs of E16 . 5 control and Runx1 CKO animals , each from different litters . Samples were labeled and hybridized to Affymetrix mouse 430 2 . 0 chips and microarray data were analyzed with Spotfire software . Only genes with a fold change greater than or equal to 1 . 5 , a p-value less than or equal to 0 . 05 were considered differentially expressed between control and Runx1 CKO DRGs and were reported . Methods for the microarray analysis of DRGs from Islet1 conditional knockout and control embryos have been described ( Dykes et al . , 2011 ) . A guinea pig polyclonal antiserum was raised against a MDYKDDDDKLVY peptide that corresponds to the N terminus of Flag-CBFβ encoded by the CbfbFlag allele using a service provided by Covance . The peptide was synthesized and conjugated at its C-terminus to KLH . Exsanguination bleeds were enriched for IgG using Protein A agarose chromatography . A sample of the IgG fraction was further affinity purified using a Flag peptide-conjugated column prepared using the Sulfolink immobilization kit for peptides ( Pierce ) . For quantifying the extent of epidermal innervation , three randomly selected regions of the epidermis were imaged for each animal . For each image , the epidermal region was defined based on TOPRO3 nuclear stain and selected as the region of interest . Images were then thresholded based on βIII-Tubulin or GFP immunostaining , and the area fraction ( the percentage of pixels above threshold in the region of interest ) was calculated using Image J and reported as a measure of epidermal innervation density . Area fraction for each image was considered an individual data point for statistical analysis . For quantifying the intensity of fluorescent images , images except those of cultured neurons were thresholded and regions of interest were defined either on a cell-by-cell basis or as populations of cells . The mean or total intensity of pixels above threshold was measured . For cultured neurons , the total intensity of pixels within each neuron was measured without thresholding . For quantifying in situ hybridization signal intensity , the same procedure was done , except that images were first converted to grayscale .	Animals detect and respond to their environment using their sensory nervous system , which forms through a complex , multi-step process . A precursor nerve cell’s fate is set early in its development , and determines the different nerve types it can become . As development progresses , sensory nerve cells develop further into distinct subtypes that perform particular tasks , such as responding to touch or pain . Nociceptors are the specialised sensory nerves that respond to potentially harmful stimuli . They form two distinct subtypes: peptidergic nerves detect potentially dangerous temperatures , whereas non-peptidergic nerves detect potentially dangerous mechanical sensations . Both subtypes originate from the same precursor nerve cell and both initially depend on an external molecule called NGF for their development and survival . During their development , non-peptidergic neurons stop responding to NGF and start producing a protein called Runx1 , considered to be the ‘master regulator’ of non-peptidergic nerve cell development . Runx1 works by forming a complex with another protein called CBFbβ , and this complex activates a program of gene expression that is specific to non-peptidergic nerves . However it was unclear how an external signal , like NGF , can coordinate with or influence a nerve cell’s internal genetic program during the nerve’s development . It was also not known whether NGF and Runx1 interact with each other . By studying non-peptidergic nerve cell development in mice that lack NGF , Runx1 and other associated proteins , Huang et al . have now established the sequence of events that regulate the development of this nerve cell subtype . Two signalling pathways converge to switch on non-peptidergic nerve cell development . An NGF-driven signalling pathway activates the production of CBFβ , while another protein binds to the Runx1 gene to switch it on . This leads to the production of the Runx1 and CBFβ proteins that complex together to activate the non-peptidergic neuronal genetic program . These findings demonstrate how two different mechanisms converge to produce the component parts of a complex , which then activates a genetic program that drives the development of a particular neuronal subtype . Whether this mechanism is involved in determining the fate of other cell types remains a question for future work .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "developmental", "biology", "neuroscience" ]	2015	Extrinsic and intrinsic signals converge on the Runx1/CBFβ transcription factor for nonpeptidergic nociceptor maturation
The brain has a remarkable capacity to acquire and store memories that can later be selectively recalled . These processes are supported by the hippocampus which is thought to index memory recall by reinstating information stored across distributed neocortical circuits . However , the mechanism that supports this interaction remains unclear . Here , in humans , we show that recall of a visual cue from a paired associate is accompanied by a transient increase in the ratio between glutamate and GABA in visual cortex . Moreover , these excitatory-inhibitory fluctuations are predicted by activity in the hippocampus . These data suggest the hippocampus gates memory recall by indexing information stored across neocortical circuits using a disinhibitory mechanism . Memories are thought to be stored across sparse and distributed neuronal ensembles in the brain ( Buzsáki , 2010; Josselyn and Tonegawa , 2020 ) . During memory recall , activity across these neuronal ensembles is selectively reinstated to recover enduring representations of the past . This reinstatement is thought to be mediated by the hippocampus , a brain region important for learning and memory ( Squire , 1992 ) . Anatomically , the hippocampus sits at the apex of a cortical sensory processing hierarchy ( Felleman and Essen , 1991 ) where inputs received by sensory cortices reach the hippocampus via the entorhinal cortex and other relay regions , which in turn make widespread cortico-cortical connections that project the hippocampal output back to neocortex ( Witter , 1993; Witter et al . , 1989 ) . This reciprocal anatomical connectivity equips the hippocampus with the necessary architecture to coordinate activity in neocortex . The hippocampus may therefore be considered to provide a ‘memory index’ , or summary sketch , for information stored across distributed cortical circuits ( Goode et al . , 2020; Teyler and DiScenna , 1985; Teyler and Rudy , 2007 ) . Consistent with this view , during memory recall , hippocampal reinstatement predicts subsequent neocortical reinstatement ( Pacheco Estefan et al . , 2019; Tanaka et al . , 2014 ) . However , the mechanism that allows the hippocampus to coordinate reinstatement across distributed neocortical circuits remains unclear . One possibility is that the hippocampus shapes computations performed by neocortical circuits by modulating the dynamic interplay between excitation and inhibition ( EI ) . At the cellular level , tight coupling between neocortical EI can be observed during both sensory stimulation and spontaneous neural activity ( Haider et al . , 2006; McCormick et al . , 2004; Okun and Lampl , 2008; Wehr and Zador , 2003 ) . This phenomenon has led to the physiological concept of EI balance , where , following changes in excitability , synaptic strength , current , or overall network activity returns to a stable set point via negative feedback ( Field et al . , 2020 ) . Evidence in humans , animal models , and theoretical models together suggests that EI balance is maintained to hold memories in a silent and dormant state ( Barron et al . , 2016; Froemke et al . , 2007; Vallentin et al . , 2016; Vogels et al . , 2011 ) , thus protecting memories from interference caused by new learning ( Koolschijn et al . , 2019; Kuchibhotla et al . , 2017 ) . During recall , however , EI balance must be transiently disturbed if memories are to be released from inhibitory control . Here , we predict that memory recall involves a transient break in EI balance , opening a window to release memories from the blanket of inhibition before network stability is re-established . Moreover , we predict that this transient break in neocortical EI balance is mediated by activity in the hippocampus . To test these predictions , here , we implemented a new imaging sequence in humans that combines functional magnetic resonance imaging ( fMRI ) with functional magnetic resonance spectroscopy ( fMRS ) ( Ip et al . , 2019; Ip et al . , 2017 ) . This sequence provides an opportunity to monitor activity in the hippocampus with fMRI while simultaneously measuring time-resolved fluctuations in neocortical glutamate and GABA using fMRS . MRS provides a unique tool to quantify the concentration of different neural metabolites ( De Graaf , 2019; Mangia et al . , 2012 ) , including glutamate and GABA , the principle excitatory and inhibitory neurotransmitters in the brain . MRS cannot dissociate between neurotransmitter and metabolic pools of glutamate and GABA ( Bak et al . , 2006; Magistretti and Allaman , 2015 ) . However , meaningful interpretation of MRS nevertheless derives from a major body of work showing an approximately 1:1 relationship between the rate of glutamine-glutamate cycling , which is necessary for glutamate and GABA synthesis , and neuronal oxidative glucose consumption , which indirectly supports neurotransmitter release among other processes ( Rothman et al . , 2003; Shen et al . , 1999; Sibson et al . , 1998 ) . Therefore , while measures of EI balance vary in both definition and granularity , MRS can provide a non-invasive marker for physiologically relevant EI at a coarse spatiotemporal scale . Correspondingly , MRS-derived glutamate and GABA reported during learning and memory paradigms show remarkable consistency with findings reported at the physiological level in animals ( Barron et al . , 2016; Castro-Alamancos et al . , 1995; Floyer-Lea et al . , 2006; Froemke et al . , 2007; Kolasinski et al . , 2019; Lunghi et al . , 2015; Trepel and Racine , 2000; Vallentin et al . , 2016 ) . Using the combined fMRI-fMRS sequence , here , we implemented a task designed to engage hippocampal-dependent recall of a visual cue . During memory recall , we report a transient increase in the ratio between MRS-derived glutamate and GABA in neocortex which is selectively predicted by the blood oxygen level-dependent ( BOLD ) signal in the hippocampus . These findings suggest the hippocampus coordinates memory recall by transiently perturbing neocortical EI balance to release memories stored across distributed neural circuits . To investigate the neuronal mechanisms that support memory recall , we designed a three-stage inference task . This task has previously been shown to involve associative memory recall in humans ( Barron et al . , 2020; Koster et al . , 2018 ) and mice ( Barron et al . , 2020 ) . Unlike some forms of associative recall , previous lesion and optogenetic studies in rodents demonstrate that associative recall required for inference is a hippocampal-dependent process ( Barron et al . , 2020; Bunsey and Eichenbaum , 1996; DeVito et al . , 2010 ) . Thus , the inference task provides an opportunity to investigate whether activity in the hippocampus mediates dynamic changes in neocortical EI during memory recall . The inference task was performed in virtual reality ( VR ) ( Figure 1A ) , an immersive and highly controlled 3D environment that has the potential to benefit from cross-species comparisons in the future ( Barron et al . , 2020 ) . The inference task was performed across 3 days and included three stages ( Figure 1B ) . In the first stage of the task , participants learned up to 80 auditory-visual associations ( ‘associative learning’ , day 1; Figure 1B , Figure 1—figure supplement 1 ) . In the second stage , which occurred approximately 24 hr later , each visual cue was paired with either a rewarding ( set 1 , monetary reward ) or neutral outcome ( set 2 , woodchip ) delivered to a wooden box in the corner of the VR environment ( ‘conditioning’ , day 2; Figure 1A–B , Figure 1—figure supplement 1 ) . Auditory cues were never paired with an outcome , providing an opportunity to assess evidence for an inferred relationship between these indirectly related stimuli . Accordingly , in the third stage of the task , we presented auditory cues in isolation , without visual cues or outcomes , and we measured evidence for inference from the auditory cues to the appropriate outcome ( ‘inference test’ , day 3; Figure 1B ) . Participants performed the inference test during an MRI scan ( Figure 1C–D , Figure 1—video 1 ) . On each trial of the inference test , participants were presented with an auditory cue , before being asked if they would like to look in the wooden box ( ‘yes’ or ‘no’ ) where they had previously found the outcomes during the conditioning stage . Participants’ responses depended upon whether they inferred the indirectly associated outcome to be rewarding or neutral . On trials where the auditory cue was associated with a visual cue paired with a rewarding outcome ( set 1 cues ) , participants were expected to select ‘yes’ if they inferred the relevant outcome ( Figure 1E ) . On trials where the auditory cue was instead associated with a visual cue paired with a neutral outcome ( set 2 cues ) , participants were expected to select ‘no’ if they inferred the relevant outcome ( Figure 1E ) . We thus categorised trials during the inference test as ‘correctly inferred’ if participants selected ‘yes’ when the auditory cue was indirectly associated with rewarding outcome or ‘no’ when the auditory cue was indirectly associated with a neutral outcome . Previous studies using this task show that in trials where participants infer the correct outcome , the associated visual cue that links the auditory cue and outcome is reinstated in the hippocampus and visual cortex ( Barron et al . , 2020 ) . Consistent with these previous findings , here , we show that participants make the correct inference only if they can later recall the relevant auditory-visual association during a surprise post-scan associative test ( Figure 1C; Figure 2A , C ) . Indeed , performance on the post-scan associative test that assessed memory for auditory-visual associations learned on day 1 predicted performance on the inference test ( Figure 2 ) . The inference task thus provides a suitable paradigm to investigate the neural mechanisms that support associative recall , in this case for auditory-visual associations . To investigate neural signatures of associative memory recall during the inference test , we implemented a novel imaging sequence ( Ip et al . , 2019; Ip et al . , 2017 ) which enabled interleaved acquisition of near-whole brain fMRI together with fMRS in primary visual cortex ( V1 ) ( Figure 3A ) . The fMRI-fMRS imaging sequence ( Figure 3A ) provided a means to simultaneously measure both haemodynamic and neurochemical changes in an event-related manner . By incorporating a temporal jitter in each trial of the experimental paradigm ( Figure 1D ) , the relationship between data acquisition and the experimental paradigm varied on a trial-by-trial basis ( Figure 4—figure supplement 1 ) . Therefore , across trials it was possible to effectively assess data at a higher temporal resolution than that given by a TR of 4 s . In the inference test , participants were required to make a binary ‘yes’/‘no’ response , with chance at 50% . To exclude trials where participants guessed , we classified trials as ‘remembered’ or ‘forgotten’ using a conservative approach . We filtered trials during the inference test post-hoc using participants’ behavioural performance from the subsequent post-scan associative test ( Figure 2B ) . Trials where participants made both the correct inference ( inference test; chance 50% ) and indicated the correct auditory-visual associations ( associative test; chance 1 . 6% ) were classified as ‘remembered’ . Trials where participants made either the incorrect inference ( inference test ) or indicated an incorrect auditory-visual association ( associative test ) were classified as ‘forgotten’ ( Figure 3B , Supplementary file 2 , Materials and methods ) . Neural signatures acquired during the ‘forgotten’ trials thus provided a condition- and stimulus-matched control for data acquired during the ‘remembered’ trials . Notably , this approach to categorising trials during the inference test controlled for false positives in the inference test , providing a conservative estimate of trials where participants remembered the auditory-visual associations . Notably , there was no significant difference between the number of trials in set 1 ( rewarding ) versus set 2 ( neutral ) for the ‘remembered’ and ‘forgotten’ conditions ( memory × set , two-way ANOVA: F ( 1 , 68 ) =0 . 67 , p = 0 . 424; Supplementary file 3 ) . Using the fMRI data from the interleaved sequence , we first identified brain regions modulated by recall of a visual cue during the inference test ( Figure 1D ) . Consistent with previous research investigating associative recall of visual cues ( Horner et al . , 2015; Wimmer and Shohamy , 2012 ) and data acquired using the same task ( Barron et al . , 2020 ) , we observed a significant increase in BOLD signal in both the hippocampus and visual cortex on ‘remembered’ versus ‘forgotten’ trials ( Figure 3C; Figure 3—figure supplement 1 ) . We then asked whether associative memory recall of a visual cue is accompanied by changes in the ratio between glutamate and GABA ( ‘glu/GABA ratio’ , see Materials and methods ) in visual cortex . We chose this ROI because recalling a visual cue is known to involve reinstating cortical representations in visual cortex ( Bosch et al . , 2014; Wheeler et al . , 2000 ) , including during inference as verified with an independent fMRI data set using the same task ( Barron et al . , 2020 ) . Using the interleaved fMRS data acquired in V1 ( Figure 3A and D ) , we quantified the concentration of glutamate and GABA normalised to total creatine ( tCr ) in an event-related manner ( Figure 3B and E ) . Notably , to assess dynamic changes in GABA , in the metabolite fitting procedure , it was not appropriate to employ default settings used to detect static estimates of GABA ( Appendix 1–supplementary note 1 ) . Importantly , these default settings constrain values of GABA relative to more stable metabolites , a process that effectively limits the dynamic range of GABA ( Figure 4—figure supplement 2 ) . Instead , here , we use unconstrained GABA estimates ( see Materials and methods ) : while this approach leads to GABA estimates that are higher than values normalised by the concentration of more stable metabolites , critically , dynamic changes in GABA can be detected ( Figure 4—figure supplement 2 ) . We used MRS-derived measures of glutamate and GABA to estimate changes in glu/GABA ratio ( Shibata et al . , 2017 ) . During associative memory recall in the inference test , we observed an increase in glu/GABA ratio in V1 when comparing ‘remembered’ versus ‘forgotten’ trials ( Figure 4A-C ) . Standard quality metrics indicated that our data quality was reliable over the course of the acquisition ( Figure 4—figure supplement 3 , Supplementary file 5 ) . To control for any biases introduced by differences in the number of ‘remembered’ versus ‘forgotten’ trials ( Supplementary file 6 ) , we compared the group mean metabolite change against a null distribution generated by permuting the identity labels ( ‘remembered’ or ‘forgotten’ ) assigned to each trial . This analysis again revealed a significant increase in glu/GABA ratio during memory recall , together with a significant decrease in GABA ( Figure 4D–F ) . These findings cannot be explained by differences in data quality measures between the ‘remembered’ and ‘forgotten’ conditions ( Figure 4—figure supplement 4 ) . In addition , the reported change in glu/GABA ratio was still observed when categorising trials into ‘remembered’ and ‘forgotten’ using performance on the inference task alone , a less conservative approach ( Figure 4—figure supplement 5 ) . The increase in glu/GABA ratio was not observed during periods immediately before or after recall ( Figure 4A–B; Figure 4—figure supplement 6 ) . Moreover , no effect between ‘remembered’ and ‘forgotten’ was observed in NAA , a neurometabolite that has overlapping peaks with GABA but is found at higher concentration ( Figure 4—figure supplement 7 ) . Notably , the observed within-subject , task-specific changes in neurochemistry were obscured when assessing the relationship between average glutamate and average GABA across subjects ( r17 = 0 . 191 , p = 0 . 433; after regressing out sex and age: r17 = 0 . 205 , p = 0 . 400 ) , consistent with previous findings ( Rideaux , 2021 ) . Thus , we propose that the reported transient increase in neocortical glu/GABA ratio reflects a mechanism for associative memory recall . As an additional control , we assessed changes in glu/GABA ratio during a subset of conditioning trials ( Figure 4—figure supplement 8A ) that were interleaved with the inference test trials during the MRI scan and shared the same temporal structure . Importantly , previous work suggests that performance on conditioning trials is not hippocampal-dependent ( Barron et al . , 2020 ) . During the conditioning trials , we observed no change in glu/GABA ratio during presentation of the visual cue or outcome , relative to the ITI period ( Figure 4—figure supplement 8B , C ) . We note that our MRS sequence does not use editing techniques which exploit known J-coupling relationships to separate signals deriving from low concentration metabolites , such as GABA , from stronger , overlapping signals ( Mullins et al . , 2014 ) . Instead , we implemented an MRS sequence without editing while taking advantage of the benefits associated with using a short TE . To further assess the sensitivity of our approach to detecting dynamic changes in GABA across task conditions , we used Monte Carlo simulations to generate MRS spectra while preserving the observed noise in our data . Using these simulations we show that the observed difference in GABA between ‘remembered’ and ‘forgotten’ conditions is significant from a null distribution that would be expected by chance ( Figure 4G ) . We next asked which brain regions coordinate this transient break in neocortical glu/GABA ratio during memory recall . The hippocampus is a promising candidate , given this brain region supports memory ( Squire , 1992 ) and shows activity modulation during the inference test ( Figure 3C ) . To test this possibility , we took advantage of our simultaneous fMRI-fMRS acquisition ( Figure 3A ) . We hypothesized that the increase in hippocampal BOLD signal observed during recall ( Figure 3C ) should predict the increase in glu/GABA ratio observed in V1 ( Figure 4B and F ) . In line with this prediction , across participants the hippocampal BOLD signal negatively predicted the relative concentration of GABA , and positively predicted the increase in glu/GABA ratio in V1 ( ‘remembered’ versus ‘forgotten’ trials; Figure 5A–B ) . This relationship between the BOLD signal and glu/GABA ratio was not observed in two control regions of interest ( ROIs ) ( Figure 5—figure supplement 1A , B ) . Furthermore , across the imaged brain volume ( Figure 3A ) , only the hippocampus significantly predicted the increase in V1 glu/GABA ratio on ‘remembered’ versus ‘forgotten’ trials ( Figure 5C ) . Finally , this relationship between the hippocampus and glu/GABA ratio was specific to the recall period during the inference test ( Figure 5D , Figure 5—figure supplement 1C , D ) . The hippocampus is thought to provide an index for memories stored across distributed neocortical circuits ( Goode et al . , 2020; Teyler and DiScenna , 1985; Teyler and Rudy , 2007 ) . However , the mechanism by which hippocampal activity coordinates with neocortex to facilitate memory recall has remained unclear . Here , using time-resolved fMRI-fMRS in humans , we show that recall of a visual cue is accompanied by a dynamic increase in the ratio between glutamate and GABA in visual cortex . This transient increase in glu/GABA ratio in visual cortex is selectively predicted by activity in the hippocampus . Accordingly , we propose the hippocampus gates recall of memories stored across distributed neocortical circuits using a disinhibitory mechanism ( Figure 5E ) . This mechanism may explain how a memory index represented by the hippocampus selectively releases otherwise dormant representations stored across distributed neocortical circuits . By simultaneously acquiring both fMRI and fMRS data , we provide a macroscopic readout of memory recall that reflects the consequence of underlying neural circuit level processes . Insight into the nature of these underlying circuit level processes can be gained from related data from animal models . For example , the neural circuit mechanisms that underlie an increase in glu/GABA ratio during recall may be informed by evidence that the ratio between excitatory and inhibitory synaptic conductances in cortical neurons fluctuate around a stable set point ( Anderson et al . , 2000; Okun and Lampl , 2008; Wehr and Zador , 2003; Wilent and Contreras , 2005 ) . This overall EI proportionality ensures that neurons and networks are neither hypo- nor hyper-excitable for prolonged periods , allowing memories to be held in a dormant state ( Barron et al . , 2016; Froemke et al . , 2007; Vallentin et al . , 2016; Vogels et al . , 2011 ) that is protected from interference caused by new learning ( Koolschijn et al . , 2019; Kuchibhotla et al . , 2017 ) . However , despite overall proportionality , the exact E/I ratio is highly dynamic and transient breaks in EI balance appear necessary for new learning and memory expression ( Letzkus et al . , 2015 ) . Here , the reported fluctuations in MRS-derived glu/GABA ratio during memory recall may therefore reflect , if indirectly , dynamic changes in EI balance . Similarly , the reported relationship between the fluctuations in glu/GABA ratio and hippocampal activity may be informed by data from animal models . Of particular relevance are studies in rodents which show that glutamatergic projections from higher-order or interconnected brain regions can target disinhibitory cortical circuits to provide selective EI modulation ( Krabbe et al . , 2019; Lee et al . , 2013; Zhang et al . , 2014 ) . For example , to enhance visual discrimination during attentional modulation , projections from the cingulate region of mouse frontal cortex modulate activity in V1 by targeting vasoactive intestinal polypeptide-expressing ( VIP+ ) interneurons , which in turn preferentially target other interneuron subtypes to release excitatory principle cells from inhibitory control ( Zhang et al . , 2014 ) . During memory recall , hippocampal projections may similarly permit memory reinstatement by targeting disinhibitory circuits in neocortex . The correlation between hippocampal activity and glu/GABA ratio reported here may therefore reflect a mechanism whereby activity in the hippocampus facilitates cortical disinhibition to release otherwise latent cortical associations from inhibitory control . This interpretation of the data is consistent with the notion that the hippocampus provides a memory index to flexibly reinstate information in extrahippocampal circuits ( Goode et al . , 2020; Teyler and DiScenna , 1985; Teyler and Rudy , 2007 ) . Moreover , our findings replicate equivalent analyses conducted on fMRI data acquired using the same task ( Barron et al . , 2020 ) and are consistent with previous studies in humans showing evidence for coordinated hippocampal-neocortical memory reinstatement ( Horner et al . , 2015; Pacheco Estefan et al . , 2019 ) . When combined with the fMRS data , our results also corroborate findings in humans showing that hippocampal glutamate and GABA can predict mnemonic control ( Nikolova et al . , 2017; Schmitz et al . , 2017 ) . Taken together , we propose a mechanism for hippocampal indexing whereby hippocampal projections control the release of mnemonic representations in sensory cortices by targeting disinhibitory circuits . Given this interpretation of the data , we emphasise that the relationship between MRS-derived measures of glutamate and GABA and physiological measures of EI balance remains complex . Rapid changes in synaptic glutamate and GABA that accompany neurotransmitter release occur on a timescale that is not possible to detect using fMRS . Moreover , only a fraction of MRS-derived neurometabolite concentration reflects neurotransmitter release . Of the different pools of glutamate and GABA ( cytoplasmic , vesicular , or extracellular ) , MRS is considered most sensitive to unconstrained , intracellular metabolic pools that reside at relatively high concentration in the neuronal cytoplasm ( Rae , 2014 ) . Indeed , changes in extracellular GABA of less than 100-fold are unlikely to be detectable using MRS ( Myers et al . , 2016 ) and post-mortem studies suggest MRS is not sensitive to intracellular pools that reside in the mitochondria or vesicles ( De Graaf and Bovée , 1990; Kauppinen and Williams , 1991 ) . Interpretation of MRS-derived glutamate and GABA is further complicated by the fact that the release and recycling of glutamate and GABA constitute major metabolic pathways ( Bak et al . , 2006; Magistretti and Allaman , 2015 ) . Yet , the metabolic and neurotransmitter pools are thought to be tightly coupled during anaesthesia , rest and certain stimulation protocols , with a 1:1 relationship reported between the rate of glutamine-glutamate cycling , which is necessary for glutamate and GABA synthesis , and neuronal oxidative glucose consumption , which indirectly supports neurotransmitter release among other processes ( Rothman et al . , 2003; Shen et al . , 1999; Sibson et al . , 1998 ) . Therefore , an increase in synaptic neurotransmission occurs together with an increase in synthesis of exogenous glutamate , which provides a precursor for GABA via the glutamate-glutamine cycle . During sensory stimulation a transient uncoupling has been observed with a short-lived mismatch between glucose utilization and oxygen consumption ( Fox et al . , 1988; Fox and Raichle , 1986 ) , particularly during stimulation protocols that alternate between high intensity and quiescent periods ( Gjedde et al . , 2002 ) . Dynamic fluctuations in fMRS-derived glutamate and GABA reported here may therefore reflect transitions to new metabolic steady states ( Stanley and Raz , 2018 ) , which could reflect ( if indirectly ) relative shifts in EI equilibrium at the physiological level . During associative memory recall , the increase in glu/GABA ratio may therefore be interpreted as an increase in synthesis of glutamate relative to degradation , with an opposing effect on GABA . This interpretation is supported by a handful of previous studies showing event-related changes in MRS glutamate ( Apšvalka et al . , 2015; Gussew et al . , 2010; Lally et al . , 2014 ) and GABA ( Cleve et al . , 2015 ) , together with a growing body of evidence reporting a relationship between MRS-derived measures of neurometabolites and behaviour ( Puts et al . , 2011; Scholl et al . , 2017; Stagg et al . , 2011 ) . Nevertheless , it remains to be established whether unconstrained glutamatergic and GABAergic pools show event-related changes that are MRS-sensitive . To validate this interpretation of event-related fMRS , it is important to leverage animal studies where more sensitive methods can be employed to relate fMRS measures to physiological parameters . Here , by implementing an inference task in VR , we operationalize memory recall using the exact same paradigm previously employed in rodents ( Barron et al . , 2020 ) . Therefore , in addition to engaging memory-dependent inference , ‘opening the box’ to find a reward in the VR environment approximated the process of rodents finding a reward from a dispenser in a 3D environment . By using VR , the findings presented here may be compared to data acquired in animal models in ongoing future research . In this manner , VR paradigms in humans may provide a basis from which to gain insight into the cellular and circuit mechanisms that underlie macroscopic measures of EI . This may prove particularly useful for establishing a more detailed understanding of the relationship between fMRS-derived measures of glutamate and GABA and physiological measures of EI balance . Previous MRS protocols typically employ a ‘block’ design , where a static measure of the concentration of glutamate and GABA is achieved by averaging the spectra across a time window that may span several minutes . This approach obscures the temporal dynamics of neurometabolites which more closely relate to fluctuations in EI reported at the physiological level . Similarly , dynamic changes in neurometabolites that accompany cognitive processes and ongoing behaviour are overlooked . Indeed , when the average concentration of Glx and GABA are considered in V1 across time , no significant relationship is observed across subjects ( Rideaux , 2021 ) , a result which we also observed when assessing average glutamate and GABA using our dataset . By contrast , with the increase in availability of ultra-high field MRI scanners and the development of more advanced sequences ( Stagg and Rothman , 2013 ) , fMRS has emerged as a viable method to detect dynamic changes in neurochemicals in both healthy and clinical populations ( Stanley and Raz , 2018 ) . Although there are currently only a handful of event-related fMRS studies , together with our data , these suggest that fMRS is highly sensitive to detecting task-relevant dynamic changes in glutamate and GABA ( Jelen et al . , 2018 ) . For example , in the lateral occipital complex , fMRS demonstrates differences in glutamate in response to presentation of objects versus abstract stimuli ( Lally et al . , 2014 ) , and in the left anterior insula fMRS reveals a transient increase in glutamate with exposure to painful stimuli ( Gussew et al . , 2010 ) . fMRS-derived glutamate is even sufficiently sensitive to detect repetition suppression effects in the lateral occipital complex ( Apšvalka et al . , 2015 ) , mirroring analogous effects reported in fMRI ( Barron et al . , 2016; Grill-Spector et al . , 2006 ) . Here , we further illustrate that within a 3 s time window delineated by the question period in the inference task , the temporal resolution of fMRS is sufficient to relate transient changes in glutamate and GABA to associative memory recall . Importantly , we compare data across two conditions ( ‘remembered’ and ‘forgotten’ ) to inherently control for: ( 1 ) between-subject differences in average GABA and glutamate which are affected by demographic ( e . g . age and gender ) ; ( 2 ) between-subject differences in spectral quality; ( 3 ) between-subject differences in tissue composition; ( 4 ) between-subject differences in the effect of other neurochemicals on measures of glutamate and GABA . Such time-resolved , within-subject , and condition-dependent fMRS may provide a promising tool to capture real-time , task-relevant changes in neurometabolites , on a timescale equivalent to task-based fMRI . Assessing whether the temporal resolution of fMRS can be further improved will likely prove an important step in refining fMRS in the future . During associative memory recall , the transient increase in glu/GABA ratio reported in our data can be accounted for by a significant decrease in the concentration of MRS-derived GABA . Notably , detecting dynamic changes in GABA is challenging for two key reasons: the concentration of GABA in human brain tissue is relatively low and the spectral peaks for GABA overlap with other , more abundant neurochemicals ( Andreychenko et al . , 2012; Govindaraju et al . , 2000; Puts and Edden , 2012 ) . While the most common approach to detecting MRS-derived GABA involves using a J-difference spectral editing technique to separate GABA peaks from overlapping peaks ( Bottomley , 1987; Mescher et al . , 1998 ) , here we use a non-edited sequence ( sLASER ) . While spectral editing may provide higher precision ( Hong et al . , 2019 ) , this occurs at the cost of a larger volume of interest , longer TEs and higher susceptibility to motion and drift artefacts due to longer acquisition times , making it less suitable for event-related fMRS ( Terpstra et al . , 2006; Trabesinger and Boesiger , 2001 ) . Moreover , direct comparisons between edited and non-edited sequences at 7 T reveal no significant difference in the concentration of GABA measurements ( Hong et al . , 2019 ) . Therefore , together with studies reporting dynamic changes in GABA with sensory stimulation ( Lin et al . , 2012; Mekle et al . , 2017 ) , our data illustrates how a non-edited sequence can provide sufficient data quality for measuring dynamic changes in MRS-derived GABA , which cannot be explained by changes in compounds at higher concentration that have overlapping peaks ( i . e . glutamate or NAA , Figure 4—figure supplement 7 ) . Indeed , Monte Carlo simulations reported here validate that non-edited sequences can be used to quantify dynamic changes in GABA ( Figure 4G; Figure 4—figure supplement 2 ) . Moreover , compared to spectral editing , our approach comes with the advantage of simultaneously measuring dynamic changes in GABA and glutamate , together with 17 other neurometabolites . To detect dynamic changes in GABA , it was necessary to disable default priors on the spectral fitting procedure that constrain GABA as a ratio to more stable metabolite concentrations ( Figure 4—figure supplement 2 , see Appendix 1—supplementary note 1 ) . As a consequence , we were able to detect dynamic changes in both glutamate and GABA across time , as illustrated using Monte Carlo simulations and permutation testing . By comparing the change in metabolite concentration between two conditions ( ‘remembered’ versus ‘forgotten’ ) , the ratio in GABA between conditions rather than absolute values was the key measure of interest . However , we note that absolute GABA estimates were higher compared to those obtained using default priors that normalise estimates relative to more stable metabolite concentrations . Importantly , the quality of our MRS data was comparable with other studies that have acquired 7 T MRS data from visual cortex ( Bednařík et al . , 2018; Hong et al . , 2019; Mekle et al . , 2017; Prinsen et al . , 2017 ) . Moreover , the quality of the glutamate estimates was in line with previous studies employing event-related fMRS to assess dynamic changes in glutamate ( Apšvalka et al . , 2015; Gussew et al . , 2010; Lally et al . , 2014 ) . Disturbances in EI balance are thought to underlie a number of neuropsychiatric conditions , including schizophrenia , autism , epilepsy , and Tourette’s syndrome ( Robertson et al . , 2016; Stanley and Raz , 2018; Taylor et al . , 2015 ) . While previous studies report inconsistencies in MRS-derived measures of glutamate and GABA in these patient populations , this may be attributed to differences in brain region , cognitive state , and imaging protocol , among other factors . Here , by using both fMRS and fMRI to reveal a signature change in glu/GABA ratio that relates to hippocampal BOLD signal , behavioural performance , and cognition , our findings present a potential target for clinical investigation . Moreover , our findings show that even in the healthy brain a transient break in EI balance is necessary to support key cognitive processes such as memory recall . In summary , using time-resolved fMRI-fMRS , we report a transient increase in glu/GABA ratio in V1 during associative recall of a visual cue . This increase in neocortical glu/GABA ratio is predicted by activity in the hippocampus . By unveiling this coordination between the hippocampus and neocortex , we show how the hippocampus may have the capacity to selectively modulate and disinhibit memories represented in neocortex . This mechanism may explain how the hippocampus plays a key role in memory recall , by indexing the release of specific memories stored across distributed neocortical circuits . This study did not generate new unique reagents . The inference test was incorporated into the fMRI-fMRS scan task . This provided an opportunity to measure neural responses to associative memory recall required for inferential judgements . The scan task included two different trial types: inference test trials ( Figure 1D ) and conditioning trials ( Figure 4—figure supplement 8A ) . For both types of trial , participants viewed a short video taken from the VR training environment ( Figure 1—video 1 ) . The videos were presented via a computer monitor and projected onto a screen inside the scanner bore . On each trial the duration of the video was determined using a truncated gamma distribution with mean of 7 s , minimum of 4 s , and maximum of 14 s . During the inference test trials , the video of the VR environment was accompanied by an auditory cue , played over MR compatible headphones ( S14 inset earphones , Sensimetrics ) . Visual cues were not displayed during these trials: the auditory cues were presented in isolation . At the end of the video , participants were presented with a question asking: ‘Would you like to look in the box ? ’ , with the options ‘yes’ or ‘no’ ( Figure 1D ) . Importantly , as described above , outcomes ( rewarding or neutral ) were delivered to the wooden box during the conditioning stage . Participants were required to make a response within 3 s using an MR compatible button box and their right index or middle fingers . No feedback was given . To infer the appropriate outcome , participants were instructed to use the learned structure of the task . After each trial ( inference or conditioning ) , a cross was presented in the centre of the screen during an ITI of varying length , determined using a truncated gamma distribution ( mean of 2 . 7 s , minimum of 1 . 4 s , maximum of 10 s ) . Trials were categorised as ‘correctly inferred’ if participants pressed ‘yes’ in response to auditory cues indirectly associated with a rewarding outcome , or pressed ‘no’ in response to auditory cues indirectly associated with a neutral outcome ( Figure 1E ) . The inference test provided an opportunity to investigate memory recall: to infer the correct outcome participants needed to recall the appropriate visual cue associated with the auditory cue ( Figure 2C–D ) . Conditioning trials were interleaved with inference test trials to minimise extinction effects . During conditioning trials , the video of the VR environment orientated towards a visual stimulus displayed on one of the four walls ( Figure 4—figure supplement 8A ) . At the end of the video , participants were presented with a still image of the associated outcome for that visual cue ( Figure 4—figure supplement 8A ) . To control for potential confounding effects of space , each video during the inference test involved a trajectory constrained to a 1/16 quadrant of the VR environment , evenly distributed across the different auditory cues . Across conditioning trials , each visual cue was presented 16 times , once in each possible spatial quadrant . Moreover , the videos were not related to the background for the relevant visual cue . Allocation of the videos to each trial was randomised separately for each participant to ensure no consistent biases . The fMRI-fMRS scan task was evenly divided across two scan blocks , each of which lasted 15 min . The fMRI-fMRS scan task was then repeated ( two more scan blocks ) using a higher quality multiband fMRI sequence ( reported elsewhere; Barron et al . , 2020 ) . To assess the relative sensitivity to detecting changes in GABA in our data , we used Monte Carlo simulations to generate synthetic spectra . The average observed spectrum ( across participants ) was used as an input to Monte Carlo simulations ( Clarke et al . , 2021 ) . For each set of conditions , we generated 2000 simulated spectra , with the SNR and line width of the simulated data matched to the SNR and line width observed in the data . The T2 values are assumed to be the same between conditions since we did not see any differences in FWHM between ‘remembered’ and ‘forgotten’ conditions in the in vivo data ( Figure 4—figure supplement 4B ) . The output of the simulations was then analysed in LCModel to quantify GABA and glutamate . These simulated spectra were used for two types of analyses . First , the simulated spectra were used to test the likelihood of observing the measured change in GABA between ‘remembered’ and ‘forgotten’ conditions by chance ( Figure 4G ) . Second , the simulated spectra were used to assess the effect of synthetically imposing changes in GABA , both with and without constraining GABA relative to the concentration of other more abundant neurochemicals ( ‘constraints on’ versus ‘constraints off’ ) ( Figure 4—figure supplement 2 ) . The imposed changes in GABA were the following multiples of the observed difference in GABA ( between ‘remembered’ and ‘forgotten’ ) : 0 , ±0 . 5 , ±1 , ±2 . To assess the sensitivity of the ‘constraints on’ and ‘constraints off’ LCModel settings to changes in imposed GABA at different SNRs , we compared the slope for each setting . Slopes were determined by fitting a general linear model ( GLM ) to the imposed versus measured GABA concentration in the simulated data . To quantify the difference in slope between categories ( Figure 4—figure supplement 2C: ‘constraints on’ versus ‘constraints off’; Figure 4—figure supplement 2D: differences in SNR ) , we randomly sampled n = 18 simulated spectra for each imposed change in GABA , for each condition of interest ( e . g . ‘constraints on’ versus ‘constraints off’; SNR 125% versus SNR 75% , etc . ) . Over 500 sets of size n = 126 ( i . e . n = 18 simulated spectra for each imposed change in GABA ) , we estimated the power to reject the null hypothesis of equal slopes between conditions . Similarly , using a parametric statistical approach , over 500 sets of size n = 126 simulated spectra , we estimated the t-statistic to reject the null hypothesis of equal slopes between conditions . Using the output of the GLMs , we assessed the difference in the univariate BOLD response between ‘remembered’ and ‘forgotten’ trials during the inference test ( as defined in Figure 3B , Trial categorisation during the inference test ) . The contrast of interest therefore involved contrasting EVs [1:4] ( ‘remembered’ ) with EVs [5:8] ( ‘forgotten’ ) , using the first GLM ( see above ) . The resulting contrast images ( ‘remembered’–‘forgotten’ ) for all participants were entered into a second-level random effects ‘group’ analysis . We set the cluster-defining threshold to p < 0 . 01 uncorrected before using whole-brain family wise error ( FWE ) to correct for multiple comparisons , with the significance level defined as p < 0 . 05 ( Figure 3C , Supplementary file 4 ) . To assess the relationship between event-related hippocampal BOLD signal and event-related fMRS measures from V1 , we used an anatomical ROI for the hippocampus ( Figure 5A ) . Capitalising on variance across participants , the relationship between the BOLD signal for ‘remembered’–‘forgotten’ within this ROI was compared with equivalent changes in glutamate , GABA , and glu/GABA ratio using a Spearman rank correlation . To assess the selectivity of these effects to the recall period ( question ) during the inference test , control analyses were performed using the output of the second and third GLMs , together with equivalent measures of glutamate , GABA , and glu/GABA ratio ( Figure 5—figure supplement 1 ) . Next , to assess the relationship between fMRS and the BOLD signal across the entire imaged brain volume , we repeated the second-level random effects ‘group’ analysis using the output of the first GLM , but now included group-level covariates for the change in glutamate and GABA for ‘remembered’–‘forgotten’ ( i . e . Figure 4A ) , along with two ‘nuisance’ regressors that accounted for unwanted variance attributed to differences in age and sex . To identify brain regions where the BOLD signal for ‘remembered’:‘forgotten’ predicted changes in glu/GABA ratio , we contrasted the EVs on the covariates for glutamate and GABA ( glutamate–GABA ) to generate a single contrast to test statistical significance . We set the cluster-defining threshold to p < 0 . 01 uncorrected before using whole-brain FWE to correct for multiple comparisons , with the significance level defined as p < 0 . 05 ( Figure 5C , Supplementary file 7 ) . To visualize the time course of fMRS across the inference test trials , we estimated a moving average , where each time bin constituted a 2 . 5 s time window shifted by 0 . 5 s in each iteration ( Figures 4C and 5D , Figure 5—figure supplement 1A , B ) . By incorporating a random jitter in each trial of the fMRI-fMRS scan task , the temporal relationship between MRS spectra acquisition and the inference test trials varied . Thus , across participants and across trials , MRS spectra were acquired in all possible 2 . 5 s time bins of the inference test trial , achieving a higher temporal resolution than the TR of 4 s ( Figure 4—figure supplement 1 ) . To ensure each time bin contained a similar number of spectra , those bins at the tail end of the jitter ( final three time bins during the video and the final two time bins of the ITI ) were enlarged to include broader time windows ( > 2 . 5 s ) . For each participant , the ‘remembered’ and ‘forgotten’ spectra were then calculated for each time bin , and the ratio estimated to give a measure of ‘remembered’:‘forgotten’ for both glutamate and GABA . For each time bin , data for a given participant was only included if the participant had more than eight spectra for both ‘remembered’ and ‘forgotten’ conditions for that bin . The number of participants per time bin therefore varied ( mean: n = 17 . 47; range: n = 12–19 ) . To visualize the time course of data acquired using fMRI , for each participant , and for each time bin during the inference test trial , the time course of the preprocessed BOLD signal was extracted from the hippocampal ROI ( Figure 5A ) and from two control ROIs defined using a 12 mm sphere within our partial epi volume ( Figure 3A ) . The first control region was positioned at the junction between parietal and occipital cortex ( ‘parietal-occipital cortex’ ) while the second control region was positioned within the brain stem ( Figure 5—figure supplement 1A , B ) . For each ROI , the obtained signal for each trial was resampled using a resolution of 400 ms and regressed against an EV indicating those trials that were ‘remembered’ . To control for differences in baseline BOLD at the start of the trial , we also included a ‘nuisance’ EV indicating whether the previous trial was ‘remembered’ . We then plotted the normalized averaged fMRI regression coefficient for ‘remembered’ versus ‘forgotten’ , using the time bins defined for the fMRS moving average ( described above ) ( Figure 5D; Figure 5—figure supplement 1A , B ) .	Memories are stored by distributed groups of neurons in the brain , with individual neurons contributing to multiple memories . In a part of the brain called the neocortex , memories are held in a silent state through a balance between excitatory and inhibitory activity . This is to prevent them from being disrupted by incoming information . When a memory is recalled , an area of the brain called the hippocampus is thought to instruct the neocortex to activate the appropriate neuronal network . But how the hippocampus and neocortex coordinate their activity to switch memories ‘on’ and ‘off’ is unclear . The answer may lie in the fact that neurons in the neocortex consist of two broad types: excitatory and inhibitory . Excitatory neurons increase the activity of other neurons . They do this by releasing a chemical called glutamate . Inhibitory neurons reduce the activity of other neurons , by releasing a chemical called GABA . Koolschijn , Shpektor et al . hypothesized that the hippocampus activates memories by changing the balance of excitatory and inhibitory activity in neocortex . To test this idea , Koolschijn , Shpektor et al . invited healthy volunteers to explore a virtual reality environment . The volunteers learned that specific sounds in the environment predicted the appearance of particular visual patterns . The next day , the volunteers returned to the environment and viewed these patterns again . After each pattern , they were invited to open a virtual box . Volunteers learned that some patterns led to money in the virtual box , while other patterns did not . Finally , on day three , the volunteers listened to the sounds from day one again , this time while lying in a brain scanner . The volunteers’ task was to infer whether each of the sounds would lead to money . Given that the sounds were never directly paired with the content of the virtual box , the volunteers had to solve the task by recalling the associated visual patterns . As they did so , the brain scanner measured their overall brain activity . It also assessed the relative levels of excitatory and inhibitory activity in visual areas of the neocortex , by measuring glutamate and GABA . The results revealed that as the volunteers recalled the visual cues , activity in both the hippocampus and the visual neocortex increased . Moreover , the ratio of glutamate to GABA in visual neocortex also increased which was predicted by activity in the hippocampus . This suggests that the hippocampus reactivates memories stored in neocortex by temporarily increasing excitatory activity to release memories from inhibitory control . Disturbances in the balance of excitation and inhibition occur in various neuropsychiatric disorders , including schizophrenia , autism , epilepsy and Tourette’s syndrome . Damage to the hippocampus is known to cause amnesia . The current findings suggest that memories may become inaccessible – or may be activated inappropriately – when the interaction between the hippocampus and neocortex goes awry . Future studies could test this possibility in clinical populations .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "neuroscience" ]	2021	Memory recall involves a transient break in excitatory-inhibitory balance
The FLT3 Internal Tandem Duplication ( FLT3ITD ) mutation is common in adult acute myeloid leukemia ( AML ) but rare in early childhood AML . It is not clear why this difference occurs . Here we show that Flt3ITD and cooperating Flt3ITD/Runx1 mutations cause hematopoietic stem cell depletion and myeloid progenitor expansion during adult but not fetal stages of murine development . In adult progenitors , FLT3ITD simultaneously induces self-renewal and myeloid commitment programs via STAT5-dependent and STAT5-independent mechanisms , respectively . While FLT3ITD can activate STAT5 signal transduction prior to birth , this signaling does not alter gene expression until hematopoietic progenitors transition from fetal to adult transcriptional states . Cooperative interactions between Flt3ITD and Runx1 mutations are also blunted in fetal/neonatal progenitors . Fetal/neonatal progenitors may therefore be protected from leukemic transformation because they are not competent to express FLT3ITD target genes . Changes in the transcriptional states of developing hematopoietic progenitors may generally shape the mutation spectra of human leukemias . Acute myeloid leukemia ( AML ) can occur at any stage of life yet the mutations that cause AML differ between childhood and adulthood , especially when one compares young children to adults ( Chaudhury et al . , 2015 ) . For example , MLL translocations and GATA1 mutations are common in infant and early childhood AML but rare in adult AML ( Andersson et al . , 2015; Horton et al . , 2013; Pine et al . , 2007 ) . Mutations in FLT3 , NPM1 , DNMT3A , TET2 and IDH1 are all common in adult AML but rare in infant and early childhood AML ( Cancer Genome Atlas Research Network , 2013; Ho et al . , 2011; Liang et al . , 2013; Zwaan et al . , 2003 ) . The genetic differences between pediatric and adult AML are not absolute , but they reflect a more general phenomenon in leukemia biology – leukemias in infants , young children , older children and adults have different genetic and epigenetic landscapes , different mechanisms of transformation and different clinical courses ( Downing and Shannon , 2002 ) . Efforts to interpret AML genomes and translate the information into useful therapies will need to account for the influences of age and developmental context on leukemia cell biology . This will require a better understanding of how normal developmental programs shape the process of leukemogenesis . The mutations that cause AML are thought to accrue first in pre-leukemic hematopoietic stem cells ( HSCs ) or committed hematopoietic progenitor cells ( HPCs ) ( Jan et al . , 2012; Welch et al . , 2012 ) , and several properties of these cells change between fetal and adult stages of life: ( 1 ) Fetal HSCs divide frequently and retain their self-renewal capacity through cumulative division cycles ( Pietras and Passegué , 2013 ) . In contrast , adult HSCs are usually quiescent , and self-renewal capacity declines with cumulative divisions ( Foudi et al . , 2009; Pietras and Passegué , 2013; Wilson et al . , 2008 ) . ( 2 ) Fetal and adult HSCs have distinct self-renewal mechanisms . For example , Sox17 is required for fetal , but not adult , HSC self-renewal ( Kim et al . , 2007 ) . Etv6 , Ash1l , Mll and Pten are all required for adult , but not fetal , HSC self-renewal ( Hock et al . , 2004; Jones et al . , 2015; Jude et al . , 2007; Magee et al . , 2012 ) . ( 3 ) Fetal and adult HSCs give rise to committed progenitors with distinct epigenetic landscapes ( Huang et al . , 2016; Xu et al . , 2012 ) and distinct lineage biases ( Benz et al . , 2012; Copley et al . , 2013; Yuan et al . , 2012 ) . These observations raise the question of whether mutations can have age-specific effects on gene expression , self-renewal , differentiation and ultimately leukemogenesis . If so , competence for transformation may be a heterochronic property of HSCs and HPCs , and this may explain why pediatric and adult leukemias have different mutations . The FLT3 Internal Tandem Duplication ( FLT3ITD ) is an example of an AML driver mutation that occurs more commonly in adults than in young children ( 30–40% of adult AML , 5–10% of AML in children <10 years old , <1% of infant AML ) ( Meshinchi et al . , 2006 ) . FLT3ITD encodes a constitutively active tyrosine kinase receptor that has been shown to activate the STAT5 , MAP-kinase ( MAPK ) , PI3-kinase ( PI3K ) , STAT3 and NF-κB signal transduction pathways in various contexts ( Choudhary et al . , 2007; Gerloff et al . , 2015; Radomska et al . , 2006 ) . Mice with a targeted Flt3ITD mutation develop myeloproliferative neoplasms ( MPN ) ( Lee et al . , 2007; Li et al . , 2008 ) , and several other mutations ( e . g . Npm1 , Tet2 and Runx1 mutations ) cooperate with Flt3ITD to drive AML in mice much as in humans ( Mead et al . , 2013; Mupo et al . , 2013; Rau et al . , 2014; Shih et al . , 2015 ) . In the absence of cooperating mutations , Flt3ITD drives adult HSCs into cycle and depletes the HSC pool ( Chu et al . , 2012 ) . This may explain why FLT3ITD mutations occur late in the clonal evolution of human AML — adult HSCs must first acquire mutations that preserve ( or ectopically establish ) self-renewal capacity in pre-leukemic progenitors—but it also raises the question of why fetal/neonatal HSCs , which have an inherently high self-renewal capacity ( He et al . , 2009 ) , do not give rise to FLT3ITD positive AML more often than is observed . To better understand how developmental context shapes myeloid leukemogenesis , we characterized the effects of Flt3ITD on HSC self-renewal , myelopoiesis , signal transduction and gene expression at several stages of pre- and post-natal development . Flt3ITD did not cause HSC depletion or myeloid progenitor expansion until after birth . This was true even in the presence of a cooperating Runx1 loss-of-function mutation . The FLT3ITD protein phosphorylated STAT5 during both pre- and post-natal stages of development while it hyper-activated the MAPK pathway only after birth . To our surprise , MAPK inhibition failed to rescue HSC depletion and myeloid progenitor expansion in adult Flt3ITD mice , and Stat5a/b deletion greatly exacerbated these phenotypes . FLT3ITD target genes , including STAT5 targets , were not induced in fetal HSCs or HPCs despite pre-natal STAT5 phosphorylation . Instead , FLT3ITD target gene activation coincided with a normal transition from fetal to adult gene expression that was evident by two weeks after birth . These temporal changes in FLT3ITD target gene expression were observed even in the setting of a cooperating Runx1 mutation . Our data establish a crucial role for developmental context in the pathogenesis of FLT3ITD-driven AML . Fetal and neonatal progenitors are protected from transformation because they are not competent to express FLT3ITD target genes . This likely explains why FLT3ITD mutations are more common in adults than young children , and it may reflect a more general role for developmental programming in leukemia pathogenesis . Since FLT3ITD occurs more commonly in adult AML patients than in young children , we hypothesized that it might have age-specific effects on self-renewal and myelopoiesis . We first tested whether Flt3 expression changes with age . We measured Flt3 transcript expression in CD150+CD48-Lineage-Sca1+c-kit+ HSCs and CD48+Lineage-Sca1+c-kit+ HPCs from 8–10 week old adult and embryonic day ( E ) 14 . 5 fetal mice by quantitative RT-PCR ( qRT-PCR ) . Flt3 was more highly expressed in HPCs than in HSCs at both ages ( Figure 1A ) , consistent with prior studies ( Buza-Vidas et al . , 2011 ) , but its expression did not change with age in either cell population ( Figure 1A ) . Flow cytometry confirmed that the FLT3 protein is expressed in both fetal and adult progenitors ( Figure 1B ) . 10 . 7554/eLife . 18882 . 003Figure 1 . Flt3ITD causes HSC depletion in adult but not fetal mice . ( A ) Flt3 transcript expression in fetal and adult HSCs and HPCs relative to fetal HSCs; n = 4–9 . ( B ) FLT3 expression in fetal and adult HSC/HPCs ( Lineage-Sca1+c-kit+ ) and unfractionated fetal liver or bone marrow cells , as determined by flow cytometry ( N = 3 ) . ( C ) HSC numbers in two tibias and femurs from adult wild type and Flt3ITD mice; n = 12–16 . ( D ) HSC numbers in fetal livers from E14 . 5 wild type and Flt3ITD mice; n = 9–20 . ( E ) Spleen HSC frequency in adult wild type and Flt3ITD/ITD mice; n = 4–5 . ( F , G ) Limiting dilution analyses using adult bone marrow ( F ) or E14 . 5 fetal liver cells ( G ) ; n = 9–10 recipients per cell dose . Wild type and Flt3ITD/ITD HSC frequencies were calculated by extreme limiting dilution analysis . ( H , I ) Frequencies of donor ( CD45 . 2 ) and competitor ( CD45 . 1 ) HSCs ( H ) and donor bone marrow cells ( I ) in primary recipients of 100 , 000 fetal liver cells; n = 15 per genotype . ( J ) Frequencies of CD45 . 2+ peripheral blood cells in secondary recipients of donor cells that originated from wild type or Flt3ITD/ITD fetal livers; n = 12–14 . ( K ) Percentage of secondary recipient mice with multilineage donor reconstitution . In all panels , error bars indicate standard deviations and n reflects biological replicates . p<0 . 05 , *p<0 . 001 by two-tailed Student’s t-test . ## p<0 . 01 by Fisher exact probability test . DOI: http://dx . doi . org/10 . 7554/eLife . 18882 . 003 Since Flt3ITD has previously been shown to deplete adult HSCs ( Chu et al . , 2012 ) , we tested whether the mutation has a similar effect on fetal HSC numbers . We measured HSC numbers in 8–10 week old adult bone marrow and E14 . 5 fetal livers from wild type , Flt3ITD/+ and Flt3ITD/ITD mice . Adult Flt3ITD/+ mice had ~50% fewer HSCs than wild type littermates , consistent with prior studies , and Flt3ITD/ITD mice had a near-complete loss of phenotypic HSCs ( Figure 1C ) . HSC depletion in the bone marrow was not accompanied by extramedullary expansion of HSCs in the spleen ( Figure 1E ) , in contrast to other leukemogenic mutations ( e . g . Pten deletion ) that cause depletion of bone marrow HSCs but marked expansion of the spleen HSC population ( Magee et al . , 2012; Porter et al . , 2016 ) . Unlike adult mice , Flt3ITD/+ and Flt3ITD/ITD fetal mice had similar numbers of HSCs as wild type littermates ( Figure 1D ) . The Flt3ITD mutation therefore depletes adult , but not fetal HSCs . We next tested whether fetal Flt3ITD/ITD HSCs are functionally impaired . We performed limiting dilution transplantation assays with either 8–10 week old adult bone marrow cells ( 600 , 000 , 100 , 000 , 50 , 000 or 10 , 000 CD45 . 2 donor cells competed with 300 , 000 CD45 . 1 adult bone marrow cells ) or E14 . 5 fetal liver cells ( 100 , 000 , 50 , 000 or 10 , 000 CD45 . 2 donor cells competed with 300 , 000 CD45 . 1 adult bone marrow cells ) . Two independent experiments were performed , and fetal and adult donor cells were transplanted at the same time in each experiment . Multi-lineage reconstitution was assessed every 4 weeks for 16 weeks following the transplants , and functional HSC frequencies were calculated by Extreme Limiting Dilution Analysis ( Hu and Smyth , 2009 ) . Adult Flt3ITD/ITD bone marrow had significantly fewer functional HSCs than adult wild type bone marrow ( p<0 . 00001 , Figure 1F ) . In contrast , wild type and Flt3ITD/ITD fetal livers had similar HSC frequencies ( Figure 1G ) . Our findings raised the question of whether fetal Flt3ITD/ITD HSCs can mature and become depleted after transplantation into adult recipient mice . To test this , we measured donor HSC chimerism in primary recipients of 100 , 000 wild type and Flt3ITD/ITD fetal liver cells ( from Figure 1G ) . Donor Flt3ITD/ITD HSCs were significantly depleted in primary recipient mice , but wild type competitor HSCs were not ( Figure 1H ) . Overall donor bone marrow chimerism was not significantly different between recipients of wild type and Flt3ITD/ITD fetal liver cells ( Figure 1I ) . Secondary transplants confirmed depletion of Flt3ITD/ITD HSCs in the marrow of primary recipients ( Figure 1J , K ) . Thus , fetal Flt3ITD/ITD HSCs are functional , but they lose repopulating activity after transplantation into adult recipient mice . We next sought to define the age at which Flt3ITD begins to deplete HSCs and expand myeloid progenitor populations . We measured HSCs , HPCs and granulocyte-monocyte progenitor ( GMP ) numbers at E14 . 5 , E16 . 5 , post-natal day ( P ) 0 and P14 . Flt3ITD/+and Flt3ITD/ITD mice had normal HSC numbers at all ages prior to birth ( Figure 2A ) . HSC depletion was evident at P14 in both the bone marrow and the spleen , though not to the extent observed in adult bone marrow ( Figure 2B ) . We observed a modest increase in Flt3ITD/+and Flt3ITD/ITD HPCs and GMPs at P0 ( Figure 2C , E ) , and this phenotype became more severe , particularly in Flt3ITD/ITD mice , by P14 ( Figure 2D , E ) . Spleen enlargement due to MPN was evident by P14 in Flt3ITD/+and Flt3ITD/ITD mice , but E14 . 5 , E16 . 5 and P0 liver sizes were not increased relative to wild type littermates ( Figure 2F ) . These data show that Flt3ITD causes HSC depletion , HPC/GMP expansion and MPN beginning at or shortly after birth . 10 . 7554/eLife . 18882 . 004Figure 2 . Flt3ITD causes HSC depletion , HPC expansion and GMP expansion at , or shortly after , birth . ( A ) Absolute HSC numbers in fetal or P0 livers for the indicated genotypes . ( B ) Absolute HSC numbers in P14 and adult bone marrow ( two hind limbs ) or P14 spleen . ( C , D ) Fetal and adult HPC numbers ( two hind limbs ) . ( E ) GMP frequencies in fetal liver or adult bone marrow . ( F ) Liver or spleen weights . In all panels , error bars indicate standard deviations; n = 6–20 biological replicates for each age and genotype . p<0 . 05; p<0 . 01; p<0 . 001 by two-tailed Student’s t-test relative to the wild type control at the same time point . DOI: http://dx . doi . org/10 . 7554/eLife . 18882 . 004 FLT3ITD mutations usually occur late during the clonal evolution of human AML . This raises the question of whether fetal/neonatal Flt3ITD mice can exhibit HSC depletion and HPC/GMP expansion when a cooperating mutation is present . To test this , we analyzed HSC , HPC and GMP frequencies in Flt3ITD/+; Runx1f/+; Vav1-Cre mice . Both mono- and bi-allelic RUNX1 loss-of-function mutations co-occur with FLT3ITD in human AML ( Schnittger et al . , 2011 ) , and Runx1 deletions synergize with Flt3ITD to cause AML in mice ( Mead et al . , 2013 ) . For the purposes of these studies , we focused on mono-allelic Runx1 deletions because bi-allelic deletions severely depleted phenotypic HSCs irrespective of the Flt3 genotype ( data not shown ) . These effects were likely due to previously described , Runx1-dependent changes in CD48 expression ( Cai et al . , 2011 ) . We evaluated HSC , HPC and GMP frequencies in ( 1 ) Flt3+/+; Runx1f/f or Runx1f/+; Cre-negative ( control ) , ( 2 ) Flt3ITD/+; Runx1f/f or Runx1f/+; Cre-negative ( Flt3ITD/+ ) , ( 3 ) Flt3+/+;Runx1f/+; Vav1-Cre ( Runx1Δ/+ ) and ( 4 ) Flt3ITD/+;Runx1f/+; Vav1-Cre ( Flt3ITD/+; Runx1Δ/+ ) littermates at E14 . 5 , P0 , P14 and P21 . HSCs were severely depleted in P14 and P21 Flt3ITD/+; Runx1Δ/+ mice relative to controls and single mutant mice ( Figure 3C , D ) . In contrast , all four genotypes of mice had similar HSC frequencies at P0 ( Figure 3B ) , and Runx1 heterozygosity increased HSC frequency at E14 . 5 irrespective of the Flt3 genotype ( Figure 3A ) . HPCs and GMPs were markedly expanded in P14 and P21 Flt3ITD/+; Runx1Δ/+ mice ( Figure 3G , H , K , L ) . These populations were only modestly expanded in compound mutant mice at P0 ( Figure 3F , J ) , and they were not expanded at all at E14 . 5 ( Figure 3E , I ) . 10 . 7554/eLife . 18882 . 005Figure 3 . Flt3ITD and Runx1 mutations cooperate to deplete HSCs and expand committed progenitor populations after birth . ( A–D ) HSC frequencies in E14 . 5 fetal liver , P0 liver , P14 bone marrow and P21 bone marrow for the indicated genotypes . ( E–H ) HPC frequencies in E14 . 5 fetal liver , P0 liver , P14 bone marrow and P21 bone marrow for the indicated genotypes . ( I–L ) GMP frequencies in E14 . 5 fetal liver , P0 liver , P14 bone marrow and P21 bone marrow for the indicated genotypes . ( M , N ) Percentages of CD45 . 2+ donor leukocytes ( M ) or CD11b+Gr1+ myeloid cells ( N ) in the peripheral blood of recipients of P0 liver or P21 bone marrow cells from control or Flt3ITD/+; Runx1Δ/+ mice . Measurements are shown at 2 and 4 weeks after transplantation . ( O , P ) Percentage of recipients with multi-lineage ( O ) or myeloid ( P ) donor reconstitution at four weeks after transplantation . In all panels , error bars indicate standard deviations . For A-L , n = 8–18 biological replicates per genotype and age . For ( M–P ) , n = 14–15 recipients from three independent donors . Statistical significance was determined with a one-way ANOVA followed by Holm-Sidak’s post-hoc test for multiple comparisons ( p<0 . 05; p<0 . 01; p<0 . 001 ) , or # p<0 . 0001 by the Fisher exact probability test . DOI: http://dx . doi . org/10 . 7554/eLife . 18882 . 005 We next tested whether compound Flt3ITD and Runx1 mutations had age-specific effects on HSC/HPC function . We transplanted 100 , 000 P0 liver cells or P21 bone marrow cells from control or Flt3ITD/+; Runx1Δ/+ littermate donors , along with 300 , 000 wild type competitor cells , into irradiated CD45 . 1 recipient mice . At two weeks after the transplants , we observed CD45 . 2+ donor-derived leukocytes in the peripheral blood of all recipients , irrespective of donor age or genotype ( Figure 3M , N ) . At four weeks after the transplants , donor chimerism was significantly and dramatically reduced in recipients of Flt3ITD/+; Runx1Δ/+ P21 donor cells as compared to recipients of control P21 donor cells and Flt3ITD/+; Runx1Δ/+ P0 donor cells ( Figure 3M ) . Indeed , only 1 of 15 recipients of Flt3ITD/+; Runx1Δ/+ P21 donor cells had multi-lineage donor chimerism ( >0 . 5% CD45 . 2+ myeloid and lymphoid cells ) ( Figure 3O ) . In contrast , all recipients of control and Flt3ITD/+; Runx1Δ/+ P0 donor cells had multi-lineage donor chimerism ( Figure 3O ) . These differences were evident even when we focused specifically on myeloid chimerism ( Figure 3N , P ) , so they were not simply a reflection of altered lineage biases in the Flt3ITD/+; Runx1Δ/+ progenitors . Altogether , these data show that Flt3ITD has developmental context-specific effects on HSC depletion , myeloid progenitor expansion and repopulating activity , even when paired with a cooperating Runx1 mutation . To better understand why Flt3ITD has developmental context-specific effects on HSCs and HPCs , we sought to better characterize the pathways that mediate FLT3ITD signal transduction in vivo . We isolated 25 , 000 HSC/multipotent progenitors ( HSC/MPPs; CD48-Lineage-Sca1+c-kit+ ) , HPCs and GMPs from adult mice by flow cytometry , and we performed Western blots to assess phosphorylation of STAT5 , STAT3 , ERK1/2 ( a MAPK pathway protein ) and AKT ( a PI3K pathway protein ) . Both STAT5 and ERK1/2 were hyper-phosphorylated in Flt3ITD HSC/MPPs and HPCs , as well as in adult Flt3ITD; Runx1Δ/Δ AML cells ( Figure 4A–C ) . In contrast , STAT3 and AKT were not hyper-phosphorylated in Flt3ITD mutant HSC/MPPs or HPCs ( Figure 4A and Figure 4—figure supplement 1A ) , and Rictor deletion ( PI3K/mTORC2 pathway inactivation ) did not rescue Flt3ITD-driven HSC depletion or MPN ( Figure 4—figure supplement 1B , C ) . These findings suggest that the STAT5 and MAPK pathways mediate FLT3ITD signal transduction in hematopoietic progenitors , but the STAT3 and PI3K pathways do not . 10 . 7554/eLife . 18882 . 006Figure 4 . FLT3ITD activates STAT5 in both fetal and adult progenitors , but it activates the MAPK pathway after birth . ( A ) Western blot showing phosphorylation of STAT5 , ERK1/2 , STAT3 and AKT in adult wild type and Flt3ITD/+ HSC/MPPs , HPCs and GMPs . ( B ) Flt3ITD/+; Runx1Δ/Δ progenitors give rise to AML in adult mice ( right panel ) that is not observed in wild type or Flt3ITD/+ bone marrow . Scale bars indicate 100 microns . ( C ) STAT5 and MAPK are hyper-phosphorylated in Flt3ITD/+; Runx1Δ/Δ AML that develops in adult mice . ( D ) STAT5 and ERK1/2 phosphorylation in wild type and Flt3ITD/+ HSC/MPPs at E14 . 5 , P0 , P14 and adulthood . ( E ) STAT5 phosphorylation in wild type and Flt3ITD/+ HPCs at E14 . 5 , P0 , P14 and adulthood . ( F ) ERK1/2 phosphorylation in wild type and Flt3ITD/+ HPCs at E14 . 5 , P0 , P14 and adulthood . Each blot is representative of two ( panels A and C ) or at least three ( panels D–F ) independent experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 18882 . 00610 . 7554/eLife . 18882 . 007Figure 4—figure supplement 1 . The PI3K/mTORC2 pathway does not mediate HSC depletion or MPN in Flt3ITD mutant mice . ( A ) Western blot showing STAT5 and AKT phosphorylation in wild type and Flt3ITD/ITD HPCs . Pten-deficient HPCs were included as a positive control for AKT phosphorylation . Rictor deletion prevented AKT hyper-phosphorylation by mTORC2 . ( B , C ) HSC numbers in two tibias and femurs ( B ) and spleen weights ( C ) in Flt3ITD/ITD; Rictorf/f; Vav1-Cre compound mutant mice ( and the indicated controls ) after conditional Rictor deletion; n = 3–12 per genotype . Error bars indicate standard deviations . DOI: http://dx . doi . org/10 . 7554/eLife . 18882 . 007 We next tested whether FLT3ITD signal transduction changes between fetal and adult stages of development . We isolated HSC/MPPs and HPCs from wild type and Flt3ITD mice at E14 . 5 , P0 , P14 and eight weeks after birth . We performed Western blots to assess STAT5 and ERK1/2 phosphorylation . STAT5 was hyper-phosphorylated in Flt3ITD mutant HSC/MPPs and HPCs at all stages of development , though the degree of STAT5 phosphorylation appeared to increase with age ( Figure 4D , E ) . ERK1/2 was only hyper-phosphorylated in post-natal Flt3ITD mutant HSC/MPPs and HPCs ( Figure 4D , F ) . Several other signal transduction proteins , including STAT3 , AKT , ribosomal protein S6 , p38 and JNK , were not hyper-phosphorylated in Flt3ITD HSC/MPPs or HPCs at any age tested , or their phosphorylation was undetectable ( data not shown ) . Our data reinforce other studies that have implicated STAT5 and MAPK as key downstream effectors of FLT3ITD signaling ( Choudhary et al . , 2007; Radomska et al . , 2006 ) . However , the data suggest that these pathways are not coupled — STAT5 is phosphorylated in fetal progenitors without concurrent MAPK pathway activation . This raises the question of whether each pathway has unique functions downstream of FLT3ITD . We used the MEK inhibitor PD0325901 to test whether MAPK pathway inhibition could prevent HSC depletion and HPC/GMP expansion in Flt3ITD mice . We administered vehicle or PD0325901 to 6-week-old wild type and Flt3ITD/+ mice ( 5 mg/kg per day for 10 days ) . This regimen effectively inhibited ERK1/2 phosphorylation in HPCs without affecting STAT5 phosphorylation ( Figure 5—figure supplement 1 ) . PD0325901-treated wild type mice had significantly more phenotypic HSCs and HPCs than vehicle treated controls ( Figure 5A , B ) . However , PD0325901 had no effect on HSC numbers , HPC numbers or GMP frequencies in Flt3ITD/+ mice ( Figure 5A–C ) . This suggests that sustained MAPK pathway signaling is not required for HSC depletion , HPC expansion and GMP expansion in Flt3ITD/+ adult mice . 10 . 7554/eLife . 18882 . 008Figure 5 . MAPK pathway inhibition does not prevent HSC depletion or committed progenitor expansion in Flt3ITD/+ mice . ( A–D ) HSC numbers ( A ) , HPC numbers ( B ) and GMP frequencies ( C ) in wild type and Flt3ITD/+ mice that were treated with vehicle or PD0325901 for 10 days beginning at six weeks after birth; n = 4–5 biological replicates per genotype and treatment . ( D–F ) HSC numbers ( D ) , HPC numbers ( E ) and GMP frequencies ( F ) in P19 wild type , Flt3ITD/+ and Flt3ITD/ITD mice whose mothers were given PD0325901 beginning at P1; n = 4–15 biological replicates for each genotype and treatment . In all panels , error bars indicate standard deviation . Statistical comparisons were made with a two-tailed Student’s t-test . p<0 . 05 relative to vehicle treated cells with equivalent genotypes; # p<0 . 05 relative to similarly treated wild type controls; ^ p<0 . 05 relative to similarly treated wild type and Flt3ITD/+ groups . DOI: http://dx . doi . org/10 . 7554/eLife . 18882 . 00810 . 7554/eLife . 18882 . 009Figure 5—figure supplement 1 . Inhibition of the MAPK pathway fails to rescue FLT3ITD-mediated HSC depletion and myeloid progenitor expansion , but Stat5a/b deletion enhances these phenotypes . A Western blot was performed with 25 , 000 HPCs from wild type and Flt3ITD/+ , vehicle and PD0325901 treated mice . PD0325901 inhibited ERK1/2 phosphorylation without affecting STAT5 phosphorylation . DOI: http://dx . doi . org/10 . 7554/eLife . 18882 . 009 We next tested whether PD0325901 could prevent the onset of the HSC depletion , HPC expansion and GMP expansion phenotypes if it was given shortly after birth . We treated nursing mothers of wild type , Flt3ITD/+ and Flt3ITD/ITD neonates with PD0325901 ( 5 mg/kg per day ) beginning at P1 . While this regimen has previously been shown to rescue MAPK pathway-dependent developmental abnormalities in Nf1 mutant neonates ( Wang et al . , 2012 ) , it did not prevent HSC depletion or HPC expansion in Flt3ITD mutant neonates ( Figure 5D , E ) , and it only partially rescued GMP expansion ( Figure 5F ) . Altogether , the data suggest that the MAPK pathway has only a minor role , if any , in causing these phenotypes . Temporal changes in MAPK pathway regulation are unlikely to account for the different effects of FLT3ITD on fetal and adult progenitors . STAT5 has been implicated as a key downstream effector of FLT3ITD in many different systems , and it is hyper-phosphorylated in Flt3ITD/+ HSCs and HPCs during fetal , neonatal and adult stages of development ( Figure 4E , F ) . This raised the question of whether genetic inactivation of Stat5a and Stat5b – with a conditional Stat5a/b allele ( Wang et al . , 2009 ) — could prevent HSC depletion , HPC expansion , GMP expansion and MPN in Flt3ITD/+ mice . To answer this question , we evaluated HSCs , HPCs , GMPs and spleen weights in 1 ) Flt3+/+; Stat5a/bf/+ or Stat5a/bf/f; Cre- ( control ) , 2 ) Flt3+/+; Stat5a/bf/+; Mx1-Cre ( Stat5Δ/+ ) , 3 ) Flt3+/+; Stat5a/bf/f; Mx1-Cre ( Stat5Δ/Δ ) , 4 ) Flt3ITD/+; Stat5a/bf/+ or Stat5a/bf/f; Cre- ( Flt3ITD/+ ) , 5 ) Flt3ITD/+; Stat5a/bf/f; Mx1-Cre ( Flt3ITD; Stat5Δ/+ ) , and 6 ) Flt3ITD/+; Stat5a/bf/f; Mx1-Cre ( Flt3ITD/+; Stat5Δ/Δ ) mice . The mice were treated with poly-inosine:poly-cytosine ( pIpC ) beginning at six weeks after birth to delete Stat5a/b , and they were analyzed four weeks later . Western blotting confirmed complete loss of STAT5 protein , and MAPK pathway activation was unaffected by Stat5a/b deletion ( Figure 6—figure supplement 1 ) . Surprisingly , Stat5a/b deletion exacerbated the HSC depletion , HPC expansion and GMP expansion phenotypes of Flt3ITD/+mice rather than rescuing them ( Figure 6A–C ) . Spleen weights were also enlarged in Flt3ITD; Stat5Δ/+ and Flt3ITD/+; Stat5Δ/Δ mice relative to control , Stat5Δ/+ , Stat5Δ/Δ and Flt3ITD/+ littermates ( Figure 5D ) . Similar results were observed when we deleted a single Stat5a/b with Vav1-cre . Only one Stat5a/b allele was deleted in these analyses because bi-allelic deletion impairs fetal erythropoiesis ( Zhu et al . , 2008 ) . Nevertheless , Flt3ITD/+; Stat5a/bf/+; Vav1-Cre mice had fewer HSCs , more HPCs , more GMPs and larger spleens than control or Flt3ITD/+ littermates at 8–10 weeks after birth ( Figure 6E–H ) . 10 . 7554/eLife . 18882 . 010Figure 6 . Stat5a/b deletion exacerbates rather than rescues HSC depletion , HPC expansion , GMP expansion and MPN in Flt3ITD/+ mice . ( A–D ) HSC numbers ( A ) , HPC numbers ( B ) , GMP frequencies ( C ) and spleen weights ( D ) in Flt3ITD/+; Stat5a/bf/f; Mx1-Cre compound mutant mice and littermate controls; n = 6–20 biological replicates per genotype . Stat5a/b was conditionally deleted six weeks after birth , and this caused a complete loss of protein expression ( figure supplement 1 ) . ( E–H ) HSC numbers ( E ) , HPC numbers ( F ) , GMP frequencies ( G ) and spleen weights ( H ) in Flt3ITD/+; Stat5a/bf/+; Vav1-Cre compound mutant mice and littermate controls; n = 8–20 biological replicates per genotype . ( I ) The data suggest that STAT5-dependent pathways promote HSC self-renewal downstream of FLT3ITD , but these effects are outweighed by STAT5-independent myeloid commitment pathways . In all panels , error bars indicate standard deviation . Statistical significance was determined with a one-way ANOVA followed by Holm-Sidak’s post-hoc test for multiple comparisons . p<0 . 05; p<0 . 01; p<0 . 001 . DOI: http://dx . doi . org/10 . 7554/eLife . 18882 . 01010 . 7554/eLife . 18882 . 011Figure 6—figure supplement 1 . Stat5a/b deletion causes a complete loss of phosphorylated and total STAT5 protein . A Western blot was performed with 25 , 000 HPCs from wild type , Flt3ITD/+ and Flt3ITD/+; Stat5Δ/Δ adult mice ( representative of two independent experiments ) . ERK1/2 phosphorylation was not affected by STAT5 deletion . STAT3 phosphorylation was not affected , and STAT1 and AKT phosphorylation were not detectable ( not shown ) . DOI: http://dx . doi . org/10 . 7554/eLife . 18882 . 011 The data reveal an unanticipated function for STAT5 in pre-leukemic , Flt3ITD-mutant progenitors . They suggest that STAT5 helps to maintain Flt3ITD-mutant HSCs in an uncommitted state and that it antagonizes Flt3ITD-driven expansion of more committed myeloid progenitor populations . Thus , FLT3ITD may simultaneously potentiate self-renewal and myeloid commitment programs via STAT5-dependent and STAT5-independent pathways , respectively ( Figure 6I ) . The changes in HSC and HPC frequencies in Flt3ITD; Stat5a/b compound mutant mice raise the question of whether FLT3ITD has STAT5-dependent and STAT5-independent effects on gene expression , and whether transcriptional changes are developmental context-specific . To answer these questions we performed two independent experiments to characterize global changes in gene expression ( Figure 7A ) . In the first experiment , we analyzed gene expression in wild type and Flt3ITD/+ HSCs and HPCs at E14 . 5 , P0 , P14 and 8–10 weeks after birth . This experiment was meant to elucidate changes in FLT3ITD target genes over time . In the second experiment , we analyzed gene expression in adult HPCs from ( 1 ) wild type , ( 2 ) Flt3ITD/+ , ( 3 ) Flt3ITD/ITD , ( 4 ) Flt3ITD; Stat5Δ/+ and 5 ) Flt3ITD/+; Stat5Δ/Δ mice . This experiment was meant to delineate which FLT3ITD targets are STAT5-dependent and which are STAT5-independent . 10 . 7554/eLife . 18882 . 012Figure 7 . FLT3ITD activates STAT5-dependent self-renewal programs and STAT5-independent commitment programs . ( A ) Overview of experimental design . ( B ) Heatmap representing genes that were differentially expressed in Flt3ITD mutant HPCs relative to wild type HPCs in both experiments 1 and 2 . Each column represents an independent sample . The gene names and dendrogram are shown in Figure 7—figure supplement 1 attached to this figure . ( C ) Self-renewal and commitment-related gene sets were generated by identifying genes that were more highly expressed ( >5 fold , adj . p<0 . 05 ) in HSCs relative to HPCs ( self-renewal ) , or HPCs relative to HSCs ( commitment ) . GSEA plots show ectopic activation of self-renewal-related genes in HPCs that express FLT3ITD , but these effects are reversed in Stat5a/b-deficient HPCs . ( D ) An independently curated self-renewal gene set ( Ivanova et al . , 2002 ) was similarly enriched in wild type and Flt3ITD/+ HPCs relative to Flt3ITD/+; Stat5Δ/Δ HPCs . ( E ) GSEA revealed enrichment of gene sets associated with increased inflammatory cytokine signaling . DOI: http://dx . doi . org/10 . 7554/eLife . 18882 . 01210 . 7554/eLife . 18882 . 013Figure 7—source data 1 . Significantly differentially expressed genes in Flt3ITD/+ HSCs and HPCs . Fold changes are shown for 270 probes that represent 254 differentially expressed genes in Flt3ITD/+ HPCs relative to controls . HSC fold change data are also shown . P-values and adjusted p-values reflect significance across the entire time course . DOI: http://dx . doi . org/10 . 7554/eLife . 18882 . 01310 . 7554/eLife . 18882 . 014Figure 7—source data 2 . Self-renewal-related and commitment-related gene sets . These gene sets were determined by identifying genes that were more highly expressed in HSCs relative to HPCs ( self-renewal ) or HPCs relative to HSCs ( commitment ) . In each case , a fold change threshold of five and an adjusted p-value threshold of <0 . 05 were used to define the lists . DOI: http://dx . doi . org/10 . 7554/eLife . 18882 . 01410 . 7554/eLife . 18882 . 015Figure 7—figure supplement 1 . FLT3ITD induces STAT5-dependent and STAT5-independent changes in gene expression . An expanded version of Figure 7B with the dendrogram and gene names attached to the heatmap . DOI: http://dx . doi . org/10 . 7554/eLife . 18882 . 015 We analyzed the data from each experiment independently , and we merged the data to identify a list of genes that were significantly , differentially expressed in FLT3ITD progenitors in both experiments . In experiment 1 , we identified 254 annotated coding genes that were differentially expressed between wild type and Flt3ITD/+ HPCs at one or more time points ( Figure 7—source data 1; adjusted p<0 . 05; fold change ≥ 2 ) . We did not identify any genes that met these stringent filtering criteria in HSCs , though statistically significant changes in gene expression were observed when specific target genes ( from the HPC list ) were individually interrogated ( p<0 . 05 , Figure 7—source data 1 ) . The differences between HSCs and HPCs may simply reflect differences in Flt3 expression ( Figure 1A ) , though it is also possible that HSCs with the strongest transcriptional responses to FLT3ITD were not captured in our microarray assays because they differentiated . Of the 254 genes that were differentially expressed in experiment 1 , 58 unique genes were also differentially expressed between wild type and Flt3ITD/+ HPCs in experiment 2 ( Figure 7B and Figure 7—figure supplement 1 ) . Thirty-three genes were expressed at higher levels in Flt3ITD HPCs relative to wild type HPCs , and 25 genes were expressed at lower levels ( Figure 7B ) . Of these , 35 normalized when Stat5a/b was deleted , but 23 did not . FLT3ITD therefore has both STAT5-dependent and STAT5-independent effects on gene expression , and these effects are more pronounced in HPCs as compared to HSCs . We tested whether FLT3ITD activates self-renewal- and commitment-related transcriptional programs via STAT5-dependent and STAT5-independent mechanisms , respectively , as predicted by our phenotypic assays ( Figure 6 ) . We generated self-renewal-related and commitment-related gene sets by comparing wild type HSCs and HPCs using the data collected in experiment 1 ( Figure 7—source data 2 ) . We then used Gene Set Enrichment Analysis ( GSEA ) to compare wild type , Flt3ITD/+ and Flt3ITD/+; Stat5Δ/Δ HPCs using data collected in experiment 2 ( Subramanian et al . , 2005 ) . Self-renewal-related genes were enriched in Flt3ITD/+ HPCs , and commitment-related genes were enriched in wild type HPCs ( Figure 7C ) . This suggests that the FLT3ITD protein can activate self-renewal mechanisms in otherwise non-self-renewing HPCs . Remarkably , these effects were strongly reversed when Stat5a/b was deleted ( Figure 7C ) . A separately curated self-renewal gene set from Ivanova et al . was similarly enriched in wild type and Flt3ITD/+ HPCs as compared to Flt3ITD/+; Stat5Δ/Δ HPCs ( Figure 7D ) ( Ivanova et al . , 2002 ) . These findings are consistent with a model in which FLT3ITD signals via STAT5 to ectopically activate self-renewal programs in HPCs , and it simultaneously promotes myeloid commitment via STAT5-independent mechanisms . To better understand the STAT5-independent mechanisms that promote myeloid commitment , we performed GSEA on Flt3ITD/+ and Flt3ITD/+; Stat5Δ/Δ HPCs with curated gene sets in the MSigDB database ( Subramanian et al . , 2005 ) . The most significantly enriched gene sets in Flt3ITD/+; Stat5Δ/Δ HPCs were generally associated with increased inflammatory cytokine signaling ( Figure 7E ) . This finding is intriguing because several prior studies have linked inflammatory cytokine signaling to loss of adult HSC self-renewal capacity and myeloid differentiation ( Baldridge et al . , 2010; Essers et al . , 2009; Pietras et al . , 2016 ) . Of note , we did not observe changes in STAT1 , STAT3 or AKT phosphorylation in Flt3ITD/+; Stat5Δ/Δ HPCs ( Figure 6—figure supplement 1 and data not shown ) . Additional studies are still needed to identify the signal transduction molecules that mediate FLT3ITD-driven myeloid commitment , and to test whether changes in cytokine-related gene expression are a cause or a consequence of differentiation in Flt3ITD/+; Stat5Δ/Δ HPCs . FLT3ITD has the capacity to activate functionally relevant signal transduction pathways , such as STAT5 , during both pre- and post-natal stages of development ( Figure 4 ) , yet HSC and HPC phenotypes were only observed after birth . This raises the question of whether pre- and post-natal progenitors have distinct transcriptional responses to FLT3ITD . We evaluated FLT3ITD target gene expression in HPCs at E14 . 5 , P0 , P14 and adulthood ( Figure 8A ) . We found that differences between wild type and Flt3ITD/+ HPCs were more evident at P14 and adulthood than at E14 . 5 or P0 ( Figure 8A and Figure 8—figure supplement 1 ) . This was true for both STAT5-dependent targets , e . g . Socs2 , and STAT5-independent targets , e . g . Ctsg ( Figure 8B ) . Thus , fetal , neonatal and adult hematopoietic progenitors have distinct transcriptional responses to FLT3ITD signaling . 10 . 7554/eLife . 18882 . 016Figure 8 . Flt3ITD-mediated changes in gene expression correlate with the normal transition from fetal to adult transcriptional programs . ( A ) Heatmap showing expression of FLT3ITD target genes at E14 . 5 , P0 , P14 and adult stages . Each column shows average fold change in Flt3ITD/+ HPCs relative wild type HPCs at the indicated time point; n = 3–4 independent arrays per genotype . The gene names and dendrogram are shown in figure supplement one attached to this figure . ( B ) Representative examples of expression of STAT5-independent ( Ctsg ) and STAT5-dependent ( Socs2 ) FLT3ITD targets . Error bars reflect standard deviation . adj . p<0 . 05 relative to wild type at the same time point , # adj . p<0 . 05 relative to Flt3ITD/+ at the same time point . ( C ) Heterochronic genes began transitioning from fetal to adult expression patterns between P0 and P14 , concordant with sensitivity to FLT3ITD . Genes that encode transcription factors and RNA binding proteins are noted to the right of the heatmap . A complete gene list is provided in Figure 8—source data 1 attached to this figure . ( D ) Representative examples of heterochronic genes that show decreased ( Lin28b ) or increased ( Esr1; Estrogen Receptor α ) expression in adult relative to fetal HSCs and HPCs . Error bars reflect standard deviation . $$ adj . p<0 . 05 relative to E14 . 5 for both HSCs and HPCs . ( E ) Principal component analysis and Euclidean distance measurements show that gene expression in P0 HSCs more closely resembles fetal HSCs than adult HSCs , and gene expression in P14 HSCs more closely resembles that of adult HSCs . Similar calculations for HPCs are shown in figure supplement 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 18882 . 01610 . 7554/eLife . 18882 . 017Figure 8—source data 1 . Heterochronic gene expression in wild type HSCs and HPCs . Source data ( log2 ) are shown for the 250 most differentially expressed probes in developing HSCs ( 228 unique genes ) . DOI: http://dx . doi . org/10 . 7554/eLife . 18882 . 01710 . 7554/eLife . 18882 . 018Figure 8—figure supplement 1 . FLT3ITD-mediated changes in gene expression correlate temporally with a transition from fetal to adult transcriptional states . An expanded version of Figure 8A with the dendrogram and gene names attached to the heatmap . DOI: http://dx . doi . org/10 . 7554/eLife . 18882 . 01810 . 7554/eLife . 18882 . 019Figure 8—figure supplement 2 . HPCs express heterochronic genes and begin to transition from fetal to adult transcriptional programs by P14 . Principal component analysis shows temporal changes in HPC gene expression before and after birth . Euclidean distance measurements and permutation testing show that P0 HPCs more closely resemble fetal HPCs than adult HPCs . P14 HPCs more closely resemble adult HPCs . DOI: http://dx . doi . org/10 . 7554/eLife . 18882 . 019 To better understand when HSCs and HPCs transition from fetal to adult transcriptional programs , we analyzed gene expression in wild type cells from experiment 1 . We identified 2627 unique genes ( from 3005 different probes ) that exhibited significant changes in gene expression in HSCs between E14 . 5 and adulthood . Of the 228 most differentially expressed transcripts ( from the top 250 probes ) , all followed a consistent trend toward increasing ( 109 genes ) or decreasing ( 119 genes ) expression with increasing age , and most were differentially expressed in both HSCs and HPCs ( Figure 8C , Figure 8—source data 1 ) . Among these genes were several that encode transcription factors and RNA binding proteins that are known to regulate HSC self-renewal , including Lin28b , Esr1 , Hmga2 and Egr1 ( Figure 8C , D ) . Principal component analysis and Euclidean distance measurements showed that P14 HSCs more closely resembled adult HSCs than fetal HSCs , and P0 HSCs more closely resembled fetal HSCs ( Figure 8E ) . Similar associations were observed for HPCs ( Figure 8—figure supplement 2 ) . The data show that HSCs and HPCs begin transitioning from fetal to adult transcriptional programs by P14 , even before they achieve quiescence ( Bowie et al . , 2006 ) . Furthermore , the data suggest that HSCs and HPCs become competent to express ( or repress ) FLT3ITD target genes as they transition from fetal to adult transcriptional programs . Flt3ITD and Tet2 loss of function mutations have recently been shown to cooperatively induce changes in gene expression and DNA methylation in adult HPCs that are not observed with either mutation alone ( Shih et al . , 2015 ) . We tested whether Flt3ITD and Runx1 mutations have similar cooperative effects on transcription and whether the effects are age-specific . We evaluated gene expression in ( 1 ) wild type , ( 2 ) Runx1Δ/+ , ( 3 ) Flt3ITD/+ and ( 4 ) Flt3ITD/+; Runx1Δ/+ HPCs at P0 and P14 . At P14 , we identified 191 genes that were significantly differentially expressed between wild type and Flt3ITD/+; Runx1Δ/+ HPCs ( adj . p<0 . 05; fold change ≥ 3 ) . At P0 only eight genes met these criteria , seven of which overlapped with the P14 gene list ( Figure 9A and Figure 9—source data 1 ) . GSEA showed significant overlap between genes that were differentially expressed in Flt3ITD/+; Runx1Δ/+ HPCs and those that Shih et al . found to be differentially expressed in Flt3ITD; Tet2Δ/Δ HPCs ( Shih et al . , 2015 ) ( Figure 9B ) . This suggests that FLT3ITD can cooperate with diverse mutations to induce a conserved set of target genes . 10 . 7554/eLife . 18882 . 020Figure 9 . Flt3ITD and Runx1 mutations cooperatively induce changes in gene expression at P14 , yet they have a much smaller effect at P0 . ( A ) Venn diagram showing overlap between genes that were significantly differentially expressed ( adj . p<0 . 05 , fold change ≥ 3 ) in Flt3ITD/+; Runx1Δ/+ HPCs relative to wild type at P14 or P0; n = 4–5 arrays per genotype and age . ( B ) GSEA shows that differentially expressed genes in Flt3ITD/+; Runx1Δ/+ HPCs overlap significantly with genes that are differentially expressed in Flt3ITD/+; Tet2Δ/Δ HPCs ( Shih et al . , 2015 ) . ( C ) Heatmap showing expression of genes that were differentially expressed in Flt3ITD/+; Runx1Δ/+ HPCs relative to wild type HPCs . Each column indicates the average fold change relative to the wild type samples from the same time point . The gene list is shown in Figure 9—source data1 attached to this figure . ( D ) Representative examples of genes that are among the most differentially expressed in Flt3ITD/+; Runx1Δ/+ HPCs relative to wild type HPCs at P14 . Most show much smaller changes in expression at P0 . Error bars reflect standard deviations , adj . p<0 . 05 . ( E ) GSEA identified several gene sets that were enriched in Flt3ITD/+; Runx1Δ/+ HPCs relative to wild type HPCs at P14 . Three of the most significantly enriched gene sets are shown for P14 and P0 . DOI: http://dx . doi . org/10 . 7554/eLife . 18882 . 02010 . 7554/eLife . 18882 . 021Figure 9—source data 1 . Flt3ITD and Runx1 mutations cooperatively induce changes in gene expression in post-natal HPCs . Genes that are differentially expressed ( Adjusted p<0 . 05 , fold change ≥ 3 ) in Flt3ITD/+; Runx1Δ/+ HPCs at P0 and P14 are shown in the indicated tables . DOI: http://dx . doi . org/10 . 7554/eLife . 18882 . 021 We used hierarchical clustering to better visualize differences in gene expression at P0 and P14 . This approach did show some differences between wild type and Flt3ITD/+; Runx1Δ/+ HPCs at P0 ( Figure 9C ) , but much greater differences were observed at P14 for most target genes ( Figure 9C , D ) . Of the 25 most differentially expressed genes in P14 Flt3ITD/+; Runx1Δ/+ HPCs , only three – Nov , Bhlhe40 and Mboat2 – were induced equally at both P0 and P14 ( Figure 9D , Figure 9—source data 1 ) . The remaining genes showed only a partial change in expression at P0 ( e . g . Socs2 ) or no change in expression ( e . g . Adgre1 , Dusp6 , Gem , Postn ) ( Figure 9D ) . GSEA also showed differences in the transcriptional programs of P0 and P14 Flt3ITD/+; Runx1Δ/+ HPCs . Several Gene Ontology and Oncogenic Signature gene sets were enriched in P14 Flt3ITD/+; Runx1Δ/+ HPCs relative to wild type controls , the most significant of which included genes that are negatively regulated by mTOR ( Majumder et al . , 2004 ) , genes that inhibit apoptosis ( Gene Ontology ) and c-myc targets ( Bild et al . , 2006 ) ( Figure 9E ) . These gene sets were not significantly enriched in P0 HSCs with the exception of the anti-apoptotic gene set , which was paradoxically enriched in wild type HPCs relative to Flt3ITD/+; Runx1Δ/+ ( Figure 9E ) . Altogether , the data show that cooperating Flt3ITD and Runx1 mutations — and likely other cooperating mutations — have developmental context-specific effects on gene expression . In the course of these studies , we discovered that Stat5a/b deletion exacerbates , rather than rescues , the HSC depletion , HPC/GMP expansion and MPN phenotypes of Flt3ITD/+ mice . Furthermore , FLT3ITD ectopically activated STAT5-dependent self-renewal programs in HPCs ( Figure 7 ) . These unanticipated findings suggest that FLT3ITD activates both STAT5-dependent and STAT5-independent signal transduction pathways and that these pathways have opposing effects on self-renewal and myeloid commitment . In this model , STAT5-independent myeloid commitment programs outweigh STAT5-dependent self-renewal programs in the Flt3ITD mutant bone marrow so the HSC pool becomes depleted ( Figure 6I and Figure 7 ) . Stat5a/b deletion can shift the balance further in favor of differentiation , though cooperating mutations may ultimately allow STAT5-dependent self-renewal programs to predominate in transformed AML cells . The STAT5-independent pathways that antagonize HSC self-renewal and promote myeloid progenitor expansion remain unclear . While the MAPK pathway was hyper-activated in postnatal Flt3ITD/+ HSCs and HPCs , MEK inhibition did not prevent , or even reduce , HSC depletion or HPC expansion in these mice ( Figure 5 ) . Other candidate pathways , including STAT3 and PI3K , were not activated by FLT3ITD ( Figure 4A ) . It is possible that low levels of signal transduction via these pathways were undetectable by Western blot but nevertheless functionally important . It is also possible that an un-interrogated pathway , such as NF-κB or CDK1 ( Gerloff et al . , 2015; Radomska et al . , 2012 ) , could promote myeloid commitment and antagonize STAT5 . Our GSEA data did show increased expression of inflammatory cytokine receptors in Stat5a/b-deficient , Flt3ITD HPCs . This raises the intriguing possibility that inflammatory cytokines could promote differentiation of Flt3ITD mutant progenitors , and perhaps AML cells . Additional genetic studies are needed to resolve whether these transcriptional changes are a cause or a consequence of enhanced lineage commitment in Flt3ITD mutant HPCs . Heterochronic genes have been implicated in cancer pathogenesis ( Shyh-Chang and Daley , 2013 ) . For example , hepatoblastomas , Wilm’s tumors and most neuroblastomas present early in life , and they often express the oncofetal proteins LIN28 or LIN28B at high levels to help maintain the primitive differentiation states of their respective anlagen ( Diskin et al . , 2012; Molenaar et al . , 2012; Nguyen et al . , 2014; Urbach et al . , 2014 ) . Adult hepatocellular carcinomas , germ cell tumors and ovarian carcinomas often ectopically activate LIN28 or LIN28B to restore oncofetal programs ( Viswanathan et al . , 2009 ) . MLL-rearranged leukemias have similarly been shown to express embryonic stem cell-related genes ( Somervaille et al . , 2009 ) , and BRAFV600E driven melanoma was recently shown to arise from melanocytes that first de-differentiate into primitive neural crest progenitors ( Kaufman et al . , 2016 ) . In each of these cases , it is easy to appreciate why maintaining or restoring a primitive cell identity might accelerate transformation—fetal cells can proliferate rapidly without differentiating or senescing . However , it is then curious as to why pediatric malignancies are relatively uncommon . Does this simply reflect greater fidelity of the genome at early ages , or are other factors at work ? Our data suggest that the transcriptional regulatory programs of fetal progenitors may , in fact , be protective against some mechanisms of transformation . Fetal and adult progenitors interpret FLT3ITD-derived signals differently , as evidenced by their distinct transcriptional responses ( Figures 7 and 8 ) , and this constrains the ability of FLT3ITD to transform fetal progenitors . Either they lack key transcriptional co-activators , or the epigenetic landscape of fetal progenitors suppresses FLT3ITD target gene activation . Further work is needed to understand the cis- and trans-regulatory elements that determine when and how individual mutations are competent to transform . If we can understand how normal developmental programs interact with genetic mutations to cause malignancies , it may be possible to target these interactions therapeutically . The Flt3ITD RRID:IMSR_JAX:011112 ( Lee et al . , 2007 ) , Runx1fRRID:IMSR_JAX:008772 ( Taniuchi et al . , 2002 ) , Stat5a/bfRRID:MMRRC_032053-JAX ( Cui et al . , 2004 ) , Rictorf RRID:IMSR_JAX:020649 ( Magee et al . , 2012 ) , Vav1-Cre RRID:IMSR_JAX:008610 ( Siegemund et al . , 2015 ) and Mx1-Cre RRID:IMSR_JAX:003556 ( Kühn et al . , 1995 ) mouse strains have all been previously described and were obtained from The Jackson Laboratory ( Bar Harbor , ME ) . These lines were all on a pure C57BL/6 background . Expression of Mx1-Cre was induced by three intraperitoneal injections of pIpC ( GE Life Sciences , Pittsburgh , PA; 10 μg/dose ) over five days beginning six weeks after birth . PD0325901 ( Cayman Chemicals , Ann Arbor , MI ) was suspended in 0 . 5% hydroxypropylmethylcellulose vehicle ( Sigma ) and administered by oral gavage as described in the text . All mice were housed in the Department for Comparative Medicine at Washington University . All animal procedures were approved by the Washington University Committees on the Use and Care of Animals . Bone marrow cells were obtained by flushing the long bones ( tibias and femurs ) or by crushing long bones , pelvic bones and vertebrae with a mortar and pestle in calcium and magnesium-free Hank’s buffered salt solution ( HBSS ) , supplemented with 2% heat inactivated bovine serum ( Gibco , Carlsbad , CA ) . Splenocytes were obtained by macerating spleens with frosted slides . Single cell suspensions were filtered through a 40 um cell strainer ( Fisher , Houston , TX ) . The cells were then stained for 20 min with fluorescently conjugated antibodies , washed with HBSS + 2% bovine serum and resuspended for analysis . Cell counts were measured by hemocytometer . The following antibodies were used for flow cytometry , all were from Biolegend ( San Diego , CA ) except as indicated: CD150 ( TC15-12F12 . 2 ) , CD48 ( HM48-1 ) , Sca1 ( D7 ) , c-Kit ( 2B8 ) , Ter119 ( Ter-119 ) , CD3 ( 17A2 ) , CD11b ( M1/70 ) , Gr-1 ( RB6-8C5 ) , B220 ( RA3-6B2 ) , CD8a ( 53–6 . 7 ) , CD34 ( eBioscience/Affymetrix , Santa Clara , CA , RAM34 ) , CD2 ( RM2-5 ) , CD45 . 1 ( A20 ) , CD45 . 2 ( 104 ) , CD127 ( A7R340 ) , CD16/32 ( 93 ) and FLT3/CD135 ( A2F10 ) . Lineage stains for all experiments included CD2 , CD3 , CD8a , Ter119 , B220 and Gr1 . Antibodies to CD4 and CD11b were omitted from the lineage stains because they are expressed on fetal HSCs at low levels . Unless otherwise indicated , we used the following surface marker phenotypes to define cell populations: HSCs ( CD150+ , CD48-Lineage- , Sca1+ , c-kit+ ) , HPCs ( CD48-Lineage- , Sca1+ , c-kit+ ) , and GMPs ( Lineage- , Sca1- , CD127- , c-kit+ , CD34+ , CD16/32+ ) . Non-viable cells were excluded from analyses by 4’ , 6-diamidino-2-phenylindone ( DAPI ) staining ( 1 μg/ml ) . When HSCs and HPCs were isolated for Western blotting or RNA collection , c-kit+ cells were enriched prior to sorting by selection with paramagnetic beads ( Miltenyi Biotec , Auburn , CA ) . Flow cytometry was performed on a BD FACSAria Fusion flow cytometer ( BD Biosciences ) . Eight to ten week old C57BL/6Ka-Thy-1 . 2 ( CD45 . 1 ) recipient mice were given two doses of 550 rad delivered at least 3 hr apart . Donor fetal liver or bone marrow cells were mixed with competitor bone marrow cells at the doses indicated in the text and injected via the retroorbital sinus . To assess donor chimerism , peripheral blood was obtained from the submandibular veins of recipient mice at the indicated times after transplantation . Blood was subjected to ammonium-chloride lysis of the red blood cells and leukocytes were stained with antibodies to CD45 . 2 , CD45 . 1 , B220 , CD3 , CD11b and Gr-1 to assess multilineage engraftment . Functional HSC frequencies were calculated and compared by using Extreme Limiting Dilution Analysis ( Hu and Smyth , 2009 ) . For secondary transplants , mice were injected with 3 million cells from the bone marrow of primary recipient mice . RNA was isolated from HSCs with RNAeasy micro plus columns ( Qiagen ) and converted to cDNA with Superscript III reverse transcriptase ( Lifetech , Carlsbad , CA ) . Quantitative RT-PCR assays were performed with Taqman Gene Expression Assays specific to mouse Flt3 and β-actin ( Lifetech ) . Analysis was performed with a Mx3005P qPCR system ( Agilent , Wilmington , DE ) . Samples were normalized based on β-actin expression . Twenty-five thousand HSC/MPPs , HPCs or GMPs were double sorted into Trichloracetic acid ( TCA ) , and the volume was adjusted to a final concentration of 10% TCA . Extracts were incubated for 15 min on ice and centrifuged at 16 , 100xg at 4°C for 10 min . Precipitates were washed in acetone twice and dried . The pellets were solubilized in 9M urea , 2% Triton X-100 , 1% DTT . LDS loading buffer ( Lifetech ) was added and the pellet was heated at 70°C for 10 min . Samples were separated on Bis-Tris polyacrylamide gels ( Lifetech ) and transferred to a PVDF membrane ( Lifetech ) . All antibodies were from Cell Signaling Technologies ( Danvers , MA ) except as indicated: P-STAT5 ( 4322 ) , Total STAT5 ( 9363 ) , P-STAT3 ( 9145 ) , Total STAT3 ( 9139 ) , P-ERK1/2 ( 4370 ) , Total ERK1/2 ( 4696 ) , P-AKT Ser473 ( 4060 ) , P-AKT T308 ( 13038 ) , Total AKT ( 4691 ) , α-TUBULIN ( 3873 ) , β-ACTIN ( Santa Cruz Bioscience , Santa Cruz , CA , clone AC-17 ) , HRP-anti-Rabbit IgG ( 7074 ) and HRP-anti-mouse IgG ( 7076 ) . Blots were developed with the SuperSignal West Femto or Pico chemiluminescence kits ( Thermo Scientific ) . Blots were stripped ( 1% SDS , 25 mM glycine pH 2 ) prior to re-probing . Bone marrow cells were isolated and spun onto glass slides using a Shandon Cytospin 3 . The slides were stained using Protocol Hema 3 Wright-Giemsa stain ( Fisher Scientific ) . All slides were reviewed by a pediatric hematologist ( JAM or ASC ) . In all cases , multiple independent experiments were performed on at least three different days to verify that the data are reproducible . Grouped data reflect biological replicates ( i . e . independent mice ) and are represented by mean +/− standard deviation . Statistical comparisons between groups were made with the two-tailed Student’s t-test except as noted in the figure legends . When multiple genotypes were compared , statistical significance was determined by performing a one-way ANOVA followed by a Holm-Sidak post-hoc test to correct for multiple comparisons . For transplantation experiments , the percentages of mice with multilineage reconstitution were compared with the Fisher’s exact test . All comparisons were performed with GraphPad Prism 6 , RRID:SCR_002798 . Ten thousand HSCs or HPCs were double sorted directly into RLT plus RNA lysis buffer ( Qiagen ) and RNA was isolated with RNAeasy micro plus columns ( Qiagen ) . Transcripts were amplified with the WTA2 kit ( Sigma ) with the Kreatech ULS RNA labeling kit ( Kreatech Diagnostics ) . Labeled cDNA was hybridized to Agilent Mouse 8 × 60 K microarrays and analyzed with an Agilent C-class scanner . Signal data were assembled and processed in Partek RRID:SCR_011860 , and samples were compared by Linear Models for Microarrays , RRID:SCR_010943 ( Ritchie et al . , 2015; Smyth , 2004 ) . Adjusted p-values were calculated by the Benjamini and Hochberg false discovery rate ( Benjamini and Hochberg , 1995 ) . Z-scores were calculated as previously described ( Cheadle et al . , 2003 ) . Hierarchical cluster analysis was performed with Cluster 3 . 0 and visualized with Java TreeView; RRID:SCR_013505 and RRID:SCR_013503 ( Eisen et al . , 1998 ) . Principal component analyses and Euclidean distance comparisons ( by permutation testing ) were performed with the R software environment . Microarray data sets have been deposited into Gene Expression Omnibus ( GSE81153 ) . GSEA was performed using gene sets that were generated as cited in the text , or with gene sets curated in the MSigDB databases; RRID:SCR_003199 ( Subramanian et al . , 2005 ) .	Leukemias are a group of blood cancers that usually arise when immature blood cells gain one or more tumor-promoting genetic mutations . However , for reasons that are not clear , the mutations that cause leukemia are different in children and adults . For example , a mutation called FLT3ITD occurs relatively often in adult leukemia but is rare in infant leukemia . This raises the question of whether the blood cells of fetuses and babies are somehow protected from the effects of the mutation . Porter et al . have now compared the effects of the FLT3ITDmutation in blood cells from adult and fetal mice . In adult mice , the FLT3ITD mutation caused immature blood cells to turn different genes on and off . By contrast , the mutation had no effect on the activity of these genes in fetal mice . Furthermore , only the adult mutant cells showed changes that indicated the early stages of leukemia: the mutant blood cells of fetuses developed as normal . Porter et al . therefore concluded that the immature blood cells of fetuses are protected from the FLT3ITDmutation . To understand why fetal and adult blood cells respond differently to the FLT3ITDmutation , further experiments are needed to investigate how various genes regulate normal blood cell development . In addition , understanding why adult blood cells react to the FLT3ITDmutation might , in the future , lead to better treatment options for leukemia .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "developmental", "biology", "cancer", "biology" ]	2016	Fetal and neonatal hematopoietic progenitors are functionally and transcriptionally resistant to Flt3-ITD mutations
The cell cycle regulator p16 is known as a biomarker and an effector of aging . However , its function in intervertebral disc degeneration ( IVDD ) is unclear . In this study , p16 expression levels were found to be positively correlated with the severity of human IVDD . In a mouse tail suspension ( TS ) -induced IVDD model , lumbar intervertebral disc height index and matrix protein expression levels were reduced significantly were largely rescued by p16 deletion . In TS mouse discs , reactive oxygen species levels , proportions of senescent cells , and the senescence-associated secretory phenotype ( SASP ) were all increased , cell cycling was delayed , and expression was downregulated for Sirt1 , superoxide dismutase 1/2 , cyclin-dependent kinases 4/6 , phosphorylated retinoblastoma protein , and transcription factor E2F1/2 . However , these effects were rescued by p16 deletion . Our results demonstrate that p16 plays an important role in IVDD pathogenesis and that its deletion attenuates IVDD by promoting cell cycle and inhibiting SASP , cell senescence , and oxidative stress . Intervertebral disc degeneration ( IVDD ) refers to the physiological and pathological process of natural degeneration and aging of the intervertebral disc , which is the basis of various clinical spinal diseases ( Silagi et al . , 2018 ) . IVDD usually results in vertebral instability , disc herniation , and spinal canal stenosis , which are commonly accompanied by low back pain with or without symptoms of nerve root or spinal cord compression . These lead to tremendous societal and economic burdens worldwide . It has been estimated that , at some point during their lifetime , 80–90% of the world’s population suffer clinical low back pain , which is positively correlated with disc degeneration in approximately 23% of these people ( Smith et al . , 2011; Walker , 2000 ) . The human intervertebral disc is a non-vascular tissue , and its annulus fibrosus ( AF ) and inner layer nucleus pulposus ( NP ) rely mainly on the penetration of the end plate to provide nutrition . In this chronically high osmotic pressure , low pH , hypoxic and low-nutrition environment , the cells are less active . This is one of the reasons for the poor self-healing ability of the disc's structure and function after tissue damage , and thus , intervertebral discs degenerate more easily than other tissues in the body ( Feng et al . , 2016 ) . Among the many factors that cause intervertebral disc degeneration is the accumulation of senescent disc cells ( most of which are NP cells ) , which has provided a novel insight into IVDD pathogenesis . Senescent NP cells generate only a small number of new cells; therefore , the number of functional cells decreases gradually . Moreover , senescent NP cells may change the disc microenvironment , creating a senescence-associated phenotype in which proinflammatory factors are overexpressed , extracellular matrix ( ECM ) is decreased , and growth factors and chemokines are downregulated ( Le Maitre et al . , 2007; Markova et al . , 2013; van Deursen , 2014 ) . However , the molecular mechanisms that underpin cell senescence in IVDD are unclear . Cell senescence is regulated by various molecular signaling pathways . One of the canonical molecules involved in cell senescence is p16 ( p16INK4a ) , which is encoded by the CDKN2A gene and belongs to the cell cycle regulatory pathway ( Serrano , 1997 ) . Senescent cells , most of which seem to express p16 ( Childs et al . , 2017 ) , accumulate with aging and are conducive to tissue dysfunction . The clearance of p16-positive senescent cells in adipose tissue , skeletal muscle and the eye has been suggested to delay aging-associated disorders in mice ( Baker et al . , 2011 ) . Specifically , the systemic clearance of p16-positive senescent cells and conditional Cdkn2a gene deletion have been shown to mitigate age-associated IVDD in mice , mostly by suppressing the senescence-associated secretory phenotype ( SASP ) , improving matrix homeostasis , and reducing apoptosis ( Novais et al . , 2019; Patil et al . , 2019 ) . However , we do not yet know how p16 drives disc cell senescence and whether other factors are present in the progression of IVDD , especially in human discs . Increasing levels of reactive oxygen species ( ROS ) , another main feature of aging , are involved in a number of age‐related pathologies . Senescence can occur under prolonged oxidative states; and thus , ROS is seen as an important mediator of the progression of cellular senescence ( Colavitti and Finkel , 2005 ) . Pathological ROS levels have been implicated in the induction of senescence-like phenotypes similar to that of p16-induced senescence . An increasing number of studies have shown that p16 might play a role in oxidative stress-associated senescence ( Gonçalves et al . , 2016; Mas-Bargues et al . , 2017 ) . Nonetheless , whether p16 contributes to intervertebral disc aging by increasing ROS is unclear . The present study aimed to highlight the influence of p16 on disc degeneration , mainly focusing on oxidative stress and human NP cell proliferation , and verified this effect in mice that have homozygous deletion of Cdkn2a . To explore the role of p16 in IVDD , p16 expression was first verified in the NP tissues of patients with various degrees of disc degeneration , as examined by histological staining ( Pfirrmann grades 2–5 , Figure 1—figure supplement 1 ) . H and E staining showed a substantially disordered tissue texture in samples with a high Pfirrmann grade , in which hypertrophic and vacuole-like cells and multinuclear giant cells were present at the end of the sample ( Figure 1A ) . Masson and Safranin O staining showed smaller amounts of proteoglycans ( PGs ) and increased levels of fibrosis in NP tissues that had a high Pfirrmann grade ( Figure 1A ) . These results confirm that severely degenerated NP tissue is correlated with a high Pfirrmann grade . In addition , immunohistochemistry ( IHC ) and western blotting ( WB ) revealed that degenerated discs with a higher Pfirrmann grade expressed a higher level of p16 than those with a lower Pfirrmann grade ( Figure 1B , C , D; with p16 expression levels and their corresponding Pffirmann grades being shown in Figure 1—figure supplement 2 ) . These results confirm that p16 accumulates in NP tissues as IVDD progresses . To uncover how p16 participates in IVDD progression , NP cells with mild degeneration were isolated from Pfirrmann grade 2 tissues and cultured in vitro . IL-1β was used to induce NP cell degeneration . Immunofluorescent ( IF ) and SA-β-gal staining clearly illustrated that IL-1β greatly increased the percentage of senescent NP cells and p16 protein expression when compared with control levels ( Figure 2A , D , E ) . In addition , the effect of altering p16 expression levels on NP cell degeneration and proliferation was investigated . p16 expression was down- and upregulated by siRNA or plasmid transfection , respectively . The transfection efficiencies of targeted siRNAs and plasmids , compared with those of null siRNA and empty plasmid , respectively , are presented in Figure 2—figure supplement 1 . p16 expression decreased after siRNA-mediated knockdown , which decreased the proportion of NP cells that demonstrated a senescent phenotype . By contrast , p16 overexpression caused a marked opposite effect ( Figure 2B , H ) . In addition to the degree of aging , proliferative ability is another indicator of cellular degeneration . Cell counting kit-8 ( CCK-8 ) analyses confirmed that NP cell proliferation was reduced by p16 overexpression and promoted by p16 knockdown when compared with the levels seen after treatment with IL-1β treatment alone ( Figure 2C ) . To determine the potential mechanism by which p16 modulates NP cell physiological behavior , flow cytometry was used to analyze ROS levels and the cell cycle . ROS levels were obviously higher when p16 expression was increased ( Figure 2F , G ) . As p16 expression gradually increased from the p16+IL-1β group to the IL-1β group and the control group , the cells presented cell-cycle arrest in the G0/G1 phase . Interestingly , considerably more NP cells progressed through G0/G1 to S phase following p16 downregulation ( Figure 2F , I ) . These results demonstrate that p16 might regulate senescence by mediating oxidative stress and promoting proliferation by accelerating the movement of cells through the G1/S checkpoint . Recent studies have reported that rapamycin prevents IVDD by inhibiting cell senescence via the mTOR signaling pathway ( Choi et al . , 2016; Ito et al . , 2017 ) . Therefore , the application of rapamycin was used to inhibit cell senescence in order to explore the interplay between this cellular event and p16 in IVDD prevention . Rapamycin antagonized the effect of IL-1β , decreasing p16 expression and the percentage of senescent NP cells ( Figure 3A , B , C , E , F ) . Flow cytometry analyses revealed significantly lower ROS levels in the IL-1β+rapamycin group than in the IL-1β-alone group , and showed that rapamycin treatment reduced the ratio of cells in the G0/G1 phase; meanwhile , the ratio of cells in S phase was higher after rapamycin treatment ( Figure 3D , G , H ) . CCK-8 assays also showed a higher level of proliferation in the IL-1β+rapamycin group when compared with the IL-1β alone group ( Figure 3I ) . These changes parallel the above described impact of p16 siRNA . These findings demonstrate that rapamycin can prevent NP cell degeneration by reducing ROS levels and mediating the cell cycle , which might be associated with p16 inhibition . To further assess whether p16 deletion plays a positive role in IVDD prevention in vivo , the Cdkn2a gene knock out ( p16 KO ) mice and the tail suspension ( TS ) method were used to establish a mouse IVDD model . After 4 weeks of TS , muscles around the spine were congested with varying degrees of injury ( Figure 4—figure supplement 1B ) . Based on the morphological and histological changes among different groups , disc height index ( DHI ) analyses showed that mouse disc heights were decreased by TS but were maintained in p16 KO mice when compared with WT mice ( Figure 4A , C ) . Furthermore , micro-magnetic resonance imaging ( MRI ) demonstrated that TS reduced water content in the disc and that p16 deletion significantly protected against this effect ( Figure 4H , Figure 4—figure supplements 2 , 3 ) . After TS , disc heights decreased and more vesicular cells appeared , and the discs in p16 KO mice exhibited obviously higher glycosaminoglycan ( GAG ) levels with or without TS than those in WT mice ( Figure 4B ) . Inflammation is a vital part of the disc degeneration process . To examine the effects of systemic p16 knockout , the levels of inflammatory factors in the NP tissues of mice were analyzed . There was a clear difference in inflammation between p16 KO and WT mice , as p16 deletion reduced the expression of TNF-α , IL-1β , and IL-6 ( Figure 4G ) . These results were confirmed when RNA expression was assessed ( Figure 4F ) . Furthermore , because NF-κB-p65 has a vital function in regulating inflammatory responses and is activated by various stimuli , including stress , we conducted analyses on NF-κB-p65 expression levels . Western blot analyses showed higher p65 levels in the TS group than in the control group . After p16 gene deletion , p65 expression decreased significantly ( Figure 4D , E ) . Matrix metallopeptidases ( MMPs ) can degrade all types of ECM proteins , decreasing the aggrecan and collagen II content of the tissue . Thus , we evaluated MMP3 , MMP9 , MMP10 , and MMP13 mRNA and protein expression levels . The levels of these MMPs mostly increased after TS , and p16 deletion partly reversed these changes ( Figure 4D , E ) . Treatment effects on the expression levels of the typical components of ECM , aggrecan , collagen I , collagen II , and collagen X , were measured by western blot or qRT-PCR . Aggrecan and collagen II , which are protective ECM components , were slightly degraded in p16 KO mice compared to WT mice; whereas collagens I and X , which are harmful ECM components , were expressed at low levels in p16 KO mice ( Figure 4D , E ) . In summary , these results suggest that p16 deletion partly postpones IVDD in mice as assessed in terms of changes in disc height , water content , inflammation , and ECM components . To further determine the potential mechanism by which p16 functions in IVDD in vivo , multiple biological indicators were explored . Specifically , we assessed the degree of senescence , proliferative capacity , oxidative stress level , and the expression of cell-cycle proteins in p16 KO and WT mice with or without TS . IHC analyses of β-gal and western blot of p19 and p53 revealed that the discs of WT mice exhibited a more senescent phenotype than those of p16–/– mice in both the TS and control groups ( Figure 5A , C , D , F ) . The proportions of PCNA- and Ki67-positive cells , the percentages of proliferative cells , and the IGF1 protein levels in the discs were substantially higher in p16 KO mice than in WT mice , even after TS ( Figure 5A ) . Interestingly , the levels of vascular endothelial growth factor ( VEGF ) , a microangiogenesis marker , were decreased in p16 KO mice , suggesting a protective function of p16 in disc degeneration ( Figure 5C , D ) . Because ROS levels decreased in human NP cells upon silencing p16 , as described above , the antioxidant enzyme gene expression and ROS levels were determined in mouse disc tissues . SOD1 , SOD2 , GPX1 , GPX3 , and CAT mRNA expression increased upon p16 deletion , and p16 KO mice had lower total ROS levels than did WT mice , even after TS ( Figure 5B , G , I ) . The mouse IVDD model also revealed the same effects on Sirt1 , SOD1 , and SOD2 protein levels ( Figure 5C , D ) . The DNA injury marker 8-hydroxy-deoxyguanosine ( 8-OHdG ) can be induced by oxidative stress . When compared to WT mice , the proportions of 8-OHdG-positive cells were greatly decreased in p16 KO mice , indicating that p16 deletion plays a protective role in the antioxidant process in the disc ( Figure 5A ) . To explore whether p16 affects proliferation by mediating the cell cycle , cell-cycle progression and cell-cycle-related proteins were analyzed by flow cytometry and western blotting . p16 KO mice showed increased progression of cells from G0/G1 into S phase compared with WT mice , with or without TS ( Figure 5B , H ) . CDK4 , CDK6 , pRb , E2F1 and E2F2 protein expression levels were upregulated in p16 KO mice compared to WT mice . Conversely , the expression level of RB protein was downregulated in p16 KO mice ( Figure 5C , E ) . These results demonstrate that p16 deletion can partially inhibit aging-related senescence by reducing disc oxidative stress injury and enhancing NP cell proliferation by promoting progression through the G1/S checkpoint . Because expression of the transcription factor NF-κB-p65 differed between p16 KO and WT mice , it was hypothesized that NF-κB-p65 might control p16 protein levels . To confirm that NF-κB-p65 controls p16 at the transcriptional level , five putative NF-κB-p65 binding sites in the CDKN2A promoter region were identified and chromatin immunoprecipitation ( ChIP ) primers were designed using Primer Premier ( Supplementary file 1 ) . First , when we assessed whether NF-κB-p65 binds to the five putative promoter sequences , only two sites were verified by ChIP as efficient binding sites . One putative promoter sequence that was bound effectively is shown in Figure 6A , and another promoter sequence with no binding is shown in Figure 6—figure supplement 1 . Using human genomic DNA as a template , the whole CDKN2A promoter segment was amplified by PCR ( lane 1 of Figure 6A ) . Clear DNA amplification was examined after immunoprecipitation without the irrelevant control IgG ( lane 2 of Figure 6A ) , and with the anti-p65 antibody ( lane 3 of Figure 6A ) . Next , the WT and binding-site mutant CDKN2A promoter sequences were cloned into the pGL4 . 23‐basic vector ( producing pGL4 . 23-wt and pGL4 . 23-mut , respectively ) , and the resulting plasmids were transiently transfected into human NP cells ( Figure 6B ) . Transfection with the empty plasmid ( pGL4 . 23 ) without the CDKN2A promoter sequence and the Renilla expression plasmid ( vector+pGL+pRL ) or with the p65 plasmid , and pGL4 . 23 without the CDKN2A promoter sequence and the Renilla expression plasmid ( p65+pGL+pRL ) served as the negative controls . Luciferase activity was significantly higher in NP cells transfected with the p65 and pGL-wt plasmids than in those transfected with empty plasmid and the pGL-wt plasmid , indicating that p65 successfully activated the CDKN2A promoter . By contrast , luciferase activity was significantly lower in NP cells transfected with the p65 and pGL-mut plasmids than in those transfected with the p65 and pGL-wt plasmids ( Figure 6C ) . The findings confirm that the CDKN2A promoter region with the predicted NF-κB-p65 binding sites is sufficient to promote transcription , providing a molecular mechanism that emphasizes p65‐dependent p16 transcriptional activation in NP cells . Although numerous studies have proven that p16 contributes to IVDD pathogenesis , few have focused on the role of p16 in humans . Here , an unbiased comparison of p16 expression in NP cells from IVDD patients with varying Pfirrmann scores showed that p16 expression is positively correlated with the degree of human disc degeneration . Along with the increase in the severity of human IVDD , NP cells showed decreased PG contents , increased fibrosis , and greater vacuolization; moreover , more multinucleated giant cells were observed , and p16 accumulated . The findings show that p16 plays a role in IVDD progression . Using NP cells harvested from patients with disc degeneration with a Pfirrmann score of 2 ( mild disc degeneration ) , in-vitro studies further explored the function of p16 in the pathology of disc degeneration . The results suggest that p16 deficiency decreases oxidative stress and DNA damage in NP cells and contributes to NP cell proliferation , which protects against IVDD by promoting cell-cycle progression . In addition , for potential therapeutic exploration , these findings have provided theoretical and experimental evidence to support the potential use of rapamycin or the upstream approach targeting NF-κB-p65 to suppress p16 expression . Thus , the current investigation has not only demonstrated the different mechanisms of p16 in IVDD but also may provide theoretical evidence to inform the exploration of effective methods to downregulate p16 in order to reverse IVDD . Although the presence of p16-positive cells indicates that an organism is in an inactive state ( Baker et al . , 2011 ) , it is unclear whether the differential expression of p16 affects human disc degeneration . To simulate the microenvironment in disc degeneration , IL-1β was used to induce NP cell senescence . Multiple analyses revealed that p16 expression increased significantly as the degree of senescence increased in NP cells . However , the senescent phenotype of NP cells became less pronounced when p16 was silenced . By contrast , increased NP cell senescence was observed when p16 levels were upregulated by plasmid transfection . These results imply that p16 not only is produced by NP cell senescence but also accelerates NP cell senescence . ROSs are mainly induced during cellular aging , but the mechanism by how p16 regulates ROS levels in NP cells is not yet clear . The ROS levels in different groups of NP cells expressing different levels of p16 showed that ROS levels increased along with p16 overexpression . This result suggests a strategy to reduce ROS in NP cells via p16 suppression . Senescence plays a basic role in regulating cell cycling by halting cell proliferation , and p16 expression was observed to be negatively correlated with human NP cell proliferation , an effect that was reversed by p16 downregulation . Previously , p16 was shown to be a cyclin-dependent kinase ( CDK ) inhibitor that is sufficient to inhibit cell proliferation and to induce aging features in mammals by restraining the cell cycle ( Boquoi et al . , 2015 ) . The in-vitro model illustrated that p16 can inhibit the progression of NP cells from G1 to S phase , which is an essential mechanism by which proliferation is regulated in human NP cells . On the basis of the results described above , it is more feasible to suppress p16 expression using a specific drug rather than siRNA transfection . After all , for patients , drug therapy is more convenient and economical than gene therapy . Rapamycin was previously shown to have antiaging effects in multiple cells and organisms by modulating oxidative stress , nutrient sensing , and the cell cycle ( Richardson , 2013; Wang et al . , 2017a ) . Rapamycin was also shown to inhibit p16 expression to some extent ( Gidfar et al . , 2017 ) . However , previous studies on the effect of rapamycin on disc degeneration focused on the role of this compound in autophagy ( Ito et al . , 2017; Tu et al . , 2018 ) . Therefore , rapamycin was applied to inhibit p16 and to explore its function in regulating ROS levels and the cell cycle . Rapamycin significantly decreased p16 expression and reversed the senescent phenotype of human NP cells . Furthermore , rapamycin decreased ROS levels in NP cells , which showed increased proliferation . Cell-cycle analyses indicated that rapamycin promoted the progression of NP cells from G1 to S phase . Therefore , rapamycin may suppress ROS levels and promote NP cell proliferation , and these effects may be related to its ability to suppress p16 . Taken together , p16 downregulation is likely to exert an antioxidant effect and promotes human NP cell proliferation , therefore playing a protective role in IVDD . To verify these results in vivo , compound mutant mice with homozygous Cdkn2a deletion were used to establish an IVDD model involving TS . As a shock absorber for the spine , the basic role of the disc is a mechanical one containing load distribution , energy dissipation , and motion permit in daily activities . Mechanical factors have been proposed as one of the mechanisms necessary for accelerating the aging progression of both human and rodent discs via altered loading in several studies , so the TS mice model is an applicable mechanical representation of both tensile force on human discs and the aging process ( Hutton et al . , 2002 ) . Simulation of weightlessness by TS changes flexion-extension , axial rotation , lateral bending and hydrostatic pressure in the disc and leads to destruction of the ECM destruction , an inflammatory response and a catabolic process that represents premature aging-related IVDD progression ( Földes et al . , 1996 ) . In the present study , p16 deletion protected against changes in disc heights and disc water contents in mice , which are the most intuitive indicators of IVDD in the clinic . Levels of aggrecan and collagen II , which protect NP cells in the ECM , were significantly increased in p16 KO mice with or without TS . By contrast , the protein levels of the fibrosis markers collagens I and X decreased after p16 deletion . To demonstrate whether p16 deletion affects inflammation in mouse discs , some inflammatory factors were further examined , and the results showed that p16 deletion reduced both the protein and the mRNA levels of the inflammatory factors . Encouragingly , the measurements of ROS levels and antioxidant activity showed that p16 deletion improved antioxidant activity in the discs and decreased ROS levels and DNA injury . Furthermore , examinations of the proliferation markers Ki67 , PCNA , and IGF-1 indicated increased proliferative capacity in NP cells after p16 deletion . Interestingly , angiogenesis , which has been shown to increase the risk of IVDD ( Kwon et al . , 2017; Zaidi et al . , 2018 ) , was reduced in p16 KO mice compared with WT mice; this might be a novel direction for future research . Finally , the analyses of the cell cycle and cell cycle-related proteins confirmed that p16 suppression activates CDK4 and CDK6 to promote Rb protein phosphorylation , enhance E2F1/E2F2 activity , and promote the progression of NP cells from G1 to S phase . Taken together , these findings demonstrate that ablation of p16 can relieve mouse disc degeneration by protecting the ECM , inhibiting fibrosis , reducing inflammation , decreasing ROS levels , and promoting proliferation by regulating the cell cycle . As p16 plays an important role in IVDD , it is essential to understand the molecular mechanism of p16 activation and dysfunction . Previous studies have shown that Bmi-1 and 1 , 25‐dihydroxyvitamin D inactivate p16 ( Chen et al . , 2019; Taylor et al . , 2015; Yamakoshi et al . , 2015 ) . However , the molecular mechanism of p16 activation is still unclear . Bmi-1 is reported to active NF-κB signaling in glioma angiogenesis , cell migration and invasion ( Jiang et al . , 2013; Sun et al . , 2014 ) . By contrast , 1 , 25-dihydroxyvitamin D takes part in the suppression of inflammation and anticancer properties by blocking NF-kB activation ( Chen et al . , 2011; Fekrmandi et al . , 2015 ) . Intriguingly , a decreased NF-kB-p65 level was observed in the discs of p16 KO mice compared with those of the WT . Therefore , further work is necessary to explore the relation between NF-kB-p65 and p16 in disc tissue . NF-κB-p65 is well known for its role in regulating inflammation , immune response , cell division and apoptosis , and has been shown to participate in IVDD progression ( Wang et al . , 2018; Wang et al . , 2017b ) . The present results support the observation that NF-κB-p65 is involved in p16 activation . Furthermore , by conducting analyses with the JASPAR database of transcription factor binding profiles , five putative NF-κB-p65 binding sites were identified in the CDKN2A promoter , and the ChIP assays have confirmed the activity of two of these putative binding sites . Finally , these two binding sites were tested in luciferase reporter gene assays , which showed that NF-κB-p65 bound at one binding site in the CDKN2A promoter , confirming that NF-κB-p65 is upstream of p16 . Taken together , from the data in the current study , a model illustrating the role and possible mechanisms of p16 in regulating IVDD can be proposed ( Figure 7 ) . NF-κB-p65 probably increases p16 expression by promoter activation . p16 deficiency regulates the antioxidative behaviors of NP cells , resulting in the suppression of ROS levels and the alleviation of both NP cellular senescence and the SASP . Subsequently , p16 deficiency promotes cellular proliferation and the production of ECM , such as collagen II and aggrecan . It can be speculated that molecular studies of the p16 in the development of IVDD may provide new solutions for preventing degenerative disc diseases . Thirty-two fresh human intervertebral disc tissue samples were harvested from patients undergoing intervertebral disc surgery at the First Affiliated Hospital of Nanjing Medical University ( patients’ information is listed in Supplementary file 2 ) . Before the operation , the informed consents of the patients were obtained . These consents included the voluntary donation of the diseased nucleus pulposus tissue extracted from the operations , and consent for the use of all specimens for scientific research and for publication of the results obtained in scientific journals . This project was implemented by the approval of the Ethics Committee of the First Affiliated Hospital of Nanjing Medical University ( registered number 2018 SR-233 ) . All the samples were divided into four groups using the Pfirrmann score , which was determined on the basis of MRI results from each patient before surgery . The NP tissues were cut into small pieces and digested at 37°C overnight with collagenase XI ( Sigma , Ohio , USA ) , dispase II ( Sigma , Ohio , USA ) and cell culture medium ( containing 2% penicillin/streptomycin and 10% fetal bovine serum , Thermo , Massachusetts , USA ) . The cell solution was centrifuged to obtain the cell pellets . NP cells were stimulated with 10 ng/mL IL-1β ( Sigma , Ohio , USA ) to establish a degeneration model ( Shen et al . , 2017 ) . We also cotreated IL-1β-treated NP cells with 50 nM rapamycin ( Gao et al . , 2018 ) ( Sigma , Ohio , USA ) to determine its function during NP cell degeneration . Human p16 plasmid vectors and siRNA were provided by Invitrogen ( Massachusetts , USA ) and GenePharma ( Shanghai , China ) . The CDKN2A gene was inserted into the pcDNA3 . 1 plasmid . On the basis of the manufacturer's instructions , NP cells were transfected with Lipo6000 ( Beyotime , Shanghai , China ) . Then cells were transfected by siRNA against CDKN2A ( Lau et al . , 2007 ) . 24 hr later , IL-1β ( 10 ng/mL ) was used to treat the cells for 4 days , and the cells were harvested for subsequent experiments . The transfection efficiency was examined by quantitative real-time polymerase chain reaction ( qRT-PCR ) , immunofluorescence ( IF ) , and western blot ( WB ) . The Cdkn2a heterozygous mice ( FVB N2 background ) were a gift from Baojie Li ( Shanghai Jiao Tong University , Shanghai , China ) and had been backcrossed on the C57BL/6J background . These mice were mated to produce Cdkn2a knock-out ( p16 KO ) and wild-type ( WT ) littermates . Animal use was approved by the Institutional Animal Care and Use Committee of Nanjing Medical University ( approval number: IACUC-1709021 ) . As an IVDD model ( Hutton et al . , 2002; Nakamura et al . , 2013 ) , TS was carried out in mice for 4 weeks . A specialized cage was made to suspend the mice ( Figure 4—figure supplement 1A ) . Forty-eight 16-week-old WT and p16 KO mice were randomly separated into two groups that were divided into two subgroups: WT control ( WT ) , p16 KO control ( p16 KO ) , tail-suspended WT ( WT+TS ) , and tail-suspended p16 KO ( p16 KO+TS ) groups . At the appropriate time point , the mice were humanely killed , and the lumbar vertebrae ( from lumbar 1 to 5 ) were removed for examination . Cultured NP cells were rinsed three times using phosphate-buffered saline ( PBS ) , fixed by 4% formaldehyde for 15 min , incubated in 0 . 25% Triton X‐100 for 15 min , and blocked by 5% bovine serum albumin ( BSA , Sigma , Ohio , USA ) in PBS for 30 min at room temperature . Then , the cells were treated with primary antibody against p16 ( ab51243 , Abcam , Cambridge , UK ) for one night at 4°C , and incubated with DyLight 488-conjugated goat anti-rabbit IgG antibody ( Abbkine , California , USA ) for 2 hr in the dark at room temperature . The cells were visualized using a fluorescence microscope ( Leica , Wetzlar , Germany ) , and nuclei were counterstained with DAPI ( Beyotime , Shanghai , China ) . Before the animals were sacrificed , X-ray images and micro-MRI scans were taken . X-rays were used to measure the disc height and vertebral body . The intervertebral disc height index ( DHI ) was obtained by calculating the average values of the posterior , middle and anterior parts of the intervertebral disc , and these values were divided by the average height of the adjacent vertebral body ( Figure 4—figure supplement 4 ) . Micro-MRI was performed using T2-weighted sections . On the basis of changes in signal intensity with four grades ( 1 , normal; 2 , minimal decrease; 3 , moderate decrease; and 4 , severe decrease ) , a revised Thompson classification was used to evaluate disc status . Mouse lumbar spines ( L3–L6 ) were decalcified for 14 days after fixation in 4% paraformaldehyde ( PFA ) solution . Human NP tissues and mouse spines were processed for paraffin embedding and sectioning into 5-μm-thick slices for histological staining or immunohistochemistry ( IHC ) , as described below . To evaluate disc degeneration , deparaffinization , hydration , and hematoxylin and eosin ( H and E ) staining were used to treat the paraffinized slices ( Wang et al . , 2009 ) so that cells and tissue morphology could be observed . Masson’s stain ( Nagatoya et al . , 2002 ) was added to analyze NP fibrosis; Safranin O ( Kiviranta et al . , 1985 ) was added to analyze proteoglycans ( PGs ) ; and senescence-associated beta-galactosidase ( SA-β-gal ) ( Ji et al . , 2012 ) was added to identify senescent cells . IHC was performed as in a previous study ( Yukata et al . , 2018 ) . Briefly , sections were treated with sodium citrate ( 10 mM , 100°C ) ( for antigen retrieval ) and H2O2 ( 10% in PBS ) ( for endogenous peroxidase inactivation ) . Next , the slices were blocked with 10% goat serum , and incubated overnight at 4°C using primary antibodies against β-galactosidase ( ab203749 , Abcam , Cambridge , UK ) , 8-hydroxy-2 deoxyguanosine ( 8-OHdG ) ( ab48508 , Abcam , Cambridge , UK ) , Ki67 ( ab15580 , Abcam , Cambridge , UK ) , and PCNA ( ab92552 , Abcam , Cambridge , UK ) . Then , biotinylated goat anti-mouse or anti‐rabbit IgG ( Sigma , Ohio , USA ) were used to treat the slices , before they were incubated with Vectastain Elite ABC reagent ( Fisher Scientific , Hampton , New Hampshire , USA ) for 30 min . 3 , 3‐diaminobenzidine was used for staining , followed by counterstaining with hematoxylin . 8-OHdG , Ki67 , and PCNA are mostly expressed in the nucleus; p16 and β-galactosidase are expressed in both the nucleus and the cytoplasm . The positive cell rate for 8-OHdG , Ki67 and PCNA is the ratio of the number of positive nuclei to the number of all hematoxylin-labeled cells . The positive cell rate for p16 and β-galactosidase is the ratio of the number of positive nuclei or/and cytoplasm to the number of all hematoxylin-labeled cells . Proteins were harvested from human NP tissues or mouse disc tissues with a Protein Extraction Kit ( Thermo , Massachusetts , USA ) . Immunoblotting was performed as in a previous study ( Miao et al . , 2008 ) , using primary antibodies , against collagen I/X ( ab34710/ab58632 , Abcam , Cambridge , UK ) , collagen II ( ab34712 , Abcam , Cambridge , UK ) , Sirt1 ( ab110304 , Abcam , Cambridge , UK ) , superoxide dismutase 1/2 ( SOD1/2 ) ( ab13498/ab13533 , Abcam , Cambridge , UK ) , matrix metalloproteinases-13 ( MMP-13 ) ( ab52915/ab39012 , Abcam , Cambridge , UK ) , nuclear factor kappa-B-p65 ( NF-κB-p65 ) ( SC-71675 , Santa Cruz , California , USA ) , insulin-like growth factor 1 ( IGF-1 ) ( ab9572 , Abcam , Cambridge , UK ) , vascular endothelial growth factor ( VEGF ) ( ab69479 , Abcam , Cambridge , UK ) , p19/53 ( SC-1665/SC-126 , Santa Cruz , California , USA ) , cyclin-dependent kinases 4/6 ( CDK4/6 ) ( ab199728/ab131469 , Abcam , Cambridge , UK ) , retinoblastoma protein/phosphorylated retinoblastoma protein ( Rb/pRB ) ( SC-74562/SC-56175 , Santa Cruz , California , USA ) , transcription factor E2F1/2 ( E2F1/2 ) ( SC-137059/SC-633 , Santa Cruz , California , USA ) , and β‐actin ( ab8226 , Abcam , Cambridge , UK ) . Immunoreactive bands were analyzed by Scion Image Beta 4 . 02 and visualized with ECL ( Beyotime , Shanghai , China ) . Total RNA was harvested from human NP cells and mouse disc tissues with TRIzol reagent ( Beyotime , Shanghai , China ) . PrimeScript RT Master Mix ( Perfect Real Time , TaKaRa , California , USA ) was used to reverse transcribe RNA to cDNA . Supplementary file 3 tabulates the qRT-PCR primer sequences . GAPDH was used for normalization . Relative mRNA expression levels were determined by the 2−ΔΔCt method . Serum samples were obtained from blood collected from the eyeballs of mice in each group . Mouse NP cells were also collected by the method described above . The levels of IL-1β , IL-6 and TNF-α in NP cell supernatants were determined using an ELISA kit ( KeyGen , Nanjing , China ) . Total ROS production , NP cell proliferation and cell-cycle progression were separately assessed using diacetyl dichlorofluorescein staining ( Sigma Aldrich , Ohio , USA ) , propidium iodide staining ( KeyGen , Nanjing , China ) , and EdU Flow Cytometry Assay Kits ( Invitrogen , Massachusetts , USA ) , respectively . Human and mouse NP single-cell suspensions were prepared in PBS , and the cells were treated with the corresponding specialized reagent . The cell pellets were incubated at 37°C for 30 min and obtained by centrifugation . Finally , the specimens were investigated by flow cytometry with a FACSCalibur flow cytometer . Cell proliferation was evaluated with a CCK-8 assay ( KeyGen , Nanjing , China ) . In brief , cells in each group ( 5000/well ) were allowed to grow for 24 , 48 , and 72 hr . Ten microliters of CCK reagent in a total volume of 100 μl was put into each well , before incubation for 3–4 hr . The absorbance at 450 nm was measured by an ELISA plate reader ( Thermo Electron , Massachusetts , USA ) . The 2000-bp region upstream of the p16 gene was selected as the promoter region according to the National Center for Biotechnology Information database ( http://www . ncbi . nlm . nih . gov/ ) . After predicting DNA-binding sites for the NF-κB-p65 transcription factor in the p16 promoter using the JASPAR core database ( Bryne et al . , 2008 ) , five putative binding sites were identified close to the transcription start site . ChIP primers targeting these sites were designed by Primer Premier . ChIP assays were carried out using a ChIP kit ( CST , Massachusetts , USA ) and a p65 antibody obtained from Abcam . The relative binding of NF-κB-p65 to p16 was assessed by PCR , followed by digital imaging of agarose gels . WT and mutant CDKN2A gene promoter segments were synthesized by Promoterbio Lab ( Taizhou , China ) and then cloned into the pGL4 . 23‐basic luciferase vector to obtain the pGL4 . 23-p16-wt and pGL4 . 23-p16-mut plasmids: control plasmid ( 0 . 1 μg ) +pGL4 . 23‐basic vector ( 0 . 1 μg ) +Renilla plasmid ( 0 . 01 μg ) , control plasmid+pGL4 . 23‐WT p16 promoter vector+Renilla plasmid , control plasmid+pGL4 . 23‐mutant p16 promoter vector+Renilla plasmid , NF-κB-p65 sequence plasmid+pGL4 . 23‐basic vector+Renilla plasmid , NF-κB-p65 sequence plasmid+pGL4 . 23‐WT p16 promoter vector+Renilla plasmid , and NF-κB-p65 sequence plasmid+pGL4 . 23‐mutant p16 promoter vector+Renilla plasmid , separately with Lipofectamine2000 ( Thermo Fisher , Massachusetts , USA ) . The cells were incubated in normal culture medium for 48 hr after transfection . Luciferase assays was implemented after the cells were collected and lysed . And then luciferase activity was standardized to Renilla luciferase activity . All analyses were carried out by SPSS software ( version 20 . 0 , USA ) . Mean ± SD was used to present the data . To compare differences between groups , one‐way ANOVA and student's t-test were used . After analysis using a chi‐square test , qualitative data are presented as percentages . All graphs were generated using GraphPad Software ( version 5 . 0 . 0 , USA ) . P values were two‐sided , and p<0 . 05 indicated statistical significance .	Neck and shoulder pain , lower back pain and leg numbness are conditions that many people will encounter as years go by . This is because intervertebral discs , the padding structures that fit between the bones in the spine , degenerate with age: their cells enter a ‘senescent’ , inactive state , and stop multiplying . A protein known as p16 , an important regulator of cell growth and division , is known to accumulate in senescent cells . In fact , in mouse fat tissue , muscles or eyes , removing the cells that contain high levels of p16 delays aging-associated disorders . However , it was still unknown whether deactivating the gene that codes p16 in senescent cells could delay disc degeneration . Here , Che , Li et al . discovered that p16 is highly present in the senescent cells of severely degenerated human intervertebral discs . The cells in the nucleus pulposus , the jelly-like and most critical tissue in the intervertebral discs , were extracted and grown in the lab under conditions that replicate the early stages of damage to the spine . Drugs and genetic manipulations were then used to decrease the amount of p16 in these cells . The experiments showed that reducing the levels of p16 results in the senescent cells multiplying more and showing fewer signs of damage and aging . In addition , the discs of mice in which the gene that codes for p16 had been deleted were less prone to degeneration compared to ‘normal’ mice in similar conditions . Overall , the work by Che , Li et al . shows that inhibiting p16 in disc cells delays the aging process and reduces the degeneration of intervertebral discs . These findings may one day be applicable to people with intervertebral disc diseases who , for example , could potentially benefit from a gene therapy targeting the cells which produce p16 .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "immunology", "and", "inflammation" ]	2020	p16 deficiency attenuates intervertebral disc degeneration by adjusting oxidative stress and nucleus pulposus cell cycle
Internal states can profoundly alter the behavior of animals . A quantitative understanding of the behavioral changes upon metabolic challenges is key to a mechanistic dissection of how animals maintain nutritional homeostasis . We used an automated video tracking setup to characterize how amino acid and reproductive states interact to shape exploitation and exploration decisions taken by adult Drosophila melanogaster . We find that these two states have specific effects on the decisions to stop at and leave proteinaceous food patches . Furthermore , the internal nutrient state defines the exploration-exploitation trade-off: nutrient-deprived flies focus on specific patches while satiated flies explore more globally . Finally , we show that olfaction mediates the efficient recognition of yeast as an appropriate protein source in mated females and that octopamine is specifically required to mediate homeostatic postmating responses without affecting internal nutrient sensing . Internal states therefore modulate specific aspects of exploitation and exploration to change nutrient selection . Nutrition is key for optimizing the evolutionary fitness of animals . Accordingly , many organisms are able to select the nutrients that fulfill their current needs . Recent work has highlighted the importance of the balance of dietary carbohydrates and proteins/amino acids ( AAs ) for overall mortality , fecundity and lifespan in most species ( Fontana and Partridge , 2015 ) ranging from Drosophila ( Grandison et al . , 2009; Lee et al . , 2008; Skorupa et al . , 2008 ) to rodents ( Solon-Biet et al . , 2014 , 2015 ) and humans ( Levine et al . , 2014 ) . The emerging picture is that there is a trade-off between reproduction and longevity driven by the protein-to-carbohydrate ratio in the diet: a low ratio extends lifespan but reduces reproductive output , while a high ratio reduces lifespan but promotes offspring production ( Simpson et al . , 2015 ) . The mechanisms by which the brain shapes behavioral output during dietary balancing to solve this ethologically relevant trade-off are still largely unknown . Significant advances have been made in our understanding of the neural circuitry underlying decision-making ( Barron et al . , 2015; Lisman , 2015 ) . But we are only beginning to understand how the internal state of an animal dictates the selection of specific actions ( Krashes et al . , 2009; Sternson , 2013 ) . This question becomes particularly relevant in value-based decision making , such as nutrient balancing , where the value of the available options is dependent on the current needs of the animal ( Itskov and Ribeiro , 2013; Ribeiro and Dickson , 2010; Simpson and Raubenheimer , 2012 ) . Thus , the behavioral strategies animals use to adapt nutrient decisions to their internal states provide an ethologically relevant framework to understand how internal states change behavior to mediate value-based decisions . The fly has emerged as an important model to study complex computational tasks due to the availability of sophisticated genetic tools ( Luo et al . , 2008; Olsen and Wilson , 2008 ) , a numerically simple nervous system , and the advent of methods to quantitatively characterize behavior . Advanced computational tools have been applied successfully in Drosophila to study for example chemotaxis ( Gomez-Marin et al . , 2011; van Breugel and Dickinson , 2014 ) , action mapping ( Berman et al . , 2014 ) , aggression and courtship ( Coen et al . , 2016; Dankert et al . , 2009 ) , fly-fly interactions ( Branson et al . , 2009; Schneider et al . , 2012 ) , and predator avoidance ( Muijres et al . , 2014 ) . This recent quantitative approach to behavioral analysis has given rise to the field of computational ethology: the use of computerized tools to measure behavior automatically , to characterize and describe it quantitatively , and to explore patterns which can explain the principles governing it ( Anderson and Perona , 2014 ) . When combined with powerful genetic approaches ( Bath et al . , 2014; Ohyama et al . , 2015 ) the fine description of behavior afforded by these methods will allow us to make significant steps forward in our understanding of the neuronal circuits and molecular pathways that mediate behavior . Flies can detect and behaviorally compensate for the lack or imbalance of proteins and amino acids in the food ( Bjordal et al . , 2014; Ribeiro and Dickson , 2010; Vargas et al . , 2010 ) and adapt their salt and protein intake to their current mating state ( Walker et al . , 2015 ) . The current nutrient state is thought to be read out directly by the nervous system through the action of nutrient-sensitive mechanisms such as the TOR and GCN2 pathways ( Bjordal et al . , 2014; Chantranupong et al . , 2015; Ribeiro and Dickson , 2010 ) . Mating acts on salt and yeast appetite via the action of male-derived Sex Peptide acting on the Sex Peptide Receptor in female reproductive tract neurons , and the resultant silencing of downstream SAG neurons ( Feng et al . , 2014; Ribeiro and Dickson , 2010; Walker et al . , 2015 ) . SAG neurons have been proposed to then change chemosensory processing to modify nutrient intake ( Walker et al . , 2015 ) . The recent development of technologies that can measure the flies’ feeding behavior quantitatively ( Itskov et al . , 2014; Ro et al . , 2014; Yapici et al . , 2016 ) gives access to the fine structure of the feeding program , and how flies homeostatically modulate this program according to their internal state . However , the further structure of foraging decisions , such as arriving at or leaving a specific food patch , and how flies balance the trade-off between exploiting a needed nutrient resource and exploring the surrounding environment to discover new resources , is still poorly understood . Understanding how internal states change the behavioral strategies of an animal should allow us to understand how the animal manages to maintain nutrient homeostasis . Here , we developed a quantitative value-based decision making paradigm to study the foraging strategies implemented by adult Drosophila melanogaster to reach protein homeostasis . We use an automated video tracking setup to characterize the exploitation and exploration of sucrose and yeast patches by flies in different dietary amino acid and mating states . We found that metabolic state and mating modulate the decisions to stop at a yeast patch and leave it . Furthermore , we describe how the internal deficit of dietary amino acids increases exploitation of proteinaceous patches and restricts global exploration and how these behaviors dynamically shift towards increasing exploration as the fly reaches satiation . Importantly , we provide two examples on how our paradigm can be used in the dissection of the genetic and neuronal mechanisms underlying nutrient decisions: First , we show that olfaction is not required to reach protein homeostasis , but that it mediates the efficient recognition of yeast as an appropriate food source in mated females . Second , we show that octopamine mediates homeostatic postmating responses , but not the effects of internal sensing of amino acid deprivation state . Our study provides a quantitative description of how the fly changes behavioral decisions to achieve homeostatic nutrient balancing as well as initial insights into the mechanisms underlying protein homeostasis . Animals are able to adapt their feeding preference towards a particular food in response to their current needs ( Dethier , 1976; Griffioen-Roose et al . , 2012; Itskov and Ribeiro , 2013; Warwick et al . , 2009 ) . However , the behavioral strategies used by animals to make feeding decisions according to their internal state are currently largely unknown . To capture how flies decide what food to eat , we built an automated image-based tracking setup ( Figure 1A ) that captures the position of a single Drosophila melanogaster in a foraging arena ( Figure 1B ) containing 9 yeast patches ( amino acid source ) and 9 sucrose patches ( carbohydrate source ) of equal concentration . 10 . 7554/eLife . 19920 . 003Figure 1 . Automated monitoring of nutrient choices using image-based tracking . ( A ) Schematic of the image-based tracking setup . ( B ) Schematic of the foraging arena , containing an inner flat circular area with 9 sucrose ( carbohydrate source ) and 9 yeast ( amino acid source ) patches . All patches had a concentration of 180 g/L of the corresponding substrate . Each food patch has an approximate diameter of 3 mm which is approximately the body length of the experimental flies . ( C ) Example of the kinematic parameters and behavior classification associated to the representative trajectory shown in ( D ) . Dashed gray horizontal lines indicate the thresholds used for behavior classification , definition of yeast and sucrose micromovements and food patch visits ( see materials and methods ) . Dashed orange rectangle marks the beginning and end of the yeast visit ( see inset in D ) . The different colors in the ethogram correspond to the behaviors labeled with the same color in ( D ) . ( D ) Representative trajectory of a fly walking in the arena . Filled circles represent food patches . Gray and colored trajectories correspond to head and body centroid position , respectively . Small arrows in between both trajectories indicate body orientation . The color code for the different behaviors is indicated by the colored labels . Inset: a yeast visit is defined as a group of consecutive yeast micromovements , in which the head distance to the center of the food patch was never >5 mm ( gray dashed line in the main trajectory ) . ( E , G ) Total duration of yeast and sucrose micromovements for virgin , n = 15 ( E ) and mated , n = 26 ( G ) female flies fed with the AA+ rich diet . ( F , H ) Distribution of yeast and sucrose micromovement durations for virgin ( F ) and mated ( H ) female flies fed with the rich diet . Bin size: 2 . 2 s . p<0 . 05 , *p<0 . 001 , significance was tested by Wilcoxon rank-sum test . In panels E and G and in the following figures in which boxplots are used , the black line represents the median , colored boxes represent inter-quartile range ( IQR ) and gray dots represent the value of the y-axis parameter for single flies . DOI: http://dx . doi . org/10 . 7554/eLife . 19920 . 00310 . 7554/eLife . 19920 . 004Figure 1—figure supplement 1 . Ground-truthing of behavior . ( A ) Normalized histogram of head speed of amino acid-deprived mated females ( AA- diet ) from two independent video tracking experiments: orange lines represent data obtained from an assay in which the arena contained 9 yeast and 9 sucrose patches; black lines represent data obtained from an assay in which the arena contained 18 agarose patches ( no food ) . For each experiment , the speed was calculated for periods inside and outside food patches . Vertical dashed gray lines in main panel and insets indicate the speed thresholds used to classify resting ( 0–0 . 2 mm/s ) , micromovement ( 0 . 2–2 mm/s ) and walking ( >2 mm/s ) . Insets are a zoom-in of the indicated regions of the main histogram . Black and orange lines indicate mean and shaded area s . e . m . ( B ) Normalized histogram of the head speed displayed during manually annotated behaviors . Time indicates the total length of the scored behaviors . ( C ) Proportion of manually annotated behaviors observed during yeast ( left ) or sucrose ( right ) micromovements . ( D ) Cumulative histogram of head positions during the first annotated proboscis extension in a yeast patch . DOI: http://dx . doi . org/10 . 7554/eLife . 19920 . 004 The distribution of the food patches was designed to promote frequent encounters with food sources , such that nutritional decisions , rather than food finding , determine the fly’s food exploitation strategies . We recorded the behavior of the fly over two hours during these nutritional decisions , and developed custom software to track the position of the fly’s body and head centroids ( all tracking data generated in this study are available for download from the Dryad repository [Corrales-Carvajal et al . , 2016] ) . We then extracted multiple kinematic parameters ( see Materials and methods for detailed list ) and computed the locomotor activity and the distance of the fly from each food patch during the whole duration of the assay ( Figure 1C and D and Video 1 ) . Upon a detailed analysis of the distribution of head speeds when the flies were inside or outside food patches ( Figure 1—figure supplement 1A ) we decided to use two speed thresholds to split the locomotor activity of the flies into three types: resting ( speed ≤ 0 . 2 mm/s ) , micromovement ( 0 . 2 mm/s < speed ≤ 2 mm/s ) and walking ( speed > 2 mm/s ) . Furthermore , slow walking bouts ( 2 mm/s < speed < 4 mm/s ) that were coupled with a rapid change in angular speed were defined as sharp turns ( 2 mm/s < speed < 4 mm/s and \|angular speed\| ≥ 125°/s ) ( Figure 1C and D ) . 10 . 7554/eLife . 19920 . 005Video 1 . Behavior classification during nutrient decisions . A 20-s-segment of the trajectory depicted in Figure 1C–D , starting on second 40 and following the same color code . The first 7 s of the video are slowed-down 0 . 5 x , as indicated by the white label at the top right corner of the video frame with the fly . DOI: http://dx . doi . org/10 . 7554/eLife . 19920 . 005 To characterize the behaviors that occur during these defined locomotor activity types , we manually annotated resting , feeding , grooming and walking events and assigned them to the corresponding speed profiles . In agreement with previous studies ( Martin , 2004; Robie et al . , 2010; Zou et al . , 2011 ) , we found that more than 80% of the speeds displayed during manually annotated resting or walking periods were below 0 . 2 mm/s or above 2 mm/s , respectively ( Figure 1—figure supplement 1B ) . Furthermore , we reasoned that micromovements could correspond to either grooming or feeding . Indeed , 70% of grooming fell in the micromovement category; while for manually-annotated feeding bouts , half of these periods were categorized as micromovements , the other half occurred at low speeds and were thus classified as resting . However , flies showed a very low rate of proboscis extension during feeding bouts at <0 . 2 mm/s ( data not shown ) and we therefore reasoned that these slow bouts had little contribution to the amount of food ingested . For this reason , we decided to use the time the fly was performing micromovements when its head was in contact with the food patch as a proxy for the time the fly spent feeding ( henceforth termed yeast micromovements or sucrose micromovements ) . To strengthen the argument that these micromovement periods within a food patch represented mostly feeding bouts and not grooming , we used the annotated video segments to quantify the percentage of feeding and grooming during a food micromovement bout . Indeed , we observed that 92 . 2% of the yeast micromovements and 70 . 6% of the sucrose micromovements corresponded to feeding bouts ( Figure 1—figure supplement 1C ) . Hence sucrose and yeast micromovements are a good way to capture the periods the fly spends feeding on a food patch . To start exploring how flies with different internal states react to the different foods , we used this metric to characterize the behavior of virgin and mated females that were previously fed a rich diet . Virgin flies displayed a preference for sucrose over yeast over the total time of the assay , while the opposite was observed in mated females ( Figure 1E and G ) . A closer look at the duration of micromovements on the two food sources , revealed very similar duration profiles between yeast and sucrose for virgin females , while a higher prevalence of long events ( ≥20 s ) on yeast when compared to sucrose was observed in mated flies ( Figure 1F and H ) . These results suggest that for mated females , yeast has a higher salience as food source , even in fully-fed conditions . These observations are in accord with previous reports showing that mating leads to a switch in yeast preference in flies ( Ribeiro and Dickson , 2010; Vargas et al . , 2010; Walker et al . , 2015 ) . Thus , the analysis of food micromovements allows us to capture previously-described changes in food preference elicited by mating . Furthermore , these results demonstrate that one way in which mating increases yeast preference is by inducing long feeding bouts , allowing us to make first conclusions about the behavioral mechanisms behind changes in food choice . A key question in nutritional neuroscience is how animals homeostatically compensate for the lack of specific nutrients ( Dethier , 1976; Itskov and Ribeiro , 2013; Simpson and Raubenheimer , 2012 ) . A concrete example of this homeostatic regulation of feeding behavior is the robust increase in preference for yeast when flies are deprived of proteinaceous food ( Ribeiro and Dickson , 2010; Vargas et al . , 2010 ) . To study the behavioral strategies underlying nutritional homeostasis , we manipulated the metabolic state of the flies by letting them feed ad libitum on a chemically defined ( holidic ) medium ( Piper et al . , 2013 ) during three days prior to the foraging assay . This holidic medium allows us to specifically manipulate amino acids ( AA ) in the diet , leaving the other macronutrients and micronutrients intact . Previous work has identified three different AA compositions having different impacts on reproduction in mated females: AA+ rich ( supporting a high rate of egg laying ) , AA+ suboptimal ( supporting a lower rate of egg laying ) and AA- ( leading to a dramatic reduction in egg laying ) ( Piper et al . , 2013; Figure 2A ) . Furthermore , to better understand how internal metabolic state and mating state interact at the behavioral level we also analyzed virgin females pre-fed these different diets . 10 . 7554/eLife . 19920 . 006Figure 2 . Flies increase yeast feeding and micromovements in response to amino acid challenges and mating . ( A ) Graphical representation of the five internal states tested and the resulting reproductive output as reported by Piper et al . ( 2013 ) , all flies were pre-fed during three days with the indicated holidic medium: ( i ) Virgin AA+ rich , ( ii ) Virgin AA+ , ( iii ) Mated AA+ rich , ( iv ) Mated AA+ suboptimal , ( v ) Mated AA− . ( B ) Effect of internal states on the total number of yeast sips obtained using flyPAD assay ( n = 32–43 ) . ( C ) Effect of internal states on the total duration of yeast micromovements obtained from the video tracking assay ( n = 15–35 ) . ( D ) Behaviors displayed by single flies in the five internal states indicated in ( A ) , during the video tracking assay . Each row represents the ethogram of a single fly , following the same color code used in Figure 1D . Yellow: yeast micromovements . Black: sucrose micromovements . Pink: micromovements outside the food patches . Blue: walking bouts . Gray: resting bouts . Green: sharp turns . ( E ) Dynamics of yeast micromovements quantified as the cumulative duration of yeast micromovements . Gray lines correspond to single flies . Thick colored lines indicate median . p<0 . 05 , p<0 . 01 , p<0 . 001 , significance was tested by Wilcoxon rank-sum test with Bonferroni correction . DOI: http://dx . doi . org/10 . 7554/eLife . 19920 . 00610 . 7554/eLife . 19920 . 007Figure 2—figure supplement 1 . flyPAD setup , sucrose sips and yeast sips dynamics . ( A ) Schematic of flyPAD arena , adapted from ( Itskov et al . , 2014 ) . ( B ) Effect of internal states on the number of sucrose sips . Experimental groups are: Virgin ( open circles ) and mated ( closed circles ) females pre-fed three types of holidic media: AA+ rich , AA+ suboptimal and AA− . The concentration of yeast and sucrose in the food patches was the same used in the video tracking assay . ( C ) Cumulative number of yeast sips of flies in the five internal state conditions indicated . Line represents the mean and the shading the s . e . m . DOI: http://dx . doi . org/10 . 7554/eLife . 19920 . 00710 . 7554/eLife . 19920 . 008Figure 2—figure supplement 2 . Sucrose micromovements . Effect of internal states on the total duration of sucrose micromovements obtained from the video tracking assay . DOI: http://dx . doi . org/10 . 7554/eLife . 19920 . 00810 . 7554/eLife . 19920 . 009Figure 2—figure supplement 3 . Fraction of yeast non-eaters and coefficient of variation for yeast micromovements . ( A ) Effect of internal states on the proportion of yeast non-eaters . A yeast non-eater is a fly for which the total duration of yeast visits was lower than 1 min . Significance was tested by a 2 x 2 Fischer’s exact test using the modified Wald method ( Agresti and Coull , 1998 ) with Bonferroni correction . ( B ) Effect of internal states on the coefficient of variation ( CV ) for yeast micromovements ( CV = SD/mean ) . DOI: http://dx . doi . org/10 . 7554/eLife . 19920 . 009 To quantify the microstructure of the feeding behavior of flies with different internal states , we used the flyPAD technology ( Figure 2—figure supplement 1A ) , which allowed us to decompose the feeding motor pattern into 'sips' ( Itskov et al . , 2014 ) . As the number of sips correlates strongly with food intake , this method enabled us to precisely measure the impact of internal states on feeding decisions . Consistent with previous observations ( Ribeiro and Dickson , 2010; Vargas et al . , 2010; Walker et al . , 2015 ) ( Figure 1E ) , virgin flies showed very little interest in yeast during the whole assay , as measured by the total number of yeast sips ( Figure 2Bi ) . Yeast feeding increased with AA deprivation ( Figure 2Bii ) , and mating ( Figure 2Biii ) . Notably , AA-challenged mated females showed a strong increase in the number of yeast sips ( Figure 2Biv and v ) with the highest rate of yeast feeding in mated flies completely deprived of AAs ( Figure 2Bv ) . We next asked whether these differences in feeding behavior could be captured using the yeast and sucrose micromovements measured using the tracking setup . Indeed , we observed that the yeast micromovements increased in the same way as the yeast sips after AA challenges in virgin and mated females ( Figure 2C ) . Importantly none of these internal state changes led to an increase in the total number of sucrose sips ( Figure 2—figure supplement 1B ) or in the total duration of sucrose micromovements ( Figure 2—figure supplement 2 ) , highlighting the dietary specificity of the manipulation and allowing us to focus our subsequent analysis on the fly’s behavior towards yeast patches . Flies are therefore capable of sensing deficits in AAs and of compensating by specifically increasing feeding and micromovements on yeast , an AA-rich substrate . Furthermore , this homeostatic response is modulated by the mating state of the fly . Our tracking approach is therefore now a validated strategy to uncover the changes in behavioral strategies elicited by different internal states and how these changes allow the animal to reach homeostasis . We investigated the dynamics of yeast micromovements by comparing the ethogram of each individual fly along the two hours of the assay and across all the internal state conditions tested ( Figure 2D , yeast micromovements are shown in yellow ) . This type of visualization revealed that the behavior towards yeast was highly variable . The observed increase in the total duration of yeast micromovements across the different internal state conditions seems to come from the combination of two factors: on one hand , the proportion of flies that showed any interest in yeast at all ( Figure 2—figure supplement 3A ) and on the other hand , the strength of the interest displayed by these flies , measured by the total duration of yeast micromovements . The behavior towards yeast was also highly variable across individuals of the same condition . For example , the total duration of yeast micromovements displayed by AA-deprived flies ranged from 5 to 59 min . Still , the initial steep increase in yeast micromovements during the first 30 min of the assay was very consistent ( Figure 2Ev and Figure 2—figure supplement 1C ) . Overall , the variability increased as a function of deprivation ( Figure 2—figure supplement 3B ) . The reaction to internal state changes is therefore highly variable across individuals . However , full AA deprivation in mated females leads to a robust population-wide effect , highlighting the importance of AAs for the animal . To feed , flies need to stay on food patches . We decided to call these events visits ( Figure 1C and inset in Figure 1D ) . A visit is defined as all consecutive bouts of micromovements on the same patch , for as long as the fly stayed in close proximity of the patch . As we observed in the total duration of yeast micromovements , the total duration of yeast visits increased as a result of mating and AA deprivation ( Figure 3A ) . Therefore , we hypothesized that the fly increases yeast intake by changing different aspects of its foraging decisions , such as approaching a patch more often , stopping at it more and/or leaving it less often . We measured approaching , stopping and leaving decisions by quantifying the number of encounters , the fraction of encounters in which the fly stops on a patch ( visits ) and the duration of visits , respectively . One easy way to increase the total time on yeast would be to approach yeast patches more often . However , none of the internal state modifications leading to an increase in yeast intake caused an increase in the total number of yeast encounters ( Figure 3—figure supplement 1A ) . Furthermore , the rate of encounters remains constant across internal states , with the exception of the mated fully AA-deprived females ( Figure 3B ) , which had a low absolute number of encounters ( Figure 3—figure supplement 1A ) . To explain the behavioral changes underlying homeostasis , we focused on the decision to stop at a yeast patch ( Figure 3C ) and leave it ( Figure 3D ) . 10 . 7554/eLife . 19920 . 010Figure 3 . Metabolic state and mating modulate the probability of stopping at a yeast patch and leaving it . ( A ) Effect of internal states on the total duration of yeast visits . Experimental groups are the ones shown in Figure 2: Mated ( filled circles ) and virgin ( open circles ) females pre-fed three types of holidic media: AA+ rich , AA+ suboptimal and AA− . ( B ) Effect of internal states on the decision to approach a yeast patch quantified as the number of yeast encounters per minute of walking outside the food patches ( rate of yeast encounters ) . ( C ) Effect of internal states on the decision to stop at a yeast patch quantified as the fraction of yeast encounters in which the fly stopped at the yeast patch . ( D ) Effect of internal states in the decision to leave a yeast patch quantified as the average duration of yeast visits . ( E ) Combination of foraging strategies ( total number of visits in x-axis and average duration of those visits in y-axis ) to reach different total durations of yeast visits ( green to blue lines ) , for individual AA-challenged mated flies: pre-fed either a suboptimal diet ( yellow circles ) or an AA- diet ( red circles ) . ns , not significant ( p≥0 . 05 ) , p<0 . 05 , p<0 . 01 , p<0 . 001 , significance was tested by Wilcoxon rank-sum test with Bonferroni correction . DOI: http://dx . doi . org/10 . 7554/eLife . 19920 . 01010 . 7554/eLife . 19920 . 011Figure 3—figure supplement 1 . Yeast encounters and probability of leaving . ( A ) Effect of internal states on the absolute number of yeast encounters . ( B ) Complementary cumulative distribution function for yeast visit durations . Single dots represent one yeast visit . All yeast visits belonging to all animals of the same internal state condition were pooled , ranked and plotted in the same color as indicated in ( Clauset et al . , 2009 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 19920 . 011 We found that in virgins , AA deprivation had a specific effect as it only modulated the decision to leave a patch , with deprived virgins showing longer visits ( Figure 3D and Figure 3—figure supplement 1B ) . Mating also modulated the decision to leave , as fully-fed mated females took longer to leave a yeast patch than virgins ( Figure 3D and Figure 3—figure supplement 1B ) , and , to a smaller degree , had a higher probability of stopping at a proteinaceous food patch upon encounter ( Figure 3C ) . Surprisingly , pre-feeding mated flies with the suboptimal diet caused a dramatic increase in the probability of stopping at a yeast patch ( Figure 3C ) . The strong effect on the decision to stop shows that flies are able to homeostatically modify their behavior in response to even subtle dietary differences that have a negative impact on their fitness ( Piper et al . , 2013 ) . This is even more striking considering that the removal of all AAs does not lead to further changes in the stopping and leaving decisions , despite its drastic impact on egg production and yeast feeding ( Figure 2 ) . We showed above that there is considerable variability across individuals in their behavioral response towards yeast . This was also the case for the strategies each mated fly chose to compensate for both AA challenges . We observed that these flies reached the same total times on yeast by mixing strategies in different ways: some flies had fewer but longer visits , while others had a higher number of visits , but each visit was shorter ( Figure 3E ) . Taken together , these data show that both metabolic and mating states significantly change the decisions to stop at a yeast patch and leave it . Furthermore , the strongest effect is observed when both states act together , as seen in AA-challenged mated females . The specific behavioral strategy each fly employs to reach homeostasis , however , varies widely . We have shown that AA deprivation leads to a 1 . 6-fold increase in yeast feeding when compared to the suboptimal diet treatment ( Figure 2B ) . Surprisingly , however , a change of this magnitude is not visible in the total duration of the yeast visits ( Figure 3A ) , nor is this homeostatic effect recapitulated in changes in specific foraging decisions ( Figure 3 ) . We therefore speculated that instead of modulating exploratory decisions , a lack of AAs could increase the motivation of the flies to exploit the yeast patch . Indeed , the time course of yeast visits ( Figure 4A and Figure 4—figure supplement 1A ) shows that AA-deprived flies displayed a sharp increase in the total duration of yeast visits during the first minutes , while flies pre-fed a suboptimal AA diet displayed a much more delayed and shallower increase in this parameter . As these early visits were also longer ( Figure 4—figure supplement 1B ) , we measured the time it took each fly to engage in its first 'long' ( ≥30 s ) visit ( Figure 4—figure supplement 1C and D ) , and found that AA-deprived flies indeed attained their first long yeast visit much sooner than flies fed a suboptimal diet ( Figure 4B ) : the median latency for AA-deprived flies was just 4 . 38 min ( IQR = 2 . 08–7 . 7 ) , which was three times shorter than the 12 . 37 min ( IQR = 19 . 87–9 . 86 ) observed in flies fed with the suboptimal diet . These results therefore suggest that AA-deprived flies are indeed more motivated to exploit yeast patches . 10 . 7554/eLife . 19920 . 012Figure 4 . The lack of dietary AAs increases exploitation and local exploration of yeast patches . ( A ) Rolling median of the total duration of yeast visits using a 5 min window and a step of 4 min for flies pre-fed a suboptimal diet ( yellow ) or AA− diet ( red ) . ( B ) Effect of AA deprivation on the time elapsed until the fly engages in the first 'long' ( ≥30 s ) yeast visit . ( C ) Histogram of the x-y relative position of all mated flies pre-fed a suboptimal diet ( left ) or a AA− diet ( right ) with respect to the center of the yeast patch ( 0 , 0 ) . The pixel color indicates the fraction of time that flies in the indicated condition spent in the corresponding location bin . ( D ) Effect of AA deprivation on the average minimum distance to the center of the yeast patch , during a yeast visit . ( E ) Effect of AA deprivation on the average area covered during a yeast visit . ( F ) Example trajectories of head position during a yeast visit for a fly of the indicated condition . Hot colors indicate higher head speeds . ( G–J ) Effect of AA deprivation on the locomotor activity of mated flies during yeast visits: ( G ) average histogram of head speeds , ( H ) body centroid speed , ( I ) angular speed and ( J ) proportion of the indicted behaviors during yeast visits . ns , not significant ( p≥0 . 05 ) , p<0 . 01 , *p<0 . 001 , significance was tested by Wilcoxon rank-sum test with Bonferroni correction . Panels B , D , E , H–J compare the indicated parameters between mated flies pre-fed a suboptimal diet ( yellow ) and mated flies pre-fed an AA− diet ( red ) . DOI: http://dx . doi . org/10 . 7554/eLife . 19920 . 01210 . 7554/eLife . 19920 . 013Figure 4—figure supplement 1 . Yeast visits dynamics and latency . ( A , B ) Rolling median of the total duration of yeast visits ( A ) and average duration of yeast visits ( B ) using a 5 min window and a step of 4 min for flies fed a AA+ rich diet ( purple ) , AA+ suboptimal diet ( yellow ) , and AA-deprived flies ( red ) . ( C , D ) Ethograms from Figure 2D showing the latency to engage in the first 'long' ( ≥30 s ) yeast visit for each fly of the indicated condition as a blue dot . DOI: http://dx . doi . org/10 . 7554/eLife . 19920 . 01310 . 7554/eLife . 19920 . 014Figure 4—figure supplement 2 . No effect in local exploration of yeast patches for flies pre-fed a suboptimal diet . ( A ) Effect of AA challenges on the average minimum distance to the center of yeast patches , during a yeast visit . ( B ) Effect of AA challenges on the body centroid speed , during a yeast visit . ( C ) Effect of AA challenges on the angular speed , during a yeast visit . DOI: http://dx . doi . org/10 . 7554/eLife . 19920 . 01410 . 7554/eLife . 19920 . 015Figure 4—figure supplement 3 . Modulation of yeast feeding program microstructure by AA challenges . ( A ) Schematic of feeding program microstructure . Two components of the feeding microstructure can be modulated to reach protein homeostasis: the number of sips inside each feeding burst ( blue shading ) and the inter-burst interval ( IBI ) . ( B ) Mean inter-burst-interval duration . ( C ) Mean number of yeast sips inside a feeding burst . ( D ) Histogram of the inter-sip-interval durations for the indicated internal states . ( E ) Histogram of the sip durations for the indicated internal states . In D and E the lines represent the mean and the shaded areas the s . e . m . DOI: http://dx . doi . org/10 . 7554/eLife . 19920 . 015 We next asked if AA deprivation could also induce differences in the way flies behaved on the yeast patches . When we plotted the distribution of the positions of the flies on the proteinaceous food patches , we observed that AA-deprived flies covered the patches more homogeneously than flies kept on a suboptimal diet , which preferred to stay at the edge of the patch Figure 4C ) . In fact , deprived flies ventured much more into the food patch as quantified by the fact that during a visit , their average minimum distance from the patch center was much smaller ( Figure 4D ) and that they covered a larger area of the resource ( Figure 4E ) . These data suggest that AA-deprived flies are not only more motivated to start exploiting a yeast patch but are also more active while on the food patch . This was further supported when we quantified locomotor activity during each visit to yeast . As visible in the two example trajectories displayed in Figure 4F , we observed that deprived flies were more active , displaying higher linear ( Figure 4G and H ) and angular speeds ( Figure 4I ) . Accordingly , these flies had fewer resting bouts and more sharp turns ( Figure 4J ) . These changes in behavior observed on the food patch were only induced by a complete lack of AAs , as there was no difference in these parameters between mated females pre-fed the rich diet versus those pre-fed the suboptimal diet ( Figure 4—figure supplement 2 ) . All these data are in agreement with an increase in yeast exploitation upon full AA deprivation . Flies lacking AAs would be more 'eager' to exploit and therefore ingest yeast , leading to a strong increase in yeast feeding as observed using the flyPAD ( Figure 2B ) . Animals homeostatically increase food intake upon food deprivation , by changing the micro-structure of their feeding motor pattern ( Davis and Smith , 1992; Itskov et al . , 2014 ) ( Figure 4—figure supplement 3A ) . As video tracking does not give us access to the fine structure of the proboscis motor program , we used the flyPAD technology to characterize the changes in the microstructure of feeding upon AA deprivation . Pre-feeding flies a suboptimal diet led to a decrease only in the inter-burst-interval ( IBI ) when compared to flies kept on a rich diet ( Figure 4—figure supplement 3B ) while the number of sips in each feeding burst did not change ( Figure 4—figure supplement 3C ) . Full AA deprivation , however , led to a strong increase in the number of sips per burst with only a mild further decrease in the IBI . These effects are very similar to those observed upon mating in yeast-deprived females , which leads to both a decrease in the inter-burst interval and an increase in the number of yeast sips per burst ( Walker et al . , 2015 ) . The volume ingested during a feeding bout is a product of the duration of that bout and the feeding rate . Therefore , we analyzed the rhythmic feeding motor pattern and observed that it was only slightly modified by dietary AA levels ( Figure 4—figure supplement 3D and E ) . The mode of the inter-sip-interval distributions decreased from 0 . 08 s in mated females pre-fed the rich diet to 0 . 07 s when pre-fed the suboptimal diet ( p=0 . 0045 , Wilcoxon rank-sum test with Bonferroni correction ) , while no further change was observed when they were pre-fed the AA− diet ( 0 . 07 s , p=1 ) . However , the mode of the sip duration distributions did not change when mated flies pre-fed a suboptimal diet were compared to females kept on a rich diet ( 0 . 12 s , p=0 . 1196 ) , but it decreased when flies were pre-fed the AA− diet ( 0 . 11 s , p=0 . 0067 ) . Taken together , while AA deprivation has minimal effects on the decision to stop at a proteinaceous food patch and leave it , this metabolic manipulation leads to drastic changes in its exploitation . The described changes in activity are likely to support an increased intake of the yeast resource , which is further promoted by a change in the feeding motor pattern of the fly . The increase in exploitation can also be interpreted as an increase in local , resource-directed exploration which could aid the micro-optimization of food intake within non-homogenous natural food patches . The data presented above clearly demonstrate that different internal states interact to modulate food exploitation . But what could be the effects of internal states on the exploratory behaviors of flies ? In order to capture how far the fly would forage to reach the next yeast patch , we calculated three types of transition probabilities: transitions to the same yeast patch , transitions to adjacent yeast patches , and transitions to distant yeast patches . We found that mated flies fed the rich diet had a high probability of transitioning to distant yeast patches ( 75% ) ( Figure 5A and D ) , and a lower probability of transitioning to adjacent food patches ( 25% ) ( Figure 5A and E ) . 10 . 7554/eLife . 19920 . 016Figure 5 . Amino acid challenges reduce global exploration and increases revisits to same yeast patch . ( A–C ) Effect of internal states on exploratory behavior of mated females pre-fed with an AA rich diet ( A ) , an AA suboptimal diet ( B ) or an AA− diet ( C ) . Example trajectories show head position during a yeast-yeast transition . Arrows and pie charts indicate the transition probabilities to visit three types of yeast patches: the same ( orange ) , an adjacent one ( blue ) or a distant one ( black ) . ( D–F ) Comparison of the transition probabilities described in ( A–C ) across the different diet treatments in mated females . ( G ) Average distance covered during transitions to yeast visits . ns , not significant ( p≥0 . 05 ) , p<0 . 05 , p<0 . 01 , p<0 . 001 , significance was tested by Wilcoxon rank-sum test with Bonferroni correction . DOI: http://dx . doi . org/10 . 7554/eLife . 19920 . 01610 . 7554/eLife . 19920 . 017Figure 5—figure supplement 1 . Dynamics of yeast-yeast transitions in single flies . Ethograms showing the yeast visits for each fly ( each row is a single fly ) along the 120 min of the video tracking assay , for the indicated condition . Colors indicate if the food patch visited previous to every yeast visit was the same ( orange ) , an adjacent ( blue ) or a distant one ( black ) . Pie charts indicate the accumulated median transition probabilities by the end of the assay , for the indicated condition ( same as Figure 5A–C ) . DOI: http://dx . doi . org/10 . 7554/eLife . 19920 . 017 Strikingly , these flies almost never returned to the yeast patch they had just visited ( Figure 5A and F ) . Fully-fed flies therefore display a high rate of global exploratory activity , traveling larger distances during their transitions ( Figure 5G ) and mainly choosing to visit distant food patches ( as in the example trace ) . Challenging flies with a suboptimal diet ( Figure 5B ) or one lacking all AAs ( Figure 5C ) significantly altered their exploratory behavior: they strongly reduced their probability of transitioning to distant yeast patches ( Figure 5D ) and increased the probability of transitioning to adjacent yeast patches ( Figure 5E ) . Further , in contrast to the fully-fed flies , AA-challenged flies showed a strong increase in their probability of returning to the same yeast patch ( Figure 5F and Figure 5—figure supplement 1 ) . As one would expect , these changes in behavior are also seen as a decrease in the average distance traveled by animals during transitions to yeast ( Figure 5G ) . Dietary AA challenges therefore lead to a switch from global to local exploration ( see example traces in Figure 5A–C ) . One of the most striking changes is the strong increase in returns to the same yeast patch upon AA deprivation . This change in exploratory strategy leads to an effective additional increase in the time on the same yeast patch without requiring a change in the decision to leave it . Taken together , these changes in exploratory strategy should enable the fly to efficiently increase the intake of yeast while minimizing the distance traveled to the next patch . It also allows the fly to focus on a resource whose quality she knows while avoiding testing food patches of unknown qualities , thereby reducing exploratory risk . If yeast exploitation and exploration are indeed regulated by the internal AA state of the fly , we hypothesized that flies should dynamically adapt their behavior as their internal state changes over the course of the assay due to satiation . To capture this effect independently from the varying yeast intake dynamics displayed by each fly , we divided the total duration of yeast micromovements of each fly into four periods , which we called 'yeast quartiles' ( Figure 6A ) . Each yeast quartile consists of 25% of the time that the fly spent in yeast micromovements , but covers a different amount of absolute time in the assay for each fly . 10 . 7554/eLife . 19920 . 018Figure 6 . Flies dynamically adapt their exploitatory and exploratory behavior as their internal AA satiation changes . ( A ) Definition of yeast quartiles based on the total duration of yeast micromovements along the two hours of the video tracking assay for an example fly . Arrows indicate start and end points of each yeast quartile . Each yeast quartile consists of 25% of the time that the fly spent in yeast micromovements , but covers a different amount of absolute time in the assay for each fly , as shown in ( B ) . ( B ) Example trajectories of head positions during each yeast quartile defined in ( A ) . Red indicates the occurrence of a yeast micromovement . ( C–H ) Effect of yeast satiation on exploration ( C–E ) and exploitation ( F–H ) parameters , for mated AA-deprived flies , quantified during the four yeast quartiles of each fly . As the flies spend more time on yeast , the values of these parameters change towards the values of flies fed with a rich diet . p<0 . 05 , *p<0 . 001 , significance was tested by Wilcoxon rank-sum test . DOI: http://dx . doi . org/10 . 7554/eLife . 19920 . 01810 . 7554/eLife . 19920 . 019Figure 6—figure supplement 1 . Exploitation parameters in AA-deprived flies revert back to fully-fed values . ( A–C ) Exploitation parameters from first yeast quartile ( Q1 ) and fourth yeast quartile ( Q4 ) of AA-deprived mated females compared to the values observed in flies pre-fed a rich and a suboptimal diet along the 2 hr of the video tracking assay . ( A ) Average minimum distance of the head to the center of the yeast patch , ( B ) angular speed during yeast visits . ( C ) Average duration of yeast visits . ns , not significant ( p≥0 . 05 ) , p<0 . 001 , significance was tested by Wilcoxon rank-sum test with Bonferroni correction . DOI: http://dx . doi . org/10 . 7554/eLife . 19920 . 019 As hypothesized , the flies displayed clear differences in their foraging behavior across the four analyzed quartiles . The effect on exploration was clearly visible in the raw tracking traces for the four quartiles ( Figure 6B ) . As the time spent on yeast increased , the average distance traveled to the next yeast patch ( Figure 6C ) and the probability of visiting a distant yeast patch increased ( Figure 6D ) , while the probability of revisiting the same yeast patch decreased ( Figure 6E ) . Accordingly , parameters related to patch exploitation such as the average minimum distance from the center of the patch ( Figure 6F ) , the angular speed on the yeast patch ( Figure 6G ) , and the average duration of the yeast visit ( Figure 6H ) reverted to the values observed in fully-fed females ( Figure 6—figure supplement 1A–C ) . These results show that flies are capable of dynamically adapting their behavioral strategies according to their current internal state and strengthen the notion that foraging strategies are modified by the AA state of the animal to homeostatically balance the intake of AA-rich foods . Starvation changes olfactory representations of food odors and these changes are thought to be required to find food efficiently ( Beshel and Zhong , 2013; Root et al . , 2011 ) . As a proof of principle of how our setup could be used to uncover the neuronal mechanisms underlying foraging decisions , we decided to analyze the role of olfaction in nutrient homeostasis . Olfactory sensory neurons in Drosophila express two main types of chemosensory receptors: Odorant Receptors ( OR ) and Ionotropic glutamate receptors ( IRs ) ( Rytz et al . , 2013; Vosshall and Stocker , 2007 ) . The OR type of olfactory receptors have been shown to significantly contribute to the olfactory detection of yeast over large distances ( Becher et al . , 2010; Christiaens et al . , 2014 ) and are known to mediate physiological responses to yeast odors ( Libert et al . , 2007 ) . We therefore focused on the function of these receptors in homeostatic yeast feeding by tracking the foraging behavior of flies lacking Orco , a co-receptor essential for OR function ( Larsson et al . , 2004 ) . Unexpectedly , we observed that in general upon AA deprivation , Orco mutants showed a similar total duration of yeast visits as controls ( Figure 7A ) . However , upon closer inspection of the time course of yeast visits , we observed that flies with impaired olfaction had a very long latency to engage in a long yeast visit when compared to the genetic controls ( Figure 7B–D , see also Figure 4A and Figure 7—figure supplement 1 ) . While Orco mutants needed around 25 min to enter into a high yeast exploitation 'mode' ( median = 25 . 58 min , IQR = 15 . 05–30 . 06 ) the genetic controls required only 5–8 min to do so ( Figure 7C ) . 10 . 7554/eLife . 19920 . 020Figure 7 . ORs mediate efficient recognition of yeast as an appropriate food source . ( A ) Orco1/1 AA-deprived flies spend as much total time visiting yeast as AA-deprived control flies ( n = 10–14 ) . ( B ) Rolling median of the total duration of yeast visits using a 5 min window and a step of 4 min . ( C ) Effect of Orco mutation on the latency to engage in the first 'long' ( ≥30 s ) yeast visit . ( D ) Behaviors displayed by Orco1/1and control flies , along the 120 min of the assay . Each row represents the ethogram of a single fly , following the same color code used in Figure 1D . Yellow: yeast micromovements . Black: sucrose micromovements . Pink: micromovements outside the food patches . Blue: walking bouts . Gray: resting bouts . Green: sharp turns . Blue circles indicate the latency ( see C ) of each fly . Arrows indicate example flies shown in ( E ) . ( E ) Top: Example trajectory of head positions of an Orco1/1 AA-deprived fly during the 23-min-long latency period ( first three panels on the left ) and during 45 min after the latency period ( fourth panel ) . Bottom: Example trajectory of head positions of a Canton S AA-deprived fly during the 4-min-long latency period ( first panel on the left ) and from the latency point up to minute 68 ( three panels on the right ) . Highlighted trajectory segments represent yeast encounters ( pink ) and yeast visits ( blue ) . ( F–G ) Exploration and locomotor activity during latency period is not affected in Orco1/1 flies as indicated by the number of yeast encounters ( F ) and the body centroid speed outside food patches ( G ) . ( H ) The first long yeast visit is longer in Orco1/1 flies than in control flies . ( I ) Probability of transition to same yeast patch is higher in Orco1/1 flies than in control flies . ns , not significant ( p≥0 . 05 ) , p<0 . 05 , *p<0 . 01 , significance was tested by Wilcoxon rank-sum test with Bonferroni correction . DOI: http://dx . doi . org/10 . 7554/eLife . 19920 . 02010 . 7554/eLife . 19920 . 021Figure 7—figure supplement 1 . Yeast dynamics of Orco mutant flies . Cumulative duration of yeast micromovements . Gray lines correspond to single flies . Thick colored lines indicate median . DOI: http://dx . doi . org/10 . 7554/eLife . 19920 . 021 Olfaction has been proposed to be important for the fly to locate food sources ( Root et al . , 2011 ) . Orco mutants , however , have plenty of encounters with yeast during the latency period . This is clearly visible in the example trace ( Figure 7E ) where pink dots mark encounters with yeast patches . In fact , the number of encounters of Orco mutant flies with yeast patches was similar to , or even higher than , that of controls ( Figure 7F ) . The increased latency also seems not to be due to an impairment in locomotion , as mutant flies walked as fast when outside the food patches as genetic controls ( Figure 7G ) . These data indicate that in our assay , Orco mutant flies easily find yeast patches but fail to efficiently engage into long yeast visits . If Orco mutant flies are inefficient in stopping at yeast patches , how do they manage to homeostatically compensate for the AA challenge ? We observed that the duration of the first long visit ( Figure 7H ) and the probability of revisiting the same yeast patch ( Figure 7I ) were greater for the Orco mutants than for the controls . These results indicate that mutant flies were either more AA deprived than controls or compensated for their sensory deficit by displaying a generally higher exploitatory motivation . Taken together these results show that , in mated females , OR-mediated olfaction is necessary for efficient recognition of yeast as an appropriate resource but is not required to locate food patches at a short range or to achieve nutritional homeostasis . Neuromodulators are thought to be important in translating internal states into behavioral output ( Taghert and Nitabach , 2012 ) . While octopamine has been shown to mediate the postmating increase in yeast feeding ( Walker et al . , 2015 ) , it has been proposed that it does not contribute to homeostatic changes in feeding behavior ( Yang et al . , 2015 ) . We therefore decided to show that our setup could be used to test possible neuromodulatory effects of octopamine on yeast foraging , using mutants for the gene encoding Tyramine β-hydroxylase ( TβH ) , an enzyme required for the biosynthesis of octopamine in the whole animal . As expected , we observed that in AA-deprived females , the drastic increase in the total duration of yeast visits after mating was greatly reduced in TβhnM18 flies ( Figure 8A and Figure 8—figure supplement 1A ) . Likewise , octopamine also seems to be required to elicit the full increase in the probability of stopping at yeast ( Figure 8B and Figure 8—figure supplement 1B ) and the full increase in the duration of yeast visits ( Figure 8C and Figure 8—figure supplement 1C ) , reiterating our previous observation that these two parameters are modulated by mating ( Figure 3 ) . To test whether octopamine was also required for mediating changes in yeast feeding behavior upon AA deprivation , we used the flyPAD technology . TβhnM18 virgin flies were able to increase the number of sips after AA deprivation to a similar extent as control flies ( Figure 8D and Figure 8—figure supplement 1D ) showing that octopamine is not involved in translating the internal state of AA deprivation into increased yeast intake . Overall , these results confirm that the decisions to stop at a yeast patch and leave it are strongly modulated by mating . They also show that octopamine mediates these postmating responses towards yeast , but is not required to sense the internal AA deprivation state . These results provide a first step towards dissecting the role of octopamine in nutrient homeostasis . 10 . 7554/eLife . 19920 . 022Figure 8 . Octopamine mediates postmating response towards yeast but not internal sensing of AA deprivation state . ( A–C ) Effect of the TβhnM18 mutation on the postmating change in foraging parameters , obtained from the video tracking assay after 1 hr: total duration of yeast visits ( A ) , probability of stopping at a yeast patch ( B ) and average duration of yeast visits ( C ) for Canton S and TβhnM18 females , both AA-deprived . Bars depict difference between median value of mated minus virgin groups for the correspondent parameter . Error bars show 5% and 95% bootstrap confidence intervals ( n = 25–33 ) . ( D ) Effect of Tβh mutation on the increase of yeast sips after AA deprivation in virgin females , quantified using the flyPAD setup . Bars depict difference between median values of AA+ ( suboptimal ) minus AA−deprived groups . Error bars show 5% and 95% bootstrap confidence intervals ( n = 26–34 ) . ns , not significant . ( A–C ) Show statistically significant differences between Canton S and TβhnM18 females , as the confidence intervals don’t overlap . DOI: http://dx . doi . org/10 . 7554/eLife . 19920 . 02210 . 7554/eLife . 19920 . 023Figure 8—figure supplement 1 . Octopamine mediates postmating response to yeast . ( A–C ) Effect of the TβhnM18mutation on the postmating change in foraging parameters , obtained using the video tracking setup: ( A ) total duration of yeast visits , ( B ) probability of stopping at a yeast patch and ( C ) average duration of yeast visits for AA-deprived Canton S and TβhnM18 virgin ( open circles ) and mated ( closed circles ) females . ( D ) Effect of TβhnM18mutation on the increase of yeast sips after AA deprivation in virgin females , using the flyPAD assay . ns , not significant ( p≥0 . 05 ) , p<0 . 05 , p<0 . 01 , p<0 . 001 , significance was tested by Wilcoxon rank-sum test with Bonferroni correction . DOI: http://dx . doi . org/10 . 7554/eLife . 19920 . 023 In order to maintain nutrient homeostasis animals need to be able to adapt their nutrient preferences to their current state . But which behavioral alterations underlie such changes in preference ? Here we use an automated video tracking setup to quantitatively capture the behavioral adaptations to AA and mating state changes that allow the animal to maintain nutrient homeostasis . We started by separating the behaviors flies display towards food into discrete decisions: the decision to approach a food patch , the decision to stop at it , and the decision to leave it ( Figure 9 ) . Strikingly , mating and AA challenges induced compensatory behaviors towards yeast patches but not sucrose patches , indicating that the fly changes its exploitation decisions in a resource specific way . Furthermore , internal state modifications impact specific decisions to a different extent . While mating had a major effect on the probability of a fly leaving a yeast patch , AA challenges strongly increased the probability of stopping at a food patch . Nevertheless , the effect of AA deprivation on the decision to stop at a food patch was strongly dependent on mating suggesting that both internal states act synergistically to increase yeast intake . Furthermore , while full AA deprivation leads to a strong increase in yeast feeding when compared to flies kept on a suboptimal diet , the described decisions were not further altered by this drastic nutritional manipulation . There was , however , a considerable decrease in the latency to visit yeast patches for a long time and a general increase in parameters related to the 'eagerness' of the fly to exploit the resource ( latency to engage on a yeast visit , locomotor activity on the patch and area of patch covered ) . Internal states , therefore , alter feeding in specific ways , allowing the fly either to spend more time on the food through the modulation of patch decisions , or to increase resource exploitation through the modulation of motivation without changing patch decisions . These specific changes allow the animal to dose its exploitatory behavior and hence the intake of nutrients over a large range ( ~17 fold ) to match its current needs . 10 . 7554/eLife . 19920 . 024Figure 9 . Model of behavioral strategies modulated by internal AA state . We propose a model in which virgin flies with high internal levels of AAs display low intake mostly ignore yeast patches upon encounter and have a high probability of leaving the yeast patch upon stopping at it . Internal AA levels decrease as a consequence of poor diets which induce a change in the leaving decision , inducing increased yeast intake . Octopamine mediates the postmating changes in the foraging decisions of stopping at the yeast patch and leaving it upon encounter . As the internal AA levels decrease in mated females , their exploration patterns switch from global exploration to local exploration and multiple returns to the same yeast patch . DOI: http://dx . doi . org/10 . 7554/eLife . 19920 . 024 The specific changes observed in the behavior upon alterations of internal state are in agreement with a modular organization of behavioral control . Such modularity has been previously described in the organization of motor output , such as locomotion ( Kiehn , 2016 ) , swimming ( Huang et al . , 2013 ) , grooming ( Seeds et al . , 2014 ) , and feeding ( Itskov et al . , 2014; Walker et al . , 2015 ) . This accumulated evidence suggests that the nervous system uses different mechanisms and hence circuits to change specific aspects of behavioral outcomes or decisions and that these changes add up to reach a specific goal . In agreement with this model , it has been shown that the impact of different starvation times on gustatory input relies on different mechanisms ( Inagaki et al . , 2012 ) . Similarly , our data show that octopamine is specifically required for mediating the changes in yeast decisions observed upon mating but not upon AA deprivation ( Figure 9 ) . Nevertheless , some decisions , such as the decision to stop at a yeast patch , seem to be synergistically gated by both the mating and the AA state of the fly . It will be interesting to dissect how different internal states act at the circuit level to change behavioral decisions: do they act differentially on specific neuronal populations , or is the observed synergism a reflection of the different internal states acting on the same set of neurons ? While at the population level the effect of internal state manipulations led to stereotypical changes in behavior , the effect of internal state on the decisions implemented varied greatly at the individual level . This effect is reminiscent of the large individual differences observed in human physiology in response to identical diets ( Zeevi et al . , 2015 ) . While such behavioral differences can stem from different metabolic states prior to the experiment , transgenerational effects in metabolism ( Öst et al . , 2014 ) , or differences in the microbiota of the flies ( Broderick and Lemaitre , 2012 ) , there is a real possibility that they also reflect idiosyncrasies in behavior and metabolic susceptibilities to internal state changes at the individual level . Indeed , upon AA challenges , we observed that some flies increased their total time on yeast by having many short yeast visits , while some flies had fewer but longer visits . It will be interesting to investigate if these differences reflect behavioral idiosyncrasies , as observed before in many animals including Drosophila ( Buchanan and de Bivort , 2015; Dingemanse et al . , 2010; Kain and de Bivort , 2012 ) . Differentiating between these two possibilities and identifying the physiological and circuit mechanisms leading to idiosyncrasies will be key to a better understanding of behavior . This is especially relevant for understanding metabolic conditions related to nutrition such as obesity . In order to balance the intake of specific nutrients the animal should be able to specifically change its decisions towards the food source which contains the nutrient it needs . Our data show that this is indeed the case , pointing to a possible important contribution of chemosensory systems to nutrient decisions . Indeed , taste processing has been shown to be changed by the mating state of the animal and to contribute to the adaptation of behavioral decisions such as food choice ( Walker et al . , 2015 ) and egg laying site selection ( Hussain et al . , 2016 ) . The contribution of olfaction to nutrient selection is less well understood . The sense of smell is thought to be mostly important for food search behavior ( Becher et al . , 2010 ) , with starvation changing olfactory sensitivity to improve the finding of a food source ( Root et al . , 2011 ) . Our data suggest that while olfactory-impaired mated flies are able to homeostatically increase yeast intake upon AA deprivation , OR-mediated olfaction still plays an important role in their capacity to do so . Interestingly , olfaction doesn’t seem to be important for locating the food but for identifying yeast as an appropriate food source . These data suggest that flies use multimodal integration to decide which food to ingest . In humans , flavor , the integration of different sensory modalities such as taste and smell , is key to the perception of food ( Verhagen and Engelen , 2006 ) . Similarly , in mosquitoes olfaction acts together with other sensory cues to initiate a meal ( McMeniman et al . , 2014 ) . Identifying the chemosensory basis for yeast feeding decisions might therefore be a powerful way to investigate the neuronal basis of flavor perception . While one would expect that internal states increase food intake by changing exploitation decisions , their effects on exploratory behaviors in our paradigm are not trivial . Exploration is key for animals to find the resources they require and to acquire information about their surrounding environment ( Calhoun et al . , 2014; Hassell and Shouthwood , 1978; Hills et al . , 2015 ) . In our paradigm , however , animals do not need to search for resources as they are readily available . A key question then becomes why animals leave a food patch at all , especially when they are deprived of AAs ( Figure 9 ) . The fact that they still do so means that there is a value in leaving the current food patch , even if that one provides the urgently required nutrients to produce offspring and has not been depleted yet . We can only speculate that there must be an advantage in taking the 'risk' of exploring unknown options and maybe identifying a better resource . Animals might often require other resources and leaving the current food patch might allow them to also explore the availability of these . Flies seem to nevertheless manage their exposure to uncertainty by tuning the spatial properties of their exploration . Their internal states not only define the probability of leaving a food patch , they also define if they will explore locally or more globally . The more deprived they are , the more local their exploratory pattern will be ( Figure 9 ) . Remarkably , while the leaving probability of flies pre-exposed to a suboptimal AA diet and a diet lacking AAs looks identical , their exploratory patterns are very different . For example , AA-deprived flies display a higher rate of returns to the same patch right after leaving it . Therefore , while the neuronal processes determining staying decisions seem not to be altered between both AA-challenged states , full AA deprivation must act on the circuits controlling exploration to strongly increase the probability of revisiting the patch the fly just left . This allows the fly to de facto stay longer on the same food patch without changing its leaving decisions . We would like to propose that the apparent separate regulation of these two aspects of the fly behavior suggests that there are two separate internal state sensing processes regulating exploitation and exploration decisions . The combination of both behavioral and circuit modules would allow the fly to trade off the requirement to exploit specific resources and the 'risk' of exposing itself to resources of unknown or lower quality . Furthermore , it is interesting to note the similarity between the revisits to the same food patch we observed upon strong AA deprivation and the 'dances' observed by Vincent Dethier in the blowfly ( Dethier , 1976 ) . Both phenomena are examples of how animals regulate their search behavior and exposure to uncertainty by modulating the local dynamics of their exploratory behavior , in a state-dependent manner . While the budget theory is a classic aspect of foraging theory , it has recently started to be reassessed . It is mainly controversial if energy-deprived animals , including humans , are more or less risk-prone ( Kacelnik and El Mouden , 2013 ) . Our data suggest that the exploratory behavior of AA-deprived animals minimizes their exposure to uncertainty . The description of how different aspects of risk management are implemented at the behavioral level opens up the opportunity to identify the circuit mechanisms by which internal states control exploration-exploitation trade-offs and therefore how animals decide how much to expose themselves to the unknown . The success of neurogenetics has relied to a large extent on the use of simple binary end-point behavioral assays to perform large-scale unbiased screens ( Leitão-gonçalves and Ribeiro , 2014; Ugur et al . , 2016; Vosshall , 2007 ) . This approach has allowed the field to make important contributions to the molecular and circuit basis of innumerable phenomena , including learning and memory ( Heisenberg , 2015 ) , chronobiology ( Konopka and Benzer , 1971 ) , innate behaviors ( Demir and Dickson , 2005; Yapici et al . , 2008 ) , and sensory physiology ( Larsson et al . , 2004 ) . While identifying these cornerstones of neuroscience was crucial , we are now in a position to start understanding how these mechanisms act at the circuit level to perform more complex computations such as the ones used during decision-making and exploration . This endeavor requires the use of a richer and dynamic description and analysis of behavior ( Gomez-Marin et al . , 2014 ) . We used a combination of computer vision ( Anderson and Perona , 2014 ) and a quantitative , automated capacitance-based behavioral assay ( Itskov et al . , 2014 ) with internal state and genetic manipulations to characterize and identify the behavioral changes allowing the animal to achieve homeostasis . It is interesting to consider that while we identify an important role of OR-mediated olfaction in nutrient decision-making , this would not have been possible using end-point analyses , as the animal manages to compensate for its sensory challenge using alternative means . The use of dynamic , quantitative descriptions of complex behavior therefore enables neuroscientists to decompose these into discrete processes , opening up the possibility to go beyond assigning circuits and molecules to general behaviors to start explaining how they act to control the generation of complex cognitive processes . Unless stated otherwise all experiments were performed with Canton S female flies . Canton S flies were obtained from the Bloomington stock center . Orco1/1 flies were a kind gift of Sofia Lavista-Llanos from the Hansson laboratory ( Larsson et al . , 2004 ) . Orco1/+ flies were obtained by crossing Canton S virgins with Orco1/1 males . TβhnM18 flies were a kind gift of Scott Waddell ( Monastirioti et al . , 1996 ) . Fly rearing , maintenance , and behavioral testing were performed at 25°C in climate-controlled chambers at 70% relative humidity in a 12 hr-light-dark cycle . All experimental and control groups were matched for age and husbandry conditions . The standard yeast-based medium ( YBM ) contained , per liter , 80 g cane molasses , 22 g sugar beet syrup , 8 g agar , 80 g corn flour , 10 g soya flour , 18 g yeast extract , 8 ml propionic acid , and 12 ml nipagin ( 15% in ethanol ) supplemented with instant yeast granules on the surface . To ensure a homogenous density of offspring among experiments , fly cultures were always set with five females and three males per vial and left to lay eggs for three days . Flies were reared in YBM until adulthood . Three different types of holidic medium were used as described previously ( Piper et al . , 2013 ) using the following formulations: 50S200NYaa ( AA+ rich ) , 50S200NHUNTaa ( AA+ suboptimal ) and 50S0N ( AA− ) . Composition of the media is as described in Piper et al . , ( 2013 ) , without food preservatives and only differ in amino acids composition . The proportion of amino acids in 50S200NYaa diet is adjusted to match that in yeast and was considered a rich diet maximizing egg laying ( Piper et al . , 2013 ) . The detailed holidic media compositions can be found in Table 1 . 10 . 7554/eLife . 19920 . 025Table 1 . Composition of holidic medium . DOI: http://dx . doi . org/10 . 7554/eLife . 19920 . 025IngredientStockAmount per literGelling agentAgar20 gSugarSucrose17 . 12 gAmino acids for 50S200NHUNTaaL-isoleucine1 . 82 gL-leucine1 . 21 gL-tyrosine0 . 42 gAmino acids for 50S200NYaaL-isoleucine1 . 16 gL-leucine1 . 64 gL-tyrosine0 . 84 gMetal ionsCaCl2 . 6H2O1000x: 250 g/l1 mlCuSO4 . 5H2O1000x: 2 . 5 g/l1 mlFeSO4 . 7H2O1000x: 25 g/l1 mlMgSO4 ( anhydrous ) 1000x: 250 g/l1 mlMnCl2 . 4H2O1000x:1 g/l1 mlZnSO4 . 7H2O1000x: 25 g/l1 mlCholesterolCholesterol20 mg/ml in Ethanol15 mlWaterWater ( milliQ ) 1 l minus combined volume to be added after autoclavingAutoclave 15 min at 120ºC . All additions below should be performed using sterile techniqueAmino acids for 50S200NHUNTaaEssential amino acid stock solution8 g/l L- arginine monohydrochloride 10 g/l L-histidine 19 g/l L- lysine monohydrochloride 8 g/l L-methionine 13 g/l L-phenylalanine 20 g/l L-threonine 5 g/l L-tryptophan 28 g/l L-valine60 . 51 mlNon-essential amino acid stock solution35 g/l L-alanine 17 g/l L-asparagine 17 g/l L-aspartic acid sodium salt monohydrate 0 . 5 g/l L-cysteine hydrochloride 25 g/l L-glutamine 32 g/l glycine 15 g/l L-proline 19 g/l L-serine60 . 51 mlSodium glutamate stock solution100 g/l L-glutamic acid monosodium salt hydrate15 . 13 mlAmino acids for 50S200NYaaEssential amino acid stock solution23 . 51 g/l L-arginine monohydrochloride 11 . 21 g/l L-histidine 28 . 70 g/l L-lysine monohydrochloride 5 . 62 g/l L-methionine 15 . 14 g/l L-phenylalanine 21 . 39 g/l L-threonine 7 . 27 g/l L-tryptophan 22 . 12 g/l L-valine60 . 51 mlNon-essential amino acid stock solution26 . 25 g/l L-alanine 13 . 89 g/l L-asparagine 13 . 89 g/l L-aspartic acid sodium salt monohydrate 30 . 09 g/l L-glutamine 17 . 89 g/l glycine 9 . 32 g/l L-proline 12 . 56 g/l L-serine60 . 51 mlSodium glutamate stock solution100 g/l L-glutamic acid monosodium salt hydrate18 . 21 mlCysteine stock solution50 g/l L-cysteine hydrochloride5 . 28 mlVitaminsVitamin solution125x: 0 . 1 g/l thiamine hydrochloride 0 . 05 g/l riboflavin 0 . 6 g/l nicotinic acid 0 . 775 g/l Ca pantothenate 0 . 125 g/l pyridoxine hydrochloride 0 . 01 g/l biotin14 mlSodium folate1000x: 0 . 5 g/l1 mlBaseBuffer10x: 30 ml/l glacial acetic acid 30 g/l KH2PO4 10 g/l NaHCO3100 mlOther nutrients125x: 6 . 25 g/l choline chloride 0 . 63 g/l myo-inositol 8 . 13 g/l inosine 7 . 5 g/l uridine8 ml To prepare the 50S200NHUNTaa diet , use the values shaded in blue . To prepare the 50S200NYaa diet , use the values shaded in orange . Groups of 9 to 11 newly hatched ( 0–6 hr old ) female flies of the indicated genotypes were transferred to vials containing fresh standard yeast-based medium ( YBM ) . Three days later , all vials were transferred to fresh standard medium and 4 males were added to half of the vials to obtain mated female flies . After two more days , all vials were transferred once again to fresh standard YBM . On the sixth day , all vials were transferred to fresh holidic media . Flies were left to feed ad-libitum for three days on the holidic media and then tested in the video tracking or flyPAD setups . Single flies were tested in individual arenas that contained two kinds of food patches: yeast ( 180 g/L ) and sucrose ( 180 g/L ) , each mixed with 0 . 75% ( tracking ) or 1% ( flyPAD ) agarose . Flies were individually transferred to the arenas by mouth aspiration and allowed to feed for 1 ( flyPAD ) or 2 ( tracking ) hours , except for the tracking experiment with TβhnM18 mutant flies , which lasted 1 hr . flyPAD data were acquired using the Bonsai framework ( Lopes et al . , 2015 ) and analyzed in MATLAB ( Mathworks , Natick , MA ) using custom-written software , as described ( Itskov et al . , 2014 ) . To avoid patch exhaustion before the end of the tracking assays , each circular patch contained 5 µL of food with a diameter of approximately 3 mm . After each assay , the tracking arenas were washed with soap , rinsed with 70% ethanol , and finally with distilled water . Videos that had more than 10% of lost frames ( due to technical problems during acquisition ) were excluded from the analysis . No further data was excluded . The experiment that compares the conditions AA+ suboptimal and AA− ( results shown in Figures 1–6 ) was performed 3 times independently , which means that an independent set of individuals ( n=15–35 , shown in the corresponding figure legend ) was reared and tested under the corresponding conditions . The experiment comparing AA+ rich vs AA− was performed two times independently . The experiments comparing TβhnM18 or Orco mutant flies with their corresponding controls were performed once with the sample size indicated in the corresponding figure legend . We confirmed that the claims made in this manuscript held for every experimental replicate . We never tested the same individual more than once . The behavioral arenas for the video tracking ( Figure 1B ) were designed and manufactured in-house using a laser-cutter and a milling machine . Material used for the base was acrylic and glass for the lid . The outer diameter of the arena was 73 mm . The inner area containing food patches was flat and had a diameter of 50 mm and a distance to the lid of 2 . 1 mm . To allow a top-view of the fly throughout the whole experiment , the outer area of the arena had 10° of inclination ( Simon and Dickinson , 2010 ) and the glass lid was coated with 10 µL of SigmaCote the night before the assays . Food patches were distributed in two concentric circles equidistantly from the edge . Furthermore , sucrose and yeast patches were alternated such that from a given food patch , there was at least one adjacent yeast and one adjacent sucrose patch . The radius of each food patch was approximately 1 . 5 mm . The minimum distance between the centers of two adjacent food patches is 10 mm . White LEDs 12V DC ( 4 . 8 watt/meter ) , were used for illumination of the arenas . They were placed under the arenas , as backlight illumination and on the walls of the behavioral box , surrounding the arenas , as shown in Figure 1A . A white cardboard arch was used to improve illumination to reflect light towards the arenas ( Figure 1A ) . Three fly arenas were recorded simultaneously from the top using a video camera ( Genie HM1400 camera , Teledyne DALSA , Canada; frame acquisition rate: 50 fps ) connected to a desktop computer using a Gigabit Ethernet connection . Body centroid positions and major axis of the fly body in each frame were extracted using custom off-line tracking algorithms written in Bonsai ( Lopes et al . , 2015 ) and Matlab ( Mathworks ) . The arena diameter in the video was measured to find the correspondence between pixels in the video and mm in the real world ( 1 pixel = 0 . 155 mm ) . The typical length of the major axis of the fly body in a video was 19 pixels ( ~3 mm ) . Video acquisition was made with slight overexposure to obtain a strong contrast between the fly and the arena . Since the fly body was the darkest object in the arena , a pixel intensity threshold was used to obtain the centroid and orientation of the fly blob . The head position was extracted using custom MATLAB ( Mathworks ) software . Head position in the first frame was manually selected . From there on , the head position is automatically propagated to the consecutive frames using a proximity rule ( Gomez-Marin et al . , 2011 ) . This rule , however , does not hold during a jump of the fly . Therefore , in addition to the proximity rule , for the intervals in-between jumps , the head position was automatically corrected using the fact that flies walk forward most of the time . Manual annotation of 510 inter-jump-intervals revealed that 98% were correctly classified . All the body and head centroid tracking data generated in this study are available for download from the Dryad repository ( Corrales-Carvajal et al . , 2016 ) . Raw trajectories of head and body centroids were smoothed using a Gaussian filter of 16 frames ( 0 . 32 s ) width . The width was chosen empirically by comparing the raw and smoothed tracks . The speed was measured from the smoothed coordinates by calculating the distance covered from the current frame and the next frame , divided by the time between them ( 0 . 02 s ) . Similarly , the angular speed was measured by calculating the difference between the heading angle from the current frame and the next frame , divided by the time between them . The heading angle for this calculation was obtained from the head and tail smoothed centroids . Walking and non-walking instances were classified applying a 2 mm/s threshold in the head speed , based on the distribution of head speed for AA-deprived flies in Figure 1—figure supplement 1A and previous studies ( Martin , 2004; Robie et al . , 2010 ) . The head speed used was also smoothed using a Gaussian filter of 60 frames ( 1 . 2 s ) to avoid rapid changes in classification around the thresholds . Sharp turns were classified when a local maximum in the angular speed exceeded a 125°/s threshold , as long as the body centroid speed was below 4 mm/s . A wider Gaussian filter ( width of 2 . 4 s ) was applied to the head speed to classify resting bouts , using a threshold of 0 . 2 mm/s . The remaining events during the non-walking segments that were not classified as resting were classified as micromovements . Manual annotation of 107 feeding events showed that when the head position was at 3 mm or less from the center of the food patch , flies were already close enough to have leg contact . Initially , encounters with a food patch were defined as the moments in which the fly crossed this 3 mm distance threshold . To avoid misclassifying the transient head movement associated with grooming or feeding around this threshold as new encounters , consecutive encounters were merged when the total displacement of the head in any direction was lower or equal than 2 pixels ( 0 . 31 mm ) during the time elapsed in-between the encounters . From each feeding event , the distance from the head of the fly to the center of the patch was also captured . Since 95% of the first proboscis extensions happened below 2 . 5 mm , this was the selected distance threshold to define yeast and sucrose micromovements ( Figure 1—figure supplement 1D ) . In this way , food micromovements were defined as the time in which flies were classified in a micromovement ( see definition in previous section ) and their head was simultaneously inside a circle of 2 . 5 mm around the food patch ( see gray dashed line in Figure 1D inset ) . The two pixels displacement rule used in the definition of encounters was also applied here to avoid definitions of false new micromovements . A visit was defined as a series of consecutive food micromovements ( already corrected for small displacements ) in which the head distance to the center of the food patch was never larger than 5 mm during the time elapsed in-between the food micromovements ( Figure 1D inset ) . 5 mm is the maximum radius of non-overlapping circles around the food patches ( see gray dashed line in main trajectory of Figure 1D ) . This 5 mm threshold was also used to merge consecutive encounters ( consecutive encounters were merged if the head distance to the center of the food patch was never larger than 5 mm during the time in-between encounters ) . In this way , for every visit there is an encounter , but there can be an encounter and no visit if the fly doesn’t stop at the food patch ( food micromovement ) . All of these parameters , unless specified otherwise , were calculated for each fly and for the whole duration of the assay .	When making decisions , animals , including humans , do not always choose the same option . One reason for this is that their “internal state” changes the value of different options . This is particularly evident when deciding what type of food to eat . Depending on which nutrients the animal needs , it will choose to eat different foods . Amino acids are key nutrients that affect health , lifespan and reproduction . Female fruit flies that have recently mated , for example , eat more amino acids in order to obtain the raw materials required to produce eggs . Despite the importance of amino acids , little was known about how animal behavior changes in response to a lack of this nutrient . Corrales-Carvajal et al . used a video tracking system to measure the time that fruit flies – some of which had a need for amino acids – spent feeding on patches of yeast ( which are rich in amino acids ) versus patches of sucrose . Recently mated females – and virgins that had been fed a diet lacking in amino acids – consumed more yeast than sucrose , whereas virgin females that were not amino acid deficient showed the opposite pattern . To bias the fly toward eating the right food for their needs , several aspects of the fly’s behavior changed , including the number and length of individual feeding bouts . These different behaviors did not all change at the same time . The pattern of exploration taken by the flies also depended on their need for amino acids . Amino acid deficient flies spent most of their time near known yeast patches . By contrast , fully fed flies adopted a riskier foraging strategy , moving away from known sources of food to explore their environment more widely . In common with humans , the flies relied upon their sense of smell to efficiently identify different types of food . Overall , the results presented by Corrales-Carvajal et al . provide us with a detailed understanding about how changes to the internal state of the fly affect its behavior . The next step will be to use the powerful genetic tools available for studying fruit flies to reveal the neural circuits and molecular mechanisms that help animals find the types of food that they need .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "neuroscience" ]	2016	Internal states drive nutrient homeostasis by modulating exploration-exploitation trade-off
Metabolic studies suggest that the absorptive capacity of the small intestine for fructose is limited , though the molecular mechanisms controlling this process remain unknown . Here we demonstrate that thioredoxin-interacting protein ( Txnip ) , which regulates glucose homeostasis in mammals , binds to fructose transporters and promotes fructose absorption by the small intestine . Deletion of Txnip in mice reduced fructose transport into the peripheral bloodstream and liver , as well as the severity of adverse metabolic outcomes resulting from long-term fructose consumption . We also demonstrate that fructose consumption induces expression of Txnip in the small intestine . Diabetic mice had increased expression of Txnip in the small intestine as well as enhanced fructose uptake and transport into the hepatic portal circulation . The deletion of Txnip in mice abolished the diabetes-induced increase in fructose absorption . Our results indicate that Txnip is a critical regulator of fructose metabolism and suggest that a diabetic state can promote fructose uptake . The dietary consumption of fructose has drastically increased in modernized societies ( Cox , 2002 ) , and growing evidence implicates fructose consumption in contributing to the worldwide increase in metabolic diseases , such as nonalcoholic fatty liver disease , obesity , and type 2 diabetes mellitus ( Johnson et al . , 2007; Ouyang et al . , 2008 ) . The uptake of fructose by the small intestine controls its availability to organs that are able to metabolize fructose , including liver ( Asano et al . , 1992 ) and kidney ( Sugawara-Yokoo et al . , 1999 ) . Following ingestion , fructose is absorbed by enterocytes in the small intestine , first entering from the intestinal lumen through glucose transporter 5 ( GLUT5 ) on the apical membrane ( Burant et al . , 1992 ) and then exiting the enterocyte into the bloodstream through glucose transporter 2 ( GLUT2 ) on the basolateral membrane ( Gould et al . , 1991 ) , although GLUT2 can translocate to the apical membrane in response to increased fructose levels in the lumen ( Gouyon et al . , 2003; Kellett and Brot-Laroche , 2005 ) . Following transport through the small intestine , fructose travels through the hepatic portal vein to the liver , where it is phosphorylated to fructose 1-phosphate by hepatic fructokinase , bypassing the energy-sensitive enzymatic activity of phosphofructokinase , which generates fructose 1 , 6-biphosphate in the glucose metabolic pathway ( Mayes , 1993 ) . Downstream metabolic intermediates of fructose , including acetyl-CoA , can be directed toward de novo lipogenesis , promoting nonalcoholic fatty liver disease ( Abdelmalek et al . , 2010 ) . The build-up of hepatic triglycerides also promotes insulin resistance and obesity by impairing insulin signaling and increasing global lipid circulation ( Wei and Pagliassotti , 2004 ) . Compared to blood glucose levels ( in the range of 5 mM in humans ) , blood fructose levels are maintained at low levels ( Douard and Ferraris , 2008 ) ( 0 . 008 to 0 . 5 mM in humans ) through the efficient clearance by the liver and , to a lesser extent , by the kidneys ( Mayes , 1993 ) . Thus , the pathogenesis of fructose-associated metabolic disease is dependent on the function of GLUT2 and GLUT5 as the primary transporters of fructose of enterocytes on the small intestine , particularly the duodenum and jejunum , which facilitate most carbohydrate absorption . Txnip , or thioredoxin-interacting protein , is an arrestin-like protein that can bind to thioredoxin protein and that regulates metabolism in mammals ( Shalev , 2014; Patwari and Lee , 2012 ) . We have previously reported that Txnip overexpression represses cellular glucose uptake while eliminating Txnip expression increases glucose uptake in peripheral tissues in both insulin-dependent and insulin-independent manners ( Parikh et al . , 2007 ) . Expression of Txnip is highly correlated with extracellular glucose concentrations which upregulate the activity of the transcription complexes chREBP/Mlx and MondoA/Mlx that bind to the carbohydrate response element ( ChoRE ) on the Txnip promoter to induce Txnip mRNA expression ( Cha-Molstad et al . , 2009; Stoltzman et al . , 2008 ) . Upregulated Txnip inhibits glucose uptake ( Patwari and Lee , 2012 ) by interacting with and altering the expression of glucose transporter 1 ( GLUT1 ) ( Wu et al . , 2013 ) . Because Txnip regulates glucose transport and fructose metabolism may be a significant factor in important metabolic diseases , we studied the effect of Txnip on fructose absorption and on fructose-associated metabolic disease . Given that Txnip is a regulator of glucose homeostasis ( Parikh et al . , 2007; Patwari et al . , 2009 ) , we sought to explore its potential to regulate fructose metabolism , specifically through uptake by the small intestine . In order to determine potential molecular interactions between Txnip and GLUT2 or GLUT5 , we performed a pulldown assay by co-immunoprecipitating TXNIP with either GLUT2 or GLUT5 . We observed an interaction between TXNIP and both GLUT2 and GLUT5 even in the absence of its C247 thioredoxin-interacting site ( Kellett and Brot-Laroche , 2005 ) ( Figure 1 ) , indicating that binding to thioredoxin is not necessary for TXNIP to interact with GLUT2 or GLUT5 . 10 . 7554/eLife . 18313 . 003Figure 1 . Txnip binds to GLUT2 and GLUT5 independently of its thioredoxin-interacting cysteine . hTXNIP-V5 or its C247S-flag mutated form were co-expressed with either hGLUT5 or hGLUT2 in HEK293 cells . Cellular lysates were captured using antibodies that were then bound to Protein A/G agarose beads . The input and captured complexes were then immunoblotted for the proteins of interest . The results indicate that Txnip binds to both GLUT2 and GLUT5 and that the Txnip mutant C247S , which abolishes the molecular interaction between Txnip and thioredoxin , can still bind GLUT5 and GLUT2 . DOI: http://dx . doi . org/10 . 7554/eLife . 18313 . 00310 . 7554/eLife . 18313 . 004Figure 1—source data 1 . Images represent the cropped Western Blot presented in the manuscript on the left-hand side accompanied by the developed film from which it was cropped on the right-hand side . The yellow highlighted regions represent the cropped regions . Bands are in the original order of the membrane unless otherwise specified by the numbering above each band . DOI: http://dx . doi . org/10 . 7554/eLife . 18313 . 004 Having established a molecular interaction between Txnip and both transporters that mediate fructose uptake , we then used Caco-2 cells transiently transfected with human TXNIP , GLUT2 , and GLUT5 overexpressing plasmid to determine if Txnip expression affects the cellular uptake of 14C-radiolabeled D-fructose . The Caco-2 cell line is a human epithelial colorectal adenocarcinoma cell line that when plated as a confluent monolayer resembles the enterocyte lining of the small intestine , both morphologically and functionally ( Hidalgo et al . , 1989; Engle et al . , 1998 ) . Measuring the retention of 14C-radiolabeled D-fructose in confluent Caco-2 cells revealed that simultaneous overexpression of TXNIP with GLUT2 or GLUT5 increased cellular 14C-radiolabeled D-fructose uptake by 161% and 162% ( p<0 . 01 , p<0 . 001; n = 6 ) respectively relative to the empty vector ( EV ) transfection control ( Figure 2a ) . Interestingly , overexpression of either fructose transporter in the absence of TXNIP overexpression was insufficient to produce a marked increase in cellular fructose uptake , suggesting that Txnip expression is required to facilitate the function of GLUT2 and GLUT5 under some circumstances . Overexpression of TXNIP with both fructose transporters produced the greatest increase in cellular fructose uptake . After normalizing the uptake in cells overexpressing TXNIP/GLUT2 or TXNIP/GLUT5 to that of cells overexpressing all three proteins , we found that the normalized ratios of uptake were not different in cells overexpressing TXNIP/GLUT2 and TXNIP/GLUT5 , suggesting that the effects of GLUT2 and GLUT5 may be similar and synergistic in Txnip-mediated fructose uptake . A reciprocal loss-of-function experiment using high-passage mouse embryonic fibroblasts ( MEFs ) isolated from WT or Txnip-null ( Txnip-KO ) mice revealed that Txnip-KO MEFs retained 51 . 3% of the fructose compared to WT MEFs ( p<0 . 001; n = 6 ) ; this reduction in cellular uptake was rescued by transient overexpression of human TXNIP , showing that the deletion of Txnip reduces cellular fructose uptake ( Figure 2b ) . An ex vivo analysis revealed a reduced uptake of 14C-radiolabeled D-fructose of 50 . 9% ( p<0 . 001; n = 6 ) in jejunum isolated from Txnip-KO mice compared to WT mice , indicating that Txnip expression promotes fructose uptake by the small intestine , particularly by the jejunum , which , in conjunction with the duodenum , is the primary site of fructose absorption ( Figure 2c ) . 10 . 7554/eLife . 18313 . 005Figure 2 . Txnip promotes cellular fructose uptake and fructose absorption by the small intestine . ( a ) Empty vector ( EV ) or TXNIP , with GLUT2 and/or GLUT5 , was transiently transfected in Caco-2 cells . Simultaneous TXNIP-overexpression with either GLUT2 or GLUT5 overexpression induced an increase in cellular fructose uptake relative to EV-transfected cells while the greatest increase in uptake was observed in cells overexpressing TXNIP with both GLUT2 and GLUT5 . There was no significant difference in uptake in cells overexpressing TXNIP with GLUT2 in comparison to cells overexpressing TXNIP with GLUT5 ( n = 6 ) . ( b ) High passage mouse embryonic fibroblasts ( MEFs ) from Txnip-KO animals had decreased cellular fructose uptake compared with cells from wild type mice , which was restored by transfecting TXNIP in the same cell line ( n = 6 ) . ( c ) Jejunum from Txnip-knockout mice is able to transport less fructose from intestinal lumen when exposed to [14C ( U ) ]-D fructose compared to small intestine from wild type mice . ( n = 4 ) . ( d ) Following administration of a [14C ( U ) ]-D fructose bolus via oral gavage , Txnip-knockout mice exhibit reduced 14C signal in the blood compared to wild type mice . Other organs , including the liver , kidney , and heart , showed the same trend of reduced 14C signal in Txnip-knockout mice relative to wild-type mice , though the signal in kidney and heart was lower than that of blood . There was no observed difference in the skeletal muscle or brain within the parameters of the experiment ( n = 4 ) . p<0 . 05 , p<0 . 01 , *p<0 . 001 . Data represent mean ± SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 18313 . 00510 . 7554/eLife . 18313 . 006Figure 2—source data 1 . Statistical analysis of Figure 2 . These tables represent the statistical analysis conducted on the raw data collected for Figure 2 using GraphPad Prism 5 . DOI: http://dx . doi . org/10 . 7554/eLife . 18313 . 00610 . 7554/eLife . 18313 . 007Figure 2—source data 2 . Statistical analysis for Figure 2—figure supplement 1 . These tables represent the statistical analysis conducted on the raw data collected for Figure 2—figure supplement 1 using GraphPad Prism 5 . DOI: http://dx . doi . org/10 . 7554/eLife . 18313 . 00710 . 7554/eLife . 18313 . 008Figure 2—figure supplement 1 . Wild type and Txnip-null mouse had no difference in organ 14C D-fructose uptake following intravenous injection . Mice were injected with 14C D-fructose through tail-vein injection . Tissues were collected and measured for 14C signal 30 min after the injection , and 14C signal was normalized to levels found in the peripheral bloodstream . There was no observed difference in 14C D-fructose signal in any of the tissues . ( n = 8 for all groups ) . DOI: http://dx . doi . org/10 . 7554/eLife . 18313 . 00810 . 7554/eLife . 18313 . 009Figure 2—figure supplement 2 . Determination of oral gavage single time point . Levels of 14C were measured in the peripheral blood collected from wild type mice following several time points after oral gavage . The maximum signal was observed in mice after 20 min , which was used in downstream oral gavage experiments . All points were significantly different from the 0 min time point ( n = 3 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 18313 . 009 To evaluate if the reduction in small intestinal fructose uptake caused by the deletion of Txnip regulates the levels of fructose in the peripheral bloodstream and other organs in vivo , we performed an oral gavage of 14C-radiolabeled D-fructose in WT and Txnip-KO mice . We administered 14C-radiolabeled fructose in a 30% solution of fructose and mannitol ( to correct fructose uptake into tissue for the adherent extracellular fluid phase ) and allowed for 20 min of digestion before measuring the retention of 14C-radiolabeled fructose in various tissues of interest . We used 20 min as the standard time point for this experiment , as it was the time point at which the peripheral blood in wild type mice contained the greatest 14C signal ( Figure 2—figure supplement 2 ) , which was also confirmed in our later experiments ( Figure 5 ) . The liver from Txnip-KO mice demonstrated a reduction of 14C-radiolabeled D-fructose to 42 . 4% compared to liver from WT mice ( p<0 . 05; n = 4 ) , suggesting that the transport of fructose from the intestinal lumen to the hepatic portal blood system was reduced in Txnip-KO mice ( Figure 2d ) ; this finding was confirmed in additional data shown below in measurements of hepatic portal vein blood ( Figure 5 ) . The reduced levels of 14C-radiolabeled fructose received and processed by the liver subsequently manifested in peripheral blood in Txnip-KO mice , which had a fructose signal that was 54 . 9% compared to that of WT ( p<0 . 01; n = 4 ) . ( Figure 2d ) . Heart and kidney tissue from Txnip-KO mice also exhibited a decrease in 14C D-fructose signal , but this signal was lower than that of the peripheral blood , indicating that the observed difference may be due to decreased circulating levels of fructose and not by uptake by these organs . Interestingly , the rectus femoris muscle from both groups demonstrated no differences within the parameters of the experiment and had a fructose signal lower than levels found in circulating blood despite previous studies reporting fructose uptake and Txnip expression in skeletal muscle ( Parikh et al . , 2007; Zierath et al . , 1995 ) ( Figure 2d ) . Brain tissue collected from the cerebrum also did not exhibit a difference in 14C D-fructose signal . These results indicate that the deletion of Txnip reduces fructose absorption and subsequent availability to other organs in vivo . To test if the absence of Txnip expression also affects fructose uptake by other organs independent of differences in intestinal uptake , we measured 14C D-fructose signal in several tissues collected 30 min following the intravenous tail injection of 14C-radiolabeled D-fructose , effectively bypassing the small intestine and hepatic portal vein circulation . After normalizing the 14C D-fructose signal of the other organs to that found in the blood , we observed no differences in any of the organs collected , including liver and kidney , both of which have the ability to absorb and process fructose ( Figure 2—figure supplement 1 ) . These results indicate that the effect of Txnip on fructose absorption may be specific to the small intestine . As such , the small intestine may be the primary regulator of fructose access to the body , or other compensatory mechanisms may exist to regulate fructose concentrations in these organs . While the physiological consequences of dietary fructose consumption remain controversial , there is increasing evidence suggesting an association between fructose consumption and metabolic diseases ( Johnson et al . , 2007; Ouyang et al . , 2008; Stanhope , 2016 ) . To determine if the negative effects of chronic fructose consumption are diminished in Txnip-KO mice , which have a marked decrease in fructose absorption , we supplemented a moderate fat diet ( containing 0 . 16% fructose ) of WT and Txnip-KO mice with 30% w/v fructose in their drinking water; this diet is similar to the Western-style diet , which is enriched with saturated fats and sugar ( Stevenson et al . , 2016 ) . We maintained mice on the fructose-supplemented diet ( FSD ) for 25 weeks before measuring differences in physiological parameters in comparison to mice on the moderate fat regular diet ( RD ) . Because male mice on a moderate fat diet develop adverse metabolic outcomes even without supplementary fructose ( Morselli et al . , 2014; Hwang et al . , 2010 ) , we analyzed the effects of the fructose-supplemented diet on female mice . WT mice on the FSD had a higher body weight of 7 . 73 ± 2 . 61 g relative to WT mice on the RD ( mean ± SEM; n = 5 , p<0 . 05 ) , while Txnip-KO mice on both diets maintained similar weights to the WT mice on the RD ( Figure 3a ) . Txnip-KO mice have reduced fasting blood glucose levels and are able to effectively maintain decreased blood glucose levels following insulin injection and reduce blood glucose levels to baseline levels following glucose injection ( Chutkow et al . , 2008 ) . The prolonged FSD caused significant systemic intolerance to intraperitoneal injections of glucose ( 2 g/kg ) as well as reduced response to insulin ( 0 . 25 milliunits/g ) in WT animals ( n = 5 ) , whereas Txnip-KO mice were resistant to these metabolic consequences caused by the FSD ( Figure 3b , c ) . Sirius Red staining of liver sections used to measure hepatic fibrosis revealed that WT mice on the FSD had a nearly five fold increased hepatic collagen content relative to the WT mice on the RD ( n = 5 , p<0 . 01 ) , while Txnip-KO mice had comparable levels to WT mice on the RD regardless of the diet they were fed ( Figure 3d ) . Liver sections graded using a previously described method for hepatocellular steatosis ( Burgess et al . , 2011 ) revealed that WT mice on the FSD scored significantly worse compared to mice on the RD ( n = 5 , p<0 . 001 ) , indicating more liver fat content and steatosis; the severity of steatosis was significantly reduced in Txnip-KO mice on the FSD ( [p<0 . 001] [Figure 3e] ) . Together , these data demonstrate that the Txnip-KO mice on the FSD had reduced severity of adverse metabolic outcomes associated with a high fructose diet relative to their WT counterparts . This phenotype may result from decreased absorption of fructose by the small intestine but could be attributed to the lack of Txnip in the various other tissues . 10 . 7554/eLife . 18313 . 010Figure 3 . Txnip-knockout mice exhibit less severe metabolic outcomes associated with chronic fructose consumption . Mice were fed a moderate fat diet supplemented with 30% ( wt/vol ) fructose water for 25 weeks . ( a ) In comparison with the control diet , fructose induced significant increases in body weight in wild type ( WT ) but not in Txnip-KO mice . ( b and c ) Intraperitoneal glucose and insulin tests were performed . WT animals developed significant glucose and insulin intolerance following prolonged high fructose diet . Txnip-KO mice were resistant to these effects . ( d and e ) Histological analysis of the liver sections . ( d ) Txnip-KO liver had observably less fibrosis after fructose consumption compared with WT . ( e ) Fructose diet induced significant steatosis in WT mice that was reduced in Txnip-KO mice . ( n = 5 per group ) . p<0 . 05 . For Glucose Tolerance Test and Insulin Tolerance Test: p<0 . 05 and p<0 . 01 WT , RD vs WT , FSD; †p<0 . 05 and ††p<0 . 01 WT , RD vs . KO , RD . #p<0 . 05 WT , RD vs . KO , FSD and p<0 . 01 vs WT , RD . Scale bar; 50 µm . Data represent mean ± SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 18313 . 01010 . 7554/eLife . 18313 . 011Figure 3—source data 1 . Statistical analysis for Figure 3 . These tables represent the statistical analysis conducted on the raw data collected for Figure 3 using GraphPad Prism 5 . DOI: http://dx . doi . org/10 . 7554/eLife . 18313 . 011 Previous studies demonstrated that Txnip expression not only regulates glucose homeostasis but also is responsive to glucose levels ( Shalev et al . , 2002 ) . To determine if Txnip is similarly responsive to chronic fructose consumption , we investigated the effects of chronic fructose consumption on the expression levels of Txnip and its potential interactions with GLUT2 or GLUT5 in the jejunum , which , along with the duodenum , is responsible for fructose absorption by the small intestine ( Holloway and Parsons , 1984 ) . Quantitative RT-PCR analysis revealed that mRNA expression of GLUT2 and GLUT5 increased significantly in the jejunum of mice on the FSD relative to those on the RD , regardless of the genotypes of the mice ( Figure 4a ) . In addition , WT mice on the FSD had higher mRNA expression of Txnip in the jejunum compared to the RD , indicating that Txnip expression is responsive to fructose consumption in a manner similar to the changes observed in GLUT2 and GLUT5 expression ( Figure 4b ) . Confocal analysis of jejunum sections probed for Txnip and GLUT2 or GLUT5 revealed that Txnip co-localized with both GLUT2 and GLUT5 in WT mice on the RD ( Figure 4c , d ) . The co-localizing signal increased in WT mice on the FSD , with the Txnip and GLUT2 co-localizing signal increasing by 12 . 9% ± 4 . 4% ( mean ± SEM; n = 5 , p<0 . 02 ) of the total GLUT2 signal and the Txnip and GLUT5 co-localizing signal increasing by 6 . 0% ± 2 . 3% ( mean ± SEM; n = 5 , p=0 . 05 ) within the total GLUT5 signal ( Figure 4b ) . 10 . 7554/eLife . 18313 . 012Figure 4 . Txnip expression is induced by fructose consumption and interacts with GLUT2 and GLUT5 . ( a–c ) Quantitative RT-PCR analysis of jejunum isolated from mice used in the fructose supplemented diet experiment . ( a and b ) Mice on the FSD had significantly higher mRNA expression of GLUT2 and GLUT5 in the jejunum ( n = 6 for WT , RD and WT , FDS groups; n = 5 for KO , RD and KO , FSD groups ) . ( c ) Mice on the FSD had higher relative mRNA expression of Txnip in the small intestine ( n = 6 for WT , RD and WT , FDS groups; n = 5 for KO , RD and KO , FSD groups ) . ( d ) Confocal analysis of small intestine sections stained for Txnip and GLUT2 revealed co-localization of the two proteins , with more co-localizing signal in FSD mice ( n = 4 for each group ) . ( e ) Confocal analysis of small intestine sections stained for Txnip and GLUT5 revealed co-localization of the two proteins , with more co-localizing signal in FSD mice ( n = 4 for each group ) . p<0 . 05 , *p<0 . 01 . AP = apical membrane , BL = basolateral membrane , white arrows point to co-localizing signal , scale bar represents 10 µM . Data represent mean ± SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 18313 . 01210 . 7554/eLife . 18313 . 013Figure 4—source data 1 . Statistical analysis for Figure 4 . These tables represent the statistical analysis conducted on the raw data collected for Figure 4 using GraphPad Prism 5 . DOI: http://dx . doi . org/10 . 7554/eLife . 18313 . 013 We then analyzed the interaction between Txnip and GLUT2 or GLUT5 interacting molecules using Fluorescent Lifetime Imaging Microscopy ( FLIM ) . Because two separate groups of mice were studied ( normal diet and fructose diet ) to determine the effect of fructose on the interaction between Txnip and GLUT2 and GLUT5 , we analyzed FLIM values between Txnip and GLUT2 and GLUT5 within each group relative to donor only controls ( Txnip only ) . We then compared these relative changes across groups . We measured FLIM values at co-localizing regions between Txnip and GLUT2 or GLUT5 , and we measured an equal number of regions across all samples . As such , an increase in protein mass resulting from the observed mRNA expression is unlikely to significantly impact the results . Rather , we expect observed differences to result from changes in the nature of the interaction between Txnip and both fructose transporters . The values for both τ1 and a1 ( % ) were calculated for Txnip within one group . The averages of these values were used for comparison across groups . After determining the donor fluorescence lifetime ( Txnip ) in the absence of the acceptor fluorophore ( Table 1 , Donor Only ) , FRET between donor and acceptor was defined by the lifetime of interacting molecules ( τ1 ) , with a1 ( % ) representing the fraction of interacting molecules . τ1 is an indirect measurement of distance between two molecules because the lifetime of the donor fluorophore diminishes with increasing proximity to the acceptor fluorophore , which can quench donor emissions . We found that Txnip interacted with both GLUT2 and GLUT5 in the small intestine of animals on the RD as indicated by a decrease in τ1 ( Table 1 ) . Under the FSD , we observed a decrease in τ1 for both GLUT2 and GLUT5 relative to samples form the RD , indicating increased proximity between Txnip and the two GLUTs ( Table 1 ) . This result indicates that the Txnip interaction with the fructose transporters responds to fructose loads . It also suggests that both fructose transporters are involved in Txnip-mediated fructose transport , consistent with our fructose uptake experiments ( Figure 2a ) . Though a1% values increased modestly for GLUT2 under the FSD relative to the RD , the contribution of this increase to fructose uptake is not clear ( Table 1 ) . Collectively , these data suggest that Txnip expression and interaction with GLUT2 and GLUT5 in the jejunum are responsive to chronic fructose consumption , and these changes may contribute to fructose absorption into the body . 10 . 7554/eLife . 18313 . 014Table 1 . Txnip/GLUT interactions determined by FLIM . DOI: http://dx . doi . org/10 . 7554/eLife . 18313 . 01410 . 7554/eLife . 18313 . 015Table 1—source data 1 . Statistical analysis for Table 1 . This table represents the statistical analysis conducted on the raw data collected for Table 1 using GraphPad Prism 5 . DOI: http://dx . doi . org/10 . 7554/eLife . 18313 . 015Regular diet ( RD ) Donor onlyGLUT2GLUT5a1 ( % ) Experiment 110035 ± 232 ± 9Experiment 210036 ± 436 ± 1Experiment 310036 ± 235 ± 3Average a1 ( % ) N/A3534 ± 1τ1 ( ps ) Experiment 12701 ± 332169 ± 432102 ± 48Experiment 22196 ± 151615 ± 601657 ± 26Experiment 32139 ± 361587 ± 301695 ± 31% of τ1 ( ps ) vs donor onlyN/A76 ± 277 ± 1Fructose-supplemented diet ( FSD ) Donor onlyGLUT2GLUT5a1 ( % ) Experiment 110049 ± 431 ± 5Experiment 210045 ± 236 ± 3Experiment 310055 ± 236 ± 3Average a1 ( % ) N/A50 ± 3†34 ± 2τ1 ( ps ) Experiment 12482 ± 351621 ± 511263 ± 99Experiment 22495 ± 621701 ± 381572 ± 53Experiment 32298 ± 721575 ± 101602 ± 34% of τ1 ( ps ) vs donor onlyN/A67 ± 160 ± 6Values represent mean percent change ± SEM . p<0 . 05 vs . regular diet . †p<0 . 01 vs . regular diet . Txnip expression is elevated in skeletal muscle of individuals with impaired glucose tolerance or type 2 diabetes , and Txnip expression is inversely correlated with insulin-stimulated glucose uptake in human insulin/glucose clamp studies ( Muoio , 2007 ) . To determine if diabetes has a similar effect on Txnip expression in the small intestine , we used the streptozotocin ( STZ ) model of type 1 diabetes in mice . All mice receiving STZ injections developed hyperglycemia , allowing us to compare the effect of this diabetic phenotype to euglycemic sodium citrate buffer ( vehicle ) -injected non-diabetic controls . Although Txnip-KO mice are hypoglycemic compared to their wild type counterparts at baseline ( Chutkow et al . , 2008 ) , STZ-injected Txnip-KO mice achieved a higher average blood glucose level relative to non-diabetic Txnip-KO mice ( Figure 5—figure supplement 1 ) . Quantitative RT-PCR revealed that WT STZ-injected mice had higher levels of Txnip mRNA expression in the jejunum compared to wild type non-diabetic mice , indicating that the diabetic state promotes Txnip expression in the small intestine ( Figure 5a ) . 10 . 7554/eLife . 18313 . 016Figure 5 . Hyperglycemia promotes Txnip expression and fructose absorption by the small intestine . ( a ) STZ-injected wild type mice have higher Txnip expression in the small intestine relative to control mice . ( n = 4 for all groups ) . ( b and c ) 14C-radiolabeled D-fructose levels were measured in WT and Txnip-KO STZ-injected and buffer controls over time . Following an oral gavage , tissues were collected at 5 , 10 , 20 , 30 , 60 min post-gavage for analysis ( n = 4 for all groups at each time point ) . ( b ) There was a significant increase in 14C fructose uptake by the whole small intestine in WT STZ-injected mice compared to the non-diabetic controls at 10 min post-gavage as well as a reduction in uptake in Txnip-KO non-diabetic mice compared to WT non-diabetic mice . There was no difference at any time point between Txnip-KO STZ-injected animals and their non-diabetic counterparts . Comparing overall levels of 14C fructose uptake , there was an increase in 14C fructose uptake in WT STZ-injected mice compared to the non-diabetic controls , while this increase was abolished in the Txnip-KO mice . ( c ) There was a significant increase in 14C fructose signal in the hepatic portal vein blood in WT STZ-injected mice relative to the non-diabetic controls at both 10 and 20 min post-gavage . There was also a decrease in 14C fructose signal in Txnip-KO non-diabetic mice compared to WT non-diabetic mice . When comparing overall levels of 14C fructose signal , there was a significant increase in WT STZ-injected mice compared to the non-diabetic controls; this difference was not observed between the Txnip-KO groups . *p<0 . 001 . For fructose absorption data: p<0 . 01 WT control vs . WT STZ , †p<0 . 05 WT control vs . KO control . Data represent mean ± SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 18313 . 01610 . 7554/eLife . 18313 . 017Figure 5—source data 1 . Statistical analysis for Figure 5 . These tables represent the statistical analysis conducted on the raw data collected for Figure 5 using GraphPad Prism 5 . DOI: http://dx . doi . org/10 . 7554/eLife . 18313 . 01710 . 7554/eLife . 18313 . 018Figure 5—source data 2 . Statistical analysis for Figure 5—figure supplement 1 . This table represents the statistical analysis conducted on the raw data collected for Figure 5—figure supplement 1 using GraphPad Prism 5 . DOI: http://dx . doi . org/10 . 7554/eLife . 18313 . 01810 . 7554/eLife . 18313 . 019Figure 5—source data 3 . Statistical analysis for Figure 5—figure supplement 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 18313 . 01910 . 7554/eLife . 18313 . 020Figure 5—figure supplement 1 . STZ-injection induces hyperglycemia in mice . Wild type mice injected with streptozotocin had a higher blood glucose level relative to control mice . Even though Txnip-KO mice have a lower basal blood glucose level relative to wild type mice , STZ-injection induced an observable increase in blood glucose levels in Txnip-KO mice . ( n = 8 ) . **p<0 . 001 . Data represents mean ± SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 18313 . 02010 . 7554/eLife . 18313 . 021Figure 5—figure supplement 2 . 13C-labeled D-fructose measurements in wild-type non-diabetic and STZ-injected mice . LC-MS analysis of 13C D-fructose in jejunum isolated from WT non-diabetic and STZ-injected mice . There was an increase in the normalized peak area of 13C D-fructose in jejunum isolated from STZ-injected mice in comparison to the non-diabetic controls . ( n = 4 for each group ) . p<0 . 05 . Data represents mean ± SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 18313 . 021 We then investigated whether STZ-injected mice had reduced rates of absorption and reduced total absorption by measuring the retention of 14C-radiolabeled D-fructose in hepatic portal vein blood and whole small intestine at different time points following an oral gavage . Relative to non-diabetic WT mice , STZ-injected WT mice demonstrated an increase in 14C signal in the whole small intestine at 10 min after the administration of the 14C-D-fructose bolus , suggesting a greater rate of absorption ( Figure 5b ) . There was no observable difference in 14C signal in whole small intestine collected from non-diabetic Txnip-KO mice relative to STZ-injected Txnip-KO mice , revealing that Txnip expression is necessary for the STZ-mediated increase in fructose absorption by the small intestine . When comparing overall levels of 14C signal over time , there was a significant difference in total fructose absorption between the STZ-injected WT and non-diabetic WT mice , while this STZ-mediated effect was abolished in the Txnip-KO mice ( Figure 5c ) . The increased levels of 14C-radiolabeled D-fructose measured in the small intestine subsequently manifested in increased observed amounts of 14C signal in hepatic portal vein blood collected from STZ-injected WT mice relative to non-diabetic controls at both 10 min and 20 min; this difference was absent when comparing STZ-injected Txnip-KO mice to non-diabetic Txnip-KO mice . Furthermore , there was an observable difference in overall 14C signal in the hepatic portal vein blood collected from STZ-injected WT mice compared to non-diabetic WT mice , while there was no difference between the Txnip-KO groups ( Figure 5c ) . In order to test that the observable differences were attributable to D-fructose and not a downstream metabolite of the administered 14C-fructose , we studied 13C-labeled D-fructose levels in the small intestine using LC-MS analysis . We collected jejunum from animals 10 min post-administration of the oral gavage and isolated the polar metabolites from the tissue for analysis . There was a marked increase in 13C-labeled D-fructose measured in the jejunum collected from STZ-injected mice relative to non-diabetic controls ( Figure 5—figure supplement 2 ) . Collectively , these data demonstrated that STZ-injected mice have overall higher jejunum Txnip expression and fructose absorption , an effect that is diminished in the absence of Txnip expression . Thus , the STZ-induced diabetic state can increase fructose absorption , and Txnip is essential for this effect . This study establishes that fructose absorption in mammals is regulated by Txnip , which interacts with both GLUT2 and GLUT5 on the cellular membrane of enterocytes in the small intestine . While it has long been known that glucose promotes fructose uptake by the small intestine , the exact mechanism for this effect remains unknown ( Riby et al . , 1993 ) . It was once believed that glucose promotes fructose uptake through a disaccharide transport system ( Riby et al . , 1993 ) . Our data in combination with other studies ( Parikh et al . , 2007; Patwari et al . , 2006 ) suggests that Txnip links glucose homeostasis with fructose transport , as diabetes induces Txnip expression , which promotes fructose absorption . Because excess absorption of fructose contributes to liver fat accumulation and hypertension , our experiments suggest that the diabetic state may contribute to these components of metabolic disease at least in part through Txnip and increased fructose transport . The ability of Txnip to regulate fructose metabolism indicates that glucose homeostasis and fructose metabolism are intertwined , possibly in a manner that potentiates metabolic diseases in settings of high fructose consumption , which is often accompanied by similar loads of glucose as sucrose and others sweeteners . The molecular mechanism through which Txnip regulates fructose absorption remains to be defined , and it is plausible that Txnip is necessary for the increase in fructose absorption in diabetic mice but is not directly causal . We demonstrated that Txnip interacts with both fructose transporters and is upregulated in response to fructose consumption . We also observed that the effect of Txnip on fructose transport is specific to the small intestine , as organs collected from intravenously injected Txnip-null animals had similar amounts of fructose to wild-type counterparts . However , defining the detailed molecular mechanism through which Txnip regulates fructose transporters is important . Elevated levels of fructose and glucose in the small intestine can cause GLUT2 to transiently translocate to the apical brush border membrane of enterocytes in order to further facilitate fructose transport , accounting for up to 60% of fructose absorption in animals consuming high amounts of sugar ( Gouyon et al . , 2003 ) . The translocation of GLUT2 is affected by activation of protein kinase C and MAP kinase ( Helliwell et al . , 2000; Kellett , 2001 ) , both of which also have the ability to activate Txnip expression and activity ( Li et al . , 2014 , 2009 ) . Past studies observed that activation of the PKC signaling pathway by phorbol 12-myristate ( PMA ) and the influx of Ca2+ through L-type channels induces cytoskeletal re-arrangement , leading to the insertion of GLUT2 protein at the brush border membrane ( Kellett and Helliwell , 2000; Morgan et al . , 2007 ) . In addition , Txnip expression is attenuated by the calcium channel blocker verapamil ( Xu et al . , 2012 ) , suggesting that normally due to the influx of calcium , Txnip could be facilitating the movement of GLUT2 to the apical membrane as a scaffold protein that directly binds to the glucose transporters . Another possibility is that the phosphorylation of Txnip may change its ability to bind to the fructose transporters and cause subsequent changes in fructose absorption . A previous study found that AMPK activation had a positive effect on GLUT1-facilitated glucose transport , which the researchers attributed to the change in the binding affinity of Txnip for GLUT1 ( Wu et al . , 2013 ) . Regulation of GLUT2 and GLUT5-mediated fructose absorption could occur through a similar mechanism in which the binding of Txnip is necessary for this process . These processes may be specific to gastrointestinal fructose absorption , as we noted that a global loss of Txnip affected fructose uptake in a variety of tissues from fed animals but not in those from intravenously injected animals . Studying the Txnip-mediated fructose absorption mechanism , thus , will require a more precise inquiry into how enterocyte Txnip expression affects fructose uptake by the digestive tract . Other possibilities exist for why this discrepancy occurs because the intravenous injection not only bypasses the small intestine but also the first-pass metabolism and uptake by the liver . Regulation of fructose absorption may be closely linked to other metabolic processes , including glucose metabolism and insulin signaling . While elevated glucose levels promote Txnip expression , elevated insulin levels decrease Txnip expression ( Parikh et al . , 2007 ) . Because chronic fructose intake is associated with impaired insulin signaling , it could facilitate the disruption of glucose homeostasis and the transition of prediabetes to type 2 diabetes through Txnip . In addition , fructose metabolism can lead to an increase in reactive oxygen species ( ROS ) and subsequent redox stress in cells ( Morgan et al . , 2007 ) . Because Txnip is responsive to the redox environment of the cell , fructose may indirectly regulate Txnip expression through this mechanism . As a result , fructose could affect Txnip in a manner that disrupts glucose metabolism and further increase fructose absorption by promoting Txnip expression , similar to how glucose can regulate Txnip to change fructose metabolism . As noted by our long-term fructose feeding experiment , fructose ingestion may facilitate the onset of metabolic disease , including insulin resistance and hepatic steatosis . The deletion of Txnip alleviated mice of these effects to an extent , suggesting that Txnip and other factors involved in the Txnip-mediated fructose absorption pathway could be potential therapeutic targets . However , much more understanding of this molecular pathway is necessary to determine if and when regulation of the pathway may be beneficial . We observed that the STZ-induced diabetic state causes a marked increase in fructose uptake by the small intestine and transport into hepatic portal vein circulation . STZ-diabetic mice showed an increase in absorption of fructose , which could contribute to metabolic dysfunction , as was the case in our long-term fructose supplemented diet experiment in which the mice developed pathologies compared to mice with relatively lower fructose consumption . STZ induces hypoinsulinemia and subsequent hyperglycemia mediated through pancreatic beta cell death , so further studies will be necessary to determine if all causes of hyperglycemia can regulate fructose absorption by the small intestine . Future studies in mice with cell-specific deletion of Txnip may reveal more detail on how Txnip regulates fructose transport . Controversy remains on the effect of the diabetic state on fructose absorption ( Stanhope , 2016 ) . One study in Japan observed an increase in serum fructose concentrations in patients with type 2 diabetes mellitus relative to healthy patients who had no diabetic symptoms ( Kawasaki et al . , 2002 ) ; these patients were all admitted to the same hospital and had their diets monitored and were fasting when blood serum was collected . However , another study in Finnish patients did not observe a difference in fructose serum levels when comparing type 2 diabetic patients to patients with no diabetic symptoms ( Pitkänen , 1996 ) . The patients in that study were not fasted prior to blood collection , and many of the diabetic patients had many diabetes-related complications affecting their health . Currently , it is difficult to determine based on available human data if the diabetic state and which specific physiological pathology of diabetes may affect fructose absorption in humans . Our data suggest that glucose homeostasis and fructose absorption interact . Given the potential importance of fructose metabolism in modern societies , more investigation into this phenomenon is warranted . All experiments were conducted in accordance with the Guide for the Use and Care of Laboratory Animals and approved by the Harvard Medical School Standing Committee on Animals . Txnip knockout mouse from a C57B1/6 background strain were developed in our laboratory ( RRID:IMSR_JAX:018313 ) ( Morselli et al . , 2014 ) . Genotyping was performed by PCR on tail DNA using the following primers: OL4F2 , 5’- CTT CAC CCC CCT AGA GTG AT –3’; P3F1 , 5’-TTT CGT TTG GGT TTT CAA GC –3’; and P3R2 , 5’-CCC AGA GCA CTT TCT TGG AC–3’ . The number of mice required for the study was determined by using the ClinCalc Sample Size Calculator: a change of 33% for all phenotypes was assumed , and an alpha of 0 . 05 and statistical power of 90% were also used as parameters to estimate the appropriate sample size . Pulldown assays were performed as previously described ( Abdelmalek et al . , 2010 ) . Briefly , indicated plasmids were transfected into HEK293T cells using Transit-293 transfection reagent ( Mirus ) . Cells were lysed in 0 . 5% Triton X-100 , 0 . 1% sodium deoxycholate , 150 mM NaCl , 50 mM Tris , 1 mM phenylmethanesulfonyl fluoride , and protease and phosphatase inhibitors , pH 7 . 8 . Immunoprecipitation of Txnip protein complexes was performed using the appropriate antibodies before binding to Protein A/G agarose beads ( Santa Cruz Biotechnology; Dallas , Texas ) . Input lysates and pulldown eluates were analyzed by SDS–PAGE and immunoblots . The following protein-specific antibodies were used: anti-V5 ( RRID:AB_307024 ) , anti-flag ( RRID:AB_298215 ) , anti-GLUT2 ( RRID:AB_641068 ) , and anti-GLUT5 ( RRID:AB_2189499 ) . The following HRP-conjugated antibodies were used: Goat Anti-Mouse IgG ( H L ) -HRP Conjugate ( RRID:AB_11125936 ) and Rabbit Anti-Mouse IgG ( Light Chain Specific ) ( D3V2A ) mAb HRP Conjugate ( RRID:AB_1549610 ) . Caco-2 cells or isolated MEFs were transfected with the indicated plasmids using Purefection Transfection Reagent ( System Biosciences ) and incubated for 24 hr . Fructose was added to normal medium conditions at a concentration of 4 . 5 g/L with the addition of 5 µCuries/mL of D-[14C]fructose ( MP Biomedicals ) . Cells were incubated with fructose for 1 hr before lysis in 0 . 5% Triton X-100 , 150 mM NaCl , 50 mM Tris , 1 mM phenylmethanesulfonyl fluoride , and protease and phosphatase inhibitors , pH 7 . 8 . Cell lysates were measured with a gamma counter ( Beckman Coulter LS 6500 ) . Fructose uptake rates in the small intestine were determined following the technique of by Karasov and Diamond ( Hwang et al . , 2010 ) . Briefly , a 1-cm segment of jejunum was everted and mounted on a grooved steel rod ( 3- mm diameter ) and preincubated at 37°C for 5 min in Ringer solution bubbled with 95% O2-5% CO2 . The sleeves were then incubated at 37°C in an oxygenated solution containing D-[14C]fructose for 2 min . The solutions were stirred at 1200 rpm during the incubation procedure to minimize unstirred layers . To reduce the radioactive label in the adherent fluid , there was a 20-s rinse in 30 ml of ice-old Ringer solution after incubation . The tissues were dissolved in a tissue solubilizer ( Solvable , Packard Instruments ) . The dissolved tissue was mixed with scintillation cocktail ( Ecolume , ICN ) , and radioactivity was measured with a liquid scintillation counter ( Beckman LS 7800 , Beckman , Fullerton , CA ) . The uptake rates of 14C D-fructose were determined at 50 mM and expressed as micromoles per milligram net weight of small intestine . Intestinal fructose transport and its biodistribution were analyzed in vivo . After intragastric administration of D-Fructose , [U-14C] ( 0 . 2 µCurie in 200 µl 30% fructose/mannitol ) with a ball tip needle or intravenous administration through tail-vein injection , mice were euthanized . Organs of interest ( blood , small and large intestine , liver , kidney , and femur muscle ) were harvested . Portal vein blood was extracted from the vein using heparinized tubing and needle . Approximately 50–150 mg of each tissue of interest was dissolved in Soluene-350 and added to the appropriate amount of Ultima Gold Scintillation Fluid . Radioactivity in tissues was then measured in the gamma-counter and results were analyzed as percentage of injected dose per gram of tissues as previously described ( Chutkow et al . , 2008 ) . Txnip deficient mice ( 6-week-old , female ) and wild type mice were divided into two groups . All animals were fed the PicoLab Mouse Diet 20 . Mice consumed approximately 30–50 g of feed per day per mouse , amounting corresponding to around 20–33 kcals per day per mouse , with no significant differences in feed intake among the different groups . One group had free access to 30% ( wt/vol ) fructose water with otherwise regular diet , and the other group had free access to plain water with regular diet for 25 weeks . Mice with access to fructose on average consumed an additional 1 kcal per day that accounted for roughly 3–5% of their total caloric intake , which is similar to the average caloric contribution of fructose in humans ( Burgess et al . , 2011 ) . Body weight , food intake , and water intake were measured . At the end of protocol , animals were sacrificed to harvest tissues . Measurements were normalized to organ weight and expressed as picomoles per mg of tissue . The DiaComp protocol for low dose streptozotocin induction in mice was used . Mice were fasted 4–6 hr prior to injections . STZ dissolved in 0 . 1 M sodium citrate ( pH 4 . 5 ) was administered to each mouse at a concentration of 50 mg/kg mouse body weight . Injections were repeated daily for 5 days . Hyperglycemia was confirmed 2 weeks post-injections using a Bayer Contour Blood Glucose Meter . STZ injections were repeated after 1 month in mice that remained euglycemic . Glucose ( 2 g/kg ) and insulin ( 2 milliunits/g ) tolerance tests were performed intraperitoneally following 4 hr fasting period to measure plasma glucose levels at 30 , 60 , 120 , and 180 min after injection . Whole blood glucose levels were assayed from tail clipping venous blood samples using an Ascensia Elite XL glucometer ( Bayer Co ) . Liver samples were collected from mice and fixed in 10% formalin . Samples were then dehydrated using a gradient of 70–100% ethanol and 100% xylene . Tissues were embedded , sectioned , and stained with H&E and Sirius Red by the Harvard Stem Cell Institute Histology Core . Relative gene expression was measured using qPCR . RNA was collected from jejunum samples from mice using the PureLink RNA Micro Kit ( Thermo Fisher Scientific ) . RNA samples were then converted to cDNA using a High-Capacity cDNA Reverse Transcription Kit ( Thermo Fisher Scientific ) . cDNA Samples were then measured using a Bio-Rad CFX384 Real-Time PCR Detection System using iTaq Universal SYBR Green Supermix ( Bio-Rad ) and the indicated primers ( Supplementary file 1 ) . All biological replicates were measured using technical triplicates . Jejunum from WT mice from the fructose diet experiment were harvested and embedded in paraffin . Sections were prepared by the Harvard SCRB Histology Facility . After rehydration and antigen recovery of the sections , slides were stained with GLUT2 ( RRID:AB_641066 ) and GLUT5 ( RRID:AB_2189502 ) with anti-rabbit secondary antibody conjugated to Alexa Fluor 594 ( RRID:AB_2534079 ) and Txnip ( RRID:AB_11033580 ) with anti-rabbit antibody conjugated to Alexa Fluor 488 ( RRID:AB_143165 ) . Slides were analyzed using an Olympus Fluoview 1500 at the Brigham and Women’s Regenerative Medicine Center . Co-localization was analyzed using the co-localization tool in the FSV-10W software using the default threshold ( intensity of 2024 ) for all samples and channels . Brightness and contrast were applied to equally to all images . All images were normalized to equal gamma levels . To analyze the interaction of Txnip with GLUT5 by Fluorescence Lifetime Microscopy ( FLIM ) , Nikon Ti-E inverted microscope was used . Becker and Hickl SPCM software with DCC was used to acquire FLIM data , and SPCImage 3 . 0 ( Becker and Hickl ) software was used for FLIM analysis . Paraffin-embedded sections of small intestine from wild type mice on either the regular or fructose-supplemented diet were used . After rehydration and antigen recovery of the sections , Txnip was detected with secondary antibody conjugated to Alexa Fluor 488 ( donor fluorophore ) and GLUT2 or GLUT5 with antibody conjugated to Alexa Fluor 594 ( acceptor fluorophore ) . Becker and Hickl SPCM software with DCC was used to acquire FLIM data , and SPCImage 3 . 0 ( Becker and Hickl ) software was used for FLIM analysis . The donor fluorophore was stimulated with a 488 nm laser , and the lifetime ( τ1 ) was determined in the presence and absence of antibody to the acceptor fluorophore . The parameter a1 ( % ) was automatically determined for a measurement of the percent of Txnip molecules interacting with GLUT2 or GLUT5 in the region being analyzed . Tissues were prepared as indicated by the Metabolite Profiling Core at the Whitehead Institute . Approximately 10–30 mg of each tissue was collected from mice and immediately frozen using liquid nitrogen . Samples were homogenized in LC/MS grade methanol at −20°C before adding LC/MS grade water and HPLC-grade chloroform . Samples were then spun down to separate the polar and organic metabolites . Samples were dried using a vacuum concentrator before submission to the Metabolite Profiling Core for analysis . All sample sizes reported refer to biological replicates . Statistical comparison between two groups was performed by unpaired Student’s t test . Statistical comparison among three or more groups was performed by one-way ANOVA with the Bonferroni post-hoc test of significance . Two-way ANOVA with the Bonferroni post-hoc test of significance was used to compare the means of groups in the fructose diet experiment and STZ-induction experiments . Statistical analysis was carried out with the Graphpad Prism 5 software ( RRID:SCR_002798 ) , and statistical significance was assigned to differences with a p value of <0 . 05 .	Fructose is a type of sugar that is found naturally in fruits , and it is closely related to glucose . The amount of fructose in our diet has increased dramatically in the last few decades . Growing evidence suggests that excessive amounts of fructose contribute to several metabolic diseases , including fatty liver disease and diabetes . Fructose is absorbed in the small intestine via transport proteins called GLUT2 and GLUT5 and then travels to the liver where it can stimulate the cells to make fats . However , it is not clear how fructose uptake is regulated in the small intestine . Glucose is taken into cells by a transport protein that is closely related to GLUT2 and GLUT5 . Another protein called thioredoxin-interacting protein ( Txnip ) interacts with the glucose transporter and regulates glucose uptake . Here , Dotimas et al . investigated whether Txnip also regulates the activities of GLUT2 and GLUT5 to control how cells absorb fructose . Initial experiments in cells showed that Txnip binds to both GLUT2 and GLUT5 and increases the amount of fructose taken up by both mouse and human cells . Cells from mutant mice that do not produce Txnip absorbed less fructose than normal cells did . Furthermore , the mutant mice had lower levels of fructose in the blood and less severe metabolic disease after consuming fructose regularly for six months . Mice with diabetes absorbed more fructose through the small intestine than normal mice , and the loss of Txnip from these mice abolished this effect . Together the findings of Dotimas et al . suggest that Txnip plays an important role in regulating fructose absorption and indicate that , at least in some circumstances , diabetes may lead to more fructose being absorbed in the small intestine . The next steps following on from this work are to understand the molecular details of how Txnip regulates fructose uptake and to determine if other forms of diabetes also show increased fructose uptake .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "biochemistry", "and", "chemical", "biology", "cell", "biology" ]	2016	Diabetes regulates fructose absorption through thioredoxin-interacting protein
Sensory experience modifies behavior through both associative and non-associative learning . In Caenorhabditis elegans , pairing odor with food deprivation results in aversive olfactory learning , and pairing odor with food results in appetitive learning . Aversive learning requires nuclear translocation of the cGMP-dependent protein kinase EGL-4 in AWC olfactory neurons and an insulin signal from AIA interneurons . Here we show that the activity of neurons including AIA is acutely required during aversive , but not appetitive , learning . The AIA circuit and AGE-1 , an insulin-regulated PI3 kinase , signal to AWC to drive nuclear enrichment of EGL-4 during conditioning . Odor exposure shifts the AWC dynamic range to higher odor concentrations regardless of food pairing or the AIA circuit , whereas AWC coupling to motor circuits is oppositely regulated by aversive and appetitive learning . These results suggest that non-associative sensory adaptation in AWC encodes odor history , while associative behavioral preference is encoded by altered AWC synaptic activity . Sensory experience shapes sensory behavior . Primary sensory neurons adjust their sensitivity and dynamic range to capture ongoing sensory information without saturating , a phenomenon illustrated by the adaptation of the retina to ambient light levels over a 1010-fold range ( Arshavsky and Burns , 2012; Fain et al . , 2001 ) . In addition , animals learn to increase , decrease , or switch their preference for sensory cues experienced in attractive or aversive contexts . Sensory adaptation and context-dependent learning affect overlapping circuits , which must preserve robust function in the face of continuously changing neuronal properties . Here , we show how an olfactory circuit implements these two processes by encoding sensory adaptation and aversive learning at distinct sites . The nematode Caenorhabditis elegans , whose nervous system is composed of 302 neurons , shows robust plasticity in olfactory , mechanosensory , thermosensory , and gustatory behaviors ( Colbert et al . , 1995; Rankin , 1991; Hedgecock and Russell , 1975; Saeki et al . , 2001 ) . Olfaction may be its most complex sense . C . elegans detects hundreds of volatile odors using dedicated sensory neurons , each of which expresses multiple G protein-coupled receptors ( GPCRs ) ( Troemel et al . , 1995 ) . For example , an olfactory neuron called AWCON ( one of two AWC neurons ) detects benzaldehyde , butanone , and isoamyl alcohol , and expresses at least five chemosensory GPCRs ( Bargmann , 2006; Lesch and Bargmann , 2010 ) . Calcium imaging and genetic studies indicate that AWC signal transduction resembles mammalian phototransduction: odors are inferred to decrease the level of cGMP , close a cGMP-dependent transduction channel , and hyperpolarize AWC ( Bargmann , 2006; Chalasani et al . , 2007 ) . When C . elegans is exposed to high concentrations of an odor in the absence of food , it gradually loses its attraction to that odor over an hour or more , and recovers over a similar timescale ( Colbert and Bargmann , 1995 ) . This process has been called adaptation , but will here be called aversive olfactory learning to reflect its long duration and the required pairing with food deprivation ( Nuttley et al . , 2002 ) , and to distinguish it from short-term sensory adaptation . Aversive learning is selective for the experienced odor , even when two odors are sensed by partly overlapping olfactory neurons ( Colbert and Bargmann , 1995 ) . The genetic requirements for aversive learning vary depending on the odor and the duration of odor exposure , and include G protein signaling pathways , ion channels , and transcriptional regulators ( Ardiel and Rankin , 2010 ) . The cGMP-dependent protein kinase EGL-4 is closely associated with aversive learning . egl-4 mutants show learning defects to multiple AWC-sensed odors , among other sensory defects ( Daniels et al . , 2000; L’Etoile et al . , 2002 ) . After prolonged odor conditioning , EGL-4 translocates from the AWC cytoplasm to the nucleus ( O'Halloran et al . , 2009; Lee et al . , 2010 ) , where it phosphorylates the heterochromatin protein HPL-2 and alters gene expression ( Juang et al . , 2013 ) . Nuclear translocation of EGL-4 is a real-time marker for AWC plasticity ( Lee et al . , 2010 ) . EGL-4 translocation requires olfactory signal transduction and endogenous cGMP signaling in AWC , but exogenous cGMP is not sufficient to modify EGL-4 localization ( O'Halloran et al . , 2012 ) , suggesting that a second coincident signal is required . The existence of a coincident context signal for learning is further supported by the observation that aversive learning only occurs when odors are paired with food deprivation ( Nuttley et al . , 2002 ) . Indeed , butanone odor becomes more attractive after pairing with food , indicating that food context drives bidirectional olfactory learning ( Torayama et al . , 2007; Kauffmann et al . , 2010 ) . The integration of food context may involve communication between neurons , as aversive learning requires insulin signals from other neurons that feed back onto AWC and reduce its activity ( Chalasani et al . , 2010; Lin et al . , 2010 ) . The circuits for aversive olfactory learning are only partially understood . First , it is unclear how information about odor and food context is represented and associated during odor conditioning . Second , the neuronal sites at which information is stored are unknown . Here , we use circuit manipulations , molecular markers , and in vivo calcium imaging to map the effects of odor history and food context on the olfactory circuit . We show that during aversive learning , the food-deprivation context engages feedback from AIA and other neurons , as well as insulin signaling . This feedback regulates EGL-4 localization and the behavioral output of AWC . In parallel , non-associative sensory adaptation shifts the dynamic range of AWC during both appetitive and aversive learning . Thus olfactory learning induces both associative and non-associative plasticity in a single sensory neuron . C . elegans is strongly attracted to butanone , an odor sensed by the AWCON olfactory neuron , but this attraction is lost after butanone is paired with food deprivation for 90 min ( Figure 1A ) . To identify neurons required for olfactory learning , we acutely silenced neurons that form pre- and postsynaptic connections with AWC in the C . elegans wiring diagram ( Figure 1B; White et al . , 1986 ) . Small groups of neurons were targeted by cell-specific expression of the Drosophila histamine-gated chloride channel HisCl1 , and silenced by administration of exogenous histamine during the conditioning period ( Pokala et al . , 2014 ) . This temporary inactivation should identify neurons that function in learning , rather than neurons with general effects on locomotor behaviors or chemotaxis strategies . 10 . 7554/eLife . 14000 . 003Figure 1 . Neurons required during odor conditioning for aversive olfactory learning . ( A ) Schematic of aversive olfactory learning assay . Adult worms are washed off food and conditioned with odor in buffer for 90 min , then washed again before being tested in a butanone chemotaxis assay for 1–2 hr . ( B ) Partial C . elegans wiring diagram showing neurons tested for effects on aversive learning and their synaptic connections with AWC and each other . ( C ) Aversive learning in animals expressing the histamine-gated chloride channel ( HisCl1 ) under cell-specific promoters , assayed with or without 10 mM histamine in the conditioning medium . Chemotaxis assays ( 1:1000 butanone dilution ) were performed on histamine-free plates . Error bars represent S . E . M . P values were generated by 2-way ANOVA for interaction of odor condition and presence of histamine ( *p<0 . 001 , p<0 . 05 ) . n = 3–61 assays , 50–200 animals/assay . ( D ) Aversive learning in animals carrying the gcy-28d::HisCl transgene and a second transgene expressing dsRNA that targets HisCl . Tested as in ( B ) . Error bars represent S . E . M . P values were generated by 2-way ANOVA for interaction of odor condition and presence of histamine ( *p<0 . 001 , p<0 . 05 , n . s . not significant ) . n = 9–15 assays , 50–200 animals/assay . ( E ) Aversive learning in animals expressing the gain-of-function potassium channel UNC-103 under the gcy-28d promoter . P values were generated by 2-way ANOVA for interaction of genotype and condition ( p<0 . 001 ) . n = 8 assays , 50–200 animals/assay . ( F ) Appetitive learning in wild-type and gcy-28d::unc-103 ( gf ) animals after conditioning with butanone and food ( 1:10 butanone dilution ) . 2-way ANOVA for interaction of genotype and condition ( n . s . not significant ) . n = 8 assays per condition , 50–200 animals/assay . DOI: http://dx . doi . org/10 . 7554/eLife . 14000 . 00310 . 7554/eLife . 14000 . 004Figure 1—source data 1 . Individual chemotaxis indices for Figure 1C–F . DOI: http://dx . doi . org/10 . 7554/eLife . 14000 . 004 Among a panel of tested strains , aversive learning was nearly eliminated in gcy-28d::HisCl1 animals that were exposed to histamine during conditioning ( Figure 1C ) . This transgene is expressed reliably in AIA ( 100% ) and AVF ( 90% ) neurons and less frequently in ASI , I1 , IL2 and M3 neurons ( see Materials and methods ) ; AIA and ASI form synapses with AWC . No defect was observed after HisCl1 silencing of several other AWC synaptic partners ( AIB , AIY , RIA ) , other sensory neurons ( AWB ) , or neuromodulatory neurons ( NSM , RIM , RIC ) . Control experiments demonstrated that histamine did not affect butanone chemotaxis or aversive learning in wild-type animals , and that the HisCl1 transgenes were innocuous in the absence of histamine ( Figure 1C ) . Thus gcy-28d::HisCl1 selectively silences neurons required during the conditioning period for aversive olfactory learning . To narrow down the relevant neurons expressing gcy-28d::HisCl1 , we took two complementary approaches: expressing HisCl1 in subsets of the gcy-28d-expressing neurons in a wild-type strain , and expressing double-stranded HisCl1 RNA in subsets of neurons to silence transgene expression in the gcy-28d::HisCl1 strain . We were unable to identify a single neuron in which HisCl1 expression could replicate the strong learning defect of the gcy-28d::HisCl1 strain ( Figure 1C ) . The most promising initial candidate in the gcy-28d set was AVF , because an unc-4::HisCl1 transgene that is expressed in AVF caused a partial learning defect ( Figure 1C ) . However , unc-4::HisCl1 is expressed in many neurons other than AVF that might contribute to its effect . Moreover , dsRNA-induced silencing of the gcy-28d::HisCl1 transgene in AVF did not rescue aversive learning , but instead caused a mild chemotaxis defect ( Figure 1D ) . These results suggest that AVF might affect chemotaxis rather than learning per se . Silencing the gcy-28d::HisCl1 transgene in AIA consistently restored aversive learning ( Figure 1D ) , a result that implicates AIA in learning . However , silencing AIA alone or in several combinations with other neurons with HisCl1 did not disrupt learning ( Figure 1C ) . Combining transgenes that express HisCl in AIA and AVF resulted in a learning defect even in the absence of histamine , an ambiguous result . These results suggest that simultaneous silencing of AIA and at least one unidentified neuron cause the aversive learning defect in the gcy-28d::HisCl1 strain . As an alternative chronic silencing method , we expressed an overactive UNC-103 potassium channel under the gcy-28d promoter . This transgene resulted in a strong defect in aversive olfactory learning ( Figure 1E ) . The gcy-28d::unc-103 ( gf ) transgene is expressed reliably in AIA interneurons and only occasionally in other neurons ( see Materials and methods ) , supporting the hypothesis that AIA promotes aversive learning . For simplicity , the neurons affected by this gcy-28d::unc-103 ( gf ) transgene will be described as 'the AIA circuit , ' with the recognition that other neurons may also contribute to its effects . Pairing butanone with food results in increased attraction to butanone , a phenomenon called appetitive olfactory learning or butanone enhancement ( Torayama et al . , 2007; Kauffman et al . , 2010 ) . Appetitive olfactory learning was unaffected by silencing the AIA circuit with the gcy-28d::unc-103 ( gf ) transgene ( Figure 1F ) , indicating a preferential requirement for the AIA circuit in aversive learning . The AIA interneurons receive synaptic input from numerous chemosensory neurons , and make reciprocal synapses onto a few , including AWC . Previous studies have implicated retrograde insulin ( ins-1 ) signaling from AIA to AWC in aversive learning to odors including benzaldehyde and isoamyl alcohol ( Lin et al . , 2010; Chalasani et al . , 2010 ) . We found that aversive learning to butanone was also lost in the ins-1 mutant , and was rescued by expressing ins-1 in the AIA neurons ( Figure 2 ) . 10 . 7554/eLife . 14000 . 005Figure 2 . Cell-specific requirements for the insulin signaling pathway in aversive olfactory learning . Aversive olfactory learning in mutants of the insulin signaling pathway , with or without cell-specific transgenes expressing cDNAs for the insulin gene ins-1 and the PI3 kinase gene age-1 . Error bars represent S . E . M . P values were generated by 2-way ANOVA for interaction of genotype and condition ( All comparisons significant at p<0 . 01 after Bonferroni correction ) . n = 4–26 assays , 50–200 animals/assay . DOI: http://dx . doi . org/10 . 7554/eLife . 14000 . 00510 . 7554/eLife . 14000 . 006Figure 2—source data 1 . Individual chemotaxis indices for Figure 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 14000 . 006 INS-1 and other insulin-like proteins act on the insulin receptor DAF-2 , which regulates the AGE-1 phosphatidyl inositol-3-kinase ( PI3K ) and downstream kinases and transcription factors ( Murphy and Hu , 2013 ) . daf-2 and age-1 null mutants are inviable , but viable age-1 reduction of function mutants were defective in aversive olfactory learning , and could be rescued by expressing age-1 in AWCON ( Figure 2 ) . Thus insulin signaling to AWC is one possible mechanism by which the AIA circuit could regulate aversive learning to butanone . To understand how the AIA circuit modifies olfactory behavior , we examined the effect of AIA silencing on reporters of AWC activity and olfactory plasticity . We began with the cGMP-dependent protein kinase EGL-4 , whose translocation from the cytoplasm to the nucleus of AWC neurons during odor conditioning is essential for aversive olfactory learning ( Lee et al . , 2010 ) . Quantitative microscopic analysis of an EGL-4::GFP transgene expressed in AWC neurons was used to determine cytoplasmic and nuclear levels of EGL-4 after pairing odor with food deprivation ( Figure 3A ) . In agreement with previous results , odor conditioning led to enrichment of EGL-4::GFP in the AWC nucleus within 90 min ( Figure 3B ) . However , when the AIA circuit was silenced , naive and butanone-conditioned animals had indistinguishable EGL-4::GFP localization in AWC , resembling naive wild-type animals ( Figure 3C ) . These results identify nuclear enrichment of EGL-4::GFP as a cell-biological readout of signaling from the AIA circuit to AWC . 10 . 7554/eLife . 14000 . 007Figure 3 . Nuclear enrichment of EGL-4 in AWC neurons after odor conditioning . ( A ) Representative images of EGL-4::GFP fluorescence and nuclear index in the AWC of naive and conditioned animals ( left ) , and the equation used to quantify the degree of EGL-4 nuclear localization ( right ) . Fnucleus , Fcytoplasm = fluorescence measured in AWC nucleus or cytoplasm of the same neuron . ( B , C , D ) Cumulative distribution ( AWC nuclear index ) for EGL-4::GFP in wild-type , gcy-28d::unc-103 ( gf ) , and age-1 ( hx546 ) animals after conditioning . An increase in AWC nuclear index is observed after conditioning wild-type ( B ) but not gcy-28d::unc-103 ( gf ) ( C ) or age-1 ( hx546 ) ( D ) . P values were generated by nonparametric Kolmogorov-Smirnov test ( p<0 . 001 ) . n = 79–90 animals per condition . DOI: http://dx . doi . org/10 . 7554/eLife . 14000 . 00710 . 7554/eLife . 14000 . 008Figure 3—source data 1 . Individual nuclear indices for Figure 3B , C , D . DOI: http://dx . doi . org/10 . 7554/eLife . 14000 . 008 The ins-1/age-1 signal from AIA to AWC is a candidate to mediate the effect of the AIA circuit on AWC EGL-4::GFP during learning . Indeed , age-1 ( PI3K ) reduction of function mutants , which were defective in aversive learning ( Figure 2 ) , also failed to relocalize EGL-4::GFP in AWC after odor conditioning ( Figure 3D ) . One potential mechanism by which aversive learning could suppress butanone attraction is downregulation of olfactory signal transduction in AWCON . Attractive odors reduce AWCON calcium , and subsequent odor removal results in a calcium increase that can overshoot before returning to baseline ( Chalasani et al . , 2007 ) . Previous studies have demonstrated a near-complete loss of AWC odor sensitivity after an hour of conditioning with isoamyl alcohol ( Chalasani et al . , 2010 ) . Using genetically-encoded calcium indicators to monitor butanone responses after aversive learning , we found that AWCON calcium responses at 1 µM butanone were greatly diminished after butanone conditioning ( Figure 4A ) . However , the AWCON responses were restored after ten minutes of recovery in buffer ( Figure 4A ) . This short-term suppression of butanone responses appears insufficient to explain aversive learning behavior , which persists through an hour-long chemotaxis assay ( Figure 1A ) . 10 . 7554/eLife . 14000 . 009Figure 4 . AWCON butanone responses shift after odor conditioning . ( A ) Average AWCON calcium responses to a 30-second pulse of 1 µM butanone in naive animals , conditioned animals , or conditioned animals after 10 min of recovery in buffer . Gray represents odor . Shaded region represents S . E . M . n = 8–27 animals . ( B ) AWCON calcium responses of naive animals and conditioned animals to a range of butanone concentrations . Animals were washed in buffer for 15 min after conditioning . Gray represents odor ( 30 s pulses ) . Shaded region represents S . E . M . n = 25–26 animals . ( C ) Average response magnitude after butanone addition ( first of three pulses , data from ( B ) ) . ( D ) Half-time of recovery after butanone removal ( last of three pulses , data from ( B ) ) . P values were generated by t-test at each odor concentration with correction for unequal variance ( p<0 . 001 , p<0 . 05 ) . Error bars represent S . E . M . DOI: http://dx . doi . org/10 . 7554/eLife . 14000 . 00910 . 7554/eLife . 14000 . 010Figure 4—source data 1 . Data and heat map showing individual responses for Figure 4B . DOI: http://dx . doi . org/10 . 7554/eLife . 14000 . 01010 . 7554/eLife . 14000 . 011Figure 4—source data 2 . Data for response magnitude and recovery time in Figure 4C , D . DOI: http://dx . doi . org/10 . 7554/eLife . 14000 . 011 To obtain a more quantitative understanding of sensory dynamics , AWCON calcium responses were examined in a high-throughput system that allowed simultaneous calcium imaging in multiple animals across multiple odor pulses ( Larsch et al . , 2013 ) . A dose-response curve of naive animals showed that AWCON calcium is suppressed by butanone over a 105-fold concentration range from 11 nM to 1 mM . This suppression of basal calcium by odor saturated between 111 nM and 1 µM butanone , was followed by a significant calcium overshoot after odor removal at 11 nM to 11 µM , and recovered to baseline only slowly from odor concentrations above 11 µM ( Figure 4B–D , black traces ) . After pairing butanone conditioning with food deprivation , the dynamic range of AWCON calcium responses was shifted toward higher concentrations ( Figure 4B , C; red traces ) . Butanone-conditioned animals had a ten-fold increase in the detection threshold , a ten-fold increase in the saturation concentration , a faster recovery after odor removal , and a reduced calcium overshoot compared to control animals conditioned in buffer alone ( Figure 4B–D ) . The shift in dynamic range persisted for at least 40 min , consistent with the time course of aversive learning . The effects of the AIA circuit on AWCON dynamic range were examined by calcium imaging in the gcy-28d::unc-103 ( gf ) strain . Naive AWCON calcium responses in this strain resembled those of wild-type animals ( Figure 5A , black traces ) . Surprisingly , butanone conditioning resulted in the same shift in the AWCON threshold , saturation concentration , and recovery dynamics as in the wild type ( Figure 5A–C , red traces ) , despite the fact that these animals did not show aversive learning at a behavioral level . This result demonstrates that the shift in AWCON dynamic range after conditioning is not sufficient for learning . 10 . 7554/eLife . 14000 . 012Figure 5 . AWCON butanone responses report odor history . ( A–F ) AWCON calcium responses are altered by aversive conditioning in mutants that do not learn . ( A–C ) gcy-28d::unc-103 ( gf ) ( D–F ) age-1 ( hx546 ) . ( G–I ) AWCON calcium responses of wild-type animals after appetitive conditioning with odor and food . Note that the well-fed naive and conditioned groups in ( G–I ) are less sensitive to odor than the food-deprived groups in Figure 4 . ( A , D , G ) Calcium responses to a concentration series with three 30 sec pulses per concentration . Gray represents odor . ( B , E , H ) Average response magnitude after butanone addition ( first of three pulses ) . ( C , F , I ) Half-time of recovery after butanone removal ( last pulse ) . n = 11–24 animals . P values were generated by t-test at each odor concentration with correction for unequal variance ( p<0 . 001 , p<0 . 05 ) . Error bars in ( B , C , E , F , H , I ) and shaded regions in ( A , D , G ) represent S . E . M . DOI: http://dx . doi . org/10 . 7554/eLife . 14000 . 01210 . 7554/eLife . 14000 . 013Figure 5—source data 1 . Data and heat map showing individual responses in Figure 5A , D , G . DOI: http://dx . doi . org/10 . 7554/eLife . 14000 . 01310 . 7554/eLife . 14000 . 014Figure 5—source data 2 . Data for response magnitude and recovery time in Figure 5B , C , E , F , H , I . DOI: http://dx . doi . org/10 . 7554/eLife . 14000 . 014 The AWCON dynamic range also shifted after butanone conditioning in age-1 PI3K mutants , which have a learning defect ( Figure 5D–F ) . In addition , conditioned age-1 mutants had pronounced AWCON oscillations after odor removal , as previously reported for ins-1 mutants ( Figure 5—Source data 1; Chalasani et al . , 2010 ) . age-1 probably affects several signaling processes , but its response supports the conclusion that a shift in AWCON dynamic range is insufficient for aversive learning . Finally , we monitored odor responses in AWCON neurons after appetitive learning . Behavioral attraction to butanone is enhanced after butanone is paired with food ( Figure 1F ) , but AWCON was less sensitive to butanone in these conditioned animals than in the matched naive control ( Figure 5G–I ) . Indeed , the AWCON calcium response was similarly affected by appetitive and aversive conditioning , with a ten-fold increase in detection threshold , an increase in the saturation concentration , and a faster recovery after odor removal compared to naive controls ( Figure 5H , I ) . Thus the shift in butanone dynamic range in AWCON represents sensory adaptation to odor history , and not odor preference . This experiment also revealed that food deprivation alters AWCON responses: food-deprived naive animals ( Figure 4B , C ) detected lower butanone concentrations than well-fed naive animals ( Figure 5G , H ) . The calcium-calmodulin dependent protein kinase CMK-1 , which shuttles between the AWC nucleus and cytoplasm in response to food deprivation , is a candidate to mediate this change in odor sensitivity ( Neal et al . , 2015 ) . The AWC neurons can drive acute behavioral responses as well as long-range chemotaxis , providing an additional assay for AWC function before and after learning . C . elegans locomotion alternates between forward runs and reversals , which are regulated by odors during chemotaxis ( Pierce-Shimomura et al . , 1999; Iino and Yoshida , 2009 ) . Odor removal increases AWCON activity and the frequency of reversals in wild-type animals , but not in chemotaxis-defective mutants , suggesting that the acute reversal assay and chemotaxis are related ( Albrecht and Bargmann , 2011 ) . Like odor removal , direct activation of AWCON with the light-activated ion channel Channelrhodopsin 2 ( ChR2 ) elicits reversals ( Gordus et al . , 2015 ) . AWCON::ChR2 bypasses odor transduction to depolarize AWC directly , thereby separating AWC sensory processing from AWC coupling to behavioral circuits . To ask how learning affects AWC behavioral output , AWCON::ChR2-expressing animals were conditioned with butanone and food deprivation and then stimulated with blue light while their locomotion was recorded . Naive animals responded to AWCON activation with an increase in reversals , as previously reported ( Figure 6A , B ) . By contrast , butanone-conditioned animals did not reverse in response to AWCON::ChR2 stimulation ( Figure 6A , B ) . This result suggests that aversive conditioning depresses AWCON coupling to target neurons that drive reversal behaviors . 10 . 7554/eLife . 14000 . 015Figure 6 . Behavioral responses to acute neuronal activation after odor conditioning . ( A , B ) Light-induced reversals in naive or conditioned wild-type animals expressing Channelrhodopsin ( ChR2 ) in AWCON . Data show the average fraction of animals executing reversals over time ( A ) , or the difference between the number of reversals initiated during and after stimulation ( 20 sec each , B ) . ( C , D ) Light-induced reversals in naive or conditioned gcy-28d::unc-103 ( gf ) animals expressing AWCON::ChR2 . Note increased basal frequency of reversals , a known property of AIA inactivation ( Chalasani et al . , 2010 ) . ( E , F ) Light-induced reversals in naive or conditioned age-1 ( hx546 ) animals expressing AWCON::ChR2 . ( G , H ) Light-induced reversals in wild-type AWCON::ChR2 animals after appetitive conditioning . ( I , J ) Light-induced reversals in AIB::ChR2 animals after aversive conditioning . Pale blue regions in ( A , C , E , G , I ) represent blue light stimulation . Shaded regions and error bars represents S . E . M . n = 7–14 assays per condition , 18–25 animals stimulated five times per assay . P values were generated by t-test with correction for unequal variance ( **p<0 . 001 , n . s . not significant ) . DOI: http://dx . doi . org/10 . 7554/eLife . 14000 . 01510 . 7554/eLife . 14000 . 016Figure 6—source data 1 . Data for induced reversal frequency in Figure 6B , D , F , H , J . DOI: http://dx . doi . org/10 . 7554/eLife . 14000 . 016 Suppression of the AWCON::ChR2 response required the AIA circuit , as gcy-28d::unc-103 ( gf ) animals responded equally strongly to AWCON::ChR2 stimulation with or without butanone conditioning ( Figure 6C , D ) . Similarly , the behavioral response to AWCON::ChR2 stimulation in age-1 mutants was not significantly altered by conditioning , matching their failure in aversive learning ( Figure 6E , F ) . Finally , we tested the effects of appetitive conditioning on AWC output using the same assay . After butanone was paired with food , AWCON::ChR2 stimulation elicited a substantially greater increase in reversals than it did in naive controls ( Figure 6G , H ) . Therefore , AWCON::ChR2-driven behavioral output demonstrates bidirectional plasticity after conditioning that mirrors aversive and appetitive learning behaviors . These results indicate that odor conditioning alters the coupling of AWC to motor circuits . To narrow down the locus at which this change might occur , we examined AIB , an interneuron that receives direct and indirect synaptic output from AWC . Channelrhodopsin activation of AIB elicits reversals in naive animals , like activation of AWC ( Gordus et al . , 2015; Piggott et al . , 2011 ) . AIB::ChR2 illumination elicited reversals equally well in naive and butanone-conditioned animals , localizing plasticity to a site upstream of or parallel to AIB ( Figure 6I , J ) . Studies of associative learning in many animals have delineated sensory pathways that represent the conditioned stimulus ( CS ) , the unconditioned stimulus ( US ) , and their convergence point within the neural circuit . In the best-known examples , sites of convergence are in higher brain areas . In Drosophila odor learning , olfactory CS information converges with electric shock or sucrose reward US information in the mushroom bodies ( Waddell , 2010 ) . In rodents , auditory CS information converges with electric shock US information in the amygdala ( Romanski et al . , 1993 ) . However , there are also examples of convergence in early sensory areas . In Aplysia , classical conditioning pairs an electric shock US with a gentle tactile CS to strengthen synapses of the tactile CS sensory neurons ( Hawkins et al . , 1983 ) . In the visually-mediated ciliary response of Hermissenda , CS and US pathways converge at two sites – first-order interneurons shared by the two pathways , and sensory neurons of the CS pathway ( Crow and Tian , 2006 ) . In butanone plasticity of C . elegans , the odor is analogous to the CS , while the food or food deprivation context could be considered the US . Learning is specific to the conditioned odor , as expected for a classical CS ( Colbert and Bargmann , 1995 ) . The AIA circuit and ins-1 , which are required for aversive but not appetitive learning , are needed to combine information about odor and food deprivation ( Figure 7 ) . We suggest that the AIA circuit senses food deprivation , and retrograde signaling to AWC merges this US information with odor-specific CS information . The AIM neurons that are implicated in long-term appetitive learning might have corresponding roles in encoding food as an appetitive US ( Lakhina et al . , 2015 ) . Organisms ranging from bacteria to mammals modify the sensitivity and dynamic range of sensory transduction after prolonged stimulation ( Fain et al . , 2001; Sourjik , 2004 ) . In AWCON , we observed a strong but transient suppression of calcium responses immediately after odor conditioning ( Chalasani et al . , 2010 ) , followed by a sustained shift in the sensitivity and dynamic range of the odor response . Aversive and appetitive conditioning elicited similar changes favoring higher odor concentrations , suggesting that this change represents sensory adaptation to the average odor intensity in the animal’s environment . The shift in AWCON dynamic range after odor conditioning may prevent saturation at higher odor concentrations , allowing detection of increasing odor concentrations against an odor background . Indeed , the increased butanone attraction after appetitive conditioning is most striking at high odor concentrations ( Torayama et al . , 2007 ) . Naive animals are only weakly attracted to these concentrations , perhaps reflecting the saturation of their AWCON calcium response . After aversive conditioning , the AWCON neuron has a diminished ability to drive reversal behavior . We suggest that this results from alterations in synaptic vesicle release by AWCON , which contains both glutamatergic and peptidergic vesicles ( Chalasani et al . , 2010 ) . Candidate mechanisms for plasticity are changes in the composition or level of the neuropeptides produced by AWC , changes in the relative rates of release of the two vesicle classes , or changes in resting potential that affect basal neurotransmitter release . Gene expression changes induced by nuclear EGL-4 or other transcriptional regulators may contribute to these functional effects ( Juang et al . , 2013; Neal et al . , 2015 ) . Sensory neuron synapses and related signaling pathways also contribute to gustatory learning in C . elegans . C . elegans migrates to salt concentrations that have been paired with food ( Kunitomo et al . , 2013; Luo et al . , 2014 ) , and learns the association over about 90 min , a time frame similar to that of olfactory learning . Salt preference changes are accompanied by synaptic changes in the ASER gustatory neuron that are induced by INS-1 , AGE-1 , and an axonal isoform of the DAF-2 insulin receptor ( Kunitomo et al . , 2013; Oda et al . , 2011; Ohno et al . , 2014; Tomioka et al . , 2006 ) . Bidirectional changes in ASER behavioral output below and above the preferred salt concentration are observed in a channelrhodopsin experiment similar to the one reported here for AWCON ( Kunitomo et al . , 2013 ) . Unlike olfactory learning , however , salt preference learning is accompanied by bidirectional shifts in the salt concentration that activates ASER in calcium imaging experiments ( Kunitomo et al . , 2013; Luo et al . , 2014 ) . Similarly , C . elegans temperature preference learning results in precise bidirectional shifts in the temperature sensitivity of the AFD thermosensory neurons ( Kimura et al . , 2004; Clark et al . , 2006; Ramot et al . , 2008; Yu et al . , 2014 ) . A key difference between olfactory learning and these other forms of plasticity is the nature of sensory preference . In thermotaxis and gustatory plasticity , animals seek out a preferred concentration at the 'setpoint' , the temperature or salt condition associated with chronic cultivation ( Hedgecock and Russell , 1975; Kunitomo et al . , 2013; Luo et al . , 2014 ) . Returning to the setpoint is the apparent homeostatic goal of the directed behavior . Olfactory plasticity , by contrast , is relatively insensitive to setpoint . It is linked to the identity of the odor , not the quantity , and the readouts of aversive and appetitive learning are observed across 100-fold concentration changes in the odor source ( Colbert and Bargmann , 1995; Torayama et al . , 2007 ) . Although odor history does change sensory representations in AWCON , this is not central to the behavioral preference; it is the context-dependent synaptic changes that appear to be instructive . Individual neurons may tune history-dependent changes in sensory detection and associative changes at sensory synapses in a variety of ways , generating different forms of sensory plasticity to solve different behavioral problems . Animals were grown at 20–22°C on Nematode Growth Medium ( NGM ) plates seeded with E . coli OP50 4–10 days before use . Wild type were Bristol strain N2 hermaphrodites . Standard molecular biology and microinjection methods ( Mello and Fire , 1995 ) were used to generate transgenic strains , listed below . StrainGenotypeFiguresCX15261kyIs617 [gcy-28d::HisCl1::SL2::GFP ( 5 ng/µl ) , myo-3::mCherry ( 5 ng/µl ) ]1CX14849kyEx4867 [ins-1 ( short ) ::HisCl1::SL2::mCherry ( 50 ng/µl ) , unc-122::GFP ( 10 ng/µl ) ]1CX16863kyIs698 [ttx-3intron7::HisCl1::SL2::GFP ( 30 ng/µl ) ]1CX15954kyEx5402 [str-3::HisCl1::SL2::GFP ( 100 ng/µl ) ]1CX15341kyEx5161 [unc-4::HisCl1::SL2::mCherry ( 50 ng/µl ) , elt2::mCherry ( 1 ng/µl ) ]1CX16862kyEx4867 [ins-1 ( short ) ::HisCl1::SL2::mCherry ( 50 ng/µl ) , unc-122::GFP ( 10 ng/µl ) ] + kyEx5402 [str-3::HisCl1::SL2::GFP ( 100 ng/µl ) ]1CX17181kyEx6003 [str-3::HisCl::SL2::mCherry ( 50 ng/µl ) , myo-3::mCherry ( 5 ng/µl ) ] + kyIs698 [ttx-3intron7::HisCl::SL2::GFP ( 30 ng/ul ) ]1CX17183kyEx6005 [unc-4::HisCl::SL2::mCherry ( 50 ng/ul ) , myo-3::mCherry ( 5 ng/ul ) ] + kyIs698 [ttx-3intron7::HisCl:SL2::GFP ( 30 ng/ul ) ] 1CX14908kyEx4924 [inx-1::hisCl1::SL2::GFP ( 30 ng/µl ) , myo-3::mCherry ( 5 ng/µl ) ]1CX14909kyEx4925 [ttx-3::hisCl1::SL2::GFP ( 50 ng/µl ) , myo-3::mCherry ( 5 ng/µl ) ]1CX16069kyEx5493 [glr-3::HisCl1::SL2::mCherry ( 50 ng/µl ) , elt-2::mCherry ( 1 ng/µl ) ]1CX16061kyEx5485 [str-1::HisCl1::SL2::GFP ( 50 ng/µl ) ]1CX15388kyEx5178 [tph-1 ( short ) ::HisCl1::SL2::mCherry PCR product ( 15 ng/µl ) ]1CX16040kyEx5464 [tdc-1::HisCl1::SL2::mCherry ( 50 ng/µl ) ]1CX16866kyIs617 [gcy-28d::HisCl1::SL2::GFP ( 5 ng/µl ) , myo-3::mCherry ( 5 ng/µl ) ] + kyEx5836 [ins-1 ( long ) ::dsRNA ( HisCl ) ( 25 ng/µl ) , unc-122::GFP ( 10 ng/µl ) ]1CX16867kyIs617 [gcy-28d::HisCl1::SL2::GFP ( 5 ng/µl ) , myo-3::mCherry ( 5 ng/µl ) ] + kyEx5837 [unc-4::dsRNA ( HisCl ) ( 100 ng/µl ) , unc-122::GFP ( 10 ng/µl ) ]1CX14599kyEx4747 [gcy-28d::unc-103 ( gf ) ::SL2::mCherry ( 30 ng/µl ) , elt-2::mCherry ( 2 ng/µl ) , pSM ( 70 ng/µl ) ]1CX7155ins-1 ( nr2091 ) 2JN1704ins-1 ( nr2091 ) ; peEx1704 [ins-1 ( short ) ::ins-1::Venus]2TJ1052age-1 ( hx546 ) 2CX17261age-1 ( hx546 ) ; kyEx6035 [str-2::age-1::SL2::GFP ( 50 ng/ul ) ]2CX16499kyIs678 [odr-3::GFP::egl-4 ( 5 ng/µl ) , elt-2::nlsGFP ( 5 ng/µl ) , pSM ( 90 ng/µl ) ]3CX16500kyIs678 [odr-3::GFP::egl-4 ( 5 ng/µl ) , elt-2::nlsGFP ( 5 ng/µl ) , pSM ( 90 ng/µl ) ] + kyEx4747 [gcy-28d::unc-103 ( gf ) ::SL2::mCherry ( 30 ng/µl ) , elt-2::mCherry ( 2 ng/µl ) , pSM ( 70 ng/µl ) ]3CX16674age-1 ( hx546 ) ; kyIs678 [odr-3::GFP::egl-4 ( 5 ng/µl ) , elt-2::nlsGFP ( 5 ng/µl ) , pSM ( 90 ng/µl ) ]3CX13914kyEx4275 [str-2::GCaMP5A ( 10 ng/µl ) , unc-122::dsRed ( 10 ng/µl ) ]4CX16213kyEx5527 [str-2::nlsGCaMP6s ( 30 ng/µl ) , elt-2::nlsGFP ( 5 ng/µl ) ]4 , 5CX16503kyEx4747 [gcy28d::unc-103 ( gf ) ::SL2::mCherry ( 30 ng/µl ) , elt-2::mCherry ( 2 ng/µl ) ] + kyEx5527 [str-2::nlsGCaMP6s ( 30 ng/µl ) , elt-2::nlsGFP ( 5 ng/µl ) ]5CX17242age-1 ( hx546 ) ; kyEx5527 [str-2::nlsGCaMP6s ( 30 ng/µl ) , elt-2::nlsGFP ( 5 ng/µl ) ]5CX14418kyEx4605 [str-2::ChR2 H134::GFP ( 50 ng/µl ) , myo-3:mCherry ( 10 ng/µl ) ]6CX16670kyEx4605 [str-2::ChR2 H134::GFP ( 50 ng/µl ) , myo-3:mCherry ( 10 ng/µl ) ] + kyEx4747 [gcy-28d::unc-103 ( gf ) ::SL2::mCherry ( 30 ng/µl ) , elt-2::mCherry ( 2 ng/µl ) ]6CX17243age-1 ( hx546 ) ; kyEx4605 [str-2::ChR2 H134::GFP ( 50 ng/µl ) , myo-3:mCherry ( 10 ng/µl ) ]6CX13210kyEx3838 [inx-1:ChR2 H134::GFP ( 30 ng/µl ) , unc-122::GFP ( 20 ng/µl ) ]6 Chemotaxis was tested on square plates containing 10 mL of chemotaxis agar ( 1 . 6% agar , 5 mM phosphate buffer pH 6 . 0 , 1 mM CaCl2 , 1 mM MgSO4 ) , poured the day before the assay . Adult animals were washed three times with S basal buffer and once with chemotaxis buffer ( 5 mM phosphate buffer pH 6 . 0 , 1 mM CaCl2 , 1 mM MgSO4 ) , and 100–200 animals placed at the center of the square plate . Two 1 µl drops of butanone diluted at 1:1000 in ethanol , or ethanol control , were spotted on each side of the plate at the beginning of the assay , with 1 µl 1 M NaN3 at the same spot to immobilize animals that reached the butanone source or ethanol source . After 1–2 hr , plates were moved to 4°C to stop the assay . The assay was quantified by counting animals that had left the origin , on each side of the plate ( #Odor , #Control ) and in the middle ( #Other ) , and calculating a chemotaxis index as [#Odor - #Control] / [#Odor + #Control + #Other] . The odor concentrations experienced by the animal during chemotaxis are estimated to span nanomolar to micromolar concentrations; the point sources provide 2 µl of 11 mM butanone , which disperses through the 110 ml volume of the chemotaxis plate for an average concentration of 200 nM . Assay conditions were modified from Colbert and Bargmann ( 1995 ) and L’Etoile et al . ( 2002 ) . One-day old adults were washed three times with S basal buffer and placed in 2 ml glass vials ( VWR 66009–556 ) containing 1 ml of S basal buffer with or without butanone diluted to a final concentration of 1 mM . Vials were laid horizontally to maintain aeration during odor conditioning . Previous studies have shown that aversive olfactory learning towards butanone is comparable after conditioning on agar plates or in liquid ( L’Etoile et al . , 2002 ) . After conditioning for 90 min , animals were washed twice with S basal and once with chemotaxis buffer before being tested in butanone chemotaxis assays . Each test population consisted of 50–200 animals , and experiments were repeated a minimum of three times . Plates that had less than 50 animals outside the origin at the end of the assay were excluded from analysis to ensure that the chemotaxis index was an accurate representation of the population . For histamine experiments , histamine dihydrochloride ( Sigma ) was added during the conditioning phase at a final concentration of 10 mM as described in Pokala et al . ( 2014 ) . Assays were modified from Torayama et al . ( 2007 ) . One-day old adults were washed three times with S basal buffer and placed on NGM plates seeded with E . coli OP50 . 12 µl of undiluted butanone was spotted onto agar plugs on the lid of the plate , and the plate was sealed with parafilm . The effective butanone concentration is approximately the same as the concentration used in aversive conditioning . Mock-conditioned groups were placed on seeded plates without butanone on the lid . After 90 min , animals were washed twice with S basal and once with chemotaxis buffer and assayed for chemotaxis to 2 µl of a 1:10 dilution of butanone ( 100-fold more than in the aversive olfactory learning assay; the average odor concentration experienced in this chemotaxis assay is 20 µM ) . Animals ranging from L2 to adults were mounted on a 2% agar pad containg 5 mM NaN3 under a glass cover slip . The HisCl1 and unc-103 ( gf ) transgenes are coupled to GFP and mCherry sequences , respectively , in bicistronic operons with an SL2 splice leader sequence before the fluorescent protein . Animals were observed under a 60X objective of a Zeiss Axioskop microscope and neurons expressing the fluorescent marker were identified based on known landmarks and morphological characteristics . Animals carrying the gcy-28d::HisCl1 transgene showed expression in AIA ( 100% ) , AVF ( ~90% ) , ASI ( ~65% , weak ) , and IL2 , I1 , or M3 pharyngeal neurons ( ~70% each ) . Animals carrying the gcy-28d::unc-103 ( gf ) transgene showed expression in AIA ( ~95% ) , AVF ( <5% ) , ASI ( <5% ) , and IL2 , I1 , or M3 neurons ( ~5% each ) . The gcy-28d::unc-103 expression pattern suggests that silencing AIA alone can disrupt aversive learning , but this appears inconsistent with the complete set of HisCl1 data in Figure 1C , because aversive learning was normal with two HisCl1 transgenes that are expressed in AIA ( ins-1 ( s ) , ttx-3 ( intron7 ) ) . Two potential explanations for this apparent discrepancy are ( 1 ) chronic AIA silencing with unc-103 ( gf ) disrupts aversive learning , but transient AIA silencing during conditioning with HisCl1 does not unless additional neurons are silenced ( 2 ) in addition to AIA , a low level of unc-103 ( gf ) in another neuron , below the detection limit of the mCherry reporter , silences that neuron in the gcy-28d::unc-103 ( gf ) strain . Plasmids driving the sense and anti-sense sequences of HisCl1 under cell-specific promoters were injected at equal concentrations . Double-stranded RNAs in C . elegans can act systemically , so control experiments were conducted with similar sense/anti-sense plasmids directed against the GFP marker in the gcy-28d::HisCl1 strain . These controls showed that the GFP dsRNA expressed from the unc-4 or ins-1 promoters silenced expression in AVF or AIA , respectively , without silencing expression in the reciprocal neuron or in other gcy-28d-expressing neurons , e . g . in the pharynx . The ins-1 promoter is also expressed weakly in ASI , and might silence HisCl1 there as well . This possibility was difficult to assess rigorously because the weak gcy-28d expression in ASI was already near the detection threshold . A plasmid with the odr-3 promoter driving GFP::EGL-4 ( Lee et al . , 2010 ) was injected into wild-type animals at 5 ng/µl . A spontaneous integrant of the resulting extrachromosomal array was backcrossed four times to wild-type N2 animals . This strain had normal chemotaxis to butanone and normal aversive olfactory learning behavior . One-day old adults were conditioned to butanone in glass vials as described above , washed three times with S basal buffer , and mounted on a 2% agar pad containing 5 mM NaN3 under a glass cover slip . Fluorescence was scored within 20 min to avoid effects of NaN3 on EGL-4 localization . The AWC neuron proximal to the objective was identified using a 40x objective of a Zeiss Axioskop microscope and a Z-stack was taken through the cell using Zeiss Axiovision software for image capture . From each stack , the image containing the central plane of the cell was selected for quantification . Fluorescence values of the cytoplasmic and nuclear regions were quantified using ImageJ software . 25–30 animals per group were imaged during each experiment , and three experiments were conducted for each group . Results were plotted as cumulative distribution plots , and statistical significance measured by the Kolmogorov-Smirnov test . Butanone conditioning results in EGL-4 translocation only in the AWCON cell ( Lee et al . , 2010 ) . The strain used here did not have a marker to distinguish AWCON from AWCOFF , because AWCON –specific promoters were not strong enough to drive EGL-4::GFP reliably , and attempts to include a second AWCON transgene as a marker with EGL-4::GFP resulted either in interference between the two transgenes , or in defects in butanone chemotaxis or aversive olfactory learning . Therefore , it should be assumed that 50% of the AWC neurons that were imaged were AWCON and 50% were AWCOFF . For the calcium imaging data in Figure 4A , custom-fabricated polydimethylsiloxane devices were used for imaging and stimulus delivery as described in Chalasani et al . ( 2007 ) . Animals were conditioned outside the device for 90 min and single animals were loaded into the device for calcium imaging . For all other calcium imaging experiments , imaging and stimulus delivery was performed as described in Larsch et al . ( 2013 ) . Devices with two separate arenas suitable for simultaneous imaging of two conditions or genotypes were flooded with S basal buffer containing 10 mM tetramisole hydrochloride to paralyze the animals , and 6–12 adult animals were loaded into each arena . For aversive conditioning , animals were conditioned in the device for 90 min with 1 mM butanone in S basal buffer , followed by 15 min of S basal buffer without odor . To ensure that butanone absorption by the PDMS device did not affect the results , a smaller number of animals were tested in a control experiment in which animals were conditioned outside the device , washed , and loaded into a fresh imaging device that had not been exposed to odor for calcium imaging . Naive and conditioned animals had calcium responses to 111 nM and 111 µm butanone that were similar to those of animals conditioned within the device in Figure 4 . For appetitive conditioning , animals were conditioned outside the device as in the behavioral assay , because introducing E . coli into the microfluidic devices contaminates the arena . 2–4 experiments were conducted for each group . Odor stimuli were delivered using a three-way valve to switch between buffer and odor flow into the arena , and a Hamilton valve to switch between different odor concentrations . The stimulation protocol consisted of three 30-second alternations between buffer and odor at each concentration , followed by one minute of buffer , and then the sequence was repeated at a 10-fold higher odor concentration for a total of six butanone concentrations ranging from 11 nM to 1 mM . Calcium responses to odors were monitored in animals expressing GCaMP5A ( Figure 4A; Akerboom et al . , 2012 ) , or nuclear-localized GCaMP6s ( Figures 4B–D , 5; Chen et al . , 2013 ) under the str-2 promoter , which is selectively expressed in AWCON neurons . Fluorescence was monitored on a Zeiss AxioObserverA1 microscope with a 2 . 5X objective , and Metamorph software was used for synchronized image capture with pulsed illumination . Fluorescence was measured using a custom ImageJ script as in Larsch et al . ( 2013 ) . Custom Matlab scripts were written to plot the fluorescence response over time as well as measure response magnitude and recovery time ( see Source code 1 ) . Integrated fluorescence values were used . ΔF/Fo was generated by dividing fluorescence values by the baseline fluorescence , which was determined to be the fluorescence at t = 1700 s ( Figure 4A ) . ΔF/Fmax was generated by normalizing each trace on a 0–1 scale , where 0 and 1 are defined as the mean of the lowest and highest 5% of pixel intensities , respectively ( all other Figures ) . AWC response magnitude was measured as the decrease in normalized fluorescence following the first odor addition in the series , which was calculated by subtracting the average fluorescence in the last 10 s during the odor presentation from the average fluorescence from the 2 s preceding the odor addition . Half-time of recovery was defined as the time for fluorescence to recover to 50% of the peak magnitude after the last odor removal in the series , and was only calculated when there was a response to odor ( defined as a reduction in fluorescence of magnitude 0 . 075 or greater ) . Averages in Figure 4D , 5C , F and I were only calculated when at least half of the individuals responded to odor . Naive animals did not recover fully from the last 1 mM pulse within the five minutes of recording ( Figure 4B , 5A , C ) , so for this concentration the half-time of recovery is a lower bound . L4-stage animals expressing the H134R variant of Channelrhodopsin 2 ( Lin et al . , 2009 ) under the str-2 ( AWCON ) or inx-1 ( AIB ) promoter were incubated overnight on plates seeded with E . coli OP50 and 50 µM retinal . The following day , adult animals were conditioned with odor and with or without food as described above . After conditioning , animals were washed twice with S basal and once with NGM buffer ( 25 mM phosphate buffer pH 6 . 0 , 1 mM CaCl2 , 1 mM MgSO4 , 52 mM NaCl ) and transferred to a 6 cm NGM plate with a square filter ring soaked in 20 mM CuCl2 to prevent animals from crawling out of the field of view . Animals received 20-second pulses of blue light ( 455 nm , 25 µW/mm2 ) every two minutes , repeated ten times , and were video-recorded; pulses six through ten were analyzed . A Pixelink camera and Streampix software were used to generate recordings . Movies were analyzed using custom Matlab scripts that tracked animal locomotion and identified reversals ( Gordus et al . , 2015 ) . Custom Matlab scripts were used to bin and plot the frequency of events over time ( see Source code 2 ) . Each group was tested on 7–14 experimental trials , with 15–25 animals in each trial .	We learn from experience . When we repeatedly encounter a signal that is coupled to either reward or punishment , we eventually learn to expect the two to occur together . This phenomenon is called associative learning . Within the brain , distinct groups of neurons process information about the signal and about reward and punishment . In addition to storing information individually ( as non-associative memories ) , the neurons communicate with one another and combine their information to create associations . Like humans and many other animals , the roundworm and model organism Caenorhabditis elegans can learn to associate odors with rewards or punishments . By teaching worms that a scent predicts either food or a lack of food , Cho et al . now show that different cells and molecules support the formation of these two associations . C . elegans detect odors using sensory neurons . Repeated exposure to an odor reduces a neuron’s sensitivity to that odor , and Cho et al . show that this occurs irrespective of whether the odor is paired with reward or with punishment . This indicates that the neuron stores information about the odor as a non-associative memory . By contrast , pairing an odor with reward has differing effects on associative learning to pairing that same odor with punishment . Pairing an odor with a reward increases a sensory neuron’s ability to communicate with target neurons – ultimately , those that control movement – whereas odor-punishment pairing reduces this ability . Further experiments showed that an insulin peptide supports learning about odors and punishments , but not about odors and rewards . The next challenge is to identify the molecules that strengthen or weaken communication between sensory neurons and target neurons after associative learning . It will also be important to identify the other neurons and molecules that detect rewards and punishments , to gain a more complete picture of how the brain acquires this information .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "neuroscience" ]	2016	Parallel encoding of sensory history and behavioral preference during Caenorhabditis elegans olfactory learning
Neural activity has been implicated in the motility and outgrowth of glial cell processes throughout the central nervous system . Here , we explore this phenomenon in Müller glia , which are specialized radial astroglia that are the predominant glial type of the vertebrate retina . Müller glia extend fine filopodia-like processes into retinal synaptic layers , in similar fashion to brain astrocytes and radial glia that exhibit perisynaptic processes . Using two-photon volumetric imaging , we found that during the second postnatal week , Müller glial processes were highly dynamic , with rapid extensions and retractions that were mediated by cytoskeletal rearrangements . During this same stage of development , retinal waves led to increases in cytosolic calcium within Müller glial lateral processes and stalks . These regions comprised distinct calcium compartments , distinguished by variable participation in waves , timing , and sensitivity to an M1 muscarinic acetylcholine receptor antagonist . However , we found that motility of lateral processes was unaffected by the presence of pharmacological agents that enhanced or blocked wave-associated calcium transients . Finally , we found that mice lacking normal cholinergic waves in the first postnatal week also exhibited normal Müller glial process morphology . Hence , outgrowth of Müller glial lateral processes into synaptic layers is determined by factors that are independent of neuronal activity . Bidirectional signaling between neurons and glia is essential for circuit formation and function throughout the nervous system . Across developmental steps from neurogenesis to circuit maturation , glia monitor their environment and in turn regulate neuronal production , migration , and differentiation , promote synapse turnover , and regulate synaptic function via neurotransmitter uptake and ion buffering ( Allen and Lyons , 2018 ) . In the vertebrate retina , for example , Müller glia exhibit neurogenic potential ( Bernardos et al . , 2007; Ji et al . , 2017; Guimarães et al . , 2018 ) , promote circuit-specific wiring via secretion of synaptogenic molecules ( Koh et al . , 2019 ) , regulate phagocytosis of neuronal debris ( Bejarano-Escobar et al . , 2017 ) , and limit neurotransmitter spillover via transporter activity ( Bringmann et al . , 2013 ) . There is extensive evidence that neuronal signaling influences glial physiology in the adult brain . Individual astroglia extend fine processes that contact thousands of synapses and express an array of neurotransmitter receptors ( Allen and Eroglu , 2017 ) . This enables glia to rapidly integrate neuronal activity , often involving intracellular calcium mobilization . Neuronal activity-evoked calcium events in astroglia range in size from small membrane-proximal microdomains to global cytosolic events mediated by various transmembrane proteins and calcium sources ( Shigetomi et al . , 2016; Bindocci et al . , 2017; Nimmerjahn and Bergles , 2015 ) . Pharmacological studies suggest compartmentalized and global calcium events are mediated by separate mechanisms and evoke different functional responses within astroglia , which lead to differential effects on neurotransmission ( Di Castro et al . , 2011; Chen et al . , 2020; Srinivasan et al . , 2015 ) . Here , we use the mouse retina as a model to explore mechanisms and a possible function of neuron-glia signaling during development . In the retina , Müller glia are the predominant glial type , tiling the entire retinal space and interacting with every retinal cell type ( Wang et al . , 2017 ) . Similar to Bergmann glia of the cerebellum ( De Zeeuw and Hoogland , 2015; Lippman et al . , 2008 ) , Müller glia exhibit a radial structure with a stalk extending from the soma , lateral processes extending from the stalk within synaptic layers , and endfeet contacting neuronal somata , axons , and vasculature . In the adult , this complex morphology enables Müller glia to mediate neurovascular coupling , maintain pan-retinal ion homeostasis , and modulate neuronal signaling in the adult ( Newman , 2003; Newman , 2015; Reichenbach and Bringmann , 2013 ) . During both mouse ( Rosa et al . , 2015 ) and zebrafish ( Zhang et al . , 2019 ) development , Müller glia undergo calcium transients in response to retinal waves , a term used to describe spontaneous bursts of depolarization that propagate across the retina . The functional relevance of wave-associated calcium signaling in Müller glia is not known . One potential role for wave-associated calcium-signaling in Müller glia is in modulating outgrowth of lateral processes , which initiates and progresses during the same developmental window as retinal waves ( Wang et al . , 2017 ) . In other brain regions and model systems , neuronal activity-evoked calcium transients lead to morphological changes in astroglial processes . For example , in the hippocampus and somatosensory cortex , perisynaptic astrocytic processes undergo spatially localized , glutamate-evoked calcium transients which are essential for activity-evoked ensheathment of synapses ( Bernardinelli et al . , 2014 ) . In the cerebellum , Bergmann glial processes are highly motile during synaptogenic periods , and their ensheathment of Purkinje cells is impaired when glial glutamate transporters are knocked out ( Lippman et al . , 2008; Miyazaki et al . , 2017 ) . Radial glia of the Xenopus optic tectum exhibit neuronal activity-evoked calcium transients and motility during visual system development ( Sild et al . , 2016 ) . In many of these systems , when glial motility is blocked , synaptic development , function , and plasticity are impaired ( Van Horn and Ruthazer , 2019 ) , highlighting the importance of glial dynamics in setting up and maintaining neural circuits . Here , we combine morphological and calcium imaging , electrophysiology , and pharmacology to characterize Müller glial lateral process outgrowth during retinal waves and to determine the impact of neuronal activity on Müller glial morphology . Our goal is to determine whether spontaneous activity driven by retinal waves influences the morphological development of Müller glial cells . After their differentiation from retinal progenitor cells , which occurs during the first postnatal week , Müller glia extend processes laterally from their stalk into the inner plexiform layer ( IPL ) ( MacDonald et al . , 2017 ) . To visualize this process , we used the Slc1a3-CreER;mTmG reporter line to sparsely label Müller glia ( Figure 1A ) . Lateral processes exhibited sublayer-specific outgrowth and distribution , and they reached an adult level of complexity soon after eye opening around postnatal day 14 ( P14 ) ( Figure 1B and D; see Supplementary files 1 and 2 for statistical comparisons across sublayer and age , respectively ) . During the first days of Müller glia differentiation , lateral processes preferentially occupied the borders of the IPL ( putative S1/S5 ) , with most outgrowth occurring in the prospective ON half of the IPL ( S3–S5 ) until about P10 . By eye opening , after P12 , processes arborized throughout the IPL and began to resemble their adult distribution , characterized by fewer processes in IPL sublayers S2/S4 than in S1/S3/S5 . These observations are consistent with previous findings using the same mouse line and immunohistochemistry in fixed tissue ( Wang et al . , 2017 ) . During outgrowth from the primary stalks , we observed that Müller glial lateral processes across the IPL were highly motile . To characterize process motility , we conducted volumetric two-photon imaging of sparsely labeled processes at a rate of roughly 1 volume every 2 min for 10-min epochs . We characterized motility that occurred during the imaging period by grouping events into distinct categories as follows: new processes that branched from stalks or from existing processes , extension of existing processes , retraction of processes without elimination , elimination of existing processes , or stable for processes that did not change length . We made several observations based on this analysis ( Figure 1E ) . First , Müller glial cell lateral processes were highly motile across the IPL throughout the entire second postnatal week . Second , there was a slight bias toward new process sprouting rather than process loss . Third , we found there were roughly equal proportions of extending and retracting processes over 10-min epochs of imaging , highlighting their rapid turnover during this period of development . Finally , we observed a sharp developmental transition toward stability a few days after eye opening ( Figure 1C , E and F; Figure 1—figure supplement 1; Supplementary file 3 ) . To assure that we could observe and quantify changes in process motility , we applied two manipulations . First , we slowed process motility with pharmacological agents that disrupt cytoskeletal proteins . Process motility was reduced by bath application of the microtubule-disrupting agent nocodazole ( 10 µM ) and the actin polymerization inhibitor cytochalasin-D ( 5 µM ) , the combination of which led to processes stabilization ( Figure 1G; Supplementary file 4 ) . It is unlikely that this loss of motility resulted from damaged cell health , as Müller glial process morphology remained normal in the presence of cytochalasin and nocodazole . Müller glia , like astroglia throughout the brain , exhibit reactive gliosis associated with hypertrophy and expression of filamentous membrane processes when exposed to damage ( Graca et al . , 2018 ) , which we did not observe here . Second , we enhanced process motility by bath application of epidermal growth factor ( EGF; one unit [100 ng]/ml; Figure 1H; Supplementary file 5 ) , which activates EGF receptors ( EGFRs ) . EGFRs are expressed by Müller glia during the second postnatal week ( Close et al . , 2006 ) and increase motility in other cells via activation of Rac- and Rho-GTPase-dependent pathways ( Pena et al . , 2018; Liu and Neufeld , 2004 ) . Taken together , these results indicate that lateral process motility is facilitated by turnover of actin filaments and microtubules and may be modulated by growth factor signaling , pathways that potentially interact with neuronal activity ( Georgiou et al . , 1999; Lavoie-Cardinal et al . , 2020 ) . Much of the morphological development of Müller glia occurs when retinal waves are present ( Figure 1 and Wang et al . , 2017 ) . Furthermore , previous studies have demonstrated that retinal waves induce increases in intracellular calcium in Müller glia ( Rosa et al . , 2015; Zhang et al . , 2019 ) . However , these studies did not assess whether there are distinct calcium compartments within Müller glia that could potentially have distinct impacts on lateral process motility . The two compartments in Müller glia we focused on were their central stalks and their lateral processes within the IPL . We conducted simultaneous two-photon calcium imaging of Müller glial stalks and processes and voltage-clamp recordings from retinal ganglion cells ( RGCs ) . Several strategies for calcium imaging from Müller glia were used . First , retinas were isolated from Slc1a3-CreER; cyto-GCaMP6fflox mice , which express genetically encoded calcium indicator selectively in Müller glia . Second , a subset of retinas were isolated from WT mice and bath-loaded with the chemical calcium dye Cal520 , which selectively labels Müller glia ( Rosa et al . , 2015; Uckermann et al . , 2004 ) . Using these two approaches , we were able to clearly identify the boundaries of Müller glial stalks , while calcium in lateral processes was assessed using regions of interest ( ROIs ) within the areas of IPL intervening glial stalks ( Figure 2A ) . Third , retinas were isolated from mice expressing a membrane-bound calcium reporter in Müller glia ( Lck-GCaMP6f in Slc1a3-CreER mice ) ( Srinivasan et al . , 2016 ) , which enabled resolution of lateral processes but poor detection of cytosolic calcium transients within stalks ( Figure 2—figure supplement 1 ) . Using all of these approaches , we verified that Müller glial stalks and lateral processes exhibited calcium transients in response to retinal waves as previously reported ( Rosa et al . , 2015 ) , as well as spontaneous , non-wave-associated calcium transients . We observed several differences in wave-associated calcium transients between stalks and lateral processes . Wave-associated stalk transients propagated throughout the stalk and involved many processes , which we refer to as a global transient . Lateral processes participated in some global transients but also exhibited wave-associated calcium transients independent of stalks ( Figure 2A and B and Figure 2—figure supplement 1 ) . Hence , lateral processes and stalks differentially participated in waves , with a greater proportion of lateral process ROIs responding to waves than stalk ROIs . In addition , wave-associated calcium transients in lateral processes were consistently observed from P9 to P12 , while those in stalks were significantly reduced in terms of ΔF/F amplitude and proportion of stalks participating by P11/12 ( Figures 2C and 3A; Supplementary file 6 ) . Finally , the subset of stalks that responded to retinal waves did so with a slight delay relative to lateral processes ( Figure 2B and D; Supplementary file 7 ) . Hence , Müller glial stalks and lateral processes can undergo spatiotemporally distinct calcium signaling events . We next sought to test whether wave-associated calcium transients in stalks versus lateral processes are mediated via activation of distinct signaling pathways . Muscarinic acetylcholine receptors ( mAChRs ) have been implicated in Müller glial responses to retinal waves ( Rosa et al . , 2015 ) , and isolated Müller glia undergo M1 mAChR-dependent responses to cholinergic agonists ( Da Silva et al . , 2008 ) . Bath application of pirenzepine ( 5 µM ) , a selective antagonist of M1 mAChRs , led to a significant reduction in wave-associated calcium transients in stalks and lateral processes ( Figure 3A ) . The magnitude of this effect was age- and compartment-dependent , as wave-associated transients in Müller glial stalks had greater sensitivity to M1 mAChR block at P9/10 than at P11/12 , and pirenzepine sensitivity was significantly greater in stalks than in lateral processes at early ages ( Figure 3B and C; Supplementary files 8–11 ) . Interestingly , we occasionally detected wave-associated stalk calcium transients that were insensitive to pirenzepine . Pirenzepine-insensitive stalk responses exhibited similar amplitude and latency to those of lateral processes ( Figure 3C and D; Supplementary file 12 ) . These data suggest that wave-associated responses in stalks are controlled by multiple mechanisms in addition to activation of M1 mAChRs . Note , pirenzepine had minimal impact on the amplitude and frequency of compound excitatory postsynaptic currents ( EPSCs ) and inter-wave interval ( IWI , Figure 3E; Supplementary file 13 ) , indicating pirenzepine had a selective effect on Müller glial M1 mAChRs , rather than on wave-generating circuits . To confirm that M1 mAChR signaling preferentially impacts wave-associated stalk transients , we enhanced acetylcholine ( ACh ) release during waves by bath-applying the GABAA receptor antagonist gabazine ( 5 µM ) ( Wang et al . , 2007 ) . As expected , the presence of gabazine increased the amplitude of EPSCs recorded from RGCs during retinal waves ( Figure 4E; Supplementary file 19 ) . Gabazine also increased the proportion of stalks and lateral process ROIs that underwent wave-associated calcium transients ( Figure 4; Supplementary file 14 ) , but with differential compartment-specific effects: there was a significantly larger gabazine-induced fold-change in proportion of stalks compared with lateral processes responding to waves , and gabazine led to a nearly twofold increase in the amplitude of wave-evoked fluorescence change in stalks , while only slightly increasing the response amplitude in lateral processes ( Figure 4C; Supplementary files 15–17 ) . Subsequent application of pirenzepine reduced the extent , magnitude , and latency of stalk responses to waves while largely sparing lateral process responses ( Figure 4B–D; Supplementary files 14–18 ) and without altering wave properties ( Figure 4E; Supplementary file 19 ) . This effect was independent of age from P9 to P12; pirenzepine suppressed wave-associated transients in stalks following gabazine-mediated upregulation of responses throughout this period . Taken together , these data indicate that M1 mAChRs mediate wave-associated global calcium transients in Müller glial stalks and support the conclusion that stalks and lateral processes are functioning as distinct calcium compartments , with lateral processes undergoing non-M1 mAChR-mediated calcium transients during retinal waves . Is Müller glial lateral process motility impacted by neuronal activity ? To test whether neuronal activity influences Müller glial lateral process motility , we assessed the impact of multiple pharmacological agents on process motility after the first postnatal week ( Figure 5 ) . First , we assessed the impact of agents that modulated mAChR signaling resulting from neural activity ( Figure 5A ) . Gabazine , which potentiated calcium transients in Müller glial processes and stalks , in part via activation of M1 mAChRs , did not change the proportion of processes undergoing motility . Similarly , there was no significant change in lateral process motility in the presence of pirenzepine , which blocks M1 mAChRs and preferentially abolished wave-associated calcium transients in Müller glial stalks ( Figure 5A and F; see Supplementary file 20 for summary statistics ) . Second , we assessed the impact of glutamatergic signaling , which is also implicated in wave-associated Müller glial calcium transients ( Rosa et al . , 2015; Zhang et al . , 2019 ) . TBOA blocks glutamate transporters and enhances wave-associated calcium transients in Müller glia by increasing glutamate spillover , while DNQX and AP5 block glutamatergic retinal waves and reduce calcium transients in Müller glia ( Rosa et al . , 2015; Blankenship et al . , 2009; Figure 3—figure supplement 1 ) . Similar to our results in gabazine and pirenzepine , the presence of TBOA or DNQX/AP5 had no significant impact on motility ( Figure 5B and F ) . These observations together indicate that lateral process motility does not arise from wave-associated calcium transients in stalks or processes . To further explore a role for ACh and glutamate in influencing motility , we tested whether the local release of neurotransmitter would alter the motility of nearby processes , as is the case for perisynaptic astrocytic processes elsewhere in the brain , where it is postulated that local elevations in neurotransmitter promote process growth and synapse coverage ( Bernardinelli et al . , 2014; Arizono et al . , 2020 ) . We locally perfused via continuous iontophoresis the AChR agonist carbachol or glutamate onto Müller glial processes in Slc1a3-CreER;mTmG retinas with sparse Cre-mediated recombination . In parallel experiments using retinas from Slc1a3-CreER;GCaMP6fflox mice , we confirmed that this method induced strong calcium transients in stalks and lateral processes . Despite this , we observed no impact of local perfusion of agonists on Müller glial lateral process motility ( Figure 5C , D and F ) . As a final test for whether intracellular calcium signaling in Müller glia is required for motility , we assessed motility after bath-loading the potent calcium chelator , BAPTA-AM . Despite blocking retinal waves and all calcium transients in Müller glia ( Figure 5—figure supplement 1 ) , the presence of BAPTA did not alter the proportion of processes exhibiting motility . Taken together these data suggest that motility of Müller glial lateral processes persists in the absence of intracellular calcium transients or neuronal activity ( Figure 5E and F ) . We note however that some of these observations were made using relatively small numbers of cells , and we cannot rule out a small effect of these pharmacological perturbations given our sample size and high cell-to-cell variability in lateral process motility . Thus far , we have assessed the role of neuronal activity and calcium signaling on Müller glial lateral process motility using live imaging during the second postnatal week , while there is lateral process outgrowth . However , during the first postnatal week , there are wave-associated calcium transients in Müller glia also mediated by activation of mAChRs ( Rosa et al . , 2015 ) . To test whether retinal wave-associated calcium signaling during this first postnatal week influences later process outgrowth during the second postnatal week , we assessed Müller glial morphology in retinas isolated from mice lacking the β2 nicotinic acetylcholine receptor subunit ( β2-nAChR-KO ) . β2-nAChR-KO mice exhibit significantly reduced cholinergic retinal waves ( Bansal et al . , 2000; Burbridge et al . , 2014; McLaughlin et al . , 2003 ) and therefore have reduced M1 mAChR-induced signaling in glial cells . Morphology of individual Müller glial cells was assessed by filling them with Alexa-488 via sharp pipette in P12 and P30 wild-type ( WT ) and β2-nAChR-KO mice . Filled cells were visualized via two-photon volumetric imaging in live tissue , and processes were traced for subsequent morphological analysis ( Figure 6A–C ) . At both P12 and P30 , individual dye-filled Müller cells exhibited lateral processes throughout the IPL , with more processes in the ON-half of the IPL than in the OFF-half prior to eye opening , in agreement with Figure 1 . Detailed assessment shows that by P12 , Müller glial lateral processes exhibited nearly the same level of complexity as observed in the adult . In addition , we found no significant difference between WT control and β2-nAChR-KO Müller glia in terms of several morphological parameters including complexity , number , area , and length of lateral processes . However , we did observe a significant reduction in number of primary branches projecting from the stalk in β2-nAChR-KO Müller glia before eye opening , while in mature retina all morphological properties of in β2-nAChR-KO Müller glia appeared normal ( Figure 6D and E; Supplementary file 21 ) . This suggests that although retinal waves may play a role in process sprouting from Müller glial stalks during development , other mechanisms produce lateral processes in normal numbers and appearance as the retina matures . Glial morphology is critical for normal development of circuits . For example , a plexus of glial processes provides a diffusional barrier for neurotransmitters and other signaling molecules ( Bringmann et al . , 2013; Syková , 2001 ) , undergoes contact-mediated signaling with neurites to modulate synapse formation and function ( Filosa et al . , 2009; Murai et al . , 2003 ) , and participates in synaptic pruning via activation of phagocytic pathways in glia ( Chung et al . , 2013 ) . Across brain regions , neuron-glia signaling impacts glial morphology in diverse ways . We found in the retina that Müller glial lateral process motility during development was not impacted by neuron-glia signaling via release of neurotransmitter . This is similar to Bergmann glia , radial glia of the cerebellum , whose process motility is developmentally regulated and not directly affected by perturbations of neuronal activity or calcium influx ( Lippman et al . , 2008 ) . Similarly in microglia , another highly motile cell type , chelation of calcium with BAPTA-AM slightly slowed but did not block process motility ( Pozner et al . , 2015 ) . In striatal astrocytes , FRET-based synaptic proximity assays have shown that although fine branches interact with synapses , these interactions are stable and unaffected by neuronal activity ( Octeau et al . , 2018 ) . These findings stand in contrast to the perisynaptic processes of hippocampal astrocytes , in which metabotropic glutamate receptor ( mGluR ) activation leads to localized increase in intracellular calcium which promotes process motility and coverage of dendritic spines ( Bernardinelli et al . , 2014 ) . This divergence between Müller glia and hippocampal astrocytes might be reflective of differences in ultrastructural interactions between astroglial processes and synapses in these two systems: perisynaptic processes of hippocampal astrocytes engage in true tripartite synapses in a brain region with high plasticity ( Ventura and Harris , 1999; Haber et al . , 2006 ) , while there is currently no published evidence revealing similar structures in the retina . Further , astrocytic processes express specific actin-binding proteins such as ezrin which enable coupling of cytoskeletal dynamics with signaling through mGluRs ( Lavialle et al . , 2011 ) , which were previously shown to have minimal contribution to retinal wave-associated responses in Müller glia ( Rosa et al . , 2015 ) . Multiple pathways independent of neuronal activity may regulate morphological dynamics in Müller glial processes . One possibility is the release of growth factors from neurons or other nearby cells , which has been implicated in morphological changes among astrocytes in other systems ( Wu et al . , 2017; Liu and Neufeld , 2007 ) . This hypothesis is supported by our finding that application of exogenous EGF enhanced process motility prior to eye opening . Interestingly , EGFR expression in Müller glia is high prior to eye opening and declines with a developmental time course that matches our observed decline in motility ( Close et al . , 2006 ) . Another intriguing possibility is that repulsive homotypic interactions between lateral processes from neighboring Müller glia underly their motility . This idea is supported by single-cell photoablation experiments in zebrafish Müller glia and mouse Schwann cells , in which processes from non-ablated cells filled in the territory vacated by ablated cells ( Williams et al . , 2010; Brill et al . , 2011 ) . Further study will be required to determine whether these or other alternative pathways mediate Müller glial lateral process motility during development . We observed two distinct calcium compartments in Müller glia activated by retinal waves . Compartmentalization of calcium within Müller glial stalks was achieved via activation of glial M1 mAChRs , while lateral processes exhibited non-M1 mAChR-mediated transients with a smaller latency than those in stalks . We observed an age-dependent reduction in stalk participation in waves , while lateral processes continued to respond through P12 . ( Note , our observation that lateral processes continue to respond to retinal waves from P9 to P12 contrasts with a previous study that reported an overall reduction in Müller glial responses to waves at these ages [Rosa et al . , 2015] . We attribute this difference to improved sensitivity of GCaMP6f over GCaMP3 ) . The reduced stalk participation observed in older animals may be due to downregulation of M1 mAChRs prior to eye opening , or to a reduction in volume release of acetylcholine during the transition to glutamatergic waves ( Ford et al . , 2012 ) . Our observation that gabazine application during glutamatergic waves caused wave-associated stalk transients to return in an M1 mAChR-dependent manner suggests that the latter is true . Gabazine application also enhanced wave-associated calcium transients in lateral processes via a mechanism independent of M1 mAChR activation . We suspect this is due to enhanced release of other excitatory neurotransmitters following relief of GABAA receptor-mediated inhibition in the presence of gabazine during retinal waves . This is similar to the action of TBOA , which blocks glutamate transporters and subsequently increases retinal wave frequency and wave-associated Müller glial calcium transients due to an increase in extracellular glutamate levels ( Rosa et al . , 2015 ) . Because astroglia and neurons both express neurotransmitter receptors and transporters , it is interesting to consider which of these cell populations is targeted by the antagonists used in this study . Current RNA-sequencing datasets suggest that Müller glia do not express GABA receptors at an appreciable level and therefore it is unlikely that gabazine directly impacts signaling in these cells . That said , Müller glia express GABA transporters which might contribute to signaling during retinal waves ( Bringmann et al . , 2013; Macosko et al . , 2015 ) . They also express glutamate transporters ( EAAT1 ) which are targeted by TBOA . At 25 μM , the concentration of TBOA used to modulate retinal waves in this study was likely too low to effectively inhibit glial EAAT1 ( IC50: 70 μM ) , and more likely affects retinal wave properties via neuronal EAATs ( Jabaudon et al . , 1999 ) . Furthermore , although we did not test whether pirenzepine acts on neuronal mAChRs , it is possible that blockade of non-Müller glial M1 mAChRs by pirenzepine indirectly affects glial responses to retinal waves , as previous studies have revealed expression of these receptors in retinal interneurons ( Macosko et al . , 2015; Strang et al . , 2010; Strang et al . , 2015 ) . However , our observation that retinal wave-associated EPSC amplitude and IWI remained normal in the presence of pirenzepine is consistent with the interpretation that pirenzepine acts directly on glial M1 mAChRs rather than on wave-generating circuits ( Figures 3E and 4E ) . Subcellular calcium compartmentalization plays a variety of roles in neuron-glia signaling and tissue homeostasis ( Wang et al . , 1997 ) . Hippocampal astrocytic processes undergo at least two types of spatiotemporally distinct calcium transients in response to nearby synaptic activity , depending on the type of synaptic activity that occurs . This calcium signaling is thought to regulate synaptic transmission within the astrocytic territory ( Di Castro et al . , 2011 ) . Similarly , compartmentalized calcium transients during startle response frequently occur in distal branches , and less frequently in somata of hippocampal astrocytes in awake mice . Distinct signaling pathways , including those involving transmembrane calcium channels or transporters , adrenergic receptors , and IP3 receptors , differentially contribute to calcium transients in branches versus somata ( Denizot et al . , 2019 ) . Another study used a computational approach to reveal how calcium influx in fine astrocytic branches can be mediated by glutamate transporter-dependent activity of Na+/Ca++ exchangers , while somatic calcium can be modulated by mGluR-dependent activation of IP3 signaling ( Oschmann et al . , 2017 ) . Compartmentalization between Müller glial stalks and processes may arise by a similar mechanism . We found that wave-associated compartmentalized calcium activity is not required for Müller glial process motility , so this activity likely plays a role in other functions of Müller glia during retinal development . These functions include calcium-dependent neurovascular coupling ( Biesecker et al . , 2016 ) , release of gliotransmitters such as ATP or D-serine ( Newman , 2003; Newman , 2015; Sullivan and Miller , 2010 ) , or secretion of synaptogenic molecules such as thrombospondin or growth factors ( Koh et al . , 2019; de Melo Reis et al . , 2008 ) . Our identification of M1 mAChR as a driver of wave-associated calcium transients in stalks provides a target for selective perturbation in Müller glia to better define a role for these transients in retinal circuit development . All mice were purchased from The Jackson Laboratory and were maintained on mixed C57BL/6 backgrounds . For motility experiments , P8-P116 Slc1a3-CreER;mTmG mice were generated by cross breeding Slc1a3-CreER mice ( strain 012586; also known as GLAST-CreER ) with Rosa26mTmG mice ( strain 007676 ) ( Wang et al . , 2017 ) . Slc1a3-CreER mice express tamoxifen-inducible Cre recombinase under control of a Müller glia-specific promoter . mTmG is a dual-fluorescence reporter line that constitutively expresses membrane-bound tdTomato , and upon Cre-mediated recombination expresses membrane-bound green fluorescent protein ( mGFP ) . For calcium imaging experiments , P9-P12 Slc1a3-CreER;GCaMP6fflox mice were generated by crossing Slc1a3-CreER mice with GCaMP6fflox ( cytosolic ) mice ( strain 024105 ) or Lck-GCaMP6fflox ( membrane-bound ) ( Srinivasan et al . , 2016 ) mice ( strain 029626 ) to enable specific and inducible calcium indicator expression in Müller glia . To verify loss of calcium transients in neurons and Müller glia following BAPTA-AM application , we include imaging results from PDGFRα;cyto-GCaMP6fflox mice ( PDGFRα-Cre: strain 013148 ) , which label Müller glia and a subset of neurons with cytosolic GCaMP6f , as well as results from Slc1a3-CreER;Lck-GCaMP6fflox retinas ( Figure 5—figure supplement 1 ) . For sharp fills of Müller glial cells with fluorescent dye , we used C57BL/6 mice as controls for comparison with mice lacking the β2 subunit of the nicotinic acetylcholine receptor ( β2-nAChR-KO ) . Cre-mediated recombination was induced via intraperitoneal injection of 4-hydroxytamoxifen ( 50:50 E and Z isomers , Sigma-Aldrich ) dissolved in sunflower seed oil . For uniform expression across the retina in neonates , injections of 0 . 5 mg tamoxifen were made 2 and 4 days before each experiment . For sparse expression to enable resolution of individual Müller glia during imaging , a single injection of 0 . 1 mg tamoxifen was made 2 days before each experiment . All animal procedures were approved by the University of California , Berkeley Animal Care and Use Committee and conformed to the NIH Guide for the Care and Use of Laboratory Animals , the Public Health Service Policy , and the SFN Policy on the Use of Animals in Neuroscience Research . For all experiments , male and female mice were anesthetized via isoflurane inhalation and decapitated . Eyes were enucleated and retinas dissected in oxygenated ( 95% O2/5% CO2 ) artificial cerebrospinal fluid ( ACSF ) at room temperature under bright field ( less than P10 ) or infrared ( P10 and above ) illumination . ACSF contained ( in mM ) 119 . 0 NaCl , 26 . 2 NaHCO3 , 11 glucose , 2 . 5 KCl , 1 . 0 K2HPO4 , 2 . 5 CaCl2 , and 1 . 3 MgCl2 . Isolated retinas were mounted ganglion cell side up on filter paper ( Millipore ) and transferred into the recording chamber of an upright microscope for imaging and electrophysiological recording . Retinas were continuously superfused with oxygenated ACSF ( 2–4 ml/min ) at 32–34°C for the duration of experiments and kept in the dark at room temperature in oxygenated ACSF when not imaging or recording . In a subset of experiments , WT retinas were bath-loaded with the organic calcium dye Cal520 ( 12 µM ) for 1 . 5–2 hr prior to performing calcium imaging . Two-photon imaging of Müller glia in the IPL was performed using a modified movable objective microscope ( MOM; Sutter Instruments ) equipped with an Olympus 60× , 1 . 0 NA , LUMPlanFLN objective ( Olympus America ) . Two-photon excitation was evoked with an ultrafast pulsed laser ( Chameleon Ultra II; Coherent ) tuned to 920 nm for all fluorophores . The microscope was controlled by ScanImage software ( http://www . scanimage . org ) . Scan parameters were ( pixels/line×lines/frame [frame rate in Hz] ) : 256×256 ( 1 . 48–2 . 98 ) , at 1–2 ms/line . When imaging GCaMP6f fluorescence in Müller glial stalks and processes , the focal plane was set to ~1/3 the distance from the ganglion cell layer to the inner nuclear layer . Volumetric imaging of motility in mGFP-expressing Müller glial processes and of surrounding tdTomato-expressing cells was achieved by acquiring sequential two-channel Z-stacks through the entire IPL , with slices 1 µm apart and averaging four frames per slice . Volumes were taken every 2 min for a total of 10 min of imaging lateral process dynamics . A similar procedure was used for imaging motility in a subset of cells after filling them with fluorescent dye . Whole-cell voltage-clamp recordings were made from whole-mount retinas while simultaneously imaging GCaMP6f fluorescence . Under infrared illumination , RGC somas were targeted for voltage-clamp recordings using glass microelectrodes with resistance of 3–5 MΩ ( PC-10 pipette puller; Narishige ) filled with an internal solution containing ( in mM ) 110 CsMeSO4 , 2 . 8 NaCl , 20 HEPES , 4 EGTA , 5 TEA-Cl , 4 Mg-ATP , 0 . 3 Na3GTP , 10 Na2Phosphocreatine , and QX-Cl ( pH 7 . 2 and 290 mOsm ) . The liquid junction potential correction for this solution was –10 mV . Signals were acquired using pCLAMP10 recording software and a MultiClamp 700A amplifier ( Molecular Devices ) , sampled at 20 kHz and low-pass filtered at 2 kHz . For pharmacology experiments , after 5–10 min of recording data in ACSF , pharmacological agents were added to the perfusion , and experimental recordings were obtained 5 min afterward . Drug concentrations were as follows: 5 µM cytochalasin-D ( Avantor ) , 10 µM nocodazole ( Sigma-Aldrich ) , 1 unit ( 100 ng ) /ml EGF ( Cytoskeleton , Inc ) , 5 µM pirenzepine ( Tocris ) , 5 µM gabazine ( Tocris ) , 25 µM DL-TBOA ( Tocris ) , 20 µM DNQX ( Tocris ) , 50 µM AP5 ( Tocris ) , and 200 µM BAPTA-AM ( Tocris ) . DL-TBOA and BAPTA-AM were prepared in 0 . 1% DMSO . For calcium chelation experiments , whole-mount retinas were incubated in BAPTA-AM or vehicle ( ACSF/0 . 1% DMSO ) for 1 . 5–2 hr , and then moved to ACSF for 30 min prior to imaging ( Shigetomi et al . , 2008 ) . To verify that BAPTA-AM loading abolished retinal waves and residual calcium activity in cytosolic and membrane-proximal compartments , we measured calcium activity in neurons and Müller glia using cyto-GCaMP6fflox ( crossed with PDGFRα-Cre ) or Lck-GCaMP6fflox ( crossed with Slc1a3-CreER ) at P10–P12 before and after BAPTA-AM loading . For focal agonist application to mGFP-expressing lateral processes in Slc1a3-CreER;mTmG retinas , carbachol ( 10 mM , Tocris ) , or glutamic acid ( 10 mM , Sigma-Aldrich ) was loaded into sharp electrodes pulled on a P-97 Micropipette Puller ( Sutter ) with a resistance of 100–150 MΩ . Electrodes also contained 2 mM Alexa-594 for verification of iontophoresis and to visualize electrodes under two-photon illumination . While imaging at 920 nm and visualizing fluorescence from mGFP and Alexa-594 , electrodes were driven into the IPL at a ~15° angle using a micromanipulator set to the ‘DIAG’ function . For each cell , the tip of the electrode was placed 1–5 µm from the arbor of lateral processes of interest . Iontophoresis of agonist or control was achieved by applying continuous current using pCLAMP10 software . To apply carbachol , which is positively charged at physiological pH , continuous current of +4 nA was used , while –4 nA current was used to apply glutamate , which is negatively charged . Current of opposite polarity to that used for each respective agonist application was applied as a control , and the order of control versus agonist application was shuffled to negate potential effects resulting from damage caused by the electrode . In a separate set of experiments , iontophoresis of agonist was performed in Slc1a3-CreER;cyto-GCaMP6fflox or Slc1a3-CreER;Lck-GCaMP6fflox retinas to verify that this method reliably evoked calcium transients in lateral processes . For a subset of experiments testing the role of retinal wave-evoked calcium transients in Müller glial motility ( Figure 5 ) , and to compare Müller glial morphology in WT and β2-nAChR-KO retinas ( Figure 6 ) , sharp electrodes were pulled as described above , and the tip was bent 15–20° using a microforge . Electrodes were loaded with 2 mM Alexa-488 in water , and iontophoresis of dye into single Müller glial cells was achieved by applying a –10 nA pulse for 500 ms in MultiClamp 700A software while the electrode was on the membrane of Müller glial endfeet in the ganglion cell layer . Electrodes were withdrawn as soon as cells started to fill with dye , and cells were imaged 1–5 min after filling ( Ding et al . , 2016 ) . Two-photon volumetric images of dye-filled lateral processes of single Müller glial cells in the IPL were acquired using the same image parameters used for imaging motility , as described above . All images were processed using custom scripts in FIJI/ImageJ ( National Institutes of Health ) ( Schindelin et al . , 2012 ) and MATLAB ( MathWorks ) . For calcium imaging movies , following non-rigid motion correction ( Pnevmatikakis and Giovannucci , 2017 ) , ROIs were semi-automatically placed over stalks using a filtering algorithm based on a Laplace operator and segmented by applying a user-defined threshold ( Dorostkar et al . , 2010 ) . The Distance Transform Watershed function ( Legland et al . , 2016 ) in FIJI was used to segregate nearby stalks in binarized images . This method defined most of the ROIs that an experienced user would recognize by eye . Manual adjustments were made to include stalks that were missed and to remove ROIs that were erroneously placed on structures other than stalks , which were defined as semiregularly spaced , punctate regions of fluorescence 2–3 µm in diameter in average intensity projection images . Lateral process ROIs were subsequently defined by randomly placing 250 2 . 5×2 . 5 µm2 squares in regions that did not overlap with stalk ROIs . Fluorescence intensity is reported as the average intensity across all pixels within the area of each ROI , and normalized as the relative change in fluorescence ( ΔF/F ) as follows:ΔFF= ( F−F0 ) /F0 , where F is the instantaneous fluorescence at any time point and F0 is the baseline fluorescence , defined as the median fluorescence value over the duration of the trace . ΔF/F traces were smoothed using a two-frame median filter and analyzed to detect calcium transients using custom MATLAB code . Traces were Z scored , and a threshold Z score of 3 was used to define calcium transients , which were then further defined as wave-associated if they occurred within 3 s of the peak of a wave-associated EPSC . Intercompartment latency was defined as the difference , in seconds , between wave-associated transients in stalks and the median wave-associated transient time for all lateral processes within the FOV , for each retinal wave . For volumetric images of Müller glial process morphology , images were bandpass filtered in XY space to reduce noise while maintaining the structure of fine processes , and registered using the Correct 3D Drift plugin in FIJI ( Parslow et al . , 2014 ) . 10-minute volumetric time series were corrected for photobleaching using the ‘Histogram matching’ setting within FIJI’s Bleach Correction plugin . Time series were collapsed into temporally color-coded images to facilitate identification of motile and stable processes in 3D . The Cell Counter tool within FIJI was used to count lateral processes and define their morphological dynamics as one of the following: extending , retracting , new process , lost process , extension followed by retraction , retraction followed by extension , and stable . Locations of process tips within the IPL were recorded and binned into one of five equally sized sections , corresponding to putative sublayers S1 through S5 , defined by their distance from the INL and GCL borders as identified using membrane-tdTomato fluorescence on somas . Volumetric images of sharp-filled Müller glia were traced using the Simple Neurite Tracer plugin ( FIJI ) ( Arshadi et al . , 2021 ) . We traced each glial stalk and subsequently any visible processes branching from the stalk , creating distinct paths for each process and preserving branch order relationships . Measurements including number of tips , total process length , total branch points , and primary branches from stalk were derived from traced paths . For Sholl and convex hull analyses , paths were converted to binary skeletons and registered to correct for XY displacement of the stalk between the GCL and INL . Sholl analysis was performed using concentric rings spaced 1 µm apart ( Ferreira et al . , 2014 ) . For the XY plane , the center of each Sholl radius was placed on the registered stalk , and intersections were counted at each radius overlayed on a maximum projection image . For the XZ and YZ planes , the center of each Sholl radius was placed on the end of the stalk closest to the INL in orthogonal projections of traced cells . Sholl radii were normalized to the total IPL thickness . Convex hull area was defined as the area of the smallest convex polygon enclosing the entire skeleton in an XY maximum projection image . Group measurements are expressed as mean ± standard error of mean ( SEM ) except when nonparametric test results are reported , in which case median , 1st , and 3rd quartiles are reported . MATLAB and RStudio were used to carry out all statistical tests . Chi-squared tests of goodness-of-fit were used to test for nonuniformity in lateral process distribution across the IPL within each age . Chi-squared tests for independent proportions were used to test for age-dependent differences in process distribution across the IPL ( Supplementary files 1 and 2 ) . When comparing the proportion of total stable processes between control and experimental manipulations , we applied Wilcoxon signed-rank tests when cells were paired between conditions , and Wilcoxon rank-sum tests for unpaired cells . When there were less than five cells in a particular condition , we pooled process counts between cells and reported chi-square test statistics for these comparisons as well ( Supplementary files 3–5 and 20 ) . To test for age- , compartment- , and drug-dependent differences in retinal wave-associated calcium transients ( Supplementary files 6; 8–11; 14–17 ) , we applied two- or three-way mixed ANOVAs followed by post hoc t-tests with Benjamini-Hochberg correction for multiple comparisons , when appropriate . Paired t-tests were used to compare fold-changes between compartments in response to pirenzepine and gabazine ( Supplementary files 9; 11; 15 and 17 ) . Intercompartment latency of wave-associated calcium transients was compared between ages and conditions using Wilcoxon rank-sum tests , and within each condition tested for significant difference from zero using Wilcoxon signed-rank tests ( Supplementary files 7; 12 and 18 ) . Wave EPSC amplitude and IWI were compared between control and drug conditions using paired t-tests ( Supplementary files 13 and 19 ) . Sholl intersection profiles between WT and β2-nAChR-KO Müller glia were compared using two-way repeated-measures mixed ANOVA to test for genotype-associated differences in complexity across Sholl radii ( Figure 6 ) . Two-sample t-tests were used for comparison of other morphological measurements between WT and β2-nAChR-KO Müller glia ( Supplementary file 21 ) .	When it comes to studying the nervous system , neurons often steal the limelight; yet , they can only work properly thanks to an ensemble cast of cell types whose roles are only just emerging . For example , ‘glial cells’ – their name derives from the Greek word for glue – were once thought to play only a passive , supporting function in nervous tissues . Now , growing evidence shows that they are , in fact , integrated into neural circuits: their activity is influenced by neurons , and , in turn , they help neurons to function properly . The role of glial cells is becoming clear in the retina , the thin , light-sensitive layer that lines the back of the eye and relays visual information to the brain . There , beautifully intricate Müller glial cells display fine protrusions ( or ‘processes' ) that intermingle with synapses , the busy space between neurons where chemical messengers are exchanged . These messengers can act on Müller cells , triggering cascades of molecular events that may influence the structure and function of glia . This is of particular interest during development: as Müller cells mature , they are exposed to chemicals released by more fully formed retinal neurons . Tworig et al . explored how neuronal messengers can influence the way Müller cells grow their processes . To do so , they tracked mouse retinal glial cells ‘live’ during development , showing that they were growing fine , highly dynamic processes in a region rich in synapses just as neurons and glia increased their communication . However , using drugs to disrupt this messaging for a short period did not seem to impact how the processes grew . Extending the blockade over a longer timeframe also did not change the way Müller cells developed , with the cells still acquiring their characteristic elaborate process networks . Taken together , these results suggest that the structural maturation of Müller glial cells is not impacted by neuronal signaling , giving a more refined understanding of how glia form in the retina and potentially in the brain .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "neuroscience" ]	2021	Excitatory neurotransmission activates compartmentalized calcium transients in Müller glia without affecting lateral process motility
Pairwise models are commonly used to describe many-species communities . In these models , an individual receives additive fitness effects from pairwise interactions with each species in the community ( 'additivity assumption' ) . All pairwise interactions are typically represented by a single equation where parameters reflect signs and strengths of fitness effects ( 'universality assumption' ) . Here , we show that a single equation fails to qualitatively capture diverse pairwise microbial interactions . We build mechanistic reference models for two microbial species engaging in commonly-found chemical-mediated interactions , and attempt to derive pairwise models . Different equations are appropriate depending on whether a mediator is consumable or reusable , whether an interaction is mediated by one or more mediators , and sometimes even on quantitative details of the community ( e . g . relative fitness of the two species , initial conditions ) . Our results , combined with potential violation of the additivity assumption in many-species communities , suggest that pairwise modeling will often fail to predict microbial dynamics . Multispecies microbial communities are ubiquitous . Microbial communities are important for industrial applications such as cheese and wine fermentation ( van Hijum et al . , 2013 ) and municipal waste treatment ( Seghezzo et al . , 1998 ) . Microbial communities are also important for human health: they can modulate immune responses and food digestion ( Round and Mazmanian , 2009; Kau et al . , 2011 ) during health and disease . Properties of the entire community ( ‘community properties’ , e . g . species dynamics , ability to survive internal or external perturbations , and biochemical activities of the entire community ) are influenced by interactions wherein individuals alter the physiology of other individuals ( Widder et al . , 2016 ) . To understand and predict community properties , choosing the appropriate mathematical model to describe species interactions is critical . A mathematical model ideally focuses only on details that are essential to community properties of interest . However , it is often unclear a priori what the minimal essential details are . We define ‘mechanistic models’ as models that explicitly consider interaction mediators as state variables . For example , if species S1 releases a compound C1 which stimulates species S2 growth upon consumption by S2 , then a mechanistic model tracks concentrations of S1 , C1 , and S2 ( Figure 1A and B , left panels ) . Note that mechanistic models used here still omit molecular details such as how chemical mediators are received and processed by recipients and how mediators subsequently act on recipients . In contrast , Lotka-Volterra ( ‘L-V’ ) pairwise models only consider the fitness effects of interactions . Specifically , L-V models assume that the fitness of an individual is the sum of its basal fitness ( the net growth rate of an individual in isolation ) and fitness influences from pairwise interactions with individuals of the same species and of every other species in the community ( ‘additivity’ assumption ) . Furthermore , regardless of interaction mechanisms or quantitative details of a community , all fitness influences are typically expressed using a single equation form wherein parameters can vary to reflect the signs and magnitudes of fitness influences ( ‘universality’ assumption ) . Thus in the example above , a pairwise model only describes how S1 increases the fitness of S2 ( Figure 1A and B , right panels ) . 10 . 7554/eLife . 25051 . 003Figure 1 . The abstraction of interaction mechanisms in a pairwise model compared to a mechanistic model . ( A ) The mechanistic model ( left ) considers a bipartite network of species and chemical interaction mediators . A species can produce or consume chemicals ( open arrowheads pointing towards and away from the chemical , respectively ) . A chemical mediator can positively or negatively influence the fitness of its target species ( filled arrowhead and bar , respectively ) . The corresponding L-V pairwise model ( right ) includes only the fitness effects of species interactions , which can be positive ( filled arrowhead ) , negative ( bar ) , or zero ( line terminus ) . ( B ) In the example here , species S1 releases chemical C1 , and C1 is consumed by species S2 and promotes S2’s fitness . In the mechanistic model , the three equations respectively state that ( 1 ) S1 grows exponentially at a rate r10 , ( 2 ) C1 is released by S1 at a rate βC1S1 and consumed by S2 with saturable kinetics ( maximal consumption rate αC1S2 and half-saturation constant KC1S2 ) , and ( 3 ) S2’s growth ( basal fitness r20 ) is influenced by C1 in a saturable fashion . In the pairwise model here , the first equation is identical to that of the mechanistic model . The second equation is similar to the last equation of the mechanistic model except that r21 and K21 together reflect how the density of S1 ( S1 ) affects the fitness of S2 in a saturable fashion . For all parameters with double subscripts , the first subscript denotes the focal species or chemical , and the second subscript denotes the influencer . Note that unlike in mechanistic models , we have omitted ‘S’ from subscripts in pairwise models ( e . g . r21 instead of rS2S1 ) for simplicity . In this example , both r21 and rS2S1 are positive . DOI: http://dx . doi . org/10 . 7554/eLife . 25051 . 00310 . 7554/eLife . 25051 . 004Figure 1—figure supplement 1 . An L-V pairwise model successfully predicts oscillations in population dynamics of the hare-lynx prey-predator community . ( A ) In a pairwise model of prey-predation proposed by Lotka and Volterra , predator reduces the fitness of prey , while prey stimulates the fitness of predator . Such dynamics can be easily simulated ( Green and Shou , 2014 ) . ( B ) Assuming random encounter between prey and predator , the pairwise model predicts oscillations in the prey and predator population sizes . ( C ) Similar oscillations have been qualitatively observed in natural populations of lynx ( predator ) and hare ( prey ) , providing support for the usefulness of pairwise models . Picture is adapted from https://biologyeoc . wikispaces . com/PopulationChanges ( BiologyEOC , 2016 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 25051 . 00410 . 7554/eLife . 25051 . 005Figure 1—figure supplement 2 . Deriving a pairwise model . ( A ) Analytically deriving a pairwise model from a mechanistic model allows us to uncover approximations required for such a transformation ( top ) . Alternatively ( bottom ) , within a ‘training window’ of the mechanistic model population dynamics , we can numerically derive parameters for a pre-selected pairwise model such that it best fits the mechanistic model . We then quantify how well such a pairwise model matches the mechanistic model under conditions different from those of the training window . ( B ) A mechanistic model of three species interacting via two chemicals ( left ) can be translated into a pairwise model of three interacting species ( center ) . S1 inhibits S1 and promotes S2 ( via C1 ) . S2 promotes S2 and S3 ( via C2 ) as well as S1 ( via removal of C1 ) . S3 promotes S1 ( via removal of C1 ) and inhibits S2 ( via removal of C2 ) . Take interactions between S2 and S3 for example: the saturable L-V pairwise model will require estimating ten parameters ( colored , right ) , some of which ( e . g . r33 in this case ) may be zero . ( C ) In the numerical method , the six monoculture parameters ( ri0 , rii , and Kii , i = 2 , 3; green and red ) are first estimated from training window T ( within a dilution cycle ) of monoculture mechanistic models ( top and middle ) . Subsequently , the four interaction parameters ( and Kij , , olive ) can be estimated from the training window T of the S2 +S3 coculture mechanistic model ( bottom ) . Parameter definitions are described in Figure 1 . Often in this work , pairwise model parameters that can be directly obtained from the mechanistic model ( e . g . species basal fitness; ) are directly obtained from the mechanistic model ( instead of being estimated ) . To estimate parameters , we use an optimization routine to minimize D¯ , the fold-difference ( hatched area ) between dynamics from a pairwise model ( dotted lines ) and the mechanistic model ( solid lines ) averaged over T and species number N: D¯=1N∑i=1N[1T∫TDi ( t ) dt]=1N∑i=1N[1T∫T\|log10 ( Si , pair ( t ) /Si , mech ( t ) ) \|dt] . Here Si , pair and Si , mech are Si calculated using pairwise and mechanistic models , respectively . Since species with densities below a set extinction limit , Sext , are assumed to have gone extinct in the model , we set all densities below the extinction limit to Sext in calculating D¯ to avoid singularities . D¯ outside the training window can be used to quantify how well the best-matching pairwise model predicts the mechanistic model . Unless otherwise stated , in all simulations to ensure that resources not involved in interactions are never limiting , a community is diluted back to its inoculation density whenever total population increases to a high-density threshold , mimicking turbidostat experiments . Too frequent dilutions will allow only small changes in population dynamics within a dilution cycle or time T , which is not suitable for estimating pairwise model parameters . Dilutions can sometimes violate conditions for convergence to reference dynamics ( Figure 3—figure supplement 4 ) . Under most cases we have tested , small variations in dilution frequency do not affect our conclusions . See Methods-Summary of simulation files for relevant Matlab codes . DOI: http://dx . doi . org/10 . 7554/eLife . 25051 . 005 L-V pairwise models are popular . L-V pairwise modeling has successfully explained the oscillatory dynamics of hare and its predator lynx ( Figure 1—figure supplement 1 ) ( Volterra , 1926; Wangersky , 1978; BiologyEOC , 2016 ) . Pairwise models have also been instrumental in delineating conditions for multiple carnivores to coexist when competing for herbivores ( MacArthur , 1970; Chesson , 1990 ) . In both cases , mechanistic models and pairwise models happen to be mathematically equivalent for the following reasons . In the hare-lynx example , both species are also interaction mediators , and therefore pairwise and mechanistic models are identical . In the second example , if herbivores ( mediators of competitive interactions between carnivores ) rapidly reach steady state , herbivores can be mathematically eliminated from the mechanistic model to yield a pairwise model of competing carnivores ( MacArthur , 1970; Chesson , 1990 ) . Pairwise models are often used to predict how perturbations to steady-state species composition exacerbate or decline over time ( May , 1972; Thébault and Fontaine , 2010; Mougi and Kondoh , 2012; Allesina and Tang , 2012; Suweis et al . , 2013; Coyte et al . , 2015 ) . Although most work are motivated by contact-dependent prey-predation ( e . g . hare-lynx ) or mutualisms ( e . g . plant-pollinator ) where L-V models could be identical to mechanistic models , these work do not explicitly exclude chemical-mediated interactions where species are distinct from interaction mediators . The temptation of using pairwise models is indeed high , including in microbial communities where many interactions are mediated by chemicals ( Mounier et al . , 2008; Faust and Raes , 2012; Stein et al . , 2013; Marino et al . , 2014; Coyte et al . , 2015 ) . Even though pairwise models do not capture the dynamics of chemical mediators , predicting species dynamics is still highly desirable in , for example , forecasting species diversity and compositional stability . For chemical-mediated interactions , L-V pairwise models are far easier to construct than mechanistic models for the following reasons . Mechanistic models would require knowledge of chemical mediators , which are often challenging to identify . Since chemical mediators are explicitly modeled , mechanistic models require more equations and parameters than their cognate pairwise models ( Figure 1 , Table ) . Pairwise model parameters are relatively easy to estimate using community dynamics or dynamics of monocultures and pairwise cocultures ( Mounier et al . , 2008; Stein et al . , 2013; Guo and Boedicker , 2016 ) . Consequently , pairwise modeling has been liberally applied to microbial communities . L-V pairwise models have been criticized when applied to communities of more than two species ( referred to as ‘multispecies communities’ ) ( Levine , 1976; Tilman , 1987; Wootton , 1993 , 2002; Werner and Peacor , 2003; Stanton , 2003; Schmitz et al . , 2004 ) . This is because a third species can influence interactions between a species pair ( ‘indirect interactions’ ) , which sometimes violates the additivity assumption of pairwise models . For example , a carnivore can indirectly increase the density of a plant by decreasing the density of an herbivore ( ‘interaction chain’; ‘density-mediated indirect interactions’ ) . A carnivore can also decrease how often an herbivore forages plants ( ‘interaction modification’ , ‘trait-mediated indirect interactions’ , or ‘higher order interactions’ ) ( Vandermeer , 1969; Wootton , 1994; Billick and Case , 1994; Wootton , 2002 ) . In interaction modification , foraging per herbivore decreases , whereas in interaction chain , the density of herbivores decreases . Interaction modification ( but not interaction chain ) violates the additivity assumption ( Methods-Interaction modification but not interaction chain violates the additivity assumption ) ( Tilman , 1987; Wootton , 1994; Schmitz et al . , 2004 ) and can cause the pairwise model to generate qualitatively wrong predictions . Indeed , pairwise models largely failed to predict biomass and species coexistence in three-species and seven-species plant communities ( Dormann and Roxburgh , 2005 ) , although reported failures of pairwise models could be due to limitations in data collection and analysis ( Case and Bender , 1981; Billick and Case , 1994 ) . Here , we examine the universality assumption of pairwise models when applied to microbial communities ( or any community that employs diverse chemical-mediated interactions ) . Microbes often influence other microbes in a myriad of fashions , via consumable metabolites , reusable signaling molecules , or a combination of chemicals ( Figure 2 ) . Can a single equation form , traditionally employed in pairwise models , qualitatively describe diverse interactions between two microbial species ? The answer is unclear . On the one hand , pairwise models have been applied successfully to diverse microbial communities . For example , an L-V pairwise model and a mechanistic model both correctly predicted ratio stabilization and spatial intermixing between two strongly-cooperating populations exchanging diffusible essential metabolites ( Momeni et al . , 2013 ) . In other examples , pairwise models largely captured competition outcomes and metabolic activities of three-species and four-species artificial microbial communities ( Vandermeer , 1969; Guo and Boedicker , 2016; Friedman et al . , 2017 ) . On the other hand , pairwise models often failed to predict species coexistence in seven-species microbial communities ( Friedman et al . , 2017 ) , although this could be due to interaction modification discussed above . 10 . 7554/eLife . 25051 . 006Figure 2 . Chemical-mediated interactions commonly found in microbial communities . Interactions can be intra- or inter-population . Examples are meant to be illustrative instead of comprehensive . DOI: http://dx . doi . org/10 . 7554/eLife . 25051 . 006 Instead of investigating natural communities where interaction mechanisms can be difficult to identify , we use in silico communities . In these communities , two species interact via mechanisms commonly encountered in microbial communities , including growth-promoting and growth-inhibiting interactions mediated by reusable and consumable compounds ( Figure 2 ) ( Stams , 1994; Czárán et al . , 2002; Duan et al . , 2009 ) . We construct mechanistic models for these two-species communities and attempt to derive from them pairwise models . A mechanistic reference model offers several advantages: community dynamics is deterministically known; deriving a pairwise model is not limited by inaccuracy of experimental methods; and the flexibility in creating different reference models allows us to explore a variety of interaction mechanisms . We demonstrate that a single pairwise equation form often fails for commonly-encountered diverse pairwise microbial interactions . We conclude by discussing when pairwise models might or might not be useful , in light of our findings . A mechanistic model describes how species release or consume chemicals and how chemicals stimulate or inhibit species growth ( Figure 1A left ) . In contrast , in pairwise models , interation mediators are not explicitly considered ( Figure 1A right ) . Instead , the growth rate of an individual of species Si is the sum of its basal fitness ( ri0 , net growth rate of the individual in the absence of any intra-species or inter-species interactions ) and fitness effects from intra-species and inter-species interactions . The fitness effect from species Sj to species Si is represented by fij ( Sj ) , where Sj is the density of species Sj . fij ( Sj ) is a linear or nonlinear function of only Sj and not of another species . When j=i , fii ( Si ) represents density-dependent fitness effect within Si ( e . g . density-dependent growth inhibition or stimulation ) . In a multi-species pairwise model , a single form of fij is used for all pairwise species interactions . For example , the most popular L-V model is linear L-V: ( 1 ) dSidt=[ ri0+∑jrijSj ]Si Here , ri0 is the basal fitness of an individual of Si , and can be positive , negative , or zero; rij is the fitness effect per Sj individual on Si . Positive , negative , or zero rij represents growth stimulation , inhibition , or no effect , respectively . An example of linear L-V is the logistic L-V pairwise model traditionally used for competitive communities: ( 2 ) dSidt=ri0[ 1−∑jSjΛij ]Si Here , nonnegative ri0 is the basal fitness of Si; positive Λij is the carrying capacity imposed by limiting shared resource ( e . g . space or carbon source ) such that a single Si individual will have a zero net growth rate when competing with a total of Λij individuals of Sj . Alternative forms of fitness effect fij ( Wangersky , 1978 ) include L-V with delayed influence , where the fitness influence of one species on another may lag in time ( Gopalsamy , 1992 ) , or saturable L-V ( Thébault and Fontaine , 2010 ) where ( 3 ) dSidt=[ ri0+∑jrijSjKij+Sj ]Si Here , ri0 is the basal fitness of an individual of Si , rij is the maximal fitness effect species Sj can exert on Si , and Kij ( >0 ) is the Sj at which half maximal fitness effect on Si is achieved . ri0 and rij can be positive , negative , or zero . Note that at a low concentration of influencer , the saturable form can be converted to a linear form . Our goal is to test whether a single equation form of pairwise model can qualitatively predict dynamics of species pairs engaging in various types of interactions commonly found in microbial communities ( e . g . Figure 2 ) . To do so , we use a combination of analytical and numerical approaches ( Figure 1—figure supplement 2 ) . Analytically deriving a pairwise model from a mechanistic model not only reveals assumptions required to generate the pairwise model , but also alleviates any concern that we may have failed to identify the optimal pairwise model parameters . When interactions become more complex ( e . g . involving multiple mediators ) , we take the numerical approach , which is typically used to infer pairwise models from experimental results ( Pascual and Kareiva , 1996 ) . In the numerical approach , we mimic experimentalists by first deciding on a pairwise model to be used , and then employing a nonlinear least squares routine to numerically identify model parameters that minimize the average difference D¯ between pairwise and mechanistic model dynamics within a training time window T ( Figure 1—figure supplement 2; Methods-Summary of simulation files ) . To evaluate how well a pairwise model predicts long-term mechanistic model dynamics , we ‘buy time’ by introducing 'dilutions' in numerical simulations of both models and quantify their difference D¯ . In this section , we analytically derive pairwise models from mechanistic models of two-species communities where one species affects the other species through a single mediator . The mediator is either reusable such as signaling molecules in quorum sensing ( Duan et al . , 2009; Jakubovics , 2010 ) or consumable such as metabolites ( Stams , 1994; Freilich et al . , 2011 ) ( Figure 2 ) . We show that a single pairwise model may not encompass these different interaction mechanisms and that for consumable mediator , the choice of pairwise model also depends on details such as the relative fitness and initial densities of the two species . Consider a commensal community where species S1 stimulates the growth of species S2 by producing a reusable ( Figure 3A ) or a consumable ( Figure 3B ) chemical C1 . We consider community dynamics where species are not limited by any abiotic resources , such as within a dilution cycle of a turbidostat experiment where all other metabolites are in excess . 10 . 7554/eLife . 25051 . 007Figure 3 . Interactions mediated via a single mediator are best represented by different forms of pairwise models , depending on whether the mediator is consumable or reusable and on the relative fitness and initial densities of the two species . S1 stimulates the growth of S2 via a reusable ( A ) or a consumable ( B ) chemical C1 . In mechanistic models of the two cases ( i ) , equations for S1 and S2 are identical but equations for C1 are different . In ( A ) , C1 can be solved to yield C1= ( βC1S1/r10 ) S10exp⁡ ( r10t ) − ( βC1S1/r10 ) S10= ( βC1S1/r10 ) S1− ( βC1S1/r10 ) S10 , assuming zero initial C1 . Here , S10 is S1 at time zero . We have approximated C1 by omitting the second term ( valid after the initial transient response has passed so that C1 has become proportional to S1 ) . This approximation allows an exact match between the mechanistic model and the saturable L-V pairwise model ( ii ) . In ( B ) , depending on the relative growth rates of the two species , and if additional requirements are satisfied ( Methods; Figure 3—figure supplement 2; Figure 3—figure supplement 3; Figure 3—figure supplement 4; Figure 3—figure supplement 5 ) , either saturable L-V or alternative pairwise model should be used . DOI: http://dx . doi . org/10 . 7554/eLife . 25051 . 00710 . 7554/eLife . 25051 . 008Figure 3—source data 1 . List of parameters for simulations in Figure 3—figure supplement 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 25051 . 00810 . 7554/eLife . 25051 . 009Figure 3—source data 2 . List of parameters for simulations in Figure 3—figure supplement 2 on interactions through a consumable mediator . DOI: http://dx . doi . org/10 . 7554/eLife . 25051 . 00910 . 7554/eLife . 25051 . 010Figure 3—source data 3 . List of parameters for simulations in Figure 3—figure supplement 3 on conditions required for convergence of the alternative pairwise model . DOI: http://dx . doi . org/10 . 7554/eLife . 25051 . 01010 . 7554/eLife . 25051 . 011Figure 3—source data 4 . List of parameters for simulations in Figure 3—figure supplement 4 on how dilution might affect the convergence of a pairwise model . DOI: http://dx . doi . org/10 . 7554/eLife . 25051 . 01110 . 7554/eLife . 25051 . 012Figure 3—figure supplement 1 . For a reusable mediator , parameter estimation after acclimation time leads to a more accurate saturable L-V pairwise model . ( A ) We use the mechanistic model for a reusable mediator to generate reference dynamics of , , and over 150 generations of community growth . Note that population fractions ( instead of population densities ) are plotted , which fluctuate less than the mediator concentration during dilutions . After an initial period of time , becomes proportional to ( inset ) . The basal fitness of S1 and S2 in pairwise models are identical to those in mechanistic models , and here and ( = 1 , 2 ) are irrelevant due to the lack of intra-population interactions . We use every 10 community doublings ( within a dilution cycle ) of reference dynamics as training windows to numerically estimate best-matching saturable L-V pairwise model parameters and . Dashed and solid rectangles represent a training window before and after acclimation , respectively . ( B ) Pairwise model parameters estimated after acclimation ( e . g . solid rectangle ) match their analytically-derived counterparts ( black dotted lines ) better than those estimated before acclimation ( e . g . dashed rectangle ) . ( C ) A pairwise model generated from population dynamics before acclimation ( top ) predicts future reference dynamics less accurately than that generated after acclimation ( bottom ) . ( D ) Quantification of the difference between pairwise and mechanistic models before ( dashed ) or after ( solid ) acclimation . All parameters are listed in Figure 3—source data 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 25051 . 01210 . 7554/eLife . 25051 . 013Figure 3—figure supplement 2 . Community trajectory approaching the -zero-isocline allows us to use the alternative pairwise model approximation . S1 releases a consumable metabolite C1 which stimulates S2 growth . In all panels , brown circles indicate the C1 and RS ( =S2/S1 ) of a community , and are separated by 1/4 of community doubling time . In the vicinity of the f-zero-isocline ( f=0 ) ( blue line ) , can be eliminated to yield a pairwise model . ( A–D ) When rS2C1>r10−r20>0 ( Methods-Deriving a pairwise model for interactions mediated by a single consumable mediator , Case II ) , a steady state ( green circle ) exists . Let us scale and against their respective steady state values to obtain R^S and C^1 . The f-zero-isocline ( blue ) and the steady state C^1=1 ( vertical solid line ) divide the phase portrait into four regions ( ① to ④ ) . ( A ) The directions of movement are marked by grey arrowheads . According to the top portion of Equation 11 , the right-hand side of Equation 14 is zero when C^1=1 . Since C^1/ ( C^1+K^S2C1 ) =1/ ( 1+K^S2C1/C^1 ) is an increasing function of C^1 , when C^1>1 , dR^S/dt > 0 ( up arrows ) , and when C^1<1 , dR^S/dt<0 ( down arrows ) . From Equation 15 , above the f-zero-isocline , dC^1/dt<0 ( left arrows ) , while below the f-zero-isocline , dC^1/dt>0 ( right arrows ) . Thus , the community moves toward the f-zero-isocline , and then moves slowly alongside ( but not superimposing ) the f-zero-isocline before reaching the steady state . A–C respectively describe community dynamics trajectories from when S10 is large and when R^S ( t=0 ) ≈1 ( Case II-2 ) , R^S ( t=0 ) ≫max ( 1 , K^S2C1−1 ) ( Case II-1 ) , or R^S ( t=0 ) ≪1/ ( 1+K^C1S2 ) ( Case II-3 ) . ( D ) R^S ( t=0 ) ≈1 but S10 is much smaller than that in A . In this case , instead of approaching the f-zero-isocline quickly as in A , the trajectory plunges sharply before moving toward the f-zero-isocline . The black dotted line marks R^S=1/ ( 1+K^C1S2 ) , the asymptotic value of f-zero-isocline . ( E–H ) When r10−r20<0 ( Methods-Deriving a pairwise model for interactions mediated by a single consumable mediator , Case III ) , there is no steady state . RS approaches infinity and C1 approaches 0 . The black dotted line marks RS=βC1S1/αC1S2 , the asymptotic value of f-zero-isocline . E , F , and G respectively describe community dynamics trajectories from C1=0 when S10 is large and when RS ( 0 ) ≈βC1S1/αC1S2 , RS ( 0 ) ≫βC1S1/αC1S2 ( Case III-1 ) , and RS ( 0 ) ≪βC1S1/αC1S2 ( Case III-2 ) . ( H ) RS ( 0 ) ≫βC1S1/αC1S2 but S10 is much smaller than that in F . Note different axis scales in different figure panels . All parameters are listed in Figure 3—source data 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 25051 . 01310 . 7554/eLife . 25051 . 014Figure 3—figure supplement 3 . Condition for the alternative pairwise model to converge to the mechanistic model in the absence of dilutions . Here are the phase portraits of Equation 43 . The olive vertical dotted lines correspond to RS=−ω/ψ , a singularity point when ω<0 . ( A ) Case II ( rS2C1>r10−r20>0 ) , ω=0 . 5 , ψ=0 . 25 . Regardless of initial , the solution converges to steady state ( in agreement with the mechanistic model ) . ( B ) Case II , ω=−1 , ψ=1 . When RS ( t=0 ) <−ω/ψ ( to the left of olive line ) , the alternative pairwise model falsely predicts extinction of S2 . ( C ) Case III ( r10<r20 ) , ω=0 . 8 , ψ=0 . 1 . Regardless of initial , the model predicts extinction of S1 ( in agreement with the mechanistic model ) . ( D ) Case III ( r10<r20 ) , ω=−9 , ψ=5 . When RS ( t=0 ) <−ω/ψ ( to the left of olive line ) , the alternative pairwise model falsely predicts steady state coexistence of the two species . All parameters are listed in Figure 3—source data 3 . DOI: http://dx . doi . org/10 . 7554/eLife . 25051 . 01410 . 7554/eLife . 25051 . 015Figure 3—figure supplement 4 . Initial conditions that require long convergence time and thus dilutions may prevent the alternative pairwise model to converge to the mechanistic model . We consider commensalism through a consumable mediator , where the producer and the consumer could reach a steady state ( Methods-Deriving a pairwise model for interactions mediated by a single consumable mediator , Case II ) . We choose a low consumption rate such that starting from equal proportions of producers and consumers , consumption can be neglected . ( A ) When the initial total population density is low ( 2 × 105 total cells/ml , and periodic 10x dilution maintains this low density ) , the community cannot approach the blue -zero-isocline in a reasonable time frame ( brown trajectory in the phase space of C^1 and R^S , the mediator concentration and the consumer-to-producer ratio normalized to their potential steady state values , respectively ) . As a result , growth and dilution lead to an alternate sustained cycle for the community ( inset ) . ( B ) In this case , accumulates proportionally to within each dilution cycle , and the ratio of to reaches a constant value ( inset ) . ( C ) Since behaves as a reusable mediator and since the community remains far from the -zero-isocline , the use of the alternative pairwise model ( dashed ) is not justified . Instead , the saturable L-V pairwise model ( dotted ) provides a better approximation . ( D ) The same community at a higher initial total density ( 2 × 108 total cells/ml ) approaches the blue -zero-isocline ( brown trajectory in the phase space ) after a few dilutions . ( E ) In the vicinity of the -zero-isocline , reaches its steady state value within each dilution cycle . ( F ) In this case , the alternative model produces a better approximation to the mechanistic model compared to a saturable L-V model . In ( C ) and ( F ) , the saturable L-V model is fitted into dynamics of the reference model after 50 generations , whereas analytical formulas are used for the alternative pairwise model . All parameters are listed in Figure 3—source data 4 . DOI: http://dx . doi . org/10 . 7554/eLife . 25051 . 01510 . 7554/eLife . 25051 . 016Figure 3—figure supplement 5 . Additional requirements for deriving a pairwise model from a mechanistic model , when S1 affects S2 via a single consumable mediator C1 where C1 ( 0 ) =0 . For details , see Methods . Here , RS=S2/S1 . S1 ( 0 ) , S2 ( 0 ) , C1 ( 0 ) , and RS ( 0 ) are the initial values of the respective variables . ( A ) The initial condition requirement for a pairwise model to converge to the mechanistic model . ( B ) The time scale required for convergence . Conditions on S1 ( 0 ) are sufficient , but may not be necessary . DOI: http://dx . doi . org/10 . 7554/eLife . 25051 . 016 When C1 is reusable , the mechanistic model ( Figure 3A , i ) can be transformed into a saturable L-V pairwise model ( compare Figure 3A , ii with Equation 3 ) , especially after the concentration of the mediator ( which is initially zero ) has acclimated to be proportional to the producer population size ( Figure 3A legend; Figure 3—figure supplement 1 ) . This saturable L-V pairwise model is valid regardless of whether the producer coexists with the consumer , outcompetes the consumer , or is outcompeted by the consumer . If C1 is consumable , different scenarios are possible ( Figure 3B; Methods ) . Case I: When supplier S1 always grows faster than consumer S2 ( the basal fitness of S1 is higher than the maximal fitness of S2 ) , C1 will eventually accumulate proportionally to S1 ( Figure 4A left; Methods-Deriving a pairwise model for interactions mediated by a single consumable mediator Case I ) . In this case , C1 may be approximated as a reusable mediator and can be predicted by the saturable L-V pairwise model ( Figure 4A right , compare dotted and solid lines ) . 10 . 7554/eLife . 25051 . 017Figure 4 . Saturable L-V and alternative pairwise models are not interchangeable . Consider a commensal community with a consumable mediator C1 . ( A ) The mediator accumulates without reaching a steady state within each dilution cycle as the consumer S2 gradually goes extinct ( Figure 3B , Case I ) . After a few tens of generations , C1 becomes proportional to its producer density S1 ( inset in left panel ) . In this case , a saturable L-V ( dotted ) but not the alternative pairwise model ( dashed ) is suitable . All parameters are listed in Figure 4—source data 1 . ( B ) The consumable mediator reaches a non-zero steady state within each dilution cycle ( Figure 3B , Case II ) . From mechanistic dynamics where initial species ratio is 1 , we use two training windows to derive saturable L-V ( dotted ) and alternative ( dashed ) pairwise models . We then use these pairwise models to predict dynamics of communities starting at two different ratios . The alternative model but not the saturable L-V predicts the mechanistic model dynamics . All parameters are listed in Figure 4—source data 2 . Note that in all figures , population fractions ( instead of population densities ) are plotted , which fluctuate less during dilutions compared to mediator concentration . DOI: http://dx . doi . org/10 . 7554/eLife . 25051 . 01710 . 7554/eLife . 25051 . 018Figure 4—source data 1 . List of parameters for simulations in Figure 4 on an interaction through a reusable mediator . DOI: http://dx . doi . org/10 . 7554/eLife . 25051 . 01810 . 7554/eLife . 25051 . 019Figure 4—source data 2 . List of parameters for simulations in Figure 4 on an interaction through a consumable mediator . DOI: http://dx . doi . org/10 . 7554/eLife . 25051 . 019 Case II: When S1 and S2 can coexist ( the basal fitness of S1 is higher than the basal fitness of S2 but less than the maximal fitness of S2 ) , a steady state solution for C1 and species ratio RS=S2/S1 exists ( Figure 4B; Methods-Deriving a pairwise model for interactions mediated by a single consumable mediator Case II , Equation 11 ) . To arrive at a pairwise model , we will need to eliminate C1 which is mathematically possible ( i . e . after community dynamics converges to the ‘f-zero-isocline’ on the phase plane of mediator C1 and species ratio RS , as depicted by blue lines in Figure 3—figure supplement 2A–D ) . However , the derived pairwise model differs from the saturable L-V model: ( 4 ) dS2dt=r20S2+rS2C1S1ωS1+ψS2S2 where constants r20 , rS2C1 , and ω=1−KS2C1/KC1S2 can be positive , negative , or zero , and ψ = ( KS2C1αC1S2 ) / ( KC1S2βC1S1 ) is positive ( see Figure 1 table for parameter definitions and see Equation 13 in Methods ) . We will refer to this equation as ‘alternative pairwise model’ , although the fitness influence term is a function of both S1 and S2 instead of the influencer S1 alone as defined in the traditional L-V pairwise model . Case III: When supplier S1 always grows slower than consumer S2 , i . e . when the basal fitness of S1 ( r10 ) is less than the basal fitness of S2 ( r20 ) , consumable C1 declines to zero concentration . This is because C1 is consumed by S2 whose relative abundance over S1 eventually exponentially increases at a rate of r20−r10 . Similar to Case II , under certain conditions ( i . e . after community dynamics converges to the f-zero-isocline as seen in Figure 3—figure supplement 2E–H ) , the alternative pairwise model ( Equation 4 ) can be derived ( Methods-Deriving a pairwise model for interactions mediated by a single consumable mediator , Case III ) . For both Case II and Case III , we analytically demonstrate that in the absence of dilutions , alternative pairwise model dynamics can converge to mechanistic model dynamics ( see Figure 3—figure supplements 3 and 5 for initial condition requirement and time scale of convergence ) . However , if initial S1 and S2 are such that the time scale of convergence is long compared to the duration of one dilution cycle ( e . g . Figure 3—figure supplement 2C and G ) , then we will have to perform dilutions and the saturable L-V model can sometimes be more appropriate than the alternative model ( Figure 3—figure supplement 4 ) . Thus in these cases , whether a saturable L-V or an alternative model is more appropriate also depends on initial conditions . The alternative model ( Equation 4 ) can be further simplified to ( 5 ) dS2/dt= ( r20+ρS1/S2 ) S2 if additionally , the half-saturation constant K for C1 consumption ( KC1S2 ) is identical to that for C1’s influence on the growth of consumer ( KS2C1 ) , and if S2 has not gone extinct . This equation form has precedence in the literature ( e . g . [Mougi and Kondoh , 2012] ) , where the interaction strength r21 reflects the fact that the consumable mediator from S1 is divided among consumer S2 . Thus , we can regard the alternative model ( Equation 4 ) or its simplified version ( Equation 5 ) as a ‘divided influence’ model . The saturable L-V model and the alternative model are not interchangeable ( Figure 4 ) . When a consumable mediator accumulates without reaching a steady state within each dilution cycle ( Figure 4A left; inset: C1 eventually becomes proportional to S1 ) , the saturable L-V model is predictive of community dynamics ( Figure 4A right , compare dotted and solid lines ) . In contrast , predictions from the alternative pairwise model are qualitatively wrong ( Figure 4A right , compare dashed and solid lines ) . When a consumable mediator eventually reaches a non-zero steady state within each dilution cycle ( Figure 4B , black ) , could a saturable L-V model still work ? The saturable L-V model derived from training window i ( initial 10 generations ) fails to predict species coexistence regardless of initial species ratios ( Figure 4B left magenta box , compare solid with dotted ) . In comparison , the saturable L-V model derived from training window ii ( at steady-state species ratio ) performs better , especially if the starting species ratio is identical to that of the training dynamics ( Figure 4B , top panel in right magenta box ) . However , at a different starting species ratio , the saturable L-V model fails to predict which species dominates the community ( Figure 4B , bottom panel in right magenta box ) . In contrast , community dynamics can be described by the alternative pairwise model derived from either window i or ii ( Figure 4B , compare dashed and solid lines in left and right magenta boxes ) . We have shown here that even when one species affects another species via a single mediator , either a saturable L-V model or an alternative pairwise model may be appropriate . The appropriate model depends on whether the mediator is reusable or consumable , how fitness of the two species compare , and initial species densities ( Figure 3; Figure 3—figure supplements 2–5 ) . Choosing the wrong pairwise model generates qualitatively flawed predictions ( Figure 4 ) . Considering that reusable and consumable mediators are both common in microbial interactions , our results call for revisiting the universality assumption of pairwise modeling . A species often affects another species via multiple mediators ( Kato et al . , 2008; Yang et al . , 2009; Traxler et al . , 2013; Kim et al . , 2013 ) . For example , a fraction of a population might die and release numerous chemicals , and some of these chemicals can simultaneously affect another individual . Here we examine the case where S1 releases two reusable chemicals C1 and C2 , both affecting the growth of S2 ( Figure 5A ) . Since the effect of each mediator can be described by a saturable L-V pairwise model ( Figure 3A ) , we ask when the two mediators can be mathematically regarded as one mediator and described by a saturable L-V pairwise model ( Figure 5B ) . 10 . 7554/eLife . 25051 . 020Figure 5 . An example of a two-mediator interaction where a saturable L-V pairwise model may succeed or fail depending on initial conditions . ( A ) One species can affect another species via two reusable mediators , each with a different potency KCi where KCi is KS2Cir10/βCiS1 ( Methods-Conditions under which a saturable L-V pairwise model can represent one species influencing another via two reusable mediators ) . A low KCi indicates a strong potency ( e . g . high release of Ci by S1 or low Ci required to achieve half-maximal influence on S2 ) . ( B ) Under what conditions can an interaction via two reusable mediators with saturable effects on recipients be approximated by a saturable L-V pairwise model ? ( C ) A community where the success or failure of a saturable L-V pairwise model depends on initial conditions . Here , KC1= 103 cells/ml and KC2= 105 cells/ml . Community dynamics starting at low S1 ( solid ) can be predicted if the saturable L-V pairwise model is derived from reference dynamics starting at low ( dotted ) . However , if we use a saturable L-V pairwise model derived from a community with high initial S1 , prediction is qualitatively wrong ( dash dot line ) . See Figure 5—figure supplement 1D for an explanation why a saturable L-V pairwise model estimated at one community density may not be applicable to another community density . Simulation parameters are listed in Figure 5—source data 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 25051 . 02010 . 7554/eLife . 25051 . 021Figure 5—source data 1 . List of parameters for simulations in Figure 5 on an interaction through two concurrent mediators . DOI: http://dx . doi . org/10 . 7554/eLife . 25051 . 02110 . 7554/eLife . 25051 . 022Figure 5—source data 2 . List of parameters for simulations in Figure 5—figure supplement 1 on an interaction through two concurrent mediators , assessed at high versus low cell densities . DOI: http://dx . doi . org/10 . 7554/eLife . 25051 . 02210 . 7554/eLife . 25051 . 023Figure 5—figure supplement 1 . Except under special conditions , a pairwise interaction through two mediators may not be represented by a single saturable L-V model . ( A ) Consider the interaction in Figure 5 . The fitness effect of S1 on S2 via C1 and C2 is rS2 , C1C2=rS2C1S1S1+KS2C1r10/βC1S1+rS2C2S1S1+KS2C2r10/βC2S1=rS2C1S1S1+KC1+rS2C2S1S1+KC2 . ( B–C ) Under special conditions the fitness effect rS2 , C1C2 ( magenta line ) can be approximated using a single saturable L-V model ( grey dash-dot line ) at all densities . These special conditions include when the potencies of two mediators , KC1 and KC2 , are similar ( B ) or the potency of one mediator is orders of magnitude stronger than the other ( C ) . Otherwise , saturable L-V pairwise models derived from a low-density community and from a high-density community can have qualitatively different parameters ( D ) . Let’s first consider the low-density case ( left black and blue bars corresponding to low total density and therefore low , respectively ) . When rS2 , C1C2 ( magenta line ) is above the ( r10−r20 ) line ( grey dashed line ) , the fitness of S2 ( rS2 , C1C2+r20 ) will be higher than the fitness of S1 . Thus , even though S1 grows at its basal fitness during a dilution cycle , S1 fraction will decrease . Thus S1 will decrease at the next dilution cycle when total density is reset to a pre-fixed level ( arrow pointing towards lower S1 ) . In contrast , when rS2 , C1C2<r10−r20 , S1 population fraction and S1 will increase at the next dilution cycle ( arrow pointing towards higher S1 ) . Thus , the dynamics will converge to a steady state ratio ( filled dot ) . Interaction coefficient of a saturable L-V ( grey dash-dot line ) is estimated to be a positive value ( =+0 . 039 ) . In contrast , in the high-density case ( right black and blue bars ) , r10>rS2 , C1C2+r20 , and S2 goes extinct . Interaction coefficient of a saturable L-V ( grey dash-dot line ) is estimated to be a negative value ( = −0 . 010 ) . As a result , a saturable L-V pairwise model with parameters estimated at high densities cannot predict communities at low densities ( Figure 5 ) . All parameters are listed in Figure 5—source data 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 25051 . 023 We assume that fitness effects from different chemical mediators on a focal species are additive . Not making this assumption will likely violate the additivity assumption essential to pairwise models . Additive fitness effects have been observed for certain ‘homologous’ metabolites . For example , in multi-substrate carbon-limited chemostats of E . coli , the fitness effects from glucose and galactose were additive ( Lendenmann and Egli , 1998 ) . ‘Heterologous’ metabolites such as carbon and nitrogen sources likely affect cell fitness in a multiplicative fashion . However , if WC and WN , the fitness influences of released carbon and nitrogen with respect to those already in the environment , are both small ( i . e . WC , WN< < 1 ) , the additional relative fitness influence will be additive: ( 1+WC ) ( 1+WN ) −1≈WC+WN . However , we need to keep in mind that even among homologous metabolites , fitness effects may not be additive ( Hermsen et al . , 2015 ) . ‘Sequential’ metabolites ( e . g . diauxic shift ) provide another example of non-additivity . Similar to the previous section , we assume that all abiotic resources are unlimited . For the two reusable mediators , depending on their relative ‘potency’ ( defined in Figure 5A legend ) , their combined effect generally cannot be modeled as a single mediator except under special conditions ( Methods-Conditions under which a saturable L-V pairwise model can represent one species influencing another via two reusable mediators ) . These special conditions include: ( 1 ) mediators share similar potency ( Figure 5—figure supplement 1B ) , or ( 2 ) one mediator dominates the interaction ( Figure 5—figure supplement 1C ) . If these conditions are not satisfied , we can easily find examples where saturable L-V pairwise models derived from a low-density community and from a high-density community have qualitatively different parameters ( Figure 5—figure supplement 1D ) . Consequently , the future dynamics of a low-density community can be predicted by a saturable L-V model derived from a low-density community but not by a model derived from a high-density community ( Figure 5C ) . Thus , even though each mediator can be modeled by saturable L-V , their joint effects may or may not be modeled by saturable L-V depending on the relative potencies of the two mediators and sometimes even on initial conditions ( high or low initial S1 ) . Similarly , when both mediators are consumable and do not accumulate ( as in Cases II and III of Figure 3B ) , the fitness effect term becomes rS2C1S1ωC1S1+ψC1S2+rS2C2S1ωC2S1+ψC2S2 . Except under special conditions ( e . g . when ωC1 and ωC2 are zero , or when ωC1/ωC2=ψC1/ψC2 , or when one mediator dominates the interaction ) , the two mediators may not be regarded as one . By the same token , when one mediator is a steady-state consumable and the other is reusable , they generally may not be regarded as a single mediator and would require yet a different pairwise model ( i . e . with the fitness effect term rS2C1S1ωC1S1+ψC1S2+rS2C2S1S1+KS2C2r10/βC2S1 ) . In summary , when S1 influences S2 through multiple mediators , rarely can we approximate them as a single mediator . Sometimes , a pairwise model derived from one community may not apply to communities initiated at different densities ( Figure 5C; Figure 5—figure supplement 1D ) . This casts further doubt on the usefulness of a single pairwise model for all pairwise microbial interactions . So far , by assuming that abiotic resources are always present in excess ( e . g . in turbidostats ) , we have not considered species competition for abiotic resources . In this section , we consider a competitive commensal community in a batch environment where S1 and S2 compete for an essential shared resource C1 supplied by the environment at a constant rate ( e . g . constant light ) , and S1 supplies an essential consumable metabolite C2 to promote S2 growth ( Figure 6A , left ) . We show that an L-V pairwise model works for some but not all communities even though these communities qualitatively share the same interaction mechanism . 10 . 7554/eLife . 25051 . 024Figure 6 . An example of a competitive commensal community where an L-V pairwise model may work or fail . ( A ) Left: Two species S1 and S2 compete for shared resource C1 . Additionally , S1 produces C2 that promotes the growth of S2 upon consumption . Right: An L-V pairwise model captures the intra- and inter-species competition as well as the commensal interaction between the two species . ( B , C ) Examples where L-V pairwise models predict the mechanistic reference dynamics well . ( D ) An example where the L-V pairwise model fails to predict the dynamics qualitatively ( note the much longer time range ) . Here , population fractions fluctuate due to changes in relative concentration of C1 compared to C2 . In all cases , the pairwise model is derived from the population dynamics in the initial stages of growth ( 150 hr in all cases ) . Simulation parameters are listed in Figure 6—source data 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 25051 . 02410 . 7554/eLife . 25051 . 025Figure 6—source data 1 . List of parameters for simulations in Figure 6 on an interaction through a consumable mediator , for species consuming a shared abiotic resource . DOI: http://dx . doi . org/10 . 7554/eLife . 25051 . 025 In our mechanistic model ( Methods-Competitive commensal interaction , Equation 47 ) , the fitness of S2 is multiplicatively affected by C1 and C2 ( Mankad and Bungay , 1988 ) . We choose parameters such that the effect from C2 to S2 is far from saturation ( e . g . linear with respect to C2 and S1 ) to simplify the problem . In our L-V pairwise model ( Figure 6A , right; Methods-Competitive commensal interaction , Equation 48 ) , intra- and inter-species competition is represented by the traditional logistic L-V model ( Equation 2; Gause , 1934; Thébault and Fontaine , 2010; Mougi and Kondoh , 2012 ) . We then introduce a linear term ( r21S1 ) to describe the fitness effect of commensal interaction . We tested various sets of mechanistic model parameters where the two species coexist in a steady fashion ( Figure 6B ) , or one species goes extinct ( Figure 6C ) , or species composition fluctuates ( Figure 6D ) . L-V pairwise models deduced from a fixed period of training time could predict future dynamics in the first two cases , but failed to do so in the third case . Thus , depending on dynamic details of communities , a pairwise model sometimes works and sometimes fails . To summarize our work , even for pairwise microbial interactions , depending on interaction mechanisms ( reusable versus consumable mediator , single mediator versus multiple mediators ) , we will need to use a plethora of pairwise models to avoid qualitative failures in predicting which species dominates a community or whether species coexist ( Figures 3 , 4 and 5 ) . Sometimes , even when different communities share identical interaction mechanisms , depending on details such as relative species fitness , interaction strength , and initial conditions , the best-fitting pairwise model may or may not predict future dynamics ( Figure 3B , Figure 3—figure supplement 4 , Figure 4 , Figure 5 , Figure 5—figure supplement 1 , and Figure 6 ) . This defeats the very purpose of pairwise modeling – using a single equation form to capture fitness effects of all pairwise species interactions regardless of interaction mechanisms or quantitative details . In a community of more than two microbial species , interaction modification can cause pairwise models to fail ( Figure 7 ) . Even if species interact in an interaction chain and thus interaction modification does not occur , various chain segments may require different forms of pairwise models . Taken together , a pairwise model is unlikely to be effective for predicting community dynamics especially if interaction mechanisms are diverse . 10 . 7554/eLife . 25051 . 026Figure 7 . Interaction chain but not interaction modification may be represented by a multispecies pairwise model . We examine three-species communities engaging in indirect interactions . Each species pair is representable by a two-species pairwise model ( saturable L-V or alternative pairwise model , purple in the right columns of B , D , and F ) . We then use these two-species pairwise models to construct a three-species pairwise model , and test how well it predicts the dynamics known from the mechanistic model . In B , D , and F , left panels show dynamics from the mechanistic models ( solid lines ) and three-species pairwise models ( dotted lines ) . Right panels show the difference metric D¯ . ( A–B ) Interaction chain: S1 affects S2 , and S2 affects S3 . The two interactions employ independent mediators C1 and C2 , and both interactions can be represented by the saturable L-V pairwise model . The three-species pairwise model matches the mechanistic model in this case . Simulation parameters are provided in Figure 7—source data 1 . ( C–F ) Interaction modification . In both cases , the three-species pairwise model fails to predict reference dynamics even though the dynamics of each species pair can be represented by a pairwise model . ( C–D ) S3 consumes C1 , a mediator by which S1 stimulates S2 . Parameters are listed in Figure 7—source data 2 . Here , S1 changes the nature of interaction between S2 and S3: S2 and S3 do not interact in the absence of S1 , but S3 inhibits S2 in the presence of S1 . The three-species pairwise model makes qualitatively wrong prediction about species coexistence . As expected , if S3 does not remove C1 , the three-species pairwise model works ( Figure 7—figure supplement 1A–B ) . ( E–F ) S1 and S3 both supply C1 which stimulates S2 . Here , no species changes ‘the nature of interactions’ between any other two species: both S1 and S3 contribute reusable C1 to stimulate S2 . S1 promotes S2 regardless of S3; S3 promotes S2 regardless of S1; S1 and S3 do not interact regardless of S2 . However , a multispecies pairwise model assumes that the fitness effects from the two producers on S2 will be additive , whereas in reality , the fitness effect on S2 saturates at high . As a result , the three-species pairwise model qualitatively fails to capture relative species abundance . As expected , if C1 affects S2 in a linear fashion , the community dynamics is accurately captured in the multispecies pairwise model ( Figure 7—figure supplement 1C–D ) . Simulation parameters are listed in Figure 7—source data 3 . DOI: http://dx . doi . org/10 . 7554/eLife . 25051 . 02610 . 7554/eLife . 25051 . 027Figure 7—source data 1 . List of parameters for simulations in Figure 7B on interaction between three species in a chain . DOI: http://dx . doi . org/10 . 7554/eLife . 25051 . 02710 . 7554/eLife . 25051 . 028Figure 7—source data 2 . List of parameters for simulations in Figure 7D on interaction modification through consumption of a shared mediator by a third species . DOI: http://dx . doi . org/10 . 7554/eLife . 25051 . 02810 . 7554/eLife . 25051 . 029Figure 7—source data 3 . List of parameters for simulations in Figure 7F on interaction modification through production of a shared mediator by a third species . DOI: http://dx . doi . org/10 . 7554/eLife . 25051 . 02910 . 7554/eLife . 25051 . 030Figure 7—source data 4 . List of parameters for simulations in Figure 7—figure supplement 1B on an interaction between three species through a shared reusable mediator affecting multiple species . DOI: http://dx . doi . org/10 . 7554/eLife . 25051 . 03010 . 7554/eLife . 25051 . 031Figure 7—source data 5 . List of parameters for simulations in Figure 7—figure supplement 1D on an interaction between three species through a shared reusable mediator produced by multiple species . DOI: http://dx . doi . org/10 . 7554/eLife . 25051 . 03110 . 7554/eLife . 25051 . 032Figure 7—figure supplement 1 . A multispecies pairwise model can work under special conditions . ( A–B ) As a control for Figure 7C , if S3 does not remove the mediator of interaction between S1 and S2 , a three-species pairwise model accurately matches the mechanistic model . Simulation parameters are provided in Figure 7—source data 4 . ( C–D ) As a control for Figure 7E , we ensured that fitness effects from multiple species are additive . In this case , a three-species pairwise model can represent the mechanistic model . To ensure the linearity and additivity of fitness effects , we have used a larger value of half saturation concentration ( KS2C1=103 μM , instead of 10−1 μM in Figure 7E–F ) . We have adjusted the interaction coefficients accordingly such that the overall interaction strength exerted by S1 and S3 on S2 is comparable to that in Figure 7E–F ( as evident by comparable population compositions ) . Since the interaction influences under these conditions remain in the linear range , the three-species pairwise model accurately predicts the reference dynamics . Simulation parameters are provided in Figure 7—source data 5 . DOI: http://dx . doi . org/10 . 7554/eLife . 25051 . 032 Multispecies pairwise models are widely used in theoretical ecology due to their simplicity . These models assume that all pairwise species interactions can be captured by a single pairwise model regardless of interaction mechanisms or quantitative details of a community ( universality assumption ) . This assumption may be satisfied if , for example , interaction mediators are always species themselves ( e . g . prey-predation in a food web ) so that pairwise models are equivalent to mechanistic models . However , interactions in microbial communities are diverse and often mediated by chemicals ( Figure 2 ) . Here , we consider the validity of universality assumption of pairwise models in well-mixed , two-species microbial communities . We have focused on various types of chemical-mediated interactions commonly encountered in microbial communities ( Figure 2 ) ( Kato et al . , 2005; Gause , 1934; Ghuysen , 1991; Jakubovics et al . , 2008; Chen et al . , 2004; D'Onofrio et al . , 2010; Johnson et al . , 1982; Hamilton and Ng , 1983 ) . For each type of species interaction , we construct a mechanistic model to generate reference community dynamics ( akin to experimental results ) . We then attempt to derive the best-matching pairwise model and ask how predictive it is . We first consider cases where abiotic resources are in excess . When one species affects another species via a single chemical mediator , either the saturable L-V or the alternative pairwise model is appropriate , depending on the interaction mechanism ( consumable versus reusable mediator ) , relative fitness of the two species , and initial conditions ( Figure 3; Figure 3—figure supplement 2 to Figure 3—figure supplement 5 ) . These two models are not interchangeable ( Figure 4 ) . If one species influences another species through multiple mediators , then in general , these mediators may not be regarded as a single mediator nor representable by a single pairwise model . For example , for two reusable mediators , unless their potencies are similar or one mediator is much more potent than the other , saturable L-V model parameters can qualitatively differ depending on initial community density ( Figure 5—figure supplement 1D ) . Consequently , a pairwise model derived from a high-density community generates false predictions for low-density communities ( Figure 5C ) , limiting the usefulness of pairwise models . We then consider a community where two species compete for a shared resource while engaging in commensalism via a single chemical mediator . We find that the best-fitting L-V pairwise model can predict future dynamics in some but not all communities , depending on parameters used in the mechanistic model ( Figure 6 ) . Thus , although a single equation form can work in many cases , it generates qualitatively wrong predictions in many other cases . In communities of more than two microbial species , indirect interactions via a third species can occur . When indirect interactions take the form of interaction chains , if each chain segment of two species engages in an independent interaction and can be represented by a pairwise model , then multispecies pairwise models can work ( Figure 7A-B ) . However , as discussed above , pairwise equation forms may vary among chain segments depending on interaction mechanisms and quantitative details of a community . When indirect interactions take the form of interaction modification , even if each species pair can be accurately represented by a pairwise model , a multispecies pairwise model may fail ( Figure 7C–F , ) . Interaction modification includes trait modification ( Wootton , 2002; Werner and Peacor , 2003; Schmitz et al . , 2004 ) , or , in our cases , mediator modification . Mediator modification is very common in microbial communities . For example , antibiotic released by one species to inhibit another species may be inactivated by a third species , and this type of indirect interactions can stabilize microbial communities ( Kelsic et al . , 2015; Bairey et al . , 2016 ) . As another example , interaction mediators are often generated by and shared among multiple species . For example in oral biofilms , organic acids such as lactic acid are generated from carbohydrate fermentation by many species ( Bradshaw et al . , 1994; Marsh and Bradshaw , 1997; Kuramitsu et al . , 2007 ) . Such by-products are also consumed by multiple species ( Kolenbrander , 2000 ) . One can argue that an extended pairwise model ( e . g . dS2dt=r20S2+rS2CS1ς+ωS1+ψS2S2 ) embodying both the saturable form and the alternative form can serve as a general-purpose model at least for pairwise interactions via a single mediator . In fact , even the effects of indirect interactions may be quantified and included in the model by incorporating higher-order interaction terms ( Case and Bender , 1981; Worthen and Moore , 1991 ) , although with many challenges ( Wootton , 2002 ) . In the end , although these strategies may lead to a sufficiently accurate phenomenological model especially within the training window , they may fail to predict future dynamics . When might a pairwise model be useful ? First , pairwise models have been instrumental in understanding ecological phenomena such as prey-predator oscillatory dynamics and coexistence of competing predator species ( Volterra , 1926; MacArthur , 1970; Case and Casten , 1979; Chesson , 1990 ) . In these cases , mechanistic models are either identical to pairwise models or can be transformed into pairwise models under simplifying assumptions . Second , pairwise models of pairwise species interactions can provide a bird’s-eye view of strong or weak stimulatory or inhibitory interactions in a community . For example , Vetsigian et al . , 2011 found that interactions between soil-isolated Streptomyces strains are enriched for reciprocity – if A inhibits or promotes B , it is likely that B also inhibits or promotes A ( Vetsigian et al . , 2011 ) . Third , pairwise models have been useful in qualitatively understanding species assembly rules in small communities ( Friedman et al . , 2017 ) . That is , qualitative information regarding species survival in competitions among a small number of species may be used to predict survival in more diverse communities within a similar time window . Fourth , a pairwise model can serve as a starting point for generating hypotheses on species interactions ( e . g . Li et al . , 2015 ) . Note that when applied to microbial communities ( Mounier et al . , 2008; Stein et al . , 2013; Marino et al . , 2014 ) , a fitting pairwise model means that the training dynamics of the community under investigation can be approximated by a theoretical community where species interactions satisfy the additivity and universality assumptions of pairwise models . Even though the theoretical community is likely different from the real community , hypothesis formulation can still be valuable . Finally , pairwise models can be useful in making predictions of limited scales . For example , Stein et al . used 2/3 of community dynamics data as a training set to derive a multispecies pairwise model , and in the best-case scenario , the model generated reasonable predictions on the remaining 1/3 of data ( Stein et al . , 2013 ) . However , as we have shown , pairwise models can generate qualitatively wrong predictions ( Figures 4–7 ) , especially if interaction mechanisms are diverse such as in microbial communities . Not surprisingly , predicting qualitative consequences of species removal or addition using a pairwise model has encountered difficulties , especially in communities of more than three species ( Mounier et al . , 2008; Friedman et al . , 2017 ) . An alternative to a pairwise model is a mechanistic model . How much information about interaction mechanisms do we need to construct a mechanistic model ? That is , what is the proper level of abstraction which captures the phenomena of interest , yet avoids unnecessary details ( Li et al . , 2015; Durrett and Levin , 1994 ) ? For example , Tilman argued that if a small number of mechanisms ( e . g . the ‘axes of trade-offs’ in species traits ) could explain much of the observed pattern ( e . g . species coexistence ) , then this abstraction would be highly revealing ( Tilman , 1987 ) . However , the choice of abstraction is often not obvious . Consider for example a commensal community where S1 grows exponentially ( not explicitly depicted in equations in Figure 8 ) and the net growth rate of S2 , which is normally zero , is promoted by mediator C from S1 in a linear fashion ( Figure 8 ) . If we do not know how S1 stimulates S2 , we can still construct an L-V pairwise model ( Figure 8A ) . If we know the identity of mediator C and realize that C is consumable , then we can instead construct a mechanistic model incorporating C ( Figure 8B ) . However , if C is produced from a precursor via an enzyme E released by S1 , then we get a different form of mechanistic model ( Figure 8C ) . If , on the other hand , E is anchored on the membrane of S1 and each cell expresses a similar amount of E , then equations in Figure 8D are mathematically equivalent to Figure 8B . This simple example , inspired by extracellular breakdown of cellulose into a consumable sugar C ( Bayer and Lamed , 1986; Felix and Ljungdahl , 1993; Schwarz , 2001 ) , illustrates how knowledge of mechanisms may eventually help us determine the right level of abstraction . 10 . 7554/eLife . 25051 . 033Figure 8 . Different levels of abstraction in a mechanistic model . How one species ( S1 ) may influence another ( S2 ) can be mechanistically modeled at different levels of abstraction . For simplicity , here we assume that interaction strength scales in a linear ( instead of saturable ) fashion with respect to mediator concentration or species density . The basal fitness of S2 is zero . ( A ) In the simplest form , S1 stimulates S2 in an L-V pairwise model . ( B ) In a mechanistic model , we may realize that S1 stimulates S2 via a mediator C which is consumed by S2 . The corresponding mechanistic model is given . ( C ) Upon probing more deeply , it may become clear that S1 stimulates S2 via an enzyme E , where E degrades an abundant precursor ( such as cellulose ) to generate mediator C ( such as glucose ) . In the corresponding mechanistic model , we may assume that E is released by S1 at a rate ζES1 and that E liberates C at a rate ηCE . ( D ) If instead E is anchored on the cell surface ( e . g . cellulosome ) , then E is proportional to S1 . If we substitute E into the second equation , then ( B ) and ( D ) become equivalent . Thus , when enzyme is anchored on cell surface but not when enzyme is released , the mechanistic knowledge of enzyme can be neglected . DOI: http://dx . doi . org/10 . 7554/eLife . 25051 . 033 In summary , under certain circumstances , we may already know that microbial interaction mechanisms fall within the domain of validity for a particular pairwise model . In these cases , a pairwise model provides the appropriate level of abstraction , and constructing such a pairwise model is much easier than a mechanistic model ( Figure 1 ) . However , if we do not know whether a pairwise model is valid , we will need to be cautious since pairwise models can fail to even qualitatively capture pairwise microbial interactions . We need to be equally careful when extrapolating and generalizing conclusions obtained from pairwise models , especially for communities where species interaction mechanisms are diverse . Considering recent advances in identifying and quantifying interactions , we advocate a transition to models that incorporate interaction mechanisms at the appropriate level of abstraction . In a pairwise model , the fitness of a focal species Si is the sum of its ‘basal fitness’ ( ri0 , the net growth rate of a single Si individual in the absence of any intra-species or inter-species interactions ) and the additive fitness effects exerted by pairwise interactions with other members of the community . Mathematically , an N-species pairwise model is often formulated as ( 6 ) dSidt= ( ri0+∑j=1Nfij ( Sj ) ) Si Here , fij ( Sj ) describes how Sj , the density of species Sj , positively or negatively affects the fitness of Si , and is a linear or nonlinear function of only Sj . Indirect interactions via a third species fall under two categories ( Wootton , 1993 ) . The first type is known as ‘interaction chain’ or ‘density-mediated indirect interactions’ . For example , the consumption of plant S1 by herbivore S2 is reduced when the density of herbivore is reduced by carnivore S3 . In this case , the three-species pairwise model ( 7 ) {dS1dt= ( r10−f12 ( S2 ) ) S1dS2dt= ( r20+f21 ( S1 ) −f23 ( S3 ) ) S2dS3dt= ( r30+f32 ( S2 ) ) S3 does not violate the additivity assumption ( compare with Equation 6 ( Case and Bender , 1981; Wootton , 1994 ) . The second type of indirect interactions is known as ‘interaction modification’ or ‘trait-mediated indirect interactions’ or ‘higher order interactions’ ( Vandermeer , 1969; Wootton , 1994; Billick and Case , 1994; Wootton , 2002 ) , where a third species modifies the ‘nature of interaction’ from one species to another ( Wootton , 2002; Werner and Peacor , 2003; Schmitz et al . , 2004 ) . For example , when carnivore is present , herbivore will spend less time foraging and consequently plant density increases . In this case , f12 in Equation 7 is a function of both S2 and S3 , violating the additivity assumption . Simulations are based on Matlab and executed on an ordinary PC . Steps are: Step 1: Identify monoculture parameters ri0 , rii , and Kii ( Figure 1—figure supplement 2C , Row 1 and Row 2 ) . Step 2: Identify interaction parameters rij , rji , Kij , and Kji where i≠j ( Figure 1—figure supplement 2C , Row 3 ) . Step 3: Calculate distance D¯ between population dynamics of the reference mechanistic model and the approximate pairwise model over a period of time outside of the training window to assess if the pairwise model is predictive . Fitting is performed using nonlinear least squares ( lsqnonlin routine ) with default optimization parameters . The following list describes the m-files used for different steps of the analysis: File name Function FitCost_BasalFitnessSource code 1 Calculates the cost function for monocultures ( i . e . the difference between the target mechanisticmodel dynamics and the dynamics obtained from the pairwise model ) FitCost_BFSatLV . mSource code 2 Calculates the cost function for communities ( i . e . the difference between the target mechanisticmodel dynamics and the dynamics obtained from the saturable L-V pairwise model ) FitCost_BFSatLV_Dp . mSource code 3 Calculates the cost function for communities ( i . e . the difference between the target mechanistic modeldynamics and the dynamics obtained from the alternative pairwise model ) DynamicsMM_WM_MonocultureDpMM . mSource code 4 Returns growth dynamics for monocultures , based on the mechanistic modelDynamicsMMSS_WM_NetworkDpMM . mSource code 5 Returns growth dynamics for communities of multiple species , based on the mechanistic modelDynamicsWM_NetworkBFSatLV . mSource code 6 Returns growth dynamics for communities of multiple species , based on the saturable L-V pairwisemodelDynamicsWM_NetworkBFSatLV_Dp . mSource code 7 Returns growth dynamics for communities of multiple species , based on the alternative pairwise modelDeriveBasalFitnessMM_WM_DpMM . mSource code 8 Estimates monoculture parameters of pairwise model ( Step 1 ) DeriveBFSatLVMMSS_WM_DpMM . mSource code 9 Estimates saturable L-V pairwise model interaction parameters ( Step 2 ) DeriveBFSatLVMMSS_WM_DpMM_Dp . mSource code 10 Estimates alternative pairwise model interaction parameters ( Step 2 ) DeriveBFSatLVMMSS_WM_DpMM_r21 . mSource code 11 Estimates saturable L-V pairwise model interaction parameters ( r21 and K21 ) in cases where we knowthat S2 is only affected by S1 , to accelerate optimizationDeriveBFSatLVMM_WM_DpMM_Dp_r21 . mSource code 12 Estimates alternative pairwise model interaction parameter ( r21 ) in cases where we know that S2 is only affected by S1 and that KS2C1=KC1S2 to accelerate optimizationDynamicsWM_NetworkBFLogLV_DI . mSource code 13 Returns growth dynamics for communities of two species competing for an environmental resourcewhile engaging in an additional interaction , based on the logistic L-V pairwise model ( Figure 6 ) C2Sp2_ARCLi_NoSatDp_FitBFLogLV_DI . mSource code 14 Estimates logistic L-V pairwise model interaction parameters for communities of two speciescompeting for an environmental resource while engaging in an additional interaction , andcompares community dynamics from pairwise and mechanistic models ( Figure 6 ) Dynamics_WM_NetworkDpMM_ODE23 . mSource code 15 Defines differential equations when using Matlab’s ODE23 solver to calculate community dynamicsCase_C1Sp2_CmnsDp_ODE23 . mSource code 16 Example of using Matlab ODE23 solver for calculating community dynamics To facilitate mathematical analysis , we assume that requirements calculated below are eventually satisfied within each dilution cycle ( see Figure 3—figure supplement 4 for an example where dilution cycles necessitated by long convergence time violate requirements for a pairwise model to converge to the mechanistic model ) . We further assume that r10>0 and r20>0 so that species cannot go extinct in the absence of dilution . See Figure 3—figure supplement 5 for a summary of this section . When S1 releases a consumable mediator which stimulates the growth of S2 , the mechanistic model as per Figure 3B , is ( 8 ) {dS1dt=r10S1dS2dt=r20S2+rS2C1C1C1+KS2C1S2dC1dt=βC1S1S1−αC1S2C1C1+KC1S2S2= ( βC1S1−αC1S2C1C1+KC1S2S2S1 ) S1 Let C1 ( t=0 ) =C10=0; S1 ( t=0 ) =S10; and S2 ( t=0 ) =S20 . Note that the initial condition C10=0 can be easily imposed experimentally by pre-washing cells . Under which conditions can we eliminate C1 so that we can obtain a pairwise model of S1 and S2 ? Define RS=S2/S1 as the ratio of the two populations . ( 9 ) dRSdt=dS2dtS1−S2dS1dtS12= ( r20+rS2C1C1C1+KS2C1 ) S2S1−S2S12r10S1= ( r20+rS2C1C1C1+KS2C1−r10 ) RS Cases II and III showed that population dynamics of the mechanistic model could be described by the alternative pairwise model . However , since the initial condition for C1 cannot be specified in pairwise model , problems could occur . To illustrate , we examine the phase portrait of the pairwise equation ( 13 ) dS2dt=r20S2+rS2C1S1ωS1+ψS2S2 where ω=1−KS2C1KC1S2 , ψ=αC1S2KS2C1βC1S1KC1S2 . From Equations 8 and 13 , ( 43 ) dRSdt=d ( S2S1 ) dt= ( r20+rS2C1S1ωS1+ψS2 ) S2S1−S2r10S1S12= ( r20+rS2C1ω+ψRS−r10 ) RS Below , we plot Equation 43 under different parameters ( Figure 3—figure supplement 3 ) to reveal conditions for convergence between mechanistic and pairwise models . If ω=1−KS2C1/KC1S2≥0 ( Figure 3—figure supplement 3A ) : When RS<RS∗ , dRS/dt is positive . When RS>RS∗ , dRS/dt is negative . Thus , wherever the initial RS , it will always converge toward the only steady state RS∗ of the mechanistic model . If ω<0 ( Figure 3—figure supplement 3B ) : ω+ψRS=0 or RS=−ω/ψ creates singularity . Pairwise model RS will only converge toward the mechanistic model steady state if ( 44 ) RS ( 0 ) >−ω/ψ If ω≥0 ( Figure 3—figure supplement 3C ) : Equation 43 dRSdt= ( r20+rS2C1ω+ψRS−r10 ) RS>0 . Thus , Equation 43 , which is based on alternative pairwise model , also predicts that RS will eventually increase exponentially at a rate of r20−r10 , similar to the mechanistic model . If ω<0 ( Figure 3—figure supplement 3D ) : RS ( 0 ) >−ω/ψ ( Equation 44 ) is required for unbounded increase in RS ( similar to the mechanistic model ) . Otherwise , RS converges to an erroneous value instead . Here , we examine a simple case where S1 releases reusable C1 and C2 , and C1 and C2 additively affect the growth of S2 ( see example in Figure 5 ) . Similar to Figure 3A , the mechanistic model is: ( 45 ) {S1=S10exp⁡ ( r10t ) dS2dt= ( r20+rS2C1S1S1+KS2C1r10/βC1S1+rS2C2S1S1+KS2C2r10/βC2S1 ) S2 Now the question is whether the saturable L-V pairwise model{S1=S10exp⁡ ( r10t ) dS2dt= ( r20+r21S1S1+K21 ) S2 can be a good approximation . For simplicity , let’s define KC1=KS2C1r10/βC1S1 and KC2=KS2C2r10/βC2S1 . Small KCi means large potency ( e . g . small KC2 can be caused by small KS2C2 which means low C2 required to achieve half maximal effect on S2 , and/or large synthesis rate βC2S1 ) . Since S1 from pairwise and mechanistic models are identical , we have ( 46 ) D¯=12T∫T\|log10⁡ ( S2 , pair ) −log10⁡ ( S2 , mech ) \|dt=12Tln⁡ ( 10 ) ∫T\|ln⁡ ( S2 , pair ) −ln⁡ ( S2 , mech ) \|dt=12Tln⁡ ( 10 ) ∫T\|∫t\| ( r20+r21S1S1+K21 ) dτ−∫t ( r20+rS2C1S1S1+KC1+rS2C2S1S1+KC2 ) dτ\|dt=12Tln⁡ ( 10 ) ∫T\|∫t[r21S1S1+K21− ( rS2C1S1S1+KC1+rS2C2S1S1+KC2 ) ]dτ\|dt D¯ can be close to zero when ( i ) KC1≈KC2 or ( ii ) rS2C1S1S1+KC1 and rS2C2S1S1+KC2 ( effects of C1 and C2 on S2 ) differ dramatically in magnitude . For ( ii ) , without loss of generality , suppose that the effect of C2 on S2 can be neglected . This can be achieved if ( iia ) rS2C2 is much smaller than rS2C1 , or ( iib ) KC2 is large compared to S1 . For the community in Figure 6A , our mechanistic model is: ( 47 ) dS1dt= ( r10+rS1C1C1C1+KS1C1 ) S1dS2dt=[r20+rS2C1 , 2 ( C1/KS2C1 ) ( C2/KS2C2 ) C1/KS2C1+C2/KS2C2 ( 1C1/KS2C1+1+1C2/KS2C2+1 ) ]S2dC1dt=β0−αC1S1rS1C1C1C1+KS1C1S1−αC1S2rS2C1 , 2C1KS2C1C2KS2C2C1KS2C1+C2KS2C2 ( 1C1KS2C1+1+1C2KS2C2+1 ) S2dC2dt=βC2S1S1−αC2S2rS2C1 , 2 ( C1/KS2C1 ) ( C2/KS2C2 ) C1/KS2C1+C2/KS2C2 ( 1C1/KS2C1+1+1C2/KS2C2+1 ) S2 Here , S1 and S2 are the densities of the two species; ri0 is the basal net growth rate of Si ( negative , representing death in the absence of the essential shared resource C1 ) ; C1 is supplied at a constant rate β0; βC2S1is the production rate of C2 by S1; αCiSj is the amount of resource Ci consumed to produce a new Sj cell . The growth of S2 is controlled by C1 and C2 . When C1 is limiting ( C1/KS2C1≪C2/KS2C2 ) , the fitness influence of the two chemicals on S2 becomes:rS2C1 , 2 ( C1/KS2C1 ) ( C2/KS2C2 ) C1/KS2C1+C2/KS2C2 ( 1C1/KS2C1+1+1C2/KS2C2+1 ) ≈rS2C1 , 2 ( C1/KS2C1 ) ( C2/KS2C2 ) C2/KS2C2 ( 1C1/KS2C1+1 ) =rS2C1 , 2C1/KS2C1C1/KS2C1+1=rS2C1 , 2C1C1+KS2C1 which is the standard Monod equation . A similar argument holds for limiting C2 . We have intentionally chosen very large KS2C2 to ensure that the fitness effect of C2 on S2 is linear with respect to C2 . This way , we minimize the number of pairwise model parameters that need to be estimated . For our L-V pairwise model , to capture intra-species competition , we usedSidt=bi0 ( 1−Siκi ) Si−diSi where non-negative bi0 represents the maximal birth rate of Si at nearly zero population density ( no competition ) , and non-negative di represents the constant death rate of Si . Positive κi is the ‘carrying capacity’ imposed by the limiting resource , and is the Si at which birth rate becomes zero . This equation can be simplified to:dSidt= ( bi0−di ) [1−Siκi ( 1−di/bi0 ) ]Si=ri0[1−SiΛi]Si . When Λi>0 ( i . e . when bi0>di ) , this resembles standard L-V model traditionally used for competitive interactions ( compare to Equation 2; Gause , 1934; Thébault and Fontaine , 2010; Mougi and Kondoh , 2012 ) . Thus , for the competitive commensal community , we have: ( 48 ) dS1dt=b10 ( 1−S1Λ11−S2Λ12 ) S1−d1S1dS2dt= ( b20+r21S1 ) ( 1−S1Λ21−S2Λ22 ) S2−d2S2 Here , birth rate of each species is reduced by competition from the two species , and Λij is the carrying capacity such that a single Si individual will have a zero birth rate when encountering a total of Λij individuals of Sj . For S2 , We used ( b20+r21S1 ) ( 1−S1Λ21−S2Λ22 ) S2 instead of b20 ( 1−S1Λ21−S2Λ22 ) S2+r21S1S2 so that when the shared resource is exhausted ( i . e . 1−S1Λ21−S2Λ22=0 ) , S2 does not keep growing due to the presence of S1 .	From the soil to our body , microbes , such as bacteria , are everywhere and affect us in many ways . Many microbes perform important roles in natural environments and for our health , but some of them can cause harm and lead to diseases . Often , microbes affect and interact with each other within large groups or communities . Because of their widespread ramifications , it is important to understand how microbial communities work . In addition to experiments , mathematical modeling offers one way to gain insight into the dynamics of microbial communities . A model commonly used to describe the interactions between organisms is the so-called ‘pairwise model’ . Pairwise models can be useful to predict the dynamics of a community in which two species physically interact , such as a predator-prey community . However , it was unknown if this model was suitable to adequately predict the dynamics of microbial species in communities . Microbes often interact via chemicals that diffuse in the environment . For example , one microbe might provide food for another microbe or release toxins to kill it . However , a pairwise model does not consider food or toxins , but only how one microbe stimulates or inhibits the growth of another . Momeni et al . simulated different scenarios commonly found in microbial communities to test whether a pairwise model could capture how , for example , chemicals released by one bacterial species would either help others to grow or stop them from growing . The results showed that for many scenarios , pairwise models cannot qualitatively represent the dynamics of a microbial community . A next step will be to work on the limitations of current experimental technologies and mathematical models to improve the understanding of microbial communities . This knowledge could be used to develop new strategies for ecosystem engineering , such as for example making soils more fertile to improve crop yields , or tackling antibiotic resistance of bacteria .	[ "Abstract", "Introduction", "Results", "Discussions", "Materials", "and", "methods" ]	[ "ecology", "computational", "and", "systems", "biology" ]	2017	Lotka-Volterra pairwise modeling fails to capture diverse pairwise microbial interactions
The histone chaperone Chromatin Assembly Factor 1 ( CAF-1 ) deposits tetrameric ( H3/H4 ) 2 histones onto newly-synthesized DNA during DNA replication . To understand the mechanism of the tri-subunit CAF-1 complex in this process , we investigated the protein-protein interactions within the CAF-1-H3/H4 architecture using biophysical and biochemical approaches . Hydrogen/deuterium exchange and chemical cross-linking coupled to mass spectrometry reveal interactions that are essential for CAF-1 function in budding yeast , and importantly indicate that the Cac1 subunit functions as a scaffold within the CAF-1-H3/H4 complex . Cac1 alone not only binds H3/H4 with high affinity , but also promotes histone tetramerization independent of the other subunits . Moreover , we identify a minimal region in the C-terminus of Cac1 , including the structured winged helix domain and glutamate/aspartate-rich domain , which is sufficient to induce ( H3/H4 ) 2 tetramerization . These findings reveal a key role of Cac1 in histone tetramerization , providing a new model for CAF-1-H3/H4 architecture and function during eukaryotic replication . Eukaryotic DNA is incorporated into chromatin structure via nucleosomes , which are units of ~146 DNA base pairs surrounding an octamer of histone proteins ( Kornberg , 1974 ) . The assembly and disassembly of nucleosomes is critical during DNA replication , as duplication of the genome requires concerted duplication of the resident histones . To facilitate this stepwise process ( Smith and Stillman , 1991 ) , eukaryotes employ histone chaperones , proteins that associate tightly with histones prior to histone deposition on DNA ( reviewed in [Ransom et al . , 2010; Das et al . , 2010; Burgess and Zhang , 2013; Gurard-Levin et al . , 2014] ) . The CAF-1 ( Chromatin Assembly Factor ( 1 ) histone chaperone is intimately engaged in replication-dependent nucleosome assembly of nascent ( Verreault et al . , 1996; Smith and Stillman , 1989; Tyler et al . , 2001 ) and parental histones H3/H4 , by way of recruitment to the replication fork through PCNA ( Proliferating cell nuclear antigen ) ( Shibahara and Stillman , 1999; Krawitz et al . , 2002; Moggs et al . , 2000; Rolef Ben-Shahar et al . , 2009 ) . At the fork , parental H3/H4 histones are inherited in a conservative manner , in which intact ( H3/H4 ) 2 tetramers are partitioned onto the replicated daughter strands ( Prior et al . , 1980; Xu et al . , 2010 ) . This model predicts CAF-1 will associate with ( H3/H4 ) 2 tetramers , rather than a single H3/H4 dimer . Multiple biophysical studies confirm that the monomeric form of CAF-1 tetramerizes H3/H4 ( Winkler et al . , 2012b; Liu et al . , 2012 ) . Other monomeric chaperones such as Asf1 bind one H3/H4 dimer ( English et al . , 2005 , 2006 ) , whereas those that accommodate ( H3/H4 ) 2 tetramers – including Nap1 , Vps75 , Mcm2 and Rtt106 – form homodimers themselves ( Bowman et al . , 2011; Su et al . , 2012; Hammond et al . , 2016; Huang et al . , 2015; Richet et al . , 2015 ) . Together , the evidence suggests that CAF-1 induces a unique mechanism of H3/H4 oligomerization . The budding yeast CAF-1 complex consists of subunits Cac1 , Cac2 , and Cac3 ( Figure 1A ) , which differ in their ability to bind H3/H4 . GST-tagged Cac1 , but not the other subunits , can co-immunoprecipitate endogenous human histones when expressed in HeLa cells ( Li et al . , 2008 ) . A single lysine to arginine substitution on H3 residue 56 abolished Cac1 binding , implicating the first alpha helix ( αN ) in this interaction . Biophysical studies with the other subunits show that the H . sapiens and D . melanogaster homologs of Cac3 can bind H3/H4 with varying affinities ( Song et al . , 2008; Murzina et al . , 2008; Nowak et al . , 2011; Schmitges et al . , 2011 ) . However , dissociation constants for Cac1 and Cac2 binding to H3/H4 have not been reported , and neither have the subunit interactions required to induce the 1:2 chaperone:histone dimer stoichiometry . 10 . 7554/eLife . 18023 . 003Figure 1 . The Cac1 subunit is sufficient for ( H3/H4 ) 2 tetramerization . ( A ) Schematic of domains in the individual CAF-1 subunits . ( B ) H3/H4 ( Py ) binding to individual CAF-1 subunits . Fluorescence anisotropy of 25 nM pyrene-labeled H3/H4 was monitored with titration of individually purified Cac1 , Cac2 , or Cac3 in Histone Buffer ( H . B . : 20 mM Tris , 150 mM KCl , 2 mM MgCl2 , 0 . 5 mM TCEP , 1% Glycerol , 0 . 05% BRIJ-35 . ) The CAF-1 complex was titrated into 5 nM H3/H4 ( Py ) . The curves were fitted using Equation 3 . ( C ) A representative EMSA separating histone:DNA species as disomes or tetrasomes . 1 . 6 µM of the indicated histone chaperone or CAF-1 subunit was incubated with 0 . 2 µM H3/H4 ( FM ) dimer , prior to addition of 0 . 4 µM 80 bp DNA . The bar graph shows the mean and standard deviation of fraction of tetrasomes formed , calculated by Equation 5 , from at least three independent experiments . ( D ) FRET of mixed labeled H3/H4 ( CPM/FM ) . Spectra were obtained for 10 nM of labeled histones incubated with 0 . 2 µM CAF-1 or DNA , or 1 µM CAF-1 subunit . The FRET Effect was calculated using Equation 4 from at least three independent experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 18023 . 00310 . 7554/eLife . 18023 . 004Figure 1—figure supplement 1 . Purified proteins used in this study . Coomassie-stained SDS PAGE of the ( A ) individual , purified CAF-1 subunits , Cac1 truncations , and ( B ) CAF-1 and CAF-1-H3/H4 complexes cross-linked by DSS or EDC . DOI: http://dx . doi . org/10 . 7554/eLife . 18023 . 004 The architecture of the CAF-1 and CAF-1-H3/H4 complexes is not known in detail . Some biochemical evidence points to specific interactions between Cac1 and Cac2 , and between Cac1 and Cac3 . Substitution of Cac1 leucine 276 to proline blocks Cac3 binding ( Krawitz et al . , 2002 ) , and GST-tagged Cac1 residues 215–429 are sufficient to bind Cac3 in vitro . In addition , the last one-third of the Cac1 human homolog , p150 , is sufficient to bind Cac2 ( Kaufman et al . , 1995 ) . Whether these interactions are critical for H3/H4 binding and oligomerization is also unknown . In this study , we investigate the CAF-1 and CAF-1-H3/H4 complexes using biophysical , structural , and functional validation approaches to gain insight into the mechanism of CAF-1 subunit interactions that promote H3/H4 tetramerization . The data indicate that Cac1 is the major subunit required and is sufficient for binding and tetramerizing H3/H4 . Mass spectrometric analyses using both hydrogen/deuterium exchange and chemical cross-linking reveal H3/H4-induced conformational changes , as well as extensive cross-links between the C-terminus of Cac1 with H3 and H4 . Although structural and biophysical analyses reveal a homo-dimerization motif at the Cac1 C-terminus , this is not sufficient for H3/H4 tetramerization . Rather , the Cac1 glutamate/aspartate ( ED ) domain is also required to tetramerize H3/H4 in vitro . These results support an architectural model for the replication-coupled CAF-1-H3/H4 complex , in which Cac1 scaffolds all protein-protein interactions and is predominantly involved in H3/H4 tetramerization . To measure the contribution of each individual CAF-1 subunit in histone binding , the individual subunits were purified for use in fluorescence anisotropy experiments with pyrene-labeled H3/H4 ( H3/H4 ( Py ) ) ( Figure 1A , B , Figure 1—figure supplement 1 ) . Each subunit was titrated independently into a fixed concentration of H3/H4 ( Py ) , inducing a concentration-dependent increase in pyrene fluorescence anisotropy that enabled binding constants to be measured ( Table 1 ) . Consistent with prior experiments , CAF-1 binds to H3/H4 ( Py ) with low nanomolar affinity ( KDapp = 5 . 3 nM ) ( Winkler et al . , 2012b; Liu et al . , 2012 ) . Although all CAF-1 subunits increased the anisotropy of H3/H4 ( Py ) at high ( 1 µM ) concentrations , only Cac1 achieved saturable binding with sub-micromolar affinity for H3/H4 ( KDapp = 97 nM ) . Cac2 and Cac3 titrations , on the other hand , did not reach saturation and displayed weak affinity for H3/H4 . These results suggest that Cac1 contributes substantially to H3/H4 binding , and the other subunits provide accessory interactions to promote a high-affinity CAF-1-H3/H4 complex . 10 . 7554/eLife . 18023 . 005Table 1 . KD values from pyrene fluorescence anisotropy of CAF-1 subunits and H3/H4 . DOI: http://dx . doi . org/10 . 7554/eLife . 18023 . 005Pyrene-labeled proteinBinding partnerKD or KDapp ( M ) H3/H4CAF-15 . 3 ± 0 . 9 × 10−9H3/H4Cac19 . 7 ± 1 . 8 × 10−8H3/H4Cac2n . c . H3/H4Cac3n . c . Cac1386Cac13862 . 6 ± 0 . 2 × 10−8Cac1454Cac14542 . 5 ± 0 . 2 × 10−8Cac1386H3/H42 . 1 ± 0 . 5 × 10−7Cac1454H3/H4n . c . Cac1386Cac21 . 3 ± 0 . 4 × 10−6n . c . not calculated Recent reports showed that CAF-1 not only binds to H3/H4 , but arranges the histones into a nucleosomal ( H3/H4 ) 2 tetramer ( Winkler et al . , 2012b; Liu et al . , 2012 ) . Since all CAF-1 subunits bind H3/H4 , albeit very weakly for Cac2 and Cac3 , potentially one or more of these subunits in combination might be required to form tetrasomes ( ( H3/H4 ) 2 tetramer bound to DNA ) . To test this possibility , we utilized an EMSA capable of resolving different stoichiometric species of fluorescein-labeled H3/H4 ( H3/H4 ( FM ) ) bound to 80 base pair nucleosome-positioning DNA ( Donham et al . , 2011 ) . Previously , this method revealed that Asf1 favors deposition of H3/H4 ( FM ) as a disome species , which consists of an H3/H4 dimer bound to DNA ( Donham et al . , 2011; Scorgie et al . , 2012 ) , whereas CAF-1 favors formation of tetrasomes ( Liu et al . , 2012 ) . Surprisingly , Cac1 alone induces tetrasome formation ( Figure 1C ) . In contrast , Cac3 does not alter the basal fraction of tetrasomes formed . Cac2 induces more disome species , but the pre-formed Cac1/Cac2 complex deposits tetrasomes , suggesting that the Cac1 subunit promotes tetrasome assembly . To confirm that Cac1-induced tetrasome formation was a direct result of ( H3/H4 ) 2 tetramerization independent of DNA , we used a mixed-fluorophore Förster resonance energy transfer ( FRET ) assay with equimolar concentrations of fluorescein ( FM ) -labeled and 7-Diethylamino-3- ( 4’-Maleimidylphenyl ) -4-Methylcoumarin ( CPM ) -labeled H3/H4 dimers ( H3/H4 ( CPM/FM ) ) . As we previously reported , H3/H4 ( CPM/FM ) exhibits CPM to FM FRET in the presence of DNA or CAF-1 ( Figure 1D ) , but not in the presence of Asf1 , which binds to a dimer of H3/H4 ( Liu et al . , 2012 ) . Neither Cac2 nor Cac3 induced FRET , but Cac1 increased the FRET signal of H3/H4 ( CPM/FM ) to a similar extent as CAF-1 or DNA . Together , these data reveal that Cac1 is sufficient for ( H3/H4 ) 2 tetramerization . To elucidate potential interaction regions between CAF-1 and H3/H4 , we subjected the CAF-1 and CAF-1-H3/H4 complexes to hydrogen/deuterium exchange ( HX ) at 10°C , followed by liquid chromatography-mass spectrometry ( LC-MS/MS ) analysis ( Hoofnagle et al . , 2003 ) . Hydrogen exchange was carried out for 30 and 60 min , followed by acid quench , pepsin digestion and mass analysis . The parent spectra ( MS1 ) were used to compute the number of deuterons incorporated per peptide , and the difference in deuteron uptake between samples was used to calculate changes in HX between bound and unbound states ( Figure 2A; Supplementary file 1A ) . Without available structural data for the CAF-1 proteins , we used the PHYRE2 server ( Kelley et al . , 2015 ) to obtain high-scoring Cac2 and Cac3 homology models for interpretation of HX ( Figure 2B ) . 10 . 7554/eLife . 18023 . 006Figure 2 . Hydrogen/deuterium exchange of CAF-1 , and CAF-1-H3/H4 complexes . ( A ) The sequences of the three individual CAF-1 subunits are shown . Each bar represents an individual identical peptide observed in the protein between the compared samples , plotted as the difference in deuteron uptake between the CAF-1 and CAF-1-H3/H4 samples ( i . e . , difference = CAF-1-H3/H4 – CAF-1 only ) . The differences in deuteron uptake at 60’ are colored according to the legend . The 'cooler' colors ( green , blue , and purple ) represent an increase in apparent protection for the peptide in CAF-1-H3/H4 compared to the CAF-1 sample , whereas the 'warmer' colors ( orange , yellow , and red ) represent decreased apparent protection . Peptide coverage was approximately 60% , 80% , and 80% for Cac1 , Cac2 and Cac3 , respectively . ( B ) Differences in HX at 60’ were mapped on PHYRE2 models of Cac2 and Cac3 . The coloring scheme is the same as for A , but amino acids with no coverage are colored dark gray to distinguish these residues from those that have coverage but did not exchange significantly . ( C ) The top panel shows five fold serial dilution analysis of strain CFY53 ( cac1 ) with the vector pCac1 introduced that was either empty , expressed wild type Cac1 or Cac1 with the indicated amino acid changes . The bottom shows five fold serial dilution analysis of strain CFY54 ( cac2 ) with the vector pCac2 introduced that was either empty , expressed wild type Cac2 or Cac2 with the indicated amino acid changes . DOI: http://dx . doi . org/10 . 7554/eLife . 18023 . 00610 . 7554/eLife . 18023 . 007Figure 2—figure supplement 1 . Peptide coverage in HX-MS . Coverage maps of Cac1 , Cac2 , and Cac3 , from side-to-side comparisons between CAF-1 and CAF-1-H3/H4 samples in the HX study . All identical peptides between multiple samples were calculated for differences in deuteron uptake at 60’ and colored accordingly , using the same scheme as in Figure 2B . DOI: http://dx . doi . org/10 . 7554/eLife . 18023 . 007 Several regions of Cac1 show changes in HX upon binding H3/H4 . In forming the CAF-1-H3/H4 complex , the C-terminal half of Cac1 showed both increased HX ( amino acids 550–591 ) and decreased HX ( amino acids 304–322 , 340–360 , and 463–473 ) ( Figure 2A , Figure 2—figure supplement 1 ) . The regions with decreased HX were candidates for potential protein-protein interactions between Cac1 and H3/H4 ( Percy et al . , 2012 ) . To evaluate the significance of these HX changes at the Cac1 C-terminus in vivo , we examined the effect of deleting the corresponding amino acids on CAF-1 function in Saccharomyces cerevisiae . Using yeast deleted for CAC1 , we introduced either empty vector , a vector expressing wild type Cac1 or mutant Cac1 with a series of deletions and substitutions ( Tables 2 and 3; Supplementary file 1B ) . CAF-1 is known to be required for resistance to DNA damaging agents that cause DNA double-strand breaks ( Linger and Tyler , 2005 ) , due to its role in chromatin assembly during DNA repair . Yeast lacking Rad52 were included to provide a control showing extreme sensitivity to DNA double-strand damage . Rad52 is essential for homologous recombination , the central repair mechanism used in yeast . Whereas cac1 yeast were sensitive to the radiomimetic zeocin , re-introduction of the vector expressing wild type CAC1 rescued cells from DNA damage ( Figure 2C ) . Of the strains that expressed Cac1 mutant proteins , the DNA damage resistance afforded by Cac1 was noticeably compromised by the 304–322 , 463–473 and 578–580 deletions in Cac1 ( Figure 2C ) . These deletions did not significantly affect the expression levels of the Cac1 protein ( Table 3 and data not shown ) . We conclude that the decreased HX in these regions upon H3/H4 binding likely reflects their importance for the physiological complex . 10 . 7554/eLife . 18023 . 008Table 2 . Yeast strains and plasmids . DOI: http://dx . doi . org/10 . 7554/eLife . 18023 . 008StrainMutationGenotypeReferencew1588-4aWTMat alpha; leu2-3 , 112; ade2-1; can1-100; his3-11 , 15; ura3-1; trp1-1; RAD5Gift from R . RothsteinCFY53cac1ΔMat alpha; leu2-3 , 112; ade2-1; can1-100; his3-11 , 15; ura3-1; trp1-1; RAD5 cac1Δ::NATThis studyCFY54cac2ΔMat alpha; leu2-3 , 112; ade2-1; can1-100; his3-11 , 15; ura3-1; trp1-1; RAD5 cac2Δ::NATThis studyCFY58cac3ΔMat alpha; leu2-3 , 112; ade2-1; can1-100; his3-11 , 15; ura3-1; trp1-1; RAD5 cac3Δ::NATThis studyJKT004rad52ΔMAT a rad52::TRP1; trp1-1; ura3-1; can1-100; ADE; bar1::LEU2; his3-11; GALRamey et al . ( 2004 ) PlasmidCharacteristicsReferencepRS315 ( EV ) CEN6 ARSH4 LEU2Sikorski and Hieter ( 1989 ) pCac1pRS315-Cac1This studypCac2pRS315-Cac3This studypCac3pRS315-Cac3This studypCac1Δ233-237pRS315-Cac1 aa 233-237 deletedThis studypCac1Δ280-284pRS315-Cac1 aa 280 to 284 deletedThis studypCac1Δ304-322pRS315-Cac1 aa 304 to 322 deletedThis studypCac1Δ340-360pRS315-Cac1 aa 340-360 deletedThis studypCac1Δ428-432pRS315-Cac1 aa 428-432 deletedThis studypCac1K442E/R443E/K444EpRS315-Cac1 with the mutation K442E/R443E/K444EThis studypCac1Δ463-473pRS315-Cac1 aa 463 to 473 deletedThis studypCac1Δ497-501pRS315-Cac1 aa 497 to 501 deletedThis studypCac1Δ574-584pRS315-Cac1 aa 574-584 deletedThis studypCac1Δ578-580pRS315-Cac1 aa 578 to 580 deletedThis studypCac1Δ576-606pRS315-Cac1 aa 576-606 deletedThis studypCac1Δ578-580pRS315-Cac1 aa 578 to 580 deletedThis studypCac2Δ1-15pRS315-Cac2 aa 1 to 15 deletedThis studypCac2E70KpRS315-Cac2 with the mutation E70KThis studypCac2D91K/D92KpRS315-Cac2 with the mutation D91K/D92KThis studypCac2S206A/A207GpRS315-Cac2 with the mutation S206A/A207GThis studypCac2V273A/P275A/S276A/G277ApRS315-Cac2 with the mutation V273A/P275A/S276A/G277AThis studypCac2I274A/S276ApRS315-Cac2 with the mutation I274A/S276AThis studypCac2D248K/E285KpRS315-Cac2 with the mutation D248K/E285KThis studypCac2R295EpRS315-Cac2 with the mutation R295EThis studypCac2K306A/N307A/R308ApRS315-Cac2 with the mutation K306A/N307A/R308AThis studypCac2L316A/K318ApRS315-Cac2 with the mutation L316A/K318AThis studypCac2L316E/K318EpRS315-Cac2 with the mutation L316E/K318EThis studypCac2Δ371-373pRS315-Cac2 aa 371 to 373 deletedThis studypCac2M417A/H418A/E420ApRS315-Cac2 with the mutation M417A/H418A/E420AThis studypCac2Δ425-468pRS315-Cac2 aa 425-468 deletedThis studypCac2Δ445-468pRS315-Cac2 aa 445-468 deletedThis studypCac2K447E/K448EpRS315-Cac2 with the mutation K447E/K448EThis studypCac3K284A/K285A/E286ApRS315-Cac3 with the mutation K284A/K285A/E286AThis studypCac3Δ306-309pRS315-Cac3 deleted aa 306 to 309This studypCac3Δ287-290pRS315-Cac3 deleted aa 287 to 290This study10 . 7554/eLife . 18023 . 009Table 3 . Yeast mutants and phenotypes observed . DOI: http://dx . doi . org/10 . 7554/eLife . 18023 . 009MutantRationale for mutantZeocin resistanceProtein expressionCac1Δ233-237Cross-link to Cac3 ( Figure 3B ) SensitiveNoCac1Δ280-284Cross-link to Cac3 ( Figure 3A , B ) SensitiveNoCac1Δ304-322HX change with H3/H4 and cross-link to Cac2 ( Figure 2A , 3B ) Very sensitiveYesCac1Δ340-360HX change with H3/H4 ( Figure 2A ) SensitiveNoCac1Δ428-432In ED-rich Region ( Figure 1A ) Not sensitiveYesCac1K442E/K443E/K444ECross-link to H3 ( Figure 3B ) Not sensitiveYesCac1Δ463-473HX change and cross-link to H3/H4 ( Figure 2A , 3B ) SensitiveYesCac1Δ497-501Cross-link to Cac2 ( Figure 3B ) Little sensitiveYesCac1Δ574-584HX change and cross-link to H3/H4 ( Figure 2A , 3B ) Little sensitiveYesCac1Δ578-580HX change and cross-link to H3/H4 ( Figure 2A , 3B ) SensitiveYesCac1Δ575-606HX change and cross-link to H3/H4 ( Figure 2A , 3B ) Very sensitiveNoCac2Δ1-15HX change with H3/H4 ( Figure 2A ) Very sensitiveYesCac2E70KLoop next to Cac2 N-terminal loop ( Figure 2A ) Not sensitiveYesCac2D91K/D92KCross-link to Cac1 ( Figure 3A ) SensitiveYesCac2S206A/A207GHX change with H3/H4 ( Figure 2A ) SensitiveYesCac2V273A/P275A/S276A/G277AHX change with H3/H4 ( Figure 2A ) SensitiveYesCac2I274A/S276AHX change with H3/H4 ( Figure 2A ) Not sensitiveYesCac2D284K/E285KCross-link to Cac1 ( Figure 3A ) Not sensitiveYesCac2R295ELoop between Cac2 blades 5 and 6 ( Figure 2A ) SensitiveNoCac2K306A/N307A/R308AHX change with H3/H4 ( Figure 2A ) Not sensitiveYesCac2L316A/K318AHX change with H3/H4 ( Figure 2A ) SensitiveYesCac2L316E/K318EHX change with H3/H4 ( Figure 2A ) SensitiveYesCac2Δ371-373Loop next to Cac2 N-term loop and blade 6 ( Figure 2A ) SensitiveYesCac2M417/H418A/E420AC-terminal loop in Cac2Not sensitiveYesCac2Δ425-468HX change with H3/H4 ( Figure 2A ) Not sensitiveYesCac2Δ445-468C-terminal loop in Cac2Not sensitiveYesCac3K284A/K285A/E286ACross-link to Cac1 ( Figure 3B ) Not sensitiveYesCac3Δ287-290Cross-link to Cac1 ( Figure 3B ) SensitiveYesCac3Δ306-309Cross-link to Cac1 ( Figure 3A ) SensitiveYes Several predicted loops in Cac2 also show changes in HX upon H3/H4 association . The Cac2 subunit is confidently homology-modeled as a WD-repeat β propeller structure , with an intrinsically disordered C-terminus ( not shown ) . The HX data reveal that the N-terminal loop and propeller blades 5 and 6 are protected from exchange following H3/H4 binding ( Figure 2A and B ) . To determine whether these amino acids also influence the physiological response to DNA damage in yeast , we designed mutations ( Tables 2 and 3; and Supplementary file 1B ) in the loop regions , including the N-terminal loop ( amino acids 7–17 ) , a loop in blade 5 ( amino acids 206–207 ) , two loops in blade 6 ( amino acids 273/275/276/277 and 316/318 ) , and the disordered C-terminus ( amino acids 425–468 ) , in order to minimize disruption of the Cac2 structure . Vectors expressing wild type Cac2 or Cac2 bearing mutations within these regions were introduced into the cac2 strain . Only the Δ1–15 , V273A/P275A/S276A/G277A , L316E/L318E , and Δ425–468 mutations displayed various degrees of sensitivity to zeocin-induced DNA damage ( Figure 2C ) . These mutations did not significantly alter the expression levels of Cac2 . From these data , we conclude that Cac2 regions at the N-terminus and in blade 6 identified in the HX analysis are functionally important in vivo . Cac3 exhibits little change in HX with H3/H4 association . Cac3 was modeled from the human homolog RbAp46 , which is also a WD-repeat β-propeller structure . This Cac3 model also contains an N-terminal alpha helix , which is used by RbAp46 for binding the N-terminal helix of histone H4 ( Murzina et al . , 2008 ) . The HX data show no significant changes in deuteration in Cac3 when the CAF-1 complex is compared to CAF-1-H3/H4 ( Figure 2A and B ) . Taken together , the HX and in vivo results indicate that H3/H4 binding to CAF-1 primarily promotes HX changes to Cac1 and Cac2 , but not Cac3 , in regions that are important for CAF-1 function in vivo . Cac1 is sufficient to tetramerize H3/H4 and also appears to act as a scaffold for the other CAF-1 subunits . However , the HX experiment reports Cac1 regions with deuteration changes that could be due either to direct protein-protein interactions , or indirect effects resulting from allosteric conformational changes or conformational dynamics . To provide more direct information about the physical interactions that shape CAF-1-H3/H4 architecture , we used chemical cross-linking with DSS ( disuccinimidyl suberate ) or EDC ( 1-ethyl-3- ( 3-dimethylaminopropyl ) carbodiimide hydrochloride ) coupled to mass spectrometry ( CX-MS ) ( Walzthoeni et al . , 2013 ) . Identification of linked residues can indicate regions involved in protein-protein interactions , and possibly identify tertiary and quaternary interactions . To identify cross-linked peptides , we digested the CAF-1 or CAF-1-H3/H4 complexes with trypsin , trypsin and GluC , or trypsin and LysC , followed by LC-MS/MS . We used the Protein Prospector package ( Chu et al . , 2010 ) from UCSF to identify cross-linked peptides . The collective data from the searches enabled us to create a map of linkages for the CAF-1 subunits and H3/H4 ( Figure 3A and B; Supplementary file 1C ) . In addition to inter-subunit cross-links , many intra-protein cross-links were observed ( Figure 3 , Figure 3—figure supplement 1 ) . Such cross-links are especially prevalent in Cac1 within the first fifty amino acids , residues 118–334 , and residues 460–593 . These may reveal folded regions in Cac1 , since unfolded , flexible domains likely cannot make stable cross-links ( Leitner et al . , 2010 ) . 10 . 7554/eLife . 18023 . 010Figure 3 . Chemical cross-linking of CAF-1 and CAF-1-H3/H4 complexes . ( A ) CAF-1 or ( B ) CAF-1-H3/H4 complexes were covalently cross-linked with DSS or EDC , then digested and run on an LTQ-Orbitrap . Cross-linked peptides were analyzed using Protein Prospector . The primary sequences are depicted in gray bars , with each gray circle marking 50 amino acid segments . DSS cross-links are shown in purple and EDC cross-links are in red . DSS leaves a 11 . 4 Å spacer arm between covalently-linked amine groups . EDC treatment results in a zero length cross-link between amine and carboxyl groups . The inter-subunit cross-links are represented as solid lines and cross-links to H3 and H4 are shown as dotted lines . ( C ) Analysis of Cac3 mutants in yeast . Cac3 mutants were subjected to zeocin-induced DNA damage response in vivo . The panel shows five fold serial dilution analysis of strain CFY58 ( cac3 ) with the vector pCac3 introduced that was either empty ( EV ) , expressed wild type Cac1 , or Cac1 with the indicated amino acid changes . DOI: http://dx . doi . org/10 . 7554/eLife . 18023 . 01010 . 7554/eLife . 18023 . 011Figure 3—figure supplement 1 . Intra-Cac1 cross-links . ( A ) DSS ( purple ) and EDC ( red ) cross-links detected within the Cac1 protein in the CAF-1 complex and ( B ) CAF-1-H3/H4 complex . DOI: http://dx . doi . org/10 . 7554/eLife . 18023 . 011 Within the CAF-1 and CAF-1-H3/H4 complexes , cross-links were observed between subunits Cac1 and Cac2 , Cac1 and Cac3 , but not between Cac2 and Cac3 . Cross-links between the Cac1 C-terminus to Cac2 ( Figure 3A and B ) are consistent with the last third of Cac1 binding to Cac2 ( Kaufman et al . , 1995 ) . In addition , cross-links from Cac2 to the middle domain of Cac1 were also observed , suggesting that modular interactions scaffold Cac1-Cac2 binding . The Cac1 cross-links to Cac3 ( Cac1K235 to Cac3K287 and Cac1K282 to Cac3K307 ) ( Figure 3A and B ) complement a prior observation in which a Cac1 L276P mutation abrogates binding to Cac3 ( Krawitz et al . , 2002 ) . To determine whether these Cac3 residues influence CAF-1 function in vivo , we deleted the CAC3 gene in S . cerevisiae , and introduced either an empty vector , a vector expressing wild type Cac3 or Cac3 bearing deletions ( ∆306–309 or ∆287–290 ) in predicted loops ( Figure 3C and Table 2 ) . Although wild type CAC3 rescued the DNA damage phenotype resulting from zeocin treatment , the deletions both failed to do so . This result suggests that these Cac3 residues are important for CAF-1 structure and function . Many cross-links were observed between CAF-1 and H3 . Multiple residues within the Cac1 C-terminus ( 577–593 ) are proximal to both the H3 N-terminal tail lysines ( 4 and 18 ) and a C-terminal lysine 122 . The predicted N-terminal helix of Cac3 also cross-links to the H3 N-terminus . Notably , Cac1 lysines 444 and 577 cross-link to the αN and α1 helices of H3 ( Cac1K444 to H3K64 , Cac1K577 to H3K64 , and Cac1K444 to H3E59 ) ( Figure 3B ) . These residues are close in space to H3-K56 and reinforce reports that mutation or acetylation of H3-K56 alters binding to Cac1 ( Winkler et al . , 2012b; Li et al . , 2008 ) . The C-terminal half of Cac1 is also involved in cross-linking to histone H4 . Cac1 residues 577–588 cross-link to the H4 N-terminal tail and to the core through H4K79 . K79 is immediately adjacent to the H4 α2 helix , which also cross-links to other Cac1 residues ( H4E53 to Cac1K317 and H4K59 to Cac1E464 ) ( Figure 3B ) . Importantly , the Cac1 residues involved in histone cross-linking are located within peptides that exhibit significant changes in HX upon H3/H4 binding ( Figure 2A ) . Notably , these peptides include residues 317 and 464 , the deletions of which compromise CAF-1-dependent resistance to DNA damage in vivo ( Figure 2C ) . Thus , the C-terminal half of Cac1 is involved in cross-links to both the tail and core domains of H3 and H4 . Although chemical capture experiments generate useful models for protein-protein interactions , the presence or absence of captured cross-links , or abundance of cross-links , are not indicative of equilibrium binding constants , which should be measured through biophysical experiments . As the Cac1 C-terminus has extensive cross-links with H3/H4 , we reasoned that this entire region might be responsible for H3/H4 binding and tetramerization . To test this hypothesis and identify such a minimal region , serial N-terminal deletions were designed , which included the ED domain through the C-terminus . Truncated proteins beginning at residues 386 ( Cac1386 ) , 421 ( Cac1421 ) , 454 ( Cac1454 ) , and 457 ( Cac1457 ) were expressed and purified from E . coli . Cac1386 , Cac1421 , and Cac1457 were compared for the ability to induce histone tetramerization in the H3/H4 ( FM ) DNA deposition EMSA , as well as in the H3/H4 ( CPM/FM ) FRET assay ( Figure 4A , B , Figure 4—figure supplement 1 ) . Both biophysical methods reveal that amino acids 386–606 within Cac1 are competent for histone tetramerization , whereas Cac1421 and Cac1457 are not . These results suggest that Cac1 residues 386–457 – which overlap much of the ED domain – may be responsible for a significant part of the binding interactions between Cac1 and H3/H4 . To test this , we used fluorescence anisotropy with pyrene-labeled Cac1386 and Cac1454 ( Cac1386 ( Py ) and Cac1454 ( Py ) , respectively ) to monitor pyrene anisotropy changes with H3/H4 binding ( Figure 4C ) . In this assay , Cac1386 ( Py ) binds H3/H4 ( KDapp = 210 nM; Table 1 ) with a slightly weaker affinity than full-length Cac1 binding to H3/H4 ( Py ) ( KDapp = 97 nM ) . Cac1454 ( Py ) binding to H3/H4 , on the other hand , was significantly weaker ( Table 1 ) . Together , these data reveal that Cac1386 binds H3/H4 and is sufficient to promote histone tetramerization . 10 . 7554/eLife . 18023 . 012Figure 4 . The C-terminus of Cac1 binds and tetramerizes H3/H4 . ( A ) EMSA evaluating tetrasome formation by Cac1 N-terminal truncations Cac1386 , Cac1421 and Cac1457 in H . B . The graph shows the mean and standard deviation from at least three independent experiments . Arrows point to complexes of DNA bound to H3/H4 dimers ( D ) or tetramers ( T ) , respectively . ( B ) Change in FRET Effect of H3/H4 ( CPM/FM ) induced by 2 µM Cac1386 , Cac1421 or Cac1457 . The Cac1 spectrum is included from Figure 2B for reference . ( C ) Fluorescence anisotropy of Cac1386 ( Py ) or Cac1454 ( Py ) titrated with H3/H4 in H . B . The schematic indicates two labeled residues on Cac1386 ( cysteines 440 and 454 ) , and one on Cac1454 . DOI: http://dx . doi . org/10 . 7554/eLife . 18023 . 01210 . 7554/eLife . 18023 . 013Figure 4—figure supplement 1 . Histone deposition assay of Cac1 truncations in Minimal Buffer ( M . B . ) . 1 . 6 µM of each Cac1 truncation was incubated with 0 . 2 µM H3/H4FM , then allowed to interact with 0 . 4 µM 80 bp DNA . The EMSA ( upper panel ) is representative of at least four independent experiments that were used for comparisons in the bar graph ( lower panel ) . DOI: http://dx . doi . org/10 . 7554/eLife . 18023 . 013 Although intact CAF-1 does not dimerize during the H3/H4 tetramerization process ( Liu et al . , 2012; Winkler et al . , 2012b ) , previous work found that the H . sapiens and X . laevis homologs of the Cac1 subunit alone can dimerize in vivo through a sequence located C-terminal to the conserved ED domain ( Quivy et al . , 2001; Gérard et al . , 2006 ) . Although this putative dimerization sequence is apparently not conserved in yeast , purified S . cerevisiae Cac1 elutes from size exclusion chromatography at a molecular weight consistent with a dimer , confirmed by multiangle light scattering ( Winkler et al . , 2012b ) . To gain structural insight into the putative Cac1 dimer , we crystallized E . coli-expressed Cac1457 , a region that forms many stable intra-Cac1 chemical cross-links ( Figure 3—figure supplement 1 ) . This region also remains as a stable proteolytic fragment after expression and purification of Cac1 from baculovirus-infected Sf9 cells ( Figure 5—figure supplement 1 ) . The structure was determined using molecular replacement to a resolution of 2 . 9 Å ( Table 4 ) with coordinates from a recently determined structure of Cac1C ( 522–600 ) ( Zhang et al . , 2016 ) . Even though the entire 457–606 protein was present in the crystal ( Figure 5A and data not shown ) , only amino acids 520–600 were visible in our electron density , and the 'wing' ( 576–581 ) was disordered . As expected , the structure we determined is virtually identical to the reported Cac1C winged helix ( WH ) domain , with an overall root mean squared deviation ( r . m . s . d . ) of 1 . 14 Å for all atoms ( Zhang et al . , 2016 ) . Indeed , we observed a crystallographic homo-dimer . Amino acids 572–600 – corresponding to the last beta strand , loop , and alpha helix – form a symmetrical head-to-tail homo-dimerization interface ( Figure 5A—figure supplement 2 and Supplementary file 1D ) that encompasses 651 Å2 of solvent-inaccessible surface area . 10 . 7554/eLife . 18023 . 014Figure 5 . The Cac1 C-terminal winged helix ( WH ) domain can form a homodimer . ( A ) Crystal structure of amino acids 520–600 at a resolution of 2 . 9 Å ( PDB ID 5JBM ) , shown as two crystallographically related monomers colored separately ( light gray and dark gray ) . The inset shows major interacting residues buried in half of the homodimer interface , which is arranged in a head-to-tail symmetry with identical interactions on both halves . ( B ) Homo-dimerization of the Cac1 C-terminus quantified by titrating unlabeled Cac1386 or Cac1454 into 10 nM of labeled Cac1386 ( Py ) or Cac1454 ( Py ) , respectively . The pyrene anisotropy of Cac1386 ( Py ) or Cac1454 ( Py ) increases in Minimal Buffer ( M . B . : 20 mM HEPES , 150 mM NaCl , 1 mM DTT , pH 7 . 5 ) , but homo-dimerization does not occur in H . B . ( C ) Binding affinity of the Cac1386 ( Py ) -Cac2 interaction . Pyrene fluorescence anisotropy of 10 nM Cac1386 ( Py ) titrated with increasing concentration of Cac2 in Histone Buffer ( H . B . ) . The KD was determined to be 1 . 3 µM ( Table 1 ) . ( D ) Pyrene fluorescence spectra of Cac1386 ( Py ) alone , Cac1454 ( Py ) alone , and Cac1386 ( Py ) bound to 2 µM H3/H4 or 13 µM Cac2 . The excimer band that peaks at 465 nm is indicated . DOI: http://dx . doi . org/10 . 7554/eLife . 18023 . 01410 . 7554/eLife . 18023 . 015Figure 5—figure supplement 1 . Purification of full-length Cac1 and Cac1457 from baculovirus-infected Sf9 cells . ( A ) Cac1 elutes from a 120 mL Sephadex 200 column in 3 peaks . ( B ) Western blotting for the Strep II epitope present on the Cac1 C-terminus . Peak 1 is full-length Cac1 , whereas Peak 3 is truncated from the N-terminus . ( C ) MALDI identified the C-terminal regions as residues 457–606 ( expected mass 18285 . 3; observed mass 18276 . 4 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 18023 . 01510 . 7554/eLife . 18023 . 016Figure 5—figure supplement 2 . Structural analysis of the Cac1457 WH domain . ( A ) The Cac1 WH domain monomer and one of the symmetry mates are depicted in both a ribbon and surface representation , in two orientations . The Cac1 monomers are colored light gray and dark gray , respectively , with the electrostatic potential shown mapped onto the surface , colored from red to blue , indicating negatively charged to positively charged regions . ( B ) The Cac1 WH domain monomer is depicted in both a ribbon and surface representation . The putative dimerization interface faces to the right . HX changes are colored in orange to represent an increase in HX with H3/H4 bound to CAF-1 ( Figure 3 ) . Amino acids that cross-link to H3/H4 are labeled and colored in red . DOI: http://dx . doi . org/10 . 7554/eLife . 18023 . 01610 . 7554/eLife . 18023 . 017Table 4 . Data collection and refinement statistics . DOI: http://dx . doi . org/10 . 7554/eLife . 18023 . 017Wavelength1 . 0 ÅResolution range – data collection29 . 43–2 . 91 ( 3 . 01–2 . 91 ) Space groupP 41 2 2Unit cell ( Å ) ( deg ) 58 . 850 58 . 830 97 . 929 90 90 90Total reflections26 , 419 ( 5001 ) Unique reflections4117 ( 393 ) Multiplicity6 . 42 ( 6 . 63 ) Completeness ( % ) 99 . 3 ( 99 . 7 ) Mean I/sigma ( I ) 12 . 9 ( 1 . 7 ) Wilson B-factor92 . 56R-meas0 . 099 ( 0 . 557 ) Resolution range - refinement29 . 43–3 . 00 ( 3 . 107–3 . 00 ) Reflections used in refinement3761 ( 365 ) Reflections used for R-free360 ( 42 ) R-work0 . 233 ( 0 . 408 ) R-free0 . 275 ( 0 . 324 ) Number of non-hydrogen atoms654 Macromolecules653 Protein residues81RMS ( bonds ) 0 . 007 ÅRMS ( angles ) 0 . 93 degRamachandran favored ( % ) 88Ramachandran allowed ( % ) 12Ramachandran outliers ( % ) 0Rotamer outliers ( % ) 4 . 3Clashscore6 . 85Average B-factor48 . 8Number of TLS groups3Statistics for the highest-resolution shell are shown in parentheses . Friedel mates were averaged when calculating data collection statistics . In order to examine dimerization of the Cac1 C-terminus in solution , we monitored the fluorescence anisotropy of Cac1386 ( Py ) or Cac1454 ( Py ) with titration of the same unlabeled Cac1 . In a minimal buffer ( M . B . ) , titration of unlabeled Cac1386 into Cac1386 ( Py ) increased anisotropy with a dissociation constant of 26 nM ( Figure 5B and Table 1 ) . Likewise , unlabeled Cac1454 titrated into Cac1454 ( Py ) increased pyrene anisotropy with a similar equilibrium constant ( KD = 25 nM ) . The results are consistent with Cac1 forming a dimer through the C-terminus . However , we also found that the Cac1386 ( Py ) - Cac1386 and Cac1454 ( Py ) - Cac1454 dimers are destabilized by the histone buffer ( H . B . ) that is typically used in H3/H4 equilibrium binding experiments ( Donham et al . , 2011; Scorgie et al . , 2012; Karantza et al . , 1996; Banks and Gloss , 2004; Winkler et al . , 2012a; Winkler et al . , 2012b ) . Importantly , our histone tetramerization assays performed in H . B . ( Figure 4A and B , Figure 4—figure supplement 1 ) indicate that monomeric Cac1386 is sufficient to tetramerize ( H3/H4 ) 2 but Cac1454 is not , even though both harbor the putative dimerization domain . Therefore , Cac1 dimerization is not necessary to promote ( H3/H4 ) 2 tetramerization , consistent with the functional stoichiometry of CAF-1 . In addition to H3/H4 binding , the Cac1 C-terminus has been reported to bind to DNA with a dissociation constant of ~2 µM ( Zhang et al . , 2016 ) and found here to possess many cross-links to Cac2 ( Figure 3A and B ) . We also found that Cac1386 ( Py ) directly binds Cac2 with a KD of 1 . 3 µM in H . B . ( Figure 5C and Table 1 ) . The weak binding of the Cac1 C-terminus to multiple partners , along with the histone-induced HX changes and simultaneous cross-links to both H3/H4 and Cac2 , suggests that this region has binding and conformational plasticity . The HX , CX , and anisotropy measurements of Cac1 with H3/H4 ( Figures 2A , 3B and 4C ) indicate that structural changes take place C-terminal to the Cac1 ED domain . To examine this possibility , we used pyrene as a photophysical probe for conformational changes . Cac1386 ( Py ) harbors two pyrene-labeled cysteines ( residues 440 and 454 ) in this region ( Figure 4C ) , whereas Cac1454 ( Py ) only has one . Two pyrenes that are within 10 Å exhibit a characteristic 'excimer' band in the 465 nm region of the fluorescence spectrum ( Birks et al . , 1963 ) . This band was observed for monomeric Cac1386 ( Py ) in H . B . but not for Cac1454 ( Py ) , as expected ( Figure 5D ) . To detect conformational changes in the vicinity of these cysteines , we monitored the Cac1386 ( Py ) excimer band with a saturating concentration of either H3/H4 or Cac2 . We found that binding of H3/H4 removes this excimer band , whereas Cac2 does not change the spectrum . These results indicate that direct binding of H3/H4 , but not Cac2 , promotes a structural change in Cac1 near the ED domain . Our investigation into CAF-1 inter-subunit interactions has revealed a central role for the large Cac1 subunit in CAF-1 function . Previous studies show that both yeast and metazoan homologs of the large subunit directly interact with the other subunits , but no direct interaction between the mid-sized and small subunits has been reported ( Tyler et al . , 2001; Kaufman et al . , 1995 ) . These observations are born out in the chemical cross-linking experiments ( Figure 3A and B ) , which show extensive cross-links between Cac1-Cac2 and Cac1-Cac3 , but not between Cac2-Cac3 . The Cac1 residues ( lysines 235 and 282 ) that cross-link to Cac3 flank a Cac1 L276P mutation that abolished Cac3 binding , and are also located within a region ( residues 215–429 ) known to associate with Cac3 in vitro and in vivo ( Krawitz et al . , 2002 ) . Just N-terminal to the Cac1-Cac3 cross-links , Cac1 harbors the PCNA interacting peptide box ( PIP-box; residues 225–232 ) ( Rolef Ben-Shahar et al . , 2009 ) . Notably , extensive intra-Cac1 cross-links surround these functional sites . Therefore , we refer to residues 118–334 as the 'middle domain' of Cac1 ( Figure 6 , and Supplementary file 1C ) . Our architectural model places the middle domain at the center of all protein-protein interactions within the CAF-1 and CAF-1-H3/H4 complexes , leading to the idea that it coordinates many functions during replication , such as recruitment to the fork through PCNA and stabilizing CAF-1 architecture . The cross-links detected between Cac1 and Cac2 indicate a more extensive binding mode than previously seen . The human Cac1 and Cac2 homologs – p150 and p60 , respectively – require the C-terminal one-third of p150 for binding ( Kaufman et al . , 1995 ) . In addition to Cac2 cross-links at the Cac1 C-terminus , we observed several Cac2 cross-links to the Cac1 N-terminus and middle domain ( Figure 3A and B ) . As Cac1386 ( Py ) binds Cac2 very weakly ( Table 1 and Figure 5C ) , both the N-terminus and middle domain also likely contribute to Cac2 binding . Interestingly , we observed different Cac1-Cac2 cross-links in the presence of H3/H4 , with most occurring in Cac1 regions that also cross-link to H3/H4 . Cac2 undergoes many changes in HX with H3/H4 binding to CAF-1 ( Figure 3A and B ) , some of which coincide with sites of Cac1 cross-links . For example , the C-terminus of Cac2 cross-links just N-terminal to the WH domain , and becomes more accessible to HX with H3/H4 binding ( residues 427–443 , Figures 3A and 2A ) . In contrast , Cac2 residues 284 and 285 cross-link to the WH domain but are among the most protected from HX with H3/H4 ( Figures 2A , 3A and B ) . This suggests that conformational changes in Cac2 take place upon histone binding . C-terminal deletions of human p150 have detrimental effects on p150-mediated chromatin assembly and binding to p60 ( Kaufman et al . , 1995 ) . Our data support the importance of the Cac1 C-terminus for these functions ( Figures 4A , B and 5C ) , as well as in the CAF-1-dependent response to DNA double-strand breaks ( Figure 2C ) . The C-terminus exhibits binding plasticity for context-dependent interactions with Cac2 , H3/H4 , and DNA ( Zhang et al . , 2016 ) ( Figures 4C and 5C ) . The cross-links from the H3/H4 N-terminal tails to the WH domain ( E569 , K577 , D579 , K583 , D588 , and E593 ) ( Figure 3B; Supplementary file 1C ) coincide with Cac1 residues that have been reported to interact with DNA . Mutations to residues in the same positively-charged surface ( K560E , K564E , K568E , R573E and R582E ) had detrimental effects on WH-DNA binding and functional consequences in vivo ( Zhang et al . , 2016 ) . Thus , the WH domain exhibits binding plasticity for different partners . These dynamic changes in Cac1 conformation are supported by distance-dependent changes in the pyrene spectrum of labeled cysteines 440 and 454 when bound to H3/H4 ( Figure 5D ) , as well as H3/H4-dependent HX changes at Cac1 residues 550–591 ( Figure 2A ) . The increase in HX upon H3/H4 binding is unexpected for a Cac1 region that also cross-links to histones . The weak affinity of Cac1454 ( Py ) to H3/H4 ( KDapp = n . c . ; Table 1 ) , however , suggests that the HX changes may not be due to direct binding , but rather to allosteric conformational changes or conformational mobility dependent on other binding site ( s ) on Cac1 ( Engen , 2009 ) . Together , these results shape a functional model , wherein Cac1 serves as a central scaffold with a modular architecture , which is subject to conformational changes upon H3/H4 binding . Of the three CAF-1 subunits , Cac1 is uniquely capable of both binding to H3/H4 and tetramerizing histones ( Figure 1B–D ) . CAF-1 primarily interacts with H3/H4 through the last third of Cac1 ( Figures 3B and 6 ) , with the majority of these Cac1-H3/H4 cross-links occurring in the ED domain and the C-terminus to the cores and N-terminal tails of both H3 and H4 . However , the Cac1-H3/H4 binding constant is more than an order of magnitude weaker than that of intact CAF-1 ( Winkler et al . , 2012b; Liu et al . , 2012 ) ( Table 1 ) . Therefore , the other subunits contribute to H3/H4 binding through direct interactions as seen in the cross-linking study ( Figure 3B ) , and also possibly indirectly through interactions that stabilize Cac1 . Several functional consequences for H3/H4 mutants linked to CAF-1 have been observed in yeast , including mutants that derepress rDNA , telomeric , and mating loci silencing ( reviewed in [Li and Zhang , 2012] ) . CAF-1 is known to co-purify with specific post-translational modifications ( PTMs ) , namely H4 acetylated lysines 5 , 8 , 12 , 16 , and H3 acetylated lysine 56 and methylated ( me1 , me2 , and me3 ) lysine 79 ( Zhou et al . , 2006; Masumoto et al . , 2005 ) . Mutations to the acetylatable lysines in H4 have a detrimental effect on CAF-1-dependent H3/H4 incorporation into chromatin ( Glowczewski et al . , 2004 ) . Moreover , a H3 K14R mutant , which cannot be acetylated , decreases levels of bound Cac2 and has functional consequences in rDNA silencing and aging ( Xu et al . , 2016 ) . However , deletion of the H3 and H4 N-terminal tails did not alter CAF-1-H3/H4 binding affinity ( Winkler et al . , 2012b ) , nor histone deposition by CAF-1 ( Shibahara et al . , 2000 ) . The cross-links from the H3/H4 tails to the WH domain , together with the weak Cac1454 ( Py ) – H3/H4 affinity ( Figures 3B , 4C and Table 1 ) , are consistent with these prior observations . As the WH domain interacts with histone tails , our model reveals how histone N-terminal PTMs could modulate CAF-1 function in vivo . In contrast to the tails , the core domains of both histone H3 and H4 cross-link to multiple regions in Cac1 , including the middle domain , WH domain , and regions near the ED domain . The cross-links in the vicinity of H3 K56 provide an explanation for the modulatory effect of H3 K56 acetylation on CAF-1 binding ( Li et al . , 2008; Winkler et al . , 2012b ) . The direct cross-links of Cac1-Cac3 occur in a region near where Cac3 homologs from multi-cellular organisms bind helix α1 of H4 ( Nowak et al . , 2011; Murzina et al . , 2008 ) . However , interactions previously observed between the H3 N-terminus with the top surface of the beta propeller ( Schmitges et al . , 2011 ) do not coincide with any cross-links observed here . The D . melanogaster homolog of Cac3 binds to the N-terminal peptides of either H3 or H4 with tight to modest affinity , respectively ( KD = 35 nM for H4 , KD = 2 µM for H3 ) ( Nowak et al . , 2011 ) . That both of these interactions are tighter than our KD for Cac3 binding to H3/H4 ( Py ) ( Figure 1B ) raises the possibility that yeast Cac3 might bind H3/H4 differently than other eukaryotic homologs . The weak Cac2 and Cac3 interactions with histones measured here , together with the inability to tetramerize H3/H4 , is consistent with the dearth of specific Cac2-H3/H4 and Cac3-H3/H4 cross-links . Importantly , our cross-links are largely consistent with a recent report examining CAF-1 and CAF-1-H3/H4 cross-links ( Kim et al . , 2016 ) . In both studies , the Cac1 middle domain cross-links to Cac2 and Cac3 , and the Cac1 C-terminus cross-links to Cac2 and H3 . We report an additional cross-link between the middle domain to H4 whereas Kim , et al . observe the middle domain cross-linking to H3 , consistent with the conclusion that this domain scaffolds binding to Cac2 , Cac3 , and H3/H4 . The electron microscopy data also support the idea that Cac1 is a platform for these interactions , as Cac1 is physically associated with Cac2 , Cac3 , and H3/H4 ( Kim et al . , 2016 ) . We observe more histone cross-links to the Cac1 C-terminus whereas Kim , et al . observe more to the middle domain . These differences may be attributed to experimental and computational differences . Specifically , cross-links are sensitive to cross-linking reaction times and amounts or concentrations of proteins and cross-linking agents . It is not possible to compare the differences between the relevant conditions in the two studies , as these methods are not well-detailed in the Kim et al . study ( Kim et al . , 2016; Leitner et al . , 2014 ) . Importantly , we emphasize that chemical cross-linking is a capture method rather than an equilibrium one; thus , the abundance of cross-links is not suggestive of equilibrium binding constants . Because cross-links report on amino acid proximity and have little or no relation to the binding energy of protein-protein interactions , we employed biophysical approaches under equilibrium conditions to validate the hypothesis that the Cac1 C-terminus interacts with histones . Full-length Cac1 binds H3/H4 with a KDapp of 97 nM ( Figure 1C ) , similar to the KDapp observed for Cac1386-H3/H4 ( 210 nM; Figure 4C ) . The bulk of the Cac1-H3/H4 interactions that contribute to lower the free energy of binding , therefore , indeed occur at the C-terminus of Cac1 . We also used a different strategy for determination of cross-linked peptides . Kim , et al . used xQuest , which analyzes a mix of unlabeled and isotopic cross-linkers , and relies on false discovery rates for statistical confidence ( Leitner et al . , 2014 ) . False discovery rates are more useful for larger sample sizes , while mixing in isotopic cross-linkers reduces the signal intensity and increases the search space . In contrast , we used the Batch-Tag Web function in Protein Prospector , which provides scores based on y and b ion matches for the overall cross-linked peptide complex , one cross-linked peptide only , and the other cross-linked peptide only . This analysis is useful for minimizing false positives in cross-linking experiments , as high-scoring peptides are often cross-linked to poorer-scoring peptides ( Trnka et al . , 2014 ) . Knowing the score of the poorer-scoring peptide , along with confirming ion matches in the spectrum , are critical determinants for a high confidence assignment . This approach compares favorably to approaches that do not show the score of the poorer-scoring peptide ( Trnka et al . , 2014 ) . Whereas the Cac1 WH domain can form a homodimer , this form is not required for ( H3/H4 ) 2 tetramerization ( Figures 4A , B and 5A ) . The putative dimerization domain consists of residues 572–600 at the C-terminus , and does not appear to extend to amino acid 386 , as Cac1386 ( Py ) and Cac1454 ( Py ) have similar dimerization constants ( Table 1 and Figure 5B ) . The dimer form , however , was not observed in H . B . Thus , the Cac1386 and Cac1454 monomer/dimer equilibrium can be significantly affected by changes in buffer composition , similar to the H3/H4 dimer/tetramer forms ( Donham et al . , 2011 ) . This observation highlights the susceptibility of highly charged proteins , such as histones and their chaperone counterparts , to buffer modifications that impact protein oligomerization states . Since CAF-1 is a monomer in complex with H3/H4 , the dimerization observed for Cac1 alone is likely masked by other subunit interactions in intact CAF-1 . Dissection of Cac1 function identified a specific region in the C-terminus that is required to tetramerize ( H3/H4 ) 2 ( Figure 4A and B ) . In the HX and CX analysis , lysines 444 and 464 cross-link to H3 and H4 , respectively ( Figure 3C ) , and residues 463–473 are protected from exchange with H3/H4 ( Figure 2A ) . Importantly , the region required for tetramerization , between amino acids 386 and 421 , overlaps the highly acidic ED domain ( residues 383–436 ) . Our studies reveal its role in CAF-1-induced ( H3/H4 ) 2 tetramer formation , consistent with the observation that deletion of the p150 ED domain has chromatin assembly defects ( Kaufman et al . , 1995 ) . Therefore , the Cac1 ED domain interacts with H3/H4 , which is similar to how other histone chaperones use acidic patches/surfaces for histone binding ( Das et al . , 2010 ) . Collectively , the data support a model in which the CAF-1-H3/H4 architecture is organized by Cac1 through modular protein-protein interactions within the middle , ED , and C-terminal regions ( Figure 6 ) . The CAF-1 conformation may be significantly different with and without histones , as evidenced by the different inter-subunit cross-linking patterns between CAF-1 and CAF-1-H3/H4 samples ( Figure 3A and B ) . In addition , extensive changes in HX occur throughout Cac1 with H3/H4 binding , even in regions not in direct contact with H3/H4 ( Figure 2A ) . Together , these results suggest that H3/H4 binding induces large structural changes in CAF-1 . The routes that histones follow before reaching the nucleosome are guided by histone chaperones , which position the histones for proper interactions with enzymes , nucleosome remodelers , other histone chaperones , and DNA ( Liu and Churchill , 2012 ) . DNA replication is coupled to histone chaperone function , as passage of the replisome stalls or slows with depletion of the histone chaperones FACT , Asf1 , or CAF-1 in vivo and in vitro ( Groth et al . , 2007; Hoek and Stillman , 2003; Schlesinger and Formosa , 2000 ) . The conservative model of ( H3/H4 ) 2 tetramer inheritance during replication ( Prior et al . , 1980; Xu et al . , 2010; Hoek and Stillman , 2003 ) , an important mechanism for epigenetic inheritance of histone marks , is explained by the tetramer form of histones maintained by CAF-1 . In contrast to other known mechanisms , monomeric CAF-1 is capable of binding ( H3/H4 ) 2 tetramers through a minimal bipartite region in the monomeric Cac1 subunit . The Cac1 subunit , then , is responsible for assembling ( H3/H4 ) 2 tetramers , localizing CAF-1-H3/H4 to replication forks through PCNA , and scaffolding the architecture ( Shibahara and Stillman , 1999; Krawitz et al . , 2002 ) ( Figure 6 ) . These biological functions likely require many protein-protein interactions through different Cac1 domains . The interactions identified by this study indeed appear to affect CAF-1-H3/H4 structure and function , as mutations designed to disrupt them impair the CAF-1-dependent DNA damage response in vivo ( Figures 2C and 3C ) . Thus , our model presents a novel mechanism of H3/H4 binding by a histone chaperone , illuminating a unique tetramerization pathway experienced by histones during replication . The expression and purification of X . laevis H3 and H4 with point substitutions H3 C110A and H4 T71C were carried out as before ( Scorgie et al . , 2012 ) . The procedure for labeling residue 71C in histone H4 with CPM ( Invitrogen ) , FM ( Invitrogen ) , or N- ( 1-Pyrenyl ) maleimide ( Sigma ) were as previously reported . Briefly , each fluorophore was individually incubated at 15x molar excess with H4 protein in denaturing buffer ( 20 mM HEPES , 6 M guanidine HCl , 0 . 5 mM TCEP , pH 7 . 25 ) . Excess fluorophore was removed by centrifugation through Sephadex G-15 beads ( Sigma ) . The remaining labeled H4 was then assembled with H3 by extensive dialysis into high salt buffer ( 10 mM Tris , 2 M NaCl , 1 mM EDTA , 0 . 5 mM TCEP , pH 7 . 5 ) . Soluble protein was finally isolated through size exclusion chromatography . Purification of S . cerevisiae CAF-1 from baculovirus-infected Sf9 cells was carried out as previously described ( Liu et al . , 2012 ) . Briefly , Sf9 cells were co-infected for 48 hr with viral stocks for each CAF-1 subunit ( Cac1 with a C-terminal Strep II epitope , Cac2 with a C-terminal His6x epitope , and Cac3 with a C-terminal FLAG epitope ) , each with an MOI of 1 . The cell pellets were homogenized in 10 mM Tris pH 7 . 4 , 350 mM NaCl , 1 mM DTT , 10 µg/mL DNase I , along with inhibitors for proteases ( EDTA-free tablet; Roche ) and phosphatases ( 1 mM Na3VO4 and 10 mM NaF ) . For Cac1 and Cac3 ( both with C-terminal Strep II epitopes ) purifications , the same procedure was followed . After Cac1 purification , MALDI MS was used to analyze the peaks pooled from SEC . The Cac2 purification from Rosetta 2 ( DE3 ) pLysS cells ( Novagen ) was performed as previously described ( Liu et al . , 2012 ) . Cac1 N-terminal deletions to be inserted into the pGEX6P-1 plasmid possessed a S503E mutation , in order to reflect the phosphorylated S503 observed in our mass spectrometry experiments ( data not shown ) . This mutation was introduced by site-directed mutagenesis ( Quikchange II XL kit; Agilent Technologies ) using the following primers: Forward primer: 5' – C ATC GTCT CTA CCA TCC AAA AGA AGT AAT GAG GAC TTA CAG GCA CAG AC – 3' Reverse primer: 5' – GT CTG TGC CTG TAA GTC CTC ATT ACT TCT TTT GGA TGG TAG AGA CGA TG – 3' The N-terminal deletions were then generated with the following primers , which included forward primers with a BamHI restriction digestion site , as well as a reverse primer overlapping the C-terminal Strep II tag with an EcoRI digestion site: Forward primers: Residues 386–606: 5’- CGA GGA TCC TCT GAC GTT GAA TGG GTT AAT G – 3’ Residues 421–606: 5’- GTT GGA TCC GGA GAG TTT GAC GGG TTT CTA G – 3’ Residues 454–606: 5’- CGC GGA TCC TGC CTA AAA TCC AAT TTT GAA AAC – 3’ Residues 457–606: 5’- CGG GGA TCC TCC AAT TTT GAA AAC TTA TCA GAG GAA – 3’ Reverse primer: 5'- GGT GAA TTC CTA CTT TTC GAA CTG CGG GTG -3' For crystallography , the following reverse primer was used to clone Cac1457 without the C-terminal Strep II tag: 5’- ATG CGG CCG CTT ACA AAG ACG GGG TTG GCA TAT TTG -3’ The inserts were then ligated by T4 ligase into pGEX6P-1 , which provides an N-terminal GST tag upstream of a PreScission protease cleavage site . After sequence verification , the plasmids were transformed into Rosetta 2 ( DE3 ) pLysS cells for protein expression . 20 mL cultures were grown overnight prior to inoculation into 3 L of total culture . When the optical density at 600 nm reached 0 . 4 , protein expression was induced with 1 mM IPTG and allowed to incubate for 3 hr at 32°C . The pellets were then harvested and flash frozen before protein purification . Subsequently , the pellets were resuspended in 20 mM Tris , 1 M NaCl , 2 mM DTT , pH 7 . 4 with 10 µg/mL DNase I and protease ( EDTA-free tablet; Roche ) and phosphatase ( 1 mM Na3VO4 and 10 mM NaF ) inhibitors . The lysate was sonicated , and then clarified by centrifugation . The resulting supernatant was bound to gluthathione Sepharose beads ( Thermo Fisher ) for 2 hr at 4°C before cleavage with PreScission protease overnight . The cleaved protein was bound to a StrepTactin Sepharose column ( GE Healthcare ) and washed extensively . This StrepTactin column step was skipped when purifying Cac1457 for crystallography . The protein was eluted from StrepTactin beads with 10 mM Tris , 350 mM NaCl , 2 mM DTT , 2 . 5 mM d-Desthiobiotin , pH 7 . 4 , and finally purified through a Superdex 75 column in 20 mM HEPES , 150 mM NaCl , 1 mM DTT , pH 7 . 4 . Cac1 truncations Cac1386 and Cac1454 were labeled with N- ( 1-Pyrenyl ) maleimide by incubation with 50x molar excess fluorophore at 4°C overnight in 20 mM HEPES , 150 mM NaCl , 0 . 5 mM TCEP , pH 7 . 4 . Excess dye was removed through G-15 Sephadex beads . For all fluorescence spectroscopy assays , a FluoroLog-3 fluorometer ( Horiba ) with a thermostat set at 20°C was used . For fluorescence anisotropy measurements of H3/H4 ( Py ) , 750 µL of 25 nM H3/H4 ( Py ) was equilibrated in Histone Buffer ( H . B . : 20 mM Tris , 150 mM KCl , 2 mM MgCl2 , 1% glycerol , 0 . 5 mM TCEP , 0 . 05% BRIJ-35 , pH 7 . 5 ) . CAF-1 or the appropriate CAF-1 subunit was then titrated into H3/H4 ( Py ) . For CAF-1 titrations , 5 nM of H3/H4 ( Py ) was used . For fluorescence anisotropy measurements of Cac1386 ( Py ) and Cac1454 ( Py ) , 750 µL of 10 nM labeled protein were incubated in the same buffer , then titrated with Cac1386 , Cac1454 , or H3/H4 . With the polarizers in place , pyrene was excited at 345 nm ( slit width: 6 nm ) and measured at 375 nm ( slit width: 13 nm ) . The anisotropy was calculated according to Equation 1: ( 1 ) r= ( IVV−G∗IVH ) / ( IVV+2G∗IVH ) , where G is the grating factor ( G = IHV / IHH ) . Before and after titration of Cac2 or H3/H4 into Cac1386 ( Py ) , the pyrene fluorescence spectra were also collected with polarizers in place by exciting pyrene at 345 nm ( slit width: 5 nm ) , and scanning the emission from 360–550 nm ( slit width: 7 nm ) . If the anisotropy experiment resulted in a significant degree of fluorophore quenching ( >10% of pyrene fluorescence intensity ) – as was the case for H3/H4 ( Py ) binding to CAF-1 , Cac1 , Cac2 , and Cac3 – the following equation was used to obtain a corrected anisotropy value ( Dandliker et al . , 1981 ) : ( 2 ) r= ( ( ( ( A−Af ) / ( Ab−A ) ) ∗ ( Qf/Qb ) ∗ ( Ab ) ) +Af ) / ( 1+ ( ( A−Af ) / ( Ab−A ) ∗ ( Qf/Qb ) ) ) , where A is the observed anisotropy , Af is the anisotropy of free H3/H4 ( Py ) , Ab is the anisotropy of saturated H3/H4 ( Py ) , Qf is the fluorescence intensity of free H3/H4 ( Py ) , and Qb is the fluorescence intensity of saturated H3/H4 ( Py ) . The rationale for choice of binding equation was based on Figure 7 , with several considerations . First , we made the assumption that the H3/H4 tetramerization equilibrium constant ( D + D’ to T ) was negligible under the FA experimental conditions . This is a reasonable assumption because we have shown previously ( Donham et al . , 2011; Liu et al . , 2012 ) that H3/H4 exists as dimers under our experimental buffer conditions . Here , we find no anisotropy change of H3/H4 ( Py ) with addition of unlabeled H3/H4 up to a 10 µM concentration ( data not shown ) . Thus , the tetramerization KD is very weak relative to the CAF-1-H3/H4 KD . Second , we cannot measure K1 and K2 independently , and so report an apparent KDapp , which is best fit with a one-site ligand depletion model . 10 . 7554/eLife . 18023 . 019Figure 7 . Potential equilibrium for CAF-1 association with H3/H4 . D and T indicate dimers or tetramers of H3/H4 , respectively . DOI: http://dx . doi . org/10 . 7554/eLife . 18023 . 019 The apparent dissociation constants were calculated in GraphPad Prism ( v . 5 . 0d ) by Equation 3 to account for ligand depletion: ( 3 ) Fi=1+ ( Fmax ) ∗ ( ( ( KD+[A∗]+[B]i ) −sqrt ( ( ( KD+[A∗]+[B]i ) 2 ) − ( 4∗[A∗]∗[B]i ) ) ) /2∗[A] ) ) ) , where i indicates the varying concentrations of unlabeled protein B that were titrated into the labeled protein A* . For the mixed fluorophore FRET experiments , 750 µL of 10 nM of H3/H4 was prepared with half of the histone population labeled with the FRET donor CPM at H4 Cys71 , and the other half labeled with the FRET acceptor FM at H4 Cys71 , equilibrated in H . B . The Förster radius for the CPM-FM pair is 52 Å ( Wu and Brand , 1994 ) . To evaluate FRET , CPM was excited at 385 nm , and the emission spectrum recorded from 400–600 nm . In parallel , FM was excited at 491 nm , and the emission recorded from 500–600 nm . The FRET Effect was calculated from the “Enhanced fluorescence of acceptor” method described by Clegg ( Clegg , 1992 ) , which calculates the efficiency of FRET by observing the enhanced emission signal of the acceptor fluorophore upon donor excitation: ( 4 ) FRETEffect=F385/F491 Under our FRET system , F385 is the extracted emission of FM when excited at 385 nm , while F491 is the emission of FM when directly excited at 491 nm . The F385 curve is extracted by first fitting the donor-only spectrum ( H3/H4 ( CPM ) ) excited at 385 nm to the dual-labeled spectrum ( H3/H4 ( CPM/FM ) ) excited at 385 nm . The fitted curve is then subtracted from the dual-labeled spectrum , producing the extracted acceptor spectrum . The FRET Effect is obtained when the extracted F385 spectrum is normalized by the direct excitation of acceptor ( F491 ) , which is especially valuable when an H3/H4 binding partner quenches the FM signal . All presented values derived from fluorescence data were obtained from at least three independent experiments . 0 . 2 µM H3/H4 ( FM ) was incubated with 1 . 6 µM of the indicated histone chaperone or CAF-1 subunit in 10 mM Tris , 150 mM NaCl , 0 . 5 mM TCEP , pH 7 . 5 for at least 20 min on ice . For assays involving truncated Cac1 , 2 µM of the Cac1 truncation was first incubated in either the above buffer or with H . B . for 30 min on ice prior to binding H3/H4 ( FM ) . After histone binding , 0 . 4 µM 80 bp Widom DNA was then introduced into the reaction for at least 20 min on ice . The disome/tetrasome species were separated by electrophoresis in 0 . 2x TBE ( 1x = 89 mM Tris , 89 mM boric acid , 2 mM EDTA , pH 8 . 0 ) 59:1 acrylamide:bis-acrylamide native gels for 150 min at 70 V . Fluorescein fluorescence was detected by scanning on a Typhoon 9400 imager ( GE Healthcare ) ( excitation: 488 nm , emission: 526 nm ) . The Integrated Density Value ( IDV ) of the disome and tetrasome bands were quantified by ImageQuant ( GE Healthcare ) , and the fraction of tetrasomes formed was calculated by Equation 5: ( 5 ) FractionTetrasome= ( IDVTetrasomes−IDVTetrasomeBackground/2 ) / ( ( IDVDisomes−IDVDisomeBackground ) + ( IDVTetrasomes−IDVTetrasomeBackground/2 ) ) For homology modeling , the PHYRE2 ( Kelley et al . , 2015 ) server ( Structural Bioinformatics Group , Imperial College ) was used . The amino acid sequences of Cac2 and Cac3 without epitope tag sequence were input into the PHYRE2 web program ( Kelley et al . , 2015 ) . 'Intensive' rather than 'Normal' modeling was used; 'Intensive' mode allows for use of multiple structural templates and ab initio modeling . Both Cac2 and Cac3 are confidently modeled as WD-repeat β propellers , with Cac2 possessing an additional disordered C-terminus ( not shown ) and Cac3 possessing an N-terminal helix . CAF-1 and CAF-1-H3/H4 complexes were purified and buffer exchanged into 50 mM KH2PO4 , 100 mM KCl , 5 mM DTT , pH 7 . 2 . The samples were exchanged in D2O ( 99 . 9% , Cambridge Isotope Laboratories ) at 10°C for varied lengths of time , then quenched with 100 mM dibasic potassium phosphate HCl , pH 2 . 4 at 0°C . The samples were then immediately injected into a 0°C chamber , digested using an online immobilized pepsin column and subjected to reversed-phase UPLC ( Ultra high Performance Liquid Chromatography ) for high resolution separation . To determine the extent of deuteration by mass , electrospray ionization and MSe data acquisition were implemented on a Synapt G2 ( Waters Corp ) q-TOF ( Quadrupole-Time of Flight ) mass spectrometer , which enabled high resolution and accurate mass analysis of both precursor and fragment ions . A peptide list was generated by analyzing a mock experiment using undeuterated buffers resulting in unlabeled protein and searching the resulting data using the ProteinLynx Global Server ( Waters Corp ) search algorithm . DynamX v3 . 0 software ( Waters Corp ) allows automated detection of counterpart peptides identified in the unlabeled analyses within raw data files containing MS data for all deuterated peptides at each time point . DynamX was used to assign individual isotope distributions and compute weighted average mass values for deuterated peptides . All isotope assignments by DynamX were manually validated . No back exchange correction was used , because only relative changes in deuteration were calculated , which is expected to be identical for both raw and corrected deuteration levels ( Wales and Engen , 2006 ) . The genotypes of yeast strains and plasmids are described in Table 2 and Supplementary file 1B . The sequences of primers used for mutagenesis are listed in Supplementary file 1B . Following confirmation by sequencing , the mutated plasmids were transformed into strains deleted for the endogenous CAC genes . The empty vector was the parent plasmid pRS315 . Resistance to zeocin was determined by five fold dilution analysis of 1 OD 600 nm logarithmically growing cultures of yeast strains onto plates with and without the indicated amounts of zeocin . Following 2–3 days of growth at 30°C , the yeast plates were photographed . For DSS cross-linking , DSS ( Thermo Pierce ) was prepared to 10 mM in DMSO . 40 µg of the CAF-1 and CAF-1-H3/H4 complexes were each allowed to incubate with 500 µM DSS for 30 min at room temperature in 40 µL of 20 mM HEPES , 150 mM NaCl , pH 7 . 4 . For EDC cross-linking , EDC was prepared to 75 mM in 50 mM KH2PO4 , 100 mM KCl , pH 6 . 0 . 40 µg of CAF-1 or CAF-1-H3/H4 were incubated with 25 mM EDC in the EDC buffer for 90 min at room temperature . As a negative control , an uncross-linked sample was included . The cross-linking reactions were quenched by addition of 50 mM Tris , pH 7 . 4 , for an additional 15 min . Subsequently , 2 µL of 1% ProteaseMAX surfactant ( Promega ) was added . The total volume was brought up to 93 . 5 µL with 50 mM NH4HCO3 . Cysteines were reduced with 1 µL of 0 . 5 M DTT at 55°C for 20 min , then alkylated with 2 . 7 µL of 0 . 55 M iodoacetamide for 15 min at room temperature in the dark . Prior to protease treatment , 1 µL ProteaseMAX surfactant was added to ensure a denaturing environment . Trypsin ( Promega #5111 ) or trypsin/LysC ( Promega #V507A ) was prepared with 50 mM acetic acid to a final concentration of 1 µg/µL , while GluC ( Promega #V1651 ) was prepared to 1 µg/µL with 50 mM NH4HCO3 . Trypsin alone , trypsin/LysC , or trypsin and GluC were added to the reaction at a 1:20 protease:protein ( w/w ) ratio , and allowed to digest for 3 hr at 37°C . The reaction was then spun down and TFA ( Trifluoroacetic Acid ) was added to a final volume/volume concentration of 0 . 5% . The sample was desalted and concentrated with a C18-embedded ZipTip ( Millpore ) . The peptides were then vacuum-concentrated using a SpeedVac ( Thermo ) and resuspended in 20 µL of 0 . 1% FA ( Formic Acid ) . Samples were analyzed on the LTQ Orbitrap Velos mass spectrometer ( Thermo Fisher Scientific ) coupled with an Eksigent nanoLC-2D LC system . For sample injection , 8 µL of sample was loaded onto a trapping column ( ZORBAX 300SB-C18 , 5 × 0 . 3 mm , 5 µm ) and washed with 2% ACN ( acetonitrile ) , 0 . 1% FA at a flow rate of 10 µL/min for 10 min . The trapping column was then switched online with the nano-pump at a flow rate of 600 nL/min . Peptides were separated on an in-house made 100 µm i . d . × 150 mm fused silica capillary packed with Jupiter C18 Resin ( Phenomex; Torrance , CA ) over a 45 min gradient from 6% - 40% ACN . The flow rate was adjusted to 350 nL/min after 10 min to increase the effective separation of the peptides . Data acquisition was performed using Xcalibur ( version 2 . 1 ) software . Higher energy collisional dissociation was used to produce the fragment ions in the linear ion trap from the precursor ions , which were measured in the Orbitrap mass analyzer . For every MS scan , the nineteen most intense ions were selected for fragmentation , and masses selected for fragmentation were then excluded for a duration of 120 s after a repeat count of 3 . Orbitrap obtained raw files were converted to de-isotoped , centroided peak lists using an in-house script ( PAVA , UCSF ) . Cross-linked peptides were identified through the Protein Prospector package ( Chu et al . , 2010 ) . The uploaded spectra were searched for constant modifications including DSS and carbamidomethylated cysteine using the Batch-Tag Web function . Batch-Tag identifies small , unspecified peptide modifications , adapted in cross-linking scenarios to search for large modifications predicted to be cross-linked peptides . Variable modifications included oxidized methionine , and phosphorylated serine , threonine , and tyrosine . Tryptic peptides were included , allowing for two missed cleavages . The precursor mass tolerance was set to 12 ppm ( parts per million ) , while fragment mass tolerance was 25 ppm . The Search Compare function was used to filter spectra; cross-linked peptide spectra were confirmed with a high score value ( >20 ) , low expect value ( <1 × 10–5 ) , and multiple b and y main sequence ion matches ( Supplementary file 1C ) . Cac1457 without the C-terminal Strep tag II was expressed and purified as described above . The protein was concentrated to 7 . 5 mg/mL and extensively dialyzed into 20 mM HEPES , 50 mM NaCl , 2 mM TCEP , pH 7 . 4 . Protein crystals grew in 0 . 2 M sodium formate , 14% PEG 3350 , 0 . 1 M Tris , pH 8 . 3 using the sitting drop vapor diffusion method . Data were collected at the Molecular Biology Consortium beamline 4 . 2 . 2 at the Advanced Light Source at Lawrence Berkeley National Laboratory , and processed using d*trek ( Pflugrath , 1999 ) ( Table 4 ) . Molecular replacement was performed using PHASER ( McCoy et al . , 2005 ) implemented in Phenix ( Adams et al . , 2010 ) with PDB ID 5EJO ( Zhang et al . , 2016 ) as the search model . The placed model was refined through iterative refinement and model building using Phenix . refine and Coot ( Emsley and Cowtan , 2004 ) until the Rfree value was sufficiently low for a model with good geometry , stereochemistry and structural characteristics as determined by PDB validation analyses ( Read et al . , 2011 ) ( Table 4 ) . The final structure has PDB ID 5JBM . The buried surface area and the contacts between Cac1457 and the indicated symmetry related molecule were calculated using Areaimol and Contact programs implemented in the CCP4 suite of programs ( Bailey , 1994 ) , respectively . Superpose ( Maiti et al . , 2004 ) was used to measure R . m . s . d values . Electrostatics calculations were performed with CHARMM ( Jo et al . , 2008 ) , and PyMol ( Schrödinger , LLC , 2014 ) was used to generate structure figures .	The DNA of a human , yeast or other eukaryotic cell is bound to proteins called histones to form repeating units called nucleosomes . Every time a eukaryotic cell divides , it must duplicate its DNA . Old histones are first removed from the nucleosomes before being re-assembled onto the newly duplicated DNA along with new histone proteins , producing a full complement of nucleosomes . A group of proteins called the chromatin assembly factor 1 ( or CAF-1 for short ) helps to assemble the histones onto the DNA . CAF-1 is made up of three proteins , and binds to two copies of each of the histones known as H3 and H4 . These are the first histones to be assembled onto the nucleosomes . It was not clear how the components of CAF-1 are organized , or how CAF-1 recognizes histones . Liu et al . have now investigated the structure of CAF-1 and its interactions with the H3 and H4 histones by studying yeast proteins and cells . Yeast is a good model system because yeast CAF-1 is smaller and easier to isolate than human CAF-1 , yet still performs the same essential activities . Using a combination of biochemical and biophysical techniques , Liu et al . found that one of the three proteins that makes up yeast CAF-1 – called Cac1 – forms a scaffold that supports the other CAF-1 proteins and histones H3 and H4 . Moreover , a specific part of Cac1 is able to bind to these histones and assemble two copies of each of them to prepare for efficient nucleosome assembly . Further experiments revealed the specific areas where the CAF-1 proteins interact with each other and with the histones , determined how strong those interactions are , and confirmed that these interactions play important roles in yeast . Overall , the results presented by Liu et al . provide new insights into the structure of CAF-1 bound to H3 and H4 . In order to understand in detail how CAF-1 helps to assemble histones onto DNA , future work needs to capture three-dimensional snapshots of the different steps in this process . Further investigation is also needed to discover how CAF-1 cooperates with other factors that promote DNA duplication .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "chromosomes", "and", "gene", "expression", "structural", "biology", "and", "molecular", "biophysics" ]	2016	The Cac1 subunit of histone chaperone CAF-1 organizes CAF-1-H3/H4 architecture and tetramerizes histones
How multicellular organisms respond to and are impacted by severe hypothermic stress is largely unknown . From C . elegans screens for mutants abnormally responding to cold-warming stimuli , we identify a molecular genetic pathway comprising ISY-1 , a conserved uncharacterized protein , and ZIP-10 , a bZIP-type transcription factor . ISY-1 gatekeeps the ZIP-10 transcriptional program by regulating the microRNA mir-60 . Downstream of ISY-1 and mir-60 , zip-10 levels rapidly and specifically increase upon transient cold-warming exposure . Prolonged zip-10 up-regulation induces several protease-encoding genes and promotes stress-induced organismic death , or phenoptosis , of C . elegans . zip-10 deficiency confers enhanced resistance to prolonged cold-warming stress , more prominently in adults than larvae . We conclude that the ZIP-10 genetic program mediates cold-warming response and may have evolved to promote wild-population kin selection under resource-limiting and thermal stress conditions . Temperature shifts pervasively affect numerous biological processes in all organisms . Heat shock stimuli activate expression of many heat-shock inducible genes through the sigma-32 factor and the evolutionarily conserved transcription factor HSF ( Heat Shock Factor ) in bacteria and eukaryotes , respectively ( Gomez-Pastor et al . , 2018; Yura et al . , 1993 ) . Coordinated expression of heat shock-induced chaperone proteins facilitates cellular proteostasis and adaptation to temperature upshift ( Mahat et al . , 2016; Solís et al . , 2016 ) . In contrast to heat shock response , how organisms respond to cold shock is still largely unknown ( Al-Fageeh and Smales , 2006; Choi et al . , 2012; Yenari and Han , 2012; Zhu , 2016 ) . Although extensive RNA expression profiling studies have identified many protein-coding genes and non-coding RNAs that are regulated by cold shock via both transcriptional and post-transcriptional mechanisms ( Al-Fageeh and Smales , 2006; Giuliodori et al . , 2010; Kandror et al . , 2004; Zhou et al . , 2017 ) , master regulators of cold shock response and cold-regulated genes ( counterparts of HSF ) have long been elusive and mechanisms of cold shock response in multicellular organisms remain poorly characterized . At the organismic level , warm-blooded mammals normally keep body temperature at about 37°C and initiate multiple homeostatic mechanisms to maintain body temperature upon exposure to hypothermia ( Bautista , 2015; Morrison , 2016; Tansey and Johnson , 2015; Vriens et al . , 2014 ) . In humans , therapeutic hypothermia ( 32–34°C ) has been widely used to treat ischemic disorders and proposed to activate multifaceted cellular programs to protect against ischemic damages ( Choi et al . , 2012; Polderman , 2009; Yenari and Han , 2012 ) . By contrast , cold-blooded animals including most invertebrates experience varying body temperature depending on the environment , but can nonetheless elicit stereotypic behavioral , physiological and transcriptional response to chronic hypothermia or transient cold shock ( Al-Fageeh and Smales , 2006; Garrity et al . , 2010 ) . Like many other types of stress , prolonged severe hypothermia can lead to the death of organisms , in most cases likely because of failure in adapting to the stress , or alternatively through stress-induced phenoptosis , namely genetically programed organismic death ( Longo et al . , 2005; Skulachev , 1999; 2002 ) . Although phenoptosis has been phenotypically documented in many cases , its evolutionary significance and genetic mechanisms remain unclear and debated ( Longo et al . , 2005; Sapolsky , 2004 ) . We previously discovered a C . elegans genetic pathway that maintains cell membrane fluidity by regulating lipid desaturation in response to moderate hypothermia ( 10–15°C ) ( Fan and Evans , 2015; Ma et al . , 2015 ) . Expression of the gene fat-7 , which encodes a lipid desaturase , is transcriptionally induced by 10–15°C but not by more severe hypothermia ( i . e . cold shock at 0–4°C ) , which impairs C . elegans reproduction and growth , and elicits distinct physiological and behavioral responses ( Garrity et al . , 2010; Lyons et al . , 1975; Ma et al . , 2015; Murray et al . , 2007 ) . However , as severe hypothermia arrests most of cell biological processes , strong transcriptional responses to cold shock e . g . 0–4°C likely only manifest during the organismic recovery to normal ambient temperature . We thus hypothesize that a genetic pathway differing from that operating under moderate hypothermia exposure controls the transcriptional response to severe hypothermia/cold shock followed by warming in C . elegans . In this work , we performed transcriptome profiling to first identify genes that are regulated by exposure to cold shock followed by recovery at normal temperature . We then used GFP-based transcriptional reporters in large-scale forward genetic screens to identify a genetic pathway consisting of isy-1 and zip-10 , the latter of which responds to cold-warming ( CW ) and mediates transcriptional responses to CW . Unexpectedly , we found strong zip-10 induction promotes organismal death while deficiency of zip-10 confers resistance to prolonged CW stress , more prominently in adults than young larvae . We propose that CW activates a ZIP-10 dependent genetic program favoring C . elegans phenoptosis and postulate that such programmed organismic death may have evolved to promote wild-population kin selection under thermal stress conditions . To identify new mechanisms of C . elegans response to severe hypothermia , we performed RNA sequencing ( RNA-seq ) of wild-type C . elegans populations after 2 hrs exposure to 4°C cold shock followed by recovery at 20°C for 1 hr . We used such CW conditions in an attempt to identify genes that specifically and rapidly respond to CW rather than those that respond to general organismic deterioration after long cold exposure . After differential expression analyses of triplicate samples , we identified 604 genes that are significantly up- or down-regulated by such CW conditions ( Figure 1—source data 1 , Figure 1A and Figure 1—figure supplement 1A ) . Gene ontology analysis indicates that the CW-regulated genes are involved in biological processes including lipid metabolisms , autophagy , proteostasis and cell signaling ( Figure 1—figure supplement 1B ) . We generated transgenic C . elegans strains in which GFP is driven by promoters of the top-ranked CW-inducible genes . In this work , we focus on asp-17 as a robust CW-inducible reporter gene owing to its low baseline expression level and high-fold induction by CW , features that permitted facile isolation of full-penetrance mutants after random mutagenesis ( see below ) with both abnormal asp-17p::GFP expression and altered organismic tolerance to prolonged cold stress . C . elegans asp-17 encodes an aspartyl-like protease with unknown molecular functions . Like other CW-inducible genes , asp-17 up-regulation is more prominently induced by severe than moderate hypothermia followed by recovery from cold shock ( Figure 1B and C ) . Among the aspartyl-like protease family members , we found that only asp-17 was robustly and specifically induced by CW ( Figure 1D ) . The up-regulation of endogenous asp-17 by CW can be recapitulated by an integrated GFP reporter driven by the endogenous asp-17 promoter , indicating transcriptional regulation of asp-17 by CW ( Figure 2—figure supplement 1A ) . We varied CW treatment conditions and found that the induction of asp-17 strictly required the warming phase after cold shock ( Figure 1E ) . However , heat shock at 32°C did not increase asp-17 expression ( Figure 1E ) , consistent with previous large-scale transcriptome profiling studies in C . elegans ( Brunquell et al . , 2016 ) . Single-molecule fluorescent in situ hybridization ( smFISH ) identified the CW-induced asp-17 predominantly in intestinal cells ( Figure 1F ) . Since CW activates numerous other genes in addition to asp-17 , we sought to use asp-17p::GFP as a robust readout reporter to identify the upstream genetic pathway and transcriptional regulators that control asp-17 induction by CW . We performed a forward genetic screen using EMS-induced random mutagenesis of a parental strain carrying a genome-integrated asp-17p::GFP reporter and isolated over 30 mutants with constitutive asp-17p::GFP expression in the absence of CW ( Figure 2—figure supplement 1B and C ) . We molecularly cloned one mutant dma50 that exhibited fully penetrant and constitutively strong expression of asp-17p::GFP ( Figure 2A , B and Figure 2—figure supplement 1D–1F ) . Compared with wild type , dma50 strongly up-regulated asp-17::GFP in the intestine ( Figure 2B ) . By single nucleotide polymorphism ( SNP ) -based linkage analysis of the intestinal asp-17p::GFP phenotype , we mapped dma50 to a genetic interval on Chromosome V and used whole-genome sequencing to identify candidate causal gene mutations ( Figure 2—figure supplement 1D , E ) . Based on phenocopying by feeding RNAi against the candidate genes and transformation rescue of the asp-17p::GFP phenotype , dma50 defines a previously uncharacterized C . elegans gene isy-1 ( Figure 2A–F and Figure 2—figure supplement 1F ) . isy-1 ( Interactor of SYF1 in yeast ) encodes a protein with strong sequence similarity to an evolutionarily highly conserved family of RNA-binding proteins in eukaryotes ( Figure 2A and Figure 2—figure supplement 2A–C ) ( Dix et al . , 1999; Du et al . , 2015 ) . dma50 caused substitution of a negatively charged glutamate , which is completely conserved in the ISY protein family , to a positively charged lysine in the predicted coiled-coil region of C . elegans ISY-1 ( Figure 2A ) . An isy-1p::isy-1::GFP translational reporter indicated a rather ubiquitous distribution of ISY-1::GFP in many tissues including intestinal nuclei ( Figure 2C ) . The strong intestinal asp-17p::GFP expression caused by dma50 was fully rescued by transgenic expression of wild-type isy-1 ( + ) , single-copy integration of a mCherry-tagged isy-1 ( + ) allele , or isy-1 ( + ) expression driven by the intestine-specific ges-1 promoter ( Figure 2E and F ) . In addition , the ges-1-driven transgenic expression of sense plus antisense isy-1 RNAi fully recapitulated the dma50 phenotype ( Figure 2D ) . Endogenous expression of asp-17 was also drastically up-regulated in isy-1 mutants ( Figure 2—figure supplement 2D ) . Thus , these results identify isy-1 as a causal cell-autonomous regulator of asp-17 . Human ISY1 is critical for certain microRNA processing while yeast ISY1 is a likely component of the spliceosome ( Dix et al . , 1999; Du et al . , 2015; Galej et al . , 2016 ) . We found that CW-induced asp-17 up-regulation was further enhanced in isy-1 mutants compared with wild type ( Figure 3A ) , suggesting that ISY-1 normally restricts transcriptional activity of asp-17 . To determine the mechanism by which ISY-1 regulates transcription of asp-17 , we sought to identify transcription factors ( TF ) that meet two criteria: a ) , its mRNA or protein products are altered in isy-1 mutants , and b ) , it is genetically epistatic to isy-1 , that is , its loss-of-function ( LOF ) can suppress isy-1 LOF ( thus also likely required for asp-17 induction by CW ) . We performed RNA-seq from triplicate samples of wild-type hermaphrodites and isy-1 mutants , from which we analyzed differentially expressed TF-encoding genes in isy-1 mutants and found that a bZIP-type transcription factor-encoding gene zip-10 met both criteria ( Figure 3B , Figure 3—source data 1 ) . zip-10 mRNA was drastically up-regulated in isy-1 mutants , whereas levels of closely related bZIP family genes , such as zip-11 , were unaffected ( Figure 3C ) . Importantly , genetic deletion of zip-10 completely abrogated the ability of isy-1 RNAi to activate asp-17p::GFP ( Figure 3D ) . These results indicate that ISY-1 regulates asp-17 by controlling the level of zip-10 mRNAs . Next , we examined how the ISY-1/ZIP-10/ASP-17 pathway is regulated by CW . CW did not apparently alter levels of endogenous isy-1 mRNAs or mCherry-tagged ISY-1 proteins under the endogenous isy-1 promoter ( Figure 2—figure supplement 2D and E ) . By contrast , we found that CW induced drastic up-regulation of ZIP-10 proteins from a tagged zip-10p::zip-10::EGFP::FLAG allele in an integrated transgenic strain ( Figure 3E ) . Although EGFP fluorescence was invisible in animals carrying such transgenes ( likely because it is sandwiched by zip-10 and FLAG ) , the striking induction of ZIP-10::EGFP::FLAG was completely blocked by RNAi against zip-10 or GFP , confirming the transgene specificity ( Figure 3F ) . The baseline level of ZIP-10::EGFP::FLAG was close to the detection limit of western blot under normal conditions , but nonetheless is strongly up-regulated upon RNAi against isy-1 ( Figure 3F ) . Similar to that of asp-17 , the induction of zip-10p::zip-10::EGFP::FLAG strictly required the warming phase of CW and occurred rapidly but transiently after warming during CW ( Figure 3G ) . CW strongly up-regulated asp-17 expression in both wild type and isy-1 mutants , which exhibited abnormally high zip-10 mRNA levels ( Figure 3H ) . Furthermore , zip-10 deletion completely abrogated the up-regulation of asp-17 levels by CW ( Figure 3I ) . We also examined the ZIP-10 dependency of other CW-inducible genes identified by RNA-seq and found that at least cpr-3 also required ZIP-10 , but other CW-inducible genes including srr-6 and F53A9 . 1 , did not ( Figure 3I ) . These results demonstrate that ISY-1 suppresses asp-17 by decreasing zip-10 levels whereas CW up-regulates ZIP-10 protein abundance to promote asp-17 expression . How is zip-10 regulated by ISY-1 and CW ? Loss of ISY-1 function affected neither general intron splicing , based on an intronic GFP reporter assay , nor specific splicing of zip-10 , although both CW and isy-1 mutations strongly up-regulated zip-10 mRNA levels ( Figure 3—figure supplement 1A–1E ) . We constructed a GFP transcriptional reporter driven by the endogenous zip-10 promoter and found it was markedly up-regulated by isy-1 RNAi ( Figure 3—figure supplement 2A ) . While non-thermal stresses such as hypoxia and starvation did not increase ZIP-10 levels , CW drastically increased abundance of ZIP-10 in both cytosol and nucleus without affecting abundance of other house-keeping proteins , including HSP90 , tubulin and histone H3 ( Figure 3—figure supplement 2B and C ) . CW up-regulation of ZIP-10 required warming and was enhanced by more prolonged cold shock ( Figure 3—figure supplement 2D ) . Since CW can markedly increase zip-10 mRNA levels but to a lesser extent than the isy-1 mutation ( Figure 3—figure supplement 1D and E ) , we tested whether ZIP-10 proteins might be regulated by CW through translational control and mRNA stability . RNAi against genes encoding eIF5 and a component of the Ccr4-Not complex did not apparently alter ZIP-10 levels ( Figure 3—figure supplement 2E ) . Together , these results indicate that CW and ISY-1 regulate zip-10 primarily at the transcriptional level . Human ISY1 facilitates the processing of primary transcripts encoding certain families of microRNAs ( Du et al . , 2015 ) . Both zip-10 and asp-17 are up-regulated in a mutant C . elegans strain deficient in the microRNA mir-60 ( Kato et al . , 2016 ) . We thus tested whether mir-60 mediates the regulation of zip-10 by ISY-1 . Immunoprecipitation of mCherry-tagged ISY-1 followed by quantitative PCR ( QPCR ) revealed specific binding of primary transcripts encoding mir-60 as well as a protein-coding gene cebp-1 ( Figure 4A–4C ) . Although neither isy-1 nor mir-60 levels were affected by CW , we found CW slightly increased mir-60 binding to ISY-1 , perhaps as a feedback mechanism to limit over-activation of zip-10-dependent genes after CW treatment ( Figure 4B ) . Importantly , mature mir-60 levels were drastically decreased in isy-1 mutants while loss of mir-60 led to up-regulation of zip-10 and zip-10-dependent subset of CW-inducible genes , including asp-17 and cpr-3 , but not many other CW-inducible genes ( Figure 4D and E and Supplementary file 1 ) . The 3’ untranslated region ( Utr ) of zip-10 appeared not to be regulated by CW or isy-1 RNAi ( Figure 4F ) . However , isy-1 RNAi caused an abnormally high baseline level of ZIP-10 in the absence of CW and enabled further heightened ZIP-10 up-regulation in response to CW , followed by its down-regulation over an extended period of warming ( Figure 4G ) . These results indicate that CW regulates transcription of zip-10 ( and thereby that of asp-17 ) , while ISY-1 controls expression of zip-10 via mir-60 , likely through microRNA processing and regulation of additional upstream transcriptional zip-10 regulators that respond to CW . We compared the genes differentially regulated by CW and those by isy-1 ( dma50 ) mutants and found 246 genes , including the two ZIP-10-dependent targets asp-17 and cpr-3 , that are commonly regulated by both conditions ( Figure 5A , Figure 5—source data 1 ) . Global transcriptome changes between these two conditions are also significantly correlated ( Figure 5B ) ( correlation coefficient R as 0 . 54 , significance P value as 0 ) . We used the bioinformatics tool MEME ( Bailey et al . , 2009 ) to identify motifs present in the promoters ( ~600 bp upstream of transcription start sites ) of the commonly regulated gene subset and identified a single enriched motif characterized by AT-rich sequences ( Figure 5C ) . The gene most enriched with this motif is asp-17 , the promoter of which contains 16 such motifs ( Figure 5C ) . ZIP-10 is a bZIP-type transcription factor predicted to contain N-terminal low sequence-complexity domains and a C-terminal DNA-binding and glutamine-rich transactivation domain ( Figure 3—figure supplement 2F–H ) . To test whether the asp-17 promoter with the identified AT-rich motifs can be bound directly by ZIP-10 , we performed chromatin immunoprecipitation ( ChIP ) experiments and detected asp-17 promoter sequences in the FLAG-tagged ZIP-10 chromatin complex only under CW conditions ( Figure 5D ) . These results indicate that ZIP-10 directly binds to and activates the asp-17 promoter in the genetic program regulated by ISY-1 and CW . The striking regulation of zip-10 and asp-17 by CW and ISY-1 prompted us to examine the organismic phenotype of various mutants upon prolonged CW stress . A majority of wild-type C . elegans adults died upon prolonged CW stress ( e . g . 2–4°C for over 24 hrs ) ( Ohta et al . , 2014 ) . We found that asp-17 or zip-10 loss-of-function mutants exhibited markedly higher survival rates than wild type under the same prolonged CW stress condition ( Figure 5E ) . Consistent with a role of wild-type zip-10 in promoting organismic death , inducible zip-10 over-expression by mild transient heat shock , mediated by the hsp-16 promoter , promoted animal death even in the absence of CW ( Figure 5F ) . By contrast , other ectopically induced zip genes including zip-11 and zip-2 did not affect animal death , while a mutation specifically disrupting the glutamine-rich transactivation domain of ZIP-10 abolished the death-promoting effect ( Figure 5F and Figure 3—figure supplement 2H ) . Although zip-10 is genetically epistatic to isy-1 in the regulation of asp-17 , we found that isy-1 mutants are also markedly resistant to prolonged cold stress . This paradox was resolved after we observed that many downstream target genes of the stress-coping transcription factors HIF-1 , HSF-1 and DAF-16 are up-regulated in isy-1 mutants , and LOF of at least daf-16 could partly suppress cold tolerance by isy-1 RNAi ( Figure 5G and H ) . Since ISY-1 regulates zip-10 via mir-60 ( Figure 4 ) supporting a role of ISY-1 in specific microRNA processing ( Du et al . , 2015 ) , we performed small RNA library sequencing of wild type animals and isy-1 mutants and identified specific members of microRNAs that were differentially regulated , including mir-60 and additional microRNAs predicted to target stress-coping TFs ( Figure 5—figure supplement 1A–F ) . Thus , isy-1 mutants likely exhibit pleiotropic phenotypes caused by abnormal activation of multiple TFs in addition to ZIP-10 . In contrast to ZIP-10 dependent genes ( asp-17 and cpr-3 ) , the HIF-1/HSF-1/DAF-16 target genes were not apparently induced by CW ( Figure 1—source data 1 ) . Furthermore , unlike HSF-1 or DAF-16 that are induced by other types of stress stimuli , ZIP-10 is more strongly induced by CW in adults than in larvae ( Figure 5—figure supplement 2C ) , suggesting phenoptosis-promoting effects of zip-10 more specifically for adults . Indeed , the phenotypic difference in cold tolerance between wild type animals and zip-10 mutants manifested more prominently in developmentally more mature-stage and older animals ( Figure 5—figure supplement 2D ) . These results indicate that CW specifically activates a ZIP-10-driven and developmental stage-modulated transcriptional genetic program to promote the organismic death , or phenoptosis , of C . elegans ( Figure 5I ) . From a genetic screen for C . elegans mutants with altered transcriptional response to CW , we identified isy-1 and subsequently discovered the CW and ISY-1-regulated transcription factor ZIP-10 as a key mediator of the transcriptional response to CW . A thermal stress-responding TF might be expected to promote adaptation of animals towards the stressor , causing its LOF mutants to be sensitive to the stress . Unexpectedly , we found zip-10 mutants are markedly resistant to prolonged cold stress . However , unlike other stress-responding TFs that activate genes largely beneficial for physiological homeostasis and thus animal health under stress conditions ( Baird et al . , 2014; Dempersmier et al . , 2015; Hwang and Lee , 2011; Kandror et al . , 2004; Kumsta et al . , 2017; Landis and Murphy , 2010 ) , identified transcriptional targets of ZIP-10 include at least two Cathepsin-type proteases , CPR-3 and ASP-17 ( Figure 4E ) . In contrast to aspartyl-type proteases which are largely unknown in cellular functions , caspase-type proteases are well-known apoptotic cell death executioners while CPR-4 , a Cathepsin CPR-3 paralogue , has been shown to inhibit cell deaths in C . elegans ( Metzstein et al . , 1998; Peng et al . , 2017; Peter , 2011 ) . Ectopic expression of zip-10 and its targets promotes organismic deaths , in contrast to the effect of zip-10 or asp-17 deficiency on cold tolerance ( Figure 5E and F ) . As duration of cold shock affects levels of ZIP-10 and transient CW does not trigger phenoptosis , the pro-death role of the zip-10 genetic program likely depends on multiple factors , including the duration and severity of cold exposure . Notably , apoptotic cell death-promoting effects have also been described for specific members of mammalian bZIP TFs ( Chüeh et al . , 2017; Hartman et al . , 2004; Ritchie et al . , 2009 ) . The specific and robust induction of ZIP-10 by CW , the opposing cold-tolerance phenotypes caused by zip-10 loss-of-function and gain-of-function genetic manipulations , as well as the pro-death roles of ZIP-10 targets support the notion that the zip-10 pathway is activated by severe CW to promote phenoptosis . How do ISY-1 and CW regulate the zip-10 pathway ? We found that the zip-10 promoter activity responds to the loss of ISY-1 , which normally maintains mir-60 levels and thereby regulates zip-10 transcription likely through the processing of small RNAs . Severe cold stress also leads to accumulation of another class of small RNA risiRNA , which is important for maintaining rRNA homeostasis ( Zhou et al . , 2017 ) . Whether ISY-1 might also affect risiRNA processing remains to be characterized . Constitutive up-regulation of ZIP-10 targets in isy-1 mutants and the lack of evidence for regulation of ISY-1 by CW supports ISY-1 as a gate-keeper for the ZIP-10-driven transcriptional response to CW ( Figure 5I ) . Regulation of zip-10 is primarily transcriptional based on evidence we present in this study; further studies are required to discern to what extent mir-60 might directly act at the zip-10 locus or more indirectly impact the transcription of zip-10 , e . g . by post-transcriptionally inhibiting translation of a transcriptional activator . Up-regulation of the activity of the zip-10 promoter by CW indicates that additional cold-responding sensors and effectors upstream of ZIP-10 remain to be identified , by signaling mechanisms perhaps similar to the well-characterized cold-responding pathways found in other organisms ( Dempersmier et al . , 2015; Kandror et al . , 2004; Zhu , 2016 ) . Precisely how zip-10 is regulated by CW in coordination with ISY-1 to promote C . elegans death under prolonged CW stress awaits further investigation . The roles of ZIP-10 and a dedicated genetic program in promoting organismic death are surprising but would make sense in light of the evolutionary kin selection theory . Kin selection refers to the evolutionary process promoting the reproductive success of an organism's kin despite a cost to the organism's own reproduction ( Hamilton , 1963; Smith , 1964 ) . Dedicated genetic programs may have evolved to promote kin selection at the population level . Although the concept and potential mechanisms of programed organismic death , or phenoptosis , are debated , examples of kin selection and stress-induced organismic deterioration have been widely documented in many organisms ( Longo et al . , 2005; Sapolsky , 2004; Skulachev , 1999; 2002 ) . Laboratory conditions for hermaphroditic C . elegans clearly no longer exert selection pressure for genetic programs underlying phenoptosis or kin selection . However , our mathematic modeling of an exemplar situation of population growth for wild-type and zip-10 deficient animals under food-limiting and CW stress conditions supports the phenoptosis or kin selection hypothesis for the zip-10 pathway ( Figure 5—figure supplement 2A and B ) . Experimentally , we found that both the CW-induced zip-10 expression and the death-promoting effect of ZIP-10 occurred more prominently in older adults than in larvae ( Figure 5—figure supplement 2C and D ) . Extending from the kin selection theory , we postulate that the evolutionary advantage of programmed organismic death might manifest in the wild , where resources for growth and reproduction are limited and environments can change drastically . As such , the selective death of adult animals would benefit young and reproductively more privileged populations to facilitate the spreading of genes by young populations under resource-limiting and high-stress conditions . Our work provides an unprecedented example of stress-induced phenoptosis in C . elegans and identify a specific transcription factor in a genetic program that likely evolved to promote kin selection during animal evolution . These findings therefore bear broad implications for understanding thermal stress response , programmed organismic death ( phenoptosis ) and evolutionary biology . C . elegans strains were maintained with standard procedures unless otherwise specified . The N2 Bristol strain was used as the reference wild type , and the polymorphic Hawaiian strain CB4856 was used for genetic linkage mapping and SNP analysis ( Brenner , 1974; Davis et al . , 2005 ) . Forward genetic screen for constitutive asp-17p::GFP reporter-activating mutants after ethyl methanesulfonate ( EMS ) -induced random mutagenesis was performed as described previously ( Ma et al . , 2012; 2015 ) . Single-copy integration of isy-1p::isy-1::mCherry transgene was generated using the MosSCI method ( Frøkjaer-Jensen et al . , 2008 ) . To generate asp-17 null alleles in C . elegans , we used CRISPR-Cas9 to induce double-stranded breaks and subsequent non-homologous end joining caused a deletion of asp-17 . Feeding RNAi was performed as previously described ( Kamath and Ahringer , 2003 ) . Transgenic strains were generated by germline transformation as described ( Mello et al . , 1991 ) . Transgenic constructs were co-injected ( at 10–50 ng/μl ) with dominant unc-54p::mCherry or rol-6 markers , and stable extrachromosomal lines of mCherry+ or roller animals were established . Genotypes of strains used are as follows: daf-16 ( mu86 ) I , mir-60 ( n4947 ) II; isy-1 ( dma50 ) V , zip-10 ( ok3462 ) V , asp-17 ( dma99 ) V , dmaIs10[asp-17p::GFP; unc-54p::mCherry] X , dmaIs21[zip-10p::GFP; unc-54p::mCherry]; wgIs634[zip-10p::zip-10::EGFP::FLAG + unc-119 ( + ) ] , oxTi302 [eft-3p::mCherry::tbb-2 3'UTR + Cbr-unc-119 ( + ) ] , dmaSi1[isy-1p::isy-1::mCherry , unc-119 ( + ) ] , dmaEx95[ges-1p::isy-1 ( + ) ; rol-6 ( + ) ] , dmaEx99[isy-1 genomic DNA ( 2 ng/ul ) ; rol-6 ( + ) ] , nEx102[ges-1p::isy-1 ( + ) ; rol-6 ( + ) ] , nEx103[ges-1p::isy-1 ( + ) ; rol-6 ( + ) ] , dmaEx104[ges-1p::mCherry::3utr ( zip-10 ) , rol-6 ( + ) ] , dmaEx123[hsp-16p::zip-10; rol-6 ( + ) ] , dmaEx124[hsp-16p::zip-10; rol-6 ( + ) ] , dmaEx131[zip-10p::GFP; unc-54p::mCherry] . Control N2 animals and the isy-1 mutants were maintained at 20°C . For cold stress , N2 animals were exposed to 4°C for 2 hrs followed by 1 hr recovery at 20°C . Upon sample collection , the animals were washed down from NGM plates using M9 solution and subjected to RNA extraction using the RNeasy Mini Kit from Qiagen . 1 µg total RNA from each sample was used for sequencing library construction . Each treatment included three biological replicates . The NEBNext rRNA Depletion Kit was used for rRNA depletion . After rRNA depletion , the Agencourt RNAClean XP Beads from Beckman Coulter were used for RNA purification . Then , the NEBNext Ultra Directional RNA Library Prep Kit for Illumina was used for RNA fragmentation , first strand and second strand cDNA synthesis and double-stranded cDNA end repair . Double strand cDNAs were purified using the Agencourt AMPure XP from Beckman Coulter and ligated to adaptors of the NEBNext Multiplex Oligos for Illumina . Finally , the Q5 Hot Start HiFi PCR Master Mix was used for PCR enrichment of the adaptor-ligated DNA . The concentration and quality of the constructed sequencing libraries were measured by using the Agilent High Sensitivity DNA Kit and a Bioanalyzer 2100 from Agilent Technologies . The libraries were submitted to 100 bp paired-end high throughput sequencing using Hiseq-3000 by the Center for Advanced Technology ( CAT ) of the University of California , San Francisco . RNA-seq data analysis was performed using a super computer system equipped with multiple processors . The raw reads were trimmed and filtered by the prinseq-lite software ( 0 . 20 . 4 ) ( Schmieder and Edwards , 2011 ) . Reads longer than 30 bp and with a minimum quality score higher than 15 were kept and used for subsequent analyses . The filtered left and right read sets were compared by the Pairfq script to separate paired and single reads . The clean reads were mapped to the C . elegans genome sequence using Hisat2 ( 2 . 0 . 5 ) ( Kim et al . , 2015 ) with default parameters . The number of mapped reads were counted by featureCounts from the Subread package ( 1 . 5 . 0 ) ( Liao et al . , 2014 ) . Differential gene expression analysis was performed using the DESeq2 package ( Love et al . , 2014 ) . Adjusted p-value≤0 . 05 was used as the threshold to identify the differentially expressed genes . Gene ontology and KEGG pathway enrichment analyses for the differentially expressed genes were conducted using the Cytoscape plugins BiNGO ( Maere et al . , 2005 ) and ClueGO ( Bindea et al . , 2009 ) , respectively . Plots for the mapped reads were generated by IGVtools ( Thorvaldsdóttir et al . , 2013 ) . 50 µl pellet animals were resuspended in 250 µl lysis buffer of Quick-RNA MiniPrep kit ( Zymo Research , R1055 ) then lysed by TissueRuptor ( Motor unit ‘8’ for 1 min ) . Total RNA was extracted following the instruction ( Zymo Research , R1055 ) . 2 µg RNA/sample was reverse transcribed into cDNA ( BioTools , B24408 ) . Real-time PCR was performed by using Roche LightCycler96 ( Roche , 05815916001 ) system and SYBR Green ( Thermo Fisher Scientific , FERK1081 ) as a dsDNA-specific binding dye . qRT-PCR condition was set to 95°C for denaturation , followed by 45 cycles of 10 s at 95°C , 10 s at 60°C , and 20 s at 72°C . Melting curve analysis was performed after the final cycle to examine the specificity of primers in each reaction . Relative mRNA was calculated by ∆∆CT method and normalized to actin . Primers for qRT-PCR: asp-17 ( Forward , ATGTTCCGCTGACTGCGAAG; Reverse , TTTCATTCATTTCATCCCAC ) , F53A9 . 1 , Forward , ACTACGGAAACGGAGGATAC; Reverse , TGGCCGTGATGATGATGATG ) , srr-6 ( Forward , CTCCAAGTCCTGAAGTCGTG; Reverse , GTAGGGATGGATTGAACTCG ) , isy-1 ( Forward , AGATGCTGAGCGATTCAGAC; Reverse , CTTTCGATAGTCCGTACCAC ) , zip-10 ( Forward , TCGAGATGCTCTTCAACTG; Reverse , CTAACTGCTTGCCGGAG ) , cpr-3 ( Forward , GTAGTGGAGCAGTAACAGGTG; Reverse , CAGTTTGAATTTCGGTGACGG ) , act-3 ( Forward , TCCATCATGAAGTGCGACAT; Reverse , TAGATCCTCCGATCCAGACG ) . Transgenic ( isy-1p::isy-1::mCherry and zip-10p::zip-10::EGFP::FLAG ) animals were cold shocked ( 4°C ) for 0 , 1 , 2 or 4 hrs , followed by recovery at 25°C for 1 hr . Animals were harvested and washed three times with M9 and 20 µl pellet animals were lysed directly in Laemmli Sample Buffer and used for western blot analysis . Proteins were resolved by 15% SDS-PAGE ( Bio-Rad , 4561084 ) and transferred to a nitrocellulose membrane ( Bio-Rad , 1620167 ) . Proteins were detected using antibodies against Flag ( Sigma , F3165 ) , mCherry ( M11217 , Life Technologies ) , Tubulin ( Sigma , T5168 ) , H3 ( Abcam , ab1791 ) or HSP90 ( Proteintech , 13171–1-AP ) . For subcellular fractionation , 50 µl pellet animals were resuspended in 150 µl 1 X cell lysis buffer ( Cell Signaling Technology , 9803S ) with protease inhibitor cocktail ( BioTools , B14002 ) and 10 µM PMSF , and incubated for 10 min on ice . Animals were lysed by TissueRuptor ( Qiagen , 9001271 ) with Motor unit ‘6’ for 30 s on ice . After incubation on ice for 5 min and centrifugation at 5 , 000 rpm at 4°C for 2 min , the supernatant was collected as the cytoplasmic part . The nuclear pellet was washed three times with lysis buffer and resuspended in 150 µl RIPA buffer ( Thermo Fisher Scientific , P89900 ) for 30 min on ice , spun at 12 , 000 rpm for 15 min , and the supernatant was collected as nuclear extract . Tubulin and H3 were separately used as cytoplasm and nuclear loading control . For RNAi experiments , zip-10p::zip-10::EGFP::FLAG animals were bleached , and the eggs were laid onto RNAi plates . Animals were harvested as L4/young adults and subject to western blot analysis as described above . ChIP-QPCR assay was carried out as before with modifications . Briefly , CW-treated animals ( 4°C for 4 hrs , recovered at 25°C for 1 hr ) and control ( 25°C ) animals were harvested and washed by 1 X PBS . The pellet animals were resuspended in cross-linking buffer ( 1% formaldehyde in 1 X PBS ) followed by homogenization using TissueRuptor with Motor unit ‘4’ for 1 min at room temperature . The process was then stopped by addition of glycine ( 125 mM final concentration ) . After washing and discarding the supernatant , the pellet was resuspended in lysis buffer and lysed by TissueRuptor with Motor unit ‘6’ for 1 min on ice , with lysate kept on ice for additional 3 min , and then repeated three times . The lysate was centrifuged to collect the supernatant and one percent of the aliquot was used as ‘Input’ . Lysate was precleared by adding salmon sperm DNA/protein-A agarose beads ( Bioworld , 20182011–1 ) , rotating at 4°C for 1 hr . After centrifugation , supernatant was divided equally and added with 50 µg Flag antibody ( Sigma , F3165 ) and mouse IgG ( Santa Cruz Biotechnology , sc-2025 ) , respectively . The samples were incubated and rotated overnight at 4°C . Next , salmon sperm DNA/protein-A agarose beads were added for 2 hrs at 4°C . The beads-antibody-TF-DNA complex was washed extensively and the complex and input were diluted with proteinase K buffer . The samples were then incubated at 55°C for 4 hrs and then at 65°C overnight to reverse crosslink . DNA was extracted by phenol-chloroform-isoamylalcohol ( Sigma-Aldrich , 77617 ) . asp-17 promoter was measured by QPCR and calculated by the percent input method . Primers for ChIP-QPCR: asp-17 promoter ( Forward , TTCGCTGCACCTATATGTTG; Reverse , CCGCTAATACCCTTATCAC ) . RNA immunoprecipitation ( RIP ) -QPCR assay was carried out as before with modifications to accommodate our reagents ( Kershner and Kimble , 2010 ) . Briefly , synchronous day-1 isy-1p::isy-1::mcherry animals were divided into two groups . One group is control ( 25°C ) and the other is cold-warming ( 4°C for 4 hrs , recovered at 25°C for 1 hr ) . Animals were harvested and washed by M9 buffer until the supernatant was clear , and then washed once in buffer A and twice in lysis buffer . About 250 μl worm pellets were frozen in liquid nitrogen twice and homogenized using TissueRuptor with Motor unit ‘4’ for 1 min on ice . The lysate was kept on ice for 15 min and centrifuged to collect the supernatant and 1% of the aliquot was kept as ‘Input’ . Equal amount of supernatant was added with RFP-Trap_MA ( Chromotek ) and rotated for 4 hrs at 4°C . IP magnetic agarose beads were washed and 10% of IP beads were boiled for 6 min in 2X Laemmli Sample Buffer . RNA was eluted from remaining beads using 200 μl lysis buffer of Quick-RNA MiniPrepkit ( Zymo Research , R1055 ) and extracted following the instruction . RNA was quantified with a Nanodrop device . 500 ng RNA was reverse transcribed into cDNA and quantified by the percent input method . Primers for RIP-qPCR: Primary mir-60 Forward TCGAAAACCGCTTGTTCTTG , Reverse CGATTTCTCAAGTCTTGAACTAG; cebp-1 Forward GATCCTTCGCAAGACAAGAC , Reverse CACATTGTCGGTAGGAACGTC . Animals were cultured under non-starved conditions for at least 4 generations at 25°C before cold tolerance assay . For cold tolerance assay of L1-stage animals , bleach-synchronized populations were kept at 4°C for 96 hrs and then recovered for 4 hrs at 25°C . For cold tolerance assay of adults , animals were raised at 25°C from hatching with excessive bacteria food on agar plates . Well-fed L4 stage animals were transferred to new plates and kept at 25°C overnight to reach day-1 adulthood . To cold shock the animals , agar plates were spread with equal distance on a thin plastic board and transferred to a constant 4°C cold room for 48 hrs or the indicated duration . After cold shock , animals were then moved to 25°C for recovery for 4 hrs before scoring survival rates . Animals were scored as dead if they showed no pumping and movement upon light touch with the body necrosis subsequently confirmed . smFISH of C . elegans and imaging were performed as previously described ( Ji and Oudenaarden , 2005 ) . For fluorescence imaging , spinning-disc confocal and digital automated epifluorescence microscopes ( EVOS , Life Technologies ) were used to capture images of animals after RNAi or CW treatments . Synchronous population of worms were randomly picked and treated with 1 mM levamisole water solution to paralyze the animals . The animals were mounted on an agar pad on a slide and aligned for imaging . Identical conditions and settings were used for both control and test groups . For quantification of fluorescence images , the animals in the images were outlined and signals were quantified by ImageJ software . The intensity of an individual animal was obtained by dividing the total signal by the area of that animal . The average intensity of the control group was set to be 1 . 0 , to which all other intensities were normalized . Graphpad Prism software was used to plot the data . For small RNA sequencing , total RNA was isolated by the Quick-RNA MiniPrep kit ( Zymo Research , R1055 ) that yields total RNA including small RNAs ranging 17–200 nt . RNA samples extracted from triplicate N2 animals and isy-1 mutants were submitted to Beijing Genomics Institute for small RNA library construction and sequencing . The low-quality reads were filtered and clean reads were mapped to the C . elegans genome using Bowtie2 program ( Langmead and Salzberg , 2012 ) . MiRDeep2 ( Friedländer et al . , 2012 ) was used to characterize known and predict novel miRNAs . The small RNA expression level was calculated as TPM ( transcript per million ) . Differentially expressed small RNAs were detected by DESeq2 ( Love et al . , 2014 ) . The threshold for differentially expressed sRNAs was adjusted p-value≤0 . 05 and the absolute value of Log2ratio ≥1 . Targets of miRNAs were predicted by TargetScan ( Jan et al . , 2011 ) , RNAhybrid ( Krüger and Rehmsmeier , 2006 ) and miRanda ( John et al . , 2004 ) using default parameters . Data were analyzed using GraphPad Prism Software ( Graphpad , San Diego , CA ) and presented as means ± S . D . unless otherwise specified with p values calculated by unpaired Student's t-tests , one-way or two-way ANOVA ( comparisons across more than two groups ) and adjusted with Bonferroni's corrections .	Life on earth faces constant changes in temperature . Most warm-blooded animals like humans can maintain a fairly stable body temperature , but cold-blooded animals can experience drastic shifts in body temperature . For example , the body temperature of the worm Caenorhabditis elegans can vary greatly depending on its surroundings . This species has evolved an exquisite set of temperature-sensing machineries that can react even to subtle fluctuations , which enables the worm to adjust its behaviour . However , drastic shifts in temperature can cause significant changes within the organism . Transient exposure to heat can activate genes that help cells to repair damaged proteins , while cold shock can influence the production of proteins in the cell . Although C . elegans can tolerate short periods of stress , an extended exposure to extreme temperatures can kill the worm . Until now , it was not known how C . elegans responds to cold shock followed by warmer temperatures , also referred to as cold-warming . To address this question , Jiang et al . created random mutations in C . elegans and isolated the worms that responded to cold-warming differently . The results revealed a molecular pathway that turns on genes in response to cold-warming . Jiang et al . found that two genes and their proteins , ISY-1 and ZIP-10 , control which other genes are switched on or off in response to this temperature change . When the worms were exposed to cold-warming over a long period , the pathway remained active and many of the worms died , in particular older animals . These findings suggest that this genetic program might have evolved to help younger animals survive better when stress conditions are high and food resources limited . More work is needed to explore this new pathway and its implication in the heat-cold shock mechanisms . The affected genes are often the same across different organisms and can therefore be engineered to benefit research and medical applications in unexpected ways . For example , patients suffering a heart attack or brain injury are exposed to colder temperature to prevent the risk of tissue injuries once the blood flow goes back to normal . Therefore , the findings of this study may help us to understand how human cells respond to and are protected by low temperature .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "chromosomes", "and", "gene", "expression" ]	2018	A genetic program mediates cold-warming response and promotes stress-induced phenoptosis in C. elegans
Human lung adenocarcinomas ( LUAD ) contain mutations in EGFR in ∼15% of cases and in KRAS in ∼30% , yet no individual adenocarcinoma appears to carry activating mutations in both genes , a finding we have confirmed by re-analysis of data from over 600 LUAD . Here we provide evidence that co-occurrence of mutations in these two genes is deleterious . In transgenic mice programmed to express both mutant oncogenes in the lung epithelium , the resulting tumors express only one oncogene . We also show that forced expression of a second oncogene in human cancer cell lines with an endogenous mutated oncogene is deleterious . The most prominent features accompanying loss of cell viability were vacuolization , other changes in cell morphology , and increased macropinocytosis . Activation of ERK , p38 and JNK in the dying cells suggests that an overly active MAPK signaling pathway may mediate the phenotype . Together , our findings indicate that mutual exclusivity of oncogenic mutations may reveal unexpected vulnerabilities and therapeutic possibilities . Large-scale sequencing of cancer genomes has provided a unique opportunity to survey and interpret the genotype of common and rare tumors . These efforts have revealed mutations in well-known tumor suppressor genes and proto-oncogenes; in genes with normal functions not previously associated with neoplasia ( such as RNA splicing and chromatin modification ) ; and in genes unlikely to have any role in carcinogenesis ( putative ‘passenger mutations’ ) ( Kandoth et al . , 2013; Hoadley et al . , 2014 ) . In several tumor types , genomic studies have revealed alterations in specific genes or signaling pathways that are highly associated with tumor origins , such as mutations affecting HIF-1 signaling in renal clear cell carcinoma ( Cancer Genome Atlas Research Network , 2013 ) , in the Wnt signaling pathway in colorectal carcinoma ( Cancer Genome Atlas Network , 2012 ) , and , more broadly , in the growth factor receptor-RAS-PIK3CA or–AKT pathways in a variety of cancers including lung adenocarcinoma ( Kandoth et al . , 2013; Cancer Genome Atlas Research Network , 2014 ) . These studies have been vital for understanding the genetic mechanisms driving tumorigenesis and revealing new targets for therapeutic intervention . However , these initial analyses are just beginning to explore more complex issues such as the co-incidences and temporal sequences of mutations , which may reveal processes driving tumor evolution and influence new strategies for targeted therapy ( Wong et al . , 2014 ) . For example , numerous investigators have noted the apparent ‘mutual exclusivity’ of oncogenic alleles of well-known proto-oncogenes in certain types of cancer , but , aside from a few instances ( Petti et al . , 2006; Sensi et al . , 2006 ) , without experimentally verified explanations . One of the first and most apparent of these mutually exclusive mutational combinations involves two well-studied proto-oncogenes , KRAS and EGFR , in human lung adenocarcinomas ( LUAD ) . This finding is especially provocative because of the frequencies of mutant alleles of KRAS and EGFR occurring separately in LUAD: ∼30% for KRAS mutations and ∼15% for EGFR mutations ( Cancer Genome Atlas Research Network , 2014 ) . The explanation generally provided for the mutual exclusivity is that the products of these two loci operate in the same or overlapping signaling pathways and hence are functionally redundant . However , this idea has not been experimentally tested , and there is reason to question its validity . For example , since they are differently positioned in a signaling network ( EGFR senses and transmits external signals from the cell membrane; RAS serves as a cytosolic node ) , the consequences of the mutations would not be expected to be identical . Hence mutations in both genes might be expected to offer a selective advantage over a mutation in just one of them . Furthermore , and tellingly , KRAS mutations are rarely , if ever , encountered in mutant EGFR-driven tumors that become resistant to treatment with tyrosine kinase inhibitors ( TKIs ) , when a ‘downstream’ mutation in KRAS would be expected to provide a strong selective advantage ( Yu et al . , 2013 ) . One way to reconcile these observations is to propose that oncogenic mutation of both of these genes in the same cell confers an unexpected inhibitory phenotype , such as synthetic lethality , that is selected against during tumor development . In this study , we aimed to determine whether an inhibitory phenotype can account for the mutual exclusivity of KRAS and EGFR mutations in LUAD and to characterize the specific cellular effects that result from co-expression of both mutant alleles in lung cancer cells . Through analysis of available sequences of human tumor samples , generation of transgenic mouse models that express mutant KRAS and EGFR in the lung epithelium , and functional tests of these mutant genes in cultured tumor cells , we conclude that synthetic lethality is responsible for the mutually exclusive nature of activating mutations in these genes . We propose that this type of gene interaction may reveal vulnerabilities in lung and other cancers that suggest novel therapeutic strategies . Although oncogenic mutations in KRAS and EGFR occur in a significant fraction of human LUAD , they are rarely—if ever—observed together in the same tumor . Since these two genes are in the same pathway and activate similar downstream targets , it is generally assumed that there is no selective pressure to favor cells with both mutations over cells with a mutation in one of them . If complete functional redundancy explains the mutual exclusivity of these mutations , assessing a large panel of samples might reveal tumors with both genes mutated . To this end , we reviewed published somatic mutation data for EGFR and KRAS from 662 LUAD that were profiled as part of four separate studies ( Ding et al . , 2008; Imielinski et al . , 2012; Seo et al . , 2012; Cancer Genome Atlas Research Network , 2014 ) . We excluded data from studies that profiled only specific exons or mutation hotspots in these two genes , as restricted sequencing might not detect recently characterized oncogenic mutations such as those in exon 20 of EGFR and codon 61 of KRAS or changes in codons not associated with gene activation . In total , 186 ( 28 . 2% ) and 110 ( 16 . 6% ) tumors had somatic mutations that changed the amino acid sequences ( non-silent mutations ) of KRAS and EGFR proteins , respectively—frequencies similar to those previously reported ( Heist and Engelman , 2012 ) . Moreover , all mutations in KRAS and 93 . 6% of EGFR mutations were identified as oncogenic ( see ‘Materials and methods’ , Supplementary file 1 ) . Attempts to estimate the expected frequency of finding coincident mutations in any two genes in a tumor cell population are influenced by a variety of potentially confounding factors , such as the size and transcriptional activity of the genes; mutation rates for cells in the tumor lineage and for individual genes; the point in the evolution of a tumor when a mutation occurs ( since that will affect the percentage of cells containing the mutant allele ) ; and the selective advantage or disadvantage of the mutation for the growth and survival of the cell in which it occurred ( Lawrence et al . , 2013 , 2014 ) . These factors make predictions complex . In the simplest terms , if we define mutations in these genes as independent events with similar selective advantage that occur at a specific frequency across the population of tumors—and assume that they are not influenced by the factors stated above—the expected co-incidence of the mutations can be derived from a simple calculation ( used in Figure 1 , panel A ) . On these terms , we would anticipate that that approximately 31 tumors in the cohort presented in Figure 1 would contain mutations in both genes . However , we observed only a single tumor in which this was the case; moreover , the EGFR mutation ( R574L ) in this tumor ( which also harbored a KRASG12V mutation ) has not been previously identified as oncogenic , lies outside the kinase domain and hence is unlikely to be activating ( Figure 1A ) . A left-tailed 2X2 Fisher's Exact Test indicates a significant negative association between the mutations ( p = 8 . 62e−17 ) , confirming the mutually exclusive nature of these mutant alleles . Furthermore , even when analyzing tumors from smokers and never-smokers separately—which are known to have different frequencies of mutations in KRAS and EGFR—a statistically significant negative association between the co-occurrence of mutations in these genes was still observed ( p = 4 . 43e−10 and p = 0 . 002 , respectively ) ( Figure 1 ) . Our failure to identify any tumors with co-existing oncogenic mutations of these two genes in this combined data set strengthened the possibility that the observed mutual exclusivity might imply more than functional redundancy—for example , a toxic effect of co-existing oncogenic mutations . 10 . 7554/eLife . 06907 . 003Figure 1 . Mutations in KRAS and EGFR in lung adenocarcinoma are mutually exclusive . ( A ) Coincidence of EGFR and KRAS mutations in 662 lung adenocarcinomas ( LUAD ) . Numbers are presented for all tumors combined and for tumors found in smokers or never smokers . ‘Expected’ values are for co-occurrence of mutations based on the frequency of mutation of either gene alone ( including assumptions about selective advantage and other features as described in the text ) . ( e . g . , in all LUAD , KRAS is mutant in 28 . 2% and EGFR is mutant in 16 . 6% , so the expected number of co-occurrences is 30 . 9 cases or 0 . 282 × 0 . 166 × 662 . ) . p-values were calculated by a left-tailed 2 × 2 Fisher's Exact Test to assess significance of negative association . The single tumor with co-incident non-silent mutations contained a non-activating EGFRR574L mutation . ( B–H ) Frequencies of coincidence of somatic silent ( synonymous point mutations ) or non-silent mutations ( non-synonymous point mutations and small insertions or deletions that affect protein coding sequence ) in coding sequences of other genes co-occurring with non-silent mutations in proto-oncogenes ( EGFR , KRAS , PIK3CA , or BRAF ) in 520 LUAD or other cancer types as noted below for individual panels . In these panels , each point represents a unique gene but data points overlap when multiple genes are mutated in the same number of tumors . Red dots represent genes for which non-silent mutations are positively associated with proto-oncogene mutations in the tumors ( Right side Fisher's Exact Test p ≤ 0 . 01 ) ; green dots represent genes for which mutations are negatively associated with mutant proto-oncogenes ( Left side Fisher's Exact Test p ≤ 0 . 01 ) . Yellow dots represent some important genes that are co-mutated at the expected rate . Genes of special interest are indicated by the arrows . ( B ) Co-occurrence of silent mutations in other genes with non-silent KRAS mutations in LUAD . ( C ) The similar analysis for non-silent mutations co-occurring with KRAS mutations . ( D , E ) As in panel C , but with mutant EGFR and PIK3CA in LUAD . ( F–H ) Additional analyses for mutations co-existing with mutant KRAS in colorectal adenocarcinoma ( COAD ) , F; with mutant EGFR in glioblastoma multiforme ( GBM ) , G; and with mutant BRAF in skin cutaneous melanoma ( SKCM ) , H . For presentation purposes , not all gene points are displayed . Complete lists of significant gene pairs from each of analysis can be found in Supplementary file 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 06907 . 003 Because the co-existence of mutations may be affected by numerous factors ( e . g . , mutation rates , sensitivity of detection , etc , as detailed above ) and because relevant data sets are relatively small , we looked more broadly for combinations that might provide a selective advantage or disadvantage , including mutual exclusivities , in LUAD . To do this , we devised a method that explores whole-exome DNA sequencing data on a global scale . Tumors were stratified based on the somatic mutation status of a known proto-oncogene ( e . g . , KRAS or EGFR , as well as other genes ) . The number of tumors containing mutations in both the specified proto-oncogene and every other gene in the genome was then calculated on a gene-by-gene basis ( Figure 1B–H ) . In the absence of selective pressures resulting from a second mutation , it would be anticipated that the more times a gene is mutated in the sample set , the more often it should co-occur in a tumor with a mutation in the proto-oncogene . Plotting these values on the y-axis against the total number of tumors with a mutation in those second genes on the x-axis would be expected to reveal a linear relationship . However , if selective pressures are involved , so that dual mutation enhances or restricts cell growth or survival , the value for a given gene will fall either above or below this line . If the additive effect confers a selective advantage , and a mutation of the second gene occurs more frequently in combination with the mutant proto-oncogene than expected , the result is displayed as a red dot ( Figure 1; see legend for details ) . If a mutation in the second gene occurs less frequently than expected in combination with the mutant proto-oncogene , it is presumed to provide a selective disadvantage and the result is shown as a green dot in Figure 1 . From these so-called ‘exclusivity plots’ , left and right 2 × 2 Fisher's Exact Tests can be performed between mutations in each gene and the proto-oncogene to confirm the degree of enhanced or diminished co-occurrence and , correspondingly , the potential degree of positive and negative selection conferred by the combinations . However , the confidence in any selective effect , reflected by a low p-value , should not be confused with the strength of the effect , since the number of mutations will influence the p-value . Therefore , since the numbers of mutations for the great majority of genes are relatively small in our data sets , we can place only a modest degree of confidence on the selective effects . However , in some cases , for example , TP53 plus KRAS mutations , there may be a highly significant but still relatively minor deviation from expected rates of co-occurrence . The deviations of greatest interest are therefore caused by those combinations of mutations , such as KRAS plus EGFR , that produce a major deviation from expectation at a high level of confidence because both genes are frequently mutant ( Figure 1 and Supplementary file 2; discussed further in ‘Materials and methods’ ) . To test the validity of this approach , we first assessed the frequency of synonymous mutations in many genes and their co-occurrence with non-silent mutations in KRAS—non-synonymous point mutations and small insertions and deletions—in LUAD . ( Details of the analysis are provided in ‘Materials and methods’ . ) Synonymous mutations—point mutations that do not change the amino acid sequence of a protein , also referred to here as ‘silent’ mutations—should be neutral ( i . e . , offer no selective advantage or disadvantage ) . Thus the more frequently a given gene contains a silent mutation , the more frequently such mutations should co-occur with a KRAS mutation , so that the number of co-mutations should increase in a linear fashion as described above . To test these possibilities , we used whole-exome sequence data from an expanded set of 520 LUAD profiled as part of The Cancer Genome Atlas ( TCGA ) initiative ( including 229 that were used in Figure 1A ) . Of these tumors , 145 had a somatic , non-silent mutation in KRAS . Silent mutations in any other gene co-occurred with a KRAS mutation as expected based on their respective frequencies ( Figure 1B ) . For example , TTN , one of the largest genes in the genome , is known to have a very high mutation rate in lung cancer ( Lawrence et al . , 2014 ) . This gene incurred frequent silent mutations in lung adenocarcinoma ( indicated by a yellow dot ) that overlapped with mutations in KRAS , much as anticipated ( Figure 1B ) . When this analysis was repeated , however , with non-silent mutations in genes , numerous statistically significant outliers were uncovered . For these genes , the number of mutations co-occurred more frequently or less than expected in concert with non-silent mutations in KRAS ( Figure 1C , Supplementary file 2 ) . EGFR is the gene with a co-mutation rate most significantly different from the expected number , demonstrating a negative association with mutant KRAS . Likewise , when this analysis was reversed by stratifying samples based on non-silent EGFR mutations ( n = 67 ) , KRAS is the only major outlier , demonstrating a similar negative association ( Figure 1C ) . The apparently negative association of KRAS and EGFR mutations is not a general phenomenon among known lung cancer proto-oncogenes . For example , PIK3CA mutations co-occurred as frequently as anticipated with either KRAS or EGFR mutations ( as indicated by the yellow dot in Figure 1C , D ) . Likewise , non-silent mutations in other genes , including known oncogenes , were never observed to have a negative association with PIK3CA mutations in the same sample set , although several such mutations had a slight positive association ( Figure 1E ) . Numerous genes that showed a similar frequency of non-silent mutations as EGFR and KRAS among the 520 samples co-occurred with mutations in KRAS or EGFR , within the range expected by chance in the absence of selective pressures ( Figure 1C , D ) . This indicates that potential confounding factors , such as mutation rates and sensitivities of sequencing methods , are unlikely to be responsible for the data implying the exclusivity of mutations in KRAS and EGFR; a more general skewing of the data would be anticipated if this was the case . Furthermore , although one tumor contained mutations in both KRAS and EGFR , the KRAS mutation , D33E , is not known to be activating . ( Note that the tumor with KRASG12V and EGFRR574L in Figure 1A was not part of this expanded dataset . ) This suggests that mutations in both genes can occur and be detected by sequencing , but that activated alleles of both genes very rarely or never co-exist . Our analysis also revealed that mutations in a few genes—such as STK11 ( also known as LKB1 ) and ATM , genes known to co-operate with mutant KRAS during lung tumorigenesis ( Ji et al . , 2007; Efeyan et al . , 2009 ) —are positively associated with KRAS mutations ( Figure 1C ) . The identification of genes known to provide a selective advantage to cancer cells driven by mutant KRAS provides strong evidence of the ability of this approach to uncover gene combinations under selective pressure during tumor evolution . Together , the analysis in Figure 1 revealed that no other mutated genes in lung adenocarcinoma demonstrate such a strong mutually exclusive relationship , as do mutant KRAS and EGFR , further suggesting negative selection against the combination during lung tumorigenesis . To determine whether mutual exclusivity between oncogenic alleles is limited to KRAS and EGFR in lung adenocarcinoma or whether it is a more general phenomenon affecting other cancer lineages , we examined sequencing data from other cancer types with the methodology employed in Figure 1 , seeking pairs of mutant proto-oncogenes that are disfavored or incompatible with tumorigenesis . First , we focused on cancer types known to have frequent activating mutations in KRAS or EGFR: colorectal adenocarcinoma ( COAD ) and glioblastoma multiforme ( GBM ) , respectively ( Cancer Genome Atlas Research Network , 2008; Cancer Genome Atlas Network , 2012 ) . ‘Exclusivity plots’ for COAD revealed mutations in BRAF to be negatively associated with mutations in KRAS , to a degree similar to that we observed for co-existing mutations of EGFR and KRAS in lung adenocarcinoma ( Figure 1F ) . Likewise , mutations in NF1 , a suppressor of RAS activity , were negatively associated with EGFR mutations in GBM , a tumor type in which activating deletions in the extracellular domain of EGFR are typically found ( Figure 1G ) . Looking beyond combinations involving mutant EGFR or KRAS , we found that BRAF mutations in skin cutaneous melanoma ( SKCM ) were negatively associated with NRAS ( Figure 1H ) . These findings suggest that certain tumor cells might not tolerate mutations in combinations of certain genes whose products operate within the EGFR-RAS-RAF signaling pathway . The analysis presented in Figure 1 supports the idea that certain pairs of mutually exclusive mutations may be selected against during tumorigenesis in some cell lineages , especially the combination of EGFR and KRAS oncogenic mutations in LUAD . To test this hypothesis experimentally , we took advantage of genetically engineered mice that develop LUAD when TetO-regulated transgenes encoding murine Kras-G12D ( KrasG12D ) or human EGFR-Deletion Exon 19 ( EGFRDEL ) are induced by doxycycline , which activates the rtTA transcriptional regulator expressed in type II airway epithelial cells through a Clara Cell Secretory Protein promoter ( CCSP ) ( Fisher et al . , 2001; Politi et al . , 2006 ) . We attempted to express both transgenic oncogenes in the same cells by feeding doxycycline to tri-transgenic mice ( CCSP-rtTA;TetO-KrasG12D;TetO-EGFRDEL , referred to as CCSP;KrasG12D:EGFRDEL ) that encode both of the oncogenic proteins and rtTA . If co-expression of mutant Kras and mutant EGFR confers a selective advantage , tumors should appear earlier and progress more quickly than in bi-transgenic animals ( CCSP;KrasG12D or CCSP;EGFRDEL ) harboring only one oncogenic transgene . If the combination has no advantage or disadvantage because of functional identity of mutant KRAS and EGFR , tumorigenesis should be no different than observed in lines that contain only one of the oncogenes . If the combination is deleterious , tumors should be prevented or grow slowly . We found that LUAD appeared in tri-transgenic mice after doxycycline-induction with the properties ( time of detection , rate of progression , and histological appearance ) observed in bi-transgenic CCSP;KrasG12D mice ( Figure 2A , see Supplementary file 2 for genotypes of all animals ) . This result would appear to support the hypothesis that the two mutant oncogenes are functionally redundant . However , the majority of tumors found in the tri-transgenic animals expressed only one of the TetO-oncogenes , as judged by RT-PCR ( Figure 2B ) . Moreover , the levels of oncogenic RNA in tumors from tri-transgenic mice were similar to those observed in tumors from their bi-transgenic littermates . When RNA from both oncogenic transgenes was detected in occasional samples , the relative amounts varied , suggesting that the samples contained more than one tumor clone or adjacent non-transformed cells , as examined in greater detail below . 10 . 7554/eLife . 06907 . 004Figure 2 . Co-induction of mutant Kras and mutant EGFR in the mouse lung epithelium leads to the development of LUAD that express a single oncogene . ( A ) Survival curves for mono-transgenic mice expressing only rtTA from the clara cell specific promoter ( CCSP , black line ) ; bi-transgenic mice expressing mutant EGFR ( CCSP;EGFRDEL , green line ) ; bi-transgenic mice expressing mutant Kras ( CCSP;KrasG12D , red line ) ; and tri-transgenic mice ( CCSP;EGFRDEL;KrasG12D , blue line ) . ( B ) Levels of transgene-specific RNA from normal lung tissue ( for CCSP only ) and tumor nodules ( for oncogene containing transgenics ) as measured by gel electrophoresis of products of RT-PCR reactions primed by oligonucleotides specific for KrasG12D , EGFR and rtTA RNA , prepared from animals encoding only rtTA ( CCSP; black ) , KrasG12D ( CCSP;KrasG12D; red ) , EGFRDEL only ( CCSP;EGFRDEL;green ) and EGFRDEL plus KrasG12D ( CCSP;EGFRDEL;KrasG12D;blue ) . Mice are indicated by the initial number; the number following the dash represents the nodule or tissue number from that mouse . ( C ) EGFRDEL and KrasG12D RNA levels measured by qRT-PCR from normal lung ( for CCSP mice only ) and from tumor nodules ( for all mice containing transgenic oncogenes ) . Data for each nodule are represented in a ‘heat map’ with the expression levels shown relative to the average of the respective bi-transgenic animals for each transgene ( see ‘Materials and methods’ ) . Mice and nodules are identified as above . ( D–I ) Histological appearance ( H&E staining ) and IHC detection of relevant proteings ( human EGFR , phospho-ERK1/2 and SPC ) in selected mouse lung nodules also examined in panel C . Mouse and nodule numbers are indicated as before along with the respective genotype for each mouse; the genotype is also indicated by the color of the line below each label by the color scheme for genotype according to the key used in panel C . ( J ) IHC for human EGFR in a lung lobe from a tri-transgenic mouse , indicating regions of distinct staining within the same region of tumor . DOI: http://dx . doi . org/10 . 7554/eLife . 06907 . 004 The findings in Figure 2 could be explained by a low frequency of dual induction of expression of two oncogenic transgenes in lung cells . However , our laboratory has previously demonstrated in a mouse model of breast cancer that two oncogenic transgenes , TetO-KrasG12D and TetO-Myc , can be co-expressed in mammary tumors at levels similar to those observed in tumors from transgenic animals containing only one of the transgenes ( Podsypanina et al . , 2008 ) . Furthermore , in other tri-transgenic mice carrying the same CCSP-rtTA transgene , most if not all tumors expressing the oncogenic TetO-EGFRL858R transgene also expressed a TetO-Cre transgene , as judged by elimination of a floxed Erbb3 allele ( Song et al . , 2015 ) . These findings suggest that our observations here reflect the incompatibility of mutant Kras and EGFR and not an inability to express both doxycycline-dependent transgenes . Since TetO-KrasG12D is the more potent oncogenic transgene in our mouse model , we supposed that if only one oncogenic transgene were expressed it would be the Kras transgene . But we initially found that two lung tumors from one tri-transgenic mouse preferentially expressed Kras and two tumors from a different tri-transgenic mouse expressed only EGFR ( Figure 2B ) . Expanding the use of the qRT-PCR assay to examine over 90 tumor nodules from 14 mice , 8 of which were tri-transgenics , confirmed that tumors from tri-transgenic mice almost always contained RNA from a single transgenic oncogene , either EGFR or Kras , and the amounts of RNA were similar to those found in tumors from the bi-transgenic animals ( Figure 2C ) . In some tri-transgenic mice ( e . g . , mouse number 687 ) , all nodules contained only EGFR transgenic RNA while nodules from some others expressed only Kras transgenic RNA ( e . g . , mouse number 685 ) . Furthermore , there were animals in which a different transgene , Kras or EGFR , was expressed in individual tumor nodules ( e . g . , mouse number 798 ) , implying that tumors from mice carrying oncogenic alleles of both Kras and EGFR have evolved by selecting against cells that express both mutant transgenes , further suggesting that expression of both these genes is detrimental . As shown in Figure 2 , some tumor nodules appear to contain both Kras and EGFR transgenic RNAs , but we could not determine unequivocally whether this means that a few tumors expressed both oncogenes or that multiple tumors may have grown too closely together to separate during macrodissection . To address the latter possibility , we used immunohistochemistry ( IHC ) to ask whether EGFR and Kras transgenes are expressed in the same cells by assessing levels of human EGFR and by measuring phosphorylated ERK ( p-ERK ) as readout of MAP kinase ( MAPK ) signaling . This analysis , however , cannot definitely address co-expression as the intensity of p-ERK signaling may be enhanced by induction of either transgene alone . Tumors from bi-transgenic animals with the EGFR transgene scored positive for both human EGFR and p-ERK ( Figure 2E ) , while tumors from bi-transgenic Kras animals scored positive only for p-ERK ( Figure 2D ) . Some individual tumor nodules from tri-transgenic animals contained human EGFR and were positive for p-ERK ( Figure 2G ) , but nodules that appeared to lack human EGFR scored positive for p-ERK , implying that they express only the Kras oncogene ( Figure 2F ) . Furthermore , we confirmed that individual tumor nodules from one tri-transgenic animal might or might not contain human EGFR ( Figure 2H ) and if they did not , they still contained p-ERK , implying expression of Kras ( Figure 2I ) . Finally , an entire lung lobe from a tri-transgenic animal revealed adjacent regions of adenocarcinoma with different patterns of human EGFR staining , suggesting that distinct cell populations expressed either Kras or EGFR ( Figure 2J ) . The proximity of these regions likely explains the detection of both transgenic RNAs by qRT-PCR in some instances . Together , the data from tri-transgenic animals suggest negative selection against co-expression . The findings with transgenic mice bearing inducible Kras and EGFR oncogenes imply that co-production of mutant EGFR and mutant KRAS is not tolerated in the same cell . To confirm and further explore this conclusion , we directly induced the expression of mutant KRAS or mutant EGFR in established lines of human lung adenocarcinoma cells known to be driven by either mutant EGFR ( PC9 cells; in-frame deletion in exon 19 ) or by mutant KRAS ( H358 cells; G12C ) ( Arao et al . , 2004; Sunaga et al . , 2011 ) . For this purpose , we cloned mutant KRAS ( G12V ) , mutant EGFR ( L858R ) , and ( as a control ) GFP into doxycycline-inducible vectors ( pInducer , [Meerbrey et al . , 2011] ) and established stable cell lines that express mutant KRAS or GFP in PC9 cells and mutant EGFR or GFP in H358 cells in response to doxycycline ( Dox ) . In this single-vector system , a tetracycline-responsive promoter controls the genes while the tetracycline transactivator , rtTA , is expressed by the constitutive Ubc promoter ( Meerbrey et al . , 2011 ) . Cells treated for 24 hr with Dox ( 100 ng/ml ) produced the appropriate proteins , suggesting that this system could allow us to describe the consequences of oncogene co-expression ( Figure 3A ) . 10 . 7554/eLife . 06907 . 005Figure 3 . Co-expression of mutant KRAS and mutant EGFR decreases viability of human lung adenocarcinoma cell lines . ( A ) Induced expression of transduced genes in human lung cancer cell lines . PC9 and H358 cell lines were transduced with the indicated tetracycline-responsive plasmids . Lysates of cells cultured in the presence or absence of doxycycline ( Dox ) for 24 hr were prepared and assayed for the indicated protein expression by western blotting , as described in Materials and Methods . ( B ) Co-expression of mutant KRAS and EGFR reduces cell viability . Cells were grown in the presence or absence of Dox for up to 7 days and tested for viability by alamar blue . Averaged values from three independent experiments were normalized and plotted for each cell line relative to untreated cells ( no Dox ) at the indicated time points . Error bars represent ± standard deviation ( SD ) for each point . p-values were calculated between the + and − Dox states of individual cell lines at each time point using a two-tailed , unpaired t test with Welch's correction . , and * represent significance values <0 . 01 , <0 . 001 and <0 . 0001 , respectively . ( C ) Co-expression of mutant KRAS and EGFR reduces cell number . Cells were grown as in panel B and counted on day 7 . Average cell number from three independent experiments were normalized and plotted for each cell line relative to cells expressing TetO-GFP in the absence of Dox . Error bars represent ± standard error of the mean ( SEM ) and p-values were calculated between the + and − Dox states as described in A . ( D ) Erlotinib protects PC9-TetO-KRASG12V cells from the toxic effects of oncogene co-expression . Cells were grown with or without erlotinib ( Erl ) and/or Dox for 7 days and cell viability determined by alamar blue . Results from three independent experiments were normalized and plotted for each cell line relative to untreated cells ( no Dox ) with error bars representing ± SEM . p-values were calculated between the groups indicated as in A . DOI: http://dx . doi . org/10 . 7554/eLife . 06907 . 005 All four of the modified cell lines were maintained in the presence or absence of Dox for 7 days , and cell viability assessed on multiple days by incubating with the vital dye , alamar blue ( Figure 3B ) . By day 7 , there was a marked decrease in viable PC9 or H358 cells producing the additional mutant oncoprotein . Cells that produce GFP after exposure to Dox were not affected in this assay ( Figure 3B , C ) . As a result , the numbers of cells in these cultures were significantly decreased by 7 days after induction ( Figure 3C ) . These results support the hypothesis that co-expression of mutant EGFR and mutant KRAS is incompatible with cell survival . To determine whether the toxic effects of the co-production of mutant EGFR and mutant KRAS depend on the tyrosine kinase activity of EGFR , we added erlotinib , an EGFR TKI used to treat human LUAD ( Pao et al . , 2004 ) , to PC9 cells induced to express either the KRAS mutant or GFP . In the absence of mutant KRAS , erlotinib markedly decreased the number of PC9 cells , regardless of whether GFP was induced by Dox . However , erlotinib significantly protected PC9 cells from the toxic effects of induced mutant KRAS expression at day 7 ( Figure 3D ) . Thus , the induction of mutant KRAS ( but not induction of GFP ) rescues PC9 cells from the lethal effects of erlotinib , implying that the toxicity of co-expression of mutant KRAS and EGFR depend on the kinase activity of mutant EGFR . In addition , these findings confirm the expected ability of mutant RAS to render EGFR-mutant adenocarcinoma cells resistant to TKI's and further enhance the significance of not observing oncogenic KRAS mutations in human tumors resistant to erlotinib ( Ohashi et al . , 2012 ) ( see ‘Discussion’ ) . Having established that co-expression of mutant KRAS and EGFR oncogenes affects cell survival , we asked whether known mechanisms of cell death , such apoptosis and autophagy , were responsible for the observed consequences of co-expression . On day 5 after induction of the second oncogene in either cell line , both H358 and PC9 cells showed reduced viability and exhibited increases in common indicators of apoptosis , cell surface staining for Annexin V and 7-AAD permeability ( Figure 4A ) . The increase in these markers of apoptosis was more evident in PC9 cells producing mutant KRAS than in H358 cells producing mutant EGFR . This was also true when measuring PARP cleavage , a marker of caspase activity ( Figure 4B ) . This suggests that conventional apoptosis may play a role in the decreased viability in both scenarios of oncogene co-expression , but is more pronounced in the modified PC9 than in H358 cells . 10 . 7554/eLife . 06907 . 006Figure 4 . Effects of mutant KRAS and EGFR co-expression . ( A ) Measurement of apoptosis after co-expression of mutant oncogenes . The indicated cells were grown for 5 days with or without Dox and assessed for apoptosis by flow cytometry for Annexin V and 7-AAD . Data are plotted as either Annexin V positive/7-AAD negative ( AV+/7-AAD− , early apoptosis , black bars ) or Annexin V/7-AAD positive ( AV/7-AAD+ , late apoptosis , gray bars ) cells . ( B ) PARP cleavage after co-expression of mutant oncogenes . Cells were grown as in panel A ( 5 days ) and lysates assayed for cleavage of PARP protein by western blotting . Cells treated with staurosporine ( Stau ) for 24 hr were used as a positive control . ( C ) LC3A lipidation with co-expression of mutant oncogenes . Cells were grown as above ( 5 days ) and assayed for autophagy by measuring the levels of LC3-II , the faster migrating band of LC3 ( where applicable ) . Cells treated with chloroquine ( Chlor ) for 16 hr were used as a positive control . ( D ) Cell cycle analysis in cells co-expressing mutant oncogenes . Cells were grown for 3 and 5 days as indicated after induction of a second oncogene , and cell cycle status was determined by propidium iodide staining and flow cytometry . The fraction of cells in G1 , S and G2/M are gated as indicated . The percentages of cells in each cell cycle stage are plotted as stacked bars . Data from A and D are plotted as averages from cells grown in triplicate wells ± SEM all data are representative of multiple independent experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 06907 . 006 Autophagy is a process whereby intracellular proteins and whole organelles are catabolized in a special double-membraned structure termed the autophagosome . A key component of the autophagosome is LC3 , light-chain 3 , a microtubule-associated protein that is uniquely modified by phosphatidylethanolamine ( LC3-II ) ( reviewed in [Mizushima , 2007] ) . We observed an increase in the modified LC3-II ( the faster migrating protein in the gels shown in Figure 4C ) at day 5 in PC9s and to a lesser , but appreciable extent in H358 cells . This suggests that autophagy may also be involved in the loss of cell viability in these cells . To further characterize the effect of expressing both mutant KRAS and mutant EGFR , we followed cell cycle progression by measuring DNA content by propidium iodide staining . In H358 cells producing mutant EGFR , we did not observe a major difference in cell cycle stages ( G1 , S , or G2/M ) at days 3 or 5 after Dox addition ( Figure 4D ) . However , in PC9 cells producing mutant KRAS , we found an approximately twofold increase in cells at G2/M treated with Dox ( Figure 4D ) . Thus , to varying levels , oncogene co-expression affects apoptosis , autophagy and cell cycle . Some combination of these processes and those described below , contribute to cell toxicity . During the course of these studies , we observed distinctive changes in cell morphology after doxycyline-mediated induction of a second oncogene in PC9 and H358 lung cancer cell lines . Both H358-TetO-EGFRL858R and PC9-TetO-KRASG12V cells displayed ruffling of the cell membrane and large phase-translucent vacuoles ( Figure 5A ) , both more prominent with time . These features are similar to what has been described in glioblastoma cell lines undergoing a process recently termed methuosis ( Chi et al . , 1999; Overmeyer et al . , 2008 ) ( see ‘Discussion’ ) . To further explore the morphological changes in cells co-expressing EGFR and KRAS oncogenes , we observed the cells by transmission electron microscopy ( EM ) ( Figure 5B ) . The presence of large , single membrane , vacuoles were detectable in cells induced to express a second mutant oncogene . Microvilli were often present on the vacuolar membrane of these cells ( Figure 5B ) . 10 . 7554/eLife . 06907 . 007Figure 5 . Co-expression of mutant KRAS and EGFR induces morphological changes and increased macropinocytosis in lung adenocarcinoma cells . ( A ) Changes in morphology induced by co-expression of mutant oncogenes . Cells were grown for 5 days after induction of expression of a second oncogene , and phase contrast images were taken with indicated objectives . Arrows show a vacuolated cell and arrowheads show membrane ruffling . Images are representative of cell morphology during the analyses ( as early as day 1 and as late as day 7 after Dox ) . ( B ) Vacuolization in cells co-expressing mutant oncogenes . Cells were cultured for 5 days after induction of a second oncogene and analyzed by transmission EM ( TEM ) . Scale bars on lower right . Far right panels for each cell type shows a vacuolar structure bound by a single membrane ( black arrowhead ) . ( C ) Increased macropinocytosis with co-expression of mutant oncogenes . Cells were cultured for 3 days in the presence or absence of Dox and assayed for macropinocytosis by uptake of fluorescent 70 kDa Dextran for 30 min . Dextran uptake was measured by flow cytometry . A histogram of data from one experiment and the average MFI ( median fluoresence intensity ) from three independent experiments ± SEM are also shown . The shaded light gray histogram represents fluorescence intensity of cells that received neither Dextran nor Dox . p-values were calculated between the + and − Dox states using a two-tailed , unpaired t test with Welch's correction with * representing significance values <0 . 01 . DOI: http://dx . doi . org/10 . 7554/eLife . 06907 . 007 Prompted by our observations and published studies , we asked whether lung cancer cell lines expressing the two mutant oncogenes displayed features of macropinocytosis ( Overmeyer et al . , 2008; Commisso et al . , 2013; Kitambi et al . , 2014 ) . 3 days after Dox addition , PC9-TetO-KRASG12V and H358-TetO-EGFRL858R cells were incubated with fluorescent 70 kDa dextran to gauge their capacity to take up this macromolecule . Based on flow cytometry 30 min after incubation , both cell lines co-expressing the two oncogenes showed increased macropinocytosis relative to the untreated lines that naturally express mutant KRAS or mutant EGFR but not both ( Figure 5C ) . Thus , uncontrolled fluid phase uptake may be one of the pathological consequences of oncogene co-expression . To identify signaling pathways that may be responsible for the decreased cell viability and morphological changes induced by co-expression of mutant KRAS and EGFR , we generated gene expression profiles of the PC9 and H358 lung adenocarcinoma cells engineered to conditionally express mutant KRAS or EGFR . 24 hr after addition of Dox to PC9-TetO-KRASG12V and H358-TetO-EGFRL858R cells or to control lines with TetO-GFP we harvested RNA . Gene expression profiles were then compared to profiles from untreated cells . The 24 hr time point was chosen to measure changes in cell signaling likely to be caused directly by induction of the co-expressed oncogene rather than by reactive changes that could be attributed to secondary events occurring at later time points . Genes differentially expressed in cells treated or not treated with Dox were identified for each cell line . Those genes showing significant differences ( ANOVA Corrected p < 0 . 01 , compared to reciprocal no Dox control ) were examined in similar tests with cells induced to express TetO-GFP to determine those specifically affected by co-expression of the two oncogenes ( ‘Materials and methods’ ) . In total , 152 probe sets corresponding to 144 unique genes detected differential expression in both H358 and PC9 cells in response to mutant EGFR and KRAS ( Figure 6A , Supplementary file 3 ) . 10 . 7554/eLife . 06907 . 008Figure 6 . Co-expression of mutant KRAS and EGFR increases MAP kinase ( MAPK ) signaling . ( A ) Modified PC9 and H358 lung adenocarcinoma cells ( see Figure 3 ) were cultured in the presence or absence of doxycycline ( Dox ) for 24 hr and analyzed for global gene expression changes using Affymetrix microarrays ( see ‘Materials and methods’ for details ) . Microarray probes differentially expressed in each TetO cell line upon the addition of Dox were identified ( corrected p < 0 . 01 , compared to control without Dox ) , and those genes specifically induced by Dox in either TetO-KRAS or TetO-EGFR cells and not the TetO-GFP control cells were determined . The Venn diagram indicates the resulting number of gene probes identified in each cell line , including the 152 unique probes specifically modulated by expression of mutant KRAS and EGFR in both PC9 and H358 cells . ( B ) Gene Set Enrichment Analysis ( GSEA ) of the genes specifically regulated upon mutant KRAS and EGFR co-expression in both PC9 and H358 lung adenocarcinoma cells identified three oncogenic signatures that were significantly upregulated ( FDR q-value <0 . 01 ) upon co-expression; the top two are indicative of KRAS signaling ( see ‘Materials and methods’ and Supplementary file 5 ) . The displayed enrichment plot is for the most significant gene set ( q-value = 0 , Normalized Enrichment Score = 2 . 29 ) demonstrating enrichment for genes related to the upregulation of mutant KRAS . ( C ) Ingenuity Pathway Analysis ( IPA ) of the KRAS + EGFR induced gene set was performed and the top ten significantly regulated canonical pathways in which these genes are involved are displayed ( see ‘Materials and methods’ ) . P38 MAPK signaling was identified as the most significant upregulated pathway from this analysis; ERK/MAPK signaling was the second . ( D ) A highly simplified diagram of the EGFR/RAS signaling pathway is illustrated; the components assessed by western blot highlighted in blue . ( E ) Increased MAPK signaling in cells co-expressing mutant oncogenes . The indicated cells were cultured for 3 days with or without Dox; lysates were assayed by western blotting for the indicated proteins and phospho-proteins . Where relevant , the phosphorylated Tyrosine ( Y ) or Threonine ( T ) residue being measured is shown . Data are representative of three independent experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 06907 . 008 To determine the signaling pathways likely to be activated based on the differential RNA profiles , we performed Gene Set Enrichment Analysis ( GSEA ) ( Subramanian et al . , 2005 ) and observed that the most significant enrichment occurred for RNAs encoding downstream targets of oncogenic KRAS ( Figure 6B , Supplementary file 4 ) . Furthermore , Ingenuity Pathway Analysis ( IPA ) revealed p38 MAPK and ERK/MAPK signaling as the most significantly up-regulated canonical pathways ( Figure 6C ) . As MAPK signaling is immediately downstream of both EGFR and KRAS ( Figure 6D ) , we asked whether further activation of this pathway might occur in response to expression of both mutant oncogenes . To gain a better understanding of EGFR-RAS-MAPK signaling in the two cell lines , we assayed molecules in this pathway for phosphorylation status as a surrogate for signaling activity ( Figure 6D ) . 3 days after the addition of Dox , EGFR was constitutively phosphorylated in PC9 cells , as expected , whereas EGFR was not appreciably phosphorylated in H358 cells , except when mutant EGFR was induced ( Figure 6E ) . Phosphorylation of AKT at T308 was strongly increased in PC9 cells expressing mutant KRASG12V but less so in H358 cells expressing mutant EGFR . We observed a similar pattern for phospho-ERK ( p44/42 ) and phospho-p38 , the latter not being associated with normal EGFR signaling ( Figure 6E ) . The most consistent change in both cell lines was the phosphorylation of JNK , a stress-activated kinase . The activation of JNK can occur through the Rho family of GTPases , and the Rho protein family has also been implicated in RAS-dependent increases in macropinocytosis ( Bhanot et al . , 2010 ) . The kinase that phosphorylates both JNK and p38 , MKK4 , has been shown to be activated by the small molecule vacquinol-1 , an inducer of macropinocytosis ( Kitambi et al . , 2014 ) ; this is discussed further below . Taken together , our data suggest a general mechanism that is similar to the collective observations of other groups: increased activity of p38 and JNK MAPK pathways , increased macropinocytosis , and vacuolization lead to cell death . We show here that this cellular state can be created by the co-incidental activity of two mutant oncogenes like KRAS and EGFR . Thus , our data could provide an explanation for the observed mutual exclusivity . Our findings argue strongly that the well-known mutual exclusivity of KRAS and EGFR mutations in LUAD is due to synthetic lethality of the two mutant oncogenes , rather than to functional redundancy . Because multiple rounds of mutation are followed by clonal selection during tumor evolution , selection against cells with lethal combinations of mutations will produce mutual exclusivity . This general concept could have important implications for the design of therapeutic strategies , as well as tumor evolution . Recent efforts to design new treatments have focused on mutant genes that , when inhibited , decrease the viability of cancer cells ( Sawyers , 2009 ) . In contrast , combinations of mutations that are not found in tumor types may suggest synthetically lethal effects that could be exploited therapeutically—for instance , by using signaling agonists to provoke events that are lethal only in the presence of another mutant oncogene . In this study , we focused on one synthetically lethal relationship—oncogenic mutations of KRAS and EGFR in LUAD . We have used three approaches to ascertain the relationship . i . First , by analyzing available DNA sequences from large numbers of LUAD , we confirmed that activating mutations in these two genes are mutually exclusive , regardless of the smoking history of patients ( Figure 1A ) . Furthermore , we found that this kind of relationship is rare . No other mutated pairs of genes demonstrated a similarly negative association in lung adenocarcinoma , further arguing that the mutual exclusivity of KRAS and EGFR mutations is due to negative selection . By applying the same ‘exclusivity analysis’ to mutation data from other tumor types , however , we identified or confirmed other pairs of mutations—mutant KRAS and BRAF in COAD , mutant EGFR and NF1 in GBM , and mutant BRAF and NRAS in SKCM ( Petti et al . , 2006; Sensi et al . , 2006 ) —that may be synthetically lethal in those tumors . Additionally , this analysis confirms previous co-operative mutations ( e . g . , KRAS and LKB1 ) and reveals potentially new , testable gene pairs that are selected for during tumor formation ( Figure 1 ) . However , we note that this analysis was confined to alleles with altered sequences in coding domains and did not include gene rearrangements and amplifications . Thus we cannot draw conclusions about whether certain kinds of combinations , such as amplification of one proto-oncogene and sequence mutation of another , create synthetic lethalities . Moreover , the data sets we examined were derived by profiling cancers from patients who had not received medical treatment; hence the relationship of drugs or drug resistance to synthetic lethality has not been examined , and we would not have detected the recently reported amplification of wild type KRAS in a COAD with a mutant BRAF gene ( Ahronian et al . , 2015 ) . ii . Second , we obtained strong support for the concept of synthetic lethality of mutant KRAS and EGFF proteins in LUAD by attempting to co-express doxycycline-inducible transgenes encoding the two proteins in mouse models . Although each oncogenic transgene could initiate lung tumorigenesis on its own ( Fisher et al . , 2001; Politi et al . , 2006 ) , if both were present in the same mouse , the resulting tumor nodules expressed only one of the two transgenes ( Figure 2 ) ; this implies selection against cells that maintain expression of both . iii . Finally , to observe the proposed synthetic lethality more directly , we engineered an inducible second oncogene into human lung adenocarcinoma cell lines carrying a mutant EGFR or KRAS gene ( Figures 3–6 ) . We found that expression of the introduced second oncogene was detrimental in cells expressing both oncogenes . By taking advantage of cell lines that can be induced to express the second oncogene in a lethal combination , we have been able to study the manner and mechanisms of cell death in this context . We were motivated to pursue these issues to seek insight into the factors governing the synthetic lethality of mutant EGFR and KRAS and to identify opportunities to exploit this knowledge for therapeutic benefit . We found that the expression of both oncogenes caused the death of lung cancer cells within a week , mostly in a non-canonical manner . While we observed some markers of classical apoptosis and autophagy and small changes in cell cycle distribution in one line , the dominant features were cell enlargement and vacuolization , activation of stress-associated protein kinases ( as judged by phosphorylation of JNK and p38 MAPK ) , and increased macropinocytosis , with a subsequent loss of cell viability . Macropinocytosis and vacuolization have been observed previously in cells with augmented RAS signaling , nutrient acquisition , and cell death ( Chi et al . , 1999; Overmeyer et al . , 2008; Commisso et al . , 2013 ) , although not as an explanation of synthetic lethality caused by multiple oncogenes . Expression of H-RasG12V in glioblastoma and gastric cancer cell lines generates vacuoles associated with increased cell death ( Chi et al . , 1999 ) , a phenomenon dubbed ‘methuosis’ ( Maltese and Overmeyer , 2014 ) . These findings were subsequently linked to induction of uncontrolled macropinocytosis though activation of the Rac1 GTP-binding protein ( Bhanot et al . , 2010 ) . More recently , a small molecule screen for inhibitors of glioblastoma cell lines identified a quinine derivative ( vacquinol-1 ) that induced cell vacuoles , membrane ruffling , and increased macropinocytosis , similar to the effects of RAS activation alone ( Kitambi et al . , 2014 ) . This process was found to be dependent on MKK4 , part of the MAP2K family , that can phosphorylate both p38 and JNK ( Kitambi et al . , 2014 ) . As Rac1 is known to activate both p38 and JNK kinase pathways ( Coso et al . , 1995; Minden et al . , 1995; Olson et al . , 1995 ) these findings imply the existence of a common mechanism whereby RAS-mediated activation of Rac1 stimulates p38 and JNK , prompting uncontrolled macropinocytosis , vacuolization , and subsequent cell death . Despite these supportive studies , however , the findings that implicate JNK and p38 pathways in the mechanism of synthetic lethality induced by mutant KRAS and EGFR co-expression are only correlative at this point . Additional studies will be needed to determine whether activation of one or both of these pathways is required for cell death and whether it causes or results from macropinocytosis and vacuolization . Parts of this scenario , however , may not always be detrimental . Commisso et al . have shown that KRASG12V induces macropinocytosis unaccompanied by vacuolization and cell death ( Commisso et al . , 2013 ) , suggesting that macropinocytosis may augment amino acid uptake , as required by cells growing rapidly under the influence of RAS signaling . In the context of active EGFR and KRAS , it is possible that both mutant alleles together increase RAS-MAPK-mediated signaling beyond a tolerable threshold , resulting in increased activation of Rac1 and subsequent stimulation of p38 and JNK , kinases that are not typically activated to high levels by mutant KRAS or EGFR alone . Indeed , our data demonstrate increased activation of KRAS signaling and activation of stress kinases p38 and JNK upon co-induction of mutant KRAS and EGFR . Furthermore , these signaling events may be accompanied by , uncontrolled macropinocytosis , resulting in the accumulation of macropinosomes in the cytosol that overwhelm the capacity of the cells to adapt to the presence of large vacuoles . Thus , unlike a situation in which macropinocytosis favors cell survival , this would result in cell death . Future work will be required to determine the biochemical details of this process , such as the role of Rac1 , and to test whether the observed mutual exclusivity of other combinations of mutant genes in other tumor types are attributable to such mechanisms . Our findings also have potential implications for understanding mechanisms of resistance to EGFR-targeted therapies in lung adenocarcinoma . While inhibitors of EGFR have produced dramatic responses when used to treat EGFR-mutant tumors , acquired resistance to these compounds inevitably emerges ( reviewed in [Chong and Janne , 2013] ) . Tumor cells acquire resistance to these compounds through multiple mechanisms , including a second-site mutation ( T790M ) in EGFR , amplification of MET or ERBB2 , and conversion to a small cell lung cancer phenotype ( Yu et al . , 2013 ) . Our findings and those from other groups ( Sharifnia et al . , 2014 ) demonstrate that mutant KRAS can also confer resistance to TKIs in EGFR mutant cell lines . Yet mutations in KRAS have not been found to explain acquired resistance to EGFR TKIs in patients ( Ohashi et al . , 2012; Yu et al . , 2013 ) . Synthetic lethality could explain this observation . Acquired resistance is generally caused by the selection of a minor clone initially present in a heterogenous tumor , in which the resistance mutation exists in a subset of cells prior to TKI treatment . However , if expression of both mutant EGFR and mutant KRAS is detrimental , co-mutated cells are unlikely to survive during tumor evolution; therefore , they are unlikely to explain TKI resistance in an EGFR-mutant tumor . Taken together , our findings offer convincing evidence that the mutual exclusivity of mutant KRAS and EGFR in lung adenocarcinoma is dictated by synthetic lethality . These findings may have major implications for the future development of new agonistic treatment strategies for a substantial fraction of lung cancers driven by mutant KRAS or EGFR . To investigate the association between KRAS and EGFR mutations in human LUAD , sequence data was obtained from four different sources ( Ding et al . , 2008; Imielinski et al . , 2012; Seo et al . , 2012; Cancer Genome Atlas Research Network , 2014 ) . All studies assessed the entire coding region of these genes in tumors and in matched normal specimens , and they provide information regarding the somatic mutation status ( single nucleotide changes , small insertions and deletions ) for 662 LUAD in total . Data for three of these studies ( Ding et al . , 2008; Imielinski et al . , 2012; Cancer Genome Atlas Research Network , 2014 ) were downloaded from the cBioPortal for Cancer Genomics under three headings: ‘Lung Adenocarcinoma ( Broad , Cell 2012 ) ’ , ‘Lung Adenocarcinoma ( TSP , Nature 2008 ) ’ and ‘Lung Adenocarcinoma ( TCGA , Provisional ) ’ ( Cerami et al . , 2012; Gao et al . , 2013 ) . Data from Seo et al . was obtained from Supplementary file 3 ( Seo et al . , 2012 ) . Somatic mutations were investigated for their potential impact on protein function and status in the Catalogue of Somatic Mutations in Cancer ( COSMIC ) using MutationAssessor as previously described ( Cancer Genome Atlas Research Network , 2011 ) . Mutations in EGFR and KRAS were classified as oncogenic if they were predicted to have a substantial impact on protein function ( insertion/deletion mutations or a High/Medium MutationAssessor Predicted Functional Impact Score for non-synonymous mutations ) or were previously reported in additional tumors in COSMIC . Smoking status was obtained for each dataset and samples classified as ‘Smokers’ or ‘Never Smokers’ , based on the categorization used in each individual study . To assess the relationship between mutations in two genes , 2 × 2 Fisher's Exact Tests were computed using an online software tool ( http://www . langsrud . com/fisher . htm ) . To predict a negative association between mutant genes , a left-tailed p-value ( which assesses the negative association between variables ) was computed; a value ≤0 . 01 was considered significant . For the exome-wide analyses , variant level data was downloaded from TCGA Data Portal ( Version 0 . 4 . 0 , 2013-07-16 , https://tcga-data . nci . nih . gov/tcga/ ) . Redundant samples were removed to yield data for 520 unique tumor samples . All variants not mapping to coding regions were removed , and the resulting data files were parsed to stratify samples according to non-silent mutation ( insertions/deletions , non-synonymous nucleotide variants ) status of a given proto-oncogene . The silent ( synonymous nucleotide variant ) and non-silent mutation status for all other genes was then determined separately on a sample by sample basis and compared to those samples with non-silent mutations to determine the number of times the gene is mutated in a tumor with or without the mutated proto-oncogene . The resulting values for each gene ( the number of samples with mutations in a gene and the proto-oncogene , number of samples with mutation in the gene and not the proto-oncogene , number of samples without mutations in the gene and with mutation in proto-oncogene , and number of samples without mutation in either ) were then compared using the Fisher's Exact Test function in R ( http://www . r-project . org/ ) . Both ‘greater’ and ‘lesser’ p-values were calculated in order to determine positive and negative associations with a p-value of ≤0 . 01 considered significant . However , it should be noted that these analyses only take into account gene mutations ( nucleotide level variants and small insertions/deletions ) and not copy number or methylation changes which potentially affect gene function . All plots were created using Prism software ( GraphPad , La Jolla , CA ) . The same analyses were repeated for COAD , GBM and SKCM using the processed MAF files described in Kandoth et al . ( 2013 ) . All animals were kept in specific pathogen-free housing with abundant food and water under guidelines approved by the NHGRI Institutional Animal Care and Use Committee . The mouse models used in this study have been previously described ( Fisher et al . , 2001; Politi et al . , 2006 ) . All mice have been re-derived and backcrossed onto a FVB/N background ( Taketo et al . , 1991 ) which are more prone to develop spontaneous lung tumors later in life ( >14 months ) than the previous mixed background for these models ( Mahler et al . , 1996 ) . Tail DNA was isolated using the DNeasy Blood & Tissue Kit ( Qiagen , Venlo , Netherlands ) according to the manufacturer's protocol . Detection of the rtTA activator , KrasG12D , and EGFRDEL transgenes was performed as described previously ( Fisher et al . , 2001; Politi et al . , 2006 ) . Bi-transgenic CCSP-rtTA;TetO-KrasG12D mice were bred with bi-transgenic CCSP-rtTA;TetO-EGFRDEL mice to generate experimental animals . Doxycycline was administered by feeding mice with doxycycline-impregnated food pellets ( Harlan-Teklad 625 ppm , Indianapolis , IN ) . Mice were sacrificed when showing obvious signs of distress and lung tissues were processed and analyzed for the presences of lung tumors as described below . All survival curves were generated using Prism software ( GraphPad ) , with only mice that presumably died from lung cancer indicated in the percent survival . Tissue samples were homogenized and RNA was extracted using the RNeasy Mini kit ( Qiagen ) according to the manufacturer's instructions . DNase I ( Qiagen ) treatment was performed to eliminate any contaminating transgene DNA . RT–PCR reactions were carried out using the High-Capacity cDNA Reverse Transcription Kit ( Life Technologies , Carlsbad , CA ) . Custom TaqMan Gene Expression Assays ( Life Technologies ) were designed based on the sequences amplified by the genotyping primer sets ( Fisher et al . , 2001; Politi et al . , 2006 ) and quantitative RT–PCR reactions were performed using standard TaqMan reagents and protocols on a 7900 HT Fast Real-Time PCR system ( Life Technologies ) . The ΔΔCt method was used for relative expression quantification using the average cycle threshold for B-actin RNA ( Life Technologies Mm01324804_m1 ) to normalize gene expression levels between samples . Resulting fold change for each sample was compared against the average expression of bi-transgenic control mice and the values plotted using GENE-E software ( http://www . broadinstitute . org/cancer/software/GENE-E/index . html ) . Animals were sacrificed with a lethal dose of CO2 per institutional guidelines . The whole lung or lung nodules were excised and either flash-frozen or directly immersed in 4% paraformaldehyde in PBS ( whole lungs ) . All tissues were fixed in 4% paraformaldehyde overnight at room temperature , placed in 70% ethanol , and sent for paraffin embedding and sectioning ( Histoserv , Germantown , MD ) . Hematoxylin and eosin stain ( H&E ) were performed to assess histology and confirm the presence of tumors . The antibodies used for IHC were anti-human-EGFR K1492 ( Dako , Carpinteria , CA ) , anti-prosurfactant protein C AB3428 ( EMD Millipore , Billerica , MA ) , and anti-phospho- MAPK Thr202/Tyr204 4376 ( Cell Signaling Technology , Danvers , MA ) as previously described ( Fisher et al . , 2001; Politi et al . , 2006 ) . PC9 ( PC-9 ) and H358 ( NCI-H358 ) cells were obtained from Dr Romel Somwar at Memorial Sloan-Kettering Cancer Center . Cells were maintained in RPMI-1640 medium ( ATCC , Manassas , VA ) supplemented with 10% Tetracycline-free FBS ( Clontech , Mountain View , CA ) and 1% penicillin-streptomycin solution ( Lonza , Basel , Switzerland ) . Cells were cultured at 37°; air; 95%; CO2 , 5% . Cell lines were authenticated by multiplex PCR ( Genewiz , South Plainfield , NJ ) . Where indicated , doxycycline hyclate ( Sigma-Aldrich , St . Louis , MO ) was added at the time of cell seeding at 100 ng/ml . Erlotinib ( Cell Signaling ) was added at the time of cell seeding at 1 μM . Human mutant EGFR ( Addgene plasmid #11012 , [Greulich et al . , 2005] ) , human mutant KRAS ( Addgene plasmid #12544 , [Khosravi-Far et al . , 1996] ) were subcloned into pENTR/D-TOPO ( Life Technologies ) and transferred by Gateway LR Clonase II enzyme mix ( Life Technologies ) to pInducer20 ( gift from S Elledge , Harvard ) . Plasmids were sequence-verified . Lentivirus was generated using 293T cells ( ATCC ) , psPAX2 #12260 ( Addgene , Cambridge , MA ) and pMD2 . G ( Addgene plasmid #12259 ) . Both polyclonal cell lines and single cell-derived clonal cell lines were used . Cells were lysed in buffer ( 9803 , Cell Signaling ) containing complete protease/phosphatase inhibitor cocktail ( 78410 , Life Technologies ) . Lysates were sonicated , cleared by centrifugation , and protein concentration determined by BCA protein assay kit ( Life Technologies ) . Samples were denatured by boiling in loading buffer ( 7722 , Cell Signaling ) . 30 μg of lysates were loaded on 4–20% Novex tris-glycine gels ( Life Technologies ) , transferred to Immobilon ( PVDF ) membranes ( EMD Millipore ) , blocked in TBST ( 0 . 1% Tween-20 ) and 5% milk . Primary incubation with antibodies was overnight at 4° , followed by appropriate HRP-conjugated secondary ( Santa Cruz Biotechnology , Dallas , TX ) and detected using ECL plus ( 32132 , Life Technologies ) or Femto chemiluminescent substrate ( 34096 , Life Technologies ) . Antibodies used were obtained from Cell Signaling and raised against the following proteins: phospho p-38 ( 4511 ) , p38 ( 8690 ) , p-p44/p42 ( 4370 ) , p44/p42 MAPK ( 4695 ) , p-SAPK/JNK ( 4668 ) , SAPK/JNK ( 9258 ) , p-EGFR ( 2234 ) , EGFR ( 2232 ) , LC3A ( 4599 ) , p-Akt ( 13038 ) , Akt ( 9272 ) , PARP ( 9542 ) , GAPDH ( 2118 ) . Additionally , we used antibodies against KRAS ( F234 , Santa Cruz Biotechnology ) and GFP ( A-21311 , Life Technologies ) . To determine the number of viable cells over a 7-day time course , cells were seeded in triplicate in 6-well plates at 20 , 000 cells/well ( PC9 derivatives ) or 40 , 000 cells/well ( H358 derivatives ) . Cells were seeded into doxycycline ( 100 ng/ml ) and/or erlotinib ( 1 μM ) . Media ( with or without doxycycline or erlotinib ) were not replenished during the 7 days . At indicated time points , media was aspirated and replaced with media containing Alamar Blue ( Life Technologies ) . Fluorescence intensities from each well were read in quadruplicate on a FluoStar Optima instrument ( BMG Labtech , Cary , NC ) and data plotted in Prism ( GraphPad ) . Cells were counted using Trypan Blue ( Lonza ) and automated cell counter Countess ( Life Technologies ) . Cells were seeded in the same format as for the 7-day time course . On day 5 , media was collected and adherent cells removed by Accutase ( eBioscience , San Diego , CA ) and cells stained with PE-Annexin V and 7-AAD ( BioLegend , San Diego , CA ) . Cells were analyzed on a BD FACS Calibur and data processed using FlowJo software ( FlowJo , Ashland , OR ) . Cells were seeded in the same format as for cell viability studies . Over a 7-day time course , cell culture media was aspirated and adherent cells removed by Accutase ( BioLegend ) . Cells were washed in PBS and fixed in 66% ethanol on ice , incubated overnight at 4° , pelleted , washed in PBS , and stained in propidium iodide/RNase staining solution ( 4087 , Cell Signaling ) . Cells were analyzed on a BD FACScan BD Biosciences , San Jose , CA ) with a Cytek dxp8 ( Cytek , Fremont , CA ) upgrade and data processed using FlowJo software . Phase contrast images were taken on a AxioObserver . A1 ( Zeiss , Thornwood , NY ) . For EM , cells were processed and embedded as described ( Gonda et al . , 1976 ) . Cells cultured in a 6-well plate were fixed in 0 . 1 M cacodylate buffer containing glutaraldehyde ( 2% vol/vol ) for 1 hr at room temperature , washed three times in cacodylate buffer ( 0 . 1 M , pH 7 . 4 ) , and post fixed in osmium tetroxide ( 1% vol/vol ) for 1 hr at room temperature . The cells were stained in 0 . 5% wt/vol uranyl acetate ( 0 . 5% vol/vol ) in acetate buffer ( 0 . 1 M , pH 4 . 5 ) for 1 hr at room temperature . Cells were gradually dehydrated in a series of ethanol solutions ( 35% , 50% , 75% , 95% , and 100% ) . Cells were washed in a pure epoxy resin three times after 100% ethanol , and then embedded in the resin . The resin was cured in an oven ( 55°C ) for 48 hr . The cured resin blocks were separated from the plate by submerging in liquid nitrogen . The cells were examined under an inverted microscope and areas selected for preparing 80 to 90 nm sections , which were then mounted on 200 copper mesh grids and counter-stained in uranyl acetate and lead citrate . The grids were carbon coated in a vacuum evaporator . The grids were examined and imaged in the electron microscope operated at 80 kv and digital images were captured by a CCD camera . Cells were seeded in 12-well plate in phenol red-free RPMI ( Life Technologies ) and 10% Tetracycline-free FBS ( Clontech ) and doxycycline ( 100 ng/ml ) , as indicated . Cells were stimulated with dextran , Oregon Green 488; 70 kDa ( D7173 , Life Technologies ) at 1 mg/ml for 30 min , washed in cold PBS and cells removed by Accutase ( eBioscience ) . Cells were analyzed on a BD FACScan with a Cytek dxp8 upgrade and data were processed using FlowJo software . H358 and PC9 cells and their derivatives were grown in 6-well plates , treated with or without 100 ng/ml dox in duplicate for 24 hr , and total RNA was extracted using the RNeasy Mini kit ( Qiagen ) as described above . Sample quality was assessed using an Agilent Bioanalyzer ( Agilent , Santa Clara , CA ) and subsequent sample preparation , array hybridization , and data acquisition was performed by the National Human Genome Research Institute's Microarray Core facility . GeneChip Human Gene 2 . 0 ST microarrays ( Affymetrix , Santa Clara , CA ) were used according to the manufacture's protocols . Raw data ( Affymetrix CEL files ) were normalized by robust multiarray analysis ( Irizarry et al . , 2003 ) and subsequently analyzed to detect genes differentially expressed between dox-treated and untreated cultures , using the ANOVA function in Partek Genomics Suite software ( Partek , St . Louis , MO ) ; a Benjamini–Hochberg corrected p value of <0 . 01 was considered significant . Overlap between differentially expressed genes among the cell lines was determined using VENNY software ( Oliveros , 2007 ) and those significantly deregulated upon the addition of dox in both H358-TetO-EGFRL858R and PC9-TetO-KRASG12V cells and not their respective TetO-GFP controls were selected for further analyses . DNA probes that detected significant differences were mapped to genes using Partek Genomics Suite; those not mapping to annotated genes were removed . An average fold-change for dox treated vs untreated H358-TetO-EGFRL858R and PC9-TetO-KRASG12V cells was calculated and these values , along with the corresponding probe IDs , were uploaded to IPA software ( Ingenuity Systems , Qiagen ) and analyzed using the Canonical Pathways function . Canonical pathways with a z-score ≥0 . 25 and a p value ≤0 . 05 were considered activated , whereas those with a z-score ≤ −0 . 25 and a p value ≤0 . 05 were considered inactivated . GSEA ( Subramanian et al . , 2005 ) was also performed on the same gene list using the pre-ranked GSEA function and the transcription factor targets gene sets within the C6: Oncogenic Signatures Collection from the Molecular Signature Database ( Subramanian et al . , 2005 ) . Genes were ranked according to their average fold change for the dox treated TetO-Oncogene cells ( described above ) and GSEA was run with default settings except ‘Min size: exclude smaller set = 10’ .	A person develops cancer when changes in a cell's DNA ( called mutations ) allow the cell to grow rapidly and spread around the body . The mutated genes are often involved in controlling the growth of cells , such as two genes called EGFR and KRAS , which are associated with forms of lung cancer . In a type of lung cancer called adenocarcinoma , the KRAS gene is mutated in about one-third of tumors and the EGFR gene is mutated in about 15% . However , the two mutations rarely or never occur in the same tumor . This could be because the effects of the mutations overlap , so that cells with both mutations have no advantages over cells with just one . Alternatively , it is possible that having both mutations may be harmful to tumor cells . Here , Unni , Lockwood et al . analyzed genetic data from over 600 lung tumors and confirmed that none of them have cancer-causing mutations in both KRAS and EGFR . Then , Unni , Lockwood et al . carried out experiments using genetically engineered mice with mutated forms of both KRAS and EGFR that are activated by a drug called doxycycline . As expected , the mice developed lung tumors when exposed to the drug , but these tumors didn't grow any faster than mouse tumors that had mutations in only one of the genes . In the mice with both mutant genes , only one of the two genes was actually active in most of the tumor cells . Unni , Lockwood et al . manipulated human lung tumor cells in the laboratory so that the cells had mutated versions of both genes . These cells developed serious abnormalities and died , which may be due to the over-activation of a communication pathway within the cells called MAPK signaling . The next challenges are to understand why the combination of these two mutant genes kills these cancer cells and to look for other combinations of mutations that can be toxic to cancer cells . In the future , it might be possible to develop drugs that can mimic the effects of these gene mutations to treat cancers .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "cancer", "biology" ]	2015	Evidence that synthetic lethality underlies the mutual exclusivity of oncogenic KRAS and EGFR mutations in lung adenocarcinoma
The bar-headed goose is famed for migratory flight at extreme altitude . To better understand the physiology underlying this remarkable behavior , we imprinted and trained geese , collecting the first cardiorespiratory measurements of bar-headed geese flying at simulated altitude in a wind tunnel . Metabolic rate during flight increased 16-fold from rest , supported by an increase in the estimated amount of O2 transported per heartbeat and a modest increase in heart rate . The geese appear to have ample cardiac reserves , as heart rate during hypoxic flights was not higher than in normoxic flights . We conclude that flight in hypoxia is largely achieved via the reduction in metabolic rate compared to normoxia . Arterial Po2 was maintained throughout flights . Mixed venous PO2 decreased during the initial portion of flights in hypoxia , indicative of increased tissue O2 extraction . We also discovered that mixed venous temperature decreased during flight , which may significantly increase oxygen loading to hemoglobin . Flapping flight in birds is the most metabolically costly form of locomotion in vertebrates ( Butler and Bishop , 2000 ) . These costs are exacerbated with increasing elevation as the air becomes less dense , reducing oxygen available to support metabolism and requiring changes to the wing kinematics of forward flying birds ( Feinsinger et al . , 1979; Dudley and Chai , 1996; Ellington , 1984; Pennycuick , 2008 ) . Our understanding of the adaptations that support high-altitude flight in birds is growing , particularly in the bar-headed goose , Anser indicus ( Storz et al . , 2010; Scott et al . , 2015 ) . This species migrates biannually across the Himalayan Mountains and Tibetan Plateau , wintering in India and breeding in China and Mongolia , typically flying through passes 5 , 000 to 6 , 000 m above sea level , where partial pressures of oxygen are only half of those at sea level . They have been documented flying at altitudes as high as 7 , 290 m ( Bishop et al . , 2015; Hawkes et al . , 2013 ) . The physiological adaptations to hypoxia that have been previously described in bar-headed geese have been examined in birds at rest or running ( Fedde et al . , 1989; Scott et al . , 2015; Hawkes et al . , 2014; Scott and Milsom , 2007 ) . Such adaptations are distributed throughout the oxygen transport cascade , the steps involved in oxygen transfer from atmosphere to mitochondria ( ventilation , lung oxygen diffusion , circulation and tissue oxygen extraction ) ( Scott et al . , 2015 ) . Direct and integrated physiological measures of oxygen transport during flight , on the other hand , are extremely limited ( Butler and Bishop , 2000; Ward et al . , 2002 ) , and none have been made under hypoxic conditions . Ward et al . ( 2002 ) documented that the metabolic cost of flight in bar-headed geese in normoxia at sea-level in a wind tunnel was roughly 12 times resting metabolic rate . This was associated with an approximately two-fold increase in heart rate . Based on extrapolation from wind tunnel heart rate data , flight metabolic rate for birds migrating at an altitude around 6 , 000 m in the wild was calculated to be approximately 15 times resting metabolic rate ( Bishop et al . , 2015 ) . This further increase in metabolic cost is concordant with the increased biomechanical costs of flying in the thinner air at high altitude ( requiring increased flight speeds to offset reductions in lift; Pennycuick , 2008 ) but may also arise in part from increased metabolic demands on the cardiorespiratory system associated with flight in hypoxia . Bar-headed geese trained to run on a treadmill did not show a significant change in metabolic rate between normoxia and severe hypoxia , however the increase in metabolic rate from rest to running was only ~2 . 5 fold ( Hawkes et al . , 2014 ) . Whether or not hypoxia increases the metabolic cost of flight remains to be determined . Based on these observations , we aimed to determine ( 1 ) how the metabolic challenge of flight differs between normoxia and normobaric hypoxia , and ( 2 ) whether bar-headed geese are capable of wind tunnel flight in severe normobaric hypoxia equivalent to altitudes of roughly 9 , 000 m ( 0 . 07 FiO2 ) , the maximum altitude at which they have been anecdotally reported to fly ( Swan , 1961 ) . The respiratory exchange ratio RER ( V˙CO2/ V˙O2 = RER ) could only be measured in normoxia due to unreliable V˙O2 values in hypoxia . There was a significant effect of activity on RER ( F2 , 301 . 95=54 . 37 , p<0 . 0001 , ICC=0 . 254 ) . RER in flight ( EMM of 1 . 00 ± 0 . 034 ) was significantly higher than pre-flight ( EMM of 0 . 87 ± 0 . 035; t=7 . 026 , p<0 . 0001 ) and rest ( EMM of 0 . 80 ± 0 . 035; t=10 . 073 , p<0 . 0001 ) . RER in pre-flight was also significantly higher than at rest ( t=3 . 453 , p=0 . 0019 ) . V˙CO2 differed significantly based on oxygen level in flight ( F2 , 549 . 54=74 . 155 , p<0 . 0001 , ICC=0 . 145 ) , but not at rest or during pre-flight ( p>0 . 468 ) . Within flight data , V˙CO2 in normoxia ( EMM=223 . 8 ± 4 . 8 mL CO2 min−1 kg−1 ) was significantly higher ( t=−8 . 047 , p<0 . 0001 ) than V˙CO2 in moderate hypoxia ( EMM=193 . 0 ± 5 . 1 mL CO2 min−1 kg−1 ) . V˙CO2 dropped significantly in severe hypoxia ( EMM=142 . 5 ± 8 . 3 mL CO2 min−1 kg−1 ) compared to moderate hypoxia ( t=−6 . 562 , p<0 . 0001 ) . Individual minimum metabolic rate ( the lowest steady state V˙CO2 of all flights for each bird ) was not different in normoxia and moderate hypoxia ( paired t-test; t=0 . 157; p=0 . 883 ) . Heart rate did not differ significantly between O2 levels in flight ( F2 , 441 . 13=1 . 237 , p=0 . 2914 , ICC = 0 . 166 ) , but O2 level had a marginally significant effect on heart rate pre-flight ( F2 , 441 . 53=3 . 077 , p=0 . 0471 ) . However , during post-hoc testing , no comparisons were significant within pre-flight ( p>0 . 12 ) . There was a significant effect of O2 level on heart rate at rest ( F2 , 439 . 20=7 . 688 , p=0 . 0005 ) , because the resting heart rate in severe hypoxia ( EMM = 149 . 7 ± 11 . 9 beats min−1 ) was significantly higher ( t = −2 . 569 , p=0 . 0316 ) than in normoxia ( EMM = 128 . 3 ± 9 . 2 beats min−1 ) and moderate hypoxia ( t = 3 . 817 , p=0 . 0005 ) . Resting heart rate in moderate hypoxia ( EMM = 107 . 3 ± 10 . 1 ) did not differ significantly from normoxia ( t = 2 . 077 , p=0 . 1151 ) . There was a significant effect of oxygen level on CO2 pulse for flight data ( F2 , 450 . 00 = 31 . 845 , p<0 . 0001 , ICC = 0 . 162 ) but not on pre-flight or rest data ( p>0 . 61 ) . CO2 pulse in normoxic flight ( EMM = 0 . 722 ± 0 . 021 mL CO2 beat−1 kg−1 ) was significantly higher ( t = −5 . 818 , p<0 . 0001 ) than CO2 pulse in moderate hypoxic flight ( EMM = 0 . 627 ± 0 . 022 mL CO2 beat−1 kg−1 ) . CO2 pulse in moderate hypoxic flight was significantly higher ( t = −3 . 666 , p=0 . 0008 ) than in severe hypoxia ( EMM = 0 . 514 ± 0 . 034 mL CO2 beat−1 kg−1 ) . Due to the non-independence of our repeated measurements across individual birds , we cannot calculate correlation statistics such as r2 . In comparing the correlation of heart rate versus metabolic rate we generated a linear mixed model for the combined data and found heart rate was a significant predictor of metabolic rate ( df = 444 . 7 , t = 37 . 535 , p<0 . 0001 , ICC = 0 . 143 ) . However , when we added activity as a fixed effect , heart rate was no longer a significant predictor of metabolic rate in flight ( df = 442 . 9 , t = 0 . 244 , p=0 . 808 , ICC = 0 . 127 ) , only during pre-flight ( df = 446 . 2 , t = −5 . 113 , p<0 . 0001 , ICC = 0 . 106 ) and rest ( df = 444 . 9 , t = 18 . 652 , p<0 . 0001 , ICC = 0 . 184 ) . This indicates that , when pooled , the data are bimodal ( flight and preflight/rest ) , but within the flight data , there is a large variation in CO2 production at any level of heart rate and vice versa ( Figure 1 , and by individual in Figure 1—figure supplement 1 and Figure 1—figure supplement 2 ) . Two other studies ( Ward et al . , 2002; Hawkes et al . , 2014 ) have measured metabolic rates and heart rates in resting and exercising bar-headed geese . We pooled our complete data set with those values ( Figure 1B ) and also compared the distribution of our heart rate data to those measured in migrating wild geese ( Bishop et al . , 2015; Figure 2 ) . The comparison shows a remarkable agreement in the peaks of the heart rate measurement distribution of geese flying below 2 , 300 meters in the wild ( Bishop et al . , 2015 ) and in the present study ( but note possible survivor bias for severe hypoxia data ) . We measured mixed venous PO2 in normoxia , moderate hypoxia , and severe hypoxia and report values for several time points across the flight ( pre-flight , start , steady state , end , and in recovery ) . There was a significant effect of both oxygen level ( F2 , 79 . 197=22 . 3439 , p<0 . 0001 ) and timepoint ( F4 , 79 . 113=5 . 0645 , p=0 . 0011 ) on venous PO2 , but not the interaction O2 leveltimepoint ( F8 , 79 . 127=0 . 9865 , p=0 . 453 ) . Venous PO2 did not significantly differ between exposed oxygen levels during pre-flight ( preflight normoxia EMM = 47 . 68 ± 2 . 52 mmHg , preflight moderate hypoxia EMM = 44 . 47 ± 3 . 16 mmHG , preflight severe hypoxia EMM = 38 . 68 ± 4 . 21 mmHg ) , but then was maintained in normoxia ( start EMM = 50 . 00 ± 2 . 52 mmHg ) while dropping in both levels of hypoxia such that both moderate hypoxia ( start EMM = 34 . 71 ± 3 . 16 mmHg , t = −4 . 360 , p=0 . 0001 ) and severe hypoxia ( start EMM = 33 . 61 ± 4 . 21; t = −3 . 705 , p=0 . 0012 ) were significantly different from normoxia at the start of flight , but did not differ from each other ( t = −0 . 236 , p=1 . 0 ) . During the steady state portion of the flight , PO2 in normoxia ( steady state EMM = 42 . 30 ± 2 . 49 mmHg ) dropped slightly so moderate hypoxia ( steady state EMM = 33 . 59 ± 3 . 40 mmHg ) was marginally non-significant ( t = −2 . 373 , p=0 . 0600 ) while PO2 in severe hypoxia ( steady state EMM = 29 . 61 ± 4 . 21 ) remained significantly different from normoxia ( t = −2 . 881 , p=0 . 0152 ) . That pattern held through the end of the flight ( end normoxia EMM = 41 . 48 ± 2 . 40 mmHg , end moderate hypoxia EMM = 33 . 54 ± 3 . 40 , end severe hypoxia EMM = 29 . 25 ± 4 . 21 mmHg ) , but in recovery PO2 in normoxia ( recovery EMM = 50 . 60 ± 2 . 45 mmHg ) increased more than PO2 in hypoxia so that both moderate hypoxia ( recovery EMM = 38 . 13 ± 3 . 16 mmHg; t = −3 . 588 , p=0 . 0017 ) and severe hypoxia ( recovery EMM = 34 . 85 ± 4 . 21 mmHg; t = −3 . 581 , p=0 . 0017 ) were significantly lower ( Figure 3 , Supplementary file 2 ) . There was no significant effect of oxygen level on venous blood temperature ( F2 , 72 . 225=1 . 7253 , p=0 . 1854; Figure 3B ) nor the interaction of O2 leveltimepoint ( F8 , 71 . 045=0 . 3347 , p=0 . 9497 ) . There was a significant main effect of timepoint ( F4 , 71 . 036=11 . 4269 , p<0 . 0001 ) , which held within each oxygen level ( normoxia: F4 , 71 . 17=6 . 333 , p=0 . 0002; moderate hypoxia: F4 , 71 . 01=3 . 547 , p=0 . 0107; severe hypoxia: F4 , 71 . 01=3 . 497 , p=0 . 0115 ) . The effect of timepoint was due to a drop in venous temperature between preflight and the steady state portion of the flight: the minimum drop was 1 . 22 °C and the maximum drop was 2 . 72 °C . In normoxia , venous temperature was significantly higher in preflight ( preflight EMM = 41 . 37 ± 0 . 402°C ) compared to steady state ( steady state EMM = 39 . 99 ± 0 . 441°C; t = 4 . 342 , p=0 . 0005 ) and the end of the flight ( end EMM = 40 . 38 ± 0 . 421°C; t=−3 . 404 , p=0 . 0110 ) , as well as in steady state compared to recovery ( recovery EMM = 41 . 04 ± 0 . 403°C; t = 3 . 311 , p=0 . 0146 ) and the start of the flight ( start EMM = 41 . 10 ± 0 . 406°C; t=−3 . 443 , p=0 . 0097 ) . In moderate hypoxia , venous temperature in steady state ( steady state EMM = 39 . 998 ± 0 . 509°C ) was significantly lower than both preflight ( preflight EMM = 41 . 473 ± 0 . 509°C; t = 3 . 139 , p=0 . 0247 ) and the start of the flight ( start EMM = 41 . 373 ± 0 . 509°C; t=−2 . 926 , p=0 . 0460 ) . In severe hypoxia , however , no pairwise comparisons among timepoints were significant ( p>0 . 07 ) . We successfully measured arterial ( carotid ) and mixed venous PO2 at all three levels of oxygen for one bird ( bird 45 ) : normoxia ( venous: four flights , arterial: two flights ) , moderate hypoxia ( venous: four flights , arterial: three flights ) , and severe hypoxia ( venous: three flights , arterial: four flights , Supplementary file 1 ) . During preflight , there was a significant effect of oxygen exposure level on arterial PO2 ( H ( 2 ) =6 . 0 , p=0 . 014 ) from a mean of 72 . 1 ± 0 . 42 mmHg ( median 72 . 1 mmHg ) in normoxia , to a mean of 56 . 5 ± 5 . 4 mmHg ( median 56 . 5 mmHg ) in moderate hypoxia , and a mean of 36 . 7 ± 0 . 54 mmHg ( median 36 . 5 mmHg ) in severe hypoxia ( Supplementary file 3 ) . In post-hoc testing , no comparisons among preflight arterial PO2 were significant ( p>0 . 05 ) . In general , preflight arterial PO2 levels were maintained throughout flights . There was a significant effect of oxygen exposure level on arterial PO2 during steady state flight ( F2 , 6=23 . 1294 , p=0 . 002 ) , such that arterial PO2 in normoxia ( 85 . 8 ± 10 . 8 mmHg ) was significantly higher ( Q = 7 . 079 , p=0 . 006 ) than arterial PO2 in moderate hypoxia ( 47 . 0 ± 4 . 3 mmHg ) , and severe hypoxia ( 36 . 3 ± 2 . 7 mmHg , Q = 9 . 537 , p=0 . 001 ) . Example PO2 recordings during normoxic and hypoxic flights are shown in Figure 4 . Note the rapid rise in mixed venous temperature immediately after landing , followed by a rapid recooling , and slow rewarming phase . This was consistently observed during flight recovery . We measured blood gas variables in resting birds during periods separate from flight trials ( see rest data in Supplementary files 2 and 3 ) . There was a significant effect of oxygen level on venous Po2 at rest ( F2 , 17 . 33=27 . 775 , p<0 . 0001 ) . Both moderate hypoxia ( EMM=28 . 96 ± 3 . 18 mmHg; t=−5 . 579 , p=0 . 0001 ) and severe hypoxia ( EMM=23 . 36 ± 3 . 44 mmHg; t=7 . 001 , p<0 . 0001 ) were significantly lower than normoxia ( EMM=46 . 51 ± 3 . 18 mmHg ) , but not significantly different from each other ( t=-1 . 694 , p=0 . 3238 ) . Venous temperature did not differ significantly between O2 levels at rest ( F2 , 4=3 . 2428 , p=0 . 1455 ) . Flights in our study were relatively short compared to wild flights by conspecifics at high altitude ( mean duration for flight over the Himalayas is 8 hr; Hawkes et al . , 2013 ) but are comparable to those of a previous wind tunnel study . Ward et al . ( 2002 ) experienced the same difficulty ( Pers . Comm . ) , with only two of their five bar-headed geese achieving flight in the wind tunnel , and flights remaining relatively short . Given that birds underwent considerable training , including outdoor flights , and that wind tunnel flights were short even in normoxia , it would appear that the birds were reluctant to fly for long once instrumented in the conditions of the wind tunnel . Flow turbulence in the tunnel , the presence of the experimenters and the presence of the mask and tubing all will have increased flight costs and may have contributed to this ( Hedenström and Lindström , 2017 ) . Although wing-beat frequencies of our birds were higher than those of bar-headed geese in the wild ( Bishop et al . , 2015 ) , values were similar between normoxic vs . hypoxic and instrumented vs . uninstrumented flights ( Supplementary file 4; Whale , 2012 ) . Despite possible instrumentation effects or the short flight durations , flights were repeatable , of similar length under all conditions , and most importantly , produced stable levels of the measured variables , allowing us to make robust comparisons between flight in normoxia vs . hypoxia , thus examining the effects of hypoxia on flight physiology under similar conditions . Determining how these results relate to the multi-hour migratory flights of this species at high altitude will require further work measuring physiological variables in the wild , or during longer flights in both normobaric and hypobaric conditions . It was challenging to obtain reliable measures of V˙O2 during flight in hypoxia , likely because of small fluctuations in gas mixing , given the dynamics of flight in the wind tunnel while wearing the mask . Similar problems were encountered in a previous study ( Hawkes et al . , 2014 ) . While small minute-to-minute fluctuations in FiO2 will average out and not alter the hypoxic challenge to the bird , they do have a destabilizing effect on the calculation of V˙O2 . Because incurrent CO2 levels remained close to zero throughout , any increase in CO2 must come from the bird and therefore our V˙CO2 data were considered robust . Because the RER averaged 0 . 988 ± 0 . 01 during flight in normoxia and flight durations between normoxia and moderate hypoxia were not significantly different , we have made the assumption that V˙CO2 and V˙O2 can be used interchangeably under this condition . As opposed to indicating carbohydrate use during flight , an RER near 1 may reflect hyperventilatory CO2 loss . We would expect during longer flights that RER would fall close to 0 . 7 , assuming the birds are preferentially metabolizing lipids . This is supported by our data as RER falls to 0 . 921 ± 0 . 02 for flights longer than six minutes . Heart rate and metabolic rate of bar-headed geese at rest in normoxia in this study were remarkably similar to those obtained by Ward et al . ( 2002 ) , as was the mean respiratory exchange ratio ( RER ) . However , wingbeat frequencies measured during flight in normoxia in the previous study were lower , as was V˙CO2 ( by approximately 29% ) . Heart rates during flight , however , were lower in the current study suggesting that our birds were working harder but were employing larger increases in cardiac output and/or pulmonary exchange ( Figure 1B ) . The discrepancy in heart rates may also be due to methodological differences , as we found it necessary to visually verify each heart rate peak while Ward et al . relied on periodic averages . Birds in the present study flew at a slightly lower range of flight speeds compared to birds in the wild [12 . 5 to 15 . 0 m s−1 versus 17 . 1 m s−1 for wild bar-headed geese migrating at <1 , 000 meters altitude ( Hawkes et al . , 2013 ) . The 2 . 5-fold difference in heart rate measured between birds at rest and in-flight in the present study is also similar to that of wild birds flying at low altitudes ( Bishop et al . , 2015 ) ( Figure 2 ) . Bishop et al . documented an increase in heart rate with increasing altitude . As heart rate in the current study was unaffected by hypoxia ( Table 1 ) , this suggests that the increases in heart rate measured in wild bar headed geese migrating at altitudes above 2 , 300 meters may be a consequence of flight dynamics in hypobaria , rather than hypoxia . Finally , heart rate was highly variable at any level of CO2 production and vice versa . This was also the case for the relationship between heart rate and wing-beat frequency in wild birds , although mean values were well correlated ( Bishop et al . , 2015 ) . Ward et al . ( 2002 ) also concluded that their wind tunnel data could not be used directly to calculate the metabolic rate of wild migratory geese from measurements of heart rate alone . In hypoxia both at rest and preflight in the wind tunnel , V˙CO2 fell by 22 and 26% for FiO2=0 . 105 and by 10 and 29% for FiO2=0 . 07 . In the one bird for which we have data at all O2 levels , arterial PO2 fell to 56 . 5 ± 5 . 4 and 36 . 7 ± 0 . 54 mmHg preflight for FiO2=0 . 105 and FiO2=0 . 07 , respectively . Based on the data from Meir and Milsom ( 2013 ) and assuming a body temperature of 41°C and an arterial pH of 7 . 4 , this would lead to a fall in O2 saturation pre-flight from around 92% ( 0 . 21 FiO2 ) to 84% ( 0 . 105 FiO2 ) and 67% ( 0 . 07 FiO2 ) , roughly equivalent to the decrease in metabolic rate . For birds flying in FiO2 = 0 . 105 , V˙CO2 was 16% lower than in birds flying in normoxia . Heart rates in moderate hypoxia were not significantly different from those under any state ( rest , pre-flight , flight ) in normoxia , while the estimated O2 pulse decreased in proportion to the V˙CO2 . The greater than 8-fold increase in O2 pulse from rest to flight in normoxia was maintained in moderate hypoxia . Thus the moderately hypoxic birds appear to have met the hypoxic challenge by a combination of a reduced metabolism , maintaining heart rate , and maintaining the increase in CO2/estimated O2 pulse . For the one bird for which we have adequate data flying in FiO2 = 0 . 07 , V˙CO2 was 20% lower under this severe hypoxic condition than in normoxia . Again , heart rates were not significantly different when flying in hypoxia . This bird did have a higher heart rate ( 333 . 6 ± 11 beats min−1 ) despite the lower V˙CO2 ( 157 . 4 ± 8 . 4 ml CO2 kg−1 min−1 ) in normoxia than the other birds , likely also contributing to its exceptional performance . The reduction in metabolism in hypoxia observed in the current study could represent O2 limitation , selective suppression of metabolism to specific tissues or increased efficiency of flight pattern and thus O2 utilization . Alternatively , the reduction measured in metabolic rate could be concordant with the onset of anaerobic metabolism . Although we cannot reject this possibility as lactate was not measured in this study , we consider it unlikely as there was no sign of an oxygen limitation , because: 1 ) the birds could still increase V˙CO2 by 14 to 23-fold during flight , 2 ) reductions in metabolic rate also occurred under rest and preflight conditions , and 3 ) the birds sustained flights of similar durations at constant levels of arterial PO2 . It is quite possible that while flying under the more metabolically challenging conditions of hypoxia , the birds are minimizing energy supply to less essential processes ( e . g . digestion , birds are known to undergo atrophy of gut tissue prior to migration; McWilliams et al . , 2004; Piersma and Gill , 1998 ) , or that they may be altering their flight behavior and biomechanics to fly with maximal efficiency . This is supported by the metabolic data as the individual minimum metabolic rates ( the lowest steady state V˙CO2 of all flights for each bird ) were not different between normoxia and moderate hypoxia . Only the overall average metabolic rate differs , indicating that birds may employ more or less efficient flight strategies in normoxia , but shift towards using only the most efficient strategies when oxygen limited . Wing-beat frequencies of bar-headed geese in this study were similar in both normoxia and hypoxia . This is consistent with results from both ruby-throated hummingbirds ( Archilochus colubris ) and the South American hummingbird ( Colibri coruscans ) , a montane species capable of hovering at altitudes over 6000m ( Chai and Dudley , 1996; Berger , 1974 ) . Despite a constant wing beat frequency , flight biomechanics of the geese in our study were altered in response to hypoxia , with increased upstroke duration ( T ) and decreased upstroke wingtip speed ( Utip ) , upstroke plane amplitude ( FSP ) , and mid-upstroke angle of inclination ( a ) ( Supplementary file 4; Whale , 2012 ) As the downstroke produces the majority of lift and all forward thrust , by increasing the ratio of the duration of upstroke to downstroke , the duration of activation of the pectoralis major muscle group is decreased ( responsible for the majority of downstroke power ) . We therefore hypothesize that bar-headed geese reduce oxygen demand in hypoxic flight by limiting oxygen supply to less essential metabolic processes and/or maximizing the mechanical efficiency of flight . We obtained the first measurements of arterial and venous PO2 and temperature records in this species , and that of any equivalently sized bird , during flight . Both mixed venous and arterial PO2 values decreased progressively with decreasing levels of FiO2 , as expected . In general , levels of arterial Po2 were maintained throughout flights ( Figure 4 ) although there was some variability in individual flights . Mixed venous PO2 , on the other hand , tended to decrease during the initial portion ( first minute ) of flights in hypoxia ( Figure 3 and Figure 4 ) , indicative of increased tissue O2 extraction . The arterial PO2 of geese flying at 0 . 105 FiO2 was similar to that of geese running on a treadmill in a previous study at 0 . 07 FiO2 ( Figure 4; Hawkes et al . , 2014 ) . When directly comparing at the same level of hypoxia ( 0 . 07 FiO2 for both studies ) , arterial PO2 during flight was about 20% lower than while running Hawkes et al . ( 2014 ) . Arterial values in the range measured in 0 . 07 FiO2 are strikingly low ( Supplementary files 1 and 3 ) , particularly given the need to support the metabolically costly activity of flight . Interestingly , these values are equivalent to the mean minimum arterial PO2 values obtained near the end of dives in elephant seals , and are similar to the range exhibited by diving emperor penguins ( Ponganis et al . , 2007; Fedak et al . , 1981 ) . These PO2 values correspond to quite different blood oxygen saturation ( SO2 ) values between these species , however , due to the inherent differences between the flight environment and breath-hold diving and their subsequent effects on the O2-Hb dissociation curve . For example , the associated hyperventilation ( decreased CO2 ) and decrease in temperature ( below ) in the flying goose correspond to a much higher arterial O2 content for the same low levels of PO2 experienced between these species ( Meir and Milsom , 2013 ) . As near complete utilization of the available O2 store ( venous PO2 values near zero at the end of dives ) certainly contributes to the success of elite divers like elephant seals and emperor penguins ( Ponganis et al . , 2007; Meir et al . , 2009 ) , the capacity to effectively maximize O2 resources in the O2-limited environment of high altitude flight would also afford a distinct advantage . With venous O2 values decreasing to only around 25–30 mmHg in the present study , even under extreme hypoxia , these high fliers may yet retain a venous O2 reserve , also suggesting that these birds were not O2 limited in hypoxic flight . Venous PO2 values as low as 2–10 mmHg have been reported during dives in elite divers like elephant seals and emperor penguins ( Ponganis et al . , 2007; Meir et al . , 2009 ) , or in hypoxemic extremes of race horses performing strenuous exercise ( Manohar et al . , 2001; Butler et al . , 1993; Bayly et al . , 1989 ) . Hopefully , further gains made in the field of bio-logging systems directly measuring PO2 or SO2 will elucidate these variables in wild , migrating birds in the future . One of our key findings was the consistent fall in venous temperature during flight ( Figure 3 and Figure 4 ) . As the temperature probes were inserted through the jugular vein and advanced to the level of the heart , these records should reflect true mixed venous temperature . The decrease in temperature may indicate cooling of blood flowing through the buccal/pharyngeal cavity via evaporative water loss from the respiratory passages and/or restriction of blood flow to the gut . As it has been demonstrated that the blood of the bar-headed goose has high thermal sensitivity ( Meir and Milsom , 2013 ) , this drop in temperature could enhance O2 loading considerably as the blood subsequently cools in the lung ( Meir and Milsom , 2013 ) . For example , when converting venous PO2 values ( Figures 3 , 4 , Supplementary file 2 ) into Hb-O2 saturation ( Meir and Milsom , 2013 ) , the corresponding temperature drop in hypoxia results in a substantial increase in O2 content , indicating an even larger venous reserve than that inferred from the PO2 values alone . As blood travels away from the lung toward the exercising tissue , it would be expected to warm , enhancing O2 unloading . Temperature profiles also reveal a transient spike in temperature immediately following each flight , perhaps due to a release of warm blood from exercising muscle or other areas . This characteristic spike is followed by a second bout of cooling , and then a slow warming to levels at rest ( Figure 4 ) . Measurements of temperature at the lung and at the muscle in wild , migrating birds would help determine if modulating blood temperature might increase oxygen flux during flight . Previous studies have documented wild , migrating geese flying regularly between 5 , 000–6 , 000 m above sea-level and as high as 7 , 290 m ( Bishop et al . , 2015; Hawkes et al . , 2013 ) , with earlier anecdotal reports suggesting that these birds may even fly as high as the summit of Mt . Makalu ( 8 , 485 m; Swan , 1961 ) . Based on the ability of geese flying under hypoxic conditions in the present study in a wind tunnel , we believe that , although these geese routinely use lower mountain passes during their migration , their suite of physiological adaptations could support flight even at extreme altitudes . We suggest that this would largely be possible via a reduction in metabolism in hypoxia , while maintaining the heart rate and relative-increase in O2 pulse also measured in flight in normoxia . Interestingly , blood temperature dynamics may also play a critical role in enhancing O2 loading in this species during its exceptional migration . To facilitate wind tunnel training , geese were imprinted on the experimenters . Briefly , bar-headed goose ( Anser indicus ) eggs were obtained from the Sylvan Heights Bird Park ( Scotland Neck , North Carolina ) . Geese ( twelve bar-headed geese in 2010 , seven in 2011 ) were imprinted on a human foster parent ( J . U . M . in 2010 and J . M . Y . in 2011 ) during the first several weeks at the waterfowl park and then transported to Vancouver , B . C . , Canada ( in accordance with U . S . Department of Agriculture Animal and Plant Health Inspection Services and Canadian Food Inspection Agency protocols/inspections ) . They were then housed in the University of British Columbia Animal Care Facility with constant access to water ( small ponds ) and ad libitum mixed grain pellets supplemented with lettuce for the duration of the project . Birds were familiarized with dummy respirometry masks and backpack systems soon after hatching . Because the wind tunnel was undergoing repair when the first year's ( 2010 ) birds fledged , they were initially taken on outdoor training flights alongside their foster parent on a bicycle , and later on a motor scooter , to facilitate development of flight muscle and physiological capacity ( Video 1 and Figure 5—figure supplement 1 ) . Geese were flown in the University of British Columbia ( UBC ) Department of Mechanical Engineering’s boundary layer 30 m open-circuit wind tunnel ( http://mech . ubc . ca/alumni/aerolab/facilities/ ) . Airspeed in the test section ( 1 . 6 m high x 2 . 5 m wide x 23 . 6 m long ) between 3 to 20 m s−1 was calibrated using a pitot tube system built into the tunnel . Experimental flights took place primarily during times that corresponded to spring and fall migration of wild bar-headed geese ( Jan . 2011-Nov . 2012 ) . The range of wind tunnel flight speeds selected was similar to that measured during natural migratory flight ( 14 to 21 m s−1; Hawkes et al . , 2013; Hawkes et al . , 2011 ) and the speed selected for each individual was that which allowed steady , stationary , and prolonged flight . Three birds flew at 12 . 5 m s−1 , one bird at 13 . 75 m s−1 , and three birds at 15 m s−1 . As the wind tunnel is an open-loop system continuously drawing in outside air while operating , temperature in the wind tunnel was equivalent to ambient local outdoor temperature ( range: 3–21°C ) for each flight . The foster parent stood against the wall at the front of the flight section of the wind tunnel to encourage the bird to sustain flight . Another investigator initially lifted the bird into the air stream from behind , then supported the tubing running from the mask to the data acquisition system , holding them 2–3 feet above and behind the bird to allow free movement of the flying bird ( Figure 5 and Video 2 ) . Seven bar-headed geese ( 2 . 21 ± 0 . 26 kg ) managed steady , stationary , and prolonged flight in the wind tunnel while fully instrumented . We measured heart rate ( fH ) , the rate of oxygen consumption ( V˙o2 ) and the rate of CO2 production ( V˙CO2 ) under conditions at rest and during flight in bar-headed geese in both normoxia and two levels of hypoxia ( moderate: 0 . 105 and severe: 0 . 07 FiO2 equivalent to altitudes of roughly 5 , 500 m and 9 , 000 m respectively ) . Subcutaneous electrodes were inserted dorsally proximal to the spine: one at the level of the axilla and the second near the pelvis . The electrodes were connected to a custom-built 3-channel PO2/temperature/electrocardiogram ( ECG ) digital recorder ( UFI , Morro Bay , CA , USA ) ( Meir et al . , 2008; Ponganis et al . , 2007; Fedak et al . , 1981; Meir et al . , 2009; Ponganis et al . , 2009 ) , which sampled at 100 Hz ( Meir et al . , 2008; Ponganis et al . , 2007 ) . V˙o2 and V˙CO2 were measured using mask respirometry . Two ports in the mask drew ambient , normoxic air in and over the nares via space at the top of the mask and introduced oxygen-free nitrogen from behind the nares such that it mixed with the ambient air to provide a hypoxic gas mix flowing over the nares for the goose to breathe ( 0 . 105 and 0 . 07 FiO2 ) . The airflow rate through the mask was 70 l min−1 during flight and 10 l min−1 at rest , which was sufficient to prevent any leakage from the mask , tested using nitrogen dilutions [16] . A subsample ( 200 ml min−1 ) of air from the mask was drawn through a Sable Systems Field Metabolic System ( FMS ) ( Sable Systems , Las Vegas , NV , USA ) , via a desiccant membrane dryer ( AEI Technologies , Pittsburgh , PA , USA , Figure 5 and Video 2 ) , which was calibrated at the start and end of each trial . Masks were custom-made on a Plaster of Paris cast of the head of a deceased bar-headed goose using heat-moldable dental mouth-guard compound ( Thermo-Forming Material , Clear-Mouthguard , . 040’ , Henry Schein , Canada ) , which was softened with a heating gun and stretched over the cast to create a light-weight , form-fitted mask that could be secured with a thin elastic strap below the base of the skull . The mask covered the beak and forehead of the goose but did not cover the eyes . The tubes sampling air and delivering nitrogen to the mask exited the tunnel to the respirometry set-up at an access point ( Figure 5 ) . Gas ( air or the hypoxic gas mixture ) was drawn from the mask by a dry rotary vane vacuum pump ( 4 . 5 cfm , 115 VAC , Cole-Parmer , Montreal , Quebec , CA , equipped with vacuum gauge and vacuum relief valve ) controlled with a digital mass flow controller ( Sierra Instruments Smart-Trak100 , 0–200 SLPM , Accuracy: + 1% of full scale , BG Controls , Port Coquitlam , BC , Canada ) . A 10 μM nylon net filter ( Millipore , Billerica , MA , USA ) was used to prevent dust , down , or debris from entering the flow controller . Rotameters ( Acrylic Flowmeter , FL-2042 , 3 to 30 l min−1 , FL-2044 , 10 to 100 l min−1 , Omega , Laval , QC , Canada ) were used to generate flow rates of oxygen-free N2 sufficient to produce 0 . 105 and 0 . 07 FiO2 in the respirometry mask for both flight and rest , using the Plaster of Paris goose head mold for the calibrations . The gas analyzer was calibrated to account for sensor drift using: 1 ) two point calibration for CO2 , 0% and 1 . 0% CO2 balance air ( Praxair Canada , Scarborough , ON , Canada ) ; 2 ) a single point calibration for O2 at a baseline of 20 . 95% for dried room air at experimental flow rates since zero is extremely stable ( Fedak et al . , 1981 ) . To determine whether arterial blood oxygen was maintained in flight , and the extent of the venous blood O2 reserve remaining following tissue O2 extraction , arterial and venous blood Po2 and temperature were measured at rest and during flight using intravascular Po2 electrodes ( Licox Po2 microprobe , Canada Microsurgical Ltd . , Burlington , ON , Canada ) and thermistors ( Yellow Springs Instruments model 555 , Fisher Scientific , Edmonton , AB , Canada ) , introduced using aseptic surgical technique under general isoflurane anesthesia , with meloxicam as an analgesic . Only one site , either arterial or venous , was targeted per surgery and subsequent flights ( n=5 birds ) . For venous deployments , Po2 electrodes and thermistors were inserted percutaneously via the right jugular vein using a peel-away catheter over needle ( Arrow 15 Ga , Teleflex Medical , Markham , Ontario , Canada; similar to methods described in Meir et al . , 2008; Ponganis et al . , 2007; Meir et al . , 2009; Ponganis et al . , 2009 ) . Electrodes were inserted to lie close to the heart to sample mixed venous blood ( ranging from 9 to 13 . 5 cm from insertion site to cannula tip , depending on the bird and insertion site ) . For arterial deployments , Po2 electrodes were inserted in the aorta via the carotid artery using peel-away catheters ( 3 . 5 FR Peel-Away Denny Sheath Introducer Set , Cook Medical Inc , Bloomington , IN , USA or Arrow 17 Ga , Teleflex Medical ) after exposing the vessel via a shallow incision . The thermistor could not be deployed simultaneously with the arterial Po2 electrode due to aortic size . Cannulae were coiled and secured with a purse string suture at the insertion site , and covered with medical tape . Bioclusive transparent film dressing ( Henry Schein , Melville , NY , USA ) was placed over the insertion site of the electrodes and Elastinet stocking placed over the neck to protect the insertion site , secured by Bioclusive at each end . Animals recovered overnight from surgical procedures before experimental sessions in the wind tunnel were conducted . At the start of each experiment the Po2 electrode and thermistor were attached to the custom-built recorder ( see Ponganis et al . , 2007; Meir et al . , 2009; Ponganis et al . , 2009 ) . The Po2 electrode and thermistor , calibration procedures and verification testing have been described previously ( Meir et al . , 2008; Ponganis et al . , 2007; Meir et al . , 2009; Ponganis et al . , 2009 ) . At the end of the experiments the cannulae were removed and the animals inspected by veterinary surgeons and recovered in outdoor aviaries . V˙o2 was calculated as the difference between the fractional concentrations of O2 in dry inspired ( FiO2 ) and expired ( FeO2 ) air ( Lighton , 2008 ) as follows:V˙O2=V˙STPD ( FiO2−FeO2 ) −FiO2 ( FeCO2−FiCO2 ) ( 1−FiO2 ) Where V˙STPD is the flow rate of the gas being drawn from the mask . V˙CO2 was calculated as:V˙CO2=V˙STPD ( FeCO2−FiCO2 ) +FiCO2 ( FiO2−FeO2 ) ( 1+FiCO2 ) The start and end of each flight was determined from the data trace by an obvious change in CO2 production . Data were used only if a stable plateau in CO2 production had been reached . FeO2 and FeCO2 were determined as the average across this entire portion of the trace . Stable data were obtained under all conditions for V˙CO2 , however it was not possible to gather reliable V˙O2 data in hypoxia ( as in other studies: Hawkes et al . , 2014 ) . Therefore , V˙O2 data are not reported for flights in hypoxia . Respiratory exchange ratios ( RER ) were calculated by dividing V˙CO2 by V˙O2 and could therefore only be calculated for data collected in normoxia . ECG data were analyzed using peak detection software to automatically mark R-waves ( all data were then visually verified ) . Heart rate was counted as the mean during the same period used above for respirometry analysis . All Po2 values were temperature corrected for construction of Po2 profiles as previously described ( Ponganis et al . , 2007 ) . For arterial deployments in which temperature data could not be obtained , temperature was assumed to be stable at baseline body temperature ( 41°C ) . CO2 pulse ( an indication of how much CO2 is transported in the blood by each heartbeat ) was calculated by dividing the rate of CO2 production ( V˙CO2 ) by heart rate ( fH ) :CO2 pulse=V˙CO2fH According to the Fick equation , this is equivalent to the product of stroke volume ( SV ) and the arterial-venous CO2 difference ( CaCO2-CvCO2 ) . CO2 pulse=V˙CO2fH=SV ( CaCO2−CvCO2 ) As stroke volume ( SV ) and the arterial-venous CO2 difference were not measured in our study , however , our data cannot differentiate between the two . O2 pulse was also estimated during normoxic flights to calculate putative V˙O2 during hypoxic flights , assuming an RER of 1 . 0 to convert V˙CO2 into V˙O2 . Data were plotted using Origin2016 software ( OriginLab , Northampton , MA , USA ) . To statistically investigate our data while accounting for repeated measures among individual birds we used a linear mixed model approach with a random effect of the individual bird . This accounts for within-subject variance by assigning each bird an individual intercept . We generated individual models for each dependent variable ( duration , RER , V˙CO2 , heart rate , CO2 pulse , blood PO2 , venous temperature ) and compared main effects of oxygen level ( partial pressure of oxygen ) and activity or time point , as well as the interaction ( oxygen level*activity ) and compared estimated marginal means post-hoc assuming significance at p<0 . 05 . We used the afex package in RStudio ( R version 3 . 5 . 1 ) for generating the models , the emmeans package for post-hoc comparisons with Bonferroni adjustment where appropriate , and calculated the adjusted intraclass correlation coefficient ( ICC ) by dividing the variance of the random intercept by the sum of the random effect variances ( a value closer to 1 indicates a greater effect of the individual bird ) . We report estimated marginal means ( EMM ) in the results where indicated , and descriptive statistics in Table 1 . For comparisons involving one individual bird , dependent variables were compared using SigmaPlot software ( Systat Software Inc , San Jose , CA , USA ) based on oxygen level , activity , or time point using t-tests or one-way ANOVAs with post-hoc Tukey tests where appropriate . When normality of data was not achieved , groups were compared using Kruskal-Wallis one-way ANOVA on ranks with post-hoc Dunn’s test assuming significance at p<0 . 05 . R script and data files ( including source data for figures , although figures were not generated in R ) were deposited in Dryad ( doi:10 . 5061/dryad . fg80hp6 ) .	The bar-headed goose is famous for reaching extreme altitudes during its twice-yearly migrations across the Himalayas . These geese have been tracked flying as high as 7 , 270 meters up , and mountaineers have anecdotally reported seeing them fly over summits around Mount Everest ( that are over 8 , 000 meters tall ) . At these heights , the air is so thin that it contains only about 30–50% of the oxygen available at sea-level . Bar-headed geese have several adaptions that help them exercise in low oxygen conditions . For example , they have larger lungs than most other birds their size , and their red blood cells contain a version of hemoglobin that binds oxygen much more tightly . To date , however , there has been no work that has comprehensively measured how the bar-headed goose adapts its physiology to fly under low oxygen conditions . As such , it remains unclear whether these birds would even be able to fly where the oxygen is as limited as it is above the summits of the world’s highest mountains . This is partly because it is extremely challenging to make these kinds of recordings from flying geese , and partly because there are few wind tunnels in the world suitable to carry out such experiments . To better understand how the bar-headed goose accomplishes its remarkable , high altitude migration , Meir et al . raised bar-headed geese from eggs , with experimenters acting as the birds’ foster parents . The birds took their first flights either in a 30-meter wind tunnel at an engineering department in the University of British Columbia or , if the wind tunnel was unavailable , alongside a bicycle or a motor scooter . Once trained , the geese then flew in the wind tunnel wearing a backpack that contained the sensors needed to record their physiology . The birds also wore a breathing mask that could simulate the limited oxygen availability at altitudes of roughly 5 , 500 and 9 , 000 meters , and measure the oxygen consumed and the carbon dioxide produced by the geese . Meir et al . found that bar-headed geese could indeed fly at these simulated extreme altitudes in the wind tunnel , and that the birds largely achieved this by reducing their metabolism to match low oxygen conditions . The recordings show that the geese did not increase their heart rate when flying in reduced oxygen compared with normal flights , suggesting that their hearts were not working at maximum capacity despite the extreme conditions . Meir et al . also discovered that the blood in the birds’ veins cooled when flying , and in some cases by more than 2°C . Since hemoglobin’s affinity for oxygen changes with temperature , this may help increase the amount of oxygen that these birds can load into their blood at the lung when in flight . These measurements suggest that the anecdotes of bar-headed geese flying over some of the highest mountains in the world are indeed physiologically plausible . The findings will be valuable to researchers studying animals living at extreme altitudes . They may also be relevant to those looking to understand how humans respond to situations where oxygen is limited , such as during medical conditions like a heart attack or stroke , or procedures like organ transplants .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "ecology", "evolutionary", "biology" ]	2019	Reduced metabolism supports hypoxic flight in the high-flying bar-headed goose (Anser indicus)
Allosteric HIV-1 integrase ( IN ) inhibitors ( ALLINIs ) are a promising new class of antiretroviral agents that disrupt proper viral maturation by inducing hyper-multimerization of IN . Here we show that lead pyridine-based ALLINI KF116 exhibits striking selectivity for IN tetramers versus lower order protein oligomers . IN structural features that are essential for its functional tetramerization and HIV-1 replication are also critically important for KF116 mediated higher-order IN multimerization . Live cell imaging of single viral particles revealed that KF116 treatment during virion production compromises the tight association of IN with capsid cores during subsequent infection of target cells . We have synthesized the highly active ( - ) -KF116 enantiomer , which displayed EC50 of ~7 nM against wild type HIV-1 and ~10 fold higher , sub-nM activity against a clinically relevant dolutegravir resistant mutant virus suggesting potential clinical benefits for complementing dolutegravir therapy with pyridine-based ALLINIs . Multifunctional HIV-1 integrase ( IN ) is an important therapeutic target . HIV-1 IN strand transfer inhibitors ( INSTIs ) , which have become a first-line therapy to treat HIV-1 infected patients , block the catalytic function of the viral protein during the early phase of infection and thus prevent integration of viral cDNA into human chromosomes ( Hazuda , 2012; Hazuda et al . , 2000; McColl and Chen , 2010 ) . Clinical applications of first generation INSTIs raltegravir ( RAL ) and elvitegravir ( EVG ) resulted in evolution of drug resistant phenotypes containing IN substitutions in the vicinity of the inhibitor binding sites ( Cooper et al . , 2008; Garrido et al . , 2012 ) . Second generation dolutegravir ( DTG ) , bictegravir ( BIC ) and investigational drug cabotegravir ( CAB ) exhibit significantly higher genetic pressure for the evolution of drug resistance and yield complex resistance pathways with IN substitutions in both the proximity of and significantly distanced from inhibitor binding sites ( Cahn et al . , 2013; Eron et al . , 2013; Malet et al . , 2018; Margolis et al . , 2015; Smith et al . , 2018; Tsiang et al . , 2016 ) . More recently , allosteric HIV-1 integrase ( IN ) inhibitors ( ALLINIs; also referred to as non-catalytic site integrase inhibitors ( NCINIs ) ; LEDGINs or INLAIs ) , which are mechanistically distinct from INSTIs have been developed ( Christ et al . , 2010; Fader et al . , 2014b; Le Rouzic et al . , 2013; Sharma et al . , 2014; Tsiang et al . , 2012 ) . Unlike INSTIs , ALLINIs are highly potent during virion maturation . They induce higher-order IN multimerization in virions and consequently inhibit IN-RNA interactions , which in turn yields eccentric noninfectious virions with the ribonucleoprotein complexes ( RNPs ) mislocalized outside of the translucent capsid ( CA ) cores ( Balakrishnan et al . , 2013; Desimmie et al . , 2013; Fontana et al . , 2015; Gupta et al . , 2014; Jurado et al . , 2013; Kessl et al . , 2016; van Bel et al . , 2014 ) . ALLINIs also display a secondary , albeit significantly reduced , activity during early steps of HIV-1 infection where these compounds interfere with HIV-1 IN binding to its cognate cellular cofactor LEDGF/p75 ( Christ and Debyser , 2013; Christ et al . , 2012; Christ et al . , 2010; Kessl et al . , 2012; Le Rouzic et al . , 2013; Sharma et al . , 2014; Tsiang et al . , 2012 ) . The key pharmacophore in all potent ALLINIs includes both a carboxylic acid and tert-butoxy group , which extend from a core aromatic ring system ( Figure 1A ) ( Christ et al . , 2010; Jurado and Engelman , 2013; Sharma et al . , 2014; Tsiang et al . , 2012 ) . The original series of compounds contained a quinoline core , exhibited relatively modest potency , and comparatively low genetic barrier to drug resistance ( Christ et al . , 2010; Fader et al . , 2014b; Feng et al . , 2013; Tsiang et al . , 2012 ) . Follow up SAR studies have extended in two directions: i ) efforts that retain the quinoline core while varying the substituted groups have resulted in highly potent inhibitors such as BI224436 with antiviral EC50 of ~14 nM ( Fader et al . , 2014b ) ( Figure 1A ) ; ii ) other studies successfully explored different core ring structures ( Demeulemeester et al . , 2014; Fader et al . , 2016; Patel et al . , 2016a; Patel et al . , 2016b; Sharma et al . , 2014 ) to also synthesize highly potent ALLINIs . We and others have rationally developed pyridine-based compounds that exhibit markedly improved antiviral activities and significantly enhanced genetic pressure for the evolution of resistance compared with archetypal quinoline-based compounds ( Fader et al . , 2016; Hoyte et al . , 2017; Sharma et al . , 2014 ) . For example , our lead racemic pyridine-based compound , KF116 ( Figure 1A ) , exhibited ~24 nM antiviral activity and remained fully potent with respect to A128T IN mutant virus that confers significant resistance to archetypal quinoline-based ALLINIs . Instead , triple IN substitutions ( T124N/V165I/T174I ) , which substantially compromise viral replication , were needed to confer effective resistance to KF116 ( Hoyte et al . , 2017; Sharma et al . , 2014 ) . Furthermore , extensive in vitro absorption , distribution , metabolism , and excretion ( ADME ) and animal pharmacokinetic studies of quinoline- and pyridine-based inhibitors performed by Boehringer Ingelheim ( BI ) have indicated the superiority of the pyridine-based compounds as ‘a clinically viable starting point’ ( Fader et al . , 2016; Fader et al . , 2014a; Fenwick et al . , 2014 ) . IN is comprised of three structurally distinct domains: N-terminal domain ( NTD ) , catalytic core domain ( CCD ) and C-terminal domain ( CTD ) ( Chiu and Davies , 2004; Engelman and Cherepanov , 2014 ) . Biochemical and x-ray crystallography experiments with individual domains revealed that ALLINIs selectively bind to a v-shaped pocket located at the CCD dimer interface ( Fader et al . , 2014b; Feng et al . , 2013; Sharma et al . , 2014 ) . Mechanistic studies with various ALLINI core ring structures have shown that the cores do not directly interact with IN . Instead , they enable the conserved pharmacophore carboxylic acid and tert-butoxy groups to accurately position within the v-shaped pocket allowing hydrogen bonding with backbone amides of IN residues Glu-170 and His-171 and the side chain of Thr-174 , respectively . Mass spectrometry-based protein foot-printing experiments ( Feng et al . , 2016; Shkriabai et al . , 2014 ) with full-length IN identified that in addition to the CCD , the CTD plays a critical role for inhibitor induced hyper-multimerization of IN . In turn , these findings enabled molecular modeling studies , which suggested that ALLINIs directly promote inter‐subunit interactions between the CCD dimer and the CTD of another IN dimer ( Deng et al . , 2016 ) . Furthermore , free energy calculations revealed that in the absence of the inhibitor the v-shaped binding cavity on the CCD dimer is occupied with thermodynamically unstable water , which minimizes potential CCD-CTD interactions . In contrast , ALLINI binding to the CCD dimer effectively fills the v-shaped pocket by displacing the water molecules and thus markedly enhances the binding interface for incoming CTD . More recently , the crystal structure of full-length HIV-1 IN Y15A/F185H in complex with quinoline-based ALLINI GSK1264 was reported ( Gupta et al . , 2016 ) , which confirmed both the earlier biochemical results and predicted model of a multimer interface involving CCD-inhibitor-CTD interactions . The crystal structure revealed two IN Y15A/F185H dimers bridged by two quinoline-based GSK1264 inhibitors through head ( CCD-CCD dimer ) to tail ( CTD of another dimer ) interactions . However , technical limitations of the crystallographic experiments ( Gupta et al . , 2016 ) have left the following important questions unanswered: ( i ) what is the relevant oligomeric state of WT IN in the context of ALLINI induced higher-order multimerization ? The crystal structure used mutant IN with Y15A and F185H substitutions , which alters the in vitro multimeric state of IN ( IN Y15A/F185H and WT IN are predominantly dimer and tetramer , respectively ) . The Y15A substitution impairs HIV-1 replication ( Takahata et al . , 2017 ) , however Y15A/F185H substitutions were critical to slow down the inhibitor induced IN multimerization allowing for the capture of smaller complexes amenable to x-ray crystallography . Therefore , it is unclear which oligomeric state ( s ) of WT IN is ( are ) the authentic target for these inhibitors . ( ii ) Is there a role for the NTD in the inhibitor induced higher-order IN multimerization ? The NTD structure could not be resolved in the x-ray structure of the IN Y15A/F185H + ALLINI GSK1264 complex . Therefore , a role for the NTD in inhibitor induced higher-order IN multimerization remains to be elucidated . ( iii ) How do distinct ALLINI scaffolds induce higher-order IN multimerization ? Pyridine-based compounds are structurally distinct from their quinoline-based counterparts . Therefore , it is critically important to understand similarities and differences for how pyridine- vs quinoline-based ALLINIs induce hyper-multimerization of WT IN . To address the above questions , we have investigated key structural determinants of WT IN for ALLINI induced higher-order multimerization . Our experiments focused on analyzing the mode of action of lead pyridine-based ALLINI KF116 , while using the highly potent quinoline-based BI224436 in side-by-side comparisons ( Figure 1A ) . We demonstrate that KF116 selectively and BI224436 preferentially target WT IN tetramers . Furthermore , we have synthesized the highly active ( - ) -KF116 enantiomer , which exhibits substantially enhanced , sub-nM potency with respect to the therapeutically important DTG resistant virus containing IN N155H/K156N/K211R/E212T substitutions ( Malet et al . , 2018 ) . These findings suggest potential clinical benefits for combining KF116 and DTG therapies to limit HIV-1 options for developing drug resistant variants in patients . In vitro preparations of full-length WT HIV-1 IN yield a mixture of tetrameric , dimeric and monomeric forms . To examine which of these forms are targeted by ALLINIs , we have separated different oligomeric forms of WT IN by size exclusion chromatography ( SEC ) ( Figure 1B ) . To prevent re-equilibration of separated species , equimolar concentrations of tetramers , dimers and monomers were immediately incubated with KF116 and formation of the inhibitor induced higher-order IN multimers were monitored by dynamic light scattering ( DLS ) . DLS is an optical method for studying the diffusion behavior of macromolecules in solution ( Stetefeld et al . , 2016 ) . While unliganded IN does not yield any detectable signal due to relatively small sizes of fully soluble monomeric , dimeric and tetrameric forms , ALLINI induced higher-order IN oligomers are readily detected by DLS ( Sharma et al . , 2014 ) . The results in Figure 1C show that KF116 specifically induced higher-order oligomerization of tetramers but not dimers or monomers . More detailed kinetic analysis ( Figure 1—figure supplement 1A ) revealed that within 1 min after addition of KF116 to IN tetramers , higher-order protein multimers with particle diameter sizes of ~200 nm were formed . Particle diameter sizes increased further to ~1 , 000 nm in a time dependent manner indicating an equilibrium shift towards higher-order multimers ( Figure 1—figure supplement 1A ) . In sharp contrast , the inhibitor failed to alter the multimeric form of dimers or monomers even after a 60 min incubation ( Figure 1C and Figure 1—figure supplement 1B ) . The absence of any higher-order multimerization of these preparations provides corollary evidence that the separated lower oligomeric forms of IN do not detectably re-equilibrate into tetramers over the incubation time . In parallel reactions , we have analyzed the quinoline-based BI224436 , which exhibited a broader specificity for tetramers and dimers but did not have any effects on IN monomers ( Figure 1D and Figure 1—figure supplement 1C–D ) . Addition of BI224436 to IN tetramers yielded higher-order multimers within 1 min ( Figure 1—figure supplement 1C ) , whereas a longer incubation time ( at least 30 min ) for BI224436 +IN dimers was needed to detect higher-order IN multimers ( Figure 1—figure supplement 1D ) . Collectively , these findings indicate that KF116 selectively and BI224436 preferentially targets IN tetramers . The crystal structure of IN ( Y15A/F185H ) +GSK1264 ( Gupta et al . , 2016 ) revealed that two IN ( Y15A/F185H ) dimers are bridged by two quinoline-based inhibitors ( i . e . , 2:2 stoichiometry ) . Here we determined the stoichiometry for KF116 and BI224436 interactions with WT IN . For this , we have quantified dose dependent effects of ALLINI induced IN aggregation ( Figure 2 and Figure 2—figure supplement 1 ) . The results in Figure 2 indicate a 2:4 ratio for KF116:IN suggesting that two molecules of KF116 bridge two IN tetramers . In contrast , BI224436:IN interactions exhibited a 2:2 or 4:4 ratio ( Figure 2 and Figure 2—figure supplement 1 ) , similar to the prior crystal structure ( Gupta et al . , 2016 ) . Previous studies from us and others ( Deng et al . , 2016; Gupta et al . , 2016; Shkriabai et al . , 2014 ) have elucidated the importance of CCD-ALLINI-CTD interactions but a role for the NTD in inhibitor induced higher-order multimerization has not been examined . The NTD was disordered and could not be resolved in the crystal structure of IN Y15A/F185H + GSK1264 ( Gupta et al . , 2016 ) . Therefore , we compared how KF116 and BI224436 affected full-length WT IN and its various truncated constructs . DLS results in Figure 3A and B , show that KF116 and BI224436 promoted higher-order multimerization of only full-length IN , whereas all truncated protein constructs ( CCD , NTD-CCD and CCD-CTD ) exhibited marked resistance . In addition to DLS experiments , which monitored higher-order protein multimerization within the first 60 mins after the addition of ALLINIs to IN , we conducted aggregation assays that detected insoluble aggregates formed after ~16 hr incubation of IN with the inhibitors ( Figure 3C , D and Figure 3—figure supplement 1 ) . Addition of KF116 or BI224436 aggregated full-length WT IN in a dose dependent manner , whereas the CCD and NTD-CCD , both of which lacked the CTD , remained fully soluble even after incubation with 20 µM inhibitors ( Figure 3C , D and Figure 3—figure supplement 1 ) . These observations are consistent with the published results from our and other groups indicating that both the CCD and CTD are required for ALLINI induced higher-order IN oligomerization ( Deng et al . , 2016; Gupta et al . , 2014; Gupta et al . , 2016; Shkriabai et al . , 2014 ) . Interestingly , the CCD-CTD , which contained all inhibitor interacting interfaces but which lacked the NTD , also exhibited substantial resistance to both KF116 and BI224436 induced aggregation ( Figure 3C , D ) . Of note , only residual aggregation of CCD-CTD was detected at higher KF116 and BI224436 concentrations ( >10 µM ) , whereas these inhibitor concentrations fully precipitated full-length WT IN . Collectively , these findings ( Figure 3 and Figure 3—figure supplement 1 ) indicate that in addition to the CCD and CTD , the NTD is critically important for the inhibitor induced higher-order IN multimerization . Unlike the CCD and CTD , the NTD does not directly interact with the inhibitors but this domain is essential for functional tetramerization of IN ( Hare et al . , 2009 ) . Specifically , the NTD from one dimer engages with the CCD of another dimer to help form stable tetramers . In addition , the NTD interacts with the α-helical linker ( 200-222 ) connecting the CCD with the CTD . Therefore , to probe the significance of IN tetramerization for ALLINI induced higher-order IN multimerization , we targeted these protein regions ( the NTD and the CCD-CTD linker ) with site directed mutagenesis to specifically compromise functional IN tetramerization without affecting the direct sites of ALLINI binding . The H12A IN substitution , which destabilizes binding of the architecturally important Zn2+ , markedly altered IN oligomerization , yielding a mixture of monomers and higher-order oligomers ( Figure 4 and Figure 4—figure supplement 1 ) . H12A IN was susceptible to BI224436 but exhibited marked resistance to KF116 ( Figure 4 and Figure 4—figure supplement 2 ) . The K14A change destabilized NTD-CCD interactions needed for functional IN tetramerization and resulted in IN dimers ( Figure 4 and Figure 4—figure supplement 1 ) . KF116 was completely inactive , whereas BI224436 remained active against K14A IN ( Figure 4 and Figure 4—figure supplement 2 ) . Y15A IN , which yielded a mixture of dimers and monomers , was resistant to both KF116 and BI224436 . Y15A/F185H IN , which was utilized in the x-ray structure with quinoline-based compounds ( Gupta et al . , 2016 ) and which yielded dimers , was susceptible to quinoline-based BI224436 but this mutant was completely resistant to pyridine-based KF116 ( Figure 4 and Figure 4—figure supplement 2 ) . Next , we targeted the CCD-CTD α-helical linker by site directed mutagenesis ( Figure 4 and Figure 4—figure supplements 1–4 ) . We incorporated a single Pro residue between T210 and K211 to introduce a kink in the α-helix connecting the CCD with the CTD . T210 +Pro IN was completely resistant to both KF116 and BI224436 indicating that the linear α-helix is critically important for accurate orientation of the CCD-inhibitor-CTD interfaces . Similarly , deletion of A205 within the α-helix rendered IN resistant to both KF116 and BI224436 . The substitutions in the NTD or the CCD-CTD connecting α-helix , both of which compromised functional tetramerization of IN , also impaired virus infectivity and inhibited IN catalytic activity in vitro ( Figure 4—figure supplements 3 and 4 ) ( Hare et al . , 2009 ) . Furthermore , these substitutions also adversely affected LEDGF/p75 binding with lesser effects seen for the T210 +Pro substitution , which is significantly distanced from LEDGF/p75 binding sites in the NTD or CCD ( Figure 4—figure supplement 3 ) . The control N222K substitution in the CCD-CTD connecting α-helix formed tetramers , retained WT levels of HIV-1 replication in cell culture , IN catalytic activity and LEDGF/p75 binding in vitro , and was fully susceptible to both KF116 and BI224436 induced higher-order IN multimerization ( Figure 4 and Figure 4—figure supplements 1–4 ) . Collectively , the experiments with select substitutions in the NTD and CCD-CTD linker , which allowed us to compromise functional IN tetramerization without directly affecting ALLINI binding sites , revealed a consistent correspondence between the effects of the substitutions on IN tetramer formation and the ability of KF116 to induce higher-order multimerization . Moreover , our findings further highlighted both key similarities and differences between pyridine- and quinoline-based compounds . Published structural studies ( Hare et al . , 2009 ) have revealed key interactions that govern IN tetramerization . For example , an ionic bond between E11 in the NTD of one dimer and K186 within the CCD of another dimer is important for functional tetramerization of IN ( Hare et al . , 2009 ) ( Figure 5A ) . Individual E11K and K186E IN substitutions fully compromise protein tetramerization ( without affecting IN dimerization ) due to charge-charge repulsions . However , mixing two mutant proteins ( E11K and K186E INs ) in vitro allows for partial reconstitution of functional IN tetramers through reversed charge-charge interactions ( Hare et al . , 2009 ) . Indeed , the results in Figure 5—figure supplement 1 show that individual E11K and K186E substitutions compromised the catalytic activities of IN and its ability to bind LEDGF/p75 , whereas these functions of IN were substantially restored through reconstituting the two inactive proteins ( E11K + K186E INs ) . Therefore , we used this powerful trans-complementation system for testing the significance of IN tetramerization for KF116 and BI224436 activity . DLS results in Figure 5B revealed that addition of KF116 to either E11K or K186E IN dimers did not induce higher-order protein multimerization . However , when two resistant IN mutants ( E11K and K186E ) were mixed together to allow reconstitution of IN tetramers and then incubated with KF116 under identical reaction conditions , we observed formation of higher-order protein multimers ( Figure 5B ) . These findings provide another line of evidence that KF116 exhibits marked selectivity for IN tetramers . In parallel experiments , BI224436 induced higher-order multimerization of dimeric K186E and reconstituted tetrameric E11K + K186E INs ( Figure 5C ) , which provide further support for a broader specificity of BI224436 for tetramers and dimers . However , E11K IN was resistant to BI224436 suggesting that this substitution , in addition to destabilizing IN tetramers , could also potentially affect the proper orientation of the NTD required for BI224436 induced higher-order IN multimerization of the IN dimers . Accordingly , we also note that E11K was significantly more defective for IN catalytic activity than was K186E IN ( Figure 5—figure supplement 1 ) . We used our experimental findings and available structural information to create molecular models for KF116 and BI224436 induced higher-order multimerization of WT IN . The full-length IN tetramer and dimer models were assembled based on the cryo-EM and x-ray crystal structures of lentiviral ( HIV-1 and MVV ) intasomes ( tetramer model ) and IN domains ( dimer model ) ( Ballandras-Colas et al . , 2017; Chen et al . , 2000; Gupta et al . , 2016; Hare et al . , 2009; Passos et al . , 2017; Wang et al . , 2001 ) . Next , the HADDOCK program , a data-driven protein-protein docking algorithm ( Dominguez et al . , 2003; van Zundert et al . , 2016 ) , was used to create molecular models for KF116 mediated tetramer-tetramer interactions ( Figure 6A ) . For this the crystal structure of KF116 bound to HIV-1 IN CCD dimer was used to establish the head-to-tail interactions between KF116 bound to the CCD-CCD dimer of one tetramer and CTD of another tetramer ( Sharma et al . , 2014 ) . Docking simulations using HADDOCK generated top ranked symmetric head-to-tail architecture , with the two tetramers connected by two CCD-KF116-CTD interfaces ( Figure 6A , B ) . This model is fully consistent with the experimentally observed stoichiometry of 2:4 for KF116:IN . The KF116 mediated buried surface area ( BSA ) between two tetramers was ~2000 Å2–3200 Å2 . Initial HADDOCK simulations yielded the BSA of ~2000 Å2 and further optimization of the symmetric interactions resulted in an improved BSA of 3200 Å2 . The optimization was performed by specifying all interface residues from the initial HADDOCK docked structure as the input for the second HADDOCK run . Strikingly , when we used IN dimers instead of tetramers as building blocks , the HADDOCK program failed to yield symmetric dimer:KF116:dimer assemblies . These findings are consistent with experimental results demonstrating marked selectivity of KF116 for IN tetramers . For comparison , we have also built multimer models containing BI224436 ( Figure 6—figure supplement 1 ) . For this , we have first solved the x-ray structure of BI224436 bound to IN CCD ( Figure 6C and Supplementary file 1 ( Table S1 ) ) . In common with all other potent ALLINIs , the carboxylic acid moiety of BI224436 maintains important hydrogen bonding interactions with the backbone amides of E170 and H171 . In addition , the hydrogen bonding interaction of the ether oxygen of the tert-butoxy and the side chain of T174 is conserved . However , there are the following important differences between BI224436 and KF116 binding ( Figure 6C ) : i ) the bulky nature of the tricyclic ring of BI224436 fully fills the hydrophobic pocket of the CCD dimer and extends closer to W132 , which caps the pocket . For comparison , KF116 contains a much smaller benzyl substitution at this position; ii ) the benzimidazole group , which is both unique and essential for KF116 activity , projects outside of the v-shaped pocket , whereas BI224436 is more deeply positioned within the pocket . Next , we used our x-ray structure ( Figure 6C ) to build models of BI224436 interactions with full-length IN ( Figure 6—figure supplement 1 ) . Symmetric multimers can be generated regardless of whether IN tetramers or dimers are used as building blocks suggesting that unlike KF116 , BI224436 can recruit both tetramers and dimers for higher-order IN multimerization . These in silico findings are fully consistent with the experimental results indicating that unlike KF116 , which is highly selective for IN tetramers , BI224436 exhibits a broader specificity for tetramers and dimers ( Figure 1 and Figure 1—figure supplement 1 ) . Our molecular models ( Figure 6A and Figure 6—figure supplement 1 ) are also consistent with experimental data showing the importance of the NTD for inhibitor induced higher-order IN oligomerization . Specifically , in the symmetric tetramer-KF116-tetramer model ( Figure 6A ) while the NTD does not directly engage the inhibitor , this domain plays two key architectural roles . First , the NTD of one dimer interacts with the CCD of another dimer to stabilize IN tetramers ( Hare et al . , 2009 ) . Second , the NTD interacts with the linear α-helix ( 200-222 ) connecting the CCD with CTD , which in turn could affect correct orientation of the CTD for inhibitor induced head-to-tail interactions . This latter interaction of the NTD with the CCD-CTD linker is also seen in the context of symmetric tetramer-BI224436-tetramer and dimer-BI224436-dimer assemblies ( Figure 6—figure supplement 1 ) . Thus , these modeling results are fully consistent with our experimental results indicating that NTD could contribute to both KF116 and BI224436 induced higher-order IN multimerization . Previously , we have reported antiviral activity of ~24 nM for racemic KF116 in single replication cycle assays ( Sharma et al . , 2014 ) . We have now synthesized ( - ) and ( + ) -KF116 enantiomers and assayed their antiviral activities during multiple rounds of HIV-1 replication in MT-4 cells . ( - ) -KF116 exhibited an IC50 of ~7 nM , which was ~30 times more potent than its ( + ) counterpart ( Figure 7A and Figure 7—figure supplement 1A ) . Next , we evaluated the metabolic stability of ( - ) -KF116 using rat and human liver microsomes ( Figure 7—figure supplement 1B ) . We probed in vitro Cytochrome ( CYP ) P450 activity in the presence of co-factor NADPH ( Wempe and Anderson , 2011; Wempe et al . , 2012a; Wempe et al . , 2012b ) and monitored ALLINI stability by LC-MS . In vitro half-life measurements and calculated intrinsic clearance values in Figure 7—figure supplement 1B show that control compounds Verapamil , Domperidone and Chlorpromazine were metabolized as expected while ALLINIs displayed excellent metabolic stability toward CYP oxidation with ( - ) -KF116 exhibiting superior properties compared with racemic KF116 and quinoline-based BI224436 . Second generation INSTIs such as DTG , which bind at the IN catalytic site in the presence of viral DNA , display a high genetic barrier to resistance . Therefore , the drug resistance phenotypes emerging in the clinic in response to second generation INSTIs reveal complex resistance profiles with IN substitutions often seen outside of the inhibitor binding site . For example , a recent clinical study revealed that failure of DTG treatment in patients was observed with concomitant appearance of IN N155H/K211R/E212T substitutions on the background of the K156N polymorphic mutation ( Malet et al . , 2018 ) . N155 and K156 are within the CCD , in close proximity to the IN active site . In contrast , K211 and E212 are significantly distanced from the DTG binding site and instead these residues are located in the CCD-CTD connecting α-helix implicated by our modeling and site directed mutagenesis studies as critically important for KF116 induced higher-order IN multimerization ( Figures 4 and 6 ) . Therefore , we wanted to examine ( - ) -KF116 activity with respect to the DTG resistant virus . Interestingly , KF116 displayed sub-nM activity and was ~10 fold more active against HIV-1 containing IN N155H/K156N/K211R/E212T substitutions compared with the WT virus ( Figure 7B ) . In control assays , KF116 was similarly active against WT HIV-1 and the mutant virus containing only the CCD ( N155H/K156N ) substitutions ( Figure 7B ) . Based on our biochemical assays highlighting the significance of the CCD-CTD connecting α-helix for functional tetramerization of IN and KF116 activities , we wanted to also test how K211R/E212T substitutions affected oligomerization of recombinant IN . Our SEC data ( Figure 7C ) show that recombinant IN ( N155H/K156N/K211R/E212T ) is almost exclusively tetramer , whereas IN ( N155H/K156N ) , which is similar to WT IN ( Figure 7C ) , forms a mixture of tetramers , dimers and monomers . These findings provide further evidence for a strong correlation between IN tetramers and KF116 activity ( Figure 7B and C ) in the context of therapeutically important DTG resistant mutant viruses . ALLINI treatment induces higher-order multimerization of IN in virions , which compromises the ability of IN to bind the viral RNA genome and causes mislocalization of RNPs outside of the conical core made of capsid proteins ( CA ) ( Jurado and Engelman , 2013; Kessl et al . , 2016 ) . However , it is not clear whether mislocalized IN complexes in ALLINI-treated samples remain associated with the viral CA core and whether these complexes traffic together to the nucleus . To address this question , we employed live-cell imaging to visualize IN co-trafficking with the viral core . HIV-1 pseudoviruses were co-labeled with IN-mNeonGreen ( INmNG ) and cyclophilinA ( CypA ) fused to DsRed ( CypA-DsRed ) , a marker of the viral CA protein ( Francis et al . , 2016; Francis and Melikyan , 2018 ) . This co-labeling strategy enables real-time tracking of viral complexes in living cells and visualization of distinct steps of HIV-1 entry , uncoating , nuclear import and integration ( Francis et al . , 2016; Francis and Melikyan , 2018 ) . We have previously shown that productive uncoating ( loss of CypA-DsRed/CA ) that leads to infection occurs at the nuclear pore , prior to delivery of the IN complexes into the nucleus . By contrast , premature loss of CypA-DsRed/CA in the cytoplasm exposes the remaining IN complexes and leads to their proteasomal degradation ( Francis and Melikyan , 2018 ) . Fluorescent viruses were produced in the presence of ALLINIs , or , in control experiments , in the presence of DMSO or the INSTI RAL . The fate of single INmNG- and CypA-DsRed-labeled complexes during virus entry was examined in HeLa-derived cells depleted of CypA ( PPIA-/- ) by live-cell imaging . Cells depleted of CypA were chosen to reduce premature uncoating of the incoming post-fusion CA cores ( Francis and Melikyan , 2018 ) and thereby extend the window of observation for assessing the fate of INmNG in living cells . The viral particles produced in the presence of ( +/- ) -KF116 , ( - ) -KF116 and BI224436 were compared with DMSO and RAL controls ( Figure 8 and Figure 8—figure supplement 1 ) . Imaging of single viral complexes bound to a cover glass before or after 5 min of virus entry into target cells showed similar extents of colocalization between CypA-DsRed and INmNG puncta in all samples ( Figure 8—figure supplement 1 ) . However , imaging of these virions in the same target cells at 90 min post-infection revealed a marked drop in the number of co-labeled particles ( primarily due to a loss of IN signal ) for viruses produced in the presence of ALLINIs , but not for control viruses prepared in the presence of DMSO or RAL ( Figure 8A–B and Figure 8—figure supplement 1 ) . We next analyzed the pattern of INmNG loss from CypA-DsRed/CA labeled cores upon entry of ALLINI pretreated HIV-1 pseudoviruses . Single particle tracking revealed that DMSO-treated control viruses undergo uncoating manifested in a loss of CypA-DsRed from INmNG labeled complexes ( Figure 8C ) . In agreement with our previous studies showing proteasomal degradation of post-uncoating IN complexes in the cytoplasm ( Francis et al . , 2016; Francis and Melikyan , 2018 ) , we observed a gradual decay of the INmNG signal after the loss of CypA-DsRed signal from the viral complexes produced in the presence of the control DMSO solvent ( Figure 8C ) . In contrast , ALLINI treated virions lost the INmNG signal prior to the loss of CypA-DsRed/CA ( Figure 8D ) . These findings suggest that ALLINI induced INmNG aggregates are readily accessible to proteasomal degradation in the cytoplasm of target cells due likely to their mislocalization outside of the protective CA cores ( Figure 8—figure supplement 2 ) , which in turn precludes productive infection . We present several lines of evidence demonstrating that IN tetramers are the authentic targets for the lead pyridine-based KF116: ( i ) the inhibitor selectively induced higher-order multimerization of WT IN tetramers but not its lower oligomeric states; ( ii ) the NTD and α-helical CCD-CTD linker ( residues 200–222 ) , which do not directly interact with ALLINIs but which are essential for functional tetramerization of full length IN , are also critically important for KF116 induced higher-order IN multimerization; ( iii ) single amino acid substitutions that compromise IN tetramerization also adversely affect KF116 activity; ( iv ) trans-complementation of two KF116 resistant IN dimers allows a partial reconstitution of IN tetramers , which in turn become susceptible to KF116 induced higher order multimerization; ( iv ) a positive correlation between increased antiviral activity of KF116 against the clinically relevant DTG resistant mutant virus and enhanced propensity of the corresponding mutated IN to form tetramers . While the overall mechanism of action of pyridine- and quinoline-based ALLINIs are very similar , our findings highlight important differences between KF116 and BI224436 . The most notable is that BI224436 exhibits a broader specificity for both IN tetramers and dimers unlike KF116 , which is highly selective for IN tetramers . This correlates well with a prior NMR study that indicated that pyridine based ALLINIs exhibited higher binding affinity for IN tetramers compared to dimers ( Fader et al . , 2014a ) . Furthermore , BI224436 , unlike KF116 , is active against IN Y15A/F185H dimers . This was likely a key factor that enabled the generation of a successful x-ray structure for a quinoline-based GSK1264 ( which is very similar to BI224436 ) bound to IN Y15A/F185H ( Gupta et al . , 2016 ) . This structure provided breakthrough information about ALLINI induced head-to-tail interactions , where GSK1264 is seen sandwiched between a CCD dimer and the CTD of another dimer . Initial structural clues about differential interactions of pyridine- vs quinoline-based inhibitors with various oligomeric forms of WT IN come from our crystal structures of these compounds bound to the CCD dimer ( Figure 6C ) . Superimposition of IN CCD co-crystal structures with KF116 or BI224436 revealed both overlapping interactions as well as distinct contacts for each inhibitor with the CCD ( Figure 6C ) . BI224436 ( similarly to its very close analog GSK1264 ) is positioned deeper inside the v-shaped pocket ( Figure 6C ) , which could allow symmetrical CCD-BI224436-CTD interactions between both IN dimers and tetramers . In contrast , the bulky benzimidazole group , which is both unique and essential for KF116 activity , extends from the v-shaped pocket ( Figure 6C ) and could force the incoming CTD to accept a slightly different orientation compared to BI224436 or GSK1264 . This , in turn may hamper the dimer-dimer symmetry seen with GSK1264 ( Gupta et al . , 2016 ) but instead optimally position two IN tetramers for symmetrical CCD-KF116-CTD interactions ( Figure 6A ) . We show that in addition to direct CCD-ALLINI-CTD interactions , the NTD and α-helical CCD-CTD linker ( 200-222 ) are critically important for ALLINI induced higher-order multimerization . These protein segments do not directly interact with KF116 or BI224436 but are essential for functional tetramerization of IN . Within each IN tetramer , the NTD interacts with the CCD-CTD α-helical linker as well as the CCD of another dimer unit . These dual interactions allow the NTD to effectively stabilize the conformation of the CCD-CTD α-helical linker as well as the entire tetramer against thermal fluctuations . In turn , such conformational rigidity could facilitate symmetric tetramer-ALLINI-tetramer assemblies . This notion is supported by our mutagenesis experiments ( Figure 4 ) demonstrating that substitutions in the NTD and α-helical CCD-CTD linker ( 200-222 ) that compromise assembly of stable tetramers also adversely affect KF116 induced higher-order oligomerization . Since both KF116 and BI224436 are highly potent during virion maturation and only IN tetramers can be targeted by both compounds , we hypothesize that the predominant form of IN in virions resides as a tetramer . It should also be noted that IN tetramers bind viral RNA and are the essential building blocks for super molecular assembly of lentiviral intasomes ( Ballandras-Colas et al . , 2017; Kessl et al . , 2016; Passos et al . , 2017 ) . These functionally critical IN tetramers are selectively and preferentially targeted by KF116 and BI224436 , respectively ( Figure 1 and Figure 1—figure supplement 1 ) . ALLINI induced higher-order multimerization of IN in virions inhibits IN interactions with the viral RNA and results in mislocalization of RNPs outside of the translucent CA cores ( Fontana et al . , 2015; Jurado et al . , 2013; Kessl et al . , 2016 ) . While these eccentric virions can effectively enter the target cells , RNPs and IN are prematurely degraded during the early phase of infection ( Madison et al . , 2017 ) . To gain further insight into the mechanism of action of ALLINIs we have used INmNG and CypA-DsRed labeling of CA and monitored co-trafficking of these viral proteins during the early phase of infection . As expected ( Francis et al . , 2016; Francis and Melikyan , 2018 ) , in the absence of inhibitors , we observed the initial decay of CypA-DsRed signal likely due to the uncoating of CA cores followed by subsequent , substantially slower decay of INmNG signal . In sharp contrast , the ALLINI treated viruses exhibited a remarkable loss of INmNG signal from single particle post-fusion viral cores that were yet to complete uncoating . The schematic summary of these experimental observations are depicted in Figure 8—figure supplement 2A . In turn , the pre-mature loss of IN from the viral CA cores during infection of target cells is likely to be a consequence of aberrant packaging of ALLINI induced higher-order IN multimers within virions ( Figure 8—figure supplement 2B ) . Collectively , our current findings together with published results ( Kessl et al . , 2016 ) suggest that binding of functional IN tetramers to the viral RNA genome are important for effective packaging of these viral components within conical CA cores and formation of the infectious virions ( Figure 8—figure supplement 2 ) . In contrast , ALLINI treatment results in aggregation and separation of IN from RNPs . As a result , both IN-ALLINI aggregates and RNPs lacking IN are mislocalized outside of the protective CA cores and yield eccentric , non-infectious virions ( Figure 8—figure supplement 2B ) . The present studies argue for the further development of KF116 for its potential translation into clinical applications . Arguably , the most exciting finding is that KF116 exhibits enhanced , sub-nM activity against therapeutically important DTG resistant HIV-1NL4-3 ( IN N155H/K156N/K211R/E212T ) . These substitutions have recently been identified in a patient receiving DTG monotherapy and follow up cell culture based virology experiments confirmed substantial resistance of HIV-1NL4-3 ( IN N155H/K156N/K211R/E212T ) to DTG ( Malet et al . , 2018 ) . Our interest in this particular resistant phenotype has been prompted by the presence of the two K211R/E212T substitutions in the α-helical CCD-CTD linker . Differing from artificial mutations ( ΔA205 and T210 + Pro ) introduced in this segment , which both impaired proper IN tetramerization and conferred marked resistance to KF116 ( Figure 4 ) , the K211R/E212T substitutions instead stabilized IN tetramers and resulted in enhanced antiviral activity of KF116 ( Figure 7 ) . These results argue strongly for the positive correlation between functional IN tetramerization and the ability of KF116 to induce higher-order IN oligomerization . It should also be noted that a previous study has shown that combinations of ALLINIs and INSTIs exhibited additive or even synergetic effects to inhibit HIV-1 in cell culture ( Christ et al . , 2012 ) . Our observation that ( - ) -KF116 exhibits substantially enhanced , sub-nM activity with respect to DTG resistant HIV-1NL4-3 ( IN N155H/K156N/K211R/E212T ) provides further support for potential clinical benefits of combining these two inhibitors to treat HIV-1 infected patients . New therapies that target the novel function of HIV-1 IN are critical for successful treatment of HIV-1 in patients and our work will help to inform future studies as ALLINIs translate from the lab to the clinic . Racemic ( +/- ) -KF116 was synthesized as described previously ( Sharma et al . , 2014 ) . Synthesis of ( + ) and ( - ) enantiomers of KF116 is detailed in the supporting information ( Supplementary file 1 ) . BI224436 was purchased from MedChemExpress . Hexa-His tagged recombinant WT full length IN , NTD-CCD , CCD and CCD-CTD domains were expressed in BL21 ( DE3 ) E . coli . Full length IN and CCD-CTD were purified by nickel and heparin column as described previously ( Cherepanov , 2007; Kessl et al . , 2012 ) . NTD-CCD and CCD domains were purified by nickel column ( Cherepanov , 2007; Kessl et al . , 2012 ) followed by SEC using HiLoad 16/600 Superdex column ( GE healthcare ) with the elution buffer of 20 mM HEPES ( pH 7 . 5 ) , 0 . 5M NaCl , 10% glycerol , 0 . 5 mM EDTA and 2 mM DTT at 0 . 5 mL/min flow rate . All the recombinant mutant proteins were generated by introducing substitutions into Hexa-His tagged pNL4-3 wild type IN by site directed mutagenesis and were purified similar to WT IN . Recombinant WT and mutant INs were analyzed on Superdex 200 10/300 GL column ( GE Healthcare ) with running buffer containing 20 mM HEPES ( pH 7 . 5 ) , 1 M NaCl , 10% glycerol and 5 mM BME at 0 . 3 mL/min flow rate . The proteins were diluted to 10 µM with the running buffer and incubated for 1 hr at 4°C followed by centrifugation at 10 , 000 g for 10 min . Different multimeric forms of IN were identified based on the following standards: bovine thyroglobulin ( 670 , 000 Da ) , bovine gamma-globulin ( 158 , 000 Da ) , chicken ovalbumin ( 44 , 000 Da ) , horse myoglobin ( 17 , 000 Da ) and vitamin B12 ( 1 , 350 Da ) . The assays were performed on a Malvern Zetasizer Nano ZS as described previously ( Sharma et al . , 2014 ) . IN domains were analyzed at 10 µM , whereas , full length WT and mutant INs were analyzed at 200 nM in the presence of 1 µM KF116 or BI224436 . Kinetic analysis was carried out at specified time points . 5 µM of full length IN or NTD-CCD , CCD and CCD-CTD domains were incubated with increasing concentrations of KF116 and BI224436 in 20 µL reaction buffer containing 20 mM HEPES ( pH 7 . 5 ) , 1 M NaCl and 5 mM β-mercaptoethanol ( BME ) . The reactions were incubated at 4°C overnight and centrifuged at 10 , 000g for 10 min . The supernatant and the pellet fractions were analyzed by SDS-PAGE as described ( Hoyte et al . , 2017 ) . Stoichiometry for KF116 and BI224436 induced IN aggregation was determined by adding increasing concentrations of inhibitors to 12 . 65 µM full length WT IN in 20 µL reaction buffer containing 20 mM HEPES ( pH 7 . 5 ) , 1 M NaCl and 5 BME . The reactions were incubated at 4°C overnight and centrifuged at 10 , 000g for 10 min . Supernatant and pellet fractions were analyzed by SDS-PAGE as described ( Hoyte et al . , 2017 ) . Aggregated IN from the pellet fraction was quantitated by ImageJ software . The data points were plotted by piecewise linear regression in OriginLab software and the stoichiometry for KF116 and BI224436 induced aggregation of IN were calculated as described ( https://s3-us-west-2 . amazonaws . com/oww-files-public/7/7b/FAQ_Stoichiometry_V03-2 . pdf ) . The HIV-1 IN CCD ( residues 50–212 ) containing the F185H mutation was expressed and purified as described ( Sharma et al . , 2014 ) . The protein was concentrated to 8 mg/ml and crystallized using hanging-drop vapor diffusion method with a crystallization buffer consisting of 100 mM sodium cacodylate pH 6 . 5 , 100 mM ammonium sulfate , 10% ( w/v ) PEG 8000 , and 5 mM DTT . Crystallization drops were prepared using an equal volume of protein and well solution . Crystallization trays were prepared on ice at room temperature and then transferred to 4°C for storage . Crystals formed within one week and continued to grow thereafter in size . Crystal data were collected on a Rigaku Micromax-007 at 100 K . Data were integrated and scaled using HKL3000 ( Minor et al . , 2006 ) and Scalepack ( Otwinowski and Minor , 1997 ) . Phaser ( McCoy et al . , 2007 ) in the PHENIX suite ( Adams et al . , 2010 ) was used to run molecular replacement using Protein Data Bank code 4O55 as a search model ( Sharma et al . , 2014 ) . Phenix . refine ( Afonine et al . , 2012 ) was used for data refinement , and manual refinement was done in Coot ( Emsley et al . , 2010 ) . The coordinates are deposited in the Protein Data Bank under accession codes 6NUJ . The data and refinement statistics are given in Supplementary file 1 ( Table S1 ) . Inhibitor induced higher order HIV-1 IN multimer models were generated by the HADDOCK program ( Dominguez et al . , 2003; van Zundert et al . , 2016 ) using IN tetramers and dimers as building blocks . To generate tetramer and dimer structures we used the cryo-EM structure of the HIV-1 intasome ( Passos et al . , 2017 ) and the crystal structure of the full-length HIV-1 IN containing ALLINI GSK1264 ( Gupta et al . , 2016 ) , respectively . To include the ALLINI molecules in the starting structures , KF116 or BI224436 were docked into the canonical ALLINI binding sites based on the respective ligand binding mode seen in the crystal structures of KF116 ( Sharma et al . , 2014 ) and BI224436 ( Figure 6C ) bound to IN CCD dimer . Among the several docking poses generated by HADDOCK , the top poses that exhibited substantially higher buried surface area ( BSA ) compared with the rest of the poses were used for the analysis . Catalytic activities of WT and mutant INs were analyzed in the presence of LEDGF/p75 using homogeneous time-resolved fluorescence ( HTRF ) ( Slaughter et al . , 2014; Tsiang et al . , 2012 ) . HTRF offers an advantage of time resolved measurement of the energy transfer between donor and acceptor fluorophores , where a time delay between donor excitation and FRET measurement minimizes the background noise signal ( Degorce et al . , 2009 ) . Briefly , 100 nM of each IN was incubated with 100 nM LEDGF/p75 , 50 nM Cy-5 labeled donor DNA and 10 nM biotinylated target acceptor DNA in 20 mM HEPES ( pH7 . 5 ) , 1 mM DTT , 10 mM MgCl2 , 10% glycerol , 0 . 05% Brij-35 and 0 . 1 mg/ml BSA . End detection was based on europium-streptavidin antibody that binds to the biotinylated DNA and brings donor europium cryptate closer to acceptor Cy5 fluorophore in integrated DNA . This proximity results in energy transfer to yield a fluorescent signal that was recorded by PerkinElmer Life Sciences Enspire multimode plate reader . The data was plotted as percentage activity where the maximum HTRF signal from WT IN was set to 100% and the HTRF counts of different mutant INs were converted to relative percent activities . Recombinant hexa-His tagged mutant INs were analyzed for their binding to LEDGF/p75 by affinity pull-down assay as described ( Hoyte et al . , 2017 ) . Briefly , 1 µM of each IN was added to 1 µM of tag-less LEDGF/p75 in buffer containing 300 mM NaCl , 2 mM MgCl2 , 20 mM Imidazole , 0 . 2% ( v/v ) Nonidet P40 , 50 mM HEPES ( pH7 . 5 ) and 2 mM BME . The reaction mixture was incubated for 30 min at room temperature and added to the pre-equilibrated Nickel-Sepharose six fast flow beads ( GE Healthcare ) . The bound fractions were later analyzed by SDS-PAGE and quantitated by ImageJ software . Plasmids encoding Vpr-INmNeonGreen ( INmNG ) and CypA-DsRed have been described previously ( Francis et al . , 2016; Francis and Melikyan , 2018 ) . The HIV-1-based packaging vector pR9ΔEnv was from Dr . C . Aiken ( Vanderbilt University ) . TZM-bl PPIA -/- cells depleted of CypA using CRISPR/Cas9 technology have been previously described ( Francis and Melikyan , 2018 ) . Fluorescently labeled pseudoviruses were produced and characterized , as described previously ( Francis et al . , 2016 ) . Briefly , HEK293T/17 cells grown in 6-well culture plates were transfected with the following plasmids: HIV-1 pR9ΔEnv ( 2 μg ) , VSV-G ( 0 . 2 μg ) , Vpr-INmNeonGreen ( 0 . 5 μg ) and , where indicated , CypA-DsRed ( 0 . 5 μg ) using the JetPrime Transfection reagent ( VWR , Radnor , PA ) . Six hours after transfection , the medium was replaced with fresh DMEM/10% FBS without phenol red , and the cells were cultured for additional 36 hr at 37°C , 5% CO2 . Viral supernatant was collected , filtered through a 0 . 45 μm filter , aliquoted and stored at −80°C until use . Where indicated , virus production was carried out in the presence of 5 µM of ( +/- ) -KF116 , ( - ) -KF116 , BI224436 or RAL . TZM-bl PPIA-/- cells ( 5 × 105 ) were grown on 35 mm glass-bottom dishes ( MatTek Corp . ) . Tracking of single HIV-1 particles pseudotyped with VSV-G and co-labeled with INmNG/CypA-DsRed in live cells was performed , as previously described ( Francis et al . , 2016 ) . In brief , cell nuclei were stained for 10 min with 2 μg/ml Hoechst-33342 followed by spinoculation with pseudoviruses ( 10 pg of p24 per 5 × 105 TZM-bl PPIA-/- cells , MOI 0 . 008 ) at 1500 × g , 4°C for 30 min . The cells were washed twice , and virus entry was initiated on a temperature- and CO2-controlled microscope stage by adding pre-warmed live-cell imaging buffer ( Life Technologies , Invitrogen ) supplemented with 10% FBS . 3D time-lapse imaging was carried out with a Zeiss LSM880 AiryScan confocal microscope using a C-Apo 63x/1 . 4NA oil-immersion objective . A suitable field of view was selected , and full cell volume was imaged every 10 s using confocal or AiryScan fast mode by acquiring 11–17 Z-stacks spaced by 0 . 5 μm , using a low power of 405 , 488 and 561 nm lasers for Hoechst-33342 , NeonGreen and DsRed , respectively . A DefiniteFocus module ( Carl Zeiss ) was utilized to correct for axial drift during image acquisition . Fixed time-point imaging was performed using AiryScan Optimal mode at room temperature using sequential imaging of multiple fields of view , each typically containing ~7 cells . After acquisition , the images was processed by applying auto-threshold parameters . Acquired image series were converted to maximum intensity projections and analyzed using the ICY image analysis software ( icy . bioimageanalysis . org ) . Single particle tracking was performed using INmNG as a reference channel for DMSO treated control samples and CypA-DsRed as reference for ALLINI treated infections to demonstrate uncoating and INmNG signal disappearance , respectively . Single virus fluorescence and colocalization analysis was performed using the ROI detector and Colocalization studio plugin from ICY software . Single particle intensity analyses was performed after background subtraction and by normalizing to the initial fluorescence intensity to 100% . The N155H/K156N and N155H/K156N/K211R/E212T substitutions were cloned into the IN coding region of pNL4-3 . Luc . E-R+ ( Connor et al . , 1995 ) using PCR-site directed mutagenesis ( Agilent ) and verified by dideoxy sequencing . Additional IN substitutions at K14A , Y15A , T210 +Pro and N222K were similarly introduced into pNL4 . 3 ( Adachi et al . , 1986 ) . HeLa and HEK293T cells were purchased from ATCC , while TZM-bl and MT-4 cells were obtained from NIH AIDS reagent program . The identity of all cell lines have been authenticated by STR profiling done at ATCC . Cell lines were tested at periodic intervals for Mycoplasma contamination by using Mycoscope Mycoplasm PCR detection kit and there has been no evidence of Mycoplasma . HEK293T cells were cultured in Dulbecco’s modified eagle medium ( Gibco ) , 10% FBS ( Sigma-Aldrich ) and 1% Penicillin Streptomycin ( Gibco ) at 37°C and 5% CO2 . MT-4 cells were cultured in Roswell Park Memorial Institute ( RPMI ) 1640 medium ( Gibco ) supplemented with 10% FBS and 1% Penicillin Streptomycin . Virus stocks were produced as previously described ( Sharma et al . , 2014 ) . For viral infectivity experiments , virions produced from HIV-1NL4-3 , HIV-1NL4-3 IN ( K14A ) , HIV-1NL4-3 IN ( Y15A ) , HIV-1NL4-3 IN ( T210 +Pro ) and HIV-1NL4-3 IN ( N222K ) molecular clones were generated by transfecting HEK293T cells as previously described ( Feng et al . , 2013 ) . Twenty-four hours post-transfection , the culture supernatant was replaced with fresh complete medium after washing once with complete medium . Forty-eight hours post-transfection , the virus containing supernatant was collected , and filtered through 0 . 45 µm filter . TZM-bl cells ( 2 × 105 ) were then infected with HIV-1 virions equivalent to 100 ng of virus determined by HIV-1 Gag p24 ELISA ( Zeptometrix ) following manufacturer’s protocol in the presence of 8 µg/ml Polybrene ( Sigma ) . Two hours post-infection the culture supernatant was removed , and the cells were washed and replaced with fresh medium . Viral infectivity was determined using the procedure previously described ( Feng et al . , 2013 ) . EC50 values of ( + ) and ( - ) enantiomers of KF116 against replication competent HIV-1NL4-3 were assayed in MT-4 cells . 50 µl of 2x test concentration of the diluted compounds were added to RPMI 1640 cell culture medium with 10% FBS added to each well of a 96-well plate ( nine concentrations ) in triplicate . MT-4 cells were infected with HIV-1NL4-3 at a multiplicity of infection ( MOI ) of 0 . 01 for 3 hr . Fifty microliters of infected cell suspension in culture medium with 10% FBS ( ~1 . 5×104 cells ) were then added to each well containing 50 µl of diluted compound . The plates were then incubated at 37°C 5% CO2 for 5 days . After 5 days of incubation , 100 µl of CellTiter-Glo reagent ( catalog no . G7571; Promega Biosciences , Inc ) was added to each well containing MT-4 cells . Cell lysis was carried out by incubation at room temperature for 10 min , and chemiluminescence was read . The EC50 values of ( - ) -KF116 and BI224436 against WT , N155H/K156N and N155H/K156N/K211R/E212T viruses were determined in single replication cycle as described previously ( Sharma et al . , 2014 ) . Briefly , HEK293T cells were transfected with WT and mutant pNL4-3 . Luc . Env- and pCG-VSV-G ( Brown et al . , 1999 ) plasmids to produce respective viruses in the absence and presence of indicated concentrations of ALLINIs . The virus supernatants were collected 24 hr after drug treatment and p24 concentrations were determined by HIV-1 Gag p24 ELISA ( ZeptoMetrix ) following manufacturer's protocol . The target HeLa cells were then infected with HIV-1 virions equivalent to 10 ng of HIV-1 p24 , in the presence of indicated concentrations of drugs . Two hours post-infection the culture supernatant was removed , washed once with complete medium , and then fresh complete medium was added with the inhibitor concentration maintained . The cells were cultured for 48 hr and the cell extracts were prepared using reporter lysis buffer ( Promega ) . Luciferase activity was determined using a commercially available kit ( Promega ) . The fitted dose-response curves were generated to calculate EC50 using OriginLab software . The metabolic stability of ( +/- ) -KF116 , ( - ) -KF116 and BI224436 was evaluated in rat and human liver microsomal incubations by probing in vitro Cytochrome ( CYP ) P450 activity in the presence of co-factor NADPH . Stock DMSO solutions were prepared and LC/MS-MS methods were developed . Verapamil , domperidone and chlorpromazine were used as controls . Liver microsomal incubations ( male Sprague-Dawley rat , p=500 , 0 . 2 mg protein/mL; and human mixed gender , p=50 , 1 . 0 mg protein/mL ) were performed in the presence of NADPH ( 1 . 0 mM ) , taking samples as a function of time and analyzing drug concentration ( Wempe and Anderson , 2011; Wempe et al . , 2012a; Wempe et al . , 2012b ) . In this assay , the loss of parent drug is monitored over-time to produce in vitro half-life values . These half-life results were used to calculate intrinsic clearance values ( CLint , reported in units of µL/min/mg protein ) , as previously described using the following equation: ( 0 . 693/in vitro T1/2 ) ● ( µL incubation/mg microsomal protein ) ( Obach , 1999 ) .	HIV-1 inserts its genetic code into human genomes , turning healthy cells into virus factories . To do this , the virus uses an enzyme called integrase . Front-line treatments against HIV-1 called “integrase strand-transfer inhibitors” stop this enzyme from working . These inhibitors have helped to revolutionize the treatment of HIV/AIDS by protecting the cells from new infections . But , the emergence of drug resistance remains a serious problem . As the virus evolves , it changes the shape of its integrase protein , substantially reducing the effectiveness of the current therapies . One way to overcome this problem is to develop other therapies that can kill the drug resistant viruses by targeting different parts of the integrase protein . It should be much harder for the virus to evolve the right combination of changes to escape two or more treatments at once . A promising class of new compounds are “allosteric integrase inhibitors” . These chemical compounds target a part of the integrase enzyme that the other treatments do not yet reach . Rather than stopping the integrase enzyme from inserting the viral code into the human genome , the new inhibitors make integrase proteins clump together and prevent the formation of infectious viruses . At the moment , these compounds are still experimental . Before they are ready for use in people , researchers need to better understand how they work , and there are several open questions to answer . Integrase proteins work in groups of four and it is not clear how the new compounds make the integrases form large clumps , or what this does to the virus . Understanding this should allow scientists to develop improved versions of the drugs . To answer these questions , Koneru et al . first examined two of the new compounds . A combination of molecular analysis and computer modelling revealed how they work . The compounds link many separate groups of four integrases with each other to form larger and larger clumps , essentially a snowball effect . Live images of infected cells showed that the clumps of integrase get stuck outside of the virus’s protective casing . This leaves them exposed , allowing the cell to destroy the integrase enzymes . Koneru et al . also made a new compound , called ( - ) -KF116 . Not only was this compound able to tackle normal HIV-1 , it could block viruses resistant to the other type of integrase treatment . In fact , in laboratory tests , it was 10 times more powerful against these resistant viruses . Together , these findings help to explain how allosteric integrase inhibitors work , taking scientists a step closer to bringing them into the clinic . In the future , new versions of the compounds , like ( - ) -KF116 , could help to tackle drug resistance in HIV-1 .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "microbiology", "and", "infectious", "disease" ]	2019	HIV-1 integrase tetramers are the antiviral target of pyridine-based allosteric integrase inhibitors
Expression of inflammatory genes is determined in part by post-transcriptional regulation of mRNA metabolism but how stimulus- and transcript-dependent nuclear export influence is poorly understood . Here , we report a novel pathway in which LPS/TLR4 engagement promotes nuclear localization of IRAK2 to facilitate nuclear export of a specific subset of inflammation-related mRNAs for translation in murine macrophages . IRAK2 kinase activity is required for LPS-induced RanBP2-mediated IRAK2 sumoylation and subsequent nuclear translocation . Array analysis showed that an SRSF1-binding motif is enriched in mRNAs dependent on IRAK2 for nuclear export . Nuclear IRAK2 phosphorylates SRSF1 to reduce its binding to target mRNAs , which promotes the RNA binding of the nuclear export adaptor ALYREF and nuclear export receptor Nxf1 loading for the export of the mRNAs . In summary , LPS activates a nuclear function of IRAK2 that facilitates the assembly of nuclear export machinery to export selected inflammatory mRNAs to the cytoplasm for translation . Eukaryotic cells produce mRNA in the nucleus through a series of events including 5’ capping , 3’-end processing and splicing , which are coupled with transcription . Once these processes are complete , mRNA is exported from the nucleus to the cytoplasm where it can be translated to generate proteins . Nxf1 ( also known as TAP ) is the key mRNA export receptor , which only binds processed mRNA . Upon completion of mRNA processing , Nxf1 is recruited to the mRNA along with the TREX complex to promote the nuclear export the mRNA . The nuclear export adaptor ALYREF , a subunit of the TREX complex , plays a critical role in integrating the signals provided by mRNA processing and in triggering nuclear export receptor Nxf1 loading for the export of the target mRNAs ( Hung et al . , 2010; Viphakone et al . , 2012; Zhou et al . , 2000 ) . ALYREF-Nxf1 interaction provides a mark on mRNA to signify that nuclear RNA-processing events are complete , and the processed mRNA is ready for export to the cytoplasm . Several other proteins have also been implicated as nuclear export adaptors , including the shuttling SR ( serine- and arginine-rich ) proteins 9G8 , SRp20 and SRSF1 ( Huang et al . , 2003 ) . However , it remains unclear how these nuclear export adaptors interact with different classes of mRNAs to achieve sequence-specific , stimulus-dependent export of target transcripts . Toll-like receptor ( TLR ) signaling regulates the expression of chemokines and cytokines at both transcriptional and posttranscriptional levels ( Anderson , 2008 ) . Cytokine and chemokine mRNAs have short half-lives because of conserved cis-elements in their three prime untranslated regions ( 3’ UTRs ) . Much effort has been devoted to understand how the conserved cis-elements within the 3’ UTR can be recognized by RNA-binding proteins ( including SRSF1 ) that function to mediate mRNA decay ( Fenger-Grøn et al . , 2005; Lykke-Andersen and Wagner , 2005; Mayr , 2016; Sun et al . , 2011 ) . Nevertheless , whether and how cytokine and chemokine mRNAs are regulated during nuclear export in response to TLR stimulation is unknown . TLRs transduce signals through the adaptor molecule MyD88 and IL-1R-associated kinase ( IRAK ) family members , including IRAK1 , IRAK2 and IRAK4 ( Kawai and Akira , 2011 ) . IRAK4 is the upstream kinase of IRAK1 and IRAK2 ( Kim et al . , 2007; Lin et al . , 2010 ) . IRAK1 is necessary for TAK1-dependent NFκB activation for the transcription of chemokines and cytokines ( Cui et al . , 2012; Yao et al . , 2007 ) . IRAK2 is an atypical kinase that mediates posttranscriptional regulation of inflammatory transcripts . Deletion of IRAK2 impairs the production of inflammatory cytokines and chemokines in macrophages in response to TLR stimulation ( Wan et al . , 2009; Yin et al . , 2011 ) without affecting transcription of inflammatory genes . However , the precise mechanism by which IRAK2 controls the posttranscriptional regulation of inflammatory cytokine and chemokine production is an evolving area of investigation . Here , we report that LPS induces IRAK2 nuclear localization to facilitate nuclear export of a subset of inflammatory mRNAs ( including Cxcl1 , Tnf and Cxcl2 ) to the cytosol for protein translation . IRAK2 kinase activity is required for LPS-induced RanBP2-mediated IRAK2 sumoylation , which facilitates IRAK2 nuclear translocation . By array analysis , we identified an SRSF1-binding motif enriched selectively in mRNA targets dependent on IRAK2 for nuclear export . IRAK2 phosphorylates SRSF1 and thereby reduces SRSF1 binding to the target mRNAs . On the other hand , SRSF1 knockdown resulted in increased nuclear export of the target mRNAs . Importantly , LPS induced the interaction of IRAK2 with the nuclear export factors ALYREF and Nxf1 recruiting them to the target mRNAs . While the depletion of SRSF1 allowed nuclear export adaptor ALYREF binding to the target mRNAs , LPS/IRAK2-induced recruitment of nuclear export receptor Nxf1 to the target transcripts was abolished by the knockdown of ALYREF . Thus , SRSF1-mediated nuclear sequestration of target mRNAs might be achieved by blocking the binding of ALYREF and Nxf1 to the mRNAs . Taken together , our results suggest that while SRSF1 binding renders target mRNAs sensitive for nuclear export , LPS promotes a nuclear function of IRAK2 that mediates the removal of SRSF1 and facilitates the assembly of nuclear export machinery to export the inflammatory mRNAs to the cytosol for protein translation . We and others have previously reported that IRAK2 plays a critical role in the production of pro-inflammatory cytokines and chemokines in response to TLR stimulation . However , the detailed molecular mechanism is not completely understood . IRAK2 is an atypical kinase due to the amino acid substitution of key catalytic residues ( Asp -> Asn333; Asp ->His351 Figure 1A ) , which is supported by the demonstration of catalytic activities of atypical kinases such as KSR2 ( Brennan et al . , 2011 ) and CASK ( Mukherjee et al . , 2008 ) . The recombinant wild-type IRAK2 as well as IRAK2 mutants ( ATP-binding site mutant KK235AA and catalytic site mutant H351A ) were purified by using a bacterial expression system . While the purified recombinant wild-type IRAK2 was able to autophosphorylate and phosphorylate myelin basic protein ( MBP ) , the IRAK2 mutants ( KK235AA and H351A ) displayed minimum activity ( Figure 1B ) . Wild-type IRAK2 , but not IRAK2 kinase-inactive mutants ( ATP-binding site mutant KK235AA; catalytic site mutants H351A and N333A ) , restored LPS-mediated induction of inflammatory cytokines and chemokines in IRAK2-deficient bone marrow-derived macrophages ( Figure 1C and E ) , suggesting that the kinase activity of IRAK2 is necessary for ligand-induced pro-inflammatory gene expression . To further investigate the role of IRAK2 in LPS-induced inflammatory gene expression , we recently generated IRAK2 kinase-inactive knockin mice in which the ATP-binding site was mutated ( KK235AA , Figure 1—figure supplement 2 ) . LPS-induced production of inflammatory cytokines and chemokines was greatly reduced in the macrophages from IRAK2 kinase-inactive knockin mice compared to that of the wild-type control mice ( Figure 1D and F ) . Notably , IRAK2 is modified in bone marrow-derived macrophages in response to LPS stimulation ( Figure 1E ) . Wild-type IRAK2 , but not IRAK2 kinase-inactive mutants ( ATP-binding site mutant KK235AA; catalytic site mutants H351A and N333A ) , restored IRAK2 modification in IRAK2-deficient bone marrow-derived macrophages ( Figure 1E ) . Interestingly , we found that the modified form of IRAK2 was translocated into the nucleus in a ligand-dependent manner ( Figure 1F ) . Immunofluorescent staining showed that IRAK2 ATP-binding site mutant KK235AA and catalytic site mutants ( H351A and N333A ) , failed to translocate into the nucleus , suggesting that IRAK2 kinase activity is important for its nuclear translocation ( Figure 1G ) . In support of this , LPS-induced IRAK2 nuclear translocation was abolished in the macrophages from IRAK2 kinase-inactive knockin mice ( Figure 1F ) . Interestingly , not only LPS stimulation leads to IRAK2 modification and nuclear localization , endogenous TLR4 ligand oxidized low-density lipoprotein ( ox-LDL ) , TLR7 ligand R848 and TLR9 ligand CpGB stimulation can also induce IRAK2 modification and subsequent nuclear translocation ( Figure 1—figure supplement 4 ) . Nuclear localization signal ( NLS ) is a short peptide motif that mediates the nuclear import of proteins by binding to importin complex . We identified a putative NLS in the IRAK2 kinase domain ( Aa 357–366: HPDNKKTKYT ) , which was mutated by substituting the three lysine residues ( K361 , K362 and K364 ) with alanines . Immunofluorescent staining showed that the IRAK2 NLS mutant indeed failed to translocate into the nucleus ( Figure 1G and Figure 1—figure supplement 3 ) . IRAK2 NLS mutant also failed to restore LPS-induced inflammatory cytokine and chemokine production in IRAK2-deficient macrophages ( Figure 1C and E ) , indicating that IRAK2 nuclear translocation is required for IRAK2-mediated inflammatory gene expression . While we showed that modified IRAK2 was translocated into the nucleus ( Figure 1F ) , IRAK2 NLS mutant failed to be modified ( Figure 1E ) , implicating a critical link between IRAK2 nuclear translocation and modification . We then investigated how LPS induced IRAK2 nuclear translocation . Mass spectrometry ( MS ) analysis showed that IRAK2 interacts with importin-β ( Figure 2A and Figure 2—figure supplement 1 ) , and co-immunoprecipitation experiment implicates the formation of IRAK2-RanBP2 complex ( Figure 2A ) . While importin-β is one of the transport receptor mediating translocation of molecules into the cell nucleus , RanBP2 is a major nucleoporin that extends cytoplasmic filaments from the nuclear pore complex and contains phenylalanine–glycine repeats that bind nuclear transport receptors of the importin beta family ( Wu et al . , 1995; Yokoyama et al . , 1995 ) . Interestingly , we found that IRAK2 ATP-binding site mutant KK235AA and catalytic site mutants H351A/N333A retained the interaction with importin-β , whereas NLS mutant failed to form a complex with importin-β ( Figure 2A ) . These results suggest that the IRAK2 kinase activity is not required for IRAK2’s NLS to engage the importin complex . On the other hand , IRAK2 KK235AA and H351A/N333A kinase-inactive mutants and as well as NLS mutant lost interaction with RanBP2 ( Figure 2A ) . It has been reported that LPS stimulation induces the phosphorylation of IRAK2 at S136 and T140 ( Weintz et al . , 2010 ) . We found that mutation at these two phosphorylation sites ( IRAK2 S136A/T140A double mutant ) abolished LPS-induced interaction of IRAK2 with RanBP2 and its nuclear localization ( Figure 2—figure supplement 2 ) , although the IRAK2 S136A/T140A double mutant still retained the interaction with importin-β . These findings suggest that the activation of IRAK2 may result in auto-phosphorylation at S136A and T140A , which in turn mediates the interaction with RanBP2 . In support of this , it is indeed the modified form of IRAK2 was specifically co-immunoprecipitated by RanBP2 ( Figure 2—figure supplement 3 ) . Furthermore , the phos-tag gel electrophoresis confirmed that the phosphorylated IRAK2 preferentially binds to RanBp2 ( Figure 2—figure supplement 3 ) . Notably , the C-terminal of RanBP2 carries an active SUMO E3 ligase domain , which was shown to modulate cytoplasmic and nuclear transport of macromolecules ( Pichler et al . , 2002 ) . The modified IRAK2 band induced by LPS stimulation was about 20 kDa bigger than the unmodified IRAK2 . Sumoylation typically gives rise to a 20 KD size shift in the SDS-PAGE gel , even though it is only 11 KD in molecular mass ( Werner et al . , 2012 ) . Importantly , IRAK2 is characterized as a sumoylated protein in recent proteomics study ( Lamoliatte et al . , 2014 ) . Thus , we hypothesized that IRAK2 is sumoylated upon LPS stimulation via its interaction with RanBP2 , which might play an important role in IRAK2 nuclear translocation . LPS stimulation indeed induced the interaction between IRAK2 and RanBP2 ( Figure 2B ) . We found LPS-induced IRAK2-RanBP2 interaction was abolished in the macrophages from IRAK2 kinase-inactive knockin mice ( Figure 2B ) , confirming that the kinase activity of IRAK2 is required for its interaction with RanBP2 . Furthermore , SUMO1 was detected in IRAK2 immunoprecipitates under denaturing condition , these results indicate that IRAK2 is conjugated by SUMO1 in response to LPS stimulation ( Figure 2C ) . Consistent with the fact that IRAK2 kinase-inactive mutant failed to interact with RanBP2 , LPS-induced IRAK2 sumoylation was abolished in the macrophages from IRAK2 kinase-inactive knockin mice ( Figure 2C ) . To identify the critical lysine residues required for sumoylation , the lysine residues in the sumoylation consensus sequence ( ψKxE ) were mutated to arginines . IRAK2 K123/182/592R triple mutant lost sumoylation , and failed to translocate into the nucleus ( Figure 2D–E and Figure 2—figure supplement 4 ) , suggesting that IRAK2 sumoylation plays an essential role for IRAK2 nuclear translocation . To determine whether RanBP2 is the E3 ligase for IRAK2 , we performed in vitro sumoylation assay using purified recombinant Aos1/Uba2 ( E1 ) , UBC9 ( E2 ) , RanBP2 ( E3 ) , and SUMO1 . By in vitro sumoylation assay , we found that wild-type IRAK2 was sumoylated , whereas the K123/182/592R triple mutant failed to be modified ( Figure 2F and Figure 2—figure supplement 5 ) . The modified form of IRAK2 disappeared after treatment with SUMO-specific isopeptidase Ulp1 , confirming the observed modification of IRAK2 was indeed due to sumoylation ( Figure 2G ) . Interestingly , the interaction between IRAK2 and RanBP2/SUMO1 was also sensitive to Ulp1 treatment ( Figure 2—figure supplement 6 ) . To further assess the importance of RanBP2 in LPS-induced IRAK2 modification , RanBP2 was knocked down in macrophages . LPS-induced IRAK2 sumoylation was abolished in the absence of RanBP2 , confirming that RanBP2 is the E3 ligase for IRAK2 sumoylation ( Figure 2H ) . Co-immunoprecipitation experiment showed that the sumoylated form of IRAK2 preferentially binds to RanBP2 ( Figure 2—figure supplement 3 ) . Furthermore , LPS-induced IRAK2 nuclear translocation was abolished in RanBP2 knock-down cells , supporting the critical role of IRAK2 sumoylation for its nuclear translocation ( Figure 2I and Figure 2—figure supplement 3 ) . We next investigated the functional importance of IRAK2 in the nucleus . Notably , IRAK2 is required for the posttranscriptional regulation of inflammatory genes in response to LPS stimulation ( Figure 3—figure supplement 1; [Wan et al . , 2009] ) . The subdivision of eukaryotic cells into nuclear and cytoplasmic compartments enables spatial separation of transcription and translation . We hypothesized that LPS-induced IRAK2 nuclear localization may facilitate nuclear export of the inflammatory mRNAs to cytosol for translation after their transcription in the nucleus . To test this hypothesis , we isolated nuclear and cytoplasmic RNA from untreated and LPS-treated wild-type and IRAK2-deficient macrophages , followed by array analysis . We identified a group of mRNAs ( including Cxcl1 , Tnf and Cxcl2 ) accumulated in the nucleus in the absence of IRAK2 , suggesting that IRAK2 is required for the exportation of these mRNAs ( Figure 3A and Figure 3—figure supplement 2 ) . The same mRNAs ( including Cxcl1 , Tnf and Cxcl2 ) were also accumulated in the nucleus of IRAK2 kinase inactive macrophages , suggesting that the kinase activity of IRAK2 is required for the exportation of these mRNAs ( Figure 3B ) . To identify the potential role of RNA-binding proteins in IRAK2-mediated regulation of mRNA nuclear export , we applied RNA-READ motif scanner , a regression-based framework which searches for previously defined RNA cis-motifs ( Ray et al . , 2013 ) whose involvement in the regression significantly improves the fitting to the data than ones based on the background distribution of the sequence alone ( Materials and methods ) . As a result , we found that the binding sites of SRSF1 ( KGRWGSM , K: G/U; R:G/A; W:A/U; S:G/C; M:A/C ) are significantly enriched in the 3’UTRs of the positive set ( LPS-induced mRNAs that were accumulated in the nucleus of IRAK2-deficient macrophages compared to that in wild-type macrophages ) versus the negative set ( LPS-induced mRNAs whose cytosol/nuclear distribution in the macrophages was not affected by IRAK2 deficiency ) , and this motif significantly improves the fitting to the data than ones based on the background distribution ( Likelihood ratio test p-value<0 . 0004 ) ( Figure 3C ) . SRSF1 , also known as SF2/ASF is a serine- and arginine-rich protein , which plays important roles in mRNA metabolism . We hypothesized that the impact of IRAK2 on the nuclear exportation of the mRNAs ( enriched with SRSF1-binding motifs ) might be through IRAK2-mediated modulation on SRSF1 . Interestingly , we detected SRSF1-IRAK2 interaction in the nucleus by duo-link assay ( Figure 3D and Figure 3—figure supplement 3 ) , which was abolished for IRAK2 kinase-inactive and sumoylation mutants . Furthermore , IRAK2 was able to phosphorylate SRSF1 in the in vitro kinase assay ( Figure 3E ) . In support of this , we found that LPS indeed induced SRSF1 serine/threonine phosphorylation , which was diminished in IRAK2-kinase-inactive nucleus ( Figure 3F ) . Previous studies have shown that SRSF1 binds RNA in its hypo-phosphorylated form ( Sanford et al . , 2005; Xiao and Manley , 1997 ) , suggesting that IRAK2-mediated SRSF1 phosphorylation probably drives the dissociation of SRSF1 from the mRNA targets , promoting their nuclear export . By RNA immunoprecipitation , we indeed found that SRSF1-bound transcripts ( Cxcl1 and Tnf ) were increased in IRAK2-deficient and IRAK2-kinase-inactive macrophages compared to that in wild-type cells ( Figure 3G and Figure 3—figure supplement 4 ) . Taken together , these results suggest that IRAK2 may directly phosphorylate SRSF1 in the nucleus in response to LPS stimulation , which facilitates the dissociation of SRSF1 from the LPS-induced transcripts , and the subsequent nuclear export of these mRNAs . Notably , SRSF1 also plays a critical role in regulating genes that are constitutively expressed , such as Ctnnb1 ( β-catenin ) ( Fu et al . , 2013 ) . Interestingly , SRSF1-bound transcript Ctnnb1 was not affected by either LPS stimulation or IRAK2 deficiency ( Figure 3G ) , indicating that IRAK2 may only regulate a specific subset of SRSF1-targeted mRNAs . The next question is how IRAK2 mediates nuclear export of the LPS-induced mRNAs bound by SRSF1 . Nuclear export of the mature mRNA is an active process that is integrated with many of the nuclear processing steps , facilitated by RNA-binding proteins and transport factors ( Wickramasinghe and Laskey , 2015 ) . Importantly , ALYREF nuclear export factor was actually among the proteins based on MS analysis of IRAK2 immunoprecipitation ( Figure 2—figure supplement 1 ) . By coimmunoprecipitation , we found that LPS stimulation induced the interaction of IRAK2 with ALYREF in the nucleus ( Figure 4A ) . Duo-link assay showed that ALYREF interacts with wild-type IRAK2 , but not kinase-inactive and sumoylation mutants ( Figure 4B and Figure 4—figure supplement 1 ) . These results indicate that the kinase activity of IRAK2 and subsequent sumoylation are required for IRAK2’s interaction with ALYREF in the nucleus , which is probably due to the fact that the kinase-inactive and sumoylation mutants failed to translocate into the nucleus ( Figure 1G ) . Since LPS-induced the interaction of IRAK2 with SRSF1 ( Figures 3D–F and 4A ) , we wondered whether IRAK2-SRSF1 and IRAK2-ALYREF belong to one complex . Interestingly , ALYREF was absent in the SRSF1 immunoprecipitates , while SRSF1 was also not detected in ALYREF immunoprecipitates ( Figure 4A ) . On the other hand , IRAK2 was detected in either SRSF1 or ALYREF immunoprecipitates in response to LPS stimulation ( Figure 4A ) . These results suggest that LPS-induced IRAK2’s interaction with SRSF1 and ALYREF are mutually exclusive . The fact that IRAK2 was able to phosphorylate SRSF1 ( Figure 3F ) led us hypothesize that IRAK2-mediated SRSF1 phosphorylation might drive SRSF1 off the mRNA targets , allowing ALYREF’s binding to the target mRNAs . In support of this , RIP analysis showed that LPS stimulation induced the binding of IRAK2 and ALYREF to the mRNAs of Cxcl1 , Cxcl2 and Tnf , and ALYREF’s binding to the mRNAs was substantially reduced in IRAK2-deficient and IRAK2-kinase-inactive macrophages ( Figure 4C–D ) . Interestingly , ALYREF bound Cxcl1 and Tnf transcripts were increased in SRSF1 knock-down cells , even in IRAK2 KO macrophages , indicating that ALYREF’s binding to Cxcl1 and Tnf is no longer IRAK2-dependent in the absence of SRSF1 ( Figure 4E and G and Figure 4—figure supplement 2 ) . Similar to SRSF1 , ALYREF also binds to constitutively expressed Ctnnb1 mRNA . Importantly , ALYREF-bound Ctnnb1 mRNA was not affected by LPS stimulation , IRAK2 deficiency or SRSF1 knockdown ( Figure 4E ) . These results confirmed that the role of IRAK2 for ALYREF’s binding to the target mRNAs is probably to remove SRSF1 from the targets . Consistently , whereas IRAK2 deficiency resulted in nuclear accumulation of Cxcl1 and Tnf mRNA , Cxcl1 and Tnf mRNA was efficiently exported to the cytosol in SRSF1 knockdown cells in the absence of IRAK2 ( Figure 4F–G ) . The metazoan TREX complex is recruited to mRNA during nuclear RNA processing and functions in exporting mRNA to the cytoplasm ( Cheng et al . , 2006 ) . Recent studies suggest that the recruitment of ALYREF to TREX allows for mRNA export by driving nuclear export receptor Nxf1 ( also known as TAP ) into a conformation capable of binding mRNA ( Viphakone et al . , 2012 ) . In addition to the LPS-induced interaction of IRAK2 with ALYREF , we found that LPS stimulation also induced the interaction of IRAK2 with Nxf1 ( Figure 5A ) . IRAK2 failed to interact with Nxf1 in the absence of ALYREF , suggesting that IRAK2-ALYREF interaction is required for ALYREF to facilitate the assembly of nuclear export complex to the target mRNAs , including the recruitment of Nxf1 ( Figure 5A ) . In support of this , LPS stimulation induced Nxf1’s binding to the mRNAs in an ALYREF- and IRAK2-dependent manner ( Figure 5B–D ) . To further assess the importance of ALYREF in LPS-induced IRAK2-dependent mRNA nuclear export , we measured nuclear and cytoplasmic mRNA levels in ALYREF knockdown cells . Consistent with the RIP analysis , we found that LPS-induced mRNAs of Cxcl1 , Cxcl2 and Tnf accumulated in the nucleus in the absence of ALYREF ( Figure 5E ) , indicating the importance of this nuclear export factor in IRAK2-mediated nuclear export of the inflammatory mRNAs . By enzyme-linked immunosorbent assay ( ELISA ) analysis , we showed that LPS-induced secretion of CXCL1 and TNF was dramatically reduced in ALYREF knockdown cells ( Figure 5F ) . Since the LPS-induced mRNAs that are dependent on IRAK2 for nuclear export were enriched with SRSF1-binding motif in the 3’UTRs ( Figure 3C ) , we decided to directly test the importance of the SRSF1-binding sites in the target mRNAs for their nuclear export . Using Cxcl1 3’UTR ( nt 781–901 , which contains stimulus-sensitive motifs ( Datta et al . , 2010; Hartupee et al . , 2007 ) , we validated the SRSF1 binding to this transcript by RNA electrophoretic mobility shift assay ( REMSA ) ( Figure 6A ) . We identified several putative SRSF1-binding motifs in this region of Cxcl1 3’UTR . Deletion analysis and REMSA showed the importance of nt 800–855 in Cxcl1 3’UTR for SRSF1 binding . We generated luciferase reporter constructs by cloning the 3’UTR of Cxcl1 with or without deletion of the SRSF1-binding region downstream of Luciferase reporter ( under the control of CMV promoter ) . These reporter constructs were introduced into IRAK2-deficient cells ( IRAK deficient 293-IL1R cells; [Yao et al . , 2007] ) with or without co-transfection with IRAK2 . While SRSF1 RIP precipitated the luciferase reporter mRNA containing the 3’UTR of Cxcl1 ( nt 721–940 ) , IRAK2 expression reduced SRSF1 binding to the target mRNA ( Figure 6B ) . However , SRSF1 RIP failed to precipitate the luciferase mRNAs with the mutated 3’UTR of Cxcl1 ( Δ790–840 and Δ829–835 ) ( Figure 6B ) . We then assessed the SRSF1 site-mutant constructs by luciferase reporter assay . For the cells transfected with wild-type Cxcl1 luciferase construct ( with wild-type 3’ UTR of Cxcl1 721–940 ) , IRAK2 strongly promoted the luciferase activity ( Figure 6C ) . Consistently , the cytosolic luciferase reporter mRNAs ( with wild-type 3’ UTR of Cxcl1 721–940 ) were increased in the IRAK2-transfected cells compared to the cells without IRAK2 ( Figure 6D ) . IRAK2 had much reduced ability to promote luciferase activity in cells transfected with mutant Cxcl1 luciferase constructs ( Cxcl1Δ790–840 and Cxcl1Δ829–835 ) ( Figure 6C ) . Removal of the functional SRSF1-binding site trapped the luciferase reporter mRNAs ( Cxcl1Δ790–840 and Cxcl1Δ829–835 ) in the nucleus and IRAK2 can no longer promote nuclear export of these mutant reporter mRNAs ( Cxcl1Δ790–840 and Cxcl1Δ829–835 ) ( Figure 6D ) . These results suggest that while SRSF1 binding may sequester the luciferase reporter mRNAs in the nucleus , the SRSF1-binding sequence is actually required for nuclear export promoted by IRAK2 . Similar results were obtained in wild-type and IRAK2-deficient macrophages . LPS stimulation induced luciferase activity from the cells transfected with constructs containing wild-type 3’UTR of Cxcl1 ( nt 721–940 ) , which was abolished in IRAK2-deficient macrophages ( Figure 6E ) . However , LPS-induced luciferase activity was substantially reduced in wild-type , IRAK2-deficient or IRAK2-kinase inactive macrophages transfected with mutant luciferase constructs ( Δ790–840 and Δ829–835 ) ( Figure 6E ) . These results confirm that IRAK2-mediated nuclear export of the target mRNAs is dependent on the SRSF1-binding sites in the target mRNA . Consistent with the fact that IRAK2 is required for LPS-induced ALYREF’s binding to the target mRNAs , IRAK2 promoted ALYREF RIP of wild-type luciferase reporter mRNAs ( containing wild-type 3’UTR of Cxcl1 nt 721–940 ) ( Figure 6F ) . Interestingly , ALYREF reduced the binding to the mutant luciferase reporter mRNAs ( Δ790–840 and Δ829–835 ) even in the presence of IRAK2 ( Figure 6F ) . These results suggest that the SRSF1-binding site seems to be required for ALYREF’s binding to the target mRNAs . In support of this , ALYREF can directly bind to the probe that contains the SRSF1-binding site in the in vitro RNA-binding assay ( Cxcl1 nt 800–855 , Figure 6G and Figure 6—figure supplement 1 ) . Taken together , our results implicate that the SRSF1-binding site renders the LPS-IRAK2 sensitivity of the target mRNA for the nuclear export possibly via the IRAK2-mediated removal of SRSF1 and recruitment of nuclear export factors ALYREF to the target mRNAs . Mirroring the IRAK2 knockout cells , the LPS-induced inflammatory mRNAs were accumulated in the nucleus in IRAK2 kinase-inactive knockin macrophages compared to that in wild-type cells ( Figure 3B ) . Consistent with the defect in promoting nuclear export of target mRNAs , LPS-induced ALYREF RIP of inflammatory mRNAs was abolished in IRAK2 kinase-inactive knockin macrophages ( Figure 4F ) . Finally , ELISA analysis validated the importance of IRAK2 kinase activity for the production of LPS-induced CXCL1 , CXCL2 and TNF ( Figure 1D ) . Since IRAK2 kinase-inactive knock-in macrophages mirrored IRAK2 knockout cells in all the ex vivo experiments in response to LPS stimulation , we then further investigated the impact of this IRAK2 mutation on LPS-response in vivo . We found that IRAK2 kinase-inactive knockin mice showed substantially increased survival after LPS-induced septic shock compared to that of the wild-type mice ( Figure 7A ) ( Mao et al . , 2013 ) . Likewise , LPS-induced serum CXCL1 and TNF levels were much reduced in IRAK2 kinase-inactive knockin mice compared to that of the wild-type mice ( Figure 7B ) . While TLR-induced inflammatory gene expression is essential for host defense against infections of pathogens , dysregulated production of cytokines and chemokines is detrimental to the host , resulting in septic shock and other inflammatory diseases . Nuclear and cytoplasmic compartments enable spatial separation and regulation of transcription and translation . In this study , we report that LPS/TLR4 engagement activates a nuclear function of IRAK2 that facilitates the assembly of nuclear export machinery to enable export the inflammatory mRNAs to the cytosol for protein translation . IRAK2 kinase activity is required for RanBP2-mediated IRAK2 sumoylation and nuclear translocation . Array analysis identified an SRSF1-binding motif enriched in the mRNAs that are accumulated in the nucleus of IRAK2-deficient and IRAK2-kinase inactive macrophages . Nuclear IRAK2 then phosphorylates SRSF1 and reduces SRSF1 binding to the target mRNAs . At the same time , LPS induces interaction between IRAK2 and nuclear export adaptor ALYREF to replace SRSF1 on the target mRNAs . The recruited ALYREF in turn bridges the interaction between the mRNA targets and nuclear export receptor Nxf1 , triggering nuclear export receptor Nxf1 loading for the export of the mRNAs . ( Figure 7C ) Therefore , LPS-induced IRAK2 activation provides a critical check point for the production of inflammatory cytokines and chemokines by controlling the nuclear and cytoplasmic distribution of the LPS-induced transcripts . IRAK2 is known as an atypical kinase due to the amino acid substitution of key catalytic residues ( Asp -> Asn333; Asp ->His351 ) . Notably , atypical kinases such as KSR2 ( Brennan et al . , 2011 ) and CASK ( Mukherjee et al . , 2008 ) have been reported to carry catalytic activity . We found recombinant IRAK2 was able to autophophorylate and phosphorylate MBP and SRSF1; mutations in the ATP-binding site and catalytic residues impaired the function of IRAK2 . IRAK2 kinase-inactive knockin mice showed substantially increased survival after LPS-induced septic shock compared to that of the wild-type mice . Likewise , LPS-induced serum CXCL1 and TNF levels were much reduced in IRAK2 kinase-inactive knockin mice compared to that of the wild-type mice . Taken together , these results indicate that the IRAK2 kinase activity is critical for the function of IRAK2 in LPS-induced pro-inflammatory response . While LPS-induced IRAK2 modification was previously reported , we now identified RanBP2 as the E3 ligase for IRAK2 sumoylation . We also for the first time report that LPS induces IRAK2 nuclear translocation , which seems to be dependent on LPS-induced IRAK2 sumoylation . IRAK2 mutants defective in sumoylation were retained in the cytoplasm . Likewise , LPS-induced IRAK2 nuclear translocation was abolished in RanBP2 knock-down cells , supporting the critical role of IRAK2 sumoylation for its nuclear translocation . Consistently , RanBP2 was shown to modulate cytoplasmic and nuclear transport of macromolecules . RanBP2 is a major nucleoporin that extends cytoplasmic filaments from the nuclear pore complex and contains phenylalanine–glycine repeats that bind the transport receptor importin-β . Indeed , IRAK2 was also shown to interact with importin-β , which requires the NLS on IRAK2 . It is intriguing that although IRAK2 kinase inactive mutants still interact with importin-β , they were unable to interact with RanBP2 and failed to translocate into the nucleus . While these results suggest a potential impact of IRAK2 kinase activity on RanBP2 , IRAK2 autophosphorylation may affect its interaction with RanBP2 . Our data showed that mutations in putative IRAK2 phosphorylation sites S136/T140 ( Weintz et al . , 2010 ) substantially reduced the interaction of IRAK2 with RanBP2 . It suggests that the activation of IRAK2 may result in auto-phosphorylation at S136A and T140A , which in turn mediates the interaction with RanBP2 . Future studies are required to elucidate the detailed mechanism for how phosphorylation of IRAK2 affects its interaction with RanBP2 and subsequent sumoylation . In addressing the functional impact of LPS-induced IRAK2 nuclear translocation , we found that IRAK2 plays a critical role in promoting nuclear export of LPS-induced transcripts . By array analysis , a SRSF1-binding motif was found to be enriched in the mRNAs targets that are dependent on IRAK2 for nuclear export . As discussed above , recombinant IRAK2 was able to phosphorylate SRSF1 and LPS-activated IRAK2 can phosphorylate SRSF1 and thereby reduces SRSF1 binding to the target mRNAs in macrophages . These results suggest that SRSF1 binding endows target mRNAs with sensitivity for LPS to promote nuclear export , and that LPS induces nuclear function of IRAK2 to mediate the removal of SRSF1 from the target mRNAs . Importantly , Mass Spec analysis showed that IRAK2 also interacts with nuclear export adaptor ALYREF in addition to importin-β and RanBP2 . Notably , ALYREF and SRSF1 actually belong to the same family of proteins called the shuttling SR ( serine- and arginine-rich ) proteins , sharing a similar RNA-binding domain . Indeed , we found SRSF1 and ALYREF were able to bind to the same SRSF1-binding motif; and the removal of SRSF1 allowed nuclear export adaptor ALYREF binding to the target mRNAs , suggesting that SRSF1-mediated nuclear sequestration of target mRNAs might be achieved by blocking the binding of ALYREF to the mRNAs . Importantly , previous studies have reported that ALYREF ( a subunit of the so-called TREX complex for nuclear export ) makes contact with nuclear export receptor Nxf1 , serving as a bridge between mRNA and nuclear export receptor Nxf1 . Thus , nuclear IRAK2 mediates the removal of SRSF1 and facilitates the assembly of nuclear export machinery to export the inflammatory mRNAs to the cytosol for translation . Interestingly , about 10–30% LPS-induced inflammatory transcripts were exported out of the nucleus in the absence of IRAK2 ( Figure 3B ) , which was consistent with the residual ALYREF’s binding to these transcripts in IRAK2-/- cells ( Figure 4D ) . These results implicate that there might be yet another mechanism besides IRAK2 that allows ALYREF binding to the target mRNAs for nuclear export . Future studies are required to identify additional players that modulate TLR-induced nuclear/cytosol distribution of inflammatory transcripts . IRAK2-deficient mice were previously described ( Wan et al . , 2009 ) . IRAK2 kinase-inactive knockin ( K235A and K236A ) mice were generated by co-microinjection of in vitro-translated Cas9 mRNA and gRNA into the C57BL/6 zygotes . The gRNA sequence used to generate the knockin mice is GATCACTAATACGACTCACTATAGGCCTCCCTGAGCTTCTTGAGTTTTAGAGCTAGAAAT and GATCACTAATACGACTCACTATAGGTTCGCCTTCAAGAAGCTCGTTTTAGAGCTAGAAAT . All procedures using animals were approved by the Cleveland Clinic Institutional Animal Care and Use Committee ( Protocol Number: 2014–1229 and 2017–1814 ) . LPS ( Escherichia coli 055:B5 ) was purchased from Sigma-Aldrich ( St . Louis , MO ) . Oxidized low density lipoprotein ( ox-LDL ) was purchased from ThermoFisher ( Waltham , MA ) . R848 and CpGB were purchased from Invivogen . Anti-IRAK2 ( ab62419 , RRID:AB_956084 ) and ant-Importin-β ( ab2811 , RRID:AB_2133989 ) antibodies were purchased from Abcam ( United Kindom ) . Antibodies to RanBP2 ( sc-74518 , RRID:AB_2176784 ) , ALYREF ( sc-32311 , RRID:AB_626667 ) , Actin ( sc-1615 , RRID:AB_630835 ) and Tubulin ( sc-8035 , RRID:AB_628408 ) were purchased from Santa Cruz ( Dallas , TX ) . Antibodies to SRSF1 ( 14902 ) , NXF1 ( 12735 ) , Histone ( 4499 ) and SUMO1 ( 4930 ) were purchased from Cell Signaling ( Danvers , MA ) . SRSF1 recombinant protein ( ab219488 ) was purchased from Abcam . Myelin Basic Protein ( MBP ) ( 31314 ) was purchased from Active Motif ( Carlsbad , CA ) . Bone-marrow derived macrophages were obtained from the bone marrow of tibia and femur by flushing with DMEM . The cells were cultured in DMEM supplemented with 20% fetal bovine serum ( FBS ) , penicillin G ( 100 µg/ml ) , streptomycin ( 100 µg/ml ) with M-CSF ( 50 ng/ml ) for seven days before the experiments . Hela cells ( RRID: CVCL_0030 ) were purchased from ATCC ( Manassas , VA ) and authenticated by STR analysis . The Heal cells were cultured in DMEM supplemented with 10% FBS , penicillin G ( 100 µg/ml ) and streptomycin ( 100 µg/ml ) and negative for mycoplasma contamination test using a PCR detection method . His-tagged IRAK2 WT , K123/182/592R , KK235AA , H351A mutant and Ally were sub-cloned into PET28a vector and expressed in E . Coli ( BL21 ) . The recombinant protein was purified using Ni Sepharose 6 Fast Flow column ( GE HealthCare ) . Size exclusion chromatography on a Superdex s200 high resolution column ( GE HealthCare , United Kingdom ) was used as a final step for purification . Recombinant Aos1/Uba2 , UBC9 , RanBP2RB3-4 and SUMO1 were purified as previously described ( Werner et al . , 2012 ) . IRAK2 WT , KK235AA and H351A ( 100 nM ) was incubated with myeline basic protein ( MBP ) ( 10 nM ) in the kinase assay buffer containing 25 mM Tris ( pH 7 . 5 ) , 5 mM β-glycerophosphate , 2 mM DTT , 0 . 1 mM Na3VO4 , 10 mM MgCl2 supplemented with 100 nM ATP and 1 ul [γ-32P]-ATP ( PerkinElmer ) ( 10 µCi ) at 37°C for 30 min . The samples were subjected to SDS-PAGE followed by autoradiograph . IRAK2 WT and K123/182/592R ( 200 nM ) was incubated with 50 nM Aos1/Uba2 , 100 nM Ubc9 , 24 nM RanBP2RB3-4 , 9 μM SUMO1 , and 1 mM ATP in a modified sumoylation assay buffer containing 20 mM PIPES ( pH 6 . 8 ) , 150 mM NaCl , 1 mg/ml ovalbumin , 0 . 05% Tween 20 , 1 mM DTT , and 1 μg/ml of each aprotinin , leupeptin , and pepstatin at 30°C . Samples were analyzed by western blot . RanBP2 level was determined prior to the reaction . SUMO protease Ulp1 ( 12588018 ) was purchased from ThermoFisher . Sumolyated-IRAK2 was incubated with SUMO protease in the buffer containing 50 mM Tris-HCl , pH 8 . 0 , 0 . 2% NP-40 , 150 mM NaCl , 1 mM DTT at 30°C for 30 min . Cells were harvested and lysed on ice in a lysis buffer containing 0 . 5% Triton X-100 , 20 mM Hepes pH 7 . 4 , 150 mM NaCl , 12 . 5 mM -glycerophosphate , 1 . 5 mM MgCl2 , 10 mM NaF , 2 mM dithiothreitol , 1 mM sodium orthovanadate , 2 mM EGTA , 20 mM aprotinin , and 1 mM phenylmethylsulfonyl fluoride for 20 min , followed by centrifuging at 12 , 000 rpm for 15 min to extract clear lysates . For immunoprecipitation , cell lysates were incubated with 1 μg of antibody and A-sepharose beads at 4 degree overnight . After incubation , the beads were washed four times with lysis buffer and the precipitates were eluted with 2x sample buffer . Elutes and whole cell extracts were resolved on SDS-PAGE followed by immunobloting with antibodies . Nuclear fractionation was performed using NUCLEI EZ PREP kit purchased from Sigma-Aldrich in accordance with the manufacturer’s instruction . Nuclear pellets were suspended in 30 μl of nuclear extraction buffer ( 20 mM HEPES , 400 mM NaCl , 1 mM EDTA , 1 mM EGTA in water , pH 7 . 9 ) containing freshly prepared 1 mM DTT and protease inhibitor cocktail . After 1 . 5 hr incubation on ice bath with intermittent vortexing , extracts were centrifuged and supernatant was collected for immunoprecipitation . siGENOME SMARTpool siRNAs were purchased from Dharmacon ( Lafayette . CO ) for RanBP2 , ALYREF and SRSF1 knockdown in macrophages . The targeted sequences are listed below: RanBP2 , GCACAUGUUGUUAAACUUA , GAGACGAGAGCAAGUAUUA , GAAUUAAACCCAACGCAAA , GAGCUUUACCGUUCAAAUA; ALYREF , CGAAACAACUUCCCGACAA , UCAUUAAGCUGAACCGGAG , UGAAUUUGGGACAUUGAAA , AGACCUGCACAGAGCAUAA; SRSF1 , GAAAGAAGAUAUGACGUAU , GCACUGGUGUCGUGGAGUU , UAUGUUACGCUGAUGUUUA , GAAGCUGGCAGGACUUAAA . siGENOME Non-targeting siRNA Pools were used for the control groups . Amaxa Cell Line Nucleofector Kit V ( LONZA , Switzerland ) was used to transfect macrophages following manufacturer’s instructions . Supernatants from cell cultures were collected and measured for the level of mouse cytokines CXCL1 and TNF using Duoset ELISA kits ( R&D system , Minneapolis , MN ) according to manufacturer’s instructions . The RIP assay was performed following the protocols as previous described ( Herjan et al . , 2013; Keene et al . , 2006 ) . Briefly , 107 macrophages were left untreated or treated with LPS ( 1 µg/ml ) for 1 hr . Cells were washed three times with ice cold PBS and suspended in the lysis buffer . The lysate was centrifuged and the supernatant was immunoprecipitated overnight at 4 Celsius degree , using Dynabeads ( Invitrogen ) preincubated with 20 µg anti-IRAK2 , SRSF1 , ALYREF , Nxf1 antibodies or anti-IgG antibody . RNA was purified from immunoprecipitates with Trizol ( Invitrogen ) according to the manufacturer’s instructions and treated with RNase-free DNase , the cDNAs were synthesized and 10% of the reverse transcriptase product was subjected to quantitative real-time PCR . RIP data analysis: Ct value of each RIP RNA fraction was normalized to the Input RNA fraction Ct value for the same qPCR Assay ( ΔCt ) to account for RNA sample preparation differences . Then the normalized RIP fraction Ct value ( ΔCt ) was adjusted for the normalized background ( anti-IgG ) [non-specific ( NS ) Ab] fraction Ct value ( ΔΔCt ) . The fold enrichment [RIP/non-specific ( NS ) ] was calculated by linear conversion of the ΔΔCt . Below are the formulas used for the calculation: ΔCt [normalized RIP]=Ct [RIP] – ( Ct [Input] – Log2 ( fraction of the input RNA saved ) ) ) ; ΔΔCt [RIP/NS] = ΔCt [normalized RIP] – ΔCt [normalized NS]; Fold Enrichment = 2 ( -ΔΔCt [RIP/NS] ) . Increasing amounts of purified protein and labeled probes ( 10 fmol , see in vitro transcription ) were combined in the binding buffer for 30 min . The final REMSA-binding buffer concentrations were 140 mM KCl , 10 mM HEPES pH 7 . 9 , 5% glycerol , 1 mM DTT and 0 . 33 mg/ml tRNA . The reaction was further supplemented with 15 μg salmon sperm DNA to reduce non-specific interactions from the lysate . Complexes were resolved on either 4% or 6% non-denaturing polyacrylamide gels . REMSA-radiolabeled 3’ UTR RNA probes were synthesized from BamHI linearized plasmids templates with T7 RNA polymerase using 1 mM GTP , 1 mM ATP , 1 mM CTP , 0 . 005 mM UTP and 25 μCi of 32P-labeled UTP for 3 hr at 37°C . Probes were DNAase I treated for 20 min and then phenol:chloroform extracted . The aqueous phase was passed through a Micro Bio-Spin P30 column according to manufacturer’s instructions ( BioRad ) . PLA was performed using the Duolink In Situ Green Kit purchased from Sigma-Aldrich ( DUO92101 ) in accordance with the manufacturer’s instruction . Briefly , transfected cells were washed once with ice cold PBS , followed by fixation with 4% paraformaldehyde for 15 min at room temperature . Fixed cells were then washed three times with PBS and permeabilized with 0 . 3% Triton X-100 containing PBS for 10 min . Permeabilized cells were blocked with 5% normal goat serum for 1 hr at room temperature . The cells were then incubated with primary antibodies diluted in 10% normal goat serum supplemented with 0 . 1% Tween at 4°C overnight . Following the incubation , the cells were washed three times with PBS and then incubated with two PLA probes ( Duolink In Situ PLA Probes Anti-rabbit PLUS and Anti-Mouse MINUS , Sigma-Aldrich ) for 1 hr at 37°C . After probe incubation , the samples were incubated in ligation solution for 1 hr at 37°C . After ligation , cells were washed with Wash Buffer A and incubated in the amplification solution for 2 hr at 37°C . Cells were then serially washed twice in 1 × Wash Buffer B , 0 . 01 × Wash Buffer B once , and PBS once , followed by incubation with secondary antibodies for 1 hr at room temperature . Finally , cells were washed three times with PBS and mounted in Duolink In Situ Mounting Medium supplemented with DAPI . Fluorescence images were obtained with a confocal microscope . Total RNA was extracted from spinal cord with TRIzol ( Invitrogen , Carlsbad , CA ) according to the manufacturer's instructions . 1 µg total RNA for each sample was reverse-transcribed using the SuperScript II Reverse Transcriptase from Thermo Fisher Scientific . The resulting complementary DNA was analyzed by real-time PCR using SYBR Green Real-Time PCR Master Mix . All gene expression results are expressed as arbitrary units relative to expression Actin . Cytoplasmic and nuclear RNA were extracted from WT and IRAK2-deficient macrophages and subjected to microarray analysis . Targets preparation was performed on a Biomek FXP ( Beckman Coulter ) using a GeneChip HT 3′IVT Express Kit ( Affymetrix , Santa Clara , CA ) in accordance with the manufacturer’s instruction . Labeled cRNA were hybridized on an Affymetrix GeneChip HT-MG-430PM-96 ( Affymetrix ) . Array hybridization , washing , and scanning were performed on GeneTitan ( Affymetrix ) . Three independent biological replicates were analyzed in each experiment . Probe signals were subjected to presence calling and normalized to derive relative transcript abundance . The ratio ( R ) between the nuclear ( N ) and cytoplasmic ( C ) abundance of each transcript is calculated ( R = N/C ) to quantify the nuclear retention of the transcript . The nuclear/cytoplasmic ratio of the transcript in the LPS-treated IRAK2 knockout BMDMs ( RKO ) is then divided by that of the same transcript in the corresponding wild-type BMDMs ( RWT ) . The resulting value is used as an index ( I = RKO/ RWT ) to measure the impact of IRAK2 deficiency on the nuclear retention of the transcript . Transcripts induced by LPS treatment are ranked based on the index I and the top 70 transcripts are used as the positive set for motif enrichment analysis . To define the sequences , mouse ( Mus Musculus ) genes were downloaded from Ensembl BioMART in June 2014 and represent the GRCm38 . p2 release of gene models . When there are multiple isoforms for the same gene we used the longest isoform to define its mature mRNA sequence and the genomic locus covered by its pre-mRNA sequence . We performed RNA-READ algorithm ( unpublished work , XL , HDL , and QM , in preparation ) to perform motif enrichment test . Specifically , we performed a likelihood ratio test to assess whether any of the previously defined RBP motifs ( Ray et al . , 2013 ) from our collection could better distinguish the positive set from the negative set when provided to a regression algorithm that also had access to a control set of features that consisted of all the dinucleotides contained within the corresponding motif as well as the length of the target sequence; the construction of these regression models is described below . The comparisons between the motif and the control features were restricted to specific regulatory region of the transcripts ( i . e . , 3’ UTR , 5’UTR or the coding region ) . We scored each regulatory region using a given motif by summing the accessibility of all the target sites , where a target site was defined as a perfect match to the IUPAC representation of the motif and the accessibility of a target site was defined as the average single base accessibility of the bases in the site . A score of zero was assigned to those regulatory region did not contain a motif match . The single base accessibility was assessed using RNAplfold ( Bernhart et al . , 2006 ) as described previously ( Li et al . , 2010 ) . We used the parameters with W = 200 , L = 150 and U = 1 . Although the analysis was applied in specific regulatory region , the entire transcript was input into RNAplfold to ensure correct folding of the bases close to the start codon and stop codon . We used the glmnet . R package ( Friedman et al . , 2010 ) to apply Lasso penalized logistic regression . In the Lasso regression , the hyper-parameter lambda ( i . e . the regularization strength ) was selected through a five-fold cross-validation procedure , from the lambda sequence computed by glmnet using the default settings of nlambda and lambda . min . ratio . The final value for lambda was the one ( from the sequence ) with the smallest average generalization error across the five folds . We then used this value of lambda with the ‘glmnet . fit’ object on the entire dataset to compute the weights for the features . The features with non-zero weights were selected as contributing most to the prediction . After the non-zero weight features were defined , we trained two standard logistic regression models: one using all non-zero weight features ( including the motif ) and one that contained only the non-zero weighted control features , and then assessed whether there was a significant difference in predictive power between these two nested models using a log-likelihood ratio test . Ten pairs of Wildtype and IRAK2 KI mice were injected intraperitoneal with 10 mg of LPS/kg , and survival was monitored for 72 hr . Blood for determination of plasma TNF and CXCL1 was obtained from the tail vein 4 hr after challenge . The significance of differences between two groups was determined by Student's t-test ( two-tailed ) . A P value less than 0 . 05 was considered significant . The survival curves were created by the method of Kaplan and Meier . For statistical comparison , survival curves were analyzed using the log rank test .	The innate immune system is the body’s first line of defense against invading microbes . Some immune cells carry specific receptor proteins called Toll-like receptors that can identify microbes and the signals they emit . As soon as the receptors have detected a threat – for example through sensing oily molecules that make up the cell membranes of microbes – they produce signaling proteins called cytokines and chemokines to alert other immune cells . The DNA in the cell’s nucleus carries the instructions needed to make proteins . To produce proteins , including cytokines and chemokines , the information first has to be transferred into mRNA templates , which carry the instructions to the sites in the cell where the proteins are made . Cytokine and chemokine mRNAs are generally short-lived , but previous studies in 2009 and 2011 have shown that an enzyme called IRAK2 can stabilize them to make them last longer . IRAK enzymes are activated by the Toll-like receptors after a threat has been detected . However , until now it was not known whether IRAK2 also helps to transport the mRNAs of cytokines and chemokines out of the cell nucleus . Using immune cells of mice , Zhou et al . – including some of the researchers involved in the previous studies – discovered that IRAK2 helped to export the mRNAs of cytokines and chemokines from the immune cell nucleus into the surrounding cell fluid . The Toll-like receptors recognized the oily molecules of the microbes and consequently activated IRAK2 , which lead to IRAK2 being moved into the cell nucleus . Once activated , IRAK2 helped to assemble the export machinery that moved selected mRNAs out of the nucleus to build the proteins . To do so , IRAK2 stopped a destabilizing protein from binding to the mRNA , so that instead the export machinery could transport the mRNA of the cytokines and chemokines out of the cell nucleus . A next step will be to test whether IRAK2 is required to guide exported mRNA tothe sites in the cell where the proteins are made . This new insight could help to develop new treatments for various diseases . For example , diseases in which the immune system attacks the cells of the body , rather than invaders , can be caused by too many cytokines and chemokines . Since IRAK2 helps to control the availability of cytokines and chemokines it may in future be used as a new drug target .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "cell", "biology", "immunology", "and", "inflammation" ]	2017	IRAK2 directs stimulus-dependent nuclear export of inflammatory mRNAs
Ribosomes can read through stop codons in a regulated manner , elongating rather than terminating the nascent peptide . Stop codon readthrough is essential to diverse viruses , and phylogenetically predicted to occur in a few hundred genes in Drosophila melanogaster , but the importance of regulated readthrough in eukaryotes remains largely unexplored . Here , we present a ribosome profiling assay ( deep sequencing of ribosome-protected mRNA fragments ) for Drosophila melanogaster , and provide the first genome-wide experimental analysis of readthrough . Readthrough is far more pervasive than expected: the vast majority of readthrough events evolved within D . melanogaster and were not predicted phylogenetically . The resulting C-terminal protein extensions show evidence of selection , contain functional subcellular localization signals , and their readthrough is regulated , arguing for their importance . We further demonstrate that readthrough occurs in yeast and humans . Readthrough thus provides general mechanisms both to regulate gene expression and function , and to add plasticity to the proteome during evolution . Upon encountering a stop codon , ribosomes can terminate translation with remarkable fidelity , yet they do not always do so . Stop codon readthrough , the decoding of a stop codon as a sense codon by the ribosome , plays important regulatory roles . Most immediately , readthrough diversifies the proteome by creating a pool of C-terminally extended proteins . In this capacity , it is essential to a variety of plant and animal viruses ( Cimino et al . , 2011; Li and Rice , 1989; Napthine et al . , 2012; Skuzeski et al . , 1991; Yoshinaka et al . , 1985; reviewed in Beier and Grimm , 2001; Firth and Brierley , 2012 ) . In eukaryotic host genes , readthrough is functionally important insofar as it may suppress pathological phenotypes caused by premature stop codons ( Kopczynski et al . , 1992; Fearon et al . , 1994 ) , antagonize nonsense-mediated decay ( Keeling et al . , 2004 ) , and , by changing the C-terminal sequence of a given protein , modulate its activity ( Torabi and Kruglyak , 2012 ) , stability ( Namy et al . , 2002 ) , and/or localization ( Freitag et al . , 2012 ) . In yeast , the efficiency of translation termination is modulated by [PSI+] , an epigenetic state resulting from prion-like aggregates of Sup35p , the yeast homologue of the translation termination factor eRF3 ( reviewed in Tuite and Cox , 2007 ) . Various yeast strains exhibit [PSI+]-dependent fitness advantages , implying that increased readthrough activates useful genetic diversity that is ordinarily masked by stop codons ( True and Lindquist , 2000; Halfmann et al . , 2012 ) . In addition , a small baseline level of readthrough appears to be beneficial in wild [psi−] yeast strains , as alleles of various factors controlling termination efficiency are under balancing selection ( Torabi and Kruglyak , 2011 ) . However , a broad understanding of the biological roles of readthrough in eukaryotes remains elusive due to a lack of experimental data . To date , only a handful of eukaryotic host genes have been experimentally demonstrated to undergo readthrough in wild-type or prion-free organisms ( Geller and Rich , 1980; Xue and Cooley , 1993; Klagges et al . , 1996; Steneberg et al . , 1998; Namy et al . , 2002; Jungreis et al . , 2011; Freitag et al . , 2012; Torabi and Kruglyak , 2012; Yamaguchi et al . , 2012 ) . Compelling evidence that readthrough is broadly important in eukaryotes came with the development of algorithms ( CSF and PhyloCSF ) that use orthologous nucleotide sequences from related organisms to identify protein-coding regions of a reference genome based upon signatures of amino acid conservation ( Lin et al . , 2007 , 2011 ) . Using this approach , 283 readthrough events were predicted in Drosophila melanogaster , six of which they confirmed experimentally ( Lin et al . , 2007; Jungreis et al . , 2011 ) . While these algorithms provide a powerful means to identify ancient and phylogenetically conserved readthrough events , they are limited in their ability to detect evolutionarily recent events . Nor can bioinformatic approaches identify a priori the tissues or cell types in which readthrough occurs , measure the fraction of ribosomes that read through a given stop codon , or determine whether any of these processes are regulated: such questions demand experimental approaches . To this end , we present a modified ribosome profiling protocol—based on the deep sequencing of ribosome-protected footprint fragments ( Ingolia et al . , 2009 ) —that enables analysis of translation at a genome-wide level in D . melanogaster . Application of the Drosophila ribosome profiling strategy allows annotation of the Drosophila proteome using empirical data . By examining the physical locations of ribosomes along mRNAs , we discover that readthrough is far more pervasive than expected: we identify more than 300 readthrough events not predicted by phylogenetic approaches . We provide evidence that these novel extensions are of recent evolutionary origin , and show using specific examples that both the novel and conserved extensions can produce stable protein products , be produced in a regulated manner , and contain functional subcellular localization signals . We further demonstrate that readthrough occurs at many loci in [psi−] yeast and in primary human foreskin fibroblasts , arguing that readthrough is both a ubiquitous feature of eukaryotic translation and a novel mechanism to regulate gene expression . Stop codon readthrough thus adds plasticity to the proteome during development , and provides an evolutionary mechanism for extant genes to acquire new functions . In order to study translation and , more specifically , stop codon readthrough in D . melanogaster , we sought to develop a robust ribosome profiling assay for this organism . We initially developed our protocol in S2 cells , a macrophage-like lineage derived from late-stage Drosophila embryos . In previous studies , ribosome-protected fragments or ‘footprints’ were generated by digesting eukaryotic polysome lysates with RNase I ( Ingolia et al . , 2009 , 2011 ) . In contrast to yeast and mammalian cell lines , we found that Drosophila ribosomes are highly sensitive to RNase I , potentially due to their unusual rRNA sequences and structures ( Figure 1—figure supplement 1A; Hancock et al . , 1988; Jordan , 1975; Jordan et al . , 1976; Pavlakis et al . , 1979 ) . By contrast , we found that Drosophila ribosomes tolerate micrococcal nuclease ( MNase ) over a wide range of concentrations ( Figure 1—figure supplement 1B–D ) . In contrast to RNase I , MNase has a strong 3′ A/T bias . This gives rise to a small amount of positional uncertainty with P-site mapping in MNase datasets , and prevents us from achieving the sort of sub-codon resolution seen in ribosome profiling datasets generated with RNase I . Nonetheless , replicate experiments established that our measure of translation rate ( the ribosome footprint density , defined as the number of ribosome-protected fragments per kilobase of coding region per million aligning reads in the dataset; RPKM ) , is highly reproducible and insensitive to changes in buffer conditions ( Figure 1—figure supplement 1E , Figure 1—figure supplement 2A , B; full data in supplementary table 1 at Dryad: Dunn et al . , 2013 ) . Focusing on coding regions that had a minimum of 128 reads , we observed strong correlation between replicates ( r2 = 0 . 998; Figure 1—figure supplement 2 ) and an inter-replicate standard deviation of 1 . 07-fold , comparable to our protocols in yeast and mammalian cells . Furthermore , our measurements are robust to the number of isoforms per gene , the fraction of sequence-degenerate positions in a gene , gene length , A/T content , and distribution of ribosome density within a gene ( Figure 1—figure supplement 3 ) . In early ( 0–2 hr ) Drosophila embryos , the vast majority of transcripts are maternally supplied and therefore regulated by post transcriptional processes , such as poly- or deadenylation , capping or de-capping , localization , degradation , and control of translation initiation . The early Drosophila embryo has thus been an important system for the study of post-transcriptional and specifically translational regulation ( reviewed in Lasko , 2011 ) . To enable the broad analysis of these processes , we developed a sample harvesting strategy that captures the translational state of early embryos with minimal perturbation . Specifically , we developed a cryolysis protocol in which embryos are collected directly from egg-laying dishes into liquid nitrogen , homogenized while frozen , and thawed in the presence of translation inhibitors to prevent post-lysis translation . Notably , we omit dechorionation and rinsing , steps which could induce cold shock , anoxia , and related translational artifacts . We collected replicate samples of 0–2 hr embryos , and subjected them to ribosome profiling and RNA-seq of poly ( A ) -selected mRNA . A subset of ribosomes partition into heavy polysomes ( Figure 1A ) , consistent with reports that a distinct subset of messages is well-translated at this stage ( Qin et al . , 2007 ) . Ribosome density measurements from replicate embryo collections are correlated nearly as well ( r2 = 0 . 984; Figure 1B; supplementary table 1 at Dryad: Dunn et al . , 2013 ) as measurements from technical replicates from a single culture of S2 cells ( r2 = 0 . 998; Figure 1—figure supplement 1E ) . The Drosophila embryo thus provides a system in which experimental noise approaches the precision of our measurements , a fact that will facilitate detection of even small expression differences between wild-type and mutant fly strains . 10 . 7554/eLife . 01179 . 003Figure 1 . Development and validation of a ribosome profiling assay for Drosophila melanogaster . ( A ) Aliquots of polysome lysate from 0–2 hr embryos were fractionated on 10–50% sucrose gradients with or without prior micrococcal nuclease digestion . Digestion of exposed mRNA between ribosomes collapses the polysome peaks into the monosomal ( 80S ) peak . The area under the monosome peak in the digested sample is 1 . 04-fold the combined area under the monosome and polysome peaks in the undigested sample , indicating quantitative recovery . ( B and C ) Measurements of translation are reproducible between replicates samples of 0–2 hr embryos . Pearson correlation coefficients ( r2 ) are shown for total ribosome-protected footprint counts in coding regions for all genes sharing at least 128 summed footprint counts between replicates ( B ) , or translation efficiency measurements for all genes sharing 128 summed mRNA fragment counts between replicates ( C ) . Histogram of log10 fold-changes in translational efficiency for each gene between two embryo replicates , along with normal error curve ( C , inset ) . ( D–F ) Pooled data for genes containing at least 128 summed mRNA counts between both embryo replicates . Median-centered histograms of translation efficiency ( pink ) and mRNA abundance ( blue ) ( D ) . Translational efficiency vs mRNA abundance for each gene ( E ) . Ribosome density vs mRNA abundance for each gene ( F ) . Source data may be found in supplementary table 1 ( at Dryad: Dunn et al . , 2013 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01179 . 00310 . 7554/eLife . 01179 . 004Figure 1—figure supplement 1 . Digestion with micrococcal nuclease yields a robust ribosome profiling assay . ( A ) Digestion of polysomes with RNase I degrades ribosomes . A lysate was made from S2 cells using a previous version of our protocol . Aliquots of this lysate were digested with increasing amounts of RNase I , and resolved on 10–50% sucrose gradients . As amounts of RNase I increase , the heights of all peaks—including the monosomal ( 80S ) peak—decrease before polysomes are fully resolved to monosomes . ( B ) as in ( A ) , but using micrococcal nuclease ( MNase ) and our current protocol . From 0 . 5 to 2 U MNase/μg total RNA , monosomes are resolved with no reduction in the size of the monosome peak . This result indicates that Drosophila ribosomes are stable to MNase over a broad range of concentrations , whereas the mRNA between ribosomes is digested . ( C ) Ribosome protection assay . A 320 nucleotide fragment of enolase ( FlyBase accession: FBgn0000579 ) was amplified using oligos oJGD123 & oJGD124 ( Supplementary file 2 ) . A body-labeled probe against this sequence was transcribed from this template using α32P-UTP and the T7 MaxiScript kit ( Ambion ) . S2 cell lysates were prepared as in methods and aliquoted . Aliquots were digested as in methods , except with 0 , 0 . 5 , 1 , 2 , 3 or 4 U MNase/μg total RNA . Monosomes were sedimented through a sucrose cushion , resuspended in 600 μl 10 mM Tris pH 7 . 0 , and their RNAs extracted as in ‘Materials and methods’ . Concentrations were determined using a NanoDrop spectrophotometer . 5 μg of each sample was hybridized to 50 , 000 CPM of probe overnight at 42°C . Single-stranded regions were digested with RNase A/T1 and the remaining footprint: probe duplexes detected using the mirVana micro-RNA detection kit ( Ambion ) , resolved on a 15% TBE-urea gel ( Invitrogen ) , and visualized on a Storm phosphorimager ( Molecular Dynamics by GE Healthcare Bio-Sciences , Pittsburgh , PA ) . For size markers , we end-labeled the Novex 10 bp dsDNA ladder ( Invitrogen ) with 32P . Over two-fold range of nuclease concentrations , the ∼30 nt peak corresponding to ribosome-protected footprints remains constant in size and intensity , indicating a lack of degradation consistent with the unchanged monosome peak height across this range of digestion conditions in ( B ) . Also visible is a roughly 60 nt band which we infer to be protected by adjacent ribosomes ( disomes ) that sterically exclude the nuclease . This interpretation is consistent with the presence of a small disome peak in digested samples ( c . f . panels B and D , and Figure 1A ) . ( D ) A polysome lysate was prepared from S2 cells and resolved in 10–50% sucrose gradients , with or without prior digestion with 3 U MNase/μg total RNA ( E ) A culture of S2 cells was split into aliquots and processed using our current protocol as if they were independent samples . Total counts aligning to the coding region of each gene were tabulated in each replicate . Genes sharing at least 128 footprint counts between replicates ( red ) are well-correlated , demonstrating the assay is robust ( see full discussion in Figure 1—figure supplement 2 ) . Source data may be found in supplementary table 1 ( at Dryad: Dunn et al . , 2013 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01179 . 00410 . 7554/eLife . 01179 . 005Figure 1—figure supplement 2 . Effects of buffer conditions upon reproducibility . A culture of S2 cells was divided into four aliquots , and each aliquot carried through the entire ribosome profiling procedure as an independent sample . Two aliquots ( ‘150a’ and ‘150b’ ) were processed using our standard lysis buffer with 150 mM Na+ and 5 mM Mg+ and digested with 3 U MNase/μg total RNA as described in ‘Materials and methods’ . The other two ( ‘250a’ and ‘250b’ ) were processed using an earlier version of our protocol , in which our lysis buffer contained 250 mM Na+ and 15 mM Mg++ , and in which we digested lysates with 30 U MNase/μg total RNA . We then calculated ribosome density for each gene over coding regions ( A ) , 5' UTRs ( C ) and 3' UTRs ( D ) , performed pairwise comparisons between samples . For each comparison , we binned genes based upon the summed number of reads in samples A and B , and calculated the correlation coefficients ( Pearson's r ) for the RPKM values for each gene in each bin ( left column ) . The number of genes in each bin are also shown ( right column ) . Correlations between samples for coding regions are robust across buffer regions ( A ) , though some salt-dependence is visible in 5′ and 3′ UTRs ( C and D ) . ( B ) As in ( A ) , but using only 10% of the reads . The high correlation observed at our 128-minimum-count threshold is therefore not a function of the number of genes in each bin Source data may be found in supplementary table 1 ( at Dryad: Dunn et al . , 2013 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01179 . 00510 . 7554/eLife . 01179 . 006Figure 1—figure supplement 3 . Variability in ribosome footprint density measurements are not correlated with isoform number , sequence degeneracy in the locus of interest , locus length , A/T content , or evenness of coverage . Comparisons are made between S2 cell technical replicates 150a and 150b ( Figure 1—figure supplement 2 ) ( A ) Variability of log2 fold-changes in ribosome footprint densities are no greater for multi-isoform loci ( pink ) than they are for single-isoform loci ( blue ) ( B ) Correlation of the fraction degenerate positions in each locus ( ‘Materials and methods’ ) with fold-changes in ribosome density between replicates at that locus . Loci with at least 128 counts between replicates are shown in black , those with less in red . ( C ) as in ( B ) , but correlation of length with inter-replicate fold-changes . ( D ) as in ( B ) , but correlation of A/T content with inter-replicate fold-changes . ( E ) as in ( B ) , but correlation of area under Lorenz curve with inter-replicate fold-changes Source data may be found in supplementary table 1 ( at Dryad: Dunn et al . , 2013 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01179 . 00610 . 7554/eLife . 01179 . 007Figure 1—figure supplement 4 . Measurements of translation efficiency obtained via ribosome profiling are consistent with those made using semiquantitative polysome gradients . Histograms of translation efficiency for genes labeled by Qin et al . ( 2007 ) as active ( blue ) or inactive ( yellow ) in 0–2 hr embryos . All genes are shown in gray . Source data may be found in supplementary table 1 ( at Dryad: Dunn et al . , 2013 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01179 . 007 Translational control is measured by a gene’s translation efficiency , estimated as the ratio of ribosome footprint density ( from ribosome profiling ) to mRNA abundance ( from mRNA-seq ) for each gene . Translation efficiency measurements between replicate embryo collections are highly reproducible ( r2 = 0 . 946; Figure 1C ) and consistent with prior measurements made by semiquantitative methods ( Figure 1—figure supplement 4 ) . The standard deviation of fold-changes between biological replicates is 1 . 19-fold ( Figure 1C , inset ) , allowing detection of even modest changes in translation efficiency . Remarkably , we find that the range of translation efficiencies for different messages spans four orders of magnitude , a range comparable to that observed for mRNA abundance of well-counted genes ( Figure 1D ) . Moreover , translation efficiency is uncorrelated with mRNA abundance ( r2 = 8 . 29 × 10−5; Figure 1E ) and mRNA abundance predicts only one third of the variance in the rate of protein production as measured by ribosome footprint density ( Figure 1F ) . Translational regulation is therefore a major determinant of gene expression in the early embryo ( supplementary table 1 at Dryad: Dunn et al . , 2013 ) , and ribosome profiling provides a quantitative and robust means to monitor translational regulation during development . In addition to measuring gene expression , ribosome profiling maps the physical positions of ribosomes on each transcript , and thus provides a powerful tool to annotate which portions of mRNAs are translated . Consistent with our previous work in mammals ( Ingolia et al . , 2011 ) and yeast ( Ingolia et al . , 2009; Brar et al . , 2012 ) , many 5′ UTRs in Drosophila contain substantial footprint density ( Figure 2A , Figure 2—figure supplement 1; Supplementary file 1A ) covering sequences that appear to be upstream open reading frames ( uORFs; example in Figure 2C ) . 10 . 7554/eLife . 01179 . 008Figure 2 . 5’ UTRs are translated . ( A ) Histograms of ribosome footprint density , corrected by mRNA abundance , for 5’ UTRs , coding regions ( CDS ) , and 3’ UTRs in 0–2 hr embryos . ( B ) Measurements of ribosome footprint densities of 5’ UTRs agree comparably well across a range of sequencing depths , regardless of whether 80S monosomes are specifically isolated on a sucrose gradient or enriched in a cushion . For each pair of sequencing samples , Pearson correlation coefficients ( r ) of ribosome footprint density measurements for 5’ UTRs are plotted as a function of sequencing depth . ( C ) Example of ribosome density in 5’ UTRs corresponding to the locations of uORFs . Roughly ∼200 nt of the genomic locus Ino80 covering portions of the 5’ UTR ( thin gray box ) and CDS ( thick gray box ) are shown . In both 0–2 hr embryos and S2 cells , Initiation peaks are visible at the starts of uORFs starting with an ATG codon ( green box ) and a near-cognate TTG codon ( yellow box ) as well as at the annotated start codon ( beginning of thick gray box ) . Source data for panels ( A ) and ( B ) may be found in supplementary table 1 ( at Dryad: Dunn et al . , 2013 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01179 . 00810 . 7554/eLife . 01179 . 009Figure 2—figure supplement 1 . Ribosome density over start and stop codons . Ribosome density across the average gene or ‘metagene’ reveals peaks of ribosome density at start and stop codons . For this analysis we included all genes that met the following criteria: ( a ) all transcripts deriving from that gene had one annotated start codon ( left panel ) or stop codon ( right panel ) , ( b ) all transcripts deriving from that locus covered identical genomic positions over the region of interest ( ROI ) shown , ( c ) all positions within the ROI were non-degenerate ( ‘Materials and methods’ ) , and ( d ) at least 10 reads were present in the coding subregion of the ROI . For each ROI meeting these criteria ( 2800–3200 ROI per sample ) , we generated a ‘coverage vector’ tallying ribosome density at each nucleotide position . We then normalized each coverage vector to the mean number of footprint reads covering the annotated coding region in the ROI , excluding a 3-codon buffer flanking the start or stop codon to avoid bleedthrough from initiation or termination peaks . We then plotted the median value across all normalized coverage vectors at each position . Peaks are visible in the start and stop codons of embryo samples . Consistent with our previous work , stop codon peaks are missing from S2 cell samples because terminating ribosomes release during our 2-min treatment with translation inhibitors . They are present in our embryo samples , because these are flash-frozen and lysed in the presence of translation inhibitors , which block termination as well as initiation and elongation . DOI: http://dx . doi . org/10 . 7554/eLife . 01179 . 00910 . 7554/eLife . 01179 . 010Figure 2—figure supplement 2 . Read lengths are similar in 5’ UTRs and coding regions . We aggregated all ribosome-protected reads aligning to all genes with a single initiation codon , and in which all annotated isoforms cover the same genomic positions in the ROI shown . We plotted the following statistics as a function of the reads whose 5' end mapped to each position on the x-axis . Top: number of reads ( y-axis ) aligning at each position . Because the 5' end , rather than the P-site , is plotted , the peak of ribosome density is approximately 13 nucleotides 5' of the start codon ( position 0 , x-axis ) . Middle: heatmap of read lengths ( y-axis ) as a function of position . Bottom: median read length ( y-axis ) at each position . DOI: http://dx . doi . org/10 . 7554/eLife . 01179 . 01010 . 7554/eLife . 01179 . 011Figure 2—figure supplement 3 . The choice of monosome enrichment technique—sedimentation through sucrose cushions or by fractionation on sucrose gradients—minimally affects of ribosome density across 5’ UTRs and coding regions . 3’ UTR measurements are noisier in samples prepared on cushions rather than gradients . A polysome lysate was made from collected 0–2 hr embryos , digested with MNase , and split into four aliquots . Monosomes from two aliquots were sedimented through a sucrose cushion and recovered . Monosomes from the remaining two aliquots were fractionated on 10–50% sucrose gradients and collected . All four samples were then independently carried through our protocol , and footprint density was calculated over coding regions , 5' UTRs , and 3' UTRs . Pairwise comparisons were made for each sample as in Figure 1—figure supplement 2 over coding regions ( A ) , 5' UTRs ( B ) , or 3' UTRs ( C ) . Pearson correlations ( r ) for the regions are plotted as a function of sequencing depth . Source data may be found in supplementary table 1 ( at Dryad: Dunn et al . , 2013 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01179 . 011 We attribute this density to translating 80S ribosomes rather than 48S preinitiation complexes for three reasons: first , the length distribution of protected fragments in 5′ UTRs ( 25–35 nt ) is indistinguishable from the length distribution of ribosome-protected fragments in coding regions ( Figure 2—figure supplement 2 ) , while the protected footprint of a preinitiation complex is reported to be larger ( 40–70 nt; Lazarowitz and Robertson , 1977; Pisarev et al . , 2008 ) . Second , our measurements of 5′ UTR density are indistinguishable whether we enrich digested monosomes by sedimentation through a sucrose cushion ( which collects all heavy particles ) or specifically separate them from preinitiation complexes by fractionation of a sucrose gradient ( Figure 2B , Figure 2—figure supplement 3 ) . Thus , the dominant signal contributing to our measurement of footprint density in 5′ UTRs is derived from fragments protected by 80S ribosomes . Third , because initiation and termination of translation are slow compared to elongation , initiation and termination events produce peaks of ribosome density ( Figure 2—figure supplement 1; Ingolia et al . , 2011 ) . Such peaks are frequently visible at the boundaries of predicted uORF sequences ( example in Figure 2C ) , again arguing that reads aligning to 5′ UTRs represent translation events . Given the known roles of uORFs in regulating both the translation and the stability of mRNAs ( reviewed in Meijer and Thomas , 2002 ) we anticipate that our methods will facilitate future analyses of the contributions of uORFs to control of gene expression throughout fly development . Comparative analysis of the genomes of 12 sequenced Drosophila species has provided a powerful strategy for annotating protein-coding regions in Drosophila genomes ( Lin et al . , 2007 , 2011 ) . Using this approach , 283 transcripts in D . melanogaster were demonstrated to contain clear phylogenetic signatures of amino acid conservation in the region between the annotated and next in-frame stop codons . It was therefore concluded that these regions encode C-terminal protein extensions ( hereon called ‘predicted extensions’ ) , produced by stop codon readthrough events ( Lin et al . , 2007; Jungreis et al . , 2011 ) . In our data , the density of ribosomes on 3′ UTRs is several orders of magnitude lower than in coding regions and 5′ UTRs ( Figure 2A , Supplementary file 1A ) , and many genes show highly efficient termination ( example in Figure 3B ) . However , a subset of transcripts exhibit high footprint density within the predicted extensions . To determine whether the footprint density was consistent with stop codon readthrough ( as opposed to alternate explanations , like frameshift ) , we manually scored each predicted extension whose corresponding structural gene was sufficiently expressed in our embryo sample ( 158 in total ) . An extension was scored positively if there existed ribosome density in the extension , ribosome density vanished or unambiguously decreased following the first in-frame stop codon , and positions occupied by ribosomes in the putative extension evenly covered the majority of the extension’s length ( see ‘Materials and methods’ for further details ) . By these criteria , 43 of the 283 transcripts predicted to undergo stop codon readthrough contained ribosome density consistent with a readthrough event ( example Figure 3C , full data in supplementary table 2 at Dryad: Dunn et al . , 2013 ) , including one example of double readthrough ( Figure 3D ) . We expect that the many of the remaining 240 transcripts also undergo readthrough , either at levels too low to detect at our sequencing depth , or at other developmental stages ( discussed further below ) . 10 . 7554/eLife . 01179 . 012Figure 3 . A subset of genes exhibit apparent stop codon readthrough . ( A ) Venn diagram summarizing readthrough events . Of 283 predicted extensions , 256 were consistent with FlyBase genome annotation revision 5 . 43 . For 158 of these , the corresponding coding regions were expressed in 0–2 hr embryos . Of this subset , 43 exhibited clear signs of readthrough . Others were ambiguous , untranslated , or could be explained by other mechanisms ( Figure 3—figure supplement 1 ) . In addition , we identified 307 examples of readthrough that were not phylogenetically predicted . ( B ) Example of a gene that does not exhibit readthrough . Top: genomic locus with UTRs ( thin boxes ) , introns ( line ) , and coding regions ( thick boxes ) . Middle: normalized footprint density covering the locus in 0–2 hr embryos ( blue ) and S2 cells ( red ) in reads per million . Bottom: magnification of region where a putative C-terminal extension would be found . Dashed lines: annotated and next in-frame stop codons ( C ) as in ( B ) , except stop codon readthrough creates a C-terminal protein extension in RanBPM , a gene phylogenetically predicted to undergo readthrough ( D ) as in ( B ) , but an example of phylogenetically predicted double-readthrough . ( E ) Ratios of the ribosome footprint density in putative extensions to corresponding coding regions . Blue: extensions predicted to undergo readthrough . Yellow: all other possible extensions . Extensions that overlapped any annotated CDS , snoRNA , or snRNA were excluded . Boxes: IQR . Whiskers: 1 . 5IQR . ( F ) as in ( C ) , except this transcript was not predicted to undergo readthrough . ( G ) as in ( D ) , except this transcript was not predicted to undergo single or double readthrough . Source data may be found in supplementary table 2 ( at Dryad: Dunn et al . , 2013 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01179 . 01210 . 7554/eLife . 01179 . 013Figure 3—figure supplement 1 . Examples of footprint density in 3’ UTRs attributed to sources other than readthrough . ( A and B ) Sample transcripts exhibiting translation in alternate frames . ( C ) Footprint density , potentially caused by RNA binding proteins or structures , coats the 3' UTR of EF1gamma , passing through stop codons ( red triangles ) in all three frames reaching the 3' end of the transcript . Colors as in ( A and B ) , but additionally showing RNA-seq data in gray . ( D ) The 3' UTR of HIS3 . 3B contains highly localized read density consistent with the presence of an RNA binding protein or mRNA structure , but not with translation of an open reading frame . Colors as in ( C ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01179 . 013 Surprisingly , we observed that a distinct set of transcripts not predicted to undergo readthrough also exhibits substantial footprint density between the annotated and next in-frame stop codons ( Figure 3E ) . We therefore searched for C-terminal extensions among all transcripts that met the following criteria: ( a ) a minimum of 128 footprint in the corresponding CDS , ( b ) a minimum footprint read density of 0 . 2 RPKM in the extension , ( c ) a minimum readthrough rate of 0 . 001 , and ( d ) a lack of methionine codons in the first three codons of the extension , as this latter group could be explained by initiation within the extension rather than readthrough of the upstream stop codon . We additionally excluded extensions whose translation could be explained by alternately spliced transcript isoforms that omit the stop codon . Scoring this group by the same criteria used for the predicted extensions , we identified 307 additional examples of stop codon readthrough ( hereon referred to as ‘novel extensions’; see example Figure 3F ) , including another example of double readthrough ( Figure 3G ) . In addition , we identified several transcripts that contained 3′ UTR footprint density more consistent with ribosomal frameshift ( Figure 3—figure supplement 1A , B ) , or the presence of additional downstream cistrons , RNA structure , or protein binding ( Figure 3—figure supplement 1C , D ) . These were excluded from further analysis . Because footprint density generally is far lower in 3′ UTRs than in 5′ UTRs or coding regions ( Figure 2A ) , it is possible that various sources of noise ( e . g . regions of mRNA protected by RNA structures or by RNA-binding proteins ) might contribute more substantially to this density than to the density in coding regions . We therefore asked whether footprints in 3′ UTRs exhibited behaviors specific to footprints protected by 80S ribosomes . In order to distinguish whether reads mapping to extensions were either protected by ribosomes or derived from alternate sources , we compared the total number of reads aligning to extensions in samples prepared from sucrose cushions , which collect all heavy macromolecular complexes , to those in which we specifically isolated 80S ribosomes on sucrose gradients . Footprint count measurements for each extension are highly correlated between libraries made using these two sample preparation methods , indicating that these footprints are either protected by 80S ribosomes , or by another RNA binding protein that co-sediments with 80S ribosomes ( Figure 4A; r2 = 0 . 945 ) . 10 . 7554/eLife . 01179 . 014Figure 4 . Translation downstream of the stop codon is due to readthrough . ( A ) Ribosome footprint counts for each C-terminal extension are well correlated between samples prepared by sedimentation through sucrose cushions or by fractionation on sucrose gradients ( blue ) . For comparison , footprint counts for annotated coding regions in each sample type are plotted ( gray ) . The Pearson correlation coefficient ( r2 ) for C-terminal extensions is shown . ( B ) Distributions of read lengths for footprints aligning to annotated coding regions ( CDS , red ) and to C-terminal extensions ( blue ) are similar , while lengths of footprints aligning to tRNAs , snRNAs , and snoRNAs are quite different . ( C ) Meta-gene average of ribosome density at the annotated stop codons of coding regions ( red ) , or at the stop codons that terminate extensions ( blue ) . Both averages show characteristic peaks of ribosome density above the stop codon , characteristic of translation termination . ( D ) Readthrough produces detectable protein products . Bottom: schema of reporters . Reporters containing the GFP variant Venus fused to the 120 C-terminal codons and entire endogenous 3’ UTR of a gene of interested were transfected into S2 cells . To facilitate detection of readthrough products , a double-FLAG epitope was inserted upstream of the stop codon ( red ) that terminates the putative extension . Top: reporters were immunoprecipitated with anti-GFP antibodies . Immunoprecipitates were then resolved by SDS-PAGE and western blotted with anti-FLAG antibodies to detect protein products of readthrough . Blue: names of genes containing extensions predicted to undergo readthrough . Yellow: names of genes containing novel extensions . ( E ) For each nucleotide in each stop codon that undergoes readthrough , we counted the fraction of reads containing nucleotide mismatches and present the data as a histogram . Transcripts containing stop codon nucleotides with significantly elevated mismatch rates are explicitly noted . Green: transcripts containing genomic polymorphisms that mutate one stop codon to another . Red: transcripts containing genomic polymorphisms that convert stop codons to sense codons . Black: other transcripts containing significantly elevated proportions of mismatches . ( F ) as in ( E ) , but for ribosome-protected footprint data . ( G ) as in ( F ) , but the analysis was restricted to the subset of footprints that both include the sequence of the stop codon and derive from ribosomes that have already translated the stop codon ( top , green ribosome in cartoon ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01179 . 01410 . 7554/eLife . 01179 . 015Figure 4—figure supplement 1 . C-terminal extensions in Drosophila melanogaster show ribosome release typical of coding regions , but not of internal codons . For each region of interest , the total number of reads aligning to 5 codon windows immediately upstream and downstream of that codon were tabulated , and the ratio ( downstream counts/upstream counts ) plotted against the total number of counts in the upstream window . ( A ) Comparison of release scores for termination codons of annotated coding regions and form randomly-selected codons internal to ( i . e . , at least 10 codons from the annotated start or end ) annotated coding regions . ( B ) as in ( A ) , but stop codons that terminate predicted extensions are compared against those that terminate annotated coding regions . ( C ) as in ( A ) but stop codons that terminate novel extensions are compared against those that terminate annotated coding regions . Source data may be found in supplementary table 2 ( at Dryad: Dunn et al . , 2013 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01179 . 015 Because various ribosome-binding proteins protect nucleotide fragments of distinct lengths , the size distribution of protected mRNA fragments provides a powerful approach for distinguishing 80S footprints from other sources ( Ingolia et al . , manuscript in preparation ) . Footprints in C-terminal extensions exhibit a length distribution very similar to footprints in coding regions , while those derived from non-coding sources , such as snoRNAs and tRNAs , show dramatically different length distributions ( Figure 4B ) . Thus , footprints aligning to extensions appear to be protected by 80S ribosomes . Finally , we sought to determine whether the ribosomes that appear to translate extensions are engaged in active translation , as opposed to some aberrant process of stalling or slippage ( e . g . , as described in Skabkin et al . , 2013 ) . Because terminating ribosomes produce a characteristic peak of ribosome density over annotated stop codons ( Figure 2—figure supplement 1; Ingolia et al . , 2009 ) , we asked whether the stop codons that terminate the C-terminal extensions also showed this behavior . Indeed , C-terminal extensions exhibit peaks at their stop codons , clearly arguing that footprint density in C-terminal extensions is attributable to actively-translating ribosomes ( Figure 4C ) . Because this meta-gene analysis represents a group average , we also compiled individual statistics on ribosome release in a manner similar to the RRS score described by Guttman et al . ( 2013 ) . Briefly , we tabulated the ratio of the total number of reads aligning within a five codon window immediately downstream of a stop codon to the number of reads aligning to the five codon window immediately upstream of that codon , with the expectation that if ribosomes terminate at a given stop codon , the score for that codon should approach zero . We performed this calculation separately for: ( 1 ) stop codons that terminate annotated coding regions , ( 2 ) stop codons that terminate C-terminal extensions , and ( 3 ) as a negative control , randomly selected codons internal to annotated coding regions . We find that the scores of stop codons that terminate C-terminal extensions fall within the distribution of scores for stop codons that terminate annotated coding regions ( Figure 4—figure supplement 1 ) , again arguing that the read density covering putative C-terminal extensions are in fact produced by ribosomes that have undergone stop codon readthrough rather than other processes . It is possible that the population of ribosomes that read through stop codons is engaged in a pathological translation process that might not produce detectable protein products . We therefore asked whether we could detect translation products by immunoprecipitation ( IP ) and western blotting . We created reporter constructs for a panel of transcripts including five predicted and 10 novel extensions that exhibited readthrough in both 0–2 hr embryos and S2 cells . In each construct , we fused Venus ( a GFP variant ) upstream of a portion of each transcript containing the C-terminal 120 codons of the annotated coding sequence and the entire endogenous 3′ UTR . To visualize readthrough , we fused a double FLAG epitope to the C-terminus of the putative C-terminal extension . We transfected these constructs into S2 cells , immunoprecipitated the reporter at the N-terminus using anti-GFP beads , and detected the extensions by western blotting using an anti-FLAG antibody . We detected readthrough products of the correct size for eight of the reporters , arguing that at least this subset of extensions yields C-terminally extended proteins in vivo ( Figure 4D ) . While we did not seek to detect C-terminally extended proteins generated by endogenous transcripts ( e . g . , through mass spectrometry ) , we do believe our reporter constructs to be at least as faithful as those used in earlier literature , as we included substantially more nucleotide context ( 120 codons upstream of stop plus the entire endogenous 3′ UTR ) than other groups screening through candidate genes to find readthrough signals ( 2–8 codons upstream and 3–15 codons downstream of the stop codon; Fearon et al . , 1994; Harrell et al . , 2002; Namy et al . , 2002 , 2003 ) . The appearance of stop codon readthrough , both in ribosome profiling data and in IP-westerns , could result from several other processes , such as selenocysteine insertion , genomic mutation of stop codons to sense codons , or the editing of stop codons in mRNAs . We consider each of these in turn . UGA stop codons may be decoded by specialized translation machinery as the unconventional amino acid selenocysteine if the 3′ UTR contains a selenocysteine insertion ( SECIS ) element . However , UGA stop codons represent only 25% of the readthrough events we report , and none of these are annotated as selenoproteins in either FlyBase ( Marygold et al . , 2013 ) or SelenoDB ( Castellano et al . , 2008 ) . Furthermore , we were unable to detect SECIS elements in any of their 3′ UTRs using SeciSearch 2 . 19 ( Kryukov et al . , 2003 ) . Thus , at most , even unannotated selenocysteine insertion events could only account for a small fraction of the readthrough events we report . We also exclude the possibility that readthrough might result from genomic polymorphisms or RNA editing at the stop codon . Because both types of events would be represented in our data as mismatches between read alignments and the reference transcript sequence , we counted the total number of matching and mismatching reads covering each nucleotide position in each stop codon in our mRNA-seq and ribosome footprint datasets . For each dataset , we calculated a global average proportion of mismatching reads , and used the binomial test to identify stop codon nucleotides whose individual proportion of mismatches significantly deviated from the corresponding global average . Together , the mRNA and footprint datasets identified a total of 10 nucleotide positions whose mismatch rates significantly exceeded the average ( Figure 4E , F ) . Three positions contained genomic polymorphisms that changed one stop codon to another stop codon ( Figure 4E , F , green ) . Two ( Figure 4E , F , red ) contained genomic polymorphisms that converted the stop codon to a sense codon . These two transcripts were therefore excluded from further study . The remaining five positions contained a variety of mismatches each occurring at low frequency . These observations are inconsistent with the presence of a genomic polymorphism at those positions , which should cause a 50% or 100% frequency of a single mismatch , depending on whether the polymorphism is hetero- or homozygous ( Figure 4E , F , black ) . An alternate explanation for a low but elevated proportion of mismatches is RNA editing , the conversion of one nucleotide to another in an mRNA . In Drosophila , the only mechanism known to edit mRNA is the deamination of adenine to inosine , which is converted to guanine by reverse transcriptase ( Ramaswami et al . , 2013 ) . A-to-I editing thus appears in sequencing data as a preference for A-to-G transitions among mismatches . Of the five mismatching positions we could not ascribe to genomic polymorphisms , four contain thymine or guanine rather than adenine residues in the reference sequence , and therefore cannot be edited by this pathway . We therefore attribute these mismatches to sequencing error . The majority of mismatches at the single remaining position are transversions from adenine to thymine , similarly arguing that these mismatches are more likely due to sequencing error than to A-to-I editing . Formally , it is possible that a minor fraction of transcripts are edited , but that this fraction , even if small as measured in the RNA-seq or total footprint data , might account for all of the stop codon readthrough we observe . Analysis of the ribosome footprint data allows us to explore this possibility directly . Specifically , were this the case , the sequences of all the footprints deriving from ribosomes that have undergone readthrough—namely , those whose A-sites have already translated the stop codon—should contain evidence of editing ( Figure 4G , top ) . We therefore separately analyzed the footprints deriving from this specific pool of ribosomes . Our dataset provided sufficient coverage to test 419 of 450 such positions ( 93% of the total ) . Of these , only four stop codon positions exhibited significantly elevated levels of mismatch ( Figure 4G , bottom ) . All of these were identified in the mRNA and total footprint datasets above as having genomic polymorphisms ( Figure 4E–F ) . Thus , our most stringent dataset contains no positive evidence of RNA editing . Further , this dataset contains positive evidence against RNA editing . Under the null hypothesis that A-to-I editing drives readthrough , one would expect nearly all footprints ( for our purposes , conservatively assuming 90% ) in the A-site footprint dataset to contain an edited base . Under this assumption , we used a binomial test to estimate the probability of observing the proportion of A-to-G mismatches in the A-site footprint dataset at each adenine residue sufficiently covered by reads ( 217 positions , representing roughly 50% of A positions in all readthrough events reported ) . In this analysis , all positions contained significantly fewer A-to-G mismatches than expected under the hypothesis of A-to-I editing , ( Bonferonni-corrected p<<0 . 05 for all transcripts ) , indicating that A-to-I editing plays no part in any of the readthrough events we could test . Because we detected far more readthrough events in Drosophila than were predicted from phylogenetic data , we collected yeast datasets and examined them for empirical evidence of readthrough . Importantly , because the [PSI+] form of the yeast eRF3 homologue is known to promote readthrough , we limited our analysis to data collected from [psi−] strains . In contrast to MNase ( which exhibits a 3′ A/T cutting bias , yielding positional uncertainty of the ribosomal P-site , see ‘Materials and methods’ ) , RNAse I shows little cutting bias . Therefore , libraries prepared with RNase I ( e . g . , yeast and mammalian libraries ) offer superior spatial resolution along mRNAs . In such libraries , the locations of ribosome-protected footprint fragments in coding regions exhibit a characteristic three-nucleotide periodicity or phasing from which reading frames can be deduced ( Ingolia et al . , 2009 , 2011 ) . We therefore tabulated the phasing of ribosome-protected footprint fragments in all annotated coding regions , putative C-terminal extensions , and the 40 codon windows downstream of the putative extensions as an approximation of the portion of the 3′ UTR distal to the putative extension ( hereafter called ‘distal 3′ UTRs’ ) . To control for cloning biases caused by skewed nucleotide frequencies at each phase , we tabulated the phasing of randomly-fragmented mRNA fragments that were cloned using the same protocol and aligned to the same regions . Non-random phasing consistent with translation is apparent in both the coding regions and the putative extensions , but not the distal 3′ UTR ( p=3 . 98 × 10−26 , Χ2 test , footprints vs mRNA fragments in extension , dof = 2; Figure 5A ) . Importantly , the major component of phasing in the putative extensions occurs in the same reading frame as that of coding regions , indicating that readthrough ( as opposed to , e . g . , frameshift ) is a major contributor to protected fragment density in 3′ UTRs in yeast . Having found global evidence for readthrough , we manually scored a subset of yeast genes to identify individual examples of readthrough , using the same filtering and scoring criteria we used in the Drosophila datasets . We found 30 clear examples of readthrough in yeast ( examples in Figure 5B , C; full results in supplementary table 3 at Dryad: Dunn et al . , 2013 ) , demonstrating that readthrough is not unique to Drosophila . 10 . 7554/eLife . 01179 . 016Figure 5 . Readthrough occurs at specific stop codons in [psi-] yeast and in human foreskin fibroblasts . ( a ) Triplet periodicity of 28-mers from yeast data in all non-overlapping coding regions ( CDS ) , putative C-terminal extensions , and distal 3’ UTRs indicates that a signature of translation readthrough is visible in extensions on a bulk scale . Distal 3’ UTRs were estimated as 40 codon windows following putative extensions . Putative extensions and distal 3’ UTRs that overlap annotated coding regions , snoRNAs , snRNAs , tRNAs or 5’ UTRs were excluded from the analysis . ( B and C ) Examples of yeast transcripts that undergo readthrough , as in Figure 3B . ( D and E ) Examples of transcripts that undergo readthrough in human foreskin fibroblasts , as in Figure 3B . ( F ) Distribution of readthrough rates , by organism , for all extensions of sufficient length not to be covered by bleedthrough from termination peaks ( ‘Materials and methods’ ) . Dashed line: fifth percentile of readthrough rate in conserved extensions in D . melanogaster , 1 . 2% . Source data may be found in supplementary tables 2 , 3 , and 4 ( at Dryad: Dunn et al . , 2013 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01179 . 01610 . 7554/eLife . 01179 . 017Figure 5—figure supplement 1 . In yeast and humans , reads mapping to C-terminal extensions are drawn from the same length distribution as reads mapping to coding regions . ( A ) Length distributions of reads mapping to coding regions and extensions in yeast . ( B ) Length distributions of reads mapping to coding regions and extensions in human foreskin fibroblasts . DOI: http://dx . doi . org/10 . 7554/eLife . 01179 . 017 Because readthrough has been observed in two mammalian genes ( Geller and Rich , 1980; Yamaguchi et al . , 2012 ) , we collected data from primary human foreskin fibroblasts and sought evidence of readthrough in humans . We identified 42 readthrough events in the human data ( Figure 5D , E; full results in supplementary table 4 at Dryad: Dunn et al . , 2013 ) . These events are not explained by selenocysteine insertion , and , as in Drosophila , read lengths mapping to extensions in the yeast and human datasets are similar to those mapping to coding regions in these organisms ( Figure 5—figure supplement 1 ) . Thus , readthrough appears to be prevalent in all three organisms . To estimate how many of the novel extensions we detected might be translated at a biologically significant level , we estimated a threshold for biological significance as the fifth percentile of readthrough rates for the phylogenetically conserved extensions that were translated in the D . melanogaster embryo , a rate of 1 . 2% . Out of all the extensions for which we could measure readthrough rates ( i . e . , those sufficiently long not to be covered by stop codon peaks , see ‘Materials and methods’ ) , 61 . 8% of the novel extensions in Drosophila , 94 . 7% of the extensions in human foreskin fibroblasts , and 40 . 0% of the extensions in yeast exceeded this threshold , arguing that readthrough might be important in all three organisms ( Figure 5F ) . Because 307 of the 350 readthrough events we discovered were not predicted phylogenetically , we sought to determine whether any of them showed signs of protein-coding conservation through the Drosophila phylogeny . To this end , we used PhyloCSF , which reports a log-likelihood ratio reflecting the relative probabilities of observing a given alignment of orthologous nucleotide sequences under models of protein-coding or non-coding evolution ( Lin et al . , 2011 ) . By this metric , only 14 of the 307 novel extensions score positively ( Figure 6A ) , and their distribution of PhyloCSF scores was not markedly different from the global distribution ( Figure 6—figure supplement 1A ) , indicating a lack of phylogenetic evidence for amino acid conservation . 10 . 7554/eLife . 01179 . 018Figure 6 . Novel C-terminal extensions in Drosophila melanogaster show signatures of selection within the melanogaster lineage . ( A ) Scatter plot comparing readthrough rates for confirmed extensions against PhyloCSF scores . Blue: predicted extensions . Yellow: novel extensions . Datapoints with unreliably measured PhyloCSF scores or readthrough rates are not shown ( ‘Materials and methods’ ) . ( B ) Z-curve classifier suggests that novel extensions have a nucleotide character intermediate between distal 3’ UTRs and coding regions . Histograms of Z-curve scores for 81-nucleotide windows drawn from annotated coding regions ( CDS ) , distal 3’ UTRs , predicted extensions , and novel extensions . A single window was selected from each region 81 or more nucleotides long . Shorter regions were excluded from analysis , as they were empirically found to be noisy during classifier training . The Z-curve classifier was trained on windows drawn from CDS and distal 3’ UTRs as described in ‘Materials and methods’ . ( C ) Novel extensions accumulate SNPs with a stronger preference than distal 3’ UTRs . Proportion of SNPs in CDS , predicted extensions , novel extensions , and distal 3’ UTRs which would be nonsynonymous if translated in frame . SNPs were obtained from wild isolates of wild-type flies by the Drosophila Population Genomics Project , and were downloaded from Ensembl ( Flicek et al . , 2013 ) . Source data may be found in supplementary table 2 ( at Dryad: Dunn et al . , 2013 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01179 . 01810 . 7554/eLife . 01179 . 019Figure 6—figure supplement 1 . Novel C-terminal extensions in Drosophila melanogaster show signatures of selection within the melanogaster lineage . ( A ) Histogram of PhyloCSF scores for C-terminal extensions . Blue: phylogenetically predicted extensions that were confirmed in our datasets . Yellow: unpredicted extensions discovered in our datasets . Gray: global distribution of all potential extensions . The distribution of novel extensions is not substantially different from the global distribution , suggesting that many of these extensions are not phylogenetically conserved beyond melanogaster . Source data may be found in supplementary table 2 ( at Dryad: Dunn et al . , 2013 ) . ( B ) A second Z-curve classifier was trained on 81-nucleotide windows of coding regions , and 81-nucleotide windows of distal 3′ UTRs , but excluding the last 50 bases of annotated UTR to remove potential effects of polyadenylation signals upon classifier scoring . As in Figure 6B , predicted extensions overlay coding regions , and novel extensions display a significant shift in median from distal 3′ UTRs ( p=3 . 81 × 10–22 , Mann–Whitney U test ) , indicating the shift identified in Figure 6B is not due to polyadenylation signals . DOI: http://dx . doi . org/10 . 7554/eLife . 01179 . 019 The lack of detectable phylogenetic evidence of amino acid conservation among the novel extensions suggests two models: either ( 1 ) the novel extensions , on average , are selectively neutral , and occur only because they do not incur too great a fitness disadvantage , or ( 2 ) the novel extensions are under selection , but originated after the divergence of D . melanogaster from its closest sequenced relatives , making conservation in this group undetectable by cross-species tools such as PhyloCSF . To distinguish these possibilities , we used two tests to detect signs of selection for protein coding specifically within D . melanogaster . To determine whether the nucleotide sequences of novel extensions show signs of selection for protein coding potential , we implemented a Z-curve classifier , a machine-learning technique that separates coding regions from non-coding regions based upon phased differences in nucleotide k-mer frequency ( Gao and Zhang , 2004 ) . We trained the classifier to distinguish annotated coding regions from distal 3′ UTRs ( see ‘Materials and methods’ for details ) . Consistent with a long history of protein-coding selection , extensions predicted by phylogenetic conservation showed a nucleotide character indistinguishable from annotated coding regions ( Figure 6B ) . By contrast , novel extensions exhibit a nucleotide character intermediate between coding regions and distal 3′ UTRs ( p=1 . 02 × 10−23 , Mann-Whitney U test , distal 3′ UTR vs novel extensions ) , which is consistent with an evolutionary trajectory towards coding-like character from a 3′ UTR . This effect is not due to specific nucleotide signals found in distal 3′ UTRs ( p=3 . 81 × 10−22 , Figure 6—figure supplement 1B ) , and was robust across Z-curve classifiers trained on different windows drawn from distal 3′ UTRs ( see ‘Materials and methods’ ) . To obtain more direct evidence for or against protein-coding selection , we analyzed SNP data from 50 individuals of D . melanogaster from the Drosophila Population Genomics Project ( http://www . dpgp . org ) . We determined the proportion of SNPs that would be synonymous if translated in-frame in coding regions , predicted extensions , novel extensions , and distal 3′ UTRs . Novel extensions show a modest but significant preference for synonymous SNPs above the background level of distal 3′ UTRs ( Figure 6C; p=1 . 42 × 10−5 , one-sided Fisher’s exact test ) , but below that of the predicted extensions ( p=8 . 42 × 10−9 , one-sided Fisher’s exact test ) . This pattern suggests that a subset of the novel extensions is undergoing selection for protein coding , and that the contribution from this subset to the average SNP preference outweighs the contributions from other subsets of extensions that are selectively neutral or undergoing diversifying selection . Together , these results favor the hypothesis that at least a fraction of the novel extensions are of recent evolutionary origin and have come under selection within the melanogaster lineage . In order to determine whether C-terminal extensions might be functional , we sought evidence for biological regulation of readthrough rates . We therefore queried our S2 cell and embryo datasets for evidence of differential regulation of readthrough in all genes that were sufficiently expressed in both datasets and contain only one , unique annotated coding region across all transcripts . For each gene meeting these criteria , we tabulated the number of ribosome-protected footprints in the corresponding coding region and extension in each tissue type , and calculated a p value for the observed distribution of counts using Fisher’s exact test . Controlling the false discovery rate at 5% , we found nine of 182 testable transcripts to significantly change between samples , indicating that all nine should be true positives ( Table 1; full data in supplementary table 2 at Dryad: Dunn et al . , 2013 ) . Thus , readthrough is differentially regulated between Drosophila cell types . 10 . 7554/eLife . 01179 . 020Table 1 . Readthrough is differentially regulated between 0–2 hr embryos and S2 cellsDOI: http://dx . doi . org/10 . 7554/eLife . 01179 . 020Gene IDAliasEmbryo readthrough rateS2 readthrough ratePhyloCSF scorep valuelog10 fold changeDirection of changeFBgn0036824CG39027 . 15E−012 . 46E−03−241 . 076 . 55E−10−2 . 46↓FBgn0004362HmgD8 . 82E−031 . 21E−02−747 . 857 . 08E−070 . 14↑FBgn0035432ZnT63C7 . 17E−032 . 71E−02181 . 261 . 14E−060 . 58↑FBgn0010409RpL18A1 . 39E−022 . 08E−03−197 . 785 . 85E−06−0 . 83↓FBgn0039218Rpb105 . 18E−032 . 03E−02−333 . 388 . 06E−060 . 59↑FBgn0038100Paip22 . 10E−024 . 60E−03−497 . 093 . 71E−05−0 . 66↓FBgn0261790SmE7 . 55E−037 . 80E−04−530 . 289 . 60E−04−0 . 99↓FBgn0030991CG74532 . 18E−015 . 28E−02−164 . 362 . 00E−03−0 . 62↓FBgn0043796CG122192 . 85E−011 . 90E+00−27 . 832 . 11E−030 . 82↑For each transcript , the number of reads aligning to the CDS and corresponding extension were tabulated in both embryo and S2 cell datasets . p values for significant changes were calculated using Fisher’s Exact Test . The False Discovery Rate was controlled at 5% using the procedures of Benjamini and Hochberg ( ‘Materials and methods’ ) , yielding nine transcripts with significant p values . In principle , readthrough could be regulated by: ( 1 ) changes in the expression or activities of global factors ( e . g . , eukaryotic release factors , charged tRNA abundance etc ) , ( 2 ) by gene- or transcript-specific elements , like mRNA structures , or ( 3 ) by a combination of both . In the first scenario , readthrough rates for all transcripts should increase or decrease monotonically in one cell or tissue type compared to another . In the latter two scenarios , readthrough rates should increase for some transcripts , but decrease for others . We identified four significant increases and five significant decreases in readthrough rate in embryos compared to S2 cells , indicating that readthrough is at least in part regulated on a transcript-by-transcript basis . The distribution of fold-changes in readthrough rate spans several orders of magnitude , indicating that transcripts that are robustly read through in one cell type are not necessarily read through in another ( Table 1 ) . This result implies that extensions function in specific cellular or developmental contexts , consistent with earlier reports that readthrough of specific genes is regulated in metazoans ( Robinson and Cooley , 1997; Yamaguchi et al . , 2012 ) . Because we observe such a large magnitude of regulation , we believe the 350 readthrough events we report here to represent a small subset of a larger group that occur throughout the lifetime of an individual fly . We therefore expect many of the extensions that were phylogenetically predicted but not observed in our samples are in fact translated at other developmental stages in Drosophila . Finally , because transcripts with significant p values are statistically more highly counted in their extensions than those without significant p values ( p=2 . 4 × 10−3 , Mann-Whitney U test ) , we surmise that our ability to detect regulation was limited by sequencing depth and that the true number of transcripts whose readthrough rates are regulated in tissue- or condition-specific manners is in fact larger than we report . Many peptide sequences—such as signal sequences , degrons , and phosphorylation sites—affect the localization , stability , or activity of proteins . Because these sequences are frequently short and/or degenerate , a high proportion of even random peptide sequences confer function ( Kaiser et al . , 1987; Kaiser and Botstein , 1990 ) . Thus , a C-terminal extension produced by termination failure could purely by chance alter the function or behavior of its host protein , and thus come under selection . Indeed , Freitag et al . ( 2012 ) reported two readthrough events in fungi that append peroxisomal localization signals ( PTS1 ) to the C-termini of glyceraldehyde-3-phosphate dehydrogenase ( GAPDH ) and 3-phosphoglycerate kinase , enabling these typically cytosolic enzymes to function in peroxisomal metabolism . We therefore searched our full set of C-terminal extensions for short peptide signals that direct peroxisome localization , nuclear localization ( NLS ) , prenylation , or ER retention , or that resemble transmembrane domains ( see ‘Materials and methods’ ) . PTS1 signals were detected in one extension . 10 proteins not annotated as nuclear in FlyBase contain predicted NLSes in their extensions . Eight extensions contain predicted transmembrane domains and one contains a C-terminal prenylation signal . No extension contained an ER retention signal ( Table 2 ) . 10 . 7554/eLife . 01179 . 021Table 2 . C-terminal extensions contain predicted functional peptide signalsDOI: http://dx . doi . org/10 . 7554/eLife . 01179 . 021Gene IDAliasExtension coordinatesPhyloCSF scoreSignal detectedFBgn0000173benX:13892649–13892781 ( + ) −302 . 18NLSFBgn0005278Sam-S2L:113542–113647 ( + ) −195 . 30NLSFBgn0026144CBPX:7235840–7236599 ( + ) 128 . 52NLSFBgn0031897CG137842L:7206347–7208015 ( − ) 4775 . 49NLSFBgn0033712CG131632R:8209607–8209934 ( + ) −675 . 02NLSFBgn0036272CG43003L:12265284–12265557 ( − ) −193 . 87NLSFBgn0039213atl3R:20459429-20459720 ( + ) 28 . 43NLSFBgn0260934par-12R:15370912–15371608 ( + ) 654 . 90NLSFBgn0261606RpL27A2L:4457220–4457289^4457374–4457380 ( − ) −148 . 56NLSFBgn0262114RanBPM2R:6322727–6323228 ( + ) 1045 . 90NLSFBgn0031683CG42302L:5098384–5098573 ( + ) −5 . 34Transmembrane domainFBgn0033712CG131632R:8209607–8209934 ( + ) −675 . 02Transmembrane domainFBgn0035498Fit13L:4106386–4106518 ( + ) −323 . 36Transmembrane domainFBgn0036980RhoBTB3L:20374798–20374821^20374891–20374982 ( + ) 154 . 91Transmembrane domainFBgn0037321CG11723R:1221902–1222220 ( + ) −624 . 55Transmembrane domainFBgn0040813Nplp23L:13350197–13350296 ( + ) −242 . 85Transmembrane domainFBgn0053523CG335233L:5922386-5922854 ( + ) 383 . 85Transmembrane domainFBgn0263864Ark2R:12913933-12914062 ( + ) −123 . 89Transmembrane domainFBgn0039690CG19693R:25567115–25567154 ( + ) 11 . 52PTS1FBgn0035540Syx173L:4404848–4404983 ( + ) 290 . 83Farnesyltransferase signalPeptide sequences of C-terminal extensions were examined using various prediction servers ( see ‘Materials and methods’ ) . Those containing predicted features are shown here . NLS: nuclear localization signal . PTS1: peroxisome localization signal . Coordinates are 0-indexed and half-open . Splice junctions are denoted with carrots ( ‘^’ ) . Strands are indicated in parentheses . To determine whether any of the putative nuclear localization signals ( NLSes ) function in vivo , we constitutively fused C-terminal extensions containing putative NLSes to the C-terminus of a GFP-mCherry-GST reporter , which is excluded from the nucleus ( Figure 7 , left column; Chan et al . , 2007 ) . When expressed in S2 cells , three of four NLSes relocalized the cytosolic reporter to the nucleus at levels above background ( Figure 7 , columns 3–5 ) , arguing that these extensions can regulate the localization of their endogenous host proteins . Given the large number of short peptide signals ( e . g . , phosphorylation motifs , degradation motifs , ubiquitination sequences , etc ) that have been discovered , and the limited number of reporters we tested here , we likely underestimate the number of extensions that confer function . Nonetheless , our results clearly establish that C-terminal extensions can alter protein function . 10 . 7554/eLife . 01179 . 022Figure 7 . Extensions contain functional localization signals . Ordinarily , a GFP-mCherry-GST reporter is excluded from the nucleus ( first column ) . When an SV40 NLS is appended to the reporter , it is predominantly nuclear ( second column ) . Three extensions also contain functional NLSes which at least partially relocalize the reporter to the nucleus when constitutively fused to it ( remaining columns ) . First row: GFP reporter . Second row: nuclei stained with Hoechst . Third row: merged GFP and Hoechst . Fourth row: DIC . DOI: http://dx . doi . org/10 . 7554/eLife . 01179 . 022 Here we present the first comprehensive study of stop codon readthrough in a eukaryote . Using empirical data , we identified 350 readthrough events in Drosophila melanogaster , the vast majority of which were not predicted from phylogenetic signatures . We further demonstrate that readthrough occurs in yeast and humans . Our studies indicate that readthrough is far more pervasive than previously appreciated , is biologically regulated , and may append functional peptide signals to host proteins . Together , these results argue that stop codon readthrough provides an important mechanism to regulate gene expression and function . Our work further suggests that readthrough provides an important means for genes to acquire new functions throughout the course of evolution . Mechanistic studies of readthrough in various systems have implicated many factors in the modulation of readthrough rates . These include the identity of the stop codon ( Robinson and Cooley , 1997; Chao et al . , 2003; Napthine et al . , 2012 ) , nucleotide context surrounding the stop codon ( Bonetti et al . , 1995; McCaughan et al . , 1995; Cassan and Rousset , 2001; Chao et al . , 2003 ) , local or distant RNA structures ( Wills et al . , 1991; Feng et al . , 1992; Steneberg and Samakovlis , 2001; Cimino et al . , 2011; Firth et al . , 2011; Napthine et al . , 2012 ) , specific hexanucleotide sequences ( Skuzeski et al . , 1991; Harrell et al . , 2002 ) , snoRNA-mediated pseudouridylation of stop codons ( Karijolich and Yu , 2011 ) , the identity of the tRNA present in the ribosomal P-site ( Mottagui-Tabar et al . , 1998 ) , the peptide sequence of the nascent chain ( Mottagui-Tabar et al . , 1998 ) , the concentrations of endogenous suppressor tRNAs ( reviewed in Beier and Grimm , 2001 ) , and proteins that bind the ribosome or mRNA ( Keeling et al . , 2004; Hatin et al . , 2007; Green et al . , 2012 ) . With the exception of the readthrough signal identified in Tobacco mosaic virus ( Skuzeski et al . , 1991 ) , the majority of readthrough events that have been mechanistically characterized are regulated by two or more such factors , often in complex , context-specific ways . For example , downstream nucleotide contexts which promote readthrough of one stop codon can inhibit readthrough of other stop codons , and these effects can be non-linearly synergistic with upstream nucleotide contexts ( Bonetti et al . , 1995 ) . Such complexity is advantageous insofar as it allows readthrough rates to be independently regulated for each transcript , consistent with our own observations . Unsurprisingly , however , this complexity has hindered efforts to identify simple cis-acting sequence elements that deterministically predict readthrough , and underscores the importance of having a method to measure readthrough empirically in a physiological setting in vivo . By using ribosome profiling to measure readthrough rates over a variety of tissue types and developmental stages , it may be possible to decompose the regulatory complexity into individual components , and then determine the cis-acting elements that collaborate to regulate readthrough in tissue-specific manners . Just as alternative splicing provides a means for proteins to acquire new domains or functional modules , we propose , along with the Lindquist ( True and Lindquist , 2000 ) and Kellis ( Jungreis et al . , 2011 ) groups , that stop codon readthrough can provide a mechanism for proteins to evolve at the C-terminus . In this model , transcripts that contain contexts favorable to leaky termination would yield substoichiometric , C-terminally extended populations of cellular proteins . If a particular extension is deleterious , natural selection can favor mutations in the corresponding mRNA that promote efficient termination rather than readthrough . If , instead , the extension provides a fitness advantage , selection can act upon both its amino acid sequence ( to tune its function ) , as well as the nucleotide sequence of its mRNA ( to increase or otherwise regulate its readthrough rate ) . In extreme cases , where an extension is universally advantageous , a mutation that changes a stop codon to a sense codon might become fixed , resulting in a constitutively extended gene . Conceivably , the two C-terminal extensions that we discovered to contain sense codons in place of their annotated stop codons could be the end result of this process . Several lines of evidence are consistent with this evolutionary model . First , non-zero readthrough rates ( 0 . 02–1 . 4% ) have been observed even for control non-readthrough reporter constructs in [psi−] yeast ( Fearon et al . , 1994; Bonetti et al . , 1995; Namy et al . , 2002; Keeling et al . , 2004; Torabi and Kruglyak , 2011 ) and mammalian cells ( Firth et al . , 2011; Napthine et al . , 2012 ) , arguing that under typical conditions in a variety of eukaryotes , there is a small pool of C-terminally-extended proteins , originating from a wide variety of genes , available for selection to act upon . Secondly , in specific circumstances , selection appears to favor leaky termination and its extension products . Torabi et al . ( Torabi and Kruglyak , 2011 ) reported that in a panel of wild strains of [psi−] yeast , allelic combinations of SUP45 and TRM10 that promote and inhibit readthrough appear to be in balancing selection , implying that a low baseline level of readthrough is beneficial . Similarly , numerous reports have demonstrated that wild strains of yeast exhibit [PSI+]-dependent fitness advantages in a variety of stress conditions , arguing that functions conferred by C-terminal extensions can provide adaptive advantages ( True and Lindquist , 2000; Halfmann et al . , 2012 ) . Thirdly , extensions have a high probability of conferring function without prior tuning by natural selection . This point is illustrated by the studies of Kaiser and Botstein , which demonstrated that a large proportion—roughly 30%—of randomly-generated peptide sequences are functional , insofar as they can relocalize a cytosolic form of invertase to the nucleus , mitochondrion , or endoplasmic reticulum in yeast ( Kaiser et al . , 1987; Kaiser and Botstein , 1990 ) . Given the large number of short peptide signals now known ( e . g . , D-boxes , KEN-boxes , SH3 binding epitopes , phosphorylation sites , etc ) , it is likely that a far greater fraction of random peptide sequences contain at least one functional signal . Consistent with this hypothesis , we discovered C-terminal extensions that are not phylogenetically conserved nonetheless contained functional NLSes in Drosophila . Furthermore , because these short signals are modular , their addition to the C-terminus of a protein can confer novel function , without requiring modification or coevolution of the host protein . In this way , even novel C-terminal extensions arising purely from termination failure can immediately alter the behavior of their host proteins , in beneficial or deleterious ways , and thus come under selection . Over evolutionary time , this process could yield phylogenetically conserved readthrough events or , in extreme cases , constitutively extended proteins . Our model predicts that , at any given moment , ribosome profiling should detect a broad spectrum of conservation among readthrough events: at one extreme are ancient , phylogenetically conserved extensions , and , at the other , extensions of recent evolutionary origin . Between these , one would find extensions under varying degrees of age , conservation , and selection . This notion is borne out in our data: in Drosophila , a subset of readthrough events are well supported by conservative codon substitutions across the phylogeny ( Lin et al . , 2007; Jungreis et al . , 2011 ) , but a far larger set is not conserved between species . In aggregate , this non-conserved group shows weak but statistically significant signals of selection among fifty wild-type individuals of D . melanogaster ( Figure 6C ) , suggesting that a fraction of this group is undergoing protein coding selection . The remainder might include many other different groups of extensions: namely , a group of extensions undergoing diversifying selection , a group of deleterious extensions undergoing counterselection , and a group of selectively-neutral extensions subject to genetic drift . Finally , our model predicts that conserved extensions should on average exhibit higher readthrough rates than novel extensions , because only a subset of the latter group would have been selected for function and regulation . Our data is also consistent with this prediction: the median readthrough rate for the conserved extensions in Drosophila is 5 . 2% , while the median for the novel extensions is 1 . 7% . Notably , 62% of the novel extensions we identified in Drosophila , 95% of the extensions in human foreskin fibroblasts , and 40% of the extensions in yeast undergo readthrough at rates comparable to those of phylogenetically conserved extensions ( Figure 5F ) , arguing for the importance of these subsets . Broadly , our work builds upon the growing amount of evidence that eukaryotic genomes and proteomes are far more plastic than previously thought , particularly with regard to translation and coding . In addition to the large number C-terminal extensions we report , various groups have used ribosome profiling to determine that large numbers of genes are regulated by uORFs that initiate at near-cognate start codons ( Ingolia et al . , 2009; Brar et al . , 2012 ) , that many genes can be N-terminally extended in a regulated manner ( Fritsch et al . , 2012 ) , and even that many parts of mammalian genomes are decoded in multiple frames ( Michel et al . , 2012 ) . Given the preeminence of Drosophila as a developmental model and the abundance of conditional genetic tools available , we anticipate that ribosome profiling in Drosophila will be useful in deciphering the biological roles of not only readthrough , but all non-canonical translation events , throughout development . Wild-type ( y w ) flies were cultured according to standard procedures . S2 cells were cultured in Schneider’s ( Gibco by Life Technologies , Carlsbad , California ) media supplemented with 10% heat-inactivated FBS ( UCSF cell culture facility , San Francisco , California ) and antibiotics ( UCSF cell culture facility ) . S2 cells were transfected using Effectene reagent ( Qiagen , the Netherlands ) following the manufacturer’s instructions . For stable transfectants , the plasmid of interest was co-transfected at a 10:1 molar excess with pCoPuro . Stable integrants were selected and maintained in Schneider’s media supplemented as above , but additionally containing 10 μg/ml puromycin . Concentrations of total RNA in lysates were determined using the RiboGreen kit ( Molecular Probes by Life Technologies ) . For each sample , 35–100 μg total RNA was diluted 2:1 in digestion buffer ( 50 mM Tris pH 7 . 5 , 5 mM MgCl2 , 0 . 5% Triton x-100 , 1 mM DTT , 20 U/ml SuperaseIn , 20 μg/ml emetine , 15 mM CaCl2 , and 3 U micrococcal nuclease [Roche Applied Science , Indianapolis , Indiana] per μg of total RNA in the sample ) , to bring the final concentration of NaCl to 100 mM and CaCl2 to 5 mM . Samples were digested for 40 min at 25°C in a Thermomixer ( Eppendorf , Hamburg , Germany ) . Digestions were quenched by adding EGTA to a final concentration of 6 . 25 mM and placing the reactions on ice . 1 U MNase is defined as previously ( Oh et al . , 2011 ) as an increase of 0 . 005 A260 per min , measured in a Spectramax M2 plate reader ( Molecular Devices , Sunnyvale , California ) using 10 μg/ml salmon sperm DNA ( Sigma-Aldrich ) with 5 mM Ca+ and 20 mM Tris , pH 8 . 0 in a 0 . 1 ml reaction at 25°C . 10–50% sucrose gradients were prepared in polysome gradient buffer ( 250 mM NaCl , 15 mM MgCl2 , 20 U/ml SuperaseIn , 20 μg/ml emetine ) using a GradientMaster ( Biocomp Instruments , Fredericton , New Brunswick , Canada ) in polyclear centrifuge tubes ( Seton Scientific , Petaluma , California ) . Up to 200 µl of samples was applied to the top of each gradient . Gradients were resolved by spinning for 3 hr at 35 krpm at 4°C in an SW-41 rotor ( Beckmann Coulter , Brea , California ) , and fractionated using the GradientMaster . When appropriate , monosome fractions were collected , flash-frozen in liquid nitrogen , and stored at −80°C . Up to 0 . 5 ml of digested sample was layered atop 1 . 0 ml of a solution of 34% sucrose in polysome gradient buffer . Monosomes were sedimented by spinning for 4 hr at 70 krpm at 4°C in a TLA-110 rotor ( Beckmann Coulter ) . Pellets were resuspended in 600 µl 10 mM Tris , pH 7 . 0 and stored at −20°C . Lysates were prepared and footprinted as above . Unless otherwise indicated , monosomes were enriched by sedimentation through 34% sucrose cushions and resuspended in 600 µl 10 mM Tris , pH 7 . 0 . Resuspended monosomes were extracted once with 700 µl 65°C acid phenol and 40 µl 10% SDS , followed by 650 µl acid phenol and a final extraction with chloroform . RNA was precipitated for at least 2 hr at −30°C , resuspended in 10 mM Tris , pH 7 . 0 , and quantitated on a NanoDrop spectrophotometer ( Thermo Scientific , Asheville , North Carolina ) . 5–35 μg RNA was dephosphorylated for 1 hr at 37°C using T4 polynucleotide kinase ( New England Biolabs , Ipswich , Massachusetts ) in a 50 µl reaction and resolved on a 15% TBE-urea gel ( Invitrogen by Life Technologies ) . A gel slab spanning 28–34 nt ( as measured by oligoribonucleotide size standards in a neighboring lane; see Supplementary file 2 ) was excised from the gel , eluted , and precipitated . Samples were then carried through all steps of library generation ( see below ) . For each sample , 375 µl of undigested polysome lysate was diluted into 3 vol Trizol LS ( Invitrogen ) and total RNA was extracted following the manufacturer’s instructions . 20–50 μg Poly ( A ) + RNA was selected on oligo-dT25 DynaBeads ( Invitrogen ) per manufacturer’s instructions , and fragmented at 95°C in fragmentation buffer ( 2 mM EDTA , 100 mM NaCO3/NaHCO3 , pH 9 . 2 ) to a mean size of roughly 100 nt . Fragmented RNA was precipitated , dephosphorylated for 1 hr at 37°C with T4 polynucleotide kinase ( New England Biolabs ) , and resolved on a 15% TBE-urea gel . A gel slab corresponding to 55–65 nt was excised from the gel , eluted , and precipitated . Samples were then carried through all steps of library generation ( see below ) . We performed two sequential rounds of subtractive hybridization on each sample . To 5 µl cDNA the following were added: 1 µl 20× SSC , 3 µl nuclease-free water , and 1 µl of a 60 μM mixture of the biotinylated oligonucleotides oJGD132 , oJGD133 , oJGD134 , oJGD135 , oJGD136 , oJGD161 , oJGD162 , oJGD163 , and oJGD164 ( sequences in Supplementary file 2 ) mixed in a ratio of 25 . 5:1:13:17:4:6:2:11:21 . Samples were denatured for 90 s at 95°C and annealed for 20 min at 25°C . MyOne Streptavidin C1 DynaBeads ( Invitrogen ) were prepared as follows: for each sample , 45 µl of beads were aliquoted into a microcentrifuge tube and washed three times in 50 µl 2 × binding buffer ( 10 mM Tris , pH 7 . 5 , 1 mM EDTA , 2 M NaCl ) , and resuspended in 22 . 5 µl 2 × binding buffer . 10 µl equilibrated beads were added to 10 µl hybridized sample . The mixture was incubated at 20 min in a room temperature Thermomixer with shaking at 850 rpm . Beads were then separated on a magnetic manifold ( Invitrogen ) and the supernatant recovered to a microcentrifuge tube . For the second round of subtraction , 1 µl 60 μM biotinylated oligo mix and 1 µl 20X SSC were added to the supernatant from the first subtraction , and the denaturation and annealing repeated . 10 µl of equilibrated beads were pelleted on a magnetic manifold . The buffer was removed , and the beads resuspended in the mixture from the second hybridization . Samples were then incubated at 20 min in a room temperature Thermomixer with shaking at 850 rpm . The supernatant was recovered on a magnetic manifold , transferred to a microcentrifuge tube , precipitated , and resuspended in 15 µl 10 mM Tris , pH 8 . 0 . RNA concentrations were measured using the Small RNA Series II Bioanalyzer assay ( Agilent Technologies , Santa Clara , California ) . 10–15 picomoles of RNAs were ligated to 1 μg 3′ miRNA cloning linker 1 ( Integrated DNA Technologies , Coralvaille , Iowa ) for 2 hr 30 min at 25°C in ligase buffer ( 1× T4 RNA ligase 2 buffer [New England Biolabs] , 40% PEG-100 [Sigma-Aldrich] , 5% DMSO , T4 RNA ligase 2 K227Q , truncated [a kind gift from Calvin Jan] ) in a 20 µl reaction . Ligated fragments were precipitated for at least 2 hr at −30°C , purified on a 10% TBE-urea gel , eluted , and precipitated . Ligation products were then reverse-transcribed using SuperScript III ( Invitrogen ) in a 16 . 7 µl reaction using using the primer o225-link1 ( see Supplementary file 2 ) . RNA template was hydrolyzed by addition of 1/10 vol 1 M NaOH and incubation at 95°C for 20 min cDNAs were purified on a 10% TBE-urea gel ( Invitrogen ) , eluted , precipitated , and resuspended in 5 µl 10 mM Tris pH 7 . 0 . cDNAs from footprint samples were subjected to two rounds of subtractive hybridization as described above . Subtracted samples were circularized using CircLigase ( Epicentre , Madison , Wisconsin ) , following manufacturer’s instructions in a 20 µl reaction . An additional microliter of CircLigase was then added , and the circularization repeated a second time . Circularized libraries were amplified by 6–12 cycles of PCR using oNTI231 and any of four indexing primers oCJ30–33 ( Supplementary file 2 ) using Phusion polymerase ( Finnzymes by ThermoScientific ) in a 17 µl reaction . Amplification products were size-selected on 8% TBE gels ( Invitrogen ) , eluted , precipitated , and resuspended in 10 µl 10 mM Tris , pH 8 . 0 . Samples were then quantitated using the Bioanalyzer High Sensitivity DNA assay ( Agilent Technologies ) , diluted to 2 nM , multiplexed as needed , and subjected to 50–57 cycles of single-end sequencing on an Illumina HiSeq sequencer ( Illumina , San Diego , CA ) using version 3 clustering and sequencing kits with a 6-cycle index read ( Illumina ) . For all Drosophila experiments we used revision 5 . 43 of the FlyBase genome annotation and the corresponding genome assembly ( Marygold et al . , 2013 ) . Reads were demultiplexed and cleaned of 3′ cloning adapters using in-house scripts . Reads shorter than 25 nt were discarded . Remaining reads were aligned using Bowtie version 0 . 12 . 7 ( Langmead et al . , 2009 ) sequentially to Bowtie indices composed of the following sequences: ( a ) D . melanogaster rRNAs ( GenBank accession #M21017 [Tautz et al . , 1988] and from FlyBase ) , ( b ) D . melanogaster tRNAs , snoRNAs , and snRNAs ( from FlyBase ) , ( c ) cloning oligos , ( d ) the S288C yeast genome version R64-1-1 ( downloaded on 6 June 2011 from http://downloads . yeastgenome . org/sequence/S288C_reference/genome_releases/ ) , ( e ) Wolbachia ( GenBank accession #AE017196 ) , ( f ) D . melanogaster chromosome arms , and ( g ) splice junctions ( from FlyBase and , in the case of embryos—figure supplemented with junctions discovered in the pooled embryo mRNA datasets using HMMSplicer 0 . 95 [Dimon et al . , 2010] ) . For all quantitative analyses , we counted only uniquely-mapped reads . Alignments were assigned to genomic coordinates as follows . Randomly-fragmented poly ( A ) + mRNA alignments were counted along the entire length of the alignment . Each genomic position covered by a single RNA fragment was incremented 1/l , where l corresponds to the length of the alignment . Ribosome-protected footprint alignments were mapped to their estimated P-sites as follows: 12 nt were pruned from each end of the alignment , leaving a fragment n nt long ( where n = l − 2 × 12 ) . Each genomic position covered by a nucleotide remaining in the pruned alignment was then incremented by 1/n . Thus , the P-site of each 25 mer was assigned to one unique position , while the P-site of each 26-mer was spread over two positions , each incremented by 0 . 5 reads , and so on . Alignment statistics are given in Supplementary file 1B . mRNA abundance and ribosome density for each genomic feature were measured in reads per kilobase of feature length per million reads aligning to chromosomes or splice junctions in the dataset ( RPKM ) , a unit which corrects for both feature length and sequencing depth . Unless otherwise indicated , the RPKM values we report for mRNA abundance reflect the total number of RNA fragments aligning to all countable exonic positions for a given locus . For ribosome density , we report the total number of ribosome-protected footprint fragments aligning to all countable positions of a coding region ( CDS ) for a given locus . We calculate translation efficiency as the ratio of footprint RPKM in the CDS to the RNA fragment RPKM across the entire locus . When comparing mRNA fragment or footprint density between samples , we restricted our analyses to genes that had at least 128 summed counts between replicates as determined in Figure 1—figure supplement 2 . When comparing translation efficiencies between samples , we required at least 128 exonic counts of mRNA for each gene . Translation efficiencies for these regions were calculated as the ratio of footprint counts to mRNA counts in each region , for all regions with at least 128 mRNA counts . We excluded all positions that could be labeled as two or more of 5′ UTR , CDS , or 3′ UTR depending upon transcript isoform . To remove variability or bleedthrough introduced by start and stop codon peaks , we additionally excluded the following genomic positions from consideration: 9 nucleotides preceding each start codon , 15 nucleotides following each start codon , the 15 nucleotides preceding each stop codon , and the 15 nucleotides following each stop codon . For each analyses ( Figure 2—figure supplement 1 , Figure 4B ) , we identified regions of interest ( ROIs ) germane to the analysis . In Figure 2—figure supplement 1 , these included roughly 3000 ROIs each for the left and right panels , each of which met the following criteria: ( a ) all transcripts deriving from that gene had one annotated start codon ( left panel ) or stop codon ( right panel ) , ( b ) all transcripts deriving from that locus covered identical genomic positions over the region of interest ( ROI ) shown , ( c ) all positions within the ROI were non-degenerate ( see ‘Materials and methods’ ) , and ( d ) at least 10 reads were present in the coding subregion of the ROI . For coding regions in Figure 4C , we kept the same criteria as above but required only 0 . 5 reads in the coding subregion of each ROI , yielding roughly 7401 ROI for that set . For C-terminal extensions , we required only that the extension be long enough to cover the interval shown , and have 0 . 5 reads in the coding subregion , allowing us to include 123 of the 350 extensions . For each ROI , we then generated a ‘coverage vector’ tallying ribosome density at each nucleotide position . We then normalized each coverage vector to the mean number of footprint reads covering the annotated coding region in the ROI , excluding a 3-codon buffer flanking the start or stop codon to avoid bleedthrough from initiation or termination peaks . We then plotted the median value across all normalized coverage vectors at each position . To improve our sensitivity in detection , we re-aligned our footprint and mRNA datasets to a Bowtie database of spliced transcript models , allowing three mismatches ( where we previously only allowed two ) . For the first , second , and third nucleotide position in each unique , annotated stop codon , we counted the number of matching and mismatching nucleotides in each read alignment covering that position . We ignored mismatches that occurred in the first position of the read alignment , because they frequently arise from non-templated nucleotide addition by reverse transcriptase . Considering the first , second , and third positions of each stop codon separately , we calculated a global average mismatch frequency for each . We then searched for individual stop codon positions that far exceeded the corresponding global average using a binomial test , controlling the false discovery rate at 5% following the procedure of Benjamini and Hochberg ( Benjamini and Hochberg , 1995 ) . We performed this analysis separately upon each of three datasets: total mRNA , total footprints , and the subset of footprints whose P-sites had passed the nucleotide position in question , following the P-site assignment rules described above . 4–48 ml of transiently or stably transfected cells were harvested 48 or 72 hr post-transfection by centrifuging for 2 min at 1600 rpm in a tabletop centrifuge . All cell pellets were rinsed once in PBS , and flash-frozen in a bath of dry ice in ethanol . Cell pellets were thawed and lysed for at least 15 min on ice in 0 . 5–1 . 5 ml lysis buffer ( 150 mM NaCl , 50 mM Tris pH 7 . 5 , 1% Triton x-100 , 1 mM EDTA and 1× complete protease inhibitor cocktail [Roche Applied Science] ) , depending upon the pellet size . Lysates were clarified by spinning 10 min at 20 , 000 × g in a microcentrifuge and supernatants collected . GFP reporters were immunoprecipitated on 10 μl of anti-GFP beads ( Chromotek , Planegg-Martinsried , Germany ) equilibrated in IP wash buffer ( 150 mM NaCl , 50 mM Tris pH 7 . 5 , 1 mM EDTA , 0 . 05% Triton x-100 ) . The bound fraction was washed three times in IP wash buffer , and finally eluted by boiling for at least 5 min in NuPage sample loading buffer ( Invitrogen ) . Supernatants were collected and transferred to new tubes , and stored at −20°C . For western blotting , samples were resolved on 4–12% NuPage gels ( Invitrogen ) in MOPS buffer . Gel lanes were loaded such that the amounts of uncleaved GFP reporter in each lane were loaded as equally as possible . GFP was detected using a mouse anti-GFP antibody ( Roche Applied Sciences ) , and visualized using IR800 anti-mouse antibodies on a LI-COR Odyssey system ( LI-COR , Lincoln , Nebraska ) . FLAG was similarly detected on a separate gel , instead using the M2 Mouse anti-FLAG antibody ( Sigma-Aldrich ) . PhyloCSF analysis was performed on all C-terminal extensions at least five codons long , exclusive of the stop codons . Multiple species alignments were obtained from the Drosophila 12-way multispecies alignment as downloaded from the UCSC genome browser , and stitched together over regions of interest using the Phast utility maf_parse ( Hubisz et al . , 2011 ) . PhyloCSF was then used to evaluate the extension on the empirical codon model ‘12flies’ . Columns in which the D . melanogaster sequence contained gaps were ignored . Alignments that contained no sequence besides that from D . melanogaster were not evaluated . We calculated the 189-variable Z-curve as previously described ( Gao and Zhang , 2004 ) . We empirically determined that the classifier became error-prone if trained on sequences 81 nt or shorter in length . Our training set consisted of 81-nucleotide windows drawn from coding regions ( the positive set , 14 , 507 windows ) or from portions of distal 3′ UTRs that did not overlap annotated coding regions or 5′ UTRs ( the negative set , 8151 windows ) . To assay the stability of the classifier’s behavior and control for overfitting , we trained the classifier with fourfold cross-validation training on 2200 windows from the CDS set and 2200 windows from the distal 3′ UTR set , yielding an average misclassification error of 6 . 9–7 . 3% with each iteration . We repeated this analysis ( and cross-validation ) several times selecting different 81-nucleotide windows from each CDS and distal 3′ UTR , obtaining similar levels of error . The classifier was then trained on the entire training set , and used to evaluate randomly chosen 81 nt windows from observed C-terminal protein extensions that were 81 nt or greater in length . These included 26 extensions predicted by Jungreis et al . and 83 novel extensions . We downloaded SNP data from the Drosophila Population Genomics Project from Ensembl . org ( release 67; Flicek et al . , 2013 ) and counted the proportion of SNPs that , if translated in frame , would cause synonymous substitutions in coding regions , extensions , and distal 3′ UTRs . To test C-terminal extensions for differential readthrough rates , we examined all extensions which met the following criteria: ( 1 ) all annotated isoforms covering the extension contain exactly the same CDS , ( 2 ) the CDS had at least 128 total footprint reads in each of the S2 cell and embryo samples , and ( 3 ) the C-terminal extension had been scored as positive for readthrough in either the S2 and/or the embryo sample . For those extensions , we tabulated the footprint reads that aligned to the CDS and putative extension , masking out regions normally covered by start and stop codon peaks as described ( see section ‘Readthrough rates’ , above ) . For each extension , this tabulation yielded a 2 × 2 contingency table of reads aligning to the CDS and extension in the S2 cell and 0–2 hr embryo datasets . We evaluated the statistical significance of asymmetry in the contingency tables using Fisher’s exact test , and controlled the false discovery rate at 5% using the procedure of Benjamini and Hochberg ( Benjamini and Hochberg , 1995 ) . For human cells , we collected data from uninfected human foreskin fibroblasts and processed it as previously described ( Stern-Ginossar et al . , 2012 ) . Yeast samples were collected from [psi−] W303 and processed as previously described ( Ingolia et al . , 2009 ) , with the exception that a 3′ linker ligation strategy was used instead of poly ( A ) tailing for fragment capture . For phasing of yeast footprints , we counted only 28-mers , which have previously been shown to be the best-phased footprint population in that organism ( Ingolia et al . , 2009 ) . C-terminal extensions 20 amino acids or longer were scanned for transmembrane domains using TmHmm ( Krogh et al . , 2001 ) using default settings . Nuclear localization signals were predicted in extensions 20 amino acids or longer using the cNLS mapper ( Kosugi et al . , 2009 ) , with a score cutoff of 7 . 0 . Peroxisome targeting signals were predicted for all extensions 12 amino acids or longer using PTS1 Predictor ( Neuberger et al . , 2003 ) with the signal type set to ‘metazoan’ . Prenylation signals were predicted for all extensions 12 amino acids or longer using PrePS ( Maurer-Stroh and Eisenhaber , 2005 ) . In addition , we searched for ER retention signals using the consensus [KH]DEL . We searched 3′ UTRs ( including the predicted extension and entire distal 3′ UTR ) for selenocysteine insertion elements using SeciSearch 2 . 19 ( Kryukov et al . , 2003 ) with parameters set as follows: e1 = 05 , e2 = −22 , Y_filter = True , O_filter = True , B_filter = True , S_filter = True . We searched each 3′ UTR using every available SECIS Pattern ( pat_c , pat_Sep20 , pat_dm , pat_g , and pat_s ) , and considered a 3′ UTR receiving a COVE score above the recommended threshold of 15 in any of the pattern searches to contain a SECIS element . Additionally , we excluded any extensions that were annotated as selenoprotein annotations in SelenoDB ( Castellano et al . , 2008; for Drosophila , yeast , and human data ) or FlyBase ( Marygold et al . , 2013; for Drosophila ) , For transcripts with no annotated or short 3′ UTRs , we extended the 3′ UTR in uninterrupted genome coordinates until it was 1000 nucleotides in length , an in Jungreis et al . ( 2011 ) . S2 cells stably transfected with the reporter of interest were maintained at a density of 1 . 6–12 million cells/ml . Nuclei were visualized by staining with 1 μg/ml Hoechst 34580 ( Invitrogen ) for at least 5 min . Live cells were imaged in culture media on an inverted spinning disk confocal Nikon Ti microscope ( Nikon Instruments , Melville , NY ) in glass-bottom culture dishes ( MatTek , Ashland , MA ) . Images were contrast-adjusted and prepared for presentation in Adobe Photoshop ( Adobe Systems , San Jose , CA ) . Both the raw data ( as FastQ files ) and processed data ( wiggle files ) are available in NCBI’s Gene Expression Omnibus ( Edgar et al . , 2002 ) under GEO series accession number GSE49197 ( http://www . ncbi . nlm . nih . gov/geo/query/acc . cgi ? acc=GSE49197 ) . Supplementary tables 1–4 are available at Dryad ( Dunn et al . , 2013; http://dx . doi . org/10 . 5061/dryad . 6nr73 ) : Supplementary table 1: gene expression measurements in 0–2 hr embryos and S2 cells . Source data for Figures 1 and 2 , as well as their supplements . Supplementary table 2: readthrough statistics for Drosophila melanogaster . Source data for Figures 3 , 4 and 6 , as well as their supplements , and annotations of readthrough events in Drosophila melanogaster . Supplementary table 3: readthrough statistics for Saccharomyces cerevisiae . Source data for Figure 5 and annotations of readthrough events in [psi-] W303 yeast . Supplementary table 4: readthrough statistics for human foreskin fibroblasts . Source data for Figure 5 and annotations of readthrough events in human foreskin fibroblasts . Supplementary file 1 provides alignment statistics and Supplementary file 2 contains the oligonucleotides used in this study . We wrote custom scripts in Python 2 . 7 , using the following open-source libraries: Numpy 1 . 6 . 0 ( http://numpy . scipy . org ) , Scipy 0 . 11 . 0rc2 ( http://www . scipy . org ) , Biopython 1 . 59 ( Cock et al . , 2009 ) , PySam , and HTSeq 0 . 5 . 1p2 ( http://www-huber . embl . de/users/anders/HTSeq/doc/overview . html ) . Plots and genome browser snapshots were generated using Matplotlib 1 . 0 . 1 ( Hunter , 2007 ) .	For a gene to give rise to a protein , its DNA is first used as a template to produce a messenger RNA molecule . Each group of three nucleotides within the messenger RNA encodes an amino acid , and structures called ribosomes assemble the protein by joining together amino acids in the correct order . The nucleotide triplets are called codons , and some are known as stop codons because they typically instruct the ribosome to stop adding amino acids . Sometimes ribosomes interpret stop codons as amino acid insertion signals , giving rise to an extended protein with a modified structure or function . This phenomenon is known as stop codon readthrough , and is required for many viruses to complete their reproductive cycles . However , much less is known about stop codon readthrough in other organisms . Now , Dunn et al . have used a technique called ribosome profiling to analyze stop codon readthrough across the entire genome of the fruit fly Drosophila melanogaster . An enzyme was used to fragment messenger RNA , and those fragments that were specifically engaged by ribosomes—and thus likely to encode protein—were sequenced . Stop codon readthrough occurred much more often than had been expected based on previous studies . Indeed , computational analysis strongly suggests that evolution has favored this process for certain fruit fly genes . Moreover , stop codon readthrough was also observed in yeast and human cells , suggesting that it is important in many organisms , not just the fruit fly . Stop codon readthrough thus provides a novel way for organisms to tune the expression levels and functions of their genes , both throughout the lifetime of an individual , and the evolution of a species .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "evolutionary", "biology", "cell", "biology" ]	2013	Ribosome profiling reveals pervasive and regulated stop codon readthrough in Drosophila melanogaster
Deep learning has led to significant advances in artificial intelligence , in part , by adopting strategies motivated by neurophysiology . However , it is unclear whether deep learning could occur in the real brain . Here , we show that a deep learning algorithm that utilizes multi-compartment neurons might help us to understand how the neocortex optimizes cost functions . Like neocortical pyramidal neurons , neurons in our model receive sensory information and higher-order feedback in electrotonically segregated compartments . Thanks to this segregation , neurons in different layers of the network can coordinate synaptic weight updates . As a result , the network learns to categorize images better than a single layer network . Furthermore , we show that our algorithm takes advantage of multilayer architectures to identify useful higher-order representations—the hallmark of deep learning . This work demonstrates that deep learning can be achieved using segregated dendritic compartments , which may help to explain the morphology of neocortical pyramidal neurons . Deep learning refers to an approach in artificial intelligence ( AI ) that utilizes neural networks with multiple layers of processing units . Importantly , deep learning algorithms are designed to take advantage of these multi-layer network architectures in order to generate hierarchical representations wherein each successive layer identifies increasingly abstract , relevant variables for a given task ( Bengio and LeCun , 2007; LeCun et al . , 2015 ) . In recent years , deep learning has revolutionized machine learning , opening the door to AI applications that can rival human capabilities in pattern recognition and control ( Mnih et al . , 2015; Silver et al . , 2016; He et al . , 2015 ) . Interestingly , the representations that deep learning generates resemble those observed in the neocortex ( Kubilius et al . , 2016; Khaligh-Razavi and Kriegeskorte , 2014; Cadieu et al . , 2014 ) , suggesting that something akin to deep learning is occurring in the mammalian brain ( Yamins and DiCarlo , 2016; Marblestone et al . , 2016 ) . Yet , a large gap exists between deep learning in AI and our current understanding of learning and memory in neuroscience . In particular , unlike deep learning researchers , neuroscientists do not yet have a solution to the ‘credit assignment problem’ ( Rumelhart et al . , 1986; Lillicrap et al . , 2016; Bengio et al . , 2015 ) . Learning to optimize some behavioral or cognitive function requires a method for assigning ‘credit’ ( or ‘blame’ ) to neurons for their contribution to the final behavioral output ( LeCun et al . , 2015; Bengio et al . , 2015 ) . The credit assignment problem refers to the fact that assigning credit in multi-layer networks is difficult , since the behavioral impact of neurons in early layers of a network depends on the downstream synaptic connections . For example , consider the behavioral effects of synaptic changes , that is long-term potentiation/depression ( LTP/LTD ) , occurring between different sensory circuits of the brain . Exactly how these synaptic changes will impact behavior and cognition depends on the downstream connections between the sensory circuits and motor or associative circuits ( Figure 1A ) . If a learning algorithm can solve the credit assignment problem then it can take advantage of multi-layer architectures to develop complex behaviors that are applicable to real-world problems ( Bengio and LeCun , 2007 ) . Despite its importance for real-world learning , the credit assignment problem , at the synaptic level , has received little attention in neuroscience . The lack of attention to credit assignment in neuroscience is , arguably , a function of the history of biological studies of synaptic plasticity . Due to the well-established dependence of LTP and LTD on presynaptic and postsynaptic activity , current theories of learning in neuroscience tend to emphasize Hebbian learning algorithms ( Dan and Poo , 2004; Martin et al . , 2000 ) , that is , learning algorithms where synaptic changes depend solely on presynaptic and postsynaptic activity . Hebbian learning models can produce representations that resemble the representations in the real brain ( Zylberberg et al . , 2011; Leibo et al . , 2017 ) and they are backed up by decades of experimental findings ( Malenka and Bear , 2004; Dan and Poo , 2004; Martin et al . , 2000 ) . But , current Hebbian learning algorithms do not solve the credit assignment problem , nor do global neuromodulatory signals used in reinforcement learning ( Lillicrap et al . , 2016 ) . As a result , deep learning algorithms from AI that can perform multi-layer credit assignment outperform existing Hebbian models of sensory learning on a variety of tasks ( Yamins and DiCarlo , 2016; Khaligh-Razavi and Kriegeskorte , 2014 ) . This suggests that a critical , missing component in our current models of the neurobiology of learning and memory is an explanation of how the brain solves the credit assignment problem . However , the most common solution to the credit assignment problem in AI is to use the backpropagation of error algorithm ( Rumelhart et al . , 1986 ) . Backpropagation assigns credit by explicitly using current downstream synaptic connections to calculate synaptic weight updates in earlier layers , commonly termed ‘hidden layers’ ( LeCun et al . , 2015 ) ( Figure 1B ) . This technique , which is sometimes referred to as ‘weight transport’ , involves non-local transmission of synaptic weight information between layers of the network ( Lillicrap et al . , 2016; Grossberg , 1987 ) . Weight transport is clearly unrealistic from a biological perspective ( Bengio et al . , 2015; Crick , 1989 ) . It would require early sensory processing areas ( e . g . V1 , V2 , V4 ) to have precise information about billions of synaptic connections in downstream circuits ( MT , IT , M2 , EC , etc . ) . According to our current understanding , there is no physiological mechanism that could communicate this information in the brain . Some deep learning algorithms utilize purely Hebbian rules ( Scellier and Bengio , 2016; Hinton et al . , 2006 ) . But , they depend on feedback synapses that are symmetric to feedforward synapses ( Scellier and Bengio , 2016; Hinton et al . , 2006 ) , which is essentially a version of weight transport . Altogether , these artificial aspects of current deep learning solutions to credit assignment have rendered many scientists skeptical of the proposal that deep learning occurs in the real brain ( Crick , 1989; Grossberg , 1987; Harris , 2008; Urbanczik and Senn , 2009 ) . Recent findings have shown that these problems may be surmountable , though . Lillicrap et al . ( 2016 ) , Lee et al . , 2015 and Liao et al . , 2015 have demonstrated that it is possible to solve the credit assignment problem even while avoiding weight transport or symmetric feedback weights . The key to these learning algorithms is the use of feedback signals that convey enough information about credit to calculate local error signals in hidden layers ( Lee et al . , 2015; Lillicrap et al . , 2016; Liao et al . , 2015 ) . With this approach it is possible to take advantage of multi-layer architectures , leading to performance that rivals backpropagation ( Lee et al . , 2015; Lillicrap et al . , 2016; Liao et al . , 2015 ) . Hence , this work has provided a significant breakthrough in our understanding of how the real brain might do credit assignment . Nonetheless , the models of Lillicrap et al . ( 2016 ) , Lee et al . , 2015 and Liao et al . , 2015 involve some problematic assumptions . Specifically , although it is not directly stated in all of the papers , there is an implicit assumption that there is a separate feedback pathway for transmitting the information that determines the local error signals ( Figure 2A ) . Such a pathway is required in these models because the error signal in the hidden layers depends on the difference between feedback that is generated in response to a purely feedforward propagation of sensory information , and feedback that is guided by a teaching signal ( Lillicrap et al . , 2016; Lee et al . , 2015; Liao et al . , 2015 ) . In order to calculate this difference , sensory information must be transmitted separately from the feedback signals that are used to drive learning . In single compartment neurons , keeping feedforward sensory information separate from feedback signals is impossible without a separate pathway . At face value , such a pathway is possible . But , closer inspection uncovers a couple of difficulties with such a proposal . First , the error signals that solve the credit assignment problem are not global error signals ( like neuromodulatory signals used in reinforcement learning ) . Rather , they are cell-by-cell error signals . This would mean that the feedback pathway would require some degree of pairing , wherein each neuron in the hidden layer is paired with a feedback neuron ( or circuit ) . That is not impossible , but there is no evidence to date of such an architecture in the neocortex . Second , the error signal in the hidden layer is signed ( i . e . it can be positive or negative ) , and the sign determines whether LTP or LTD occur in the hidden layer neurons ( Lee et al . , 2015; Lillicrap et al . , 2016; Liao et al . , 2015 ) . Communicating signed signals with a spiking neuron can theoretically be done by using a baseline firing rate that the neuron can go above ( for positive signals ) or below ( for negative signals ) . But , in practice , such systems are difficult to operate with a single neuron , because as the error gets closer to zero any noise in the spiking of the neuron can switch the sign of the signal , which switches LTP to LTD , or vice versa . This means that as learning progresses the neuron’s ability to communicate error signs gets worse . It would be possible to overcome this by using many neurons to communicate an error signal , but this would then require many error neurons for each hidden layer neuron , which would lead to a very inefficient means of communicating errors . Therefore , the real brain’s specific solution to the credit assignment problem is unlikely to involve a separate feedback pathway for cell-by-cell , signed signals to instruct plasticity . However , segregating the integration of feedforward and feedback signals does not require a separate pathway if neurons have more complicated morphologies than the point neurons typically used in artificial neural networks . Taking inspiration from biology , we note that real neurons are much more complex than single-compartments , and different signals can be integrated at distinct dendritic locations . Indeed , in the primary sensory areas of the neocortex , feedback from higher-order areas arrives in the distal apical dendrites of pyramidal neurons ( Manita et al . , 2015; Budd , 1998; Spratling , 2002 ) , which are electrotonically very distant from the basal dendrites where feedforward sensory information is received ( Larkum et al . , 1999; 2007; 2009 ) . Thus , as has been noted by previous authors ( Körding and König , 2001; Spratling , 2002; Spratling and Johnson , 2006 ) , the anatomy of pyramidal neurons may actually provide the segregation of feedforward and feedback information required to calculate local error signals and perform credit assignment in biological neural networks . Here , we show how deep learning can be implemented if neurons in hidden layers contain segregated ‘basal’ and ‘apical’ dendritic compartments for integrating feedforward and feedback signals separately ( Figure 2B ) . Our model builds on previous neural networks research ( Lee et al . , 2015; Lillicrap et al . , 2016 ) as well as computational studies of supervised learning in multi-compartment neurons ( Urbanczik and Senn , 2014; Körding and König , 2001; Spratling and Johnson , 2006 ) . Importantly , we use the distinct basal and apical compartments in our neurons to integrate feedback signals separately from feedforward signals . With this , we build a local error signal for each hidden layer that ensures appropriate credit assignment . We demonstrate that even with random synaptic weights for feedback into the apical compartment , our algorithm can coordinate learning to achieve classification of the MNIST database of hand-written digits that is better than that which can be achieved with a single layer network . Furthermore , we show that our algorithm allows the network to take advantage of multi-layer structures to build hierarchical , abstract representations , one of the hallmarks of deep learning ( LeCun et al . , 2015 ) . Our results demonstrate that deep learning can be implemented in a biologically feasible manner if feedforward and feedback signals are received at electrotonically segregated dendrites , as is the case in the mammalian neocortex . Deep supervised learning with local weight updates requires that each neuron receive signals that can be used to determine its ‘credit’ for the final behavioral output . We explored the idea that the cortico-cortical feedback signals to pyramidal cells could provide the required information for credit assignment . In particular , we were inspired by four observations from both machine learning and biology: With these considerations in mind , we hypothesized that the computations required for credit assignment could be achieved without separate pathways for feedback signals . Instead , they could be achieved by having two distinct dendritic compartments in each hidden layer neuron: a ‘basal’ compartment , strongly coupled to the soma for integrating bottom-up sensory information , and an ‘apical’ compartment for integrating top-down feedback in order calculate credit assignment and drive synaptic plasticity via ‘plateau potentials’ ( Bittner et al . , 2015; Bittner et al . , 2017 ) ( Figure 3A ) . As an initial test of this concept we built a network with a single hidden layer . Although this network is not very ‘deep’ , even a single hidden layer can improve performance over a one-layer architecture if the learning algorithm solves the credit assignment problem ( Bengio and LeCun , 2007; Lillicrap et al . , 2016 ) . Hence , we wanted to initially determine whether our network could take advantage of a hidden layer to reduce error at the output layer . The network architecture is illustrated in Figure 3B . An image from the MNIST data set is used to set the spike rates of ℓ=784 Poisson point-process neurons in the input layer ( one neuron per image pixel , rates-of-fire determined by pixel intensity ) . These project to a hidden layer with m=500 neurons . The neurons in the hidden layer ( which we index with a ‘0’ ) are composed of three distinct compartments with their own voltages: the apical compartments ( with voltages described by the vector V0a ( t ) =[V10a ( t ) , . . . , Vm0a ( t ) ] ) , the basal compartments ( with voltages V0b ( t ) =[V10b ( t ) , . . . , Vm0b ( t ) ] ) , and the somatic compartments ( with voltages V0 ( t ) =[V10 ( t ) , . . . , Vm0 ( t ) ] ) . ( Note: for notational clarity , all vectors and matrices in the paper are in boldface . ) The voltage of the it⁢h neuron in the hidden layer is updated according to: ( 1 ) τdVi0 ( t ) dt=−Vi0 ( t ) +gbgl ( Vi0b ( t ) −Vi0 ( t ) ) +gagl ( Vi0a ( t ) −Vi0 ( t ) ) where gl , gb and ga represent the leak conductance , the conductance from the basal dendrites , and the conductance from the apical dendrites , respectively , and τ=Cm/gl where Cm is the membrance capacitance ( see Materials and methods , Equation ( 16 ) ) . For mathematical simplicity we assume in our simulations a resting membrane potential of 0 mV ( this value does not affect the results ) . We implement electrotonic segregation in the model by altering the ga value—low values for ga lead to electrotonically segregated apical dendrites . In the initial set of simulations we set ga=0 , which effectively makes it a feed-forward network , but we relax this condition in later simulations . We treat the voltages in the dendritic compartments simply as weighted sums of the incoming spike trains . Hence , for the it⁢h hidden layer neuron: ( 2 ) Vi0b ( t ) =∑j=1ℓWij0sjinput ( t ) +bi0Vi0a ( t ) =∑j=1nYijsj1 ( t ) where Wi⁢j0 and Yi⁢j are synaptic weights from the input layer and the output layer , respectively , bi0 is a bias term , and sinput and s1 are the filtered spike trains of the input layer and output layer neurons , respectively . ( Note: the spike trains are convolved with an exponential kernel to mimic postsynaptic potentials , see Materials and methods Equation ( 11 ) . ) The somatic compartments generate spikes using Poisson processes . The instantaneous rates of these processes are described by the vector ϕ0 ( t ) =[ϕ10 ( t ) , . . . , ϕm0 ( t ) ] , which is in units of spikes/s or Hz . These rates-of-fire are determined by a non-linear sigmoid function , σ⁢ ( ⋅ ) , applied to the somatic voltages , that is for the it⁢h hidden layer neuron: ( 3 ) ϕi0 ( t ) =ϕmaxσ ( Vi0 ( t ) ) =ϕmax11+e−Vi0 ( t ) where ϕmax is the maximum rate-of-fire for the neurons . The output layer ( which we index here with a ‘1’ ) contains n=10 two-compartment neurons ( one for each image category ) , similar to those used in a previous model of dendritic prediction learning ( Urbanczik and Senn , 2014 ) . The output layer dendritic voltages ( V1b ( t ) =[V11b ( t ) , . . . , Vn1b ( t ) ] ) and somatic voltages ( V1 ( t ) =[V11 ( t ) , . . . , Vn1 ( t ) ] ) are updated in a similar manner to the hidden layer basal compartment and soma: ( 4 ) τdVi1 ( t ) dt=−Vi1 ( t ) +gdgl ( Vi1b ( t ) −Vi1 ( t ) ) +Ii ( t ) Vi1b ( t ) =∑j=1ℓWij1sj0 ( t ) +bi1 where Wi⁢j1 are synaptic weights from the hidden layer , s0 are the filtered spike trains of the hidden layer neurons ( see Equation ( 11 ) ) , gl is the leak conductance , gd is the conductance from the dendrites , and τ is given by Equation ( 16 ) . In addition to the absence of an apical compartment , the other salient difference between the output layer neurons and the hidden layer neurons is the presence of the term Ii⁢ ( t ) , which is a teaching signal that can be used to force the output layer to the correct answer . Whether any such teaching signals exist in the real brain is unknown , though there is evidence that animals can represent desired behavioral outputs with internal goal representations ( Gadagkar et al . , 2016 ) . ( See below , and Materials and methods , Equations ( 19 ) and ( 20 ) for more details on the teaching signal ) . In our model , there are two different types of computation that occur in the hidden layer neurons: ‘transmit’ and ‘plateau’ . The transmit computations are standard numerical integration of the simulation , with voltages evolving according to Equation ( 1 ) , and with the apical compartment electrotonically segregated from the soma ( depending on ga ) ( Figure 3C , left ) . In contrast , the plateau computations do not involve numerical integration with Equation ( 1 ) . Instead , the apical voltage is averaged over the most recent 20–30 ms period and the sigmoid non-linearity is applied to it , giving us ‘plateau potentials’ in the hidden layer neurons ( we indicate plateau potentials with α , see Equation ( 5 ) below , and Figure 3C , right ) . The intention behind this design was to mimic the non-linear transmission from the apical dendrites to the soma that occurs during a plateau potential driven by calcium spikes in the apical dendritic shaft ( Larkum et al . , 1999 ) , but in the simplest , most abstract formulation possible . Importantly , plateau potentials in our simulations are single numeric values ( one per hidden layer neuron ) that can be used for credit assignment . We do not use them to alter the network dynamics . When they occur , they are calculated , transmitted to the basal dendrite instantaneously , and then stored temporarily ( 0–60 ms ) for calculating synaptic weight updates . To train the network we alternate between two phases . First , during the ‘forward’ phase we present an image to the input layer without any teaching current at the output layer ( I ( t ) i=0 , ∀i ) . The forward phase occurs between times t0 to t1 . At t1 a plateau potential is calculated in all the hidden layer neurons ( αf=[α1f , . . . , αmf] ) and the ‘target’ phase begins . During this phase , which lasts until t2 , the image continues to drive the input layer , but now the output layer also receives teaching current . The teaching current forces the correct output neuron to its max firing rate and all the others to silence . For example , if an image of a ‘9’ is presented , then over the time period t1-t2 the ‘9’ neuron in the output layer fires at max , while the other neurons are silent ( Figure 4A ) . At t2 another set of plateau potentials ( αt=[α1t , . . . , αmt] ) are calculated in the hidden layer neurons . The result is that we have plateau potentials in the hidden layer neurons for both the end of the forward phase ( αf ) and the end of the target phase ( αt ) , which are calculated as: ( 5 ) αif=σ ( 1Δt1∫t1−Δt1t1Vi0a ( t ) dt ) αit=σ ( 1Δt2∫t2−Δt2t2Vi0a ( t ) dt ) where Δ⁢ts is a time delay used to allow the network dynamics to settle before integrating the plateau , and Δ⁢ti=ti- ( ti-1+Δ⁢ts ) ( see Materials and methods , Equation ( 22 ) and Figure 4A ) . Similar to how targets are used in deep supervised learning ( LeCun et al . , 2015 ) , the goal of learning in our network is to make the network dynamics during the forward phase converge to the same output activity pattern as exists in the target phase . Put another way , in the absence of the teaching signal , we want the activity at the output layer to be the same as that which would exist with the teaching signal , so that the network can give appropriate outputs without any guidance . To do this , we initialize all the weight matrices with random weights , then we train the weight matrices W0 and W1 using stochastic gradient descent on local loss functions for the hidden and output layers , respectively ( see below ) . These weight updates occur at the end of every target phase , that is the synapses are not updated during transmission . Like Lillicrap et al . ( 2016 ) , we leave the weight matrix Y fixed in its initial random configuration . When we update the synapses in the network we use the plateau potential values αf and αt to determine appropriate credit assignment ( see below ) . The network is simulated in near continuous-time ( except that each plateau is considered to be instantaneous ) , and the temporal intervals between plateaus are randomly sampled from an inverse Gaussian distribution ( Figure 4B , top ) . As such , the specific amount of time that the network is presented with each image and teaching signal is stochastic , though usually somewhere between 50–60 ms of simulated time ( Figure 4B , bottom ) . This stochasticity was not necessary , but it demonstrates that although the system operates in phases , the specific length of the phases is not important as long as they are sufficiently long to permit integration ( see Lemma 1 ) . In the data presented in this paper , all 60 , 000 images in the MNIST training set were presented to the network one at a time , and each exposure to the full set of images was considered an ‘epoch’ of training . At the end of each epoch , the network’s classification error rate on a separate set of 10 , 000 test images was assessed with a single forward phase for each image ( see Materials and methods ) . The network’s classification was judged by which output neuron had the highest average firing rate during these test image forward phases . It is important to note that there are many aspects of this design that are not physiologically accurate . Most notably , stochastic generation of plateau potentials across a population is not an accurate reflection of how real pyramidal neurons operate , since apical calcium spikes are determined by a number of concrete physiological factors in individual cells , including back-propagating action potentials , spike-timing and inhibitory inputs ( Larkum et al . , 1999 , 2007 , 2009 ) . However , we note that calcium spikes in the apical dendrites can be prevented from occurring via the activity of distal dendrite targeting inhibitory interneurons ( Murayama et al . , 2009 ) , which can synchronize pyramidal activity ( Hilscher et al . , 2017 ) . Furthermore , distal dendrite targeting interneurons can themselves can be rapidly inhibited in response to temporally precise neuromodulatory inputs ( Pi et al . , 2013; Pfeffer et al . , 2013; Karnani et al . , 2016; Hangya et al . , 2015; Brombas et al . , 2014 ) . Therefore , it is entirely plausible that neocortical micro-circuits would generate synchronized plateaus/bursts at punctuated periods of time in response to disinhibition of the apical dendrites governed by neuromodulatory signals that determine ‘phases’ of processing . Alternatively , oscillations in population activity could provide a mechanism for promoting alternating phases of processing and synaptic plasticity ( Buzsáki and Draguhn , 2004 ) . But , complete synchrony of plateaus in our hidden layer neurons is not actually critical to our algorithm—only the temporal relationship between the plateaus and the teaching signal is critical . This relationship itself is arguably plausible given the role of neuromodulatory inputs in dis-inhibiting the distal dendrites of pyramidal neurons ( Karnani et al . , 2016; Brombas et al . , 2014 ) . Of course , we are engaged in a great deal of speculation here . But , the point is that our model utilizes anatomical and functional motifs that are loosely analogous to what is observed in the neocortex . Importantly for the present study , the key issue is the use of segregated dendrites which permit an effective feed-forward dynamic , punctuated by feedback driven plateau potentials to solve the credit assignment problem . To solve the credit assignment problem without using weight transport , we had to define local error signals , or ‘loss functions’ , for the hidden layer and output layer that somehow took into account the impact that each hidden layer neuron has on the output of the network . In other words , we only want to update a hidden layer synapse in a manner that will help us make the forward phase activity at the output layer more similar to the target phase activity . To begin , we define the target firing rates for the output neurons , ϕ1∗=[ϕ11∗ , . . . , ϕn1∗] , to be their average firing rates during the target phase: ( 6 ) ϕi1∗=ϕi1¯t=1Δt2∫t1+Δtst2ϕi1 ( t ) dt ( Throughout the paper , we use ϕ* to denote a target firing rate and ϕ¯ to denote a firing rate averaged over time . ) We then define a loss function at the output layer using this target , by taking the difference between the average forward phase activity and the target: ( 7 ) L1≈\|\|ϕ1∗−ϕ1¯f\|\|22=\|\|ϕ1¯t−ϕ1¯f\|\|22=\|\|1Δt2∫t1+Δtst2ϕ1 ( t ) dt−1Δt1∫t0+Δtst1ϕ1 ( t ) dt\|\|22 ( Note: the true loss function we use is slightly more complex than the one formulated here , hence the ≈ symbol in Equation ( 7 ) , but this formulation is roughly correct and easier to interpret . See Materials and methods , Equation ( 23 ) for the exact formulation . ) This loss function is zero only when the average firing rates of the output neurons during the forward phase equals their target , that is the average firing rates during the target phase . Thus , the closer L1 is to zero , the more the network’s output for an image matches the output activity pattern imposed by the teaching signal , I⁢ ( t ) . Effective credit assignment is achieved when changing the hidden layer synapses is guaranteed to reduce L1 . To obtain this guarantee , we defined a set of target firing rates for the hidden layer neurons that uses the information contained in the plateau potentials . Specifically , in a similar manner to Lee et al . , 2015 , we define the target firing rates for the hidden layer neurons , ϕ0∗=[ϕ10∗ , . . . , ϕm0∗] , to be: ( 8 ) ϕi0∗=ϕi0¯f+αit−αif where αit and αif are the plateaus defined in Equation ( 5 ) . As with the output layer , we define the loss function for the hidden layer to be the difference between the target firing rate and the average firing rate during the forward phase: ( 9 ) L0≈\|\|ϕ0∗−ϕ0¯f\|\|22=\|\|ϕ0¯f+αt−αif−ϕ0¯f\|\|22=\|\|αt−αf\|\|22 ( Again , note the use of the ≈ symbol , see Equation ( 30 ) for the exact formulation . ) This loss function is zero only when the plateau at the end of the forward phase equals the plateau at the end of the target phase . Since the plateau potentials integrate the top-down feedback ( see Equation ( 5 ) ) , we know that the hidden layer loss function , L0 , is zero if the output layer loss function , L1 , is zero . Moreover , we can show that these loss functions provide a broader guarantee that , under certain conditions , if L0 is reduced , then on average , L1 will also be reduced ( see Theorem 1 ) . This provides our assurance of credit assignment: we know that the ultimate goal of learning ( reducing L1 ) can be achieved by updating the synaptic weights at the hidden layer to reduce the local loss function L0 ( Figure 5A ) . We do this using stochastic gradient descent at the end of every target phase: ( 10 ) ΔW1=−η0∂L1∂W1ΔW0=−η1∂L0∂W0 where ηi and ΔWi refer to the learning rate and update term for weight matrix Wi ( see Materials and methods , Equations ( 28 ) , ( 29 ) , ( 33 ) and ( 35 ) for details of the weight update procedures ) . Performing gradient descent on L1 results in a relatively straight-forward delta rule update for W1 ( see Equation ( 29 ) ) . The weight update for the hidden layer weights , W0 , is similar , except for the presence of the difference between the two plateau potentials αt−αf ( see Equation ( 35 ) ) . Importantly , given the way in which we defined the loss functions , as the hidden layer reduces L0 by updating W0 , L1 should also be reduced , that is hidden layer learning should imply output layer learning , thereby utilizing the multi-layer architecture . To test that we were successful in credit assignment with this design , and to provide empirical support for the proof of Theorem 1 , we compared the loss function at the hidden layer , L0 , to the output layer loss function , L1 , across all of the image presentations to the network . We observed that , generally , whenever the hidden layer loss was low , the output layer loss was also low . For example , when we consider the loss for the set of ‘2’ images presented to the network during the second epoch , there was a Pearson correlation coefficient between L0 and L1 of r=0 . 61 , which was much higher than what was observed for shuffled data , wherein output and hidden activities were randomly paired ( Figure 5B ) . Furthermore , these correlations were observed across all epochs of training , with most correlation coefficients for the hidden and output loss functions falling between r=0 . 2 - 0 . 6 , which was , again , much higher than the correlations observed for shuffled data ( Figure 5C ) . Interestingly , the correlations between L0 and L1 were smaller on the first epoch of training ( see data in red oval Figure 5C ) . This suggests that the guarantee of coordination between L0 and L1 only comes into full effect once the network has engaged in some learning . Therefore , we inspected whether the conditions on the synaptic matrices that are assumed in the proof of Theorem 1 were , in fact , being met . More precisely , the proof assumes that the feedforward and feedback synaptic matrices ( W1 and Y , respectively ) produce forward and backward transformations between the output and hidden layer whose Jacobians are approximate inverses of each other ( see Proof of Theorem 1 ) . Since we begin learning with random matrices , this condition is almost definitely not met at the start of training . But , we found that the network learned to meet this condition . Inspection of W1 and Y showed that during the first epoch , the Jacobians of the forward and backwards functions became approximate inverses of each other ( Figure 5—figure supplement 1 ) . Since Y is frozen , this means that during the first few image presentations W1 was being updated to have its Jacobian come closer to the inverse of Y's Jacobian . Put another way , the network was learning to do credit assignment . We have yet to resolve exactly why this happens , though the result is very similar to the findings of Lillicrap et al . ( 2016 ) , where a proof is provided for the linear case . Intuitively , though , the reason is likely the interaction between W1 and W0: as W0 gets updated , the hidden layer learns to group stimuli based on the feedback sent through Y . So , for W1 to transform the hidden layer activity into the correct output layer activity , W1 must become more like the inverse of Y , which would also make the Jacobian of W1 more like the inverse of Y’s Jacobian ( due to the inverse function theorem ) . However , a complete , formal explanation for this phenomenon is still missing , and the the issue of weight alignment deserves additional investigation Lillicrap et al . ( 2016 ) . From a biological perspective , it also suggests that very early development may involve a period of learning how to assign credit appropriately . Altogether , our model demonstrates that deep learning using random feedback weights is a general phenomenon , and one which can be implemented using segregated dendrites to keep forward information separate from feedback signals used for credit assignment . Given our finding that the network was successfully assigning credit for the output error to the hidden layer neurons , we had reason to believe that our network with local weight-updates would exhibit deep learning , that is an ability to take advantage of a multi-layer structure ( Bengio and LeCun , 2007 ) . To test this , we examined the effects of including hidden layers . If deep learning is indeed operational in the network , then the inclusion of hidden layers should improve the ability of the network to classify images . We built three different versions of the network ( Figure 6A ) . The first was a network that had no hidden layer , that is the input neurons projected directly to the output neurons . The second was the network illustrated in Figure 3B , with a single hidden layer . The third contained two hidden layers , with the output layer projecting directly back to both hidden layers . This direct projection allowed us to build our local targets for each hidden layer using the plateaus driven by the output layer , thereby avoiding a ‘backward pass’ through the entire network as has been used in other models ( Lillicrap et al . , 2016; Lee et al . , 2015; Liao et al . , 2015 ) . We trained each network on the 60 , 000 MNIST training images for 60 epochs , and recorded the percentage of images in the 10 , 000 image test set that were incorrectly classified . The network with no hidden layers rapidly learned to classify the images , but it also rapidly hit an asymptote at an average error rate of 8 . 3% ( Figure 6B , gray line ) . In contrast , the network with one hidden layer did not exhibit a rapid convergence to an asymptote in its error rate . Instead , it continued to improve throughout all 60 epochs , achieving an average error rate of 4 . 1% by the 60th epoch ( Figure 6B , blue line ) . Similar results were obtained when we loosened the synchrony constraints and instead allowed each hidden layer neuron to engage in plateau potentials at different times ( Figure 6—figure supplement 1 ) . This demonstrates that strict synchrony in the plateau potentials is not required . But , our target definitions do require two different plateau potentials separated by the teaching signal input , which mandates some temporal control of plateau potentials in the system . Interestingly , we found that the addition of a second hidden layer further improved learning . The network with two hidden layers learned more rapidly than the network with one hidden layer and achieved an average error rate of 3 . 2% on the test images by the 60th epoch , also without hitting a clear asymptote in learning ( Figure 6B , red line ) . However , it should be noted that additional hidden layers beyond two did not significantly improve the error rate ( data not shown ) , which suggests that our particular algorithm could not be used to construct very deep networks as is . Nonetheless , our network was clearly able to take advantage of multi-layer architectures to improve its learning , which is the key feature of deep learning ( Bengio and LeCun , 2007; LeCun et al . , 2015 ) . Another key feature of deep learning is the ability to generate representations in the higher layers of a network that capture task-relevant information while discarding sensory details ( LeCun et al . , 2015; Mnih et al . , 2015 ) . To examine whether our network exhibited this type of abstraction , we used the t-Distributed Stochastic Neighbor Embedding algorithm ( t-SNE ) . The t-SNE algorithm reduces the dimensionality of data while preserving local structure and non-linear manifolds that exist in high-dimensional space , thereby allowing accurate visualization of the structure of high-dimensional data ( Maaten and Hinton , 2008 ) . We applied t-SNE to the activity patterns at each layer of the two hidden layer network for all of the images in the test set after 60 epochs of training . At the input level , there was already some clustering of images based on their categories . However , the clusters were quite messy , with different categories showing outliers , several clusters , or merged clusters ( Figure 6C , bottom ) . For example , the ‘2’ digits in the input layer exhibited two distinct clusters separated by a cluster of ‘7’s: one cluster contained ‘2’s with a loop and one contained ‘2’s without a loop . Similarly , there were two distinct clusters of ‘4’s and ‘9’s that were very close to each other , with one pair for digits on a pronounced slant and one for straight digits ( Figure 6C , bottom , example images ) . Thus , although there is built-in structure to the categories of the MNIST dataset , there are a number of low-level features that do not respect category boundaries . In contrast , at the first hidden layer , the activity patterns were much cleaner , with far fewer outliers and split/merged clusters ( Figure 6C , middle ) . For example , the two separate ‘2’ digit clusters were much closer to each other and were now only separated by a very small cluster of ‘7’s . Likewise , the ‘9’ and ‘4’ clusters were now distinct and no longer split based on the slant of the digit . Interestingly , when we examined the activity patterns at the second hidden layer , the categories were even better segregated , with only a little bit of splitting or merging of category clusters ( Figure 6C , top ) . Therefore , the network had learned to develop representations in the hidden layers wherein the categories were very distinct and low-level features unrelated to the categories were largely ignored . This abstract representation is likely to be key to the improved error rate in the two hidden layer network . Altogether , our data demonstrates that our network with segregated dendritic compartments can engage in deep learning . The backpropagation of error algorithm ( Rumelhart et al . , 1986 ) is still the primary learning algorithm used for deep supervised learning in artificial neural networks ( LeCun et al . , 2015 ) . Previous work has shown that learning with random feedback weights can actually match the synaptic weight updates specified by the backpropagation algorithm after a few epochs of training ( Lillicrap et al . , 2016 ) . This fascinating observation suggests that deep learning with random feedback weights is not completely distinct from backpropagation of error , but rather , networks with random feedback connections learn to approximate credit assignment as it is done in backpropagation ( Lillicrap et al . , 2016 ) . Hence , we were curious as to whether or not our network was , in fact , learning to approximate the synaptic weight updates prescribed by backpropagation . To test this , we trained our one hidden layer network as before , but now , in addition to calculating the vector of hidden layer synaptic weight updates specified by our local learning rule ( ΔW0 in Equation ( 10 ) ) , we also calculated the vector of hidden layer synaptic weight updates that would be specified by non-locally backpropagating the error from the output layer , ( ΔWBP0 ) . We then calculated the angle between these two alternative weight updates . In a very high-dimensional space , any two independent vectors will be roughly orthogonal to each other ( i . e . ΔW0∠ΔWBP0≈90∘ ) . If the two synaptic weight update vectors are not orthogonal to each other ( i . e . ΔW0∠ΔWBP0<90∘ ) , then it suggests that the two algorithms are specifying similar weight updates . As in previous work ( Lillicrap et al . , 2016 ) , we found that the initial weight updates for our network were orthogonal to the updates specified by backpropagation . But , as the network learned the angle dropped to approximately 65∘ , before rising again slightly to roughly 70∘ ( Figure 7A , blue line ) . This suggests that our network was learning to develop local weight updates in the hidden layer that were in rough agreement with the updates that explicit backpropagation would produce . However , this drop in orthogonality was still much less than that observed in non-spiking artificial neural networks learning with random feedback weights , which show a drop to below 45∘ ( Lillicrap et al . , 2016 ) . We suspected that the higher angle between the weight updates that we observed may have been because we were using spikes to communicate the feedback from the upper layer , which could introduce both noise and bias in the estimates of the output layer activity . To test this , we also examined the weight updates that our algorithm would produce if we propagated the spike rates of the output layer neurons , ϕ1 ( t ) , back directly through the random feedback weights , Y . In this scenario , we observed a much sharper drop in the ΔW0∠ΔWBP0 angle , which reduced to roughly 35∘ before rising again to 40∘ ( Figure 7A , red line ) . These results show that , in principle , our algorithm is learning to approximate the backpropagation algorithm , though with some drop in accuracy introduced by the use of spikes to propagate output layer activities to the hidden layer . To further examine how our local learning algorithm compared to backpropagation we compared the low-level features that the two algorithms learned . To do this , we trained the one hidden layer network with both our algorithm and backpropagation . We then examined the receptive fields ( i . e . the synaptic weights ) produced by both algorithms in the hidden layer synapses ( W0 ) after 60 epochs of training . The two algorithms produced qualitatively similar receptive fields ( Figure 7B ) . Both produced receptive fields with clear , high-contrast features for detecting particular strokes or shapes . To quantify the similarity , we conducted pair-wise correlation calculations for the receptive fields produced by the two algorithms and identified the maximum correlation pairs for each . Compared to shuffled versions of the receptive fields , there was a very high level of maximum correlation ( Figure 7C ) , showing that the receptive fields were indeed quite similar . Thus , the data demonstrate that our learning algorithm using random feedback weights into segregated dendrites can in fact come to approximate the backpropagation of error algorithm . Once we had convinced ourselves that our learning algorithm was , in fact , providing a solution to the credit assignment problem , we wanted to examine some of the constraints on learning . First , we wanted to explore the structure of the feedback weights . In our initial simulations we used non-sparse , random ( i . e . normally distributed ) feedback weights . We were interested in whether learning could still work with sparse weights , given that neocortical connectivity is sparse . As well , we wondered whether symmetric weights would improve learning , which would be expected given previous findings ( Lillicrap et al . , 2016; Lee et al . , 2015; Liao et al . , 2015 ) . To explore these questions , we trained our one hidden layer network using both sparse feedback weights ( only 20% non-zero values ) and symmetric weights ( Y=W1T ) ( Figure 8A , C ) . We found that learning actually improved slightly with sparse weights ( Figure 8B , red line ) , achieving an average error rate of 3 . 7% by the 60th epoch , compared to the average 4 . 1% error rate achieved with fully random weights . But , this result appeared to depend on the magnitude of the sparse weights . To compensate for the loss of 80% of the weights we initially increased the sparse synaptic weight magnitudes by a factor of 5 . However , when we did not re-scale the sparse weights learning was actually worse ( Figure 8—figure supplement 1 ) , though this could likely be dealt with by a careful resetting of learning rates . Altogether , our results suggest that sparse feedback provides a signal that is sufficient for credit assignment . Similar to sparse feedback weights , symmetric feedback weights also improved learning , leading to a rapid decrease in the test error and an error rate of 3 . 6% by the 60th epoch ( Figure 8D , red line ) . This is interesting , given that backpropagation assumes symmetric feedback weights ( Lillicrap et al . , 2016; Bengio et al . , 2015 ) , though our proof of Theorem 1 does not . However , when we added noise to the symmetric weights any advantage was eliminated and learning was , in fact , slightly impaired ( Figure 8D , blue line ) . At first , this was a very surprising result: given that learning works with random feedback weights , why would it not work with symmetric weights with noise ? However , when we considered our previous finding that during the first epoch the feedforward weights , W1 , learn to have the feedforward Jacobian match the inverse of the feedback Jacobian ( Figure 5—figure supplement 1 ) a possible answer emerges . In the case of symmetric feedback weights the synaptic matrix Y is changing as W1 changes . This works fine when Y is set to W1T , since that artificially forces something akin to backpropagation . But , if the feedback weights are set to W1T plus noise , then the system can never align the Jacobians appropriately , since Y is now a moving target . This would imply that any implementation of feedback learning must either be very effective ( to achieve the right feedback ) or very slow ( to allow the feedforward weights to adapt ) . Another constraint that we wished to examine was whether total segregation of the apical inputs was necessary , given that real pyramidal neurons only show an attenuation of distal apical inputs to the soma ( Larkum et al . , 1999 ) . Total segregation ( ga=0 ) renders the network effectively feed-forward in its dynamics , which made it easier to construct the loss functions to ensure that reducing L0 also reduces L1 ( see Figure 5 and Theorem 1 ) . But , we wondered whether some degree of apical conductance to the soma would be sufficiently innocuous so as to not disrupt deep learning . To examine this , we re-ran our two hidden layer network , but now , we allowed the apical dendritic voltage to influence the somatic voltage by setting ga=0 . 05 . This value gave us twelve times more attenuation than the attenuation from the basal compartments , since gb=0 . 6 ( Figure 9A ) . When we compared the learning in this scenario to the scenario with total apical segregation , we observed very little difference in the error rates on the test set ( Figure 9B , gray and red lines ) . Importantly , though , we found that if we increased the apical conductance to the same level as the basal ( ga=gb=0 . 6 ) then the learning was significantly impaired ( Figure 9B , blue line ) . This demonstrates that although total apical attenuation is not necessary , partial segregation of the apical compartment from the soma is necessary . That result makes sense given that our local targets for the hidden layer neurons incorporate a term that is supposed to reflect the response of the output neurons to the feedforward sensory information ( αf ) . Without some sort of separation of feedforward and feedback information , as is assumed in other models of deep learning ( Lillicrap et al . , 2016; Lee et al . , 2015 ) , this feedback signal would get corrupted by recurrent dynamics in the network . Our data show that electrontonically segregated dendrites is one potential way to achieve the separation between feedforward and feedback information that is required for deep learning . Deep learning has radically altered the field of AI , demonstrating that parallel distributed processing across multiple layers can produce human/animal-level capabilities in image classification , pattern recognition and reinforcement learning ( Hinton et al . , 2006; LeCun et al . , 2015; Mnih et al . , 2015; Silver et al . , 2016; Krizhevsky et al . , 2012; He et al . , 2015 ) . Deep learning was motivated by analogies to the real brain ( LeCun et al . , 2015; Cox and Dean , 2014 ) , so it is tantalizing that recent studies have shown that deep neural networks develop representations that strongly resemble the representations observed in the mammalian neocortex ( Khaligh-Razavi and Kriegeskorte , 2014; Yamins and DiCarlo , 2016; Cadieu et al . , 2014; Kubilius et al . , 2016 ) . In fact , deep learning models can match cortical representations better than some models that explicitly attempt to mimic the real brain ( Khaligh-Razavi and Kriegeskorte , 2014 ) . Hence , at a phenomenological level , it appears that deep learning , defined as multilayer cost function reduction with appropriate credit assignment , may be key to the remarkable computational prowess of the mammalian brain ( Marblestone et al . , 2016 ) . However , the lack of biologically feasible mechanisms for credit assignment in deep learning algorithms , most notably backpropagation of error ( Rumelhart et al . , 1986 ) , has left neuroscientists with a mystery . Given that the brain cannot use backpropagation , how does it solve the credit assignment problem ( Figure 1 ) ? Here , we expanded on an idea that previous authors have explored ( Körding and König , 2001; Spratling , 2002; Spratling and Johnson , 2006 ) and demonstrated that segregating the feedback and feedforward inputs to neurons , much as the real neocortex does ( Larkum et al . , 1999; 2007; 2009 ) , can enable the construction of local targets to assign credit appropriately to hidden layer neurons ( Figure 2 ) . With this formulation , we showed that we could use segregated dendritic compartments to coordinate learning across layers ( Figure 3 , Figure 4 and Figure 5 ) . This enabled our network to take advantage of multiple layers to develop representations of hand-written digits in hidden layers that enabled better levels of classification accuracy on the MNIST dataset than could be achieved with a single layer ( Figure 6 ) . Furthermore , we found that our algorithm actually approximated the weight updates that would be prescribed by backpropagation , and produced similar low-level feature detectors ( Figure 7 ) . As well , we showed that our basic framework works with sparse feedback connections ( Figure 8 ) and more realistic , partial apical attenuation ( Figure 9 ) . Therefore , our work demonstrates that deep learning is possible in a biologically feasible framework , provided that feedforward and feedback signals are sufficiently segregated in different dendrites . In this work we adopted a similar strategy to the one taken by Lee et al . , 2015 in their difference target propagation algorithm , wherein the feedback from higher layers is used to construct local firing-rate targets at the hidden layers . One of the reasons that we adopted this strategy is that it is appealing to think that feedback from upper layers may not simply be providing a signal for plasticity , but also a predictive and/or modulatory signal to push the hidden layer neurons towards a ‘better’ activity pattern in real-time . This sort of top-down control could be used by the brain to improve sensory processing in different contexts and engage in inference ( Bengio et al . , 2015 ) . Indeed , framing cortico-cortical feedback as a mechanism to predict or modulate incoming sensory activity is a more common way of viewing feedback signals in the neocortex ( Larkum , 2013; Gilbert and Li , 2013; Zhang et al . , 2014; Fiser et al . , 2016; Leinweber et al . , 2017 ) . In light of this , it is interesting to note that distal apical inputs in sensory cortical areas can predict upcoming stimuli ( Leinweber et al . , 2017; Fiser et al . , 2016 ) , and help animals perform sensory discrimination tasks ( Takahashi et al . , 2016; Manita et al . , 2015 ) . However , in our model , we did not actually implement a system that altered the hidden layer activity to make sensory computations—we simply used the feedback signals to drive learning . In-line with this view of top-down feedback , two recent papers have found evidence that cortical feedback can indeed guide feedforward sensory plasticity ( Thompson et al . , 2016; Yamada et al . , 2017 ) , and in the hippocampus , there is evidence that plateau potentials generated by apical inputs are key determinants of plasticity ( Bittner et al . , 2015; Bittner et al . , 2017 ) . But , ultimately , there is no reason that feedback signals cannot provide both top-down predicton/modulation and a signal for learning ( Spratling , 2002 ) . In this respect , a potential future advance on our model would be to implement a system wherein the feedback makes predictions and ‘nudges’ the hidden layers towards appropriate activity patterns in order to guide learning and shape perception simultaneously . This proposal is reminiscent of the approach taken in previous computational models ( Urbanczik and Senn , 2014; Spratling and Johnson , 2006; Körding and König , 2001 ) . Future research could study how top-down control of activity and a signal for credit assignment can be combined . In a number of ways , the model that we presented here is more biologically feasible than other deep learning models . We utilized leaky integrator neurons that communicate with spikes , we simulated in near continuous-time , and we used spatially local synaptic plasticity rules . Yet , there are still clearly unresolved issues of biological feasibility in our model . Most notably , the model updates synaptic weights using the difference between two plateau potentials that occur following two different phases . There are three issues with this method from a biological standpoint . First , it necessitates two distinct global phases of processing ( the ‘forward’ and ‘target’ phases ) . Second , the plateau potentials occur in the apical compartment , but they are used to update the basal synapses , meaning that this information from the apical dendrites must somehow be communicated to the rest of the neuron . Third , the two plateau potentials occur with a temporal gap of tens of milliseconds , meaning that this difference must somehow be computed over time . These issues could , theoretically , be resolved in a biologically realistic manner . The two different phases could be a result of a global signal indicating whether the teaching signal was present . This could be accomplished with neuromodulatory systems ( Pi et al . , 2013 ) , or alternatively , with oscillations that the teaching signal and apical dendrites are phase locked to ( Veit et al . , 2017 ) . Communicating plateau potentials to the basal dendrites is also possible using known biological principles . Plateau potentials induce bursts of action potentials in pyramidal neurons ( Larkum et al . , 1999 ) , and the rate-of-fire of the bursts would be a function of the level of the plateau potential . Given that action potentials would propagate back through the basal dendrites ( Kampa and Stuart , 2006 ) , any cellular mechanism in the basal dendrites that is sensitive to rate-of-fire of bursts could be used to detect the level of the plateau potentials in the apical dendrite . Finally , taking the difference between two events that occur tens of milliseconds apart is possible if such a hypothetical cellular signal that is sensitive to bursts had a slow decay time constant , and reacted differently depending on whether the global phase signal was active . A simple mathematical formulation for such a cellular signal is given in the methods ( see Equations ( 36 ) and ( 37 ) ) . It is worth noting that incorporation of bursting into somatic dynamics would be unlikely to affect the learning results we presented here . This is because we calculate weight updates by averaging the activity of the neurons for a period after the network is near steady-state ( i . e . the period marked with the blue line in Figure 3C , see also Equation ( 5 ) ) . Even if bursts of activity temporarily altered the dynamics of the network , they would not significantly alter the steady-state activity . Future work could expand on the model presented here and explore whether bursting activity might beneficially alter somatic dynamics ( e . g . for on-line inference ) , as well as driving learning . These possible implementations are clearly speculative , and only partially in-line with experimental evidence . As the adage goes , all models are wrong , but some models are useful . Our model aims to inspire new ways to think about how the credit assignment problem could be solved by known circuits in the brain . Our study demonstrates that some of the machinery that is known to exist in the neocortex , namely electrotonically segregated apical dendrites receiving top-down inputs , may be well-suited to credit assignment computations . What we are proposing is that the neocortex could use the segregation of top-down inputs to the apical dendrites in order to solve the credit assignment problem , without using a separate feedback pathway as is implicit in most deep learning models used in machine learning . We consider this to be the core insight of our model , and an important step in making deep learning more biologically plausible . Indeed , our model makes both a generic , and a specific , prediction about the role of synaptic inputs to apical dendrites during learning . The generic prediction is that the sign of synaptic plasticity , that is whether LTP or LTD occur , in the basal dendrites will be modulated by different patterns of inputs to the apical dendrites . The more specific prediction that our model makes is that the timing of apical inputs relative to basal inputs should be what determines the sign of plasticity for synapses in the basal dendrites . For example , if apical and basal inputs arrive at the same time , but the apical inputs disappear before the basal inputs do , then presumably plateau potentials will be stronger early in the stimulus presentation ( i . e . αf>αt ) , and so the basal synapses should engage in LTD . In contrast , if the apical inputs only arrive after the basal inputs have been active for some period of time , then plateau potentials will be stronger towards the end of stimulus presentation ( i . e . αf<αt ) , and so the basal synapses should engage in LTP . Both the generic and specific predictions should be experimentally testable using modern optical techniques to separate the inputs to the basal and apical dendrites ( Figure 10 ) . Another direction for future research should be to consider how to use the machinery of neocortical microcircuits to communicate credit assignment signals without relying on differences across phases , as we did here . For example , somatostatin positive interneurons , which possess short-term facilitating synapses ( Silberberg and Markram , 2007 ) , are particularly sensitive to bursts of spikes , and could be part of a mechanism to calculate differences in the top-down signals being received by pyramidal neuron dendrites . If a calculation of this difference spanned the time before and after a teaching signal arrived , it could , theoretically , provide the computation that our system implements with a difference between plateau potentials . Indeed , we would argue that credit assignment may be one of the major functions of the canonical neocortical microcircuit motif . If this is correct , then the inhibitory interneurons that target apical dendrites may be used by the neocortex to control learning ( Murayama et al . , 2009 ) . Although this is speculative , it is worth noting that current evidence supports the idea that neuromodulatory inputs carrying temporally precise salience information ( Hangya et al . , 2015 ) can shut off interneurons to disinhibit the distal apical dendrites ( Pi et al . , 2013; Karnani et al . , 2016; Pfeffer et al . , 2013; Brombas et al . , 2014 ) , and presumably , promote apical communication to the soma . Recent work suggests that the specific patterns of interneuron inhibition on the apical dendrites are spatially precise and differentially timed to motor behaviours ( Muñoz et al . , 2017 ) , which suggests that there may well be coordinated physiological mechanisms for determining when and how cortico-cortical feedback is transmitted to the soma and basal dendrites . Future research should examine whether these inhibitory and neuromodulatory mechanisms do , in fact , control plasticity in the basal dendrites of pyramidal neurons , as our model , and some recent experimental work ( Bittner et al . , 2015; Bittner et al . , 2017 ) , would predict . A non-biological issue that should be recognized is that the error rates which our network achieved were by no means as low as can be achieved with artificial neural networks , nor at human levels of performance ( Lecun et al . , 1998; Li et al . , 2016 ) . As well , our algorithm was not able to take advantage of very deep structures ( beyond two hidden layers , the error rate did not improve ) . In contrast , increasing the depth of networks trained with backpropagation can lead to performance improvements ( Li et al . , 2016 ) . But , these observations do not mean that our network was not engaged in deep learning . First , it is interesting to note that although the backpropagation algorithm is several decades old ( Rumelhart et al . , 1986 ) , it was long considered to be useless for training networks with more than one or two hidden layers ( Bengio and LeCun , 2007 ) . Indeed , it was only the use of layer-by-layer training that initially led to the realization that deeper networks can achieve excellent performance ( Hinton et al . , 2006 ) . Since then , both the use of very large datasets ( with millions of examples ) , and additional modifications to the backpropagation algorithm , have been key to making backpropagation work well on deeper networks ( Sutskever et al . , 2013; LeCun et al . , 2015 ) . Future studies could examine how our algorithm could incorporate current techniques used in machine learning to work better on deeper architectures . Second , we stress that our network was not designed to match the state-of-the-art in machine learning , nor human capabilities . To test our basic hypothesis ( and to run our leaky-integration and spiking simulations in a reasonable amount of time ) we kept the network small , we stopped training before it reached its asymptote , and we did not implement any add-ons to the learning to improve the error rates , such as convolution and pooling layers , initialization tricks , mini-batch training , drop-out , momentum or RMSProp ( Sutskever et al . , 2013; Tieleman and Hinton , 2012; Srivastava et al . , 2014 ) . Indeed , it would be quite surprising if a relatively vanilla , small network like ours could come close to matching current performance benchmarks in machine learning . Third , although our network was able to take advantage of multiple layers to improve the error rate , there may be a variety of reasons that ever increasing depth didn’t improve performance significantly . For example , our use of direct connections from the output layer to the hidden layers may have impaired the network’s ability to coordinate synaptic updates between hidden layers . As well , given our finding that the use of spikes produced weight updates that were less well-aligned to backpropagation ( Figure 7A ) it is possible that deeper architectures require mechanisms to overcome the inherent noisiness of spikes . One aspect of our model that we did not develop was the potential for learning at the feedback synapses . Although we used random synaptic weights for feedback , we also demonstrated that our model actually learns to meet the mathematical conditions required for credit assignment ( Figure 5—figure supplement 1 ) . This suggests that it would be beneficial to develop a synaptic weight update rule for the feedback synapses that made this aspect of the learning better . Indeed , Lee et al . , 2015 implemented an ‘inverse loss function’ for their feedback synapses which promoted the development of feedforward and feedback functions that were roughly inverses of each other , leading to the emergence of auto-encoder functions in their network . In light of this , it is interesting to note that there is evidence for unique , ‘reverse’ spike-timing-dependent synaptic plasticity rules in the distal apical dendrites of pyramidal neurons ( Sjöström and Häusser , 2006; Letzkus et al . , 2006 ) , which have been shown to produce symmetric feedback weights and auto-encoder functions in artificial spiking networks ( Burbank and Kreiman , 2012; Burbank , 2015 ) . Thus , it is possible that early in development the neocortex actually learns cortico-cortical feedback connections that help it to assign credit for later learning . Our work suggests that any experimental evidence showing that feedback connections learn to approximate the inverse of feedforward connections could be considered as evidence for deep learning in the neocortex . A final consideration , which is related to learning at feedback synapses , is the likely importance of unsupervised learning for the real brain , that is learning without a teaching signal . In this paper , we focused on a supervised learning task with a teaching signal . Supervised learning certainly could occur in the brain , especially for goal-directed sensorimotor tasks where animals have access to examples that they could use to generate internal teaching signals Teşileanu et al . ( 2017 ) . But , unsupervised learning is likely critical for understanding the development of cognition ( Marblestone et al . , 2016 ) . Importantly , unsupervised learning in multilayer networks still requires a solution to the credit assignment problem ( Bengio et al . , 2015 ) , so our work here is not completely inapplicable . Nonetheless , future research should examine how the credit assignment problem can be addressed in the specific case of unsupervised learning . In summary , deep learning has had a huge impact on AI , but , to date , its impact on neuroscience has been limited . Nonetheless , given a number of findings in neurophysiology and modeling ( Yamins and DiCarlo , 2016 ) , there is growing interest in understanding how deep learning may actually be achieved by the real brain ( Marblestone et al . , 2016 ) . Our results show that by moving away from point neurons , and shifting towards multi-compartment neurons that segregate feedforward and feedback signals , the credit assignment problem can be solved and deep learning can be achieved . Perhaps the dendritic anatomy of neocortical pyramidal neurons is important for nature’s own deep learning algorithm . The network described here consists of an input layer with ℓ neurons , a hidden layer with m neurons , and an output layer with n neurons . Neurons in the input layer are simple Poisson spiking neurons whose rate-of-fire is determined by the intensity of image pixels ( ranging from 0 - ϕmax ) . Neurons in the hidden layer are modeled using three functional compartments—basal dendrites with voltages V0b ( t ) =[V10b ( t ) , V20b ( t ) , . . . , Vm0b ( t ) ] , apical dendrites with voltages V0a ( t ) =[V10a ( t ) , V20a ( t ) , . . . , Vm0a ( t ) ] , and somata with voltages V0 ( t ) =[V10 ( t ) , V20 ( t ) , . . . , Vm0 ( t ) ] . Feedforward inputs from the input layer and feedback inputs from the output layer arrive at basal and apical synapses , respectively . At basal synapses , presynaptic spikes from input layer neurons are translated into filtered spike trains sinput ( t ) =[s1input ( t ) , s2input ( t ) , . . . , sℓinput ( t ) ] given by: ( 11 ) sjinput ( t ) =∑kκ ( t−tjkinput ) where tjkinput is the k th spike time of input neuron j is the response kernel given by: ( 12 ) κ ( t ) = ( e−t/τL−e−t/τs ) Θ ( t ) / ( τL−τs ) where τs and τL are short and long time constants , and Θ is the Heaviside step function . Since the network is fully-connected , each neuron in the hidden layer will receive the same set of filtered spike trains from input layer neurons . The filtered spike trains at apical synapses , s1 ( t ) =[s11 ( t ) , s21 ( t ) , . . . , sn1 ( t ) ] , are modeled in the same manner . The basal and apical dendritic potentials for neuron i are then given by weighted sums of the filtered spike trains at either its basal or apical synapses: ( 13 ) Vi0b ( t ) =∑j=1ℓWij0sjinput ( t ) +bi0Vi0a ( t ) =∑j=1nYijsj1 ( t ) where b0=[b10 , b20 , . . . , bm0] are bias terms , W0 is the m×ℓ matrix of feedforward weights for neurons in the hidden layer , and Y is the m×n matrix of their feedback weights . The somatic voltage for neuron i evolves with leak as: ( 14 ) τdVi0 ( t ) dt= ( VR−Vi0 ( t ) ) +gbgl ( Vi0b ( t ) −Vi0 ( t ) ) +gagl ( Vi0a ( t ) −Vi0 ( t ) ) ( 15 ) = ( VR−Vi0 ( t ) ) +gbgl ( ∑j=1ℓWij0sjinput ( t ) +bi0−Vi0 ( t ) ) +gagl ( ∑j=1nYij0sj1 ( t ) −Vi0 ( t ) ) where VR is the resting potential , gl is the leak conductance , gb is the conductance from the basal dendrite to the soma , and ga is the conductance from the apical dendrite to the soma , and τ is a function of gl and the membrance capacitance Cm: ( 16 ) τ=Cmgl Note that for simplicity’s sake we are assuming a resting potential of 0 mV and a membrane capacitance of 1 F , but these values are not important for the results . Equations ( 13 ) and ( 14 ) are identical to the Equation ( 1 ) in results . The instantaneous firing rates of neurons in the hidden layer are given by ϕ0 ( t ) =[ϕ10 ( t ) , ϕ20 ( t ) , . . . , ϕm0 ( t ) ] , where ϕi0⁢ ( t ) is the result of applying a nonlinearity , σ⁢ ( ⋅ ) , to the somatic potential Vi0⁢ ( t ) . We chose σ⁢ ( ⋅ ) to be a simple sigmoidal function , such that: ( 17 ) ϕi0 ( t ) =ϕmaxσ ( Vi0 ( t ) ) =ϕmax11+e−Vi0 ( t ) Here , ϕmax is the maximum possible rate-of-fire for the neurons , which we set to 200 Hz . Note that Equation ( 17 ) is identical to Equation ( 3 ) in results . Spikes are then generated using Poisson processes with these firing rates . We note that although the maximum rate was 200 Hz , the neurons rarely achieved anything close to this rate , and the average rate of fire in the neurons during our simulations was 24 Hz . Units in the output layer are modeled using only two compartments , dendrites with voltages V1b ( t ) =[V11b ( t ) , V21b ( t ) , . . . , Vn1b ( t ) ] and somata with voltages V1 ( t ) =[V11 ( t ) , V21 ( t ) , . . . , Vn1 ( t ) ] is given by: ( 18 ) Vi1b ( t ) =∑j=1mWij1sj0 ( t ) +bi1 where s0 ( t ) =[s10 ( t ) , s20 ( t ) , . . . , sm0 ( t ) ] are the filtered presynaptic spike trains at synapses that receive feedforward input from the hidden layer , and are calculated in the manner described by Equation ( 11 ) . Vi1⁢ ( t ) evolves as: ( 19 ) τdVi1 ( t ) dt= ( VR−Vi1 ( t ) ) +gdgl ( Vi1b ( t ) −Vi1 ( t ) ) +Ii ( t ) where gl is the leak conductance , gd is the conductance from the dendrite to the soma , and I ( t ) =[I1 ( t ) , I2 ( t ) , . . . , In ( t ) ] are somatic currents that can drive output neurons toward a desired somatic voltage . For neuron i , Ii is given by: ( 20 ) Ii ( t ) =gEi ( t ) ( EE−Vi1 ( t ) ) +gIi ( t ) ( EI−Vi1 ( t ) ) where gE ( t ) =[gE1 ( t ) , gE2 ( t ) , . . . , gEn ( t ) ] and gI ( t ) =[gI1 ( t ) , gI2 ( t ) , . . . , gIn ( t ) ] are time-varying excitatory and inhibitory nudging conductances , and EE and EI are the excitatory and inhibitory reversal potentials . In our simulations , we set EE=8 V and EI=-8 V . During the target phase only , we set gIi=1 and gEi=0 for all units i whose output should be minimal , and gEi=1 and gIi=0 for the unit whose output should be maximal . In this way , all units other than the ‘target’ unit are silenced , while the ‘target’ unit receives a strong excitatory drive . In the forward phase , I ( t ) is set to 0 . The Poisson spike rates ϕ1 ( t ) =[ϕ11 ( t ) , ϕ21 ( t ) , . . . , ϕn1 ( t ) ] are calculated as in Equation ( 17 ) . At the end of the forward and target phases , we calculate plateau potentials αf=[α1f , α2f , . . . , αmf] and αt=[α1t , α2t , . . . , αmt] for apical dendrites of hidden layer neurons , where αif and αit are given by: ( 21 ) αif=σ ( 1Δt1∫t1−Δt1t1Vi0a ( t ) dt ) αit=σ ( 1Δt2∫t2−Δt2t2Vi0a ( t ) dt ) where t1 and t2 are the end times of the forward and target phases , respectively , Δ⁢ts=30 ms is the settling time for the voltages , and Δt1 and Δt2 are given by: ( 22 ) Δt1=t1− ( t0+Δts ) Δt2=t2− ( t1+Δts ) Note that Equation ( 21 ) is identical to Equation ( 5 ) in results . These plateau potentials are used by hidden layer neurons to update their basal weights . All feedforward synaptic weights are updated at the end of each target phase . Output layer units update their synaptic weights W1 in order to minimize the loss function ( 23 ) L1=\|\|ϕ1∗−ϕmaxσ ( V1¯f ) \|\|22 where ϕ1∗=ϕ1¯t as in Equation ( 6 ) . Note that , as long as neuronal units calculate averages after the network has reached a steady state , and the firing-rates of the neurons are in the linear region of the sigmoid function , then for layer x , ( 24 ) ϕmaxσ ( Vx¯f ) ≈ϕmaxσ ( Vx ) ¯f=ϕx¯f Thus , ( 25 ) L1≈\|\|ϕ1¯t−ϕ1¯f\|\|22 as in Equation ( 7 ) . All average voltages are calculated after a delay Δ⁢ts from the start of a phase , which allows for the network to reach a steady state before averaging begins . In practice this means that the average somatic voltage for output layer neuron i in the forward phase , Vi1¯f , has the property ( 26 ) Vi1¯f≈kdVi1b¯f=kd ( ∑j=1mWij1sj0¯f+bi1 ) where kd is given by: ( 27 ) kd=gdgl+gd Thus , ( 28 ) ∂L1∂W1≈−kdϕmax ( ϕ1∗−ϕmaxσ ( V1¯f ) ) σ′ ( V1¯f ) ∘s0¯f∂L1∂b1≈−kdϕmax ( ϕ1∗−ϕmaxσ ( V1¯f ) ) σ′ ( V1¯f ) Note that these partial derivatives assume that the activity during the target phase is fixed . We do this because the goal of learning is to have the network behave as it does during the target phase , even when the teaching signal is present . Thus , we do not update synapses in order to alter the target phase activity . As a result , there are no terms in the equation related to the partial derivatives of the voltages or firing-rates during the target phase . The dendrites in the output layer use this approximation of the gradient in order to update their weights using gradient descent: ( 29 ) W1→W1−η1P1∂L1∂W1b1→b1−η1P1∂L1∂b1 where η1 is a learning rate constant , and P1 is a scaling factor used to normalize the scale of the rate-of-fire function . In the hidden layer , basal dendrites update their synaptic weights W0 by minimizing the loss function ( 30 ) L0=\|\|ϕ0∗−ϕmaxσ ( V0¯f ) \|\|22 We define the target rates-of-fire ϕ0∗=[ϕ10∗ , ϕ20∗ , . . . , ϕm0∗] such that ( 31 ) ϕi0∗=ϕi0¯f+αit−αif where αf=[α1f , α2f , . . . , αmf] and αt=[α1t , α2t , . . . , αmt] are forward and target phase plateau potentials given in Equation ( 21 ) . Note that Equation ( 31 ) is identical to Equation ( 8 ) in results . These hidden layer target firing rates are similar to the targets used in difference target propagation ( Lee et al . , 2015 ) . Using Equation ( 24 ) , we can show that ( 32 ) L0≈\|\|αt−αf\|\|22 as in Equation ( 9 ) . Hence: ( 33 ) ∂L0∂W0≈−kb ( αt−αf ) ϕmaxσ′ ( V0¯f ) ∘sinput¯f∂L0∂b0≈−kb ( αt−αf ) ϕmaxσ′ ( V0¯f ) where kb is given by: ( 34 ) kb=gbgl+gb+ga Note that although ϕ0∗ is a function of W0 and b0 , we do not differentiate this term with respect to the weights and biases . Instead , we treat ϕ0∗ as a fixed state for the hidden layer neurons to learn to reproduce . Basal weights are updated in order to descend this approximation of the gradient: ( 35 ) W0→W0−η0P0∂L0∂W0b0→b0−η0P0∂L0∂b0 Again , we assume that the activity during the target phase is fixed , so no derivatives are taken with respect to voltages or firing-rates during the target phase . Importantly , this update rule is spatially local for the hidden layer neurons . It consists essentially of three terms , ( 1 ) the difference in the plateau potentials for the target and forward phases ( αt−αf ) , ( 2 ) the derivative of the spike rate function ( ϕmaxσ′ ( V0¯f ) ) , and ( 3 ) the filtered presynaptic spike trains ( sinput¯f ) . All three of these terms are values that a real neuron could theoretically calculate using some combination of molecular synaptic tags , calcium currents , and back-propagating action potentials . One aspect of this update rule that is biologically questionable , though , is the use of the term ( αt−αf ) . This requires a difference between plateau potentials that are separated by tens of milliseconds . How could such a signal be used by basal dendrite synapses to guide their updates ? Plateau potentials can drive bursts of spikes ( Larkum et al . , 1999 ) , which can propagate to basal dendrites ( Kampa and Stuart , 2006 ) . Since the plateau potentials are similar to rate variables ( i . e . a sigmoid applied to the voltage ) , the number of spikes during the bursts , Nf=[N1f , N2f , . . . , Nmf] and Nt=[N1t , N2t , . . . , Nmt] , for the forward and target plateaus , respectively , could be sampled from a Poisson distribution with rate parameter equal to the plateau potential level: ( 36 ) Nf∼Poisson ( αf ) Nt∼Poisson ( αt ) If the distinct phases ( forward and target ) were marked by some global signal , ϕ⁢ ( t ) , that was communicated to all of the neurons , for example a neuromodulatory signal , the phase of a global oscillation , or some blanket inhibition signal , then we can imagine an internal cellular memory mechanism in the basal dendrites of the it⁢h neuron , Mi ( e . g . a molecular signal like the activity of an enzyme , the phosphorylation level of some protein , or the amount of calcium released from intracellular stores ) , which could be differentially sensitive to the inter-spike interval of bursts , depending on ϕ . So , for example , if we define: ( 37 ) ϕ ( t ) ={−1 , if in the forward phase , i . e . x=f1 , if in the target phase , i . e . x=tdMi ( t ) dt∝ϕ ( t ) Nix where x indicates the forward or target phase . Then , the change in Mi from before the bursts occur to afterwards would be , on average , proportional to the difference ( αt−αf ) , and could be used to calculate the weight updates . However , this is highly speculative , and it is not clear that such a mechanism would be present in real neurons . We have outlined the mathematics here to make the reality of implementing the current model explicit , but we would predict that the brain would have some alternative method for calculating differences between top-down inputs at different times , for example by using somatostatin positive interneurons that are preferentially sensitive to bursts and which target the apical dendrite ( Silberberg and Markram , 2007 ) . We are ultimately agnostic as to this mechanism , and so , it was not included in the current model . In order to extend our algorithm to deeper networks with multiple hidden layers , our model incorporates direct synaptic connections from the output layer to each hidden layer . Thus , each hidden layer receives feedback from the output layer through its own separate set of fixed , random weights . For example , in a network with two hidden layers , both layers receive the feedback from the output layer at their apical dendrites through backward weights Y0 and Y1 . The local targets at each layer are then given by: ( 38 ) ϕ2∗=ϕ2¯t ( 39 ) ϕ1∗=ϕ1¯t+α1t−α1f ( 40 ) ϕ0∗=ϕ0¯t+α0t−α0f where the superscripts 0 and 1 denote the first and second hidden layers , respectively , and the superscript 2 denotes the output layer . The local loss functions at each layer are: ( 41 ) L2=\|\|ϕ2∗−ϕmaxσ ( V2¯f ) \|\|22L1=\|\|ϕ1∗−ϕmaxσ ( V1¯f ) \|\|22L0=\|\|ϕ0∗−ϕmaxσ ( V0¯f ) \|\|22 where L2 is the loss at the output layer . The learning rules used by the hidden layers in this scenario are the same as in the case with one hidden layer . For each of the three network sizes that we present in this paper , a grid search was performed in order to find good learning rates . We set the learning rate for each layer by stepping through the range [0 . 1 , 0 . 3] with a step size of 0 . 02 . For each combination of learning rates , a neural network was trained for one epoch on the 60 , 000 training examples , after which the network was tested on 10 , 000 test images . The learning rates that gave the best performance on the test set after an epoch of training were used as a basis for a second grid search around these learning rates that used a smaller step size of 0 . 01 . From this , the learning rates that gave the best test performance after 20 epochs were chosen as our learning rates for that network size . In all of our simulations , we used a learning rate of 0 . 19 for a network with no hidden layers , learning rates of 0 . 21 ( output and hidden ) for a network with one hidden layer , and learning rates of 0 . 23 ( hidden layers ) and 0 . 12 ( output layer ) for a network with two hidden layers . All networks with one hidden layer had 500 hidden layer neurons , and all networks with two hidden layers had 500 neurons in the first hidden layer and 100 neurons in the second hidden layer . For all simulations described in this paper , the neural networks were trained on classifying handwritten digits using the MNIST database of 28 pixel × 28 pixel images . Initial feedforward and feedback weights were chosen randomly from a uniform distribution over a range that was calculated to produce voltages in the dendrites between -6 - 12 V . Prior to training , we tested a network’s initial performance on a set of 10 , 000 test examples . This set of images was shuffled at the beginning of testing , and each example was shown to the network in sequence . Each input image was encoded into Poisson spiking activity of the 784 input neurons representing each pixel of the image . The firing rate of an input neuron was proportional to the brightness of the pixel that it represents ( with spike rates between [0 - ϕmax] . The spiking activity of each of the 784 input neurons was received by the neurons in the first hidden layer . For each test image , the network underwent only a forward phase . At the end of this phase , the network’s classification of the input image was given by the neuron in the output layer with the greatest somatic potential ( and therefore the greatest spike rate ) . The network’s classification was compared to the target classification . After classifying all 10 , 000 testing examples , the network’s classification error was given by the percentage of examples that it did not classify correctly . Following the initial test , training of the neural network was done in an on-line fashion . All 60 , 000 training images were randomly shuffled at the start of each training epoch . The network was then shown each training image in sequence , undergoing a forward phase ending with a plateau potential , and a target phase ending with another plateau potential . All feedforward weights were then updated at the end of the target phase . At the end of the epoch ( after all 60 , 000 images were shown to the network ) , the network was again tested on the 10 , 000 test examples . The network was trained for up to 60 epochs . For each training example , a minimum length of 50 ms was used for each of the forward and target phases . The lengths of the forward and target training phases were determined by adding their minimum length to an extra length term , which was chosen randomly from a Wald distribution with a mean of 2 ms and scale factor of 1 . During testing , a fixed length of 500 ms was used for the forward transmit phase . Average forward and target phase voltages were calculated after a settle duration of Δ⁢ts=30 ms from the start of the phase . For simulations with randomly sampled plateau potential times ( Figure 5—figure supplement 1 ) , the time at which each neuron’s plateau potential occurred was randomly sampled from a folded normal distribution ( μ=0 , σ2=3 ) that was truncated ( max=5 ) such that plateau potentials occurred between 0 ms and 5 ms before the start of the next phase . In this scenario , the average apical voltage in the last 30 ms was averaged in the calculation of the plateau potential for a particular neuron . The time-step used for simulations was d⁢t=1 ms . At each time-step , the network’s state was updated bottom-to-top beginning with the first hidden layer and ending with the output layer . For each layer , dendritic potentials were updated , followed by somatic potentials , and finally their spiking activity . Table 1 lists the simulation parameters and the values that were used in the figures presented . All code was written using the Python programming language version 2 . 7 ( RRID: SCR_008394 ) with the NumPy ( RRID: SCR_008633 ) and SciPy ( RRID: SCR_008058 ) libraries . The code is open source and is freely available at https://github . com/jordan-g/Segregated-Dendrite-Deep-Learning ( Guerguiev , 2017 ) . The data used to train the network was from the Mixed National Institute of Standards and Technology ( MNIST ) database , which is a modification of the original database from the National Institute of Standards and Technology ( RRID: SCR_006440 ) ( Lecun et al . , 1998 ) . The MNIST database can be found at http://yann . lecun . com/exdb/mnist/ . Some of the simulations were run on the SciNet High-Performance Computing platform ( Loken et al . , 2010 ) . Theorem 1 shows that if we use a target ϕ~0∗=ϕ0¯f+σ ( Yϕ1¯t ) −σ ( Yϕmaxσ ( kdW1ϕ0¯f ) ) for the hidden layer , there is a guarantee that the hidden layer approaching this target will also push the upper layer closer to its target ϕ1∗ , if certain other conditions are met . Our specific choice of ϕ0∗ defined in the results ( Equation ( 8 ) ) approximates this target rate vector using variables that are accessible to the hidden layer units . If neuronal units calculate averages after the network has reached a steady state and the firing rates of neurons are in the linear region of the sigmoid function , ϕmaxσ ( V1¯f ) ≈ϕ1¯f . Using Lemma 1 , E[V1¯f]≈kdW1ϕ0¯f and E[V0a¯f]≈Yϕ1¯f . If we assume that V1¯f≈E[V1¯f] and V0a¯f≈E[V0a¯f] , which is true on average , then: ( 42 ) αf=σ ( V0a¯f ) ≈σ ( Yϕ1¯f ) ≈σ ( Yϕmaxσ ( V1¯f ) ) ≈σ ( Yϕmaxσ ( kdW1ϕ0¯f ) ) and: ( 43 ) αt=σ ( V0a¯t ) ≈σ ( Yϕ1¯t ) Therefore , ϕ0∗≈ϕ~0∗ . Thus , our hidden layer targets ensure that our model employs a learning rule similar to difference target propagation that approximates the necessary conditions to guarantee error convergence . Theorem 1 had to rely on the equivalence between the average spike rates of the neurons and their filtered spike trains . Here , we prove a lemma showing that this equivalence does indeed hold as long as the integration time is long enough relative to the synaptic time constants ts and tL .	Artificial intelligence has made major progress in recent years thanks to a technique known as deep learning , which works by mimicking the human brain . When computers employ deep learning , they learn by using networks made up of many layers of simulated neurons . Deep learning has opened the door to computers with human – or even super-human – levels of skill in recognizing images , processing speech and controlling vehicles . But many neuroscientists are skeptical about whether the brain itself performs deep learning . The patterns of activity that occur in computer networks during deep learning resemble those seen in human brains . But some features of deep learning seem incompatible with how the brain works . Moreover , neurons in artificial networks are much simpler than our own neurons . For instance , in the region of the brain responsible for thinking and planning , most neurons have complex tree-like shapes . Each cell has ‘roots’ deep inside the brain and ‘branches’ close to the surface . By contrast , simulated neurons have a uniform structure . To find out whether networks made up of more realistic simulated neurons could be used to make deep learning more biologically realistic , Guerguiev et al . designed artificial neurons with two compartments , similar to the ‘roots’ and ‘branches’ . The network learned to recognize hand-written digits more easily when it had many layers than when it had only a few . This shows that artificial neurons more like those in the brain can enable deep learning . It even suggests that our own neurons may have evolved their shape to support this process . If confirmed , the link between neuronal shape and deep learning could help us develop better brain-computer interfaces . These allow people to use their brain activity to control devices such as artificial limbs . Despite advances in computing , we are still superior to computers when it comes to learning . Understanding how our own brains show deep learning could thus help us develop better , more human-like artificial intelligence in the future .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "computational", "and", "systems", "biology", "neuroscience" ]	2017	Towards deep learning with segregated dendrites
Newts have the ability to repeatedly regenerate their lens even during ageing . However , it is unclear whether this regeneration reflects an undisturbed genetic activity . To answer this question , we compared the transcriptomes of lenses , irises and tails from aged newts that had undergone lens regeneration 19 times with the equivalent tissues from young newts that had never experienced lens regeneration . Our analysis indicates that repeatedly regenerated lenses showed a robust transcriptional program comparable to young never-regenerated lenses . In contrast , the tail , which was never regenerated , showed gene expression signatures of ageing . Our analysis strongly suggests that , with respect to gene expression , the regenerated lenses have not deviated from a robust transcriptional program even after multiple events of regeneration throughout the life of the newt . In addition , our study provides a new paradigm in biology , and establishes the newt as a key model for the study of regeneration in relation to ageing . Newts are among the few vertebrates that possess the remarkable ability to regenerate tissues , organs , and body parts , including limbs , tails and eye tissue ( Sanchez Alvarado and Tsonis , 2006 ) . Importantly , newts appear to regenerate using cells recruited locally from the site of the insult . For example , during limb regeneration , the cells at the site of amputation , such as muscle and bone cells , dedifferentiate and then redifferentiate to reconstruct the lost part ( Tsonis , 1996; Sandoval-Guzmán et al . , 2014 ) . For this process to occur , cells from the original organ must remain to provide a source for regeneration; if the entire limb is removed , no regeneration occurs . However , regeneration of the lens is different in two key ways , providing additional experimental benefits . First , regeneration is possible following complete removal of the lens , and thus whole-organ regeneration occurs . Second , the lens is regenerated from a different tissue , that is , the pigment epithelial cells ( PECs ) of the dorsal iris , via transdifferentiation rather than from the remaining lens tissue ( Henry and Tsonis , 2010; Barbosa-Sabanero et al . , 2012 ) . Because of these unparalleled regenerative traits , newts may provide answers that regenerative medicine is presently seeking ( Baddour et al . , 2012 ) . A fundamental question is whether newt regenerative ability declines with age or repeated insult . To answer this question , we undertook a long-term study of lens regeneration . Using Japanese newts ( Cynops pyrrhogaster ) , lens regeneration was followed for 16 years . During this period , lenses were removed from the same animals 18 times . Previously , it was shown that the 17 and 18 times regenerated lenses , which were obtained from the second-to-last and last collections , respectively , were virtually identical to the intact lenses removed from full-grown , 14-year-old newts produced from fertilized eggs that had never undergone a lentectomy or lens regeneration . Throughout this 16-year period , the rate and stage of regeneration was carefully evaluated , and no significant delay in the lens regeneration process was observed in any of the 18 repetitions ( Eguchi et al . , 2011 ) . At the gross anatomical level , the experimental and control lenses were of the same size and transparency . The lens fiber organization appeared normal , with the nucleus containing primary fibers and the cortex containing secondary fibers . Most importantly , the gene expression patterns of the experimental and control lenses were very similar . The genes examined included crystallins and transcription factors that regulate crystallin expression , such as Pax-6 , Sox2 , MafB , Sox1 , Prox-1 , and Delta , all of which participate in lens development and lens fiber differentiation and are thus involved in normal lens homeostasis . The study also established that the age of the animal does not affect its regenerative capacity ( see also Sousounis et al . , 2014 ) . The newts were estimated to be at least 14 years old at the onset of the project and thus would have been at least 30 years old at the end of the study . Because the reported lifespan of the Japanese newt is 25 years ( Goin et al . , 1978 ) , this group truly represents an old population . These results raise the question as to whether the repeatedly regenerated lens of a 30-year-old newt retains the biological signature of a 14-year-old's lens . Especially this is of interest if one considers the relation of regeneration and ageing . To investigate this possibility , we undertook a transcriptomic analysis of lenses that had been regenerated 19 times along with appropriate controls . The Japanese newt C . pyrrhogaster was used in this study . The experimental and control groups of newts were as follows . The experimental group ( referred to as #19 throughout ) comprised of 32-year-old newts whose lenses had been removed 19 times . These lenses were regenerated 19 times and removed 18 years after the start of the project . The control group ( referred to as #0 ) consisted of 14-year-old newts that had their original lenses ( i . e . , non-regenerated lenses ) ( Figure 1 ) . The tissue collected from the experimental group consisted of #19 lenses , #19 dorsal irises , and #19 tails ( n=5 for each tissue type ) . The dorsal iris was sampled because this tissue gives rise to the regenerated lens , which implies that the dorsal iris had also been regenerated/replenished 19 times . The tails were included as an aged tissue that had never been regenerated . The corresponding tissues were also sampled from #0 newts . In total , 30 samples were prepared for RNA sequencing and transcriptomic analysis . 10 . 7554/eLife . 09594 . 003Figure 1 . Experimental overview . Arrows depict the number of repeated lentectomies performed over a period of 18 years . Panel shows the process of lens regeneration that occurred after each lens removal highlighted as a single arrow . At the end of the experiment , lens , iris , and tail tissues were collected from both old newts that had regenerated their lenses 19 times and young newts that had never experienced lentectomy . Di: Dorsal iris; Vi: Ventral iris; L: Lens . DOI: http://dx . doi . org/10 . 7554/eLife . 09594 . 003 We generated nearly 4 . 5 billion reads , with approximately 150 million reads per sample . The reads were of high quality ( >97% passed TAILING:30 criteria using Trimmomatic [Bolger et al . , 2014] ) and included very few duplicates ( approximately 2% , as assessed using FastUniq [Xu et al . , 2012] ) . Trinity was used for de novo assembly of the reference transcriptome ( see Materials and methods ) , which was composed of 4 . 3 million contigs and isoforms ( referred to as transcripts or genes throughout ) . We used NCBI BLASTx to annotate 133 , 503 ( 73 , 233 contigs ) of the transcripts against the human reference proteome obtained from UniProt ( e-value<1E-10 ) . Remarkably , 58 , 331 of these transcripts were related to human transposons ( 43 . 7% ) . In total , we obtained 15 , 077 non-redundant annotations representing nearly 75% of all human genes . Reads were used to compute the relative abundance of transcripts in each sample . Transcripts that showed the most significant variability between samples are shown as a heat map ( Figure 2A ) . To identify highly expressed genes in the three different tissues we focused only on the annotated transcripts and considered the ones with >1000 fragment per kilobase per millions of reads ( FPKM ) . In other words , which were the genes with the highest expression in each tissue irrespective of treatment ( young or old ) ( Figure 2B , Supplementary file 1 ) . As expected alpha- , beta- , and gamma-crystallin genes ( CRY ) were found to be the highest expressed genes in lens samples . Crystallins are known to be the major structural protein of the lens ( Masters et al . , 1977 ) . The same dataset also contained the lens fiber major intrinsic protein MIP , and phakinin ( BFSP2 ) , genes highly expressed in lenses ( Figure 2B; orange ) ( Broekhuyse and Kuhlmann , 1974; Maisel and Perry , 1972 ) . Ornithine decarboxylase antizyme 1 ( OAZ1 ) , hemoglobin subunit alpha ( HBA1 ) , and cell division control protein 42 homolog ( CDC42 ) were the highest expressed genes in the iris samples ( Figure 2B; red ) . Expectedly , keratin ( KRT ) and ribosomal protein genes were the ones with the highest expression in tails ( Figure 2B; yellow ) . Keratin proteins are known to be expressed in the skin . Six genes , five coding for ribosomal proteins and one for the ferritin heavy chain , were found to be the most expressed in all tissue types ( Figure 2B; purple ) . In a different analysis , we identified genes exclusively or preferentially expressed in a particular tissue , when compared with the others . We sorted genes that were adequately expressed in a given tissue ( FPKM>100 ) and were 100-fold more expressed in one versus the other tissues ( Figure 2C , Supplementary file 1 ) . Genes in the lens dataset included crystallins , lens fiber membrane intrinsic protein LIM2 , filensin ( BFSP1 ) , and tudor domain-containing protein 7 ( TDRD7 ) , among others ( Figure 2C; orange ) . LIM2 , BFSP1 , and TDRD7 are known to be expressed in lenses ( Church and Wang , 1993; Hess et al . , 1998; Lachke et al . , 2011 ) . Iris preferentially expressed retinal pigment epithelial ( RPE ) -retinal G protein-coupled receptor ( RGR ) , a protein found in pigmented cells of the retina ( Figure 2C; red ) ( Tao et al . , 1998 ) , while tail samples expressed keratins , creatine kinase M-type ( CKM ) , resistin ( RETN ) , and alpha skeletal muscle actin ( ACTA1 ) ( Figure 2C; yellow ) , proteins found in muscle , skin , and adipose tissue ( Way et al . , 2001; Nowak et al . , 1999 ) . Many of the genes identified by these two methods are known to be enriched in the same or equivalent ( e . g . , tail to be a composition of muscle , fat , skin , and spinal cord ) tissues in other organisms including humans . These genes are also known to be involved with disease states including lens cataracts or ageing . 10 . 7554/eLife . 09594 . 004Figure 2 . Gene expression among tissue samples . ( A ) Heat map constructed from the expression profiles of the 30 sequenced samples . CT , CL , CI: #0 ( control ) tail , lens , and iris , respectively . ET , EL , EI: #19 ( experimental ) tail , lens , and iris , respectively . Genes selected based on the following parameters: p-value<=0 . 001 and log2 ( FC ) >=2 . Note the nearly uniform pattern between the #0 and #19 lens samples which indicates no differences between non-regenerated young lenses and repeatedly regenerated lenses from aged newts . ( B ) Highly expressed genes in each tissue type irrespective of age . Red , orange , and yellow colors denote genes from iris , lens , and tail samples , respectively . Comparisons are visualized using a venn graph while non-redundant annotations are highlighted using boxes including the corresponding gene names . Purple color is used for highly expressed genes in all three samples . ( C ) Genes that are preferentially expressed in a given tissue versus the others are denoted using the same color code as in B . The different tissues are indicated using a cartoon newt and an enlarged cross-sectioned eye . DOI: http://dx . doi . org/10 . 7554/eLife . 09594 . 004 Differential gene expression analysis between #0 and #19 equivalent tissues was performed using edgeR ( Robinson et al . , 2010 ) . This analysis provided us with genes that their abundance changed during ageing and repetitive lens regeneration . No genes were found to differ significantly in their expression between the #19 and #0 lens samples ( false discovery rate [FDR] <0 . 05 and fold change [FC] >2; Supplementary file 2 ) . In the iris samples , we found 311 ( 54 of these annotated ) genes with FDR<0 . 05 and FC>2 ( Figure 3A and Supplementary file 3 ) . Even greater differences in gene expression were observed for the tail samples . We found 4204 ( 780 of these annotated ) genes with FDR<0 . 05 and FC >2 ( Figure 3B and Supplementary file 4 ) . In our experimental design , tail samples were collected in order to provide a tissue that was never amputated or regenerated from the same animals where repetitive lentectomy was performed during the last 19 years . Gene expression comparisons between young #0 tails and old #19 tails were conducted to prove that amphibian gene expression signatures change over time as tissues age . To begin with , we studied the roles of the differentially regulated genes in the tail samples by assigning Gene Ontology ( GO ) terms based on their biological processes , molecular functions , and sub-cellular localization ( Supplementary file 5 ) . Enrichment analysis revealed that GO terms related to translation , electron transport chain , oxidation reduction , and mitochondrion were enriched in the group of down-regulated genes in the #19 tail samples ( Figure 3C; green bars; FDR<0 . 05 , Supplementary file 5 ) . As will be discussed later , iris samples also showed enrichment of electron transport chain genes in the down-regulation dataset . Down-regulation of electron transport chain-associated genes is a well-established signature of ageing in many vertebrate animal models and flies ( López-Otín et al . , 2013; Zahn et al . , 2006 ) . To further illustrate this , we data-mined genes with GO terms related to ageing and/or senescence , which were differentially regulated between #0 and #19 tail samples and found 16 genes ( Figure 3D ) . These data suggest that the transcriptomic profile of newt tails and irises changed over time and showed signs of ageing . Since we observed changes in the abundance of several genes in tail and iris samples , we next asked whether these changes were reflected in the transcriptomic complexity of these tissues . By sorting genes based on their relative abundance we plotted the percent contribution of each gene in a cumulative way ( Figure 3E , Supplementary file 1 ) . This method identifies how many genes are sharing the total transcriptomic output of each sample; for example , if 50% of the transcriptomic output is shared by 100 genes , the underlying profile is relatively simple and the line on the plot will appear flat ( orange; Figure 3E ) . On the other hand , if 1000 genes are sharing 50% of the transcriptomic output , the profile of the tissue is more complex having a steeper line ( red; Figure 3E ) ( Mele et al . , 2015 ) . This analysis reveals that iris is the most complex tissue followed by the tail and the lens ( Figure 3E ) . Interestingly , #19 iris and tail samples showed slightly increased complexity versus the respective #0 samples . #19 lens samples showed the lowest increase corroborating our previous data indicating no significant changes between #0 and #19 lens samples maintaining a stable gene expression profile . Using sample correlation matrix plot we further validated our gene expression data ( Figure 4A , Supplementary file 6 ) . When the genes from the #19 lenses ( EL1–EL5 ) were compared with those from the #0 lenses ( CL1–CL5 ) , a nearly uniform pattern was obtained across all 10 samples . This pattern similarity was also evidenced by the lack of segregation of the experimental and control lens samples via cladograms . These results indicate that these samples were highly correlated for their overall gene expression pattern . However , a different pattern emerged when comparing the irises and tails between the #19 and #0 groups . Areas on the sample correlation matrix plot showed the characteristic four-boxed color pattern indicating differences in the overall correlation between #19 and #0 samples ( Figure 4A ) . Similar results were obtained by using jackknifing and random 20% sampling methods ( Supplementary files 7 and 8 ) . By dissecting the within tissue correlation values and comparing them , it was evident that the five biological replicates of each tissue had high correlation values among them ( Figure 4B; solid colors and big-dotted bars ) . However , when comparisons are made between #0 and #19 samples ( Figure 4B; small-dotted bars ) , the correlation drops with lens samples showing the least decrease ( orange bars ) compared to iris ( red bars ) and tail ( yellow bars ) . We also performed a comparison between tissues and found that iris and lens tissues were the most related while lens and tail tissues the least related based on the genes expressed ( Figure 4C ) . 10 . 7554/eLife . 09594 . 005Figure 3 . Differential gene expression between #19 and #0 tissues . ( A ) Volcano plot for the #19 versus #0 iris samples . ( B ) Volcano plot for the #19 versus #0 tail samples . Differentially expressed genes ( false discovery rate [FDR]<0 . 05 and fold change [FC]>2 ) are depicted in cyan . Tail samples , which never experienced regeneration , showed the most differentially expressed genes . Iris samples , which as the source of lens regeneration have experienced some degree of regeneration/replenishment , showed a reduced number of differentially expressed genes and an intermediate ageing profile . ( C ) Selected enriched ( FDR <0 . 05 ) Gene Onthology ( GO ) terms in tail samples plotted based on their fold enrichment . Green-colored bars mark gene groups that are down-regulated in the #19 samples . Yellow line marks fold enrichment of 1 . Electron transport chain is one of the functional group enriched in the down-regulated group , a well-documented ageing signature in other vertebrates . ( D ) Genes selected for their role in ageing and/or senescence and plotted based on their fold change between #19 and #0 tail samples . Green and red bars mark down-regulated or up-regulated genes in #19 samples , respectively . Yellow line marks log2 ( FC ) of 1 . These data provide additional evidence of ageing signs in our #19 tails samples . ( E ) Transcriptomic complexity between #0 and #19 tissues . Red , orange , and yellow denotes iris , lens , and tail samples , respectively . Solid and dotted lines represent tissues from #0 and #19 newts , respectively . In this graph , genes were sorted based on their expression and plotted based on their cumulative percent contribution to the overall transcriptomal output . Iris had the most complex transcriptome by having more genes contributing to the overall output ( steeper line ) , followed by the tail and lens . Tissues from #19 newts showed slightly increased transcriptomic complexity versus their #0 counterparts ( enlarged insert ) . Lens showed the least increase in complexity which further supports that lens regeneration is a robust process that can faithfully proceed throughout life . DOI: http://dx . doi . org/10 . 7554/eLife . 09594 . 00510 . 7554/eLife . 09594 . 006Figure 4 . Sample correlations between the 30 samples . ( A ) Sample correlation matrix plot . Note the uniform red color between #19 ( experimental lens , EL ) and #0 ( control lens , CL ) , indicating high correlation between them . Iris and tail #0 and #19 samples segregate clearly creating a characteristic four-box pattern in the two edges of the plot . #19 and #0 lens samples are so similar that the cladogram clusters them together . EL4 sample exhibits the least correlation among the #19 newt lenses . ( B ) Within tissue correlation plotted as bar graphs for better visualization . Solid colored bars ( inter-#0 correlations ) and big-dotted bars ( inter-#19 correlations ) showed very high values . Intra #0 - #19 correlation values were lower except those in lens samples . These data indicate that #0 and #19 lens samples are very similar in regards to gene expression versus equivalent comparisons in iris and tail samples . ( C ) Correlations between tissues illustrated as box plots . Iris-lens gene expression correlations were the highest followed by iris-tail and lens-tail . DOI: http://dx . doi . org/10 . 7554/eLife . 09594 . 006 Sample correlation matrix plot also revealed that sample EL4 ( one from #19 old newts ) did not strongly correlate with the other lens samples . The correlation values of this sample were the lowest for the within tissue comparisons performed ( Supplementary file 6 ) . However , the values were higher than the within tissue correlations of #0 and #19 in iris and tail samples . Nevertheless we wanted to test whether this sample showed signs of ageing . To accomplish that we isolated all crystallin-associated genes expressed in the EL4 sample and compared them to the average FPKM values of the other #19 lens samples ( Supplementary file 9 ) . We chose crystallin genes because crystallins are the major structural proteins of this tissue and down-regulation is often linked to disease states in humans and mice ( Sousounis and Tsonis , 2012 ) . Our analysis revealed that 30% of the transcripts associated with crystallin genes were deregulated in the EL4 sample . However , most of them showed higher expression in the EL4 sample than the other #19 lens samples , an expression pattern that does not match a pathological profile . For example , transcript c1474631_g1 corresponding to gamma crystallin B , a highly expressed gene in the lens , showed a more than twofold up-regulation ( Supplementary file 9 ) . Overall , our analysis showed that EL4 had the weakest association among the #19 lens samples and #0 young control lenses; however , differences in gene expression were not strongly associated with ageing or disease . Iris is the source of lens regeneration in newts . After lentectomy the whole lens is removed and dorsal iris PECs transdifferentiate to lens cells . By collecting iris tissue for RNA sequencing we studied how repetitive regeneration and ageing affected its transcriptomic profile . As mentioned earlier , 311 genes were found to be differentially affected when #0 and #19 iris samples were compared ( Figure 3A , Supplementary file 3 ) . Iris , as the cellular source of lens regeneration , should have reflected , at least in part , this deregulated profile to the regenerate . Surprisingly though , the lens samples did not have any significantly deregulated genes suggesting that the age-regulated profile of the iris was corrected during the regeneration process . To further highlight these differences , we compared the regulated genes by first plotting the FPKM values of #0 and #19 iris and lens samples and applying linear regression ( Figure 5A , Supplementary file 10 ) . As expected lens FPKM values were highly correlated ( r = 0 . 9982 ) and differed between them on average 27% ( slope; m = 0 . 7293 ) ( Figure 5A; orange ) . On the other hand , iris FPKM values were not correlated and differed completely ( r = 0 . 0070 , m = 0 . 004 , Figure 5A; red ) . These data clearly indicate that deregulation of these genes were corrected in the regenerated lens . When the function of the annotated genes was investigated , we found that electron transport chain was enriched in the group of genes that were down-regulated in the #19 iris samples ( Figure 5B; green bars , Supplementary file 11 ) . As indicated above , this is a well-established ageing signature and suggests that #19 repetitive regenerated lenses did not inherit it during the transdifferentiation process . 10 . 7554/eLife . 09594 . 007Figure 5 . Correction of iris-regulated gene expression in the regenerated lens . ( A ) The 311 genes used for this analysis were differentially regulated in #0 versus #19 iris samples . Since lens is regenerated from the dorsal iris , the question arises whether or not these differences in gene expression are reflected in the regenerated lenses . Average fragment per kilobase per millions of reads ( FPKM ) values of these genes from iris and lens samples were plotted in the same graph . Red and orange colors mark the iris and lens , respectively . Linear regression analysis revealed that lens genes are more correlated ( r = 0 . 9982 ) with m = 0 . 7293 while iris genes are not correlated ( r = 0 . 0070 ) as expected . Based on the slope ( m ) values ( where m = 1 is the absolute perfect when #0 and #19 values are identical ) , these data indicate that the #0 and #19 lens FPKM values differ approximately 27% while the equivalent iris samples are completely different . ( B ) Gene Ontology enrichment analysis of genes differentially expressed between #0 and #19 iris samples . Green and red bars denote gene groups in the down-regulated and up-regulated datasets , respectively . Yellow line marks fold enrichment of 1 . Note that as with the tail samples , electron transport chain is also down-regulated in these samples , a sign of ageing . DOI: http://dx . doi . org/10 . 7554/eLife . 09594 . 007 Overall , our results point out that repeated lens regeneration employs a robust transcriptomic program that is maintained throughout life , an attribute not found in the never-regenerated tail tissue from the same animals . In addition , the fact that #19 lenses did not show down-regulation of genes related to electron transport chain , a well-established signature of ageing revealed in #19 iris and tail samples , suggests that repeated regeneration might ameliorate age-regulated gene changes . Newts have the remarkable ability to regenerate their lenses after repeated insults throughout their lifespan . In order to gain additional insights about the molecular interactions underlying this trait , we started by exploring highly and uniquely expressed genes in each of the tissues collected; iris , lens , and tail . We found that genes preferentially and highly expressed in lens or tail are known to be expressed in other vertebrates . More importantly for the lens , crystallins , phakinin , filensin , tudor domain-containing protein 7 , lens fiber major intrinsic protein MIP , and lens fiber membrane intrinsic protein LIM2 are major structural/molecular components of the lens and linked to age-related lens diseases when deregulated ( Sousounis et al . , 2014; Lachke et al . , 2011; Sousounis and Tsonis , 2012; Jakobs et al . , 2000; Ramachandran et al . , 2007; Berry et al . , 2000; Pras et al . , 2002 ) . Thus these genes are also good markers should the newt lens age . However , when gene expression patterns of the #19 and #0 lenses were compared , no significant differences were found . In addition , compared with the non-regenerated lenses from younger animals , the #19 lenses showed no differences in size , transparency , or overall fiber structure ( Figure 6 ) and during this 18-year experiment , cataracts were never observed in the regenerated lenses . This important result clearly demonstrates that repeated lentectomies and ageing have no effect on lens regeneration , development , or homeostasis . In addition to the aforementioned genetic causes of cataracts , it is well documented that lenses are severely affected by ageing with marked changes in expression of mitochondrion electron transport chain , oxidative stress and crystallin genes as well as alterations of the fiber structure and homeostasis leading to cataracts ( Su et al . , 2015; Tsentalovich et al . , 2015; Wei et al . , 2015; Linetsky et al . , 2014; Petrash et al . , 2013; Sousounis and Tsonis , 2012 ) . Based on these observations , the newt #19 lenses show a robust transcriptional program as they undergo multiple events of regeneration throughout their lives . Conceptually the process of transdifferentiation might provide robustness to the process of regeneration . In relation to this , it is interesting to note that the transdifferentiation ability of even aged human iris PECs is retained in vitro . Previous studies have shown that such a cell line from an 80-year-old human is capable of transdifferentiating into lens ( Yun et al . , 2015 ) . 10 . 7554/eLife . 09594 . 008Figure 6 . Lenses from #0 control ( left ) and #19 experimental ( right ) newts . Note that the size , fiber arrangement , and transparency of both samples are normal . DOI: http://dx . doi . org/10 . 7554/eLife . 09594 . 008 In contrast to the patterns observed in the lenses , a comparison of gene expression between the #19 and #0 tails revealed striking differences . Thousands of genes were significantly differentially regulated . Among the most highly deregulated genes in the #19 tail samples were those encoding proteins that are part of the electron transport chain ( Supplementary file 5 ) . Down-regulation of these gene-sets are part of an established ageing signature in other vertebrates ( López-Otín et al . , 2013; Zahn et al . , 2006 ) . In addition , several ageing- and senescence-related genes were found to be deregulated in these samples . These findings indicate that the tails of the #19 newts show clear hallmarks of ageing . These observations provide in our opinion strong evidence that a robust transcriptional program ensues after an insult to guarantee that the regenerative ability in newts will not be thwarted with age . Although the differences in gene expression between the #19 and #0 tails and lenses were pronounced , the irises showed an intermediate pattern , especially with respect to the number of deregulated genes . Similarly to the #19 tails , #19 iris samples down-regulated electron transport chain genes , a sign of ageing . Iris is the source of lens regeneration and ageing of this tissue may hinder the process . Since #19 lenses are comparable to #0 and that is not the case for the iris , there should be a mechanism that amends or corrects the profile from the source ( dorsal iris ) to the regenerated tissue ( lens ) . Supplementary file 9 lists genes shown to be differentially regulated in the iris samples and their potential transcriptomic correction in the lens samples . All genes that were found to differ in the iris , the source of lens regeneration , were similar in the regenerate , the lens ( Figure 5 and Supplementary files 10 , 11 ) . For instance , contig c1229960_g1 corresponding to nicotinamide adenine dinucleotide ( NADH ) -ubiquinone oxidoreductase chain 1 , shows a low expression in the control iris samples , but it is highly expressed in the experimental iris samples . However , the same newts with iris tissues that showed this expression profile had regenerated lenses with low expression of the gene , similar to that of the control iris and lens rendering the difference non-significant . The gene expression differences observed in the iris may also be attributed to the fact that not all dorsal iris PECs contribute to the regeneration of the lost lens . After lentectomy regeneration occurs via dedifferentiation of the lower dorsal tip of the iris . These dorsal iris PECs are either replenished , or cells migrate there from other locations in the iris . Regardless , given that repeated lentectomies always trigger lens regeneration , it is clear that not the whole dorsal iris is eventually transformed into lens cells and that cell proliferation continuously provides the dorsal iris with PECs . Consequently , some parts of the dorsal iris are regenerated and might employ a transcriptomic program similar to that of young controls . On the other hand , other cells might not have this ability and eventually age , thus reflecting the intermediate ageing profile of this tissue . Nevertheless , the cellular or transcriptomic correction of the ageing profile observed in iris to the regenerated lens should be critical for the integrity of the newly formed organ ( Figure 7 ) . Recently it has been shown that there is significant turnover of senescent cells during newt limb regeneration . This might explain why newts can regenerate repeatedly their lost structures throughout their lives ( Yun et al . , 2015 ) , As such , the possibility exists that senescent cells are removed from the dorsal iris to ensure the correct process of lens regeneration . 10 . 7554/eLife . 09594 . 009Figure 7 . Summary of results from our transcriptomic comparisons between #19 and #0 newts . ( A ) Tail samples that had never experienced regeneration showed a marked deregulation of electron transport chain , mitochondrion , and ribosome genes , in #19 newts all signatures of ageing . On the contrary , lenses that were regenerated 19 times over a period of 18 years , showed a transcriptomic profile comparable to never-regenerated lenses from young newts . Iris showed an intermediate profile marked by deregulation of electron transport chain-related genes . ( B ) Regeneration versus ageing in newts . Triangles indicate the amount of regeneration activity ( in light blue , decreasing from left to right ) and ageing signatures as found by our transcriptomic analysis ( in hot pink , increasing from left to right ) of sampled #19 tissues . Regeneration initiates a robust transcriptomic program that can be faithfully restarted during repeated insult with no transcriptomic deregulation or molecular signatures of ageing . In our #19 newts , lenses had been fully removed and regenerated 19 times , thus having the highest regeneration activity and showed no signs of ageing . Iris , as the source of lens regeneration , has been regenerated/replenished after transdifferentiation to lens , thus showing some activity and an intermediate profile ( the asterisk indicates that iris is not regenerated completely ) . Tails were never removed or regenerated and showed the most deregulated genes and signatures of ageing compared to the young controls . DOI: http://dx . doi . org/10 . 7554/eLife . 09594 . 009 In this study we have compared the same tissues derived from young and old animals . Thus , the differences in the expression profiles were not attributed to the histological complexity . As also discussed above , to compare the ageing status of our collected tissues , our analysis included genes ( such as the ones involved in electron transport chain ) that are known to be expressed in the majority of cell types and deregulated during ageing . The molecular pathways related to ageing have been studied extensively in other animal models , particularly worms , flies and mice . The use of databases like AGEMAP ( a gene expression database for ageing in mice ) to make comparisons among species has proven informative for the field of ageing . Many of the genes that are regulated during ageing have been associated with the mitochondrial electron transport chain ( Zahn et al . , 2007; Signer and Morrison , 2013; Gomes et al . , 2013 ) . Another major regulatory pathway involves insulin signaling , which negatively regulates the FOXO transcription factor DAF-16 . This transcription factor regulates metabolism and oxidative stress by promoting antioxidant enzymes . The up-regulation of DAF-16 could enhance longevity ( Curran et al . , 2009; Curran and Ruvkun , 2007; Murphy et al . , 2003 ) . Our results suggest that the patterns of ageing in newts are similar to those of other species , particularly those related to the mitochondrial electron transport chain . Thus , it is conceivable that these mechanisms might also be involved in regeneration in newts . Consequently , our 18-year-long experiments provide data that render the newt an indispensable model for addressing issues of regeneration and ageing . All procedures were performed as described previously ( Sousounis et al . , 2014 ) . C . pyrrhogaster was used in this study: Five 32-year-old newts whose lenses had been removed 19 times over a period of 18 years , and five 14-year-old newts that had their original lenses . Tissues collected from every newt were lenses ( n = 5 ) , dorsal irises ( n = 5 ) , and tails ( n = 5 ) . Each tissue from every newt was appropriately labeled and placed in collection tubes . Tissues were stored in RNAlater solution ( Ambion , Chicago , Illinois , USA ) until RNA isolation . RNA was extracted using an RNeasy Plus Kit ( Qiagen , Valencia , CA , USA ) according to the manufacturer’s protocol . The input RNA was quantified with a Qubit fluorometric RNA HS assay ( Life Technologies , Grand Island , NY , USA ) . The samples were then analyzed on an Agilent Bioanalyzer using an RNA Pico assay to evaluate the quality . A total of 20 ng of each sample was used to synthesize cDNA using NuGEN Ovation RNA-Seq v2 kit . Libraries were made from 100 ng of cDNA using the NuGEN Ovation Ultralow Library System and then quantified with the Qubit fluorometric DNA HS assay and the Bioanalyzer DNA HS assay . KAPA qPCR was performed to quantify and pool the libraries for sequencing . The libraries were sequenced on a HiSeq 1500 using a 2 x 75 bp high output run . Raw sequencing reads has been deposited in the NCBI’s Sequence Read Archive ( SRA ) database ( BioProject accession: PRJNA288378 ) . Due to the unusually large size of the samples and limited computational resources , Trinity 20140413p1 was used for this project , and we used multi-step in silico normalization for the sequencing reads ( Grabherr et al . , 2011 ) . As the developer suggested , max_cov was set to 50 ( personal communication ) . The final assembly result gave an N50 of 430 , and the average contig length was 409 . All assembly work was performed on PSC ( Pittsburgh Supercomputing Center ) Blacklight which is an SGI UV 1000cc-NUMA shared-memory system comprising 256 blades . The 16 cores on each blade share 128Gb of local memory . After assembly , the original reads ( not in silico normalized ) were aligned to the Trinity transcripts to obtain abundance estimates using Bowtie 2 ( Langmead and Salzberg , 2012 ) . Then , RSEM software was used to estimate the expression levels based on the resulting alignments . After estimating abundance , we obtained the expression profiles for each sample , and the edgeR Bioconductor package was used to identify differentially expressed transcripts ( Robinson et al . , 2010 ) . edgeR analysis was carried out using the protocol of identifying differentially expressed features with biological replicates and counts matrix as abundance estimation pulled from RSEM as input . For the data appearing in Figures 2A and 4A differentially expressed genes ( p-value<=0 . 001 and log2 ( FC ) >=2 in at least one comparison pair ) were used . Euclidean distance and complete linkage were used to calculate the correlation . Two subsampling methods were also applied . Jackknife resampling was first used to estimate the variance of the correlation between each pair of samples by systematically leaving out one contig expression from the expression results matrix . We also random selected 20% of contigs to make a subsample , then calculated the correlation matrix of samples . The plots are shown in the Supplementary files 7 and 8 . The de novo assembled transcriptome was annotated against the human reference proteome ( e-value<1E-10 ) using NCBI BLASTx ( Altschul et al . , 1990; Looso et al . , 2013 ) . The annotated transcripts created the newt reference proteome from which all gene names were derived . Differentially regulated transcripts ( FDR<0 . 05 and FC >2 ) were mined from the raw edgeR output files and linked to the assigned annotation using custom Perl scripts . For all the analysis using annotated transcripts we used all potential isoform annotations in the testing and reference datasets . For data appearing in Figure 2B we selected annotated transcripts expressed more than 1000 average FPKM . The Venn diagram was made with Venny ( http://bioinfogp . cnb . csic . es/tools/venny/index . html ) and modified with Photoshop ( Adobe ) . For data in Figure 2C we used transcripts expressed more than 100 average FPKM in the tissue of interest and less than 100 average FPKM in the other tissues , while the fold change between them was more than 100 . For GO enrichment , the UniProt IDs of the differentially regulated gene groups were used as 'gene lists' in the DAVID 6 . 7 online functional annotation tool ( Huang et al . , 2008 , 2009 ) . We used the newt reference proteome as the source of background genes . We performed the enrichment analysis using the three default gene ontology categories . GO terms with FDR<0 . 05 were considered enriched . To mine genes related to ageing and/or senescence we searched for gene names with GO terms that contain 'age , , 'aging' , 'ageing' or 'senescence' and crossed them with our gene-sets . For the transcriptomic complexity graphs in Figure 3E , we sorted average FPKM values from all annotated transcripts individually for each tissue . The percent contribution to the total transcriptomic output was computed by dividing the average FPKM of a certain transcript to the sum FPKM of all transcripts in that tissue . Then transcripts were plotted from the least to the most expressed in a cumulative way ( Supplementary file 1 ) . To investigate potential signs of ageing in the EL4 sample we performed the following: EL4 genes were considered that deviate from the other #19 samples with the following formula: EL¯+2σ ( EL ) <EL4<EL¯-2σ ( EL ) where EL is the FPKM value of EL1 , EL2 , EL3 , and EL5 . Generally , genes were considered that were expressed in the samples if their FPKM value was more than 2 ( Supplementary file 9 ) . Linear regression analysis appearing in Figure 5A was performed with SigmaPlot 11 . 0 and Excel .	Newts are unusual animals because they are able to regenerate injured or lost body parts . To regenerate the lens in an eye , certain cells in the iris need to change into lens cells . In 2011 , a group of researchers reported the results of a 16-year long study of lens regeneration in Japanese newts . This study found that lenses from old newts that have undergone lens regeneration many times are structurally identical to those of young individuals that still have their original lenses . Also , many genes required to make lens proteins were expressed at similar levels in the lenses of the old and young newts . Therefore , even old newts retain the ability to fully regenerate their lenses . However , it is possible that the lenses in the old newts might show more subtle signs of ageing in the form of differences in the expression of other genes . Here , Sousounis et al . – including some of the researchers from the 2011 work – used an approach called transcriptomics to examine the patterns of gene expression in this group of newts in more detail . Sousounis et al . collected cells from the lenses , irises and tails of both the old and young newts . The experiments show that the patterns of gene expression in the regenerated lenses closely resemble the patterns seen in the lenses of the young newts . In contrast , the tail cells of the old and young newts display different gene expression patterns , with those from the older newts displaying hallmarks of ageing that are absent in the younger newts . The iris cells from the old newts show a mixed gene expression profile with features characteristic of both young and aged tissue . Sousounis et al . ’s findings highlight the value of using newts as models to study the links between regeneration and ageing DOI: http://dx . doi . org/10 . 7554/eLife . 09594 . 002	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "stem", "cells", "and", "regenerative", "medicine" ]	2015	A robust transcriptional program in newts undergoing multiple events of lens regeneration throughout their lifespan
Preferably , lifespan-extending therapies should work when applied late in life without causing undesired pathologies . Reducing insulin/insulin-like growth factor ( IGF ) -1 signaling ( IIS ) increases lifespan across species , but the effects of reduced IIS interventions in extreme geriatric ages remains unknown . Using the nematode Caenorhabditis elegans , we engineered the conditional depletion of the DAF-2/insulin/IGF-1 transmembrane receptor using an auxin-inducible degradation ( AID ) system . This allowed for the temporal and spatial reduction in DAF-2 protein levels at time points after which interventions such as RNAi become ineffective . Using this system , we found that AID-mediated depletion of DAF-2 protein surpasses the longevity of daf-2 mutants . Depletion of DAF-2 during early adulthood resulted in multiple adverse phenotypes , including growth retardation , germline shrinkage , egg retention , and reduced brood size . By contrast , AID-mediated depletion of DAF-2 post-reproduction , or specifically in the intestine in early adulthood , resulted in an extension of lifespan without these deleterious effects . Strikingly , at geriatric ages , when 75% of the population had died , AID-mediated depletion of DAF-2 protein resulted in a doubling in lifespan . Thus , we provide a proof-of-concept that even close to the end of an individual’s lifespan , it is possible to slow aging and promote longevity . The goal of aging research or geroscience is to identify interventions that promote health during old age ( Kennedy et al . , 2014; López-Otín et al . , 2013; Partridge et al . , 2018 ) . Nutrient-sensing pathways that regulate growth and stress resistance play major roles as conserved assurance pathways for healthy aging ( Kenyon , 2010; López-Otín et al . , 2013 ) . One of the first longevity pathways discovered was the insulin/insulin-like growth factor ( IGF ) -1 signaling pathway ( reviewed in Kenyon , 2010 ) . Reducing insulin/IGF-1 signaling ( IIS ) increases lifespan across species ( Kenyon , 2010 ) . Mice heterozygous for the IGF-1 receptor , or with depleted insulin receptor in adipose tissue , are stress-resistant and long-lived ( Blüher et al . , 2003; Holzenberger et al . , 2003 ) , for example , and several single-nucleotide polymorphisms in the IIS pathway have been associated with human longevity ( Kenyon , 2010 ) . Moreover , gene variants in the IGF-1 receptor have been associated and functionally linked with long lifespans in human centenarians ( Suh et al . , 2008 ) . This suggests that a comprehensive understanding of this pathway in experimental , genetically tractable organisms has promising translational value for promoting health in elderly humans . However , whether or not reducing IIS during end-of-life stages can still promote health and longevity in any organism is unknown . Therefore , we turned to the model organism Caenorhabditis elegans to investigate whether reducing IIS during old age was sufficient to increase lifespan . The groundbreaking discovery that a single mutation in daf-2 , which is the orthologue of both the insulin and IGF-1 receptors ( Kimura et al . , 1997 ) , or mutations in ‘downstream’ genes in the IIS pathway , could double the lifespan of an organism was made in the nematode C . elegans ( Friedman and Johnson , 1988; Kenyon et al . , 1993 ) . Since its discovery , over 1000 papers on daf-2 have been published , making it one of the most studied genes in this model organism ( Source: PubMed ) . Genetic and genomics approaches have revealed that the DAF-2 insulin/IGF-1 receptor signaling regulates growth , development , metabolism , inter-tissue signaling , immunity , stress defense , and protein homeostasis , including extracellular matrix remodeling ( Ewald et al . , 2015; Gems et al . , 1998; Kimura et al . , 1997; Murphy and Hu , 2013; Wolkow et al . , 2000 ) . Much of our knowledge of the effects of daf-2 on aging has come from the study of reduction-of-function alleles of daf-2 . Several alleles of daf-2 have been isolated that are temperature-sensitive with respect to an alternative developmental trajectory . For instance , most daf-2 mutants develop into adults at 15°C and 20°C but enter the dauer stage at 25°C ( Gems et al . , 1998 ) , which is a facultative and alternative larval endurance stage in which C . elegans spends most of its life cycle in the wild ( Hu , 2007 ) . Under favorable conditions , C . elegans develops through four larval stages ( L1–L4 ) . By contrast , when the animals are deprived of food and experience an overcrowded environment and/or thermal stress ( above 27°C ) , the developing larvae molt into an alternative pre-dauer ( L2d ) stage . If conditions do not improve , C . elegans enter the dauer diapause instead of the L3 stage ( Golden and Riddle , 1984; Hu , 2007; Karp , 2018 ) . A major limitation in using daf-2 mutants is that several of them show L1 larval and pre-dauer stage ( L2d ) arrest ( Gems et al . , 1998 ) . Furthermore , the daf-2 alleles have been categorized into two mutant classes depending on the penetrance of dauer-like phenotypes during adulthood , such as reduced brood size , small body size , and germline shrinkage , as observed in the daf-2 class II mutants ( Arantes-Oliveira et al . , 2003; Ewald et al . , 2018; Ewald et al . , 2015; Gems et al . , 1998; Hess et al . , 2019; Patel et al . , 2008; Podshivalova and Kerr , 2017 ) . RNA interference of daf-2 can be applied , which increases lifespan without dauer formation during development and circumvents induction of daf-2 class II mutant phenotypes during adulthood ( Dillin et al . , 2002; Ewald et al . , 2018; Ewald et al . , 2015; Kennedy et al . , 2004 ) . However , the increase in lifespan by RNAi of daf-2 is only partial compared to strong alleles such as daf-2 ( e1370 ) ( Ewald et al . , 2015 ) . Furthermore , adult-specific RNAi knockdown of daf-2 quickly loses its potential to increase lifespan and does not extend lifespan when started after day 6 of adulthood ( Dillin et al . , 2002 ) , that is , after the reproductive period of C . elegans . Whether this is due to age-related functional decline of RNAi machinery or residual DAF-2 protein levels , or whether the late-life depletion of daf-2 simply does not extend lifespan remains unclear . As such , using an alternative method to reduce DAF-2 levels beyond RNAi or daf-2 mutation may allow us to more clearly uncouple the pleiotropic effects of reduced IIS during development from those that drive daf-2-mediated longevity during late adulthood . To this end , we used an auxin-inducible degradation ( AID ) system to induce the depletion of the degron-tagged DAF-2 protein with temporal precision ( Zhang et al . , 2015 ) . The Arabidopsis thaliana IAA17 degron is a 68-amino acid motif that is specifically recognized by the transport inhibitor response 1 ( TIR1 ) protein only in the presence of the plant hormone auxin ( indole-3-acetic acid; Dharmasiri et al . , 2005 ) . Although cytoplasmic , nuclear , and membrane-binding proteins tagged with degron have been recently shown to be targeted and degraded in C . elegans ( Beer et al . , 2019; Zhang et al . , 2015 ) , to our knowledge , the AID system has not been used previously to degrade transmembrane proteins , such as the DAF-2 insulin/IGF-1 receptor . We find that using AID effectively degrades DAF-2 protein and promotes dauer formation when applied early in development . Dauer-like phenotypes are present in adults when AID of DAF-2 is applied late in development . Some of these adulthood dauer traits are induced by the loss of daf-2 in neurons , but others seem to be caused by the systemic loss of daf-2 . More importantly , the post-developmental , conditional degradation of DAF-2 protein extends lifespan without introducing dauer-like phenotypes . Remarkably , we demonstrate that when more than half of the population has died at day 25 of adulthood , AID of DAF-2 in these remaining aged animals is sufficient to promote longevity . Our work suggests that therapeutics applied at even extremely late stages of life are capable of increasing longevity and healthspan in animals . To monitor and conditionally regulate protein levels of the C . elegans DAF-2 insulin/IGF-1 receptor , we introduced a degron::3xFLAG tag into the 3' end of the daf-2 open reading frame ( Figure 1—figure supplement 1 ) . This degron::3xFLAG insertion into the genome was designed to tag the DAF-2 receptor at the cytosolic part for two reasons: first , to minimize any interference by the 81-amino acids large degron::3xFLAG-tag with the DAF-2 receptor function; and second , to ensure accessibility of the degron for targeted degradation by the TIR1 ubiquitin ligase expressed in the cytoplasm ( Figure 1A ) . We endogenously tagged the DAF-2 receptor using CRISPR , and the resulting daf-2 ( bch40 ) CRISPR allele was verified by PCR ( Figure 1—figure supplement 1 ) . We performed western blot analysis against the 3xFLAG-tag and detected a specific band in daf-2 ( bch40 ) animals . This band was absent in wild type ( N2 ) and animals carrying only the eft-3p::TIR1::mRuby::unc-54 3'UTR transgene ( Figure 1B ) , which expresses TIR1 in all somatic cells , including neurons ( Tomioka et al . , 2016 ) . To promote degradation of the degron::3xFLAG-tagged DAF-2 receptor , we crossed daf-2 ( bch40 ) into TIR1-expressing C . elegans . The strain obtained from this cross will be called ‘DAF-2::degron’ throughout this paper ( i . e . , Si57 [Peft-3::TIR1::mRuby::unc-54 3'UTR+ Cbr-unc-119 ( + ) ] II; daf-2 ( bch40 [degron::3xFLAG::STOP::SL2-SV40-degron::wrmScarlet-egl-13 NLS] ) III ) . This strain showed no obvious phenotypes and exhibited a normal developmental progression at 20°C ( Figure 1—figure supplement 1 ) . To verify whether the band from the western blot was indeed DAF-2::degron::3xFLAG , we treated DAF-2::degron animals with daf-2 RNAi . The band nearly completely disappeared after 48 hr of daf-2 ( RNAi ) feeding ( Figure 1C and D , Source data 1 and Source data 2 ) . Collectively , these results suggested that the tagged transmembrane receptor DAF-2 did not interfere with normal DAF-2 function . Next , we monitored endogenous DAF-2 protein levels under different environmental conditions , such as temperature and diet . Previously , Kimura and colleagues used DAF-2 antibody immunostaining of whole animals and reported that mutant DAF-2 ( e1370 ) protein is present at 15°C but barely detected at 25°C , whereas mutant DAF-2 ( e1370 ) protein in a daf-16 null background or wild-type DAF-2 protein persists at both 15°C and 25°C ( Kimura et al . , 2011 ) . By contrast , upon 24 hr of starvation , the DAF-2 receptor is no longer detectable by using immunofluorescence in fixed C . elegans ( Kimura et al . , 2011 ) . Since the FOXO transcription factor daf-16 is the transcriptional output of daf-2 signaling ( Ewald et al . , 2018; Gems et al . , 1998 ) , these results suggest that DAF-2 protein levels may be autoregulated by IIS and might be influenced by temperature and food availability . We first asked whether our DAF-2::degron::3xFLAG tag allows quantification of endogenous DAF-2 levels . We observed comparable wild-type DAF-2::degron::3xFLAG levels across a range of temperatures ( 15–28°C; Figure 1E ) , indicating that temperature does not influence DAF-2 levels in wild type . Intriguingly , however , we found that using FLAG-HRP antibodies to monitor protein levels , DAF-2 protein almost completely disappeared after 36–48 hr of starvation ( Figure 1F and G ) . In keeping with this result , well-fed animals , for which we added 1% glucose into the bacterial diet ( OP50 ) , increased the DAF-2 protein levels ( Figure 1F and G ) . Curiously , we noted that this starvation-induced degradation of DAF-2 did not happen when , during development , C . elegans were fed another bacterial strain , HT1115 , used for RNAi ( L4440 ) . When DAF-2::degron animals were grown on L4440 and then shifted on empty NGM plates for 24 or 48 hr of starvation , DAF-2 levels did not decrease ( Figure 1H ) . This observation suggests a hypothesis that the nutritional composition of the animal’s diet prior to starvation influences DAF-2 stability , which will be interesting to test in future research . We conclude that food availability controls not only the secretion of insulin-like peptides to regulate DAF-2 activity ( Pierce et al . , 2001 ) but also DAF-2 receptor abundance . In C . elegans , cytosolic degron-tagged proteins are almost completely degraded after 30 min of auxin treatment ( Zhang et al . , 2015 ) . However , the degradation of transmembrane proteins using AID in vivo has not been previously reported . We hypothesized that C . elegans might exhibit similar kinetics of degradation of a transmembrane protein following auxin treatment . In keeping with that hypothesis , after 30 min of 1 mM auxin treatment , we observed a dramatic decrease in transmembrane DAF-2 protein abundance ( Figure 1I and J ) . Levels of DAF-2 were only slightly further reduced by continued auxin treatment , as indicated at 4 and 24 hr time points ( Figure 1I and J ) . After 24 hr of 1 mM auxin treatment , we observed only a 40% total decrease in DAF-2 protein abundance rather than a complete loss ( Figure 1C and D ) . Similar kinetics in the degradation of DAF-2::degron::3xFLAG levels were confirmed using additional FLAG and degron antibodies ( Source data 2 ) . Taken together , these results suggest that our AID system allows for the partial , rapid degradation of the transmembrane DAF-2 receptor . We wondered whether the reduction of DAF-2 levels by AID would have consequences consistent with reduced IIS . Activation of DAF-2/insulin/IGF-1 receptor induces a downstream kinase cascade to phosphorylate the transcription factors DAF-16/FOXO and SKN-1/NRF , causing their retention in the cytoplasm ( Figure 2A; Ewald et al . , 2015; Henderson and Johnson , 2001; Lin et al . , 2001; Murphy et al . , 2003; Ogg et al . , 1997; Tullet et al . , 2008 ) . Genetic inhibition of daf-2 results in less DAF-16 and SKN-1 phosphorylation and promotes nuclear translocation to induce the expression of target genes , such as sod-3 ( superoxide dismutase ) and gst-4 ( glutathione S-transferase ) , respectively ( Ewald et al . , 2015; Henderson and Johnson , 2001; Lin et al . , 2001; Murphy et al . , 2003; Tullet et al . , 2008 ) . Within 1 hr of 1 mM auxin treatment , we found that most DAF-16::GFP translocated into the nuclei in DAF-2::degron animals ( Figure 2B ) , with observable translocation already after 30 min ( Figure 2—figure supplement 1 ) . This DAF-16::GFP nuclear localization in DAF-2::degron animals was time- and auxin-concentration-dependent and did not occur in DAF-16::GFP animals with wild-type DAF-2 ( Figure 2B , Figure 2—figure supplement 1 ) . Similarly , SKN-1- or DAF-16-target gene expression of gst-4 or sod-3 was only induced upon auxin treatment in DAF-2::degron animals ( Figure 2C and D ) . Thus , expectedly , IIS is reduced upon AID DAF-2 degradation . Reduced IIS during development promotes dauer entry . Dauer formation at 15°C has been observed for a variety of strong loss-of-function daf-2 alleles , such as the class I alleles e1369 and m212 , the class II allele e979 , the null alleles m65 , m646 , m633 , and a variety of unclassified alleles discovered by Malone and Thomas ( Gems et al . , 1998; Kimura et al . , 2011; Malone and Thomas , 1994; Patel et al . , 2008 ) . For the commonly used reference alleles e1368 and e1370 , penetrant dauer formation only occurs at 25°C ( Gems et al . , 1998 ) . By contrast , knockdown of daf-2 by RNAi does not cause dauer formation at any temperature ( Dillin et al . , 2002; Ewald et al . , 2015; Kennedy et al . , 2004 ) . We hypothesized that dauer formation would not happen because the decrease of DAF-2::degron levels after auxin treatment is only around 40% . However , synchronized L1 treated with 0 . 1 or 1 mM auxin all formed dauers at 25°C ( Figure 2E and F ) . We observed dose-dependent retardation of the developmental speed when using lower auxin concentrations ( 1 , 10 , and 50 μM ) , but the offspring of retarded animals , grown on plates containing auxin , became dauers ( Figure 2E and F ) . The dependence on auxin concentration for dauer formation of L1 animals suggests a threshold of DAF-2 receptor levels for the decision or commitment to dauer diapause . Even more surprising was the observation that dauer formation was also observed at 15°C and 20°C with complete penetrance ( Figure 2G and H ) . We found that dauer formation was related to the developmental speed at a given temperature: At 15°C , it took 6 days; at 20°C , it took 4 days; and at 25°C , it took 3 days to form dauers ( Figure 2H ) . We verified that all auxin-induced DAF-2::degron dauers showed dauer-specific characteristics , such as SDS resistance , cessation of feeding , constricted pharynxes , and dauer-specific alae ( Figure 2G , Figure 2—figure supplement 1 ) , suggesting a complete dauer transformation . Thus , the AID of DAF-2 promotes complete dauer formation independent of temperature but dependent on DAF-2 protein abundance . Wild-type animals enter the pre-dauer L2d stage , where they keep monitoring their environment before completely committing to dauer formation ( Golden and Riddle , 1984; Hu , 2007; Karp , 2018 ) . Treating wild type with dauer pheromone suggested mid-L1 as the stage when the dauer decision is made ( Golden and Riddle , 1984 ) . By contrast , previous temperature-shifting experiments ( from 15°C to 25°C ) with daf-2 mutants suggested a dauer decision time window from L1 to L2 stage , before the L2d stage ( Swanson and Riddle , 1981 ) . Since AID allows for precise temporal degradation of DAF-2 , we pinpointed the dauer entry decision to the mid-L1 stage . Specifically , we shifted synchronized L1s at different time points to plates containing 1 mM auxin and counted the number of cells in developing gonads to determine the developmental stage , when 50% of the population committed to becoming dauers ( Figure 3—figure supplement 1 ) . We found that when DAF-2 levels are below a given threshold at the mid-L1 stage , the animals commit to becoming dauers . The FOXO transcription factor DAF-16 is required for dauer formation in daf-2 mutants . We crossed DAF-2::degron with DAF-16::degron ( Aghayeva et al . , 2021 ) and found that daf-16 was required for dauer formation and developmental speed alterations after DAF-2::degron depletion ( Figure 2—figure supplement 2 ) . Previous reports suggest that many daf-2 alleles show low to severe penetrance of embryonic lethality and L1 arrest at higher temperatures ( Collins et al . , 2008; Ewald et al . , 2016; Gems et al . , 1998; Patel et al . , 2008 ) . Although the constitutive dauer formation of the proposed null allele daf-2 ( m65 ) is suppressed by daf-16 null mutations , the embryonic lethality and L1 arrest are not daf-16-dependent ( Patel et al . , 2008 ) . We observed no embryonic lethality nor L1 arrest in the progeny of animals placed on 1 mM auxin as L4s . Similar results were seen using either the DAF-2::degron or DAF-2::degron; germline TIR1 strains . Higher concentrations of auxin lead to toxicity in both wild-type and DAF-2::degron animals ( Figure 2—figure supplement 2 ) . A lack of embryonic lethality could be explained by insufficient DAF-2 degradation or earlier decision stages . Taken together , the inactivation of DAF-2 by AID is 100% penetrant for dauer formation at any temperature . Still , the absence of embryonic lethality or L1 arrest at 1 mM auxin suggests that DAF-2::degron functionally is more similar to a non-conditional and severe loss-of-function mutation than a null allele . Given the strong phenotypic effects of DAF-2 AID on animal development , we next explored whether DAF-2 degradation by AID could affect the function of adult animals . Previous studies indicate that reducing IIS , either by daf-2 RNAi knockdown or in genetic mutants , increases lifespan at any temperature ( 15–25°C ) ( Ewald et al . , 2018; Gems et al . , 1998 ) . We hypothesized that AID-dependent degradation of DAF-2 would have similar effects on the lifespan of animals . We found that auxin supplementation of DAF-2::degron animals , starting from L4 , resulted in a 70–135% lifespan extension ( Figure 3A; Supplementary file 1 ) . Impressively , DAF-2 degradation using 1 mM auxin surpassed the longevity of commonly used daf-2 ( e1368 ) and daf-2 ( e1370 ) mutants ( Figure 3A , Supplementary file 1 ) . By contrast , auxin treatment at 0 . 1 or 1 mM concentration had little or no effect on wild-type lifespan ( Figure 3A; Supplementary file 1 ) . These results suggest that auxin-induced degradation of daf-2 is a powerful tool to promote longevity . Although reducing daf-2 function causes beneficial increases in longevity and stress resistance , it causes residual detrimental phenotypes in adult animals that resemble the behavioral and morphologic changes reminiscent of developing animals remodeling to enter the dauer state ( Ewald et al . , 2018; Gems et al . , 1998 ) . In class II daf-2 alleles , these phenotypes , such as small gonads , reduced brood size , reduced motility , and reduced brood size , manifest only at 25°C during adulthood but not at lower temperatures ( Ewald et al . , 2018; Gems et al . , 1998 ) . To determine whether DAF-2::degron AID animals display daf-2 class II mutant phenotypes , we quantified these characteristics at 15°C and 25°C . In placing L4 DAF-2::degron animals on auxin and at 25°C , we observed similar levels of egg retention and effects on gonad size as was seen in the daf-2 ( e1370 ) class II allele . Strikingly , these effects were temperature-dependent , as DAF-2::degron animals did not retain eggs or had reduced gonad sizes at 15°C ( Figure 3B , Figure 3—figure supplement 1 ) . Similarly , DAF-2::degron animals on auxin exhibited germline shrinkage at 25°C , albeit to a lesser degree than the daf-2 ( e1370 ) class II allele ( Figure 3C , Figure 3—figure supplement 1 ) . This phenotype was also absent at 15°C , suggesting that egg retention and germline shrinkage are temperature-sensitive traits . Another known daf-2 class II mutant phenotype at 25°C is the quiescence or immobility of class II daf-2 ( e1370 ) mutants ( Ewald et al . , 2018; Gems et al . , 1998 ) . We did not observe any immobility of auxin-treated DAF-2::degron animals at 25°C or during lifespan assays at 20°C ( Video 1 , Supplementary file 2 ) . Although the effects on body size of daf-2 ( e1370 ) class II allele are temperature-dependent , presenting at 25°C but not at 15°C ( Ewald et al . , 2015; Gems et al . , 1998; McCulloch and Gems , 2003; Figure 3—figure supplement 1 ) , auxin-induced degradation of DAF-2 starting from L4 shortened body size of 2-day-old adults at both temperatures ( Figure 3D , Figure 3—figure supplement 1 ) . Similarly , while daf-2 ( e1370 ) mutants only exhibit reduced brood sizes at higher temperatures ( Ewald et al . , 2018; Gems et al . , 1998 ) , AID of DAF-2::degron starting from L4 reduced brood size at both 15°C and 25°C ( Figure 3E ) . This suggests that smaller body and brood size manifest as non-conditional traits , in keeping with insulin/IGF-1’s role as an essential gene for these functions . In summary , these results suggest that some daf-2 class II mutant phenotypes , or pathologies , can be induced during adulthood independent of temperature and that passing through the L2d stage is not required for the daf-2 class II mutant phenotypes to emerge in adult animals . The pleiotropic effects of DAF-2 have been ascribed to tissue-specific functions of DAF-2 . DAF-2 protein levels are predominantly found in the nervous system and intestine , and to a lesser extent in the hypodermis ( Kimura et al . , 2011 ) , while daf-2 mRNA expression has also been detected in the germline ( Han et al . , 2017; Lopez et al . , 2013 ) . Mosaic loss of daf-2 in different cell lineages indicated neurons as crucial tissue to control dauer formation non-cell autonomously ( Apfeld and Kenyon , 1998 ) . Moreover , dauer formation in daf-2 ( e1370 ) can be restored by expressing wild-type DAF-2 only in neurons ( Wolkow et al . , 2000 ) . Thus , we hypothesized that select tissues might drive the daf-2 class II mutant phenotypes . To test this , we expressed TIR1 specifically in muscles , neurons , and intestine , driven by the myo-3 , rab-3 , and vha-6 promoters , respectively ( Materials and methods , Supplementary file 2 ) . TIR1 expressed from any of these three tissue-specific promoters did not result in reduced body size ( Figure 4E and Figure 3—figure supplement 1E ) . To validate that the neuronal TIR1 was functional , we crossed neuronal TIR1 into daf-16 ( ot853 [daf-16::linker::mNG::3xFLAG::AID] ) ( Aghayeva et al . , 2021 ) and observed that DAF-16::mNG was selectively degraded in neurons upon auxin treatment ( Figure 3—figure supplement 2 ) . Nonetheless , we found that depletion of DAF-2 in neurons caused egg retention and germline shrinkage ( Figure 3F and G ) . Interestingly , the germline shrinkage and egg retention phenotypes were temperature-dependent , suggesting some interaction of temperature and neuronal DAF-2 abundance . Thus , some daf-2-phenotypes appear to emerge from a single tissue , whereas others might be due to an interplay between several tissues . Previously , transgenic expression of wild-type copies of DAF-2 in neuronal or intestinal cells was shown to partially suppress the longevity of daf-2 ( e1370 ) mutants at 25°C ( Wolkow et al . , 2000 ) . Taking advantage of our unique AID system , we wanted to test whether the degradation of DAF-2 in a single tissue is sufficient to induce longevity . We found that either neuronal or intestinal depletion of DAF-2 was alone sufficient to extend lifespan , although not to the extent as when DAF-2 is degraded in all tissues ( Figure 4A; Supplementary file 1 ) . Therefore , we asked whether tissue-specific DAF-2 degradation was also sufficient for stress resistance seen in daf-2 mutant animals ( Ewald et al . , 2018; Gems et al . , 1998 ) . Auxin-induced degradation of DAF-2 in all tissues resulted in increased oxidative stress resistance , comparable to daf-2 ( e1370 ) mutants ( Figure 4B ) . We observed improved oxidative stress resistance when we depleted DAF-2 in the intestine but not in neurons ( Figure 4C and D , Figure 3—figure supplement 2 ) . This implies that reducing DAF-2 levels , specifically in the intestine , promotes longevity and stress resistance without causing dauer-like phenotypes . We have previously shown that the transcription factor SKN-1/NRF1 , 2 , 3 is localized in the nucleus at 15°C or 25°C in daf-2 ( e1370 ) mutants ( Ewald et al . , 2015 ) , suggesting that SKN-1 activation occurs under reduced insulin/IGF-1 receptor signaling conditions ( Tullet et al . , 2008 ) . Intriguingly , skn-1 activity is necessary for full lifespan extension of daf-2 ( e1370 ) mutants only at 15°C but not at 25°C ( Ewald et al . , 2015 ) . We hypothesized that skn-1 requirements for lifespan extension are masked when dauer-like reprogramming conditions are triggered at higher temperatures ( Ewald et al . , 2018 ) . Because auxin-treated DAF-2::degron AID animals exhibit daf-2 class II mutant traits during adulthood at 15°C , we asked whether the lifespan extension caused by DAF-2::degron AID upon auxin treatment at 15°C is skn-1-independent . We found that the lifespan extension in DAF-2::degron animals fully required skn-1 at 15°C . Surprisingly , however , a loss of skn-1 extended the median lifespan of DAF-2::degron animals at 25°C ( Figure 4E and F; Supplementary file 1 ) . This suggests that skn-1 may function independently from dauer-like reprogramming pathways at 15°C . Furthermore , the increased longevity seen in DAF-2::degron animals may result from a differential transcriptional program at higher temperatures compared to lower temperatures . Finally , we asked whether it would be possible to promote longevity in geriatric animals by depleting DAF-2 by AID . Previous studies using RNAi indicated that reduced daf-2 expression extended lifespan when started at day 6 of adulthood but not later ( Dillin et al . , 2002 ) , raising the question of whether daf-2-longevity induction is possible beyond the reproductive period ( days 1–8 of adulthood ) . To address this , we maintained DAF-2::degron animals on control plates and shifted them to 1 mM auxin-containing plates at day 0 ( L4 ) , and up to day 20 of adulthood ( Figure 5A ) . We found that shifting the animals past the reproductive period at day 10 and day 12 still led to an increase in lifespan by 48–72% and 49–57% , respectively ( Figure 5A , Supplementary file 1 ) . Since transferring old C . elegans to culturing plates without bacterial food can also increase lifespan past reproduction ( Smith et al . , 2008 ) , we decided to top-coat lifespan plates with auxin late in life . We observed lifespan extension of animals by supplementing auxin very late during lifespan at day 21 or 25 of adulthood , at a time at which already approximately 50–75% of the population had died ( Figure 5B–D , Supplementary file 1 ) . Remarkably , AID of DAF-2 doubled the lifespan of these animals at this late stage ( Figure 5A–D , Supplementary file 1 ) . For example , when about three-quarters of the population had ceased by day 21 ( Figure 5B ) or day 25 ( Figure 5D ) , and control-treated DAF-2::degron animals lived for just another 4 or 7 days , the auxin-treated DAF-2::degron animals lived for another 26 or 43 days , respectively ( Figure 5B and D , Supplementary file 1 ) . This demonstrates that reducing insulin/IGF-1 receptor signaling is feasible in geriatric C . elegans . Thus , our approach to selectively reduce DAF-2 protein in old animals suggests that targeting the daf-2 signaling pathway late in life may be an effective strategy to extend lifespan . For longevity interventions to be efficient without causing undesired side effects , the time point of treatment must be chosen carefully . This is especially important for pathways such as the insulin/IGF-1 pathway that is essential for growth and development ( Kenyon , 2010 ) . Although the importance of DAF-2 in regulating lifespan is well established , the consequences of late-in-life inhibition remained unknown . Here , we demonstrate that late-life degradation of DAF-2 extends lifespan . In this work , we have effectively engineered a degron tag into the endogenous daf-2 locus using CRISPR , representing the first report of auxin-induced degradation ( AID ) of a transmembrane receptor in vivo . DAF-2 receptor levels were strongly reduced via AID . Consistent with reduced insulin signaling , AID-mediated degradation of DAF-2 facilitated dauer formation , longevity , stress resistance , caused growth-related phenotypes during early adulthood , and finally increased longevity during the post-reproductive geriatric stages of life . It is well established that DAF-2/insulin/IGF-1 receptor signaling connects nutrient levels to growth and development ( Murphy and Hu , 2013 ) . This is attributed to insulin-like peptides binding DAF-2 and activating a downstream phosphorylation kinase cascade that alters metabolism ( Murphy and Hu , 2013 ) . Surprisingly , we find that DAF-2 receptor abundance is linked to nutrient availability and perhaps dietary content . Starving C . elegans leads to decreased DAF-2 receptor abundance , consistent with previous in situ antibody staining ( Kimura et al . , 2011 ) , whereas mimicking a high-energy diet by adding glucose to the bacterial food source increases DAF-2 receptor abundance . This suggests an additional layer of regulating IIS by connecting food cues to adapt metabolism via DAF-2 receptor levels , potentially as an internal representation of the environment . Environmental conditions are carefully monitored by developing C . elegans . First , larval stage ( L1 ) wild-type C . elegans that are food-deprived , exposed to elevated temperatures ( >27°C ) , and/or are crowded , enter into a pre-dauer L2d stage , where they continue monitoring their environment before committing and molting into dauers ( Hu , 2007; Karp , 2018 ) . However , previous temperature-shifting experiments ( from 15°C to 25°C ) with daf-2 mutants have indicated the existence of a ‘dauer decision time window’ between the L1 and L2 stage ( Swanson and Riddle , 1981 ) . By using AID to manipulate DAF-2 levels directly , we found that AID degradation of DAF-2 during a narrow time period in the mid-L1 stage is sufficient to induce dauer formation , suggesting that the decision to enter dauer relies upon a threshold of DAF-2 protein levels . This decision is uncoupled from temperature or food abundance . Thus , absolute DAF-2 protein abundance appears to be a key factor in the animal’s decision to enter into the dauer state during early development . Although IIS is reduced in daf-2 ( e1370 or e1368 ) mutants at lower temperatures ( Ewald et al . , 2018; Gems et al . , 1998 ) , dauer larvae are formed only when these daf-2 mutants are exposed to higher temperatures . The observation that mutant DAF-2 protein ( Kimura et al . , 2011 ) but not wild-type DAF-2 protein ( Figure 5 ) is lost at elevated temperatures suggests a model where DAF-2 mutant protein becomes unstable with increasing temperatures and might be subsequently targeted for degradation . This hypothesis might explain why strong class II daf-2 mutants , such as daf-2 ( e979 ) and daf-2 ( e1391 ) , which have much lower DAF-2 protein levels at 15°C and 25°C compared to other daf-2 mutants ( Tawo et al . , 2017 ) , exhibit higher propensities toward dauer formation at any temperature ( Gems et al . , 1998 ) . Thus , the severity of classical daf-2 mutant alleles in regard to dauer formation and adult dauer traits might be linked to DAF-2 receptor abundance . At higher temperatures , adult daf-2 class II mutants exhibit significant penetrance of undesired phenotypes ( Ewald et al . , 2018 ) . In these animals , a clear remodeling of the body and internal organs , including constriction of the pharynx and shrinkage of the germline , remains present , along with an altered neuronal morphology and electrical synapse connectome that drives behavioral changes such as quiescence , diminished foraging behavior , and altered egg-laying programs ( Arantes-Oliveira et al . , 2003; Ewald et al . , 2015; Gems et al . , 1998; Hess et al . , 2019; Bhattacharya et al . , 2019; Patel et al . , 2008; Podshivalova and Kerr , 2017 ) . We showed that L4-specific AID of the DAF-2::degron results in a non-conditional reduction of body size and brood size , whereas egg retention and germline shrinkage only occur at higher temperatures . This indicates that the non-conditional daf-2 class II traits are not a residual effect of a dauer-like developmental program but instead are side effects caused by reduced functions of DAF-2 during later stages of development . Additional temperature-sensitive daf-2 class II traits of egg retention and gonad shrinkage are mediated by loss of daf-2 in neurons only at higher temperatures . Why these traits only manifest at higher temperatures remains unclear . One explanation may be that DAF-2 levels are reduced in temperature-sensing neurons , which then elicits a systemic effect that drives germline shrinkage and egg retention . Neurons are refractory to RNAi , suggesting that daf-2 ( RNAi ) effects work through other tissues than neurons to extend lifespan . Furthermore , treating class I mutants daf-2 ( e1368 ) with daf-2 ( RNAi ) doubles their longevity without causing adult daf-2 class II traits ( Arantes-Oliveira et al . , 2003 ) , suggesting that daf-2 ( RNAi ) would lower DAF-2 receptor levels in other tissues than neurons for this additive longevity effect . Consistent with neuronal regulation of these traits is that daf-2 RNAi applied to wild type does not result in dauers but results in dauer formation when applied to neuronal-hypersensitive RNAi C . elegans strains ( Dillin et al . , 2002; Ewald et al . , 2015; Kennedy et al . , 2004 ) . We find that DAF-2 degradation in neurons or intestine increases lifespan . This is consistent with a previous finding by Apfeld and Kenyon that either mosaic loss of daf-2 in AB cell lineage that gives rise to neurons and other cells ( epidermis , seam , pharyngeal , vulval cells ) or loss of daf-2 in EMS cell lineage that gives rise to intestinal and other cells ( pharyngeal and gonadal cells ) results in increased lifespan ( Apfeld and Kenyon , 1998 ) . By contrast , our neuronal DAF-2 depletion did not result in a doubling of lifespan as seen by loss of daf-2 in the AB lineage ( Apfeld and Kenyon , 1998 ) , suggesting that loss of daf-2 in other tissues , in combination with neurons or intestine , is required to recapitulate full lifespan extension . Given that reducing DAF-2 in neurons results in daf-2 class II traits at higher temperatures , one might target intestinal DAF-2 for degradation to uncouple longevity from any undesired phenotypes . Yet , DAF-2 is essential for growth . We find the best time point for DAF-2 inhibition is rather late in life to bypass these undesired side effects to promote longevity . We find that as late as day 25 of adulthood , when almost three-quarters of the population had died , AID of DAF-2 is sufficient to increase lifespan . The only other manipulation that was able to increase the lifespan so late in life was the transfer of old C . elegans to culture plates without food ( Smith et al . , 2008 ) . However , C . elegans do not feed after reaching mid-life ( Collins et al . , 2008; Ewald et al . , 2016 ) , suggesting that it might not be the intake of calories that promotes longevity . Instead , the old C . elegans could sense the absence of food and thereby reduce DAF-2 levels to promote longevity . Along these lines , it would be interesting to determine if late-life bacterial deprivation works synergistically with DAF-2 AID or not in future studies . In mammals , mid-life administration of IGF-1 receptor monoclonal antibodies to 78-week-old mice ( a time when all mice are still alive and 6 weeks before first mice start to die ) is sufficient to increase their lifespan and improve their healthspan ( Mao et al . , 2018 ) . Other parallels between mammals and nematodes are the tissues from which lower insulin/IGF-1 receptor levels promote longevity . Mice carrying brain-specific heterozygous IGF-1 receptor knockout are long-lived ( Kappeler et al . , 2008 ) , as are the mice with adipose-specific knockout of the insulin receptor ( Blüher et al . , 2003 ) . This is reminiscent of our findings that AID of DAF-2::degron in neurons or intestine ( the major fat-storage tissue in C . elegans ) was sufficient to increase the lifespan . These similarities could reflect conserved functions , as daf-2 is considered the common ancestor to both insulin and IGF-1 receptors ( Kimura et al . , 1997 ) . Although it is known that neurons and the intestine are important for food perception and regulation of food intake , the effects of food perception or intake on insulin receptor and IGF-1 receptor levels are poorly understood . Starving rats for 3 days increases the abundance of insulin-bound insulin receptors ( Koopmans et al . , 1995 ) , possibly to increase glucose uptake . It is unknown whether prolonged starvation would lead to lower basal insulin/IGF-1 receptor levels . However , alterations of the IGF-1 receptor levels are associated with altered lifespan . For instance , heterozygous IGF-1 receptor knockout mice , which have lower IGF-1 receptor levels , have an increased lifespan ( Holzenberger et al . , 2003; Kappeler et al . , 2008; Xu et al . , 2014 ) . Also , overexpression of the short isoform of p53 ( p44 ) increases IGF-1 receptor levels and shortens the lifespan of mice ( Maier et al . , 2004 ) . Furthermore , the administration of recombinant human IGF-1 increases IGF-1 receptor abundance in murine embryonic cells ( Maier et al . , 2004 ) . Food components themselves can affect insulin receptor levels . For instance , palmitate activates PPARα to induce miR-15b , which targets insulin receptor mRNA for degradation ( Li et al . , 2019 ) . Potentially , food components could impact the murine insulin/IGF-1 receptor also via its degradation . Indeed , there is some evidence suggesting the regulation of insulin/IGF-1 receptor abundance by E3 ubiquitin ligases . For instance , the E3 ligase CHIP regulates insulin/IGF-1 receptor levels in C . elegans , Drosophila , and human cell cultures ( Tawo et al . , 2017 ) . In mice , the muscle-specific mitsugumin 53 ( MG53 ) E3 ligase targets the insulin receptor for degradation ( Song et al . , 2013 ) . High-fat diet results in the reduction of insulin receptor levels ( Li et al . , 2019 ) via higher MG53-mediated degradation ( Song et al . , 2013 ) . MG53 is upregulated under a high-fat diet in mice , and MG53-/- deficient mice are protected from high-fat diet-induced obesity , insulin resistance , and other metabolic syndrome-associated phenotypes ( Song et al . , 2013 ) . Furthermore , another E3 ligase , MARCH1 , is overexpressed in obese humans and targets the insulin receptor for ubiquitin-mediated degradation ( Nagarajan et al . , 2016 ) . Taken together , these observations suggest that food abundance controls mammalian insulin receptor levels via E3 ligase-mediated degradation . Although in C . elegans we found the opposite changes in DAF-2 receptor levels , that is , they were reduced upon starvation and increased upon high glucose feeding , our observations suggest that nutritional cues may regulate insulin/IGF-1 receptor levels via a variety of mechanisms , including ubiquitination and proteasomal degradation , across species . In summary , we have demonstrated that interventions at almost the end of life can increase lifespan . We have established that auxin-induced degradation is suitable for targeting transmembrane receptors for non-invasive manipulations during developmental and longevity in vivo studies . We reconciled a longstanding question by providing evidence that dauer-like traits are not a spill-over of reprogrammed physiology from developing L2d pre-dauers . Instead , the essential growth-related functions of DAF-2 are causing deficits when applied during development or growth phases . We have shown that tissue-specific interventions or global interventions beyond reproduction or growth extend lifespan without pathology or deficits . Degradation of DAF-2/insulin/IGF-1 receptor might not be an artificial intervention since DAF-2/insulin/IGF-1 receptor abundance is read-out to adapt metabolism to food abundance . Dissecting intrinsic DAF-2/insulin/IGF-1 receptor abundance in response to nutritional cues may impact our understanding of nutrient sensing in promoting health during old age . All strains were maintained on NGM plates and OP50 Escherichia coli at 15°C as described . The strains and primers used in this study can be found in Supplementary file 3 . Statistical analysis was either done by using RStudio or Excel . All plots have been made using RStudio ( 1 . 2 . 5001 ) . The packages ggplot2 , survminer , dplyr were required for some plots . Auxin ( indole-3-acetic acid , Sigma #I3750 ) was dissolved in DMSO to prepare a 400 mM stock solution and stored at 4°C . Auxin was added to NGM agar that has cooled down to about 60°C before pouring the plates ( Zhang et al . , 2015 ) . For lower concentrations ( 1 μM and 10 μM ) , the 400 mM stock dilution was further diluted in DMSO . Control plates contained the same amount of DMSO ( 0 . 25% for 1 mM auxin plates ) . For lifespans , plates were supplemented with FUdR to the final concentration of 50 μM . The sgRNA targeting the terminal exon of the annotated daf-2 isoform a , 5’ ( G ) TTTGGGGGTTTCAGACAAG 3’ was cloned into the PU6:: sgRNA ( F + E ) plasmid backbone , pIK198 ( Katic et al . , 2015 ) , yielding plasmid pIK323 . The initial guanine was added to aid the transcription of the sgRNA . The underlined nucleotides in the sgRNA correspond to the stop codon of the DAF-2 ( A ) protein . The CRISPR tag repair template pIK325 degron::3xFLAG::SL2::SV40::NLS::degron::wrmscarlet::egl-13 NLS was assembled using the SapTrap method ( Schwartz and Jorgensen , 2016 ) from the following plasmids: pMLS257 ( repair template-only destination vector ) , pIK320 ( wrmscarlet::syntron-embedded LoxP-flanked , reverse Cbr-unc-119 ) , pMLS285 ( egl-13 NLS N-tagging connector ) , pIK321 ( linker::auxin degron::3XFLAG::SL2 operon::SV40 NLS::linker::auxin degron C-tagging connector ) , and phosphorylated pairs of hybridized oligonucleotides oIK1182 5’TGGTCGGCTTTCGGTGAAAATGAGCATCTAATCGAGGATAATGAGCATCATCCACTTGTC 3’ , oIK1183 5’CGCGACAAGTGGATGATGCTCATTATCCTCGATTAGATGCTCATTTTCACCGAAAGCCGA 3’ , and oIK1184 5’ACGAACCCCCAAAAAATCCCGCCTCTTAAATTATAAATTATCTCCCACATTATCATATCT 3’ , oIK1185 5’TACAGATATGATAATGTGGGAGATAATTTATAATTTAAGAGGCGGGATTTTTTGGGGGTT 3’ , respectively . Modules pIK320 and pIK321 ( this study ) , compatible with the SapTrap kit , were assembled through a combination of synthetic DNA ( Integrated DNA technologies ) and molecular cloning methods . EG4322 ttTi5605; unc-119 ( ed3 ) animals were injected with a mix consisting of the sgRNA pIK323 at 65 ng/ml , tag repair template pIK325 at 50 ng/ml , pIK155 Peft-3::Cas9::tbb-2 3’UTR at 25 ng/ml and fluorescent markers pIK127 Peft-3::GFP::h2b::tbb-2 3’UTR at 20 ng/ml , and Pmyo-3::GFP at 10 ng/ml . Among the non-Unc F2 progeny of the injected animals not labeled with green fluorescence were correctly tagged daf-2 ( bch40 [degron::3xFLAG::SL2::SV40 NLS::degron::wrmscarlet::egl-13 NLS] ) animals . We were able to recover two independent CRISPR alleles daf-2 ( bch39 ) and daf-2 ( bch40 ) . pIK280 ( TIR1::mRuby::tbb-2 in a MosSCI-compatible backbone ) ( Frøkjær-Jensen et al . , 2012; Frøkjaer-Jensen et al . , 2008 ) was created by Gibson assembly ( Gibson et al . , 2009 ) from templates including pLZ31 ( Zhang et al . , 2015 ) . Promoter regions were inserted by Gibson assembly of PCR products into pIK280 to express TIR-1::mRuby in different tissues . Such plasmids were injected into EG4322 ttTi5605; unc-119 ( ed3 ) animals ( Frøkjær-Jensen et al . , 2012 ) . The strains are IFM160 bchSi59 [Pmyo-3::TIR1::mRuby::tbb-2] II; unc-119 ( ed3 ) , IFM161 bchSi60 [Pvha-6::TIR1::mRuby::tbb-2] II; unc-119 ( ed3 ) , and IFM164 bchSi64 [Prab-3::TIR1::mRuby::tbb-2] II; unc-119 ( ed3 ) . RNAi bacteria cultures were grown overnight in LB with carbenicillin ( 100 µg/ml ) and tetracycline ( 12 . 5 µg/ml ) , diluted to an OD600 of 1 , and induced with 1 mM IPTG and spread onto NGM plates containing tetracycline ( 12 . 5 µg/ml ) and ampicillin ( 50 µg/ml ) as described in Ewald et al . , 2017b . Plasmid pL4440 was used as an empty RNAi vector ( EV ) control . The daf-2 ( RNAi ) clone was a kind gift from the Blackwell lab and was sequenced for validation . Synchronized C . elegans , on their first day of adulthood , were shifted to auxin plates for different time points . About 2000–5000 adult C . elegans per condition were disrupted using beads in lysis buffer ( RIPA buffer [ThermoFisher #89900] ) , 20 mM sodium fluoride ( Sigma #67414 ) , 2 mM sodium orthovanadate ( Sigma #450243 ) , and protease inhibitor ( Roche #04693116001 ) and kept on ice for 15 min before being centrifuged for 10 min at 15 , 000 × g . For equal loading , the protein concentration of the supernatant was determined with BioRad DC protein assay kit II ( #5000116 ) and standard curve with Albumin ( Pierce #23210 ) . Samples were boiled at 37°C for 30 min , shortly spun down , and 40 μg of protein was loaded onto NuPAGE Bis-Tris 10% Protein Gels ( ThermoFisher #NP0301BOX ) , and proteins were transferred to nitrocellulose membranes ( Sigma #GE10600002 ) . Western blot analysis was performed under standard conditions with antibodies against Tubulin ( Sigma #T9026 , 1:1000 ) ( Sigma #F3165 , 1:1000 ) , FLAG-HRP ( Sigma #A8592 , 1:1000 ) , and Degron ( MBL #M214-3 , 1:1000 ) . HRP-conjugated goat anti-mouse ( Cell Signaling #7076 , 1:2000 ) secondary antibodies were used to detect the proteins by enhanced chemiluminescence ( Bio-Rad #1705061 ) . Quantification of protein levels was determined using ImageJ software and normalized to loading control ( Tubulin ) . Statistical analysis was performed using either a two-tailed or one-tailed t-test . All western blots and quantifications can be found in Source data 1 and Source data 2 . Transgenic daf-16::gfp; DAF-2::degron C . elegans were grown on plates for the indicated length of time supplemented with the corresponding concentration auxin at 20°C . For image acquisition , the animals were placed on freshly made 2% agar pads and anesthetized with tetramisole ( Teuscher and Ewald , 2018 ) . Images were taken with an upright bright-field fluorescence microscope ( Tritech Research , model: BX-51-F ) and a camera of the model DFK 23U × 236 ( Teuscher and Ewald , 2018 ) . For quantification , the animals were observed under a fluorescent stereomicroscope after the indicated amount of time has passed . sod-3p::gfp and gst-4p::gfp animals were incubated overnight at 20°C and quantified the next morning . L4 larvae were used for quantification . Statistical analysis was performed by using Fisher’s exact test for daf-16::gfp and gst-4::gfp and two-tailed t-test for sod-3::gfp . The DMSO control was compared to the ones treated with various concentrations of auxin . As described in Ewald et al . , 2012 , L4 C . elegans of wild-type N2 and DAF-2::degron were picked to plates at 15°C . After 2 days , the adult animals were shifted to new plates and were allowed to lay eggs for 2 hr . The stage of the offspring and their health was assayed 4 days later at 20°C . Statistical analysis was performed by using a two-tailed t-test . Synchronized L1 C . elegans were put on 1 mM auxin plates and incubated at 15°C , 20°C , and 25°C . At the indicated time points , the animals were washed off with M9 , shortly centrifuged down , and SDS was added for a final concentration of 1% . After 10 min of gentle agitation , the animals were put on plates and checked for survival . Adult C . elegans were placed on 1 mM auxin plates ( for ‘DAF-2::degron’ ) or DMSO plates ( for ‘DAF-2::degron’ , daf-2 ( e1368 ) and daf-2 ( e1370 ) ) and shifted to 25°C . Dauer-like offspring or size-matching controls were picked after 4 days , anesthetized in 10 mM sodium azide , and mounted on 2% agarose pads . Images were taken at 40× magnification using an inverted microscope ( Tritech Research , MINJ-1000-CUST ) and a camera of the model DFK 23U × 236 . Adult C . elegans were put on 1 mM auxin plates , or DMSO plate seeded with OP50 containing a 1:100 dilution of red fluorescent latex bead solution ( Sigma #L3280 ) and shifted to 25°C . Dauer offspring and control L2/L3 larvae on the bacterial lawn were picked after 4 days , anesthetized in 10 mM sodium azide , and mounted on 2% agarose pads . An upright bright-field fluorescence microscope ( Tritech Research , model: BX-51-F ) and a camera of the model DFK 23U × 236 were used for image acquisition . The presence of beads in the intestine was checked at 20× magnification . A two-tailed t-test was used for analysis . Bleached eggs were synchronized for 2 days at 20°C in an M9 buffer supplemented with 5 μg/ml cholesterol to yield a highly synchronous L1 population ( Teuscher et al . , 2019 ) . The synchronized L1 larvae were put on OP50 NGM plates and then switched to plates containing 1 mM auxin at different time points . Dauer and non-dauer animals were counted after 2 days at 25°C or 3 days at 20°C . Bleached eggs were synchronized for 2 days at 20°C in M9 buffer supplemented with 5 μg/ml cholesterol to yield a highly synchronous L1 population ( Teuscher et al . , 2019 ) . The larvae were put on OP50 NGM plates for 24–25 hr , washed off and anesthetized with 0 . 25 mM tetramisole , and mounted on 2% agarose pads . An upright bright-field fluorescence microscope ( Tritech Research , model: BX-51-F ) was used to count the cells in the developing gonad . Freshly hatched L1 have two germ stem cells ( Z2-3 ) , and L2 have , on average , 16 germ stem cells ( Mainpal et al . , 2015 ) . However , we counted all visible cells in the gonad ( i . e . , both somatic and germ cells ) , which are four for the freshly hatched L1 ( Z1-4 ) and 22 for animals around the L1/L2 transition ( Hubbard and Greenstein , 2000 ) . We relied on the characteristic round shapes of germ cells ( Altun and Hall , 2002 ) ( https://www . wormatlas . org/hermaphrodite/somatic%20gonad/Images/somaticfig3leg . htm ) . Using these criteria , we might have included DTC and other cells inside the gonad . L4 C . elegans maintained at 15°C were picked on 1 mM auxin or control plates and shifted to the indicated temperatures . On the second day of adulthood , animals were mounted on 2% agar pads and anesthetized with 0 . 25 mM tetramisole . Images were taken at 40× magnification on an inverted microscope ( Tritech Research , MINJ-1000-CUST ) and a camera of the model DFK 23U × 236 . Statistical analysis was performed by using a two-tailed t-test for egg retention and Fisher’s exact test for germline morphology . L4 C . elegans maintained at 15°C were picked on 1 mM auxin or control plates and shifted to the indicated temperatures . On the second day of adulthood , animals were mounted on 2% agar pads and anesthetized with 0 . 25 mM tetramisole . Images were taken at 10× magnification with an upright bright-field fluorescence microscope ( Tritech Research , model: BX-51-F ) and a camera of the model DFK 23U × 236 . Body lengths were measured by placing a line through the middle of the body , starting from head to tail , using ImageJ 1 . 51 j . Statistical analysis was performed by using a two-tailed t-test . L4 C . elegans maintained at 15°C were picked on 1 mM auxin or control plates and shifted to the indicated temperatures . C . elegans were shifted when necessary to fresh plates , and the progeny was counted after 2 days of development . Animals that crawled off the plate , dug into the agar , or bagged precociously were censored . Statistical analysis was performed by using a two-tailed t-test . Synchronized L1 C . elegans were cultured at 15°C or 20°C on OP50 and shifted at the L4 stage to NGM plates containing 50 μM FUdR and auxin or DMSO . Bursted , dried out , or escaped animals were censored , and animals were considered dead when they failed to respond to touch and did not show any pharyngeal pumping . For late-life auxin lifespan assays: L4 C . elegans were picked onto NGM plates containing 50 μM FUdR . At day 20 or 25 of adulthood , plates were top-coated either DMSO or auxin to reach a final concentration of 0 . 25% DMSO or 1 mM auxin with 0 . 25% DMSO . Log-rank was used for statistical analysis . The plots were made by using the R-package survminer or JMP 14 . 1 . All statistics can be found in Supplementary file 1 . Oxidative stress assay was modified from Ewald et al . , 2017a . C . elegans of the L1 or L4 stage were shifted to auxin or DMSO plates , washed off at the indicated time point , incubated with 5 mM sodium arsenite in U-shaped 96-well plates , and put into the wMicroTracker ( MTK100 ) for movement scoring . For statistical analysis , the area under the curve was measured , and the mean for each run was calculated . Statistical analysis was performed by using a paired sample t-test . All plots can be found in Supplementary file 2 .	The goal of geroscience , or research into old age , is to promote health during old age , and thus , to increase lifespan . In the body , the groups of biochemical reactions , or ‘pathways’ , that allow an organism to sense nutrients , and regulate growth and stress , play major roles in ensuring healthy aging . Indeed , organisms that do not produce a working version of the insulin/IGF-1 receptor , a protein involved in one such pathway , show increased lifespan . In the worm Caenorhabditis elegans , mutations in the insulin/IGF-1 receptor can even double their lifespan . However , it is unclear whether this increase can be achieved once the organism has reached old age . To answer this question , Venz et al . genetically engineered the nematode worm C . elegans so that they could trigger the rapid degradation of the insulin/IGF-1 receptor either in the entire organism or in a specific tissue . Venz et al . started by aging several C . elegans worms for three weeks , until about 75% had died . At this point , they triggered the degradation of the insulin/IGF-1 receptor in some of the remaining worms , keeping the rest untreated as a control for the experiment . The results showed that the untreated worms died within a few days , while worms in which the insulin/IGF-1 receptor had been degraded lived for almost one more month . This demonstrates that it is possible to double the lifespan of an organism at the very end of life . Venz et al . ’s findings suggest that it is possible to make interventions to extend an organism’s lifespan near the end of life that are as effective as if they were performed when the organism was younger . This sparks new questions regarding the quality of this lifespan extension: do the worms become younger with the intervention , or is aging simply slowed down ?	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "developmental", "biology" ]	2021	End-of-life targeted degradation of DAF-2 insulin/IGF-1 receptor promotes longevity free from growth-related pathologies
Apicomplexan parasites contain a conserved protein CelTOS that , in malaria parasites , is essential for traversal of cells within the mammalian host and arthropod vector . However , the molecular role of CelTOS is unknown because it lacks sequence similarity to proteins of known function . Here , we determined the crystal structure of CelTOS and discovered CelTOS resembles proteins that bind to and disrupt membranes . In contrast to known membrane disruptors , CelTOS has a distinct architecture , specifically binds phosphatidic acid commonly present within the inner leaflet of plasma membranes , and potently disrupts liposomes composed of phosphatidic acid by forming pores . Microinjection of CelTOS into cells resulted in observable membrane damage . Therefore , CelTOS is unique as it achieves nearly universal inner leaflet cellular activity to enable the exit of parasites from cells during traversal . By providing novel molecular insight into cell traversal by apicomplexan parasites , our work facilitates the design of therapeutics against global pathogens . Apicomplexan parasites of the genera Plasmodium , Babesia , Cytauxzoon and Theileria are responsible for the established and emerging global infectious diseases malaria ( Black et al . , 2010; Price et al . , 2007 ) , babesiosis ( Homer et al . , 2000 ) , cytauxzoonosis ( Sherrill et al . , 2015 ) , and theileriosis ( Shaw , 2003 ) , respectively . Although these parasites infect a range of hosts and cause different diseases , they share biological adaptations for traversal through a variety of cells within their mammalian hosts and arthropod vectors to complete their life cycle ( Homer et al . , 2000; Sherrill et al . , 2015; Shaw , 2003; Mota et al . , 2001; Yuda and Ishino , 2004; Amino et al . , 2008; Sinnis and Zavala , 2012; Ménard et al . , 2013; Gueirard et al . , 2010 ) . Traversal refers to the ability of a eukaryotic parasite to move through host or vector cells to complete essential segments of a complicated life cycle . Within the host , cell traversal by parasites is essential for successful infection as sporozoites must negotiate and cross several distinct cell types until they enter defined cells that are receptive for replication , leading to disease progression . Traversal is also essential for transmission by vectors as parasites must cross the arthropod midgut epithelia to mature into sporozoites that are the transmissible form of these apicomplexan parasites . Traversal of parasites through host and vector cells relies on a number of different biological processes mediated by various parasite proteins . Gliding motility is the characteristic motion of parasites as they move through various tissues and invade cells ( Yuda and Ishino , 2004; Ménard et al . , 2013; Harupa et al . , 2014; Moreira et al . , 2008 ) . Parasites also physically breach cell barriers during the invasion of cells ( Homer et al . , 2000; Sherrill et al . , 2015; Shaw , 2003; Mota et al . , 2001; Yuda and Ishino , 2004; Amino et al . , 2008; Kadota et al . , 2004; Risco-Castillo et al . , 2015 ) . Finally , parasites have developed strategies for immune evasion as they migrate through various cells , including phagocytic Kupffer cells ( Sinnis and Zavala , 2012; Zheng et al . , 2014 ) . Ongoing efforts to identify and elucidate the function and mechanism of parasite proteins involved in diverse cell traversal processes will advance our understanding of parasite biology , and will unveil new targets for therapeutics . Of all the apicomplexan parasites , therapeutics for Plasmodium falciparum and Plasmodium vivax are desperately needed as these parasites are leading causes of human death and major burdens to socio-economic development ( Black et al . , 2010; Price et al . , 2007 ) . Cell-traversal protein for ookinetes and sporozoites ( CelTOS ) is unique among traversal proteins as it is essential for traversal of malaria parasites in both the mosquito vector and human host and is therefore critical for malaria transmission and disease pathogenesis ( Kariu et al . , 2006 ) . Additionally , CelTOS has been identified as a promising malaria vaccine candidate referred to as Antigen 2 ( Doolan et al . , 2003 ) . Immunization of mice with recombinant CelTOS results in humoral and cellular immune responses that reduced infection , demonstrating that targeting CelTOS is a viable approach for developing a malaria vaccine ( Bergmann-Leitner et al . , 2010; Ferraro et al . , 2013; Bergmann-Leitner et al . , 2011 , 2013 ) . Even though CelTOS is critical during cell traversal by malaria parasites and is a leading transmission- and infection-blocking malaria vaccine candidate , its function remains unknown as it has no sequence similarity to proteins of known function . In this study , we determined the structure of CelTOS in order to gain insight into its function . CelTOS structurally resembles viral membrane fusion proteins and a bacterial pore-forming toxin . This observation informed our hypothesis that apicomplexan CelTOS directly binds to and breaches plasma membranes . We show that CelTOS specifically binds phosphatidic acid , a lipid predominantly found in the inner leaflet of plasma membranes , and potently disrupts defined liposomes composed of phosphatidic acid . We determine that liposome disruption is achieved by the formation of CelTOS-dependent pores . Additionally , microinjection of CelTOS into Xenopus oocytes results in membrane damage . Together , these results demonstrate CelTOS is the only known apicomplexan protein with universal inner leaflet cellular activity as it addresses phosphatidic acid found specifically on the cytoplasmic face of plasma membranes , and disrupts these membranes to enable the exit of apicomplexan parasites through diverse vector and host cells . In addition to discovering this unique role of CelTOS in cell traversal , our work on the structure , function , and mechanism of CelTOS will enable structural vaccinology ( Dormitzer et al . , 2012 ) to produce a potent protective malaria vaccine , and inform the development of other therapeutics targeting CelTOS . As CelTOS is important for multiple stages of the parasite life cycle , we examined if CelTOS was evolutionarily conserved among other apicomplexan pathogens , including those with recently sequenced genomes . We found CelTOS is conserved across various diverse branches of apicomplexan parasites including the hemosporidia ( Plasmodium spp . ) and piroplasms ( Theileria , Babesia , Cytauxzoon spp . ) , groups that are thought to have diverged more than 100 million years ago ( DeBarry and Kissinger , 2011 ) ( Figure 1A and Figure 1—figure supplement 1 ) . Hence , CelTOS represents an ancient and widespread adaptation that is common to apicomplexans that have two host life cycles alternating between asexual replication in their vertebrate hosts and a sexual cycle in their arthropod vectors . This is supported by the lack of CelTOS in the apicomplexan Toxoplasma gondii that does not require an arthropod vector for transmission . This evolutionary conservation and importance in disease pathogenesis and transmission prompted the in-depth molecular study of CelTOS that is applicable to diverse pathogenic apicomplexan parasites . 10 . 7554/eLife . 20621 . 003Figure 1 . Alignment and structure of the conserved apicomplexan protein CelTOS from the human pathogen Plasmodium vivax ( PvCelTOS ) . ( A ) Alignment of CelTOS from apicomplexan parasites Plasmodium , Babesia , Theileria , Cytauxzoon , and mapping of the structural elements . In the alignment , grey shading represents similarity , black shading represents identity . The secondary structure features are based on the crystal structure of PvCelTOS and shown in green . Protein accession codes as follows: Plasmodium vivax [UniProtKB - A5JZX5] , Plasmodium falciparum [UniProtKB - Q8I5P1] , Babesia microti [UniProtKB - I7J9D8] , Theileria parva [UniProtKB - Q4N982] , Cytauxzoon felis [PiroplasmaDB - CF003135] . ( B ) PvCelTOS is an alpha helical dimer that resembles a tuning fork . Each monomer is in green or white ribbon and surface representation . ( C ) Surface maps of CelTOS dimer showing the electrostatic surface potential colored from red ( −5 kT e−1 ) to blue ( 5 kT e−1 ) . ( D ) The CelTOS monomer shown as ribbon representation can be separated into two distinct subdomains composed of N-terminal ( α-helices 1 and 2 ) and C-terminal helices ( α-helices 3 and 4 ) . ( E ) Surface maps of CelTOS monomer reveals the inner hydrophobic surfaces composed of the dimer interface and one face of the tuning fork prongs . Coloring as in Figure 1C . DOI: http://dx . doi . org/10 . 7554/eLife . 20621 . 00310 . 7554/eLife . 20621 . 004Figure 1—source data 1 . Data collection , phasing and refinement statistics . DOI: http://dx . doi . org/10 . 7554/eLife . 20621 . 00410 . 7554/eLife . 20621 . 005Figure 1—source data 2 . Sedimentation equilibrium analytical ultracentrifugation analysis for Pf and PvCelTOS . Three independent global fits for three concentrations and two speeds demonstrate Pf and PvCelTOS are dimers in solution . The theoretical monomer molecular weight for PfCelTOS and PvCelTOS are 18 . 655 kDa and 18 . 665 kDa , respectively . DOI: http://dx . doi . org/10 . 7554/eLife . 20621 . 00510 . 7554/eLife . 20621 . 006Figure 1—figure supplement 1 . Alignment of CelTOS from Apicomplexan parasites and mapping of the structural elements . In the alignment , grey shading represents similarity , black shading represents identity . Protein accession codes for CelTOS from the following Apicomplexan parasites are as follows: Plasmodium falciparum [UniProtKB - Q8I5P1] , Plasmodium vivax [UniProtKB - A5JZX5] , Plasmodium berghei [UniProtKB - Q4YC08] , Plasmodium knowlesi [UniProtKB - B3LCG1] , Plasmodium reichenowi [UniProtKB - A0A060RVP1] , Plasmodium yoelii [UniProtKB - Q6T944] , Plasmodium chabaudi chabaudi [UniProtKB - A0A077TR60] , Plasmodium fragile [UniProtKB - A0A0D9QNC3] , Plasmodium inui [UniProtKB - W7ABI3] , Plasmodium vinckei vinckei [UniProtKB - W7B0C7] , Plasmodium vinckei petteri [UniProtKB - W7B0C7] , Babesia microti [UniProtKB - I7J9D8] , Babesia bovis [UniProtKB - A7AQQ4] , Babesia bigemina [UniProtKB - A0A061DAC2] , Babesia equi [UniProtKB - L0AZQ6] , Theileria parva [UniProtKB - Q4N982] , Theileria annulata [UniProtKB - Q4UGH3] , Theileria orientalis [UniProtKB - J7M8E0] , Theileria equi [NCBI Reference Sequence: XP_004830035 . 1] and Cytauxzoon felis [PiroplasmaDB - CF003135] . DOI: http://dx . doi . org/10 . 7554/eLife . 20621 . 00610 . 7554/eLife . 20621 . 007Figure 1—figure supplement 2 . Electron-density maps . 2Fo-Fc electron density contoured at 1 . 5σ around a representative section of PvCelTOS . DOI: http://dx . doi . org/10 . 7554/eLife . 20621 . 007 The mechanism by which CelTOS enables the traversal of apicomplexan parasites through cells within the mammalian host and arthropod vector has been elusive because CelTOS has no sequence similarity to proteins of known function . Therefore , we determined the crystal structure of CelTOS from the human pathogen P . vivax ( PvCelTOS ) in order to obtain insight into function ( Figure 1 , Figure 1—figure supplement 2 and Figure 1—source data 1 ) . As CelTOS is highly conserved across apicomplexans ( Figure 1A and Figure 1—figure supplement 1 ) , the structural inferences derived from the PvCelTOS structure will apply to apicomplexan parasites generally . CelTOS forms an alpha helical dimer that resembles a tuning fork ( Figure 1B ) . The dimer has a large buried surface area of 3003 Å2 . Consistent with the structure and large buried surface area , we established that CelTOS is an obligate dimer in solution by sedimentation equilibrium analytical ultracentrifugation ( Figure 1—source data 2 ) . Two N-terminal helices ( α-helices 1 and 2 ) of one monomer pack against the C-terminal helices ( α-helices 3 and 4 ) of a second monomer to form the dimer . The outer surface of the CelTOS dimer is mildly hydrophilic ( Figure 1C and E ) and masks inner hydrophobic surfaces created between the two tines of the CelTOS dimer and the dimer interface ( Figure 1E ) . We compared the structure of the CelTOS monomer ( Figure 1D ) against all structures in the Protein Data Bank to identify proteins with similar structure and to inform function using the DALI server ( Holm et al . , 2010 ) . Strikingly , CelTOS resembles class I viral membrane fusion glycoproteins and a bacterial pore-forming toxin with roles in membrane binding and disruption ( Figure 2 ) . The C-terminal helices of CelTOS show structural similarity to HIV-1 gp41 ( Figure 2A ) and Mycobacterium bovis ESAT-6 ( Figure 2D ) , and the N-terminal helices show similarity to Hendravirus fusion core ( Figure 2B ) ; Nipahvirus fusion subunit core ( Figure 2C ) ( Buzon et al . , 2010; Lou et al . , 2006; Renshaw et al . , 2005; Hsu et al . , 2003 ) . This informed the hypothesis that Plasmodium CelTOS may function to bind and disrupt cell membranes , consistent with a role in host and vector cell traversal by malaria parasites . While CelTOS resembles these viral and bacterial proteins , the CelTOS structure is unique and distinct as it contains two independent subdomains that both could act as membrane disruption modules . 10 . 7554/eLife . 20621 . 008Figure 2 . PvCelTOS is structurally similar to viral and bacterial proteins that disrupt membranes . All panels are structural overlays of PvCelTOS ( green ) with pathogenic membrane disrupting proteins ( orange ) : ( A ) HIV-1 gp41 PDB 3P30 ( Dali Z-score: 3 . 2 , rmsd 3 . 2 ) , ( B ) Hendravirus fusion protein PDB 3N27 ( Dali Z-score: 3 . 1 , rmsd 3 . 3 ) ( orientation of CelTOS is rotated 60° along the y-axis compared to Figure 2A ) , ( C ) Nipahvirus fusion protein PDB 1WP7 ( Dali Z-score: 3 . 3 , rmsd 2 . 9 ) ( orientation of CelTOS is rotated 60° along the y-axis compared to Figure 2A ) , and ( D ) Mycobacterium bovis ESAT-6 PDB 1WA8 ( Dali Z-score: 2 . 6 , rmsd 2 . 3 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 20621 . 008 Given this structural similarity , we examined if CelTOS could directly bind cell membrane phospholipids and if binding was specific to a particular lipid subset using spotted arrays . Both P . falciparum ( Pf ) and P . vivax ( Pv ) CelTOS demonstrated significant binding to phosphatidic acid ( PA ) ( Figure 3 ) . The specificity to this small anionic headgroup phospholipid is unique and excludes other anionic but large headgroup phospholipids . Phosphatidic acid is predominantly found on the inner leaflet of plasma membranes ( Op den Kamp , 1979 ) . Specific binding of PvCelTOS to cardiolipin was also observed . However , the relevance of this binding specificity is unclear as it is not conserved in CelTOS from both P . falciparum and P . vivax . Limited or undetectable binding was observed to phospholipids predominantly found on the outer leaflet of plasma membranes including sphingomyelin ( SM ) and phosphatidylcholine ( PC ) ( Op den Kamp , 1979 ) . These results demonstrate that CelTOS has enhanced specificity for the cytosolic face of cell membranes and suggests CelTOS possesses a specific binding pocket that preferentially accommodates PA over the headgroups of other phospholipids . 10 . 7554/eLife . 20621 . 009Figure 3 . CelTOS binds the inner leaflet cell membrane lipid phosphatidic acid . ( A ) Spotted lipid array probed with PfCelTOS demonstrates significant binding to PA compared to the negative control ( BB ) in lipid arrays . Right panel: Normalized binding intensity from eight experiments shown as mean ± s . e . m . Statistical differences were determined by one-way ANOVA with matched replicates , followed by Dunnett’s multiple comparison test; p<0 . 05 , p<0 . 01 , *p<0 . 001 . GT - Glyceryl tripalmitate , DAG - Diacylglycerol , PA - Phosphatidic Acid , PS - Phosphatidylserine , PE - Phosphatidylethanolamine , PC - Phosphatidylcholine , PG - Phosphatidylglycerol , CL - Cardiolipin , PI - Phosphatidylinositol , PI4P - Phosphatidylinositol ( 4 ) -phosphate , PIP2 - Phosphatidylinositol ( 4 , 5 ) -bisphosphate , PIP3 - Phoshatidylinositol ( 3 , 4 , 5 ) -trisphosphate , CH - Cholesterol , SM - Sphingomyelin , SF - 3-sulfogalactosylceramide , and BB - Blue blank . ( B ) Spotted lipid array probed with PvCelTOS demonstrates significant binding to PA and PS compared to the negative control ( BB ) in lipid arrays . Right panel: Normalized binding intensity from three experiments shown as mean ± s . e . m . Statistical differences were determined by one-way ANOVA with matched replicates , followed by Dunnett’s multiple comparison test; p<0 . 05 , p<0 . 01 , *p<0 . 001 . Acronyms as in Figure 3A . DOI: http://dx . doi . org/10 . 7554/eLife . 20621 . 009 Further to identifying its lipid-binding specificity , we established that CelTOS disrupts membranes in a liposome disruption assay ( Figure 4 and Figure 4—figure supplement 1 ) ( Saito et al . , 2000 ) . Liposomes that mimic cell membranes were created containing carboxyfluorescein at self-quenching concentrations using 1-palmitoyl-2-oleoyl-sn-glycero-3-phosphate ( POPA ) , 1-palmitoyl-2-oleoyl-sn-glycero-3-phospho-L-serine ( POPS ) and 1-palmitoyl-2-oleoyl-sn-glycero-3-phosphocholine ( POPC ) . Liposomes containing 8:2 POPC:POPA mixtures ( PA liposomes ) or 8:2 POPC:POPS mixtures ( PS liposomes ) represent the inner leaflet of plasma membranes , while pure POPC ( PC liposomes ) represents the outer leaflet . 10 . 7554/eLife . 20621 . 010Figure 4 . CelTOS disrupts liposomes containing phosphatidic acid ( PA liposomes ) . ( A ) Time dependence of liposome disruption at various concentrations of PfCelTOS , a representative plot of three independent experiments is presented . The solid lines represent the fit of liposome disruption to Equation 2 and the dashed lines represent the raw data of liposome disruption . ( B ) The time constant , Tau , determined from the fit to Equation 2 , is plotted against the concentration of PfCelTOS for three technical replicates is shown as mean ± s . e . m . ( C ) Time dependence of liposome disruption at various concentrations of PvCelTOS . ( D ) The time constant , Tau , plotted against the concentration of PvCelTOS . DOI: http://dx . doi . org/10 . 7554/eLife . 20621 . 01010 . 7554/eLife . 20621 . 011Figure 4—figure supplement 1 . CelTOS minimally disrupts liposomes composed of phosphatidylserine and phosphatidylcholine , and the negative control protein EcIsPF does not disrupt the liposomes composed of the aforementioned lipids and phosphatidic acid . ( A , B ) Time dependence of PS liposome disruption at various concentrations of PfCelTOS and PvCelTOS respectively , a representative plot of three independent experiments is presented . The solid lines represent the fit of liposome disruption to Equation 2and the dashed lines represent the raw data of liposome disruption . ( C ) Time dependence of PC liposome disruption with 10 µM PfCelTOS and PvCelTOS respectively , a representative plot of three independent experiments is presented . ( D ) Time dependence of PA , PS and PC liposome disruption with 10 µM E . coli IspF , a control protein with similar mass , charge and purification protocol as CelTOS . DOI: http://dx . doi . org/10 . 7554/eLife . 20621 . 011 Upon incubation of either Pf or PvCelTOS with PA liposomes , carboxyfluorescein was released , fluorescence dequenched , and the time-dependent increase in liposome disruption was measured ( Figure 4 ) . Nanomolar concentrations of CelTOS disrupted PA liposomes consistent with a potent role in the disruption of PA containing membranes ( Figure 4 ) . In contrast , minimal disruption of PS and PC liposomes by CelTOS was observed and this required micromolar concentrations of CelTOS ( Figure 4—figure supplement 1A–C ) . Liposome disruption was specific to CelTOS as addition of the control protein E . coli IspF , that has similar mass , isoelectric point and purification method as CelTOS , had no effect ( Figure 4—figure supplement 1D ) . These results demonstrate that the CelTOS protein alone is necessary and sufficient for membrane disruption , specifically targets inner leaflet phosphatidic acid lipid compositions , and has a conserved function across Plasmodium spp . We investigated the mechanism by which CelTOS disrupts liposomes . CelTOS could disrupt liposomes either by generalized solubilization and mixed micelle formation similar to detergents , or by forming defined protein-dependent pores within the lipid bilayer . A key test of pore formation is whether leakage of liposome contents can be prevented by plugging the pore using a molecule of the correct Stoke’s radius ( Saito et al . , 2000 ) . We examined the CelTOS-dependent release of carboxyfluorescein from liposomes in the presence of 20 μM dextran molecules of various molecular weights: 66 . 9 kDa ( Stoke’s radius:~5 . 8 nm ) , 148 kDa ( Stoke’s radius:~8 nm ) , 500 kDa ( Stoke’s radius: 14 . 7 nm ) , 1500–2800 kDa ( Stoke’s radius ranging from ~25 to ~60 nm ( Tang et al . , 2016 ) ; Stoke’s radius of 2000 kDa Dextran: 27 nm [Armstrong et al . , 2004] ) . Remarkably , only the 500 kDa dextran blocked carboxyfluorescein release from liposomes disrupted with PfCelTOS and PvCelTOS , respectively ( Figure 5A and B ) . Dextran molecules of higher and lower molecular weights and Stoke’s radius had no effect on carboxyfluorescein release , similar to experiments without dextran . Therefore , only the 500 kDa dextran has the correct Stoke’s radius to occupy the pore , while the remaining dextran molecules are either too small and diffuse through the pore or too large and may not enter the pore . The 500 kDa dextran significantly lowered the maximum amount of carboxyfluorescein released disrupted ( A , in Equation 2 ) compared to experiments with molecules of higher and lower molecular weights , or without dextran ( Figure 5—figure supplement 1 ) . This suggests that CelTOS forms pores of uniform size in liposomes and that the dextran-pore complex might be slowly reversibly tying up the protein in a non-functional pore . We visualized the CelTOS-pore in liposomes by transmission electron microscopy of negative stained liposomes ( Figure 5C ) . The average diameter of the pores , reported as mean ± s . d . , formed by PfCelTOS and PvCelTOS were 45 . 85 ± 15 . 55 nm and 58 . 62 ± 21 . 86 nm respectively ( Figure 5—figure supplement 2 ) . The dextran-blocking and electron microscopy studies suggest CelTOS forms a pore that opens a portal for parasite exit from invaded cells . The observed portal is large and strongly suggests that the membrane structure of CelTOS may continue to evolve after the initial carboxyfluorescein release . 10 . 7554/eLife . 20621 . 012Figure 5 . CelTOS forms pores in lipid membranes and a 500 kDa Dextran ( Stoke’s radius = 14 . 7 nm ) blocks the release of carboxyfluorescein from PA liposomes disrupted by PfCelTOS and PvCelTOS . Time dependence of PA liposome disruption by ( A ) PfCelTOS or ( B ) PvCelTOS at 250 nM in the presence of dextran molecules of various molecular weights , a representative plot is presented . The solid lines represent the fit of liposome disruption to Equation 2 and the dashed lines represent the raw data of liposome disruption . ( C ) Transmission electron microscopy of negative stained liposomes alone ( left ) , liposomes treated with PfCelTOS ( middle ) , and liposomes treated with PvCelTOS ( right ) reveals CelTOS-dependent pores ( red arrows ) . DOI: http://dx . doi . org/10 . 7554/eLife . 20621 . 01210 . 7554/eLife . 20621 . 013Figure 5—figure supplement 1 . Quantitation of the dextran block experiment . Reduction in the maximum amount of liposomes disrupted ( A from Equation 2 ) in the presence of dextran molecules after incubation with 250 nM of ( a ) PfCelTOS or ( b ) PvCelTOS . The maximum amount of liposomes disrupted of three technical replicates is shown as mean ± s . e . m . Statistical differences were determined for seven replicates by one-way ANOVA with matched replicates , followed by Dunnett’s multiple comparison test . DOI: http://dx . doi . org/10 . 7554/eLife . 20621 . 01310 . 7554/eLife . 20621 . 014Figure 5—figure supplement 2 . Diameters of pores in liposomes formed by CelTOS . Diameters of pores in liposomes formed by PfCelTOS ( 45 . 85 ± 15 . 55 nm ) and PvCelTOS ( 58 . 62 ± 21 . 86 nm ) determined by transmission electron microscopy . Diameters are reported as mean ± s . d . for 23 PfCelTOS-pores and 18 PvCelTOS-pores . DOI: http://dx . doi . org/10 . 7554/eLife . 20621 . 014 The structural and activity data above suggest CelTOS functions within the cytosol of host and vector cells to disrupt cell plasma membranes from the cytoplasmic face . To test this hypothesis , we microinjected purified CelTOS into the cytosol of Xenopus oocytes and examined the effects on membrane integrity and cell survival ( Figure 6A and B , and Figure 6—figure supplement 1 ) . Microinjection allows for direct examination of the cellular effect of proteins . Only minor membrane phenotypes due to the injection process were observed in oocytes injected with buffer and in those injected with the negative control protein , EcIspF . In contrast , microinjection of Pf or PvCelTOS caused significant increase in the number of cells that had damaged membranes ( Figure 6A and B , and Figure 6—figure supplement 1 ) . 10 . 7554/eLife . 20621 . 015Figure 6 . CelTOS disrupts cells by directly binding the cytosolic face of cell plasma membranes . ( A ) Microinjection of Xenopus oocytes with CelTOS results in membrane damage . Cells were incubated at room temperature and examined 24 hr post-injection . Five representative images from 24 individually microinjected oocytes per condition are shown . Scale bar represents 0 . 2 mm . ( B ) Percent of oocytes with damaged membranes from six independent experiments each with 24 technical replicates . Statistical differences were determined by Kruskal-Wallis and Dunn’s multiple comparison tests . The graph is shown as mean ± s . e . m . ( C ) RAW 264 . 7 macrophages used in a cytotoxicity assay demonstrate CelTOS is unable to disrupt cells from the extracellular surface . Membrane damage was measured by uptake of a membrane-impermeable fluorescent dye . Statistical differences were determined for three independent experiments each with three technical replicates by Kruskal-Wallis and Dunn’s multiple comparison test . The graph is shown as mean ± s . e . m . DOI: http://dx . doi . org/10 . 7554/eLife . 20621 . 01510 . 7554/eLife . 20621 . 016Figure 6—figure supplement 1 . Complete images for a representative experiment of Xenopus oocyte microinjection . Images were collected 24 hr after treatment of oocytes , and the treatment is denoted of the left . Scale bar represents 0 . 2 mm . DOI: http://dx . doi . org/10 . 7554/eLife . 20621 . 016 We examined if CelTOS can disrupt cell membranes from the extracellular face using RAW 264 . 7 macrophages . The cell membrane of Xenopus oocytes is protected from the extracellular environment by a vitelline membrane . This makes oocytes unsuitable for studying the effect of CelTOS on the outer surface of cell membranes . RAW macrophages have an exposed cell membrane and are an especially relevant model as they are similar to liver macrophages ( Kupffer cells ) , which are traversed by malaria parasites in a CelTOS-dependent process . No membrane disruption , measured by uptake of a membrane-impermeable fluorescent dye , was observed for macrophages in tissue culture when CelTOS was added to the extracellular media ( Figure 6C ) . This is fully consistent as CelTOS does not bind phospholipids found predominantly in the outer leaflet of plasma membranes , rendering it incapable of disrupting plasma membranes from the extracellular face ( Figure 3 , Figure 6C and Figure 4—figure supplement 1C ) . This study provides unprecedented insight into the structure , function and mechanism of a critical protein involved in host and vector cell traversal by apicomplexan parasites . We have identified CelTOS as the only parasite protein known to breach plasma membranes from the cytoplasmic face to enable the exit of parasites from cells during traversal . The data from this study support the model depicted in Figure 7 . CelTOS is released from parasite secretory organelles and adheres to the surface of the malaria parasite prior to and during the traversal of host and vector cells ( Bergmann-Leitner et al . , 2010 , 2011 ) . Our results show that when the parasite is extracellular , CelTOS has limited cell membrane disrupting activity due to its low preference for lipids on the outer surface of plasma membranes . Once the parasite invades the cytosol of host and vector cells , CelTOS binds with high specificity to lipids in the inner leaflet of the plasma membrane . Binding leads to localized membrane disruption and pore formation that creates a portal to ultimately direct the exit of the parasite from the host or vector cells to complete traversal . 10 . 7554/eLife . 20621 . 017Figure 7 . Model for CelTOS in cell traversal . ( A ) During the pre-erythrocytic stage , malaria parasite sporozoites traverse various host cells including Kupffer cells and hepatocytes . CelTOS forms pores and disrupts the cell membranes of these cells to allow exit of the sporozoites to complete traversal . The inset shows CelTOS is localized to the surface of sporozoites and disrupts plasma membranes by directly binding to phosphatidic acid ( PA ) , colored brown , in the inner leaflet to create a pore that enables sporozoite exit . EEF – exo-erythrocytic form . ( B ) During the mosquito stage , malaria parasite ookinetes traverse the mosquito midgut epithelium to reach the basal lamina where it develops into oocysts . CelTOS forms pores and disrupts vector cell membranes to direct ookinete exit during traversal through the mosquito midgut epithelium . The inset shows CelTOS is localized to the surface of ookinetes and disrupts plasma membranes by directly binding to phosphatidic acid ( PA ) , colored brown , in the inner leaflet to create a pore that enables ookinete exit . DOI: http://dx . doi . org/10 . 7554/eLife . 20621 . 017 Traversal of apicomplexan parasites across cells of their arthropod vectors and vertebrate hosts is required for transmission and disease progression ( Homer et al . , 2000; Shaw , 2003; Mota et al . , 2001; Yuda and Ishino , 2004; Amino et al . , 2008; Sinnis and Zavala , 2012 ) . Parasites utilize gliding motility during traversal and have the ability to breach plasma membranes from the extracellular space to invade cells . Parasite proteins involved in gliding motility ( Yuda and Ishino , 2004; Tardieux and Ménard , 2008 ) , and the breaching of plasma membranes from the extracellular space to enable entry ( Kadota et al . , 2004; Risco-Castillo et al . , 2015; Ishino et al . , 2005 ) have been reported . The current literature supports a model where perforin-like proteins ( PLPs ) mediate entry into various pre-erythrocytic cells including the PLP-1 dependent disruption of transient vacuoles created during entry of hepatocytes ( Risco-Castillo et al . , 2015; Ishino et al . , 2005 ) . This suggests a mechanism in which PLPs engage outer leaflet lipids for disruption , although direct studies probing the biochemical function and lipid specificity of PLPs have not been described . Further , PLPs have been shown to play a role in egress of parasites from the parasitophorous vacuole and red blood cells ( Garg et al . , 2013; Wirth et al . , 2014 ) . However , proteins involved in breaching the plasma membrane of invaded cells during traversal , from the cytoplasmic face , to enable the exit of parasites are unknown . We have discovered that secreted apicomplexan CelTOS attacks plasma membranes from the cytoplasmic face . CelTOS had previously been identified as a malaria antigen , and CelTOS-deficient parasites are unable to traverse cells of their host and vector ( Kariu et al . , 2006; Doolan et al . , 2003 ) . The results of our study indicate that CelTOS functions to breach plasma membranes from the cytoplasmic face to enable the exit of parasites from invaded host and vector cells during traversal . This study challenges the paradigm that motility alone is sufficient for parasite exit , as it is now clear that active membrane disruption by CelTOS is necessary for parasite release . This work identifies CelTOS-dependent pore formation leading to inner-leaflet membrane disruption as a new active molecular process during traversal , and presents novel avenues for preventing traversal by inhibiting inner leaflet membrane disruption . As CelTOS is conserved across several large groups of apicomplexan parasites including Plasmodium spp . , Cytauxzoon , Theileria and Babesia , this finding informs fundamental pathogen biology applicable to multiple diseases and microorganisms of global importance with particular relevance to cell infection , traversal and membrane disruption . While CelTOS resembles viral membrane fusion proteins and a bacterial pore-forming toxin , our findings define a new class of membrane disruption proteins . CelTOS has a distinct structural architecture with two subdomains that independently resemble membrane binding and/or disrupting proteins and could simultaneously act during disruption ( Figure 2 ) . Further , CelTOS employs a unique lipid specificity for phosphatidic acid to achieve nearly universal inner leaflet cellular activity . This new class is also supported by the fact that primary sequence similarity searches did not identify membrane disruption proteins as relatives of CelTOS . Both the N-terminal and C-terminal helices in CelTOS resemble the heptad repeats that form helices in class I viral fusion proteins ( Figure 2A–C ) ( Sackett and Shai , 2002; Harrison , 2015; LeDuc and Shin , 2000 ) . Viral heptad repeats undergo conformational changes that enhance the destabilization of host membranes post-insertion of the viral fusion peptide that enables lipid mixing during fusion ( 38 ) . It is plausible that the N- and C-terminal helices in CelTOS may also undergo similar conformational changes during membrane disruption . However , CelTOS employs a distinct membrane disruption mechanism from viral proteins as lipid mixing is not required for CelTOS to target inner leaflet lipids and disrupt plasma membranes . In addition , CelTOS is structurally similar to the pore-forming toxin M . tuberculosis ESAT-6 ( Figure 2D ) ( Hsu et al . , 2003; Los et al . , 2013; Smith et al . , 2008 ) . However , these toxins generally recognize a host cell target or receptor embedded within the target membrane , and ESAT-6 is secreted in complex with a chaperone that maintains it in an inactive state . Upon binding the membrane target , the toxin sheds the chaperone , oligomerizes and embeds into the membrane to form a pore on the angstrom length scale for the passage of ions and water ( Los et al . , 2013 ) . In contrast to a membrane-embedded target or receptor , CelTOS specifically binds phosphatidic acid within the inner leaflet of host or vector cell membranes . The dimeric nature of CelTOS may serve as a chaperone-like function to retain CelTOS in an inactive state prior to membrane binding as the dimer masks the inner hydrophobic faces of CelTOS . We provide evidence from dextran block experiments and transmission electron microscopy that CelTOS forms a pore of uniform size in liposomes . Malaria continues to be a global health problem , with half of the world’s population at risk . The lack of an effective vaccine against malaria , and the continuous evolution of drug-resistant parasites make malaria control difficult . Targeting the liver-stage ( Good and Doolan , 1999 ) and mosquito-stages ( mCGo , 2011 ) are of high priority for malaria vaccine development . Immunization of human volunteers with radiation-attenuated sporozoites induces sterile immunity ( Good and Doolan , 1999; Hoffman et al . , 2002 ) , and all volunteers showed strong immune responses to CelTOS ( referred to as Antigen 2 ) that were correlated with protection ( Doolan et al . , 2003 ) . In contrast , circumsporozoite protein , which is the major component of the most advanced malaria vaccine RTS , S ( Crompton et al . , 2010 ) , was only recognized by a third of the volunteers and mostly in those who were not protected ( Doolan et al . , 2003 ) . Individuals living in malaria endemic regions mount IFN-γ responses to CelTOS suggesting that CelTOS is a target of protective immunity ( Anum et al . , 2015 ) . CelTOS is also unique as it is a promising multi-stage malaria vaccine target ( Bergmann-Leitner et al . , 2010; Ferraro et al . , 2013; Bergmann-Leitner et al . , 2011 , 2013 ) for both infection-blocking and transmission-blocking vaccines currently in clinical trials . We show that CelTOS has conserved critical function in both Plasmodium falciparum and Plasmodium vivax that infect humans . The high-sequence similarity between Pf and PvCelTOS suggest that CelTOS may elicit cross-species protection . Therefore , CelTOS could be used alone or in combination with other antigens already in development to produce an effective multicomponent vaccine which targets both the mosquito and human stages and may be cross-species protective ( Doolan et al . , 2003; Crompton et al . , 2010; Crosnier et al . , 2011 ) . CelTOS-based vaccines have demonstrated the ability to elicit protection from malaria infection . However , protection is not absolute suggesting improvements in the immunogenicity , protectivity and formulation of CelTOS-based vaccines are necessary . Structure-function studies not only provide mechanistic insight into host-pathogen interactions ( Tolia et al . , 2005; Batchelor et al . , 2011; Vulliez-Le Normand et al . , 2012; Doud et al . , 2012; Malpede et al . , 2013; Batchelor et al . , 2014; Wright et al . , 2014; Lin et al . , 2012 ) , but also provide the necessary framework for improved vaccine design through structural vaccinology ( Dormitzer et al . , 2012; Chen et al . , 2013 , 2015 , 2016 ) . Immune recognition depends on antigen conformation and on surface accessibility of individual residues . The structure and architecture of CelTOS provides necessary insight to inform immune recognition , and enables protein engineering to retain the structural fold during immunogen design . The functional and mechanistic analyses presented here will enable assaying of immune responses for their ability to disrupt CelTOS function and provide an accurate molecular correlate for protection . Collectively , this work enables rational approaches to block CelTOS function , and rational approaches for vaccine development of global and emerging infectious diseases . CelTOS homologs in diverse apicomplexan parasites were identified using jackhammer and examination of the EuPathDB , and aligned in ClustalW . Codon-optimized PfCelTOS amino acids 25–182 , PvCelTOS amino acids 36–196 and EcIspF amino acids 1–159 with C-terminal hexahistidine tags , were expressed in E . coli . Tagged proteins were purified by Nickel-NTA chromatography and gel filtration using a Superdex 200 10/300 GL column ( GE Healthcare Life Sciences , Pittsburg , PA ) . PvCelTOS crystals were grown at 17°C by hanging-drop vapor diffusion after mixing 1 µl of protein at 20 mg/ml with 1 µl of reservoir containing 1 . 6 M Ammonium dihydrogen phosphate , 0 . 08 M Tris pH 8 . 5 and 20% glycerol and streak seeding PvCelTOS crystals previously obtained from the same crystallization condition . Native PvCelTOS crystals were harvested and soaked for ~1 min in cryo-protectant solution containing: 1 . 6 M Ammonium dihydrogen phosphate , 0 . 08 M Tris pH 8 . 5 and 30% glycerol . Crystals of the PvCelTOS Platinum- derivative were obtained by soaking native CelTOS crystals for 1–2 min in cryo-protectant solutions containing: 10 µM Potassium tetracyanoplatinate ( II ) hydrate , 1 . 6 M Ammonium dihydrogen phosphate , 0 . 08 M Tris pH 8 . 5 and 30% glycerol . Crystals of the CelTOS Mercury- derivative were obtained by soaking native CelTOS crystals for 1–2 min in cryo-protectant solutions containing: 1 µM mercury ethylphosphate , 1 . 6 M Ammonium dihydrogen phosphate , 0 . 08 M Tris pH 8 . 5 and 30% glycerol . The structure was phased in SHARP/autoSHARP ( Vonrhein et al . , 2007 ) as multiple isomorphous replacement with anomalous scattering . Phases were extended by solvent flattening in SHARP/autoSHARP . The structure was refined in PHENIX ( Adams et al . , 2002 ) with NCS restraints . Map and model manipulation was aided by the CCP4 program suite ( Collaborative Computational Project No . 4 , 1994 ) . MolProbity ( Davis et al . , 2007 ) places this structure in the top 100th percentile of structures with comparable resolution . 98 . 44% and 1 . 56% of residues lie in the favored and allowed regions of the Ramachandran plot , respectively . Structurally similar PDBs were identified using the Dali server ( Holm et al . , 2010 ) , and alignments and the normalized rmsd values obtained using lsqman ( Kleywegt , 1996 ) . Buried surface area of the dimer was determined using PDBePISA ( Krissinel and Henrick , 2007 ) . Atomic coordinates and structure factors for the reported crystal structure were deposited into the Protein Data Bank under PDB code 5TSZ ( http://www . pdb . org ) . Sedimentation equilibrium experiments were conducted with a Beckman/Coulter XL-A analytical ultracentrifuge ( Beckman/Coulter , Indianapolis , IN ) using an An60Ti rotor at 10°C and λ 286 nm . PfCelTOS and PvCelTOS were purified by gel filtration using a Superdex 200 10/300 GL column ( GE Healthcare Life Sciences ) in 150 mM NaCl and 10 mM HEPES pH 7 . 4 . Data were collected for speeds 12 , 000 rpm and 15 , 000 rpm with PfCelTOS at 25 µM , 36 µM and 46 µM , and PvCelTOS at 29 µM , 41 µM and 53 µM . A partial specific volume of 0 . 730919 and 0 . 737102 was calculated using Sednterp for PfCelTOS and PvCelTOS , respectively . Data were analyzed using a single component model with UltraScan II version 9 . 9 . Average molecular weight is reported as mean ± s . d . Membrane lipid strips ( Echelon Biosciences Inc . , Salt Lake City , UT ) were probed with proteins according to the manufacturer specified protocol . Membrane strips were blocked at room temperature for 1 hr in blocking buffer ( 10 mM Tris pH 8 . 0 , 150 mM NaCl , 0 . 1% Tween 20% and 3% BSA ) and washed three times for 5 min with wash buffer ( 10 mM Tris pH 8 . 0 , 150 mM NaCl and 0 . 1% Tween 20 ) . PfCelTOS and PvCelTOS were incubated in blocking buffer at a concentration of 20 µM . Mouse anti-His monoclonal antibody ( Invitrogen , Waltham , MA ) diluted 1:500 in blocking buffer was used as the primary antibody , and Peroxidase AffiniPure Goat Anti-Mouse IgG ( Jackson Laboratories ) diluted 1:10 , 000 in blocking buffer as the secondary antibody . Binding was detected with ECL Prime Western Blotting Detection Kit ( GE Healthcare Biosciences ) , imaged using Image Reader FLA-5000 series phosphorimager ( Fuji PhotoFilm , Japan ) and quantified using Image Gauge version 4 . 23 ( Fuji PhotoFilm ) . Experiments were performed with eight replicates , along with one negative control where CelTOS was omitted . Spot intensities were normalized to background and the negative control strip . Data were analyzed by one-way ANOVA with Dunnett’s multiple comparison tests to compare the intensity of each lipid spot to the negative control on that strip using Prism 6 ( GraphPad Software , La Jolla , CA ) . The normalized binding intensity of each spot was plotted as mean ± s . e . m . The membrane lipid screen is described in more detail at Bio-protocol ( Jimah et al . , 2017a ) . Liposomes used for experiments were composed of 1-palmitoyl-2-oleoyl-sn-glycero-3-phosphocholine ( POPC ) alone or in combination with either 1-palmitoyl-2-oleoyl-sn-glycero-3-phosphate ( POPA ) or 1-palmitoyl-2-oleoyl-sn-glycero-3-phospho-L-serine ( POPS ) ( Avanti Polar Lipids , Alabaster , AL ) . Chloroform-dissolved lipids were dried under N2 gas followed by vacuum for 3 hr . To hydrate the lipids , equal volumes of ether and a solution composed of 10 mM HEPES pH 7 . 4 , 150 mM KCl and 20 mM carboxyfluorescein were added to the lipids , the mixture sonicated and ether evaporated using a roto-vap . Liposomes were sized by extrusion using 200 nm polycarbonate Track-Etched filters ( Whatman/Nucleopore ) . Size exclusion chromatography with Sephadex G 25–300 ( Sigma Aldrich , Saint Louis , MO ) was used to separate unincorporated carboxyfluorescein from liposomes . In all experiments , varying concentrations of proteins were incubated with liposomes containing 250 nM of lipids , and the time-dependence of liposome disruption was monitored using a Cary Eclipse Fluorescence Spectrophotometer ( Varian Inc/ Agilent , Santa Clara CA ) by observing the carboxyfluorescein fluorescence emission at 512 nm upon excitation at 492 . Three replicate measurements were performed for each protein . The percent liposome disruption was calculated as: ( 1 ) %Disruptiontime=[ ( F512ofliposome+protein−F512ofliposome ) / ( F512ofliposome+triton−F512ofliposome ) ]∗100 where 0 . 2% Triton X-100 was added to normalize each sample to complete dequenching as previously described , and data fitted to Equation 1 ( Saito et al . , 2000 ) . For kinetic and pore-forming analyses , the time dependence of liposome disruption was also fitted to the following Equation 2 ( Saito et al . , 2000 ) : ( 2 ) %LiposomesDisrupted=A ( 1−e− ( time/tau ) ) +m∗time where A is the maximum percentage of liposomes disrupted , Tau is the time constant for the exponential component , and m is the slope of the linear component . For the liposome disruption assays with dextran , 20 µM dextran was added to liposomes , and then 250 nM of PfCelTOS or PvCelTOS was added , and liposome disruption monitored . The maximum amount of liposome disrupted , A , was obtained by fitting the data to Equation 2 . Data was analyzed using Prism 6 ( GraphPad Software , La Jolla , CA ) by performing a one-way ANOVA with Dunnett’s multiple comparison test . Data with seven experimental replicates of each preparation included as mean ± s . e . m . The membrane lipid screen is described in more detail at Bio-protocol ( Jimah et al . , 2017b ) . Mature female Xenopus laevis frogs were purchased from Xenopus Express ( Brooksville , FL ) . All animal protocols followed guidelines approved by the Washington University School of Medicine and the National Institutes of Health . Frogs were anesthetized with a 0 . 1% tricaine solution buffered with 0 . 1% NaHCO3 prior to survival surgery in which a portion of the ovary is removed . Stage V-VI oocytes were isolated and maintained at 18 in modified Barth’s solution of the following composition: 88 mM NaCl , 2 . 4 NaHCO3 , 1 mM KCl , 0 . 3 mM Ca ( NO3 ) 2 , 0 . 4 mM CaCl2 , 0 . 8 mM MgSO4 and 10 mM HEPES/Tris pH 7 . 4 , and supplemented with 50 mg/l gentamicin , 6 mg/l ciprofloxacin , and 100 mg/l streptomycin sulfate/100 , 000 units/l penicillin G sodium ( Life Technologies , Carlsbad , CA ) . One day after isolation , oocytes were microinjected with 50 nl ( 0 . 1 nmol , approximately 100 µM final concentration ) of EcIspF , PfCelTOS or PvCelTOS , prepared fresh as 2 mM stock solutions in Protein buffer composed of 10 mM HEPES pH 7 . 4 and 150 mM KCl; non-injected oocytes , oocytes injected with 50 nl of protein buffer , and oocytes injected with EcIspF , served as controls . Twenty-four oocytes per condition were individualized in 96-well plates , and monitored for membrane damage . Oocytes were monitored 24 hr post-infection under an Olympus SZX10 microscope ( Olympus Corporation , Waltham , MA ) , and images captured with Infinity Lite camera using Infinity capture version 6 . 00 software ( Lumnera Corporation ) . Results are shown as mean ± s . e . m . of six separate experiments with independent protein preparations and with oocytes from different donor frogs . Data were analyzed using Prism 6 ( GraphPad Software , La Jolla , CA ) by performing Kruskal-Wallis followed by Dunn’s multiple comparison tests . RAW 264 . 7 cells ( ATCC ) liberated using CellStripper reagent ( Corning , Corning , NY ) were filtered through a 40 µm mesh to yield a single-cell suspension . Cells were seeded onto clear-bottomed , white-walled 96-well plates ( Corning 3610 ) at 2×105 cells per well in DMEM 10% FBS and allowed to adhere overnight . Wells were washed with PBS and assay reagents added . The assay was conducted in 5% FBS with phenol-red-free DMEM . Recombinant proteins were added to 100 µM final concentration . Triton X-100 at a final concentration of 0 . 05% v/v served as a positive control . Total reaction volume was 100 µl . The plates were incubated at 37°C , 5% CO2 for 3 . 5 hr prior to the addition of 50 µl 1x CellTox Green reagent ( Promega , Madison , WI ) and incubated for a further 30 min . Fluorescence at 485 nm was read using a Cytation 3 Cell Imaging Multimode Reader ( Biotek , Winooski , VT ) . Results are shown as the mean ± s . e . m . of three separate experiments with independent RAW 264 . 7 cells each with three technical replicates . Data were analyzed using Prism 6 ( GraphPad Software , La Jolla , CA ) by performing Kruskal-Wallis and Dunn’s multiple comparison tests . Liposomes were treated with either 250 µM PfCelTOS or PvCelTOS . Suspension of liposomes were then allowed to absorb onto freshly glow discharged formvar/carbon-coated copper grids for 10 min . Grids were washed in dH2O and stained with 1% aqueous uranyl acetate ( Ted Pella Inc . , Redding , CA ) for 1 min . Excess liquid was gently wicked off and grids were allowed to air dry . Samples were viewed on a JEOL 1200EX transmission electron microscope ( JEOL USA , Peabody , MA ) equipped with an AMT megapixel digital camera ( Advanced Microscopy Techniques , Woburn , MA ) . Pore diameters were determined using ImageJ 1 . 48V .	Half of the world’s population are at risk of contracting malaria: a disease caused by parasites that are spread by mosquito bites . Yet antimalarial drugs are becoming less and less effective because many of the parasites have grown to be resistant to them . Furthermore , and although some vaccines are already being trialed , there is not an effective vaccine that has been approved to control the disease . As such , many researchers hope that understanding more about the critical stages in the life cycle of the parasites will unveil new targets for antimalarial drugs and vaccines . To complete their life cycle , malaria parasites must move through various cells in the human and mosquito . Parasites that lack a protein called CelTOS can enter these cells , but remain stuck inside . It was not known why this was the case , in part because the sequence of building blocks called amino acids that make up CelTOS is unlike that of other proteins which do have a known activity or function . Jimah et al . sought to understand how CelTOS allows malaria parasites to pass through cells by solving its three-dimensional structure . This is because a protein’s function often depends more on the protein’s structure than the specific sequence of its amino acids; such that two proteins that have a similar shape are very likely to work in similar ways . Jimah et al . solved the structure of CelTOS using a technique called X-ray crystallography and found that it resembles proteins known to bind and disrupt cell membranes . Further experiments showed that CelTOS from malaria parasites specifically binds to a fatty molecule that is predominantly found in the inner face of cell surface membranes . It was also confirmed that CelTOS forms pores that disrupt cell membranes . Together , these findings suggest that CelTOS breaches the cell membranes from the inside of infected human and mosquito cells to enable the parasites to exit . While a vaccine based on the CelTOS protein is already being tested in clinical trials , the findings of Jimah et al . should enable this vaccine to be refined if it is not effective , or redesigned based on knowledge of its structure , function and mechanism . This new knowledge could also provide clues as to which kinds of molecules might neutralize the CelTOS protein’s activity and therefore might lead to new antimalarial drugs .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "structural", "biology", "and", "molecular", "biophysics", "microbiology", "and", "infectious", "disease" ]	2016	Malaria parasite CelTOS targets the inner leaflet of cell membranes for pore-dependent disruption
Pharmacological inhibition of chromatin co-regulatory factors represents a clinically validated strategy to modulate oncogenic signaling through selective attenuation of gene expression . Here , we demonstrate that CBP/EP300 bromodomain inhibition preferentially abrogates the viability of multiple myeloma cell lines . Selective targeting of multiple myeloma cell lines through CBP/EP300 bromodomain inhibition is the result of direct transcriptional suppression of the lymphocyte-specific transcription factor IRF4 , which is essential for the viability of myeloma cells , and the concomitant repression of the IRF4 target gene c-MYC . Ectopic expression of either IRF4 or MYC antagonizes the phenotypic and transcriptional effects of CBP/EP300 bromodomain inhibition , highlighting the IRF4/MYC axis as a key component of its mechanism of action . These findings suggest that CBP/EP300 bromodomain inhibition represents a viable therapeutic strategy for targeting multiple myeloma and other lymphoid malignancies dependent on the IRF4 network . Multiple myeloma is an aggressive and incurable hematologic malignancy characterized by the proliferation of abnormal plasma cells ( Mahindra et al . , 2010 ) . Myeloma is driven by transcriptional reprogramming events that prevent the differentiation of activated B cells to plasma cells and subsequently promote the proliferation of dysfunctional plasma cells ( Mahindra et al . , 2010 ) . Abnormal activity of a number of transcription factors has been implicated in multiple myeloma development , including NF-κB , MAF , MYC , and interferon regulatory factor 4 ( IRF4 ) ( Dean et al . , 1983; Keats et al . , 2007; Palumbo et al . , 1989; Shaffer et al . , 2008 ) . The oncogenic activity of these transcription factors in multiple myeloma is demonstrated by the presence of translocation events that fuse them to highly active enhancers that drive high expression ( Dean et al . , 1983; Iida et al . , 1997 ) . The IRF4 transcription factor is a critical component of the normal adaptive immune response and is required for lymphocyte activation and differentiation of immunoglobulin-secreting plasma cells ( Klein et al . , 2006; Mittrücker et al . , 1997; Sciammas et al . , 2006 ) . Downstream targets of IRF4 include factors that regulate cell cycle progression , survival , and normal plasma cell function ( Shaffer et al . , 2008 ) . While oncogenic translocations of IRF4 have been found , more frequently , myeloma and other lymphoid malignancies are dependent on dysfunctional transcriptional networks downstream of a genetically normal IRF4 locus ( Shaffer et al . , 2008 ) . While the immunomodulatory agent lenalidomide has been shown to promote IRF4 protein degradation ( Moros et al . , 2014 ) , pharmacological agents that regulate the expression of IRF4 mRNA have not been identified . Small molecule inhibition of bromodomain-containing transcriptional co-regulators have recently been shown to be a viable strategy for the suppression of otherwise un-druggable downstream transcription factors . This is best exemplified by the inhibitors of BET family bromodomains , which down-regulate MYC and BCL2 and are thus highly active in malignancies driven by these critical oncogenes ( Dawson et al . , 2011; Delmore et al . , 2011; Mertz et al . , 2011; Zuber et al . , 2011 ) . Cyclic AMP response element binding protein ( CREB ) -binding protein ( CBP ) and E1A interacting protein of 300 kDa ( EP300 ) are highly homologous bromodomain-containing transcriptional co-activators that regulate a number of important cellular events through their acetyltransferase activity ( Goodman and Smolik , 2000 ) . Genetic studies in mice and surveys of human cancer mutations and translocations have implicated CBP/EP300 in cancer , but the role of the bromodomain in the normal and pathological function of CBP/EP300 has not been extensively studied ( Kung et al . , 2000; Murati et al . , 2007; Ohnishi et al . , 2008; Pasqualucci et al . , 2011; Peifer et al . , 2012 ) . Given the importance of these genes in cancer development , CBP/EP300 bromodomain inhibition may represent an important therapeutic strategy to reprogram oncogenic signaling pathways in human malignancies . To assess the functional role of CBP/EP300 bromodomains , we made use of two chemical probes recently generated by the Structural Genomics Consortium ( Figure 1A ) ( SGC; www . thesgc . org ) ( Hay et al . , 2014 ) . SGC-CBP30 and I-CBP112 are chemically distinct tool compounds with selective affinity for the bromodomains of CBP/EP300 over other bromodomains in this protein family . Independent of CBP/EP300 , the bromodomains with the highest affinity for these molecules is the BET bromodomain family ( Hay et al . , 2014 ) . We confirmed the biochemical potency and selectivity of SGC-CBP30 and I-CBP112 using AlphaLISA with the isolated bromodomain of CBP and the first bromodomain of BRD4 ( BRD4-BD1 ) ( Figure 1B , F ) . We further addressed the selectivity of the compounds through the use of Differential Scanning Fluorimetry ( DSF ) with a panel of 19 purified bromodomains ( Figure 1—source data 1 ) . Taken together , these data are consistent with published reports regarding the selectivity of these compounds ( Hammitzsch et al . , 2015; Picaud et al . , 2015 ) . 10 . 7554/eLife . 10483 . 003Figure 1 . Characterization of CBP/EP300 bromodomain inhibitors . ( A ) Structures of SGC-CBP30 and I-CBP112 . ( B ) Representative AlphaLISA curves showing the inhibition of acetylated peptide binding to isolated CBP or BRD4 bromodomains in the presence of SGC-CBP30 and I-CBP112 . Error bars represent SEM of 3 technical replicates . ( C ) Dose-titrations of SGC-CBP30 , I-CBP112 , and CPI203 using NanoBRET with the isolated CBP bromodomain and histone H3 . 3 in 293 cells . Error bars represent SEM of three technical replicates . The calculated EC50 values are shown in F . ( D ) ZsGreen-bromodomain fusion proteins were monitored by high content imaging . Representative nuclei showing nuclear foci in the indicated assays in the presence of DMSO , SGC-CBP30 ( 5 μM ) , I-CBP112 ( 5 μM ) or CPI203 ( 0 . 33 μM ) . ( E ) Quantification of chromatin release assay . Each curve represents a titration of the indicated compound in stable cell lines expressing the indicated fusion protein ( CBP: CBP-bromodomain/BRD9; BRD4: full length BRD4 ) . Values are mean of four fields per well of two technical replicates , ± SEM . ( F ) Summary of biochemical and cellular activity of the indicated compounds . Values represent half-maximal inhibition ( IC50 ) in AlphaLISA assays ( n ≥ 2 independent replicates ) or half-maximal induction ( EC50 ) in NanoBRET ( n = 3 technical replicates ± SEM ) or chromatin release assays ( n = 2 biological replicates ± SEM ) . ND = not determined due to a failure to produce 100% inhibition compared to controls . DOI: http://dx . doi . org/10 . 7554/eLife . 10483 . 00310 . 7554/eLife . 10483 . 004Figure 1—source data 1 . Bromodomain selectivity of CBP/EP300 bromodomain inhibitors . Differential scanning fluorimetry was carried out with the indicated isolated bromodomains at 4–8 μM and the compounds at 20 μM . Shifts in melting temperature ( △Tm , °C ) and SEM for n = 3 technical replicates are shown . DOI: http://dx . doi . org/10 . 7554/eLife . 10483 . 004 To assess the potency of these probes in cells , we utilized a proximity-based assay ( NanoBRET ) , which monitors the interaction between the bromodomain of CBP and histone H3 . 3 . SGC-CBP30 and I-CBP112 showed similar dose-dependent inhibition of CBP-H3 . 3 binding , with calculated EC50 values of 0 . 28 μM and 0 . 24 μM , respectively ( Figure 1C and F ) . The BET bromodomain inhibitor CPI203 ( Devaiah et al . , 2012 ) did not display dose-dependent inhibition in this assay ( Figure 1C ) . Next , we made use of an imaging-based assay that measures the release of bromodomain-GFP fusion proteins from chromatin upon ligand binding ( Huang et al . , 2014 ) . As shown in Figure 1D , chromatin release results in aggregation of fusion proteins into finite speckles whose number and intensity increase with ligand binding . Both SGC-CBP30 and I-CBP112 promote chromatin release of CBP bromodomain fusion proteins at low micromolar concentrations as quantitated by high-content imaging ( 10-fold cell shift ) , comparable to previous results ( Figure 1E , F ) ( Hay et al . , 2014 ) . In contrast , both probe compounds release BRD4-BD1 fusion proteins from chromatin at significantly higher concentrations as compared to the selective BET inhibitor CPI203 ( Figure 1D–F ) ( Devaiah et al . , 2012 ) . Given the cellular selectivity of the compounds , we are confident that at defined concentrations of the inhibitors ( ≤2 . 5 μM SGC-CBP30 or ≤5 μM I-CBP112 ) , any observed pharmacological effects are due to on-target inhibition of CBP/EP300 bromodomains . To assess the biological activity of CBP/EP300 bromodomain inhibition , we treated a panel of cell lines of multiple myeloma and acute leukemia origin with SGC-CBP30 and I-CBP112 . As shown in Figure 2A , B , and Figure 2—figure supplement 1A , a subset of cell lines was highly sensitive to both compounds , with the most sensitive cell lines having GI50 values below 3 μM of SGC-CBP30 . Notably , 14 of the 15 most sensitive cell lines are of multiple myeloma origin ( Figure 2A ) . As effectors of multiple biological processes , CBP and EP300 play important roles in multiple phases of the cell cycle . To assess the requirement of the CBP/EP300 bromodomains in cell cycle progression , we released G0/G1 arrested LP-1 cells in the presence of either DMSO , SGC-CBP30 , or I-CBP112 . As shown in Figures 2C and Figure 2—figure supplement 1B , the progression of the cells appears normal through G2/M phase ( 8 hr ) . Only upon entry into the next cell cycle is there a noticeable alteration in cell cycle progression , with compound-treated cells accumulating in G1 at 16 and 24 hr as compared to DMSO-treated cells . Thus , it appears that the primary phenotypic effect of CBP/EP300 bromodomain inhibition is arrest in the G1 phase of the cell cycle . Consistent with these observations , growth inhibition resulting from CBP/EP300 bromodomain inhibition is accompanied by G0/G1 arrest and apoptosis in phenotypically sensitive cell lines ( Figures 2D and Figure 2—figure supplement 1C ) . As the phenotypic effects of SGC-CBP30 and I-CBP112 appeared similar , we utilized the more potent compound , SGC-CBP30 , for further experiments and made use of I-CBP112 as a distinct chemotype to confirm important observations . 10 . 7554/eLife . 10483 . 005Figure 2 . Phenotypic effects of CBP/EP300 bromodomain inhibition . ( A ) Growth inhibitory effects of SGC-CBP30 and I-CBP112 in the indicated cell lines . Cells were incubated with compounds for 6 days , and viability was measured with resazurin . Values are the mean of at least two biological replicates . Values with error can be found in Figure 2—source data 1 . ( B ) Example viability curves for LP-1 . Values represent the mean of three3 technical replicates , ± SD . ( C ) LP-1 were synchronized by double thymidine block and released into either DMSO or 2 . 5 μM SGC-CBP30 . Cells were fixed and stained with PI for cell cycle analysis at the indicated time points . Cell cycle distribution at 24 hr is shown in the table . Representative data from one of two biological replicates are shown . ( D ) LP-1 cells were treated as in ( A ) and fixed after 6 days . Viable cell number and percent increase in G0/G1 or sub-G1 over DMSO were determined by flow cytometry . Each point is the mean of three technical replicates , ± SD . See Figure 2—figure supplement 1 for additional data with I-CBP112 . DOI: http://dx . doi . org/10 . 7554/eLife . 10483 . 00510 . 7554/eLife . 10483 . 006Figure 2—source data 1 . GI50 and standard deviation for a minimum of two replicates for the data shown in Figure 2 and Figure 2—figure supplement 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 10483 . 00610 . 7554/eLife . 10483 . 007Figure 2—figure supplement 1 . CBP/EP300 bromodomain inhibition affects the viability of multiple myeloma cells . As in Figure 2 , except with I-CBP112 . ( A ) as in Figure 2A . ( B ) as in Figure 2C . ( C ) as in Figure 2D . DOI: http://dx . doi . org/10 . 7554/eLife . 10483 . 007 Recent work by many groups has demonstrated that small molecule inhibitors of BET family bromodomains are highly active in cell lines of hematopoetic origin ( Dawson et al . , 2011; Delmore et al . , 2011; Mertz et al . , 2011; Zuber et al . , 2011 ) . In contrast , our results suggest that CBP/EP300 bromodomain inhibition preferentially targets a more limited subset of hematologic cell lines , with a bias toward multiple myeloma/plasmacytoma cell lines . To gain insight into the mechanisms underlying these phenotypic differences , we carried out RNA sequencing of LP-1 cells treated with SGC-CBP30 or the pan-BET inhibitor CPI203 . To narrow our focus to direct transcriptional effects , we examined gene expression changes following short term ( 6 hr ) compound treatment . As shown in Figures 3A and Figure 3—figure supplement 1A , the transcriptional footprint of SGC-CBP30 is more circumscribed than that of CPI203 , with far fewer genes differentially expressed . Notably however , the genes differentially expressed by SGC-CBP30 are not simply a subset of those affected by CPI203 ( Figures 3A and Figure 3—figure supplement 1A; confirmed with I-CBP112 in Figure 3—figure supplement 1D ) . This suggests that the two modalities may target distinct transcriptional pathways . 10 . 7554/eLife . 10483 . 008Figure 3 . CBP/EP300 bromodomain inhibition targets IRF4 . ( A ) LP-1 cells were treated with SGC-CBP30 ( 2 . 5 μM ) or CPI203 ( 0 . 25 μM ) for 6 hr , and mRNA expression was measured using RNA sequencing . Expression values for biological replicate compound-treated samples were normalized to paired DMSO controls to obtain log2 fold change values . ( B ) Example enrichment plots for GSEA of SGC-CBP30 treated LP-1 cells . ( C ) Left , Scatter plot of P value vs . NES for multiple myeloma and IRF4 gene signatures for SGC-CBP30 ( red ) or CPI203 ( black ) treated LP-1 cells . Dashed line indicates p = 0 . 05 . Right , fraction of gene signatures significantly enriched ( p<0 . 05 ) with each treatment . Error bars indicate SEM . SGC-CBP30: 26/58; CPI203: 9/58 . *** indicates p = 0 . 0005 by unpaired 2-tailed t-test . ( D ) IRF4 target genes differentially expressed ( minimum 1 . 5 fold , p<0 . 05 ) with SGC-CBP30 , but not CPI203 . See Figure 3—figure supplement 1 for additional gene expression data and analysis . DOI: http://dx . doi . org/10 . 7554/eLife . 10483 . 00810 . 7554/eLife . 10483 . 009Figure 3—figure supplement 1 . CBP/EP300 bromodomain inhibition targets IRF4 transcriptional programs . ( A ) Venn diagrams showing the overlap of genes down- or up-regulated at least 2-fold following treatment with SGC-CBP30 or CPI203 as in Figure 3A . ( B ) Significantly enriched ( p<0 . 05 ) IRF4 gene signatures upon SGC-CBP30 treatment . ( C ) The fraction of the 309 IRF4 target genes present in the overall set of mapped genes ( 20299 genes ) or in the genes differentially expressed at least 1 . 5 fold by SGC-CBP30 ( 393 genes ) or CPI203 ( 2959 genes ) was determined . p-values were calculated by unpaired 2-tailed t-test . ( D ) Expression of the indicated mRNAs was determined by q-RTPCR following treatment of LP-1 cells with SGC-CBP30 ( 2 . 5 μM ) , I-CBP112 ( 5 μM ) , or CPI203 ( 0 . 25 μM ) for 6 hr . Relative gene expression is expressed as log2 fold change relative to expression in DMSO . ( E ) Left , As in Figure 3C , except with MYC gene signatures . Right , fraction of gene signatures significantly enriched with each treatment . Error bars indicate SEM . SGC-CBP30: 15/51; CPI203: 26/51 . * indicates p = 0 . 03 by unpaired 2-tailed t-test . DOI: http://dx . doi . org/10 . 7554/eLife . 10483 . 009 To better understand the pathways impacted by CBP/EP300 and BET bromodomain inhibition , we carried out Gene Set Enrichment Analysis ( GSEA ) ( Subramanian et al . , 2005 ) with an emphasis on transcriptional pathways that might distinguish the two modalities . As expected , gene signatures negatively correlated with CPI203 treatment were dominated by MYC-dependent transcriptional pathways ( Figure 3—figure supplement 1E ) . However , while several MYC signatures were also enriched upon treatment with SGC-CBP30 , more notable was the enrichment of signatures for pathways important in multiple myeloma ( Figures 3B and C ) , which was distinct from the effects of BET inhibition . We noted in particular the significant negative correlation ( p-value < 0 . 05 ) of 4 gene signatures containing downstream targets of IRF4 , a lymphocyte-specific transcription factor that is essential for the survival of multiple myeloma cells ( Figure 3—figure supplement 1B ) ( Shaffer et al . , 2008 ) . Consistent with this gene set enrichment , IRF4 target genes ( catalogued by Shaffer et al . ) are significantly enriched in the set of genes differentially expressed following treatment with SGC-CBP30 ( Figure 3—figure supplement 1C ) . A subset of these IRF4 target genes ( including IRF4 itself ) is significantly differentially expressed following treatment with SGC-CBP30 but not CPI203 ( Figure 3D ) , arguing that the IRF4 transcriptional axis may be selectively targeted by CBP/EP300 bromodomain inhibition . Since the regulation of the IRF4 transcriptional axis through small molecule inhibition of CBP/EP300 bromodomains would represent a promising new therapeutic strategy for multiple myeloma , we sought to better understand our initial observations . We first demonstrated by qRT-PCR that IRF4 mRNA was suppressed in a dose-dependent manner by CBP/EP300 bromodomain inhibition in both LP-1 and another multiple myeloma cell line , OPM2 ( Figure 4A and Figure 4—figure supplement 1A ) . The EC50 of IRF4 suppression in each cell line is in the range of the cellular EC50 values shown in Figure 1 and the GI50 values shown in Figure 2A , arguing for an on-target effect . Consistent with suppression at the mRNA level , IRF4 protein is reduced upon treatment with SGC-CBP30 or I-CBP112 ( Figure 4—figure supplement 1D ) . In support of a direct effect on the transcription of IRF4 , we observed that IRF4 is suppressed within 2 hr of addition of SGC-CBP30 ( Figure 4—figure supplement 1C ) , and recovers within 1 hr of removal of the compound ( Figure 4B ) . 10 . 7554/eLife . 10483 . 010Figure 4 . IRF4 is a direct transcriptional target of CBP/EP300 bromodomain inhibition . ( A ) Dose-dependent inhibition of IRF4 mRNA expression ( qRT-PCR ) with SGC-CBP30 in LP-1 and OPM2 cells following 6 hr of treatment . Values represent the mean of three biological replicates , ± SEM . ( B ) LP-1 cells were treated with SGC-CBP30 ( 2 . 5 μM ) for 4 hr , compound was removed , and cells were incubated for an additional 1 hr in fresh media . Levels of IRF4 mRNA were measured by qRT-PCR and normalized to GAPDH . Relative mRNA values normalized to DMSO at each time point represent the mean of 2 biological replicates , ± SEM . ( C ) Cells were transduced with lentivirus and lysed for Western analysis with the indicated antibodies ( 3 days post-infection ) . ( D ) IRF4 expression was determined by qRT-PCR at 3 . 5 days following the transduction of shRNA lentivirus , and mRNA was normalized to GAPDH and expressed relative to the control shLuc ( n = 3 technical replicates , ± SEM ) . ( E ) Western analysis with the indicated antibodies was carried out at 3 . 5 days post-transduction with the indicated shRNA constructs . ( F ) Cells were fixed at the indicated time points following transduction and viability was determined by flow cytometry . Percent growth is expressed relative to control shLuc at each time point . Values represent the mean of n = 3 technical replicates , ± SEM . ( G ) LP-1 cells were treated with SGC-CBP30 ( 2 . 5 μM ) for 6 hr , and the indicated antibodies were used for ChIP-seq . Sequencing traces for the IRF4 super-enhancer and the transcriptional start site are shown . See Figure 4—figure supplements 1 and 2 for additional supporting data . DOI: http://dx . doi . org/10 . 7554/eLife . 10483 . 01010 . 7554/eLife . 10483 . 011Figure 4—figure supplement 1 . CBP/EP300 bromodomain inhibition regulates the expression of IRF4 . ( A ) Cells were treated with I-CBP112 for 6 hr , and levels of IRF4 were determined as in Figure 4A . Values represent the mean of n = 3 biological replicates ± SEM . ( B ) LP-1 cells were treated with a titration of CPI203 for 6 hr , and IRF4 expression was determined by qRT-PCR and normalized to GAPDH . Values represent the mean of n = 2 biological replicates , ± SEM . ( C ) LP-1 cells were treated with DMSO , SGC-CBP30 ( 2 . 5 μM ) , or I-CBP112 ( 5 μM ) . Total RNA was prepared at the indicated time points and used for qRT-PCR . Expression of IRF4 was normalized to GAPDH and calculated relative to DMSO treated cells at each time point . Values represent the mean of n = 4 technical replicates , ± SEM . ( D ) Uninduced LP-1/IRF4 cells were treated with SGC-CBP30 ( 2 . 5 μM ) or I-CBP112 ( 5 μM ) for 24 hr , and lysates were prepared for Western analysis with the indicated antibodies . ( E ) LP-1 cells were treated with the indicated concentrations of SGC-CBP30 for 16 hr , and extracts were prepared for Western analysis with the indicated antibodies . DOI: http://dx . doi . org/10 . 7554/eLife . 10483 . 01110 . 7554/eLife . 10483 . 012Figure 4—figure supplement 2 . CBP/EP300 bromodomain inhibition does not cause global eviction of BRD4 from chromatin . ( A ) BRD4 ChIP-seq peaks were called using MACS and ranked by log2 fold change in BRD4 enrichment in LP-1 cells treated for 6 hr with 0 . 25 µM CPI203 compared to DMSO-treated cells . ( B ) EP300 ChIP-seq peaks were called using MACS and ranked by log2 fold change in EP300 enrichment in LP-1 cells treated for 6 hr with 2 . 5 µM SGC-CBP30 compared to DMSO-treated cells . ( C ) Examples of BRD4 and EP300 ChIP-seq tracks showing that CBPi does not cause global eviction of BRD4 , and that BETi does not globally reduce EP300 chromatin binding . Representative tracks of two biological replicates are shown . DOI: http://dx . doi . org/10 . 7554/eLife . 10483 . 012 To further corroborate that IRF4 suppression is due to the on-target activity of CBP/EP300 , we used RNAi to knock down either CBP or EP300 in the LP-1 cell line . As shown in Figures 4C , D and E , three unique shRNA constructs that efficiently knocked down either CBP or EP300 reduced the expression of IRF4 at the mRNA and protein level . Viability effects were observed subsequent to suppression of IRF4 ( Figure 4F ) , which is consistent with the kinetics and phenotypic effects of CBP/EP300 bromodomain inhibition . Taken together , these data argue that the suppression of IRF4 is due to on target inhibition of the CBP/EP300 bromodomains . CBP and EP300 function as transcriptional co-activators via acetylation of histones and transcription factors . The bromodomains of CBP/EP300 are required for the acetylation of histones within a chromatin context , and histone H3 lysine 18 ( H3K18 ) and histone H3 lysine 27 ( H3K27 ) have been shown to be specifically targeted by CBP/EP300 ( Jin et al . , 2011 ) . To investigate the mechanism of transcriptional suppression of IRF4 , we first examined whether CBP/EP300 bromodomain inhibition causes global reduction in histone acetylation . Following incubation of LP-1 cells with SGC-CBP30 , we did not observe any significant changes in the global levels of H3K18 or H3K27 acetylation by Western analysis ( Figure 4—figure supplement 1E ) . We looked more closely for localized changes in histone acetylation by using chromatin immunoprecipitation followed by massively parallel sequencing ( ChIP-seq ) . As shown in Figure 4G , we observed a significant reduction in both H3K18ac and H3K27ac at a previously annotated super-enhancer of IRF4 ( Chapuy et al . , 2013 ) as well as at the transcription start site . Notably , this reduction in acetylation is accompanied by a reduction in the chromatin occupancy of EP300 , suggesting that CBP/EP300 bromodomain inhibition promotes release of the protein from chromatin leading to a reduction in histone acetylation . It should be noted that broad and complete loss of EP300 was not observed , perhaps suggesting that the bromodomain of EP300 serves to localize it to restricted domains ( Figure 4G ) . Importantly , treatment with SGC-CBP30 did not result in global eviction of BRD4 , arguing against a direct effect on BET bromodomain proteins ( Figure 4—figure supplement 2 ) . We have shown that CBP/EP300 bromodomain inhibition leads to viability defects in multiple myeloma cell lines and to the suppression of IRF4 and its downstream transcriptional programs in the representative cell line LP-1 . To understand whether the suppression of IRF4 was more broadly involved in the phenotypic response to CBP/EP300 bromodomain inhibition , we profiled the transcriptional response of a panel of cell lines of varying sensitivity to SGC-CBP30 ( Figure 2A ) following a 6-hr treatment with the inhibitor . As shown in Figure 5A , the degree of suppression of IRF4 mRNA is significantly correlated with phenotypic sensitivity to SGC-CBP30 , suggesting that this pharmacodynamic response is important for the mechanism of growth inhibition . 10 . 7554/eLife . 10483 . 013Figure 5 . IRF4 suppression is correlated with phenotypic sensitivity to SGC-CBP30 , and MYC is downregulated concomitant with IRF4 suppression following CBP/EP300 knockdown or bromodomain inhibition . ( A ) The indicated cell lines were treated with SGC-CBP30 ( 2 . 5 μM ) for 6 hr , and IRF4 expression normalized to GAPDH was determined by q-RTPCR . Suppression of IRF4 ( log2 fold change relative to DMSO ) was plotted against log2 GI50 . R2 and p-value of the linear regression are shown . Cell lines indicated in red have a GI50 of less than 2 . 5 μM SGC-CBP30 ( Figure 2A ) . Source data can be found in Figure 5—source data 1 . ( B ) Lentiviral shRNA constructs were transduced into the indicated cell lines . Western analysis was carried out after 4 days , and viability ( n = 3 technical replicates ± SEM ) , was assessed after 7 days . Intensity of MYC bands relative to GAPDH bands is shown below the Western blots . ( C ) Cells were treated as in Figure 4A , and normalized expression of MYC was determined by q-RTPCR . Values represent the mean of three biological replicates , ± SEM . ( D ) LP-1 cells were transduced as in Figure 4E , and MYC protein expression was determined by Western analysis . See Figure 5—figure supplements 1 and Figure 5—source data 1 for additional data . DOI: http://dx . doi . org/10 . 7554/eLife . 10483 . 01310 . 7554/eLife . 10483 . 014Figure 5—source data 1 . Source data for Figure 5A and Figure 5—figure supplement 1B . DOI: http://dx . doi . org/10 . 7554/eLife . 10483 . 01410 . 7554/eLife . 10483 . 015Figure 5—figure supplement 1 . Suppression of the IRF4/MYC axis is important for the effects of CBP/EP300 bromodomain inhibition . ( A ) The indicated cell lines were transduced as in Figure 5B , and Western analysis and viability were assessed as in Figure 5B . ( B ) As in Figure 5A , except with MYC expression . ( C ) Cells were treated with I-CBP112 for 6 hr , and levels of MYC were determined as in Figure 4A . Values represent the mean of n = 3 biological replicates ± SEM . ( D ) Cells were treated as in Figure 4G , and sequencing traces at the IgH enhancer and the MYC transcriptional start site are shown . ( E ) As in Figure 4D , except with MYC expression . Values represent the mean of n = 3 technical replicates , ± SEM . ( F ) OPM2 cells were transduced with the indicated shRNAs . Western analysis was carried out after 4 days , and viability was assessed by flow cytometry after 7 days . Values represent the mean of n = 3 technical replicates , ± SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 10483 . 015 To better understand the events downstream of IRF4 suppression that are important for reducing proliferation and viability following CBP/EP300 bromodomain inhibition , we reduced the expression of IRF4 in a panel of multiple myeloma cell lines through shRNA transduction . The results first indicate that those cell lines that are sensitive to SGC-CBP30 ( GI50 < 2 . 5 μM ) require IRF4 for viability ( Figures 5B and Figure 5—figure supplement 1A ) . Further , consistent with published results ( Shaffer et al . , 2008 ) , knockdown of IRF4 and reduction in viability are associated with concomitant suppression of the oncogenic transcription factor c-MYC ( MYC ) . We reasoned that CBP/EP300 bromodomain inhibition may exert its phenotypic effects through the suppression of MYC downstream of IRF4 in multiple myeloma cells . While not among the most downregulated genes following the treatment of LP-1 cells with SGC-CBP30 , MYC expression was significantly reduced ( see below ) , and MYC transcriptional programs were affected ( Figure 3—figure supplement 1E ) . Further , as with IRF4 suppression , the degree of suppression of MYC mRNA in a panel of cell lines is significantly correlated with phenotypic sensitivity to SGC-CBP30 ( Figure 5—figure supplement 1B ) . To confirm the dose-dependent reduction of MYC expression , we treated LP-1 and OPM2 cells with either SGC-CBP30 or I-CBP112 ( Figures 5C and Figure 5—figure supplement 1C ) . The expression of MYC was reduced in a dose-dependent manner , with IC50 values somewhat higher than those observed for IRF4 suppression ( Figures 4A and Figure 4—figure supplement 1A ) . We also noted that H3K18ac and H3K27ac were reduced at the chromatin regions driving MYC expression following CBP/EP300 bromodomain inhibition , although loss of EP300 was less apparent , consistent with IRF4 suppression being up-stream of MYC suppression in this context ( Figure 5—figure supplement 1D ) . Further , consistent with the suppression of IRF4 at both the mRNA and protein levels ( Figures 4D , E ) , the expression of MYC was reduced following the knockdown of either EP300 or CBP in LP-1 and OPM2 cells ( Figure 5D , Figure 5—figure supplement 1E and F ) . Taken together , these data suggest that the bromodomains of CBP and EP300 are involved in the regulation of the IRF4/MYC axis in multiple myeloma cells , and the suppression of the IRF4/MYC axis may be important for the phenotypic effects of CBP/EP300 bromodomain inhibition . To further test the link between the transcriptional effects on IRF4/MYC and the phenotypic consequences of CBP/EP300 bromodomain inhibition , we generated LP-1 cell lines containing inducible IRF4 ( LP-1/IRF4 ) or MYC ( LP-1/MYC ) expression cassettes . We then treated these cell lines with SGC-CBP30 or I-CBP112 in the presence or absence of doxycycline to induce ectopic expression of IRF4 or MYC . As shown in Figure 6A and Figure 6—figure supplement 1 , in the absence of doxycycline , CBP/EP300 bromodomain inhibition induces G0/G1 arrest within 24 hr , consistent with our previous observations ( Figure 2C ) . However , upon ectopic expression of IRF4 , the cell cycle arrest is completely abrogated , indicating that suppression of IRF4 is required for the most proximal phenotypic consequence of CBP/EP300 bromodomain inhibition . While long-term viability appears to be reduced by the over-expression of IRF4 itself , there is a significant abrogation of growth inhibition and a reduced induction of apoptosis over background in the presence of ectopic IRF4 after a 6-day incubation with CBP/EP300 inhibitor ( Figure 6A ( right ) and Figure 6—figure supplement 1A ) . 10 . 7554/eLife . 10483 . 016Figure 6 . CBP/EP300 bromodomain inhibition suppresses the IRF4/MYC axis to cause viability defects . ( A ) IRF4 expression was induced in the LP1/IRF4 cell line by the addition of doxycycline . Left , lysates were prepared after 3 days and used for Western analysis with the indicated antibodies . Middle , cells were incubated for an additional 24 hr with DMSO or SGC-CBP30 ( 2 . 5 μM ) and fixed for cell cycle analysis by flow cytometry . Representative histograms of two biological replicate experiments are shown . Right , Cells were incubated for 6 days in the presence of SGC-CBP30 ( 2 . 5 μM ) . Viable cells were counted by flow cytometry and percent growth was calculated relative to the DMSO-treated condition for induced or uninduced cells . Values represent the mean of n = 3 technical replicates , ± SEM ( B ) Cells were induced as in ( A ) and were treated with DMSO or SGC-CBP30 ( 2 . 5 μM ) for 6 hr . Expression of MYC was measured by qRT-PCR , normalized to GAPDH , and expressed relative to uninduced cells treated with DMSO . Values represent the mean of n = 3 technical replicates , ± SEM . ( C ) As in ( B ) except cells were treated for 24 hr and lysed for Western analysis with the indicated antibodies . Values represent the ratio of GAPDH-normalized MYC expression relative to uninduced DMSO-treated cells . ( D ) MYC expression was induced in the LP1/MYC cell line by the addition of doxycycline . Cells were incubated for an additional 24 hr with DMSO or SGC-CBP30 ( 2 . 5 μM ) and fixed for cell cycle analysis by flow cytometry . Representative histograms of two independent experiments are shown . ( E ) RNA sequencing data from Figure 3A is expressed as the mean of the two biological replicates ( ± SEM ) normalized to DMSO-treated cells . ( F ) Model for the suppression of the IRF4/MYC axis by CBP/EP300 and BET bromodomain inhibitors . DOI: http://dx . doi . org/10 . 7554/eLife . 10483 . 01610 . 7554/eLife . 10483 . 017Figure 6—figure supplement 1 . Additional data pertaining to IRF4 and MYC reconstitution experiments in Figure 6 . ( A ) Quantification of % sub G1 following 7 d of treatment with the DMSO , SGC-CBP30 ( 2 . 5 μM ) , or I-CBP112 ( 5 μM ) in the absence ( -DOX ) or presence ( +DOX ) of ectopic IRF4 . Fold increase above DMSO treatment for each condition is shown above the bars . Values represent the mean and SEM of n = 3 technical replicates . ( B , C , and D , ) as in Figure 6A , B , and C , except with the LP-1/MYC cell line . E and F as in Figure 6 , except with I-CBP112 at 5 μM . DOI: http://dx . doi . org/10 . 7554/eLife . 10483 . 017 If the IRF4-mediated suppression of MYC is required for the phenotypic effects of CBP/EP300 bromodomain inhibition , one would expect that ectopic expression of IRF4 should block MYC suppression . Indeed , we found that the induction of IRF4 in the LP-1/IRF4 cell line both increased MYC expression ( most prominently at the mRNA level ) and prevented the suppression of MYC by SGC-CBP30 and I-CBP112 ( Figure 6A far left , 6B and 6C and Figure 6—figure supplement 1E ) . Consistent with MYC suppression being a critical downstream effect of IRF4 suppression , ectopic expression of MYC in the LP-1/MYC cell line phenocopied ectopic expression of IRF4 , rescuing cell cycle arrest and abrogating MYC suppression following CBP/EP300 bromodomain inhibition ( Figure 6D and Figure 6—figure supplement 1B , C , D , and 1F ) . While BET proteins are known to similarly target MYC in multiple myeloma , a comparison of CBP/EP300 and BET bromodomain inhibition demonstrated that these modalities target the IRF4-MYC network at different nodes , with BET inhibition having no impact on IRF4 at the doses and timepoints examined ( Figure 6E , Figure 3D , and Figure 4—figure supplement 1B ) . Our data suggest that CBP/EP300 bromodomain inhibition exerts its anti-myeloma effects in a mechanism distinct from BET inhibition via the direct transcriptional inhibition of IRF4 and the downstream suppression of IRF4 target genes such as MYC . In the current study , we demonstrate that CBP/EP300 bromodomain inhibition results in cell cycle arrest and apoptosis in multiple myeloma cell lines . Viability effects are dependent on the silencing of the transcription factor IRF4 , which results in the downstream suppression of c-MYC . CBP/EP300 bromodomain inhibition thus targets the IRF4/MYC network , which is critical for multiple myeloma cells independent of the upstream oncogenic signal . A recent publication describes the use of the CBP/EP300 bromodomain inhibitor I-CBP112 to inhibit the growth of leukemic cells ( Picaud et al . , 2015 ) . Our data pointing to the preferential activity of both SGC-CBP30 and I-CBP112 in multiple myeloma cell lines as compared to leukemic cell lines is consistent with this published work . Similar to our findings , Picaud et al . observed minor cytostatic and limited cytotoxic effects in all leukemic cell lines screened with the exception of Kasumi-1 . Only upon examining the effects of I-CBP112 on clonogenic growth did the authors observe more broad phenotypic effects . Thus , while CBP/EP300 bromodomain inhibition may have robust cytotoxic effects in multiple myeloma , our results do not exclude the possibility that this modality would have additional therapeutic utility in leukemia by targeting leukemic self-renewal . Pharmacological inhibition of CBP/EP300 bromodomains represents a viable strategy for targeting these transcriptional co-activators . Evidence from genetic studies in mice has shown that ablation of any two of the four alleles of CBP and EP300 results in embryonic lethality , and mouse embryonic fibroblasts lacking expression of CBP and EP300 cannot proliferate ( Kung et al . , 2000; Jin et al . , 2011; Yao et al . , 1998 ) . The selective viability effects and limited transcriptional footprint observed with CBP/EP300 bromodomain inhibitors suggests that this modality is milder than genetic ablation , perhaps affording an acceptable therapeutic index once drug-like molecules are optimized . Our results are more broadly consistent with recent studies using SGC-CBP30 and I-CBP112 that demonstrated selective phenotypic and transcriptional effects in distinct biological contexts ( Hammitzsch et al . , 2015; Picaud et al . , 2015 ) . Mice with heterozygous loss of Cbp are prone to the development of hematologic malignancies , and human patients with germline mutations in CREBBP develop the Rubinstein-Taybi cancer predisposition syndrome ( Kung et al . , 2000; Iyer et al . , 2004 ) . Further , recent surveys of the mutational landscape in a variety of tumors have demonstrated frequent loss of function mutations in CREBBP and EP300 ( Pasqualucci et al . , 2011; Peifer et al . , 2012; Kishimoto et al . , 2005; Mullighan et al . , 2011; Zhang et al . , 2012 ) . While this evidence implicates CBP/EP300 as tumor suppressors , evidence also supports their oncogenic activity . Rare human leukemias have been found with oncogenic fusion proteins containing either CBP or EP300 , and these oncogenic fusion proteins require the activity of CBP or EP300 ( Murati et al . , 2007; Ohnishi et al . , 2008; Yung et al . , 2011; Wang et al . , 2011 ) . Genetic ablation and pharmacological inhibition of CBP/EP300 in leukemic cell lines and primary patient samples also support the oncogenic role of CBP and EP300 ( Picaud et al . , 2015; Giotopoulos et al . , 2015 ) . Our data in multiple myeloma are consistent with an activity supporting oncogenic signaling , as either pharmacological inhibition or knockdown resulted in loss of viability . It is unclear whether CBP/EP300 bromodomain inhibition would have tumor promoting activity in normal tissues . However , concerns about inhibiting potential tumor suppressor activity of CBP/EP300 in normal tissues may be alleviated by a dosing regimen that prevents continuous target coverage in normal tissues . BET bromodomain inhibitors are highly active in hematologic malignancies , including multiple myeloma ( Delmore et al . , 2011; Mertz et al . , 2011 ) . The activity of CBP/EP300 bromodomain inhibitors in multiple myeloma potentially suggests that this modality may modify similar genes regulated by BET bromodomain inhibitors , but transcriptional profiling does not support this notion . At the doses of SGC-CBP30 utilized , CBP/EP300 bromodomain inhibition appears to have a more circumscribed transcriptional footprint than BET bromodomain inhibition . Phenotypic effects of BET bromodomain inhibition in multiple myeloma are likely due to direct suppression of MYC and BCL2 , while the effects of CBP/EP300 bromodomain inhibition appear to be via suppression of IRF4 . The distinct transcriptional effects of the two modalities suggests that combinations may be efficacious . It has in fact been shown that targeting the IRF4 network with lenalidomide and the MYC network with BET bromodomain inhibitors has synergistic effects in mantle cell lymphoma and primary effusion lymphoma ( Moros et al . , 2014; Gopalakrishnan et al . , 2015 ) . CBP/EP300 bromodomain inhibition may thus represent an alternative strategy for targeting the IRF4 transcriptional axis in these contexts . The discovery of BET bromodomain inhibitors represented a breakthrough in the ability to target what were thought to be intractable oncogenic factors . Here we have shown that CBP/EP300 bromodomain inhibitors may similarly be used to target the expression of critical oncogenic transcription factors . As dysregulated transcriptional control is central to the pathology of cancer , the ability to target oncogenic transcription networks with small molecule bromodomain inhibitors represents a promising direction for future therapeutics . Sources of cell lines and results of mycoplasma testing are provided as Supplementary file 2 . All cell lines were used within 1–2 months of thawing from original stock vials received from supplier and were not further authenticated . LP-1 cells containing doxycycline-inducible IRF4 were generated as described for the inducible LP-1/MYC cell line ( Mertz et al . , 2011 ) using the IRF4 coding sequence ( RefSeq BC015752 . 1 ) obtained from Origene Technologies , Inc . ( Rockville , MD ) as a template . Inducible cell lines were incubated with 1 μg/ml doxycycline ( Sigma-Aldrich , Inc; St . Louis , MO ) for 3 days . SGC-CBP30 ( 2 . 5 μM ) or I-CBP112 ( 5 μM ) was added for 6 hr for RNA analysis or for 24 hr , and cells were fixed for cell cycle analysis or pelleted for Western analysis , or were seeded in a 96 well plates for long -term viability testing . NanoBRET was carried out using the NanoBRET Protein:Protein Interaction System ( Promega , Inc . ; Madison , WI ) according to the manufacturer’s instructions . Briefly , HEK293 cells were transiently co-transfected with a vector for histone H3 . 3-HaloTag and a NanoLuc tagged CBP bromodomain expression construct . Transfected cells were plated in 96 -well plates in the presence or absence of ligand , then treated with dose titrations of indicated compounds . Readings were performed on an Envision Plate Reader ( Perkin Elmer , Inc . ; Waltham , MA ) and BRET readings were calculated by dividing the acceptor emission value ( 600 nm ) by the donor emission value ( 460/50 nm ) . As described previously ( Huang et al . , 2014 ) , this assay monitors the compound-dependent release and aggregation of a fusion protein consisting of a bromodomain and the fluorescent protein ZsGreen . For the BRD4 chromatin release assay , U2OS cells capable of inducibly expressing the full-length BRD4 protein in fusion with ZsG were generated using the pLVT3G/ZsGreen-BRD4/TO3G vector and maintained in blasticidin at 15 μg/ml . Consistent with published data ( Dawson et al . , 2011 ) , we did not observe global release of full length CBP fused to ZsGreen in response to compounds or bromodomain point mutations . Therefore , the bromodomain ( BD ) of CBP was cloned into full length BRD9 ( replacing the BRD9 BD ) in frame with a ZsGreen fluorescent tag ( ZsG ) . U2OS cells capable of inducibly expressing the ZsGreen-CBPBD fusion protein were generated by lentiviral delivery of the pLVT3G/BRD9-ZsG-CBPBD/TO3G vector , which contains both the inducible fusion protein and the tet transactivator . Cells were selected and maintained in the presence of 15 μg/ml blasticidin . 5000 cells/well were seeded in 384-well imaging plates in the presence of 2 μg/ml doxycycline to induce the expression of ZsG-fusion proteins . After 16 hr of incubation with doxycycline , fresh media containing serial dilution of compounds were added to the cells for 2 hr at 37°C . Cells were fixed with 4% paraformaldehyde ( PFA ) dissolved in PBS for 15 min at room temperature . Images of cells were acquired using ImageXpress Micro ( Molecular Devices , Inc . ; Sunnyvale , CA ) and processed with the Transfluor Module of MetaXpress software . Average pits per cell values were obtained from four adjacent images in each well with two technical replicates for each compound concentration . Dose-response curves were generated by plotting the average pits per cell values at each dose and EC50 values were calculated by a four-parameter non-linear regression model in GraphPad Prism . Differential scanning fluorimetry ( DSF ) was performed as described ( Niesen et al . , 2007 ) with the indicated bromodomains using the ViiA7 real time PCR system ( Life Technologies , Inc . ; Carlsbad , CA ) . Variable buffer compositions were used for the different bromodomain proteins with 12X SYPRO orange dye , 20 μM of the compounds or equivalent percentage of DMSO , and 4–8 µM of the indicated protein . A melting curve was established using a range of 25–95°C and a ramping rate of 3°C per minute . The melting temperature ( Tm ) for each sample was determined using the ViiA7 software ( version 1 . 2 . 2 ) and the ΔTm was calculated by subtracting the Tm of the control from the Tm of the compound treated sample . Cells were plated at 5000–10000 cells per well of 96-well plates containing titrations of the compounds as indicated . After incubation , the cells were incubated with 500 μg/ml resazurin ( Sigma ) in PBS for 2–8 hr , and fluorescence was measured ( Ex 530 nm , Em 590 nm ) . Cell cycle analysis was performed as described previously ( Mertz et al . , 2011 ) . For visualization , DNA content histograms were generated with GraphPad Prism . Dose-response curves were generated by plotting the normalized percent growth , percent sub-G1 and percent increase in G0/G1 at each dose values . GI50 values were determined as the concentration at which viability was 50% of the DMSO value and calculated by a four-parameter non-linear regression model in GraphPad Prism . Cell synchronization was performed as described ( Mertz et al . , 2011 ) , and cells were released into media containing DMSO , 2 . 5 μM SGC-CBP30 or 5 μM I-CBP112 . Lentiviral shRNA vector and packaging have been described previously ( Mertz et al . , 2011 ) . Cells ( 2E6 cells/ml ) were transduced with lentivirus at an MOI of 5–10 in 8 μg/ml sequebrene ( Sigma ) and centrifuged at 1000g for 2 hr . Cells were diluted to 1 × 106 cells/ml overnight . Infected cells were diluted to 2 × 105 cells/mL in 1 μg/ml puromycin and transferred to 96-well plates or TC flasks . After 3–4 days , cells in flasks were pelleted and used for qRT-PCR or Western analysis . Cells in 96-well plates were incubated for 9 days and fixed for cell cycle analysis , with passaging and fixing of aliquots as indicated . Target sequences for shRNAs were as follows: shEP300 1: 5’CGGAAACAGTGGCACGAAGAT3’; shEP300 2: 5’CGGAGGATATTTCAGAGTCTA3’; shEP300 3: 5’GCGGAATACTACCACCTTCTA3’; shCBP 1: 5’CCTCTTTGGAGTCTGCATCCT3’; shCBP 2: 5’GAGCTTCCCAAGTTAAAGAAG3’; shCBP 3: 5’GCCCATTGTGCATCTTCACGA3’ . For IRF4 knockdown , validated shRNA constructs were obtained from Sigma . Constructs shIRF4-1 , shIRF4-2 , shIRF4-3 , shIRF4-4 , and shIRF4-5 correspond to TRCN0000429523 , TRCN0000014764 , TRCN0000014765 , TRCN0000014767 , and TRCN000433892 , respectively . Total RNA was prepared with an RNeasy Mini Kit ( Qiagen , Inc . ; Hilden , Germany ) with on column DNAse digestion and submitted to Ocean Ridge Biosciences ( Palm Beach Gardens , FL ) for sequencing and mapping . The data in RPKM for each gene with compound treatment was compared to DMSO treatment , and log2 fold changes were used for further analysis . Rank ordered gene lists were used for Gene Set Enrichment Analysis ( Subramanian et al . , 2005 ) . RNA preparation , cDNA synthesis , and qRT-PCR were performed as described ( Mertz et al . , 2011 ) . For dose titration experiments , cells in 96 -well plates were lysed in lysis buffer ( 1% Triton X-100 , 0 . 01 μM glycine pH 2 . 5 ) and used directly for cDNA synthesis and qRT-PCR . Primer sequences can be found in Supplementary file 1 . Whole cell extracts were prepared by lysis in RIPA buffer + EDTA ( Boston Bioproducts , Inc . ; Ashland , MA ) with protease inhibitor cocktail ( Roche Life Sciences; Indianapolis , IN ) . Extracts were subjected to SDS-PAGE and Western analysis with MYC ( Cell Signaling ( Danvers , MA ) #5605 ) , IRF4 ( Cell Signaling #4964 or #4948 ) , GAPDH ( Life Technologies AM4300 ) , CBP ( Santa Cruz ( Dallas , TX ) sc-369 ) , EP300 ( Santa Cruz sc-584 ) , or β-actin ( Life Technologies AM4302 ) primary antibodies . For histone analysis , extracts were prepared by sulfuric acid extraction of permeablized nuclei , and extracted histones were subjected to SDS-PAGE and Western analysis with H3K18ac ( Cell Signaling #9675 ) , H3K27ac ( EMD Millipore ( Billerica , MA ) 07–360 ) , or H4 ( Abcam ( Cambridge , MA ) 31830 ) . Blots were incubated with DyLight conjugated secondary antibodies , imaged and quantified with a Licor fluorescence imager ( Licor , Inc . ; Lincoln , NE ) , or with HRP- conjugated secondary antibodies for ECL visualization . 5 x 107 LP-1 cells were treated for 6 hr with DMSO or 2 . 5 µM SGC-CBP30 at a density of 5 x 105/ml . Cells were fixed with a final concentration of 1% formaldehyde for 10 min at room temperature . Glycine was added to a final concentration of 0 . 125 M to stop crosslinking . The cells were washed twice in cold PBS followed by lysis at 4°C for 1 hr in buffer containing 10 mM Tris-HCl pH 7 . 5 , 10 mM NaCl , 5 mM MgCl2 , 0 . 2% NP-40 , + protease inhibitor cocktail ( Sigma ) . Following lysis , nuclei were recovered by centrifugation , and resuspended in buffer containing 10 mM Tris-HCl , 0 . 1 mM EDTA , 5 mM MgAc2 , 25% glycerol . An equal part 2X MNase buffer was added , containing 50 mM KCl , 8 mM MgCl2 , 2 mM CaCl2 , 100 mM Tris-HCl . Micrococcal nuclease ( Roche ) was added to 300 U/ml and chromatin was digested at room temperature for 20 min . Dilution buffer ( 0 . 1% SDS , 1 . 1% Triton-X 100 , 2 mM EDTA , 20 mM Tris-HCl pH 8 . 0 , 150 mM NaCl ) was added and nuclei were broken down by sonication . Chromatin was cleared by centrifugation and pre-cleared with protein A-conjugated magnetic beads ( Life Technologies ) . 10–20 µg pre-cleared chromatin was combined with 10 µg anti-EP300 antibody ( Santa Cruz sc-585X ) or 2 . 5 ug of anti-H3K18ac ( Cell Signaling , 9675 ) or anti-H3K27ac ( Abcam , ab4729 ) conjugated to protein A magnetic beads . IPs were performed overnight at 4°C . Immune complexes were washed twice in buffer containing 140 mM NaCl , once in buffer containing 360 mM NaCl , once in 250 mM LiCl wash buffer , and twice in TE . Samples were eluted and treated with 20 µg proteinase K ( Roche ) for 1 hr at 55°C , crosslinks were reversed for 4 hr at 65°C , and 20 µg RNase ( Sigma ) was added for 1 hr at 37°C . DNA was purified with the MinElute kit ( Qiagen ) , and libraries were prepared using the Ovation Ultralow DR Multiplex System ( NuGEN , Inc . ; San Carlos , CA ) according to the manufacturer’s recommendations . Amplified libraries were size selected and gel-purified prior to Illumina massively parallel sequencing on a HiSeq 2000 system at the MIT Biomicro Center . Biological replicates were performed for each sample , and representative images are depicted . Inhibitory activity of compounds was determined by following the inhibition of the binding of purified His-Flag-tagged bromodomains to H4-TetraAc-biotin peptide ( New England Peptide , Inc . ; Gardner , MA ) using AlphaLISA technology ( Perkin Elmer ) . Compound at varying concentrations were dispensed into 384 well Proxiplates ( Perkin Elmer ) using Echo technology ( Labcyte , Inc . ; Sunnyvale , CA ) . For CBP assays , 0 . 5 μM His-Flag-tagged CBP bromodomain ( amino acids 1082–1197 ) was incubated with 0 . 003 μM H4-TetraAc-biotin for 20 min at room temperature in 1x reaction buffer ( 50 mM HEPES pH 7 . 5 , 1 mM TCEP , 0 . 069 mM Brij-35 , 150 mM NaCl , and 0 . 1 mg/ml BSA ) . Streptavidin acceptor beads and nickel donor beads ( Perkin Elmer ) were added to 15 μg/ml with a Combi Multidrop dispenser . Plates were sealed and incubated at 90 min in the dark at room temperature , and plates were read on an Envision plate reader ( Perkin Elmer ) according to manufacturer’s instructions . For the BET assays , the protocol was similar except that BET family bromodomains were used at 0 . 03 μM ( BRD4-BD1 ) and incubated with 0 . 2 μM H4-TetraAc-biotin for 20 min in reaction buffer ( 40 mM HEPES pH7 . 0 , 1 mM DTT , 0 . 069 mM Brij-35 , 40 mM NaCl , and 0 . 1 mg/ml BSA ) . Streptavidin donor beads and Anti-Flag Acceptor beads ( Perkin Elmer ) were added to 10 μg/ml , and then plates were sealed and incubated in the dark for 60 min prior to reading on the Envision . The synthesis and characterization of CPI203 have been published previously ( Devaiah et al . , 2012 ) . SGC-CBP30 and I-CBP112 are commercially available ( Sigma ) .	Multiple myeloma is an aggressive and incurable cancer of white blood cells called B cells and plasma cells . Many of the mutations that trigger multiple myeloma interfere with genes that normally cause B cells to develop into plasma cells . Multiple myeloma cells often activate genes that are inactive in healthy cells or vice versa . They also express some genes that are active in healthy cells but at the wrong levels . These changes in gene expression are regulated by proteins that bind to DNA and other DNA-associated proteins . Proteins called CBP and EP300 are two examples of regulatory proteins , and have been implicated in promoting various cancers in humans . Both CBP and EP300 contain a region known as a bromodomain , which binds to proteins associated with DNA . Abnormal activity of the bromodomains of CBP and EP300 may thus promote the onset of cancer . Conery et al . have now treated a wide range of human blood cancer cells grown in the laboratory with two new chemicals that inhibit CBP and EP300 bromodomains . Of all the cells tested , multiple myeloma cells were the most strongly affected; these cells proliferated more slowly and died more quickly in the presence of the chemical inhibitors . Next , Conery et al . analysed the changes in gene expression in the multiple myeloma cells when they were treated with the inhibitors . The genes whose expression levels changed the most were genes that are regulated by a protein called IRF4 . This protein is important for normal B cell and plasma cell development . One notable IRF4 target gene that was down-regulated was the gene that encodes a protein called Myc , which strongly encourages cell division and growth . Conery et al . then supplemented the multiple myeloma cells with extra IRF4 or Myc while treating with the inhibitors and found that this caused the inhibitors to lose most of their effect . Neither CBP nor EP300 have previously been thought of as targets for multiple myeloma therapy . Therefore a next critical step is to find more refined chemicals to target their bromodomains and importantly to test these chemicals in preclinical trials . These studies could in turn lead to improved treatments for patients with multiple myeloma in the future .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "chromosomes", "and", "gene", "expression", "cancer", "biology" ]	2016	Bromodomain inhibition of the transcriptional coactivators CBP/EP300 as a therapeutic strategy to target the IRF4 network in multiple myeloma
The internal state of an organism influences its perception of attractive or aversive stimuli and thus promotes adaptive behaviors that increase its likelihood of survival . The mechanisms underlying these perceptual shifts are critical to our understanding of how neural circuits support animal cognition and behavior . Starved flies exhibit enhanced sensitivity to attractive odors and reduced sensitivity to aversive odors . Here , we show that a functional remodeling of the olfactory map is mediated by two parallel neuromodulatory systems that act in opposing directions on olfactory attraction and aversion at the level of the first synapse . Short neuropeptide F sensitizes an antennal lobe glomerulus wired for attraction , while tachykinin ( DTK ) suppresses activity of a glomerulus wired for aversion . Thus we show parallel neuromodulatory systems functionally reconfigure early olfactory processing to optimize detection of nutrients at the risk of ignoring potentially toxic food resources . Sensory systems undergo dramatic functional modifications when animals enter different internal states such as hunger or arousal ( Su and Wang , 2014 ) . These functional changes in neural circuits support flexible and adaptive behaviors in animals in response to their changing needs and circumstances . The mechanisms driving these neural circuit modifications are under intense investigation and are fundamental to our understanding of how neural circuits support animal cognition and behavior . Neuromodulators are an important class of molecules positioned to exert significant influence over local neural circuits through their effects on neuronal excitability and network properties ( Bargmann and Marder , 2013 ) . The study of hunger and satiety in fruit flies has identified key neuromodulators that serve to communicate an animal's nutritional status to its nervous system and thus , change its behavior . Neuromodulators such as dopamine , short neuropeptide F ( sNPF ) and NPF have been demonstrated to influence different aspects of appetitive behaviors such as taste sensitivity ( Inagaki et al . , 2012; Marella et al . , 2012; Inagaki et al . , 2014 ) , the formation and expression of appetitive memories ( Krashes et al . , 2009 ) , odor preference ( Root et al . , 2011; Beshel and Zhong , 2013 ) , and control of food intake ( Lee et al . , 2004; Yu et al . , 2004; Wu et al . , 2005; Wang et al . , 2013 ) . In natural environments , foraging and feeding behaviors expose animals to risks of predation and harmful toxins in food . For example , food deprivation increases an animal's tolerance for noxious stimuli ( Gillette et al . , 2000; Wu et al . , 2005; Inagaki et al . , 2014 ) and suppresses escape behavior at the risk of predation ( Gillette et al . , 2000; Gaudry and Kristan , 2009 ) . Thus , when evaluating potential food sources , animals must weigh both aversive and attractive sensory inputs . Their perceptions of these sensory stimuli and behavioral decisions are influenced by their own internal states and needs for energy homeostasis . In Drosophila , starvation heightens sensitivity in odorant receptor neurons ( ORNs ) that are critical for behavioral attraction to appetitive odors through neuromodulatory mechanisms ( Root et al . , 2011 ) . Starvation has also been shown to reduce behavioral avoidance to innately aversive odors ( Bracker et al . , 2013 ) . This starvation effect requires the presence of appetitive odors and has been suggested to occur at higher order levels of the Drosophila brain ( Bracker et al . , 2013 ) . Whether starvation also reduces sensitivity of ORNs in the periphery that are critical for behavioral avoidance is unknown . Here we describe a neuromodulatory mechanism in Drosophila controlling the reduction of aversive odor sensitivity during starvation at the level of the first olfactory synapse . This pathway operates in parallel and independently of the mechanisms driving increases in attractive odor sensitivity . We show here that starvation does not simply scale up or down global activity in the antennal lobe . Rather , it upregulates activity in certain sensory channels and downregulates it in others in what appears to be an optimization strategy that may serve to increase the hedonic value of food odors . Thus , individual neurons may read the same global metabolic signals and differentially respond in a manner that fine tunes local circuits towards a concerted modulation of appetitive behaviors . In Drosophila , some olfactory sensory channels are hardwired for innate behaviors ( Suh et al . , 2004; Kurtovic et al . , 2007; Semmelhack and Wang , 2009; Ai et al . , 2010; Grosjean et al . , 2011; Stensmyr et al . , 2012 ) . The hedonic value of behaviorally relevant odors is therefore encoded by glomerular activity in the early olfactory system . In particular , activation of the DM1 glomerulus , innervated by Or42b ORNs , triggers attraction to an odor , while activation of DM5 , innervated by Or85a ORNs , reduces attraction to high concentrations of vinegar ( Semmelhack and Wang , 2009 ) . Thus , activity in DM1 and DM5 represents positive and negative valence , respectively , and physiological modulation of these glomeruli should alter olfactory behaviors . To study how attractive and aversive odor input channels might be modulated in starved flies , we took advantage of the finding that the odor map changes as concentrations of odor increase ( Wang et al . , 2003 ) . At intermediate concentrations , food odors such as vinegar are attractive to hungry flies . At low or high concentrations , however , odors are ignored ( Semmelhack and Wang , 2009; Root et al . , 2011 ) . Using a single fly food odor search paradigm ( Root et al . , 2011; Zaninovich et al . , 2013 ) , we first sought to extend our earlier study which used low concentrations of vinegar ( Root et al . , 2011 ) and evaluated how behavioral attraction in starved flies changes in response to a broader range of odor concentrations from low to high . We measured the time required for starved and fed flies to locate a source of apple cider vinegar across a range of concentrations . Starved flies typically locate the odor source within minutes after entering the observation chamber ( Figure 1A ) . This behavior can be quantified by an appetitive index , which we define as the percentage of flies reaching the odor source within a 10 min observation period . In starved flies , the appetitive index rises ( from 23 to 60 ) as the vinegar concentration increases ( 0 . 5–25% ) , but then steadily declines at higher concentrations ( Figure 1B ) . At all tested concentrations ( 0 . 5–100% ) , the appetitive index in starved flies is greater than that of the fed flies . 10 . 7554/eLife . 08298 . 003Figure 1 . Starvation state fine-tunes appetitive behavior . ( A ) A single fly assay was used to measure food search behavior . The coordinates of representative fed ( left ) and starved ( right ) flies show their positions during a 10-min period in response to 5% cider vinegar . Scale bar: 10 mm . ( B ) The appetitive index of fed and starved Orco-Gal4 control flies at varying concentrations . ( C–E ) The appetitive index of receptor knockdown flies , in which the receptor RNAi is expressed in the Orco odorant receptor neurons ( ORNs ) . ( C ) Short neuropeptide F receptor ( sNPFR ) knockdown flies . ( D ) Drosophila tachykinin receptor ( DTKR ) knockdown flies . ( E ) sNPF and DTKR dual knockdown flies . ( B–E ) n = 63–129 for each condition . Error bars show s . e . m . p < 0 . 05 , p < 0 . 01 , p < 0 . 001; z-test for proportions comparing between starved and fed ( B ) , and comparing knockdown flies to the Orco-Gal4 and UAS- control group in the starvation state ( C , D , E ) . DOI: http://dx . doi . org/10 . 7554/eLife . 08298 . 00310 . 7554/eLife . 08298 . 004Figure 1—figure supplement 1 . Food search behavior in control and knockdown flies . ( A ) The appetitive index of fed control and sNPFR knockdown flies compared to a starved Orco-Gal4 control . n = 60–124 for each condition . ( B ) The appetitive index of the fed control and DTKR knockdown flies compared to a starved Orco-Gal4 control . n = 55–122 for each condition . ( C ) The appetitive index of control and sNPF knockdown flies in the fed and starved state . n = 76–118 for each condition . ( D ) The appetitive index of the fed control and sNPF-DTKR double knockdown flies compared to a starved Orco-Gal4 control . n = 56–127 for each condition . Error bars show s . e . m . p < 0 . 01; z-test for proportions comparing knockdown flies to the fed control groups ( A , B , D ) , and comparing knockdown flies to the starved control group ( C ) . DOI: http://dx . doi . org/10 . 7554/eLife . 08298 . 004 sNPF , a Drosophila homolog of the mammalian orexigenic neuropeptide Y ( NPY ) ( Barsh and Schwartz , 2002; Lee et al . , 2004 ) , enhances the olfactory sensitivity of ORNs and increases appetitive behavior at low odor concentrations ( Root et al . , 2011 ) . To what extent does sNPF signaling account for behavioral attraction at all odor concentrations ? To answer this question , we used RNAi to knock down expression of the sNPF receptor ( sNPFR ) in ORNs using Orco-Gal4 ( Larsson et al . , 2004; Vosshall and Hansson , 2011 ) . At low concentrations ( 0 . 5–1% ) of cider vinegar , knockdown of sNPFR eliminated the behavioral difference between starved and fed flies ( Figure 1C ) . However , at high concentrations ( 5–25% ) , the appetitive index of sNPFR knockdown flies remains significantly higher than that of the control fed flies . In a fed state , sNPFR knockdown had no behavioral effect compared to the control flies ( Figure 1—figure supplement 1A ) . These results suggest the existence of a parallel mechanism for starvation-mediated changes in appetitive behavior at high odor concentrations . One potential mechanism for regulating appetitive behavior at high odor concentrations is to alter a sensory neuron's neuropeptide response by modulating levels of a neuropeptide receptor ( Root et al . , 2011 ) . Most neuromodulators signal through G-protein coupled receptors ( GPCRs ) ( Greengard , 2001 ) and a small increase in GPCR expression can have a dramatic effect on neural circuit function and related behavior ( Bendesky et al . , 2011 ) . We therefore performed a transcriptome analysis using RNA-seq to identify differentially expressed GPCRs in the antennae of fed and starved flies . We focused on the typical GPCR families ( Brody and Cravchik , 2000 ) that include receptors for biogenic amines , neuropeptides , as well as classic neurotransmitters , because these types of GPCRs have the potential to alter excitability ( Greengard , 2001 ) . This analysis identified 34 GPCRs that have higher expression in the antennae of starved flies ( Table 1 ) . This group included sNPFR , a GPCR we had previously shown to undergo upregulation in Or42b ORNs after starvation ( Root et al . , 2011 ) . Other upregulated GPCRs , not yet described in ORNs , include the dopamine 2-like and dopamine-ecdysone receptors which have both been implicated in enhancing fly sugar receptor sensitivity in the starved state ( Inagaki et al . , 2012; Marella et al . , 2012 ) . Additional GPCRs shown to influence feeding behaviors include the serotonin 2A receptor ( Gasque et al . , 2013 ) , a receptor that promotes anorectic behaviors and the GABA-B receptor type 1 ( Bjordal et al . , 2014 ) , a receptor that is part of a circuit that detects amino acid imbalances . Whether these receptors are expressed in select glomeruli remains to be determined . Interestingly , dopaminergic ( Riemensperger et al . , 2005 ) and serotonergic ( Roy et al . , 2007; Dacks et al . , 2009 ) terminals have been described in the adult antennal lobe , thus making these receptor classes highly plausible targets for internal state modulation . 10 . 7554/eLife . 08298 . 005Table 1 . Differentially expressed GPCRs in the antennae of fed and starved fliesDOI: http://dx . doi . org/10 . 7554/eLife . 08298 . 005FlyBase IDGeneGene nameCount ratiop-valueFPKM starvedA . Receptors for biogenic amines and related compounds FBgn0011582DopRDopamine receptor1 . 520 . 0002 . 59 FBgn0053517D2RDopamine 2-like receptor1 . 320 . 0011 . 63 FBgn0038980oa2Octopamine receptor 21 . 290 . 00019 . 64 FBgn0038542TyrRTyramine receptor1 . 230 . 0200 . 85 FBgn00041685-HT1ASerotonin receptor 1A1 . 210 . 0008 . 27 FBgn0250910Octbeta3ROctbeta3R1 . 200 . 00017 . 00 FBgn0037546mAChR-Bmuscarinic Acetylcholine Receptor , B-type1 . 180 . 00010 . 28 FBgn0004514Oct-TyrROctopamine-Tyramine receptor1 . 160 . 0212 . 09 FBgn00870125-HT2Serotonin receptor 21 . 150 . 00010 . 50 FBgn0024944OambOctopamine receptor in mushroom bodies1 . 150 . 00037 . 92 FBgn0000037mAcR-60Cmuscarinic Acetylcholine receptor 60C1 . 140 . 0006 . 74 FBgn0035538DopEcRDopamine/Ecdysteroid receptor1 . 080 . 000133 . 42 FBgn0015129DopR2Dopamine receptor 21 . 070 . 0444 . 67 FBgn0038063Octbeta2ROctbeta2R0 . 790 . 0031 . 30B . Peptide receptors FBgn0039396CcapRCardioacceleratory peptide receptor2 . 420 . 0170 . 13 FBgn0004622Takr99DTachykinin-like receptor at 99D1 . 670 . 0290 . 28 FBgn0003255rkrickets1 . 500 . 0000 . 98 FBgn0033579CG13229–1 . 450 . 0021 . 78 FBgn0053696CNMaRCNMamide Receptor1 . 440 . 0190 . 51 FBgn0036934sNPF-Rshort neuropeptide F receptor1 . 410 . 0007 . 26 FBgn0028961AlstRAllatostatin receptor1 . 300 . 0110 . 85 FBgn0035331DmsR-1Dromyosuppressin receptor 11 . 260 . 0031 . 59 FBgn0038880SIFRSIFamide receptor1 . 200 . 0002 . 78 FBgn0259231CCKLR-17D1CCK-like receptor at 17D11 . 130 . 00068 . 50 FBgn0025631moodymoody1 . 090 . 00050 . 86 FBgn0016650FshFsh-Tsh-like receptor1 . 080 . 0217 . 74 FBgn0085410TrissinRTrissin receptor1 . 060 . 02515 . 30 FBgn0038874ETHRETHR0 . 940 . 00321 . 31 FBgn0031770CG13995–0 . 910 . 00015 . 93 FBgn0004841Takr86CTachykinin-like receptor at 86C0 . 910 . 0166 . 86 FBgn0029723Proc-RProctolin receptor0 . 890 . 0056 . 22 FBgn0030954CCKLR-17D3CCK-like receptor at 17D30 . 790 . 0008 . 62 FBgn0025595AkhRAdipokinetic hormone receptor0 . 740 . 00011 . 06 FBgn0038201Pk1rPyrokinin 1 receptor0 . 670 . 00011 . 58 FBgn0039354Lgr3Lgr30 . 510 . 0000 . 24 FBgn0039595AR-2Allatostatin receptor 20 . 390 . 0020 . 11C . Metabotropic glutamate receptor family FBgn0050361mttmangetout3 . 270 . 0000 . 54 FBgn0019985mGluRAmetabotropic glutamate receptor1 . 940 . 0001 . 14 FBgn0052447CG32447–1 . 890 . 0004 . 27 FBgn0031275GABA-B-R3GABA-B receptor subtype 31 . 270 . 0002 . 44 FBgn0051760CG31760–1 . 170 . 0006 . 36 FBgn0051660pogpoor gastrulation1 . 160 . 00034 . 68 FBgn0260446GABA-B-R1GABA-B receptor subtype 11 . 160 . 00033 . 18 FBgn0085401CG34372–1 . 100 . 0333 . 96 FBgn0027575GABA-B-R2GABA-B receptor subtype 20 . 940 . 00027 . 45Each RNA sample was from the antennae of 200 female flies ( w1118;+;Orco-Gal4/+ ) . Count ratio is the number of reads aligned to each gene between starved and satiated flies . FPKM , fragment per kilobase of exon per million mapped fragments . p-values were calculated on raw counts using the Fisher's exact test in edgeR package . We also identified 11 GPCRs that exhibited reduced expression in the antennae of starved flies ( Table 1 ) . This group included a few GPCRs described as having roles in feeding behaviors or energy storage . One example is as the adipokinetic hormone receptor , which regulates lipid and carbohydrate storage ( Bharucha et al . , 2008 ) . Another is the pyrokinin 1 receptor , which is activated by hugin , a neuropeptide that suppresses feeding ( Melcher and Pankratz , 2005 ) . Although it is not known whether these two antennal GPCRs are expressed in ORNs , such localization has been reported for the Drosophila tachykinin receptor ( DTKR ) . This receptor , also known as Takr99D , mediates presynaptic inhibition ( Ignell et al . , 2009 ) and is implicated in nutritional stress responses ( Winther and Nassel , 2001 ) . We therefore focused our attention on this receptor . To assay whether DTKR expression in ORNs is required for the post-fasting behavioral modification , we knocked down DTKR and measured the appetitive index . Starved DTKR knockdown flies do not behave differently from control starved flies in response to low concentrations ( 0 . 5–1% ) of cider vinegar . In contrast , their appetitive index at high concentrations ( 5–25% ) of cider vinegar is significantly lower than that of control flies ( Figure 1D ) . Furthermore , we found no effect of DTKR knockdown on the behavior of fed flies ( Figure 1—figure supplement 1B ) . Thus , DTKR signaling is necessary for starvation-dependent change in food search behavior at high , but not low odor concentrations . Our results indicate both sNPFR and DTKR signaling contribute to appetitive changes . Do these two pathways fully account for the starvation response ? To address this question , we explored whether removal of both signaling mechanisms transforms the behavior of starved flies into that of fed flies . Indeed , simultaneous knockdown of the sNPF peptide ( the equivalent of sNFPR knockdown , see Figure 1—figure supplement 1C ) and DTKR in Orco neurons abolished the effect of starvation , leading to behavior indistinguishable from that of fed flies ( Figure 1E and Figure 1—figure supplement 1D ) . Thus , these two modulation systems are both required to bring about the appetitive behavior observed in starved flies . To identify a circuit-level mechanism for modulation of food search behavior , we next examined whether corresponding changes in glomerular activity could be detected . We used two-photon microscopy to monitor odor-evoked activity in the second order projection neurons ( PNs ) that receive input from ORNs ( Wang et al . , 2003 ) . Flies bearing GH146-LexA and LexAop-GCaMP transgenes allow the imaging of PN dendritic calcium responses in specific glomeruli to cider vinegar ( Figure 2A ) . Given that the PN response in DM1 to low odor concentrations is sensitized by starvation ( Root et al . , 2011 ) , we compared the response to higher odor concentrations in fed and starved flies . Strikingly , starvation suppresses olfactory sensitivity of DM5 ( Figure 2B ) a glomerulus that mediates aversion ( Semmelhack and Wang , 2009 ) . Furthermore , DM5 is the only glomerulus recruited by vinegar that is modulated by starvation at high odor intensity ( see Figure 2—source data 1 for a more complete characterization of glomerular responses to vinegar in fed and starved flies ) . Testing a range of concentrations , we found that DM5 became activated at 20% saturated vapor ( SV pressure ) in fed flies , but not until 80% SV in starved flies ( Figure 2C ) . At the concentrations when DM5 is suppressed by starvation , DM1 responses saturate and do not exhibit further modulation by starvation ( Figure 2C ) . Thus , modulation of DM5 olfactory sensitivity by starvation increases the activation threshold of a glomerulus that mediates aversion . 10 . 7554/eLife . 08298 . 006Figure 2 . Starvation-dependent neuropeptide signaling modulates sensitivity of the DM1 and DM5 glomeruli . ( A ) Representative two photon images of projection neuron ( PN ) dendritic calcium responses to 80% saturated vapor ( SV ) of apple cider vinegar in starved DTKR knockdown flies . Grey-scale images show the glomerular map on three optical planes whereas the pseudocolored images show the change of fluorescence . ( B ) Peak ΔF/F in glomeruli that are activated by 80% SV cider vinegar in fed and starved flies . ( C , D ) Responses in the DM1 and DM5 glomeruli at varying concentrations in fed and starved control flies ( C ) and in flies that express DTKR-RNAi or sNPFR-RNAi in ORNs labeled by Orco-Gal4 ( D ) . Calcium signals for PN responses were imaged using GH146-LexA , LexAop-GCaMP flies in addition to the indicated transgenes . n = 5–7 for each . Error bars show s . e . m . p < 0 . 05 , p < 0 . 01 , *p < 0 . 001; Student's t-test comparing between starved and fed responses . DOI: http://dx . doi . org/10 . 7554/eLife . 08298 . 00610 . 7554/eLife . 08298 . 007Figure 2—source data 1 . Glomerular responses to vinegar in fed and starved flies . DOI: http://dx . doi . org/10 . 7554/eLife . 08298 . 00710 . 7554/eLife . 08298 . 008Figure 2—figure supplement 1 . PN responses to vinegar in flies with DTKR knockdown . Two-photon imaging of PN dendritic calcium responses . ( A ) Peak ΔF/F in glomeruli that are activated by 80% SV cider vinegar in control and DTKR knockdown flies . ( B ) Peak ΔF/F in the DM1 glomerulus in response to 0 . 2% SV cider vinegar in control and DTKR knockdown flies . ( C ) Representative images of calcium responses to 0 . 4% SV of ethyl butyrate ( 1:100 dilution in mineral oil ) . Grey-scale images show the glomerular map on the DM5 plane whereas the pseudocolored images show the change of GCaMP fluorescence . ( D ) Peak ΔF/F in the DM5 glomerulus in response to ethyl butyrate in control and DTKR knockdown flies . All flies have GH146-LexA , LexAop-GCaMP , Orco-Gal4 , and DTKRi flies have UAS-DTKR-RNAi in addition . For each condition , n = 4–9 . Error bars show s . e . m . p < 0 . 05 , *p < 0 . 01; Student's t-test comparing the response between knockdown and control groups ( A ) and comparing between starved and fed responses ( B , D ) . DOI: http://dx . doi . org/10 . 7554/eLife . 08298 . 008 In light of our RNAi results ( Figure 1D ) , we hypothesized that the starvation-dependent suppression of olfactory sensitivity in DM5 is controlled by DTKR . To test this idea , we imaged PN responses to cider vinegar while knocking down DTKR in most ORNs in flies bearing the GH146-LexA , LexAop-GCaMP , Orco-Gal4 , and UAS-DTKR-RNAi transgenes . Consistent with our hypothesis , knockdown of DTKR in ORNs abolished the starvation-dependent suppression in DM5 across a range of odor concentrations ( Figure 2D ) . DM1 responses to 80% SV vinegar , however , remained unaffected . The lack of an effect of DTKR on DM1 responses may reflect a saturation of these responses at this odor concentration that would mask further increase in response amplitude upon reduction of DTKR levels ( Figure 2—figure supplement 1A , B ) . Furthermore , starvation-dependent sensitization in DM1 was abolished by knockdown of sNPFR ( Figure 2D ) , as previously reported ( Root et al . , 2011 ) . Although two other glomeruli ( VM2 and DM2 ) also exhibited changes in activity at a high odor concentration when DTKR signaling is removed ( Figure 2—figure supplement 1A ) , it is known that these two glomeruli do not contribute to appetitive behavior ( Semmelhack and Wang , 2009 ) . We next asked whether the starvation modulation of DM5 generalizes to other odors , and found that the response of DM5 PNs to ethyl butyrate is similarly modulated by DTKR ( Figure 2—figure supplement 1C , D ) . What is the source of DTK peptide for DM5 suppression ? Previous studies have shown that a population of GABAergic local interneurons ( LNs ) labeled by the GH298-Gal4 line is immunoreactive for DTK ( Ignell et al . , 2009 ) . We investigated whether these LNs are the source of DTK for DM5 modulation , by knocking down DTK expression in LNs with RNAi . We first monitored odor-evoked activity in flies that express GCaMP in PNs and DTK-RNAi in LNs or ORNs as a negative control . Imaging DM5 responses to cider vinegar , we found that DTK peptide knockdown in LNs abolished the starvation-dependent DM5 suppression , whereas expression of DTK-RNAi in ORNs did not have any effect ( Figure 3A ) . Next , we asked if DTK expression in LNs is required for starvation-dependent behavior , and found expression of DTK-RNAi in LNs , but not ORNs , significantly reduces the appetitive index for food search behavior ( Figure 3B ) . Together , these studies demonstrate that DTKR signaling is required for the starvation-induced suppression of DM5 , which augments food search behavior . 10 . 7554/eLife . 08298 . 009Figure 3 . Tachykinin released by antennal lobe local interneurons ( LNs ) is necessary for starvation-dependent suppression of DM5 glomerular activity . ( A ) Representative traces showing ΔF/F in the DM5 glomerulus in flies that have UAS-DTK-RNAi in ORNs ( Orco-Gal4 ) or in LNs ( GH298-Gal4 ) . Bar graph depicts peak ΔF/F for each indicated genotype . Calcium signals for PN responses were imaged using GH146-LexA , LexAop-GCaMP flies in addition to the indicated transgenes . n = 5 for each condition . ( B ) The appetitive index of DTK knockdown flies in response to 5% cider vinegar . n = 87–101 for each condition . For imaging experiments , Error bars show s . e . m . p < 0 . 05 , Student's t-test ( A ) and z-test for proportions ( B ) comparing GH298-Gal4 , UAS-DTK-RNAi group to control groups or Orco-Gal4 , UAS-DTK-RNAi group . DOI: http://dx . doi . org/10 . 7554/eLife . 08298 . 009 Thus far we have shown that two parallel neuropeptide signaling systems are employed at different ends of the odor concentration spectrum to enhance appetitive behavior . The data suggest that sNPF and DTK receptors are preferentially upregulated by starvation in select glomeruli , such as DM1 and DM5 . We therefore investigated whether DTK and sNPFRs coexist in the same ORN population , or whether each receptor selectively exerts greater effect on specific glomeruli . To test this , we first investigated the effect of exogenous application of either peptide on the activity of DM1 and DM5 . We performed electrical stimulation of the olfactory nerve while measuring calcium activity in PNs before and after addition of synthetic sNPF or DTK . We found that exogenous sNPF increased the response of DM1 but not DM5 in starved flies ( Figure 4A ) . Conversely , exogenous DTK suppressed activity in DM5 but not DM1 in starved flies ( Figure 4B ) . In a previous report , we observed modulation of DM1 when DTKR levels were knocked down in most ORNs ( Ignell et al . , 2009 ) . Thus our current observation that DTK peptide administration doesn't decrease responses in DM1 may be due to saturation of DTKR already present in Or42b by endogenous levels of the peptide; whereas upregulation of DTKR in Or85a may explain its enhanced response to the DTKR peptide in this population . It is noteworthy that other glomeruli such as DM2 and VM2 exhibit some DTK sensitivity even in the satiated state , as observed in the current and previous study ( Ignell et al . , 2009 ) . Thus , DM1 and DM5 appear to be specifically sensitive to the addition of sNPF and DTK , respectively , in both a glomerular-specific and starvation-dependent manner . 10 . 7554/eLife . 08298 . 010Figure 4 . sNPF and DTK modulatory mechanisms target different sensory neurons . ( A , B ) Representative traces of calcium activity ( left ) in the DM1 and DM5 PNs in response to olfactory nerve stimulation before and after bath application of sNPF ( A ) or DTK ( B ) synthetic peptides . The percent facilitation and suppression ( right ) are measured as the percent change in peak ΔF/F after peptide addition . ( C ) Peak ΔF/F responses in the DM1 PNs ( left ) and the DM5 PNs ( right ) to cider vinegar with glomerulus specific neuropeptide receptor knockdown . ( D ) The appetitive index of starved flies with glomerulus specific neuropeptide receptor knockdown . ( E , F ) Expression of NaChBac sensitizes DM1 and DM5 ( left ) , which alters appetitive behavior in opposite directions ( right ) . For imaging experiments ( n = 5–6 ) , 0 . 2% and 80% SV cider vinegar was used to stimulate the DM1 and DM5 PNs , respectively . For the behavioral experiments ( n = 53–111 ) , 5% cider vinegar was used . Error bars show s . e . m . p < 0 . 05 , p < 0 . 01 , *p < 0 . 001; t-test for the imaging results , comparing between starved and fed responses; z-test for the behavioral results comparing knockdown groups to Gal4 or UAS control groups . DOI: http://dx . doi . org/10 . 7554/eLife . 08298 . 010 To further investigate whether this starvation-dependent neuropeptide modulation is specific to DM1 and DM5 , we imaged PN responses while knocking down neuropeptide receptor genes in their corresponding ORNs—Or42b or Or85a , respectively . Knockdown of sNPFR in Or42b neurons abolished starvation-dependent sensitization in DM1 ( Figure 4C ) and reduced the appetitive index to the same extent as when the RNAi construct was expressed in most ORNs ( Figure 4D ) . Thus , reducing sNPFR dependent modulation of Or42b activity blocks the effects of starvation in enhancing both neuronal sensitivity and behavioral attraction . According to our working model , behavioral attraction at higher odor concentrations of vinegar is the sum of the opposing effects of Or42b and Or85a . We propose that removing the sNPFR modulation of Or42b does not reduce behavioral attraction to fed levels because the weight of Or42b increases when Or85a is inhibited . This working model is supported by our observation that net behavioral attraction is not completely abolished by genetic knockdown of sNPFR . Knocking down DTKR had no effect in the same neurons for which we observed phenotypes with sNPFR-RNAi . Although DTKR could in theory be absent from this population , we prefer the hypothesis Or42b responses are saturated by the 5% vinegar that we used for these behavioral experiments . Likewise , knockdown of DTKR in Or85a neurons abolished starvation-dependent suppression of DM5 and reduced food finding to the same extent as when the RNAi construct was expressed in most ORNs . Thus , reducing DTKR dependent modulation of Or85a activity blocks the effects of starvation on both neuronal sensitivity and behavioral attraction . Expression of sNPFR-RNAi in the same neurons had no effect . Taken together , these findings indicate that sNPF and DTK modulate distinct ORNs in opposite directions , in what appears to be a push–pull optimization strategy to increase the attractive valence of an odor . Based on these results , we predicted that artificial enhancement of DM1 and DM5 should shift the appetitive behavior . In particular , sensitization of DM1 in fed flies should mimic the effect of starvation , while sensitization of DM5 in starved flies should mimic the effect of satiety . To test this hypothesis , we ectopically expressed the bacterially derived sodium channel ( NaChBac ) , which makes neurons hyperexcitable ( Nitabach et al . , 2006 ) . Targeted expression of NaChBac in Or42b neurons increased the olfactory sensitivity of DM1 in fed flies and resulted in a marked increase in appetitive index ( Figure 4E ) . Likewise , expression of NaChBac in Or85a neurons increased activity of DM5 in starved flies and was accompanied by a significant decrease in appetitive index ( Figure 4F ) . Interestingly , activation of Or85a drives down levels of behavioral attraction , but does not trigger behavioral repulsion , presumably because Or42b that is still present provides competing inputs . Thus , by increasing activity in DM1 or DM5 , we were able to directly influence foraging behavior in opposite directions in a manner that mimics behavior appropriate for the corresponding metabolic state . What is the metabolic sensor for starvation to suppress DM5's response to food odor ? Previous work has implicated insulin signaling in mediating differences between rover and sitter , naturally occurring polymorphisms in the foraging gene that lead to dramatic differences in feeding behaviors ( de Belle et al . , 1989; Kent et al . , 2009 ) . We also recently reported that insulin negatively regulates sNPFR gene expression in DM1 ORNs ( Root et al . , 2011 ) . To determine whether insulin also functions as the upstream metabolic cue regulating DTKR signaling in the Or85a/DM5 neurons , we investigated the effect of insulin signaling on the olfactory sensitivity of DM5 and appetitive behavior . We first blocked insulin signaling with wortmannin , an inhibitor of PI3K ( Weinkove et al . , 1999; Root et al . , 2011 ) . Flies fed with sugar and wortmannin exhibited a suppressed DM5 response that was accompanied by an increased appetitive index ( Figure 5A ) , thereby mimicking the starved state . Moreover , this suppression occurs through increased DTKR signaling , because targeted knockdown of DTKR in ORNs blocked the effect of wortmannin . We next asked whether constitutive activation of InR prevents starvation-like physiology and behavior . Indeed , ectopic expression of a constitutively active InR ( InR-CA ) in Or85a neurons of starved flies led to increased DM5 activity and decreased appetitive behavior ( Figure 5B ) , thereby mimicking satiety . Thus , insulin signaling in Or85a neurons gates the expression of DTKR to modulate DM5 activity and appetitive behavior . 10 . 7554/eLife . 08298 . 011Figure 5 . Insulin controls DTKR signaling . ( A ) Peak ΔF/F responses of the DM5 glomerulus to cider vinegar ( left ) and the appetitive index of flies ( right ) that were fed overnight with 4% sucrose alone or sucrose with the PI3K blocker , wortmannin . DTKRi flies contained the DTKR-RNAi in Orco-Gal4 . ( B ) Peak ΔF/F responses of the DM5 glomerulus to cider vinegar ( left ) and the appetitive index of starved flies ( right ) that expressed constitutively active insulin receptor ( InR-CA ) in the Orco ORNs or selectively in Or85a neurons . For imaging experiments , PN responses to 80% SV cider vinegar were measured using GH146-LexA , LexAop-GCaMP flies . n = 5–13 for each condition . For behavior experiments , fly responses to 5% apple cider vinegar were measured . n = 67–91 for each condition . Error bars show s . e . m . p < 0 . 05 , p < 0 . 01 , *p < 0 . 001; Student's t-test ( imaging results ) and z-test for proportions ( behavioral results ) comparing wortmannin-fed group to sugar-fed group ( A ) , and comparing the InR-CA groups to the control counterpart ( B ) . DOI: http://dx . doi . org/10 . 7554/eLife . 08298 . 011 Here we demonstrate that shifts in the internal metabolic state of an animal lead to dramatic functional changes in its olfactory circuit and behaviors . Starved flies exhibit enhanced odor sensitivity in ORNs that mediate behavioral attraction and decreased sensitivity in ORNs that mediate behavioral aversion . This functional remodeling of the olfactory map is mediated by parallel neuromodulatory systems that act in opposing directions on olfactory attraction and aversion . In our earlier study , we showed that sNPFR signaling increases sensitivity in Or42b ORNs and thus enhances behavioral attraction ( Root et al . , 2011 ) . In our current study , however , we show that sNPFR signaling does not account for all changes induced by starvation in behavioral responses to a wider range of odor concentrations . Second , we show that starvation leads to a decreased sensitivity in the Or85a ORNs , an odorant channel that mediates behavioral aversion ( Semmelhack and Wang , 2009 ) . Third , we show that DTKR signaling mediates the reduced sensitivity in the Or85a ORNs and partly accounts for enhanced behavioral attraction to high concentrations of vinegar . Fourth , we show eliminating DTKR and sNPFR signaling pathways together fully reverses the effect of starvation on behavioral attraction across all odor concentrations tested . Finally we show evidence suggesting that the same global insulin signal regulating sNPFR expression may also regulate DTKR expression . In the wild , rotten fruits early in the fermentation process are more attractive to Drosophila than fresh or highly fermented fruits ( Chakir et al . , 1996; Castrezana and Markow , 2001 ) . In the laboratory , well fed flies display very little attraction to apple cider vinegar ( Root et al . , 2011 ) . Low levels of vinegar are indicative of fresh fruit of limited nutritional value . Expanding odor sensitivity to lower concentrations of potential food odors may encourage flies to accept food sources of lower value . High odor concentrations typically accompany late stages of fermentation and are often aversive or uninteresting to flies . We show that starved flies are attracted to high concentrations of vinegar partly due to neuromodulatory mechanisms that enhance sensitivity in Or42b ORNs , an attractive odor channel , and partly through neuromodulatory mechanisms that reduce sensitivity in Or85a ORNs , an aversive odor channel . In our working model , behavioral attraction to higher odor concentrations of vinegar is the sum of the opposing effects of Or42b and Or85a ( Figure 6B , C ) . When flies face starvation , the balance of these inputs shifts to favor Or42b over Or85a inputs , as mediated by selective upregulation of sNPFR and DTKR in these ORNs , respectively ( Figure 6D , E ) . These processes could serve to encourage flies to risk ingestion of potentially toxic foods when under nutritional stress . 10 . 7554/eLife . 08298 . 012Figure 6 . How starvation changes early olfactory processing . ( A ) A schematic diagram depicting anatomical locations for Or42b and Or85a ORNs in the fly antenna as well as their corresponding glomeruli , DM1 and DM5 , respectively , in the antennal lobe . LNs release DTK peptide broadly throughout the antennal lobe . ( B ) A model for how starvation state fine-tunes ORN sensitivity via the actions of neuromodulation of Or42b/DM1 and Or85a/DM5 . In the starvation state , sNPF sensitizes the DM1 glomerular responses through additive gain modulation . Tachykinin suppresses DM5 glomerular responses through a divisive gain modulation . ( C ) The concerted effect of these two neuromodulatory systems increases behavioral attraction and expands the concentration range over which attraction to vinegar manifests . ( D ) In the DM1 glomerulus , both DTK and sNPF are available and released from the LNs and ORNs , respectively , in both the fed and starved states . DTKR is also present in these terminals in the fed state . Upon starvation , loss of insulin signaling leads to selective upregulation of sNPFR expression in the Or42b ORNs , which leads to their presynaptic facilitation . ( E ) In the DM5 glomerulus , both DTK and sNPF are available and released from the LNs and ORNs , respectively , in both the fed and starved states . Upon starvation , loss of insulin signaling leads to upregulation of DTKR expression in the Or85a ORNs , leading to their presynaptic inhibition . DOI: http://dx . doi . org/10 . 7554/eLife . 08298 . 012 Given the broad array of glomeruli that can respond to odors such as vinegar ( Figure 2—source data 1 ) , it may be surprising that the modulation of only two glomeruli is sufficient to significantly impact fly behavioral attraction . Whether these findings extend to a broad array of food associated odors and whether additional glomeruli are modulated by these neuromodulatory systems remain to be determined . In this context , we note that a recent correlational analysis predicts DM5 activity is highly correlated with behavioral attraction ( Knaden et al . , 2012 ) . However , this prediction has not been confirmed by direct testing of the DM5 glomerulus in behavioral experiments and is contradicted by more recent findings ( Gao et al . , 2015 ) , as well as the data in this paper . Thus our findings suggest that in starved flies the concentration range over which vinegar odor is attractive expands in both directions , with the acute need for caloric intake apparently outweighing considerations of food quality or risk ( Figure 6C ) . This study highlights the importance of neuromodulators in shaping local neural circuit activity to accommodate the internal physiological state of an organism . The often unique expression patterns of specific GPCRs in sensory systems highlights the flexibility conferred by this evolutionarily ancient mechanism to translate neuroendocrine signals into local shifts in neuronal excitability and network properties that ultimately lead to adaptive behaviors . sNPF shares structural and functional similarities with its vertebrate homolog , NPY ( Hewes and Taghert , 2001; Lee et al . , 2004 ) . Both neuropeptides show roles in controlling food intake and feeding behaviors in insects and vertebrates . Interestingly , NPY is also expressed in the vertebrate olfactory bulb ( Hansel et al . , 2001; Mathieu et al . , 2002; Mousley et al . , 2006 ) and is thus positioned to shape olfactory processing during shifts in appetitive states as well . sNPF's broad expression pattern in the fly brain ( Nassel et al . , 2008 ) supports the possibility it is widely used to orchestrate changes across many different neuropils to shape appetitive behaviors . Indeed , sNPF and NPF , another NPY homolog in Drosophila , have been shown in the fly gustatory system to control sweet and bitter taste sensitivity , respectively , in parallel but opposing directions ( Inagaki et al . , 2014 ) . The similar changes manifested by nutritional stress in both the olfactory and gustatory systems suggests complex networks of neuromodulators may shape sensory processing of aversive and attractive inputs differentially throughout the brain in a hunger state . DTK and DTKR share homology with substance P and its receptor NK1 , respectively ( Li et al . , 1991 ) . Interestingly , they seem to share roles in shaping the processing of stressful or negative sensory cues in both flies and mammals . For example , in rodents , emotional stressors cause long-lasting release of substance P to activate NK1 in the amygdala to generate anxiety-related behavior ( Ebner et al . , 2004 ) . In Drosophila , DTK signaling has also been shown to be critical for aggressive behaviors among male flies ( Asahina et al . , 2014 ) . In previous work , we showed Drosophila tachykinin mediates presynaptic inhibition in ORNs and detected expression in the LNs ( Ignell et al . , 2009 ) . In this current study , we map the locus of DTK's effects on behavioral responses to vinegar to the Or85a/DM5 ORNs using behavior and functional imaging . We also confirm that the source of the peptide is indeed the LNs as previous anatomical data had suggested ( Ignell et al . , 2009 ) . Thus , tachykinin's role in modulating stressful sensory inputs appears to extend to a glomerulus hardwired to behavioral aversion in the olfactory system . Our results here resonate with discoveries in the gustatory system ( Inagaki et al . , 2014 ) and show that starvation changes the perception of both attractive and aversive sensory inputs beginning at the peripheral nervous system . Through the use of parallel neuromodulatory systems , the internal state of the organism functionally reconfigures early olfactory processing to optimize its detection of nutrients at the risk of ignoring potentially toxic food resources . It is certainly likely that neuromodulatory systems also impact and reconfigure central circuits in appetitive contexts . Thus , it will be of great interest to understand the contributions of peripheral and central circuits towards modifying appetitive behaviors . All Gal4- and UAS- control flies were crossed to w1118 fly strain . The following fly stocks were used: Orco-Gal4 ( Kreher et al . , 2005 ) ; Or42b-Gal4 , Or85a-Gal4 ( II ) ( Fishilevich and Vosshall , 2005 ) ; GH146-LexA ( Lai et al . , 2008 ) , LexAOp-GCaMP ( Root et al . , 2008 ) ; UAS-sNPFR-RNAi ( Lee et al . , 2008 ) ; UAS-DTKR-RNAi and GH298-Gal4 ( Ignell et al . , 2009 ) ; UAS-DTK-RNAi ( Winther et al . , 2006 ) ; UAS-InR-CA , Or42b-Gal4 ( III ) and Or85a-Gal4 ( III ) ( Bloomington stock center #8263 , #9972 and #24461 ) ; UAS-NaChBac ( Nitabach et al . , 2006 ) . Single-fly assay was used to measure the latency of food finding as previously described ( Root et al . , 2011; Zaninovich et al . , 2013 ) . Female flies that were 2–5 days old and presumed non-virgin were used for all experiments . Single flies were introduced into chambers that were 60 mm in diameter and 6 mm in height . The chamber was illuminated by 660 nm LEDs . Flies were tracked at 2 Hz with custom software written in Labview ( V . 8 . 5 , National Instruments , Austin , TX ) , and analysis was performed with Igor Pro ( V . 6 , Wavemetrics , Inc . , Portland , OR ) using a custom macro ( Root et al . , 2011; Zaninovich et al . , 2013 ) . Apple cider vinegar was diluted in 1% low melting temperature agarose . 5 μl of cider vinegar solution was placed in the center of the chamber for all experiments . A fly was counted as having found the food when it spends 5 s or longer within a 5 mm radius of the center . The elapsed time before an individual fly reached the odor target was also recorded . All control flies were crossed with w1118 flies . Flies were starved with water for 16–24 hr prior to experiments . About 50 flies ( w1118;+;Orco-Gal4/+ ) of both sexes were kept in each vial for 3 days . Female flies were then transferred to a new food vial ( control fed flies ) or a vial with a Kimwipe saturated by water ( starved flies ) . 12 hr later , antennae were collected from these female flies . Dissection was performed in the morning at the same time to minimize circadian difference . Antennae from 200 flies were collected for each condition and total RNA was extracted using Trizol ( Invitrogen , Carlsbad , CA ) . Libraries were prepared using Illumina's mRNA sequencing kit and further purified using AMPure XP beads ( Agencourt ) . Sequencing was performed at UCSD's BIOGEM facility on an Illumina GA2 sequencer . For each of the biological conditions , over 80 million 36 bp reads were generated from two lanes . Reads were aligned to the Drosophila genome ( dm3 assembly ) using TopHat ( Trapnell et al . , 2009 ) , allowing up to three mismatches with the reference sequence . Transcripts were then assembled against FlyBase ( release 5 . 39 ) gene annotations and their abundances were calculated using Cufflinks ( Trapnell et al . , 2010 ) . In total , over 136 million reads were mapped to protein-coding genes . For differential expression analysis , raw gene counts were generated using HTSeq ( Anders , 2010 ) software and then normalized for the difference in sequencing depth between the two conditions . Probability values were calculated on raw counts using the Fisher's exact test as computed by the edgeR package ( Robinson et al . , 2010 ) ( R software environment ) . GCaMP imaging was performed as previously described ( Wang et al . , 2003; Root et al . , 2008 ) . In odor experiments , a constant airflow of 1 l/min was applied to the antennae via a pipe of 12 mm diameter . Odor onset was controlled by mixing a defined percentage of carrier air with air redirected through odor bottles as previously described ( Root et al . , 2008; Semmelhack and Wang , 2009 ) . Nerve stimulation was performed with a glass suction electrode and an S48 stimulator ( Grass , Warwick , RI ) as previously described ( Wang et al . , 2003; Root et al . , 2008 ) . Stimulation was 1 ms in duration , 10 V in amplitude , and 16 pulses ( Figure 4A ) and 45 pulses ( Figure 4B ) at 100 Hz . Starved flies were starved with water for 16–24 hr . sNPF peptide , AQRSPSLRLRF-NH2 , 98% purity ( Celtek Peptides , Franklin , TN ) and DTK peptide , APTSSFIGMR-NH2 , 98% purity ( Bio Basic Inc . , Markham , Ontario , Canada ) were each dissolved in saline to a final concentration of 10 μM . Wortmannin ( LC Laboratories , Woburn , MA ) was dissolved in DMSO at stock concentrations of 10 mM . Flies were fed overnight with 200 μl of 4% sucrose solution , or plus 25 nM wortmannin .	Animals typically need to forage for their food , but doing so is not without risk . Foraging can expose an animal to predators and harmful toxins . Many animals use odors and other chemical signals to help them locate food or to avoid harm . In some animals , such as fruit flies , different parts of the nervous system are hardwired to encourage individuals to move towards attractive odors or away from unpleasant ones . Fruit flies feed on the yeast that grows on decaying fruit . They do so by ignoring fresh fruits ( which have very little yeast ) and avoiding overly-rotten fruits ( which might contain toxic chemicals ) . To determine ripeness , flies use a fruit's vinegar levels: fresh fruits contain low levels of vinegar , while fermented fruits have high levels . Previous studies using low levels of vinegar have shown that well-fed flies largely ignore the scent , while starving flies are attracted to it . Ko et al . have built on the results of previous studies and now report that starving fruit flies are much less sensitive to unfavorable odors in high levels of vinegar and much more sensitive to favorable odors in low levels of vinegar . This behavior is due to two neuropeptides ( molecules that carry signals between neurons ) that have opposite effects on different parts of the fly's nervous system . One of the neuropeptides made the groups of neurons that respond to attractive odors more responsive , while the other suppressed the activity of neurons that normally respond to unpleasant odors . Together these changes could encourage the animals to take more risks when they are hungry , by suppressing of their ability to recognize noxious or harmful chemicals in favor of their ability to perceive attractive odors . The effect of both neuropeptides is triggered by the insulin hormone , which carries information about the metabolic state ( for example , whether it is starving or well-fed ) throughout the whole animal . Thus , individual neurons may read the same metabolic signals and then respond in different ways to fine-tune the activity of nearby circuits of neurons to alter foraging behavior in a coordinated manner . Furthermore , it is almost certain that similar changes to the sensory system could affect an animal's appetite for food . One of the next challenges will be to attempt to understand if and how appetite in humans might be controlled in a similar way .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "neuroscience" ]	2015	Starvation promotes concerted modulation of appetitive olfactory behavior via parallel neuromodulatory circuits
Translational repression and mRNA degradation are critical mechanisms of posttranscriptional gene regulation that help cells respond to internal and external cues . In response to certain stress conditions , many mRNA decay factors are enriched in processing bodies ( PBs ) , cellular structures involved in degradation and/or storage of mRNAs . Yet , how cells regulate assembly and disassembly of PBs remains poorly understood . Here , we show that in budding yeast , mutations in the DEAD-box ATPase Dhh1 that prevent ATP hydrolysis , or that affect the interaction between Dhh1 and Not1 , the central scaffold of the CCR4-NOT complex and an activator of the Dhh1 ATPase , prevent PB disassembly in vivo . Intriguingly , this process can be recapitulated in vitro , since recombinant Dhh1 and RNA , in the presence of ATP , phase-separate into liquid droplets that rapidly dissolve upon addition of Not1 . Our results identify the ATPase activity of Dhh1 as a critical regulator of PB formation . Rapid modulation of gene expression is critical for cells to respond to environmental challenges and to initiate developmental programs . Eukaryotic cells have developed a variety of mechanisms to achieve tight regulation of gene expression . This includes post-transcriptional control of messenger RNA ( mRNA ) levels by the regulation of translation or by varying the rates of mRNA degradation . Many of these post-transcriptional regulatory mechanisms , including the transition from mRNA translation to storage or decay , are not well characterized . Cytoplasmic mRNAs are marked by a 7-methylguanosine cap at the 5’ end and by a polyA tail at the 3’ end . These modifications enable interaction with translation factors , including the cap-binding complex ( eIF4F ) and the polyA binding protein ( Pab1 ) and protect the mRNA against degradation ( Coller and Parker , 2004 ) . Given their impact on both translation and mRNA decay , the status of the 5’ and 3’ ends of the mRNA , as well as the complement of proteins that bind the mRNA termini , are tightly controlled . In budding yeast , a key event for the entry of mRNAs into the degradation pathway is the removal of the polyA tail ( Muhlrad and Parker , 1992 ) , which is predominantly accomplished by the CCR4-NOT complex ( Wiederhold and Passmore , 2010 ) . While deadenylated mRNAs can also be degraded from the 3’ end by the 10-subunit exosome complex ( Chlebowski et al . , 2013 ) , mRNA decay in S . cerevisiae occurs predominantly via removal of the 5’ cap by the Dcp1-Dcp2 decapping enzyme , followed by degradation by the 5’-3’ exonuclease , Xrn1 ( Garneau et al . , 2007; Sun et al . , 2013 ) . Under certain stress conditions , such as glucose starvation or osmotic shock , protein factors involved in mRNA turnover can assemble into larger mRNP foci , known as processing bodies ( PBs ) ( Sheth and Parker , 2003; Teixeira et al . , 2005 ) . PBs are dynamic , membrane-less structures that appear to form from multivalent interactions between proteins and RNA in a liquid-liquid phase separation phenomenon ( Decker et al . , 2007; Fromm et al . , 2012 , 2014; Guo and Shorter , 2015 ) . Remarkably , PBs and several other related types of mRNP granules , including stress granules , germ granules , and neuronal transport granules , form in a number of different species and cell types , and in a variety of different biological contexts , suggesting these structures are important for cellular function ( Erickson and Lykke-Andersen , 2011; Kiebler and Bassell , 2006; Voronina , 2013 ) . There is increasing evidence that the ability to form PBs is critical for survival under various stress conditions . For example , cells unable to form PBs show a severe loss in viability in stationary phase ( Ramachandran et al . , 2011; Shah et al . , 2013 ) . Furthermore , ectopic expression of highly expressed mRNAs in cells that cannot form PBs is toxic ( Lavut and Raveh , 2012 ) . Because of their composition , PBs are postulated to be sites of mRNA storage and/or mRNA degradation ( Aizer et al . , 2014; Anderson and Kedersha , 2009; Decker and Parker , 2012 ) . Yet , how the cell regulates PB assembly and disassembly , and how PBs modulate gene expression , has remained elusive . It is likely that PB formation requires factors that can either remodel the translating mRNP complex or stimulate the formation of a decay-competent or repressed mRNP . The DEAD-box ATPase Dhh1 stimulates mRNA decay and translation repression ( Carroll et al . , 2011; Coller and Parker , 2005; Fischer and Weis , 2002; Sweet et al . , 2012 ) and is thought to function at an early step in PB formation ( Teixeira and Parker , 2007 ) , making it a good candidate to facilitate mRNA inactivation . Similar to other DEAD-box proteins , Dhh1 possesses N- and C-terminal RecA-like domains connected by a short linker , and can bind RNA with high affinity in a sequence-independent manner through the phosphate backbone ( Cheng et al . , 2005; Linder and Jankowsky , 2011; Russell et al . , 2013 ) . In vitro , Dhh1 has a significantly lower ATPase activity than other well characterized DEAD-box proteins such as eIF4A or Ded1 ( Cordin et al . , 2006; Dutta et al . , 2011; Pause and Sonenberg , 1992; Tritschler et al . , 2009 ) . This is likely due to intramolecular interactions between its N- and C-terminal RecA lobes that hold Dhh1 in a conformation that is not competent for ATP hydrolysis ( Cheng et al . , 2005; Sharif et al . , 2013 ) suggesting the ATPase activity of Dhh1 is stimulated by factors that can alter the conformation of its two RecA domains . Several recent studies have revealed that DEAD-box proteins can be stimulated or inhibited by trans-acting factors . These interacting partners appear to share a common 3D architecture , namely , the presence of a MIF4G fold – a highly alpha helical HEAT repeat-like structure found in a number of different DEAD-box-interacting proteins , including eIF4G ( with eIF4A ) , Gle1 ( with Dbp5 ) and CWC22 ( with eIF4AIII in higher eukaryotes ) ( Buchwald et al . , 2013; Montpetit et al . , 2011; Ozgur et al . , 2015a; Schütz et al . , 2008 ) . Intriguingly , DDX6 , the mammalian homolog of Dhh1 , binds directly to CNOT1 ( Not1 in S . c . ) , the central scaffold subunit of the CCR4-NOT deadenylase complex , through its MIF4G domain ( Chen et al . , 2014; Mathys et al . , 2014 ) and CNOT1 binding activates the ATPase of DDX6 ( Mathys et al . , 2014 ) . The binding surface between these two proteins is conserved between yeast and human ( Rouya et al . , 2014 ) suggesting that the interaction between Not1 and Dhh1 is also important for modulating the activity of Dhh1 in budding yeast . In this study , we examine the ATPase activity of Dhh1 in vitro and in vivo , and demonstrate that the ATPase cycle of Dhh1 is a critical regulator of PB nucleation and disassembly . Cells expressing a Dhh1 variant carrying a mutation in the DEAD motif ( E195Q , or Dhh1DQAD ) that disrupts ATP hydrolysis form constitutive granules with both the behavior and composition of PBs induced during glucose starvation . Using recombinant proteins , we show that Not1 stimulates the ATPase activity of yeast Dhh1 , similar to its function in mammals . Disruption of the interaction between Dhh1 and Not1 in vivo leads to the formation of PBs in the absence of stress , similar to the catalytically dead Dhh1DQAD allele . Furthermore , we demonstrate that Dhh1 , ATP , and RNA , are sufficient to form liquid droplets in vitro with the dynamic behavior of PBs , and that these droplets can be dissolved by addition of purified Not1 . Overall , these results reveal that the ATPase activity of Dhh1 is a critical regulator of PB dynamics . Previously , our lab demonstrated that abrogation of the ATPase activity of Dhh1 through mutation of the conserved DEAD motif ( Dhh1E195Q , henceforth Dhh1DQAD; see Supplementary file 2C for a list of all Dhh1 mutants in this study ) mislocalizes Dhh1 to large Dcp2-positive foci in the absence of stress ( Carroll et al . , 2011 ) . To differentiate whether loss of Dhh1 ATPase activity triggers formation of genuine processing bodies or whether these Dhh1DQAD-induced foci are anomalous granules , we monitored the localization of several PB components – namely Dcp1 , Edc3 , and Xrn1 – in both DHH1 and dhh1DQAD mutant cells . Similar to the PB composition in glucose starvation conditions , all three GFP-tagged proteins colocalized with Dcp2-mCherry in Dhh1DQAD-expressing cells in glucose-rich conditions ( Figure 1A ) . In contrast , the stress granule marker Pab1 did not assemble into foci in dhh1DQAD cells ( Figure 1—figure supplement 1A ) demonstrating that Dhh1DQAD granules are composed of proteins found in bona fide PBs . 10 . 7554/eLife . 18746 . 003Figure 1 . Loss of the ATPase activity of Dhh1 triggers bona fide processing body ( PB ) formation . ( A ) Known PB components localize to Dhh1DQAD foci . Cells co-expressing the indicated PB component were grown to exponential growth phase , then shifted to glucose-rich or glucose starvation conditions for 20 min and observed by confocal microscopy . Scale bar: 5 μm ( B ) Constitutive PB formation by Dhh1DQAD is rescued by the presence of wild-type Dhh1 . Dhh1-GFP or Dhh1DQAD-GFP was expressed from a CEN plasmid in DHH1 or dhh1∆ cells and were treated as in ( A ) . Scale bar: 5 μm ( C ) Loss of ATPase activity mildly disrupts degradation of a Dhh1-tethered mRNA . Dhh1 or Dhh1DQAD was co-expressed as a PP7CP fusion protein in dhh1∆ cells expressing FBA1-PP7L . FBA1 mRNA levels were measured by qPCR and normalized to ACT1 mRNA . Graphs show mean mRNA levels from three independent experiments of biological triplicate samples . Error bars represent SD . A student’s t-test comparing Dhh1 and Dhh1DQAD is shown . Asterisks indicate p<0 . 005 . ( D ) FBA1 mRNAs do not colocalize with PBs in Dhh1 or Dhh1DQAD-expressing cells , suggesting functional mRNA decay , but enrich in PBs in xrn1∆ cells . The indicated strains were grown to exponential growth phase , shifted to either glucose-rich ( 2% glucose ) or glucose starvation conditions for 20 min , fixed with paraformaldehyde and processed for smFISH . Depicted is a maximum projection of the central 10 planes of a 3D image . Insets show representative cells ( 1 . 67X magnification ) . The graph shows the quantification of a representative experiment ( n = 2 biological replicates ) . Scale bar: 5 µm . ( E ) GFA1 mRNAs do not colocalize with PBs in Dhh1 or Dhh1DQAD-expressing cells , suggesting functional mRNA decay , but enrich in PBs in xrn1∆ cells . The indicated strains were grown to exponential growth phase , shifted to either glucose-rich ( 2% glucose ) or glucose starvation conditions for 20 min , fixed with paraformaldehyde and processed for smFISH as in ( D ) . Insets show representative cells ( 1 . 67X magnification ) . The graph shows the quantification of a representative experiment ( n = 2 biological replicates ) . Scale bar: 5 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 18746 . 00310 . 7554/eLife . 18746 . 004Figure 1—figure supplement 1 . Loss of ATPase activity of Dhh1 does not trigger stress granule formation . ( A ) Dhh1DQAD does not trigger constitutive stress granule formation . Cells co-expressing Pab1-GFP were grown to exponential growth phase , then shifted to glucose-rich or glucose starvation conditions for 20 min and observed by confocal microscopy . Scale bar: 5 μm ( B ) PAT1 mRNAs do not colocalize with PBs in Dhh1 or Dhh1DQAD-expressing cells , suggesting functional mRNA decay , but enrich in PBs in xrn1∆ cells . The indicated strains were grown to exponential growth phase , shifted to either glucose-rich or glucose starvation conditions for 20 min , fixed with paraformaldehyde and processed for smFISH . Depicted is a maximum projection of the central 10 planes of a 3D image . Insets show representative cells ( 1 . 67X magnification ) . Scale bar: 5 µm . ( C ) PGK1 mRNAs do not colocalize with PBs in Dhh1 or Dhh1DQAD-expressing cells , suggesting functional mRNA decay , but enrich in PBs in xrn1∆ cells . The indicated strains were grown to exponential growth phase , shifted to either glucose-rich or glucose starvation conditions for 20 min , fixed with paraformaldehyde and processed for smFISH as in ( A ) . Insets show representative cells ( 1 . 67X magnification ) . ale bar: 5 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 18746 . 004 Despite their identification nearly 15 years ago , the precise functional role of PBs in S . cerevisiae remains poorly understood . Therefore , it is unclear whether PB formation in cells expressing dhh1DQAD is caused by a loss or gain of Dhh1 function . If Dhh1DQAD PB formation is caused by a loss of Dhh1 function , then the presence of a wild-type copy of DHH1 should abolish constitutive granule formation . To test this , we expressed wild-type Dhh1-GFP and Dhh1DQAD-GFP in either DHH1 or dhh1∆ cells in glucose-rich conditions and observed the localization of Dhh1DQAD-GFP ( Figure 1B ) . While Dhh1DQAD-GFP – but not Dhh1-GFP – robustly formed PBs in dhh1∆ cells , PBs were no longer present in cells expressing an additional DHH1 copy , indicating that Dhh1DQAD PB formation is a recessive phenotype , and that the presence of enzymatically active Dhh1 is sufficient to prevent PB formation . Given that Dhh1DQAD PB formation appeared to be due to a loss-of-function rather than a gain-of-function , one possible explanation for the constitutive formation of PBs could be a block in mRNA decay in dhh1DQADcells , similar to dcp1∆ or xrn1∆ cells ( Teixeira and Parker 2007; Sheth and Parker 2003 ) . In order to directly interrogate whether loss of the ATPase activity of Dhh1 disrupted mRNA turnover , we tested the functionality of Dhh1DQAD in mRNA decay using a previously established tether-based functional assay . We and others have observed that tethering Dhh1 to a reporter mRNA using the bacteriophage PP7 or MS2 systems is sufficient to stimulate the decay of a tethered mRNA ( Carroll et al . , 2011; Sweet et al . , 2012 ) . We expressed wild-type Dhh1 , Dhh1DQAD , or GFP as a PP7 coat protein ( PP7CP ) fusion protein in dhh1∆ cells containing a single stem loop ( PP7L ) engineered into the 3’UTR of the FBA1 gene ( Figure 1C ) and assessed steady state mRNA levels by qPCR . As expected , tethering Dhh1-PP7CP to FBA1 mRNA caused an 80% reduction of FBA1 mRNA levels compared with GFP-tethered mRNA ( Carroll et al . , 2011 ) . In comparison , tethering Dhh1DQAD showed a partial attenuation of mRNA decay , with FBA1 levels decreasing by 55% , ( Figure 1C ) ( Carroll et al . , 2011 ) , indicating that Dhh1DQAD is capable of stimulating mRNA decay . While tethering Dhh1DQAD to an mRNA demonstrated that this variant can function in mRNA decay , it does not address whether Dhh1DQAD PBs can degrade mRNAs . Therefore , we performed single molecule mRNA fluorescence in situ hybridization ( smFISH ) to examine if Dhh1DQAD PBs show hallmarks of mRNA decay . Log-phase DHH1 , dhh1DQAD , and xrn1∆ cells were shifted into glucose starvation media , and the mRNA localization of FBA1 , an essential glycolytic gene , was analyzed ( Figure 1D ) . In xrn1∆ cells , 54% of FBA1 mRNAs colocalized with a Dcp2-GFP PB marker in cells grown in glucose-rich conditions and this colocalization was further increased to 75% following glucose starvation . In contrast , only 11% of FBA1 mRNAs colocalized with PB foci in glucose-starved cells expressing wild-type DHH1 , consistent with the notion that PBs are sites of mRNA decay , rather than mRNA storage . In cells expressing dhh1DQAD , FBA1 mRNA showed a modest overlap with Dcp2-GFP – around 20% in glucose-rich conditions , and 22% following glucose starvation . Similar results were obtained in smFISH experiments with mRNAs coding for GFA1 , which functions in chitin biosynthesis ( Lagorce et al . , 2002 ) ( Figure 1E ) , PAT1 , which is involved in mRNA decapping ( Bonnerot et al . , 2000 ) ( Figure 1—figure supplement 1B ) , and the phosphoglycerate kinase PGK1 ( Hitzeman et al . , 1980 ) ( Figure 1—figure supplement 1C ) in DHH1 , dhh1DQAD , and xrn1∆ cells . Our tethering experiments , together with the difference in mRNA accumulation between Dhh1DQAD PBs and PBs in xrn1∆ cells , suggest that Dhh1DQAD PB formation is likely not due to a complete failure to degrade mRNAs . However , some mRNAs show slower turnover in the presence of Dhh1DQAD ( Carroll et al . , 2011 ) . Therefore , reduced decay kinetics may cause mRNAs to persist for longer in PBs , which may in part contribute to the formation of Dhh1DQAD PBs in the absence of stress . Our data so far indicate that loss of ATPase activity by Dhh1 triggers formation of bona fide PBs , suggesting that Dhh1 is ATP bound in PBs . We therefore tested whether ATP binding is required for PB localization of Dhh1 . Wild-type Dhh1 or a previously characterized ATP-binding mutant of Dhh1 ( Dhh1F66R , Q73A , henceforth Dhh1Q-motif ) were co-expressed along with Dcp2-mCherry and localization was monitored in glucose-rich or glucose starvation conditions . Dhh1Q-motif showed a strong defect in PB formation ( Figure 2A ) , consistent with prior observations ( Dutta et al . , 2011 ) , and also as evidenced by a reduction in PB localization of Dcp2 , Xrn1 , Dcp1 , and Edc3 ( Figure 2—figure supplement 1A–C ) demonstrating that ATP binding by Dhh1 is required for robust PB formation . We also tested the functionality of Dhh1Q-motif using our tethering assay , and saw that Dhh1Q-motif did not show any obvious defects in mRNA decay when tethered to FBA1 mRNA ( Figure 2—figure supplement 2A ) . 10 . 7554/eLife . 18746 . 005Figure 2 . ATP-bound , RNA-bound Dhh1 is required for robust PB formation . ( A ) Disruption of ATP-binding activity of Dhh1 interferes with PB formation . Wild-type Dhh1 or Dhh1Q-motif was co-expressed from a plasmid as a GFP fusion protein in dhh1∆ cells along with Dcp2-mCherry as a PB marker and grown to exponential growth phase , then shifted to either glucose-rich or glucose starvation conditions for 20 min and observed by confocal microscopy . Images were also acquired using wide-field microscopy and PB formation was quantified using Diatrack 3 . 5 particle tracking software ( see Materials and methods ) . Graphs represent average Dhh1-GFP or Dcp2-mCherry foci number per cell ( n=3 biological replicates , >800 cells per experiment ) . Error bars represent SD . A student’s t-test comparing Dhh1 and Dhh1Q-motif is shown . Asterisks indicate p<0 . 005 . Scale bar: 5 µm . ( B ) Disruption of RNA binding activity of Dhh1 interferes with PB formation . Wild-type or mutant Dhh1 was co-expressed from a plasmid as a GFP fusion protein in dhh1∆ cells along with Dcp2-mCherry as a PB marker and treated as in ( A ) . Scale bar: 5 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 18746 . 00510 . 7554/eLife . 18746 . 006Figure 2—figure supplement 1 . Loss of ATP binding and RNA binding by Dhh1 disrupts PB localization of other PB factors . Xrn1-GFP ( A ) , Dcp1-GFP ( B ) , or Edc3-GFP ( C ) was co-expressed in dhh1∆ cells along with pDHH1-TAP , pDHH1Q-motif-TAP , or pDHH13X-RNA-TAP and Dcp2-mCherry as a PB marker and treated as in Figure 2A . A student’s t-test comparing DHH1 and either dhh1Q-motif or dhh13X-RNA is shown . Asterisks indicate p<0 . 005 ( *** ) , or p<0 . 05 ( * ) . Scale bar: 5 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 18746 . 00610 . 7554/eLife . 18746 . 007Figure 2—figure supplement 2 . Disruption of RNA-binding , but not ATP-binding , affects the ability of tethered Dhh1 to promote mRNA decay . ( A ) Dhh1Q-motif is functional in mRNA decay when tethered to an mRNA . Dhh1 or Dhh1Q-motif was co-expressed as a PP7CP fusion protein in dhh1∆ cells expressing FBA1-PP7L . FBA1 mRNA levels were measured by qPCR and normalized to ACT1 mRNA . Graphs show mean mRNA levels from three independent experiments of biological triplicate samples . Error bars represent SD . A student's t-test comparing Dhh1 and Dhh1Q-motif is shown . ( B ) Loss of RNA-binding activity disrupts degradation of a Dhh1-tethered mRNA . Dhh1 or Dhh13X-RNA was co-expressed as a PP7CP fusion protein in dhh1∆ cells expressing FBA1-PP7L . FBA1 mRNA levels were measured by qPCR and normalized to ACT1 mRNA . Graphs show mean mRNA levels from three independent experiments of biological triplicate samples . Error bars represent SD . A student’s t-test comparing Dhh1 and Dhh13X-RNA is shown . Asterisks indicate p<0 . 005 . ( C ) Wild-type and Dhh1 mutant proteins are expressed to similar levels . Western blot of Dhh1 and Dhh1 mutant protein expression from cells in exponential growth phase from Figure 2A and B . Dhh1 was detected using an anti-Dhh1 antibody . Hxk1 was used as a loading control . DOI: http://dx . doi . org/10 . 7554/eLife . 18746 . 007 How does the catalytic activity of Dhh1 contribute to PB formation ? Given that ATP-bound Dhh1 binds mRNA in a sequence-independent manner with nanomolar affinity ( Dutta et al . , 2011; Ernoult-Lange et al . , 2012 ) , one plausible model is that Dhh1DQAD binds to mRNA , but is unable to dissociate from it in the absence of ATP hydrolysis , ultimately leading to constitutive PB formation . To test this possibility , we generated an RNA binding mutant of Dhh1 , Dhh13X-RNA , with alanine substitutions at three residues in the C-terminal RecA domain that are important for RNA binding ( R322A , S340A , R370A ) ( Dutta et al . , 2011 ) . Wild-type Dhh1 , Dhh1DQAD , Dhh13X-RNA , and a Dhh1DQAD/3X-RNA-GFP double mutant were co-expressed with Dcp2-mCherry in glucose-rich conditions and PB formation was monitored . While Dhh1DQAD cells formed PBs as expected ( Figure 2B , left panel ) , combining ATPase-dead and RNA-binding mutations in cis in the Dhh1DQAD/3X-RNA mutant abolished constitutive PB formation . In addition , both Dhh13X-RNA and Dhh1DQAD/3X-RNA mutants showed a strong reduction of PB formation in glucose starvation conditions ( Figure 2B , right panel ) . In addition , several other PB components showed strong defects in PB localization in cells expressing dhh13X-RNA ( Figure 2—figure supplement 1A–C ) . Notably , all Dhh1 mutant proteins were expressed to similar levels in these experiments ( Figure 2—figure supplement 2C ) . Next , we examined whether disruptions in RNA binding by Dhh1 also affected Dhh1 function in mRNA decay using our tethering assay ( Figure 2—figure supplement 2B ) . Dhh13X-RNA caused only a ~30% reduction in tethered FBA1 mRNA levels , demonstrating that disruption of the mRNA decay activity of Dhh1 per se is not sufficient to trigger PB formation . Overall , we conclude that Dhh1 in its ATP-bound state promotes PB formation and that PB assembly requires RNA binding by Dhh1 . The requirements of ATP and RNA binding by Dhh1 for robust PB formation would predict that deletion of DHH1 should also cause a reduction in PB formation . However , previous reports suggested that dhh1∆ cells did not show strong defects in PB assembly ( Buchan et al . , 2008; Teixeira and Parker , 2007 ) . To carefully assess the effects of the deletion of DHH1 on PB formation , we utilized the Diatrack particle tracking software ( Vallotton and Olivier , 2013 ) which allows for an unbiased , automated , and accurate quantitation of foci formation ( see Materials and methods ) . Analysis of greater than 1000 cells per experiment revealed a nearly 80% reduction of the Dcp2-mCherry foci number per cell in dhh1∆ compared to wild-type cells during glucose starvation ( Figure 2B ) . Together , our results demonstrate that Dhh1 is required for robust PB formation . Given that loss of Dhh1 ATPase activity drives constitutive PB assembly and abolishes Dhh1 recycling from PBs ( Carroll et al . , 2011 ) , we asked if Dhh1DQAD also affects the dynamic localization of other mRNA decay factors to PBs . To address this question , we performed fluorescence recovery after photobleaching ( FRAP ) experiments ( Figure 3A ) . Cells expressing either wild-type Dhh1 or Dhh1DQAD were shifted to glucose-free media to allow PBs to form , and the recovery of GFP-tagged mRNA decay factors within photobleached PBs was measured over time . Consistent with our previous work , Dhh1-GFP PB fluorescence recovered to roughly 80% within 1 min , while the Dhh1DQAD-GFP signal did not , suggesting that Dhh1 ATP hydrolysis is required for Dhh1 to shuttle in and out of PBs ( Carroll et al . , 2011 ) . In contrast , the dynamics of several mRNA decay factors , namely Dcp1 , Dcp2 , Edc3 , and Xrn1 , remained unchanged in cells expressing Dhh1DQAD compared with Dhh1 . The mRNA decay factors observed showed two distinct FRAP profiles: Xrn1-GFP showed dynamic PB localization in DHH1 and dhh1DQAD cells , while Dcp1 , Dcp2 , and Edc3 showed a static PB localization profile . The limited recovery seen for Dcp1 and Dcp2 is in agreement with previous FRAP measurements in mammalian cells ( Aizer et al . , 2008 , 2014 ) , and indicates that these factors are likely resident PB proteins . Thus , with the exception of Dhh1 itself , the dynamics of all tested PB components were not significantly altered by the loss of Dhh1’s ATPase activity . 10 . 7554/eLife . 18746 . 008Figure 3 . Loss of the ATPase activity of Dhh1 disrupts PB dynamics . ( A ) Loss of the ATPase activity of Dhh1 does not alter the dynamics of known PB components . Fluorescence recovery after photobleaching experiments ( FRAP ) were performed on cells expressing the indicated GFP-tagged PB component . Cells were glucose starved for 20 min to allow PBs to form , then PBs were bleached and recovery of GFP fluorescence to the PB was followed over time . Recovery of PB components is presented as an averaged data plot of FRAP recovery curves from three independent experiments ( n > 8 PBs per experiment , typically ~12 PBs per experiment ) . Error bars represent SD . ( B ) The ATPase activity of Dhh1 is required for proper PB disassembly . dhh1∆ cells expressing Dhh1-GFP or Dhh1DQAD-GFP were glucose starved for 30 min to allow PBs to form and then treated with either 50 μg/mL cycloheximide or solvent only ( DMSO ) for 2 hr and disappearance of Dhh1-GFP or Dhh1DQAD-GFP foci per cell was monitored for 2 hr . Each time point image is a maximum-projection of 8 z-stacks at a distance of 0 . 4 µm . The graph shows foci number per cell measurements for Dhh1 and Dhh1DQAD normalized to 1 to account for differences in PB formation between Dhh1 and Dhh1DQAD ( n = 3 biological replicates , >100 cells ) . Error bars represent SEM . Scale bar: 5 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 18746 . 00810 . 7554/eLife . 18746 . 009Figure 3—figure supplement 1 . Loss of ATPase activity of Dhh1 disrupts the PB dynamics of other PB components . ( A ) Loss of ATPase activity of Dhh1 affects disassembly of Dcp2 foci . Dcp2-mCherry was expressed in dhh1∆ cells co-expressing Dhh1-GFP or Dhh1DQAD-GFP from a plasmid , and cells were glucose starved for 30 min to allow PBs to form , followed by treatment with either 50 μg/mL cycloheximide or solvent only ( DMSO ) for 2 hr . Disappearance of Dcp2-mCherry foci over time is monitored ( n = 3 biological replicates ) . Each time point image is a maximum-projection of 8 z-stacks at a distance of 0 . 4 µm . Scale bar: 5 µm ( B ) Two-hour cycloheximide treatment does not disrupt cell viability . The indicated cells were glucose starved for 30 min , and then treated with or without 50 μg/mL cycloheximide and 5-fold serial dilutions were plated on SD ( -URA ) + 2% dextrose media . DOI: http://dx . doi . org/10 . 7554/eLife . 18746 . 009 FRAP experiments allowed characterization of Dhh1DQAD and wild-type Dhh1 PB dynamics on a sub-minute time scale . To examine the dynamicity of these granules over a longer period , we also treated cells with cycloheximide , which disrupts PB formation , likely by trapping mRNAs on polysomes and preventing their entry into PBs ( Kroschwald et al . , 2015; Teixeira et al . , 2005 ) . Cells expressing either Dhh1-GFP or Dhh1DQAD-GFP and Dcp2-mCherry were grown to mid-log phase and shifted to glucose-free media for 30 min to allow PBs to form , and then treated with cycloheximide for up to 2 hr and PB disassembly was monitored over time ( Figure 3B , Figure 3—figure supplement 1A ) . While Dhh1-GFP showed roughly 60% disassembly of PBs after 20 min following cycloheximide treatment versus no treatment or solvent-only ( Figure 3B , Videos 1 and 2 ) , Dhh1DQAD PB disassembly occurred significantly slower , with 60% disassembly occurring around 80 min after cycloheximide treatment ( Figure 3B , Videos 3 and 4 ) . Notably , 2 hr cycloheximide treatment did not adversely affect cell viability ( Figure 3—figure supplement 1B ) . The disassembly of Dhh1DQAD PBs following cycloheximide treatment suggests that these structures , like wild-type PBs , are RNA-dependent structures . However , the slower disassembly kinetics of Dhh1DQAD PBs , coupled with the dampened recycling of Dhh1DQAD , indicates that ATPase activity of Dhh1 is critical for normal PB disassembly , for instance , by facilitating release of Dhh1 from its mRNA client . 10 . 7554/eLife . 18746 . 010Video 1 . Cycloheximide treatment causes wild-type PB disassembly . dhh1∆ cells expressing Dhh1-GFP from a plasmid were glucose starved for 30 min to allow PBs to form , and were then treated with 50 µg/mL cycloheximide and disappearance of Dhh1-GFP foci was monitored ( 5 min intervals; movie played at 5 fps ) . Each frame represents a maximum-projection of 8 z-stacks at a distance of 0 . 4 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 18746 . 01010 . 7554/eLife . 18746 . 011Video 2 . DMSO treatment does not trigger wild-type PB disassembly . dhh1∆ cells expressing Dhh1-GFP from a plasmid were glucose starved for 30 min to allow PBs to form , and were then mock treated with DMSO and disappearance of Dhh1-GFP foci was monitored ( 5 min intervals; movie played at 5 fps ) . Each frame represents a maximum-projection of 8 z-stacks at a distance of 0 . 4 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 18746 . 01110 . 7554/eLife . 18746 . 012Video 3 . Dhh1DQAD PBs disassemble more slowly than wild-type PBs following cycloheximide treatment . dhh1∆ cells expressing Dhh1DQAD-GFP from a plasmid were glucose starved for 30 min to allow PBs to form , and were then treated with 50 µg/mL cycloheximide and disappearance of Dhh1DQAD-GFP foci was monitored ( 5 min intervals; movie played at 5 fps ) . Each frame represents a maximum projection of 8 z-stacks at a distance of 0 . 4 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 18746 . 01210 . 7554/eLife . 18746 . 013Video 4 . DMSO treatment does not trigger Dhh1DQAD PB disassembly . dhh1∆ cells expressing Dhh1DQAD-GFP from a plasmid were glucose starved for 30 min to allow PBs to form , and were then mock treated with DMSO and disappearance of Dhh1DQAD-GFP foci was monitored ( 5 min intervals; movie played at 5 fps ) . Each frame represents a maximum projection of 8 z-stacks at a distance of 0 . 4 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 18746 . 013 So far , our data reveal that the ATPase cycle of Dhh1 is a critical regulator of PB dynamics , and that Dhh1 in its ATP-bound state promotes PB formation . Interestingly , Dhh1 alone is a very poor ATPase in vitro ( Dutta et al . , 2011; Tritschler et al . , 2009 ) . However , DDX6 , the mammalian homolog of Dhh1 , can be stimulated by CNOT1 , the central scaffold subunit of the CCR4-NOT complex ( Mathys et al . , 2014 ) . Based on our data , we would therefore predict that Not1 should promote the disassembly of PBs by stimulating the ATPase cycle of Dhh1 . To test this prediction , we first examined whether S . cerevisiae Not1 , like its mammalian homolog , stimulated the ATPase activity of Dhh1 in vitro . We recombinantly expressed and purified full-length Dhh1 and Dhh1DQAD and performed in vitro ATPase assays to assess the enzymatic activity of Dhh1 in the presence or absence of polyU RNA and recombinant Not1MIF4G ( amino acids 754–1000 ) . Similar to previous observations , we could not detect an intrinsic ATPase activity for Dhh1 alone . Dhh1 was weakly stimulated by polyU RNA ( Figure 4—figure supplement 1A ) ( Dutta et al . , 2011 ) , whereas addition of Not1MIF4G alone had little effect ( Figure 4A ) . However , addition of polyU RNA and increasing concentrations of Not1MIF4G robustly stimulated the ATPase activity of Dhh1 , but not Dhh1DQAD ( Figure 4A ) . In contrast , Gle1 , another MIF4G-fold protein that stimulates the activity of Dbp5 , a related DEAD-box ATPase that functions in mRNA export ( Montpetit et al . , 2011; Snay-Hodge et al . , 1998 ) , had no effect on Dhh1 ( Figure 4—figure supplement 1B ) . Furthermore , Not1 was unable to stimulate the catalytic activity of Dbp5 ( Figure 4—figure supplement 1B ) , demonstrating that Not1 specifically activates the ATPase cycle of Dhh1 in vitro . 10 . 7554/eLife . 18746 . 014Figure 4 . The ATPase activity of Dhh1 is stimulated in vitro and in vivo by Not1 . ( A ) ATPase activity of Dhh1 is stimulated by Not1 . The ATPase activity of full-length Dhh1 or Dhh1DQAD was measured with increasing concentrations of Not1MIF4G . Graphs represent average ATPase activity ( n=3 ) . Error bars represent SD . ( B ) Disruption of Dhh1 interaction with the MIF4G region of Not1 by mutation of conserved residues in Dhh1 . TAP-tagged Dhh1 , Dhh15X-Not , or untagged Dhh1 were purified from cells in exponential growth phase using IgG-coupled magnetic beads and co-purifying Not1-3HA was detected by Western blot . Quantification of Not1 to Dhh1 ratio is plotted with SEM ( n=5 biological replicates ) . A representative Western blot is shown . A student’s t-test comparing Dhh1 and Dhh15X-Not is shown . Asterisks indicate p<0 . 01 . ( C ) Mutations in the Not1-binding surface of Dhh1 trigger constitutive PB assembly . Wild-type or mutant Dhh1 was co-expressed from a plasmid as a GFP fusion protein in dhh1∆ cells along with Dcp2-mCherry as a PB marker and grown to exponential growth phase , then shifted to either glucose-rich or glucose starvation conditions for 20 min and observed by confocal microscopy . Images were also acquired using wide-field microscopy and PB formation was quantified using Diatrack 3 . 5 particle tracking software . Graphs represent the average Dhh1-GFP foci or Dcp2-mCherry foci number per cell ( n=3 biological replicates , >800 cells per experiment ) . Error bars represent SD . Scale bar: 5 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 18746 . 01410 . 7554/eLife . 18746 . 015Figure 4—figure supplement 1 . Not1 is a specific activator of the ATPase activity of Dhh1 . ( A ) ATPase activity of Dhh1 is weakly stimulated by RNA . ATPase activity of full-length Dhh1 was measured with increasing concentrations of polyU RNA . Graphs represent average ATPase activity ( n=3 biological replicates ) . Error bars represent SD . ( B ) Not1 is a specific activator of Dhh1 . ATPase activity of full-length Dhh1 or Dbp5 was measured in the presence of the indicated protein . Graphs represent average ATPase activity ( n=3 biological replicates ) . Error bars represent SD . ( C ) Not1 binding is diminished in a ATP-binding mutant of Dhh1 . TAP-tagged Dhh1 or Dhh1Q-motif were purified from cells in exponential growth phase using IgG-coupled magnetic beads and co-purifying Not1-3HA was detected by Western blot . Quantification of Not1 to Dhh1 ratio is plotted with SEM ( n=4 biological replicates ) . A representative Western blot is shown . A student’s t-test comparing Dhh1 and Dhh1Q-motif is shown . Asterisks indicate p<0 . 01 . ( D–F ) Known PB components localize to Dhh15X-Not foci . Xrn1-GFP ( D ) , Dcp1-GFP ( E ) , or Edc3-GFP ( F ) was co-expressed in DHH1 or dhh15X-Notcells , along with Dcp2-mCherry as a PB marker . Cells were grown to exponential growth phase , then shifted to glucose-rich or glucose starvation conditions for 20 min and observed by confocal microscopy . Images were also acquired using wide-field microscopy and PB formation was quantified using Diatrack 3 . 5 particle tracking software . Graphs represent the average foci number per cell ( n = 3 biological replicates , >800 cells per experiment ) . Error bars represent SD . Scale bar: 5 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 18746 . 01510 . 7554/eLife . 18746 . 016Figure 4—figure supplement 2 . Tethered Dhh1 does not require ATPase activation by Not1 to promote mRNA decay . ( A ) Dhh15X-Not does not show a significant defect in mRNA decay when tethered to an mRNA . Dhh1 or Dhh15X-Not was co-expressed as a PP7CP fusion protein in dhh1∆ cells expressing FBA1-PP7L . FBA1 levels were measured by qPCR and normalized to ACT1 mRNA . Graphs show mean mRNA levels from three independent experiments of biological triplicate samples . Error bars represent SD . ( B ) Wild-type and Dhh1 mutant proteins are expressed to similar levels . Western blot of Dhh1 and Dhh1 mutant protein expression from cells in exponential growth phase . Dhh1 was detected using an anti-GFP antibody . Hxk1 was used as a loading control . DOI: http://dx . doi . org/10 . 7554/eLife . 18746 . 016 If Not1 also stimulates Dhh1 ATPase activity in vivo then our model would predict that disruption of the Dhh1-Not1 interaction should lead to constitutive PB formation . To interfere with the Dhh1-Not1 interaction , we generated a mutant with amino acid substitutions in conserved residues on three distinct surfaces of Dhh1 ( Dhh1R55E , F62E , Q282E , N284E , R355E , henceforth Dhh15X-Not ) that are predicted to affect binding to Not1 , based on previous structural data ( Chen et al . , 2014; Mathys et al . , 2014 ) . Indeed , Dhh15X-Not showed a marked reduction in Not1 binding in immunoprecipitation experiments compared with wild-type Dhh1 ( Figure 4B ) , indicating that these amino acid residues are important for the interaction between Dhh1 and Not1 . Given that ATP binding by Dhh1 is likely a prerequisite for PB formation , we also examined whether ATP binding by Dhh1 was needed for the interaction with Not1 . As shown in Figure 4—figure supplement 1C , Dhh1Q-motif was defective in Not1 binding , suggesting that ATP-bound Dhh1 is needed for robust interaction with Not1 . To examine the importance of the Dhh1-Not1 interaction in PB formation , we co-expressed GFP-tagged Dhh1 , Dhh1DQAD , or Dhh15X-Not along with Dcp2-mCherry , grew cells into mid-log phase and examined Dhh1 localization . Dhh15X-Not triggered Dhh1 and Dcp2 colocalization in cytoplasmic granules in glucose-rich conditions , similar to catalytically dead Dhh1DQAD ( Figure 4C ) . All Dhh1 mutant proteins were expressed to similar levels in these experiments ( Figure 4—figure supplement 2B ) . Two lines of evidence suggest that these granules are indeed bona fide PBs . First , Dhh15X-Not granules contained several other known PB proteins – including Xrn1 , Dcp1 , and Edc3 – in both glucose-rich and glucose starvation conditions ( Figure 4—figure supplement 1D–F ) . Second , Dhh15X-Not granule assembly required robust RNA binding activity , as a Dhh15X-Not/3X-RNA mutant showed a dramatic defect in PB formation in both glucose-rich and glucose starvation conditions ( Figure 4C ) . Of note , the Dhh15X-Not mutant did not show a defect in mRNA decay using our tethering assay ( Figure 4—figure supplement 2A ) , suggesting that PB formation in cells expressing Dhh15X-Not is not due to a block in mRNA degradation . Several other decay factors have been shown to interact with the C-terminal RecA domain of Dhh1 ( Sharif et al . , 2013; Tritschler et al . , 2009 ) . Thus , it was conceivable that the Dhh15X-Not mutant not only disrupted the interaction between Dhh1 and Not1 , but also with additional factors , which may contribute to PB formation . We therefore generated a NOT1 allele , not19X-Dhh1 – with substitutions at conserved amino acid residues that were previously shown to mediate interaction between CNOT1 and DDX6 ( F791A , N795A , K804A , E832R , N834A , Y835A , K962A , F967A , and E970A ) ( Chen et al . , 2014; Mathys et al . , 2014; Rouya et al . , 2014 ) . We co-expressed Not1 or Not19X-Dhh1 along with Dhh1-GFP and Dcp2-mCherry and shifted cells into media with and without glucose to evaluate PB formation . While cells expressing wild-type Not1 showed diffuse Dhh1 and Dcp2 localization , the Not19X-Dhh1 mutant triggered colocalization of Dhh1 and Dcp2 into distinct foci in both glucose-rich and glucose starvation conditions ( Figure 5 ) . While granule induction in these cells was less pronounced than in Dhh1DQADor Dhh15X-Not cells ( Figure 4C ) in glucose-rich conditions , these foci contained other known PB components ( Figure 5—figure supplement 1A–C ) , suggesting they are bona fide PBs . Additionally , Not19X-Dhh1 was expressed at wild-type Not1 levels ( Figure 5—figure supplement 1D ) . In summary , we conclude that the ATPase cycle of Dhh1 is a critical regulator of PB formation , and that Not1 regulates the ATPase activity of Dhh1 in vivo , preventing PB formation in glucose-rich conditions . 10 . 7554/eLife . 18746 . 017Figure 5 . Mutations in Dhh1-binding surface of Not1 trigger constitutive PB assembly . Not1 or Not19X-Dhh1 was co-expressed with Dhh1-GFP and Dcp2-mCherry and grown to exponential growth phase , then shifted to either glucose-rich or glucose starvation conditions for 20 min and observed by confocal microscopy . Images were also acquired using wide-field microscopy and PB formation was quantified using Diatrack 3 . 5 particle tracking software ( see Materials and methods ) . Graphs represent average Dhh1-GFP foci or Dcp2-mCherry foci number per cell ( n=3 biological replicates , >800 cells per experiment ) . Error bars represent SD . Scale bar: 5 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 18746 . 01710 . 7554/eLife . 18746 . 018Figure 5—figure supplement 1 . Not19X-Dhh1 triggers PB assembly . ( A–C ) Known PB components localize to Not19X-Dhh1 foci . Xrn1-GFP ( A ) , Dcp1-GFP ( B ) , or Edc3-GFP ( C ) was co-expressed in NOT1 , or not19X-Dhh1cells , along with Dcp2-mCherry as a PB marker . Cells were grown to exponential growth phase , then shifted to glucose-rich or glucose starvation conditions for 20 min and observed by confocal microscopy . Images were also acquired using wide-field microscopy and PB formation was quantified using Diatrack 3 . 5 particle tracking software . Graphs represent average foci number per cell ( n=3 biological replicates , >800 cells per experiment ) . Error bars represent SD . A student's t-test comparing localization between Not1 and Not19x-Dhh1 is shown . Asterisks indicate p<0 . 005 ( *** ) , or p<0 . 05 ( * ) . Scale bar: 5 μm ( D ) Not19X-Dhh1 mutant is expressed to wild-type Not1 levels . Western blot of Not1 or Not19X-Dhh1 mutant protein expression from cells in exponential growth phase . Not1 was detected using Rabbit IgG , and Dhh1 was detected using an anti-Dhh1 antibody . Hxk1 was used as a loading control . DOI: http://dx . doi . org/10 . 7554/eLife . 18746 . 018 To better understand how the ATPase cycle of Dhh1 regulates PB formation , we attempted to reconstitute granule formation in vitro . Remarkably , recombinant Dhh1 , in the presence of RNA and ATP , readily formed droplets in solution ( Figure 6A and B ) . These droplets showed hallmarks of liquid-liquid phase separation , undergoing growth and fusion events and reversible deformation ( Video 5 ) , consistent with the reported biophysical behavior of PBs ( Kroschwald et al . , 2015 ) . Dhh1 droplet formation was RNA-dependent , as no droplets formed when RNA was omitted ( Figure 6A ) , and the number and size of droplets rapidly decreased upon addition of RNase A ( Figure 6C , Video 6 ) , but not with buffer alone ( Figure 6C , Video 7 ) . 10 . 7554/eLife . 18746 . 019Figure 6 . Dhh1 PB dynamics can be recapitulated in vitro . ( A ) Formation of liquid Dhh1-droplets depends on the presence of the RNA analog polyuridylic acid ( polyU ) and increases with increasing protein concentration . Recombinant mCherry-tagged Dhh1 was diluted into a low salt buffer and incubated at 4°C for 1 hr . Formation of liquid droplets was observed by fluorescence microscopy . Scale bar: 10 μm . ( B ) Dhh1 liquid droplet formation requires ATP . Dhh1 liquid droplets were formed as in ( A ) , in the presence or absence of ATP and the creatine kinase ATP regeneration system . Scale bar: 20 μm . ( C ) Addition of Not1MIF4G or RNase A , but not buffer alone , dissolves pre-formed Dhh1 liquid droplets . Dhh1 liquid droplets were pre-formed for 20 min at 4°C , followed by the addition of 5 µM Not1MIF4G or RNase A . Scale bar: 10 μm . ( D , E ) Pre-incubation with Not1MIF4Gprevents formation of Dhh1 , but not Dhh1DQAD liquid droplets . Reactions were imaged after 1h incubation at 4°C . Scale bar: 10 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 18746 . 01910 . 7554/eLife . 18746 . 020Figure 6—figure supplement 1 . Single point mutants in the ATP binding site of Dhh1 affect PB assembly and liquid droplet formation . ( A ) Single mutations in the Q-motif of Dhh1 show minor defects in PB formation . Wild-type Dhh1 , Dhh1F66R or Dhh1Q73A were expressed from a plasmid as a GFP fusion protein in dhh1∆ cells along with Dcp2-mCherry as a PB marker and grown to exponential growth phase , then shifted to either glucose-rich or glucose starvation conditions for 20 min and observed by wide-field microscopy . PB formation was quantified using Diatrack 3 . 5 particle tracking software . Graphs represent average Dhh1-GFP or Dcp2-mCherry foci number per cell ( n=3 biological replicates , >300 cells per experiment ) . Error bars represent SD . A student’s t-test comparing Dhh1F66R or Dhh1Q73A with Dhh1 is shown . Scale bar: 5 µm ( B ) Dhh1 liquid droplet formation requires ATP binding . Recombinant mCherry-tagged Dhh1 , Dhh1DQAD or Dhh1F66R were diluted into a low salt buffer and incubated at 4°C for 1 hr in the presence of ATP and 0 . 075 mg/mL polyU . Formation of liquid droplets was observed by fluorescence microscopy . Scale bar: 10 μmDOI: http://dx . doi . org/10 . 7554/eLife . 18746 . 02010 . 7554/eLife . 18746 . 021Video 5 . Purified Dhh1 , ATP , and RNA form liquid-like droplets in vitro . Droplets were formed for 2 min with 6 . 25 µM Dhh1-mCherry and 0 . 075 mg/mL polyU in a final volume of 20 µL and imaged live in a time course ( 5 s intervals; movie played at 7 fps ) . Fusion events can be observed that lead to rounding up of the new droplet to assume a spherical shape . DOI: http://dx . doi . org/10 . 7554/eLife . 18746 . 02110 . 7554/eLife . 18746 . 022Video 6 . RNase A treatment dissolves Dhh1 liquid droplets . Droplets were formed for 20 min from 6 . 25 µM Dhh1-mCherry and 0 . 075 mg/mL polyU in a final volume of 20 µL . The imaging time course started ( 10 s intervals; movie played at 3 fps ) . After few frames , RNase A was added ( 1 . 5 µL of a 0 . 04 µg/mL stock solution , which was prepared by dilution of a 10 mg/mL stock solution in Not1MIF4G storage buffer ) to the pre-formed Dhh1 droplets . DOI: http://dx . doi . org/10 . 7554/eLife . 18746 . 02210 . 7554/eLife . 18746 . 023Video 7 . Addition of Not1MIF4G storage buffer does not affect Dhh1 liquid droplet formation . Droplets were formed for 20 min from 6 . 25 µM Dhh1-mCherry and 0 . 075 mg/mL polyU in a final volume of 20 µL . The imaging time course was started ( 10 s intervals; movie played at 3 fps ) . After a few frames , 1 . 5 µL Not1MIF4G storage buffer was added to the pre-formed Dhh1 droplets . DOI: http://dx . doi . org/10 . 7554/eLife . 18746 . 023 Next , we examined the role of ATP binding by Dhh1 in droplet formation . Despite numerous attempts , we were unable to purify Dhh1Q-motif with sufficient quality for analysis . However , we successfully purified a single Q-motif point mutant , Dhh1F66R ( Dutta et al . , 2011 ) for use in our in vitro assay . While Dhh1F66R showed only a minor defect in PB localization following glucose starvation ( Figure 6—figure supplement 1A ) , this mutant showed a dramatic loss of droplet formation in vitro ( Figure 6—figure supplement 1B ) . Additionally , Dhh1 droplets did not form in the absence of ATP ( Figure 6B ) , demonstrating that Dhh1 in its ATP-bound form promotes liquid droplet formation . Given that Not1 promotes PB disassembly in vivo by stimulating the ATPase activity of Dhh1 , we next examined whether the presence of Not1 also antagonizes Dhh1 liquid droplet formation in vitro . Consistent with our in vivo data , addition of Not1MIF4G triggered the dissolution of pre-formed Dhh1 liquid droplets ( Figure 6C , video 8 ) . Furthermore , no assembly occurred when Not1MIF4G was added before polyU during the assembly reaction ( Figure 6D ) . 10 . 7554/eLife . 18746 . 024Video 8 . Addition of Not1MIF4G dissolves Dhh1 liquid droplets . Droplets were formed for 20 min with 6 . 25 µM Dhh1-mCherry and 0 . 075 mg/mL polyU in a final volume of 20 µL . The imaging time course was started ( 10 s intervals; movie played at 3 fps ) . After a few frames , Not1MIF4G ( 1 . 5 µl of a 150 µM stock solution ) was added to the pre-formed Dhh1 droplets . DOI: http://dx . doi . org/10 . 7554/eLife . 18746 . 024 To determine whether catalytically active Dhh1 was required for droplet dissolution , we also tested the functionality of the ATPase-dead Dhh1DQAD mutant in our in vitro assay . While Dhh1DQAD formed droplets to a similar extent as wild-type Dhh1 ( Figure 6E ) , these structures did not dissolve in the presence of Not1MIF4G , supporting the specificity of the observed effect ( Figure 6E ) . Interestingly , the Dhh1DQAD droplets slightly increased in size and number upon Not1MIF4G addition . It is likely that the MIF4G domain of Not1 , like other MIF4G domains , stabilizes a conformation of the two RecA domains which facilitates nucleotide and RNA loading ( Montpetit et al . , 2011; Oberer et al . , 2005 ) , which consequently may enhance droplet formation in the absence of Dhh1's ATPase activity . Thus , while other mRNA decay factors contribute to PB formation in vivo , this demonstrates that with a minimal number of constituents , namely Dhh1 , RNA , and ATP , higher-order dynamic liquid droplets can be formed in vitro . These droplets recapitulate several properties of PBs formed in vivo such as the dependence on ATP and RNA binding by Dhh1 to form ( Figure 2A and B ) as well as the requirement of both a functional Dhh1 ATPase and the MIF4G domain of Not1 for dissolution ( Figures 4C and 5 ) . The DEAD-box ATPase Dhh1 and its orthologs play a critical role in translational repression and degradation of cytoplasmic mRNAs . However , how the catalytic activity of Dhh1 contributes to its function has not been well defined . Here , we show that the ATPase activity of Dhh1 regulates the dynamics of PBs in an RNA-dependent manner . Point mutations in Dhh1 that prevent ATP hydrolysis or disrupt the interaction surface with the ATPase activator Not1 were sufficient to trigger aberrant PB formation in vivo in the absence of stress ( Figure 4C , Figure 5 ) . Furthermore , we can recapitulate this process in vitro , as Dhh1 forms dynamic liquid droplets in the presence of RNA and ATP that are dissolved upon addition of the purified MIF4G ATPase activation domain of Not1 ( Figure 6 ) . The central scaffold of the CCR4-NOT complex , Not1 , similar to its mammalian homolog CNOT1 , is shown here to be an activator of the catalytic cycle of Dhh1 in vitro . Like other known DEAD-box cofactors , Not1 possesses a MIF4G domain ( Chen et al . , 2014; Mathys et al . , 2014; Ozgur et al . , 2015b; Rouya et al . , 2014 ) that is critical for stimulation of Dhh1 . In the absence of Not1MIF4G and RNA , we could not detect ATPase activity of Dhh1 . Both Dhh1 and DDX6 alone adopt an unusual closed conformation , characterized by extensive intramolecular contacts that are not present in other members of the DEAD-box protein family ( Chen et al . , 2014; Cheng et al . , 2005; Mathys et al . , 2014 ) . Binding of CNOT1 causes a dramatic structural rearrangement , shifting DDX6 into an ATPase-competent state ( Mathys et al . , 2014 ) . However , even the Not1-stimulated activity of Dhh1 remains low ( Figure 4A ) . Although the rate-limiting step in the catalytic cycle of DEAD-box ATPases often appears to be the release of substrate RNA and ADP/Pi ( Cao et al . , 2011; Henn et al . , 2010; Hilbert et al . , 2011; Wang et al . , 2010 ) , it is conceivable that the large conformational change that Dhh1 must undergo in order to bind both Not1 and substrate significantly contributes to the slow ATPase cycle of Dhh1 . While mutations in conserved residues that mediate the interaction between Dhh1 and Not1 triggered constitutive PB formation , these mutants did not completely recapitulate the degree of PB formation seen in cells expressing catalytically dead Dhh1DQAD . This may perhaps be due to only a partial loss of stimulation of Dhh1 by Not1 in these mutants . Unfortunately , we were unable to purify these variants as recombinant proteins , and therefore could not determine their effect on ATPase stimulation in vitro . Alternatively , there may be additional cellular factors that modulate the catalytic cycle of Dhh1 . In addition to demonstrating a critical role for ATP hydrolysis by Dhh1 in the regulation of PB formation , we also show that both RNA and ATP binding by Dhh1 are critical for PB assembly , consistent with prior observations ( Dutta et al . , 2011 ) . Neither Dhh1Q-motif nor Dhh13X-RNA mutant cells robustly form PBs following glucose starvation . In addition , mutations that disrupt RNA binding also prevent constitutive PB formation of catalytically dead Dhh1 ( Figure 2B ) . Remarkably , Dhh1 forms liquid droplets in vitro that require both ATP and RNA ( Figure 6 ) , indicating that multimeric assembly of Dhh1 in its ATP-bound state with RNA may drive PB formation . Since DDX6 can oligomerize in both an RNA-dependent and RNA-independent manner ( Ernoult-Lange et al . , 2012 ) , and given that both Dhh1 and DDX6 exist in molar excess over cytoplasmic mRNA ( Ghaemmaghami et al . , 2003; Nagaraj et al . , 2011 ) , it is conceivable that an ATP-bound conformation of Dhh1 multimerizes on RNA in vivo , thereby delivering mRNAs to PBs and seeding PB assembly . Upon ATP hydrolysis , Dhh1 could then return to the cytoplasmic pool to bind and deliver the next mRNA target . However , when ATP hydrolysis is inhibited , such as in Dhh1DQAD or Dhh15X-Not-expressing cells , Dhh1 remains associated with its mRNA client , triggering the formation of constitutive PBs . Despite an increasing understanding of PB composition , the precise functional role of PBs remains unclear . Given the large number of mRNA decay factors present in PB assemblies , as well as the accumulation of Xrn1-protected polyG-tract-containing mRNAs , PBs were initially proposed to be sites of mRNA decay ( Sheth and Parker , 2003; Teixeira and Parker , 2007 ) . However , several studies have shown that mRNAs can also stably localize within PBs , raising the question of whether these granules are sites of active mRNA decay or rather of mRNA storage ( Hocine et al . , 2013; Lavut and Raveh , 2012; Lui et al . , 2014; Zid and O’Shea , 2014 ) . It should be noted , however , that in many cases mRNAs were localized to PBs using either the MS2 or PP7 coat protein system , whereby multiple stem loops are engineered into the 3’UTR of transcripts and then visualized using fluorescently tagged coat-protein fusions that recognize these stem loops . Yet , recent data indicate that these stem loop systems may inhibit mRNA decay in budding yeast , and that primarily these stem loop structures – but not the body of transcripts – persist in PBs , and cannot be degraded by Xrn1 ( Garcia and Parker , 2015 , 2016; Haimovich et al . , 2016; Heinrich and Weis , unpublished ) . Our smFISH data are consistent with the hypothesis that active decay occurs within PBs , since none of the four tested mRNAs were enriched in wild-type and Dhh1DQAD PBs in contrast to PBs in xrn1∆ cells ( Figure 1D–E , Figure 1—figure supplement 1B–C ) . Thus , selective delivery of mRNAs to PBs could enhance their degradation because of the high local concentration of the mRNA decay machinery in PBs . Alternatively , the sequestration of mRNA decay factors and selected mRNAs into PBs could also allow for spatial separation of translation factors from mRNA , preventing translation of messages that would be unproductive during periods of stress . The formation of mRNPs – including PBs – into membrane-less organelles that behave like dynamic liquid droplets has recently emerged as a common mechanism by which cells may further compartmentalize their biochemistry ( Guo and Shorter , 2015; Kroschwald et al . , 2015; Weber and Brangwynne , 2012 ) . Furthermore , a variety of different ATP-driven protein machines have also emerged as important regulators of mRNP granule assembly . For example , stress granule ( SG ) assembly and dynamics are disrupted by loss-of-function alleles in the MCM and RVB helicase complexes , while hypomorphic alleles in the chaperonin-containing T ( CCT ) complex form more SGs ( Jain et al . , 2016 ) . Additionally , the AAA+ ATPase Cdc48 was also previously shown to facilitate clearance of SGs ( Buchan et al . , 2013 ) . Our data demonstrate that the DEAD-box ATPase Dhh1 is a critical regulator of PB disassembly in vivo and that liquid droplets containing Dhh1 multimers form in the presence of RNA and ATP , which can be dissolved upon induction of ATP hydrolysis in vitro . Two biochemical functions are critical for the role of Dhh1 in PB formation both in vivo and in vitro: ( 1 ) Dhh1’s affinity for RNA , which may facilitate delivery of mRNA substrates into PBs , and ( 2 ) ATP binding , and the tuning of Dhh1’s ATPase activity by factors such as Not1 . These features of Dhh1 may ultimately be the critical controllers of PB formation and PB turnover . PBs have been extensively studied , yet the molecular mechanisms driving PB formation and disassembly are poorly understood . Our data show that the ATP- and RNA-bound form of Dhh1 promotes PB formation , while Not1 promotes PB disassembly by stimulating the ATPase activity of Dhh1 ( for illustration , see model Figure 7 ) . Still , there are several elements of the PB and Dhh1 ATPase cycle that remain unclear . For example , what leads to PB formation under specific cellular stress conditions ? Is this driven by a stress-induced attenuation of translation or increase in mRNA turnover , which leads to an increased number of client mRNAs targeted to PBs ? Alternatively , cellular stress may dampen the ATPase activity of Dhh1 , for example by regulating the interaction between Dhh1 and Not1 , thereby shifting the equilibrium towards the ATP-bound , RNA-bound Dhh1 state . This in turn would then slow down PB disassembly , causing a build up of PB structures . 10 . 7554/eLife . 18746 . 025Figure 7 . Model: The ATPase cycle of Dhh1 controls PB assembly and disassembly . An ATP- and RNA-bound conformation of Dhh1 nucleates PB formation , while stimulation of Dhh1’s ATPase activity by Not1 promotes granule disassembly . DOI: http://dx . doi . org/10 . 7554/eLife . 18746 . 025 In addition , the polyA status of Dhh1-bound mRNAs targeted to PBs is also unknown . While deadenylation by CCR4-NOT was previously placed upstream of Dhh1 in the mRNA decay pathway ( Coller et al . , 2001; Fischer and Weis , 2002 ) , it is unclear whether deadenylation is a prerequisite for targeting mRNAs to PBs . Finally , with respect to the hydrolysis step , does Not1 facilitate recycling of Dhh1 from the PB by promoting the release of Dhh1 from RNA , similar to the function of other DEAD-box activators ( Linder and Jankowsky , 2011; Montpetit et al . , 2011 ) or does it regulate the interaction with other factors such as scaffold proteins of PBs ? Intriguingly , the ATPase cycle of Dhh1 does not seem to influence the recycling of other PB components ( Figure 3A ) , consistent with the idea that the regulated interaction between Dhh1 and mRNA shifts the balance between PB formation or disassembly . While Not1 has a well-known role as the central subunit of the major cytoplasmic deadenylase , our work defines a novel function for Not1 in PB disassembly and Dhh1 recycling , which presumably occurs at a late stage in mRNA turnover . Interestingly , there is increasing evidence that the CCR4-NOT complex functions at several other steps during gene expression outside of mRNA decay , including transcription ( Gupta et al . , 2016; Kruk et al . , 2011; Villanyi and Collart , 2015 ) and translation ( Panasenko , 2014; Preissler et al . , 2015 ) . Future work is needed to address how the activity of this multifunctional protein complex is modulated in order to regulate and coordinate multiple steps of gene expression . The strains used in this study are derivatives of W303 and are described in Supplementary file 1 . Yeast deletion strains and C-terminal epitope tagging of ORFs was done by PCR-based homologous recombination , as previously described ( Longtine et al . , 1998 ) . Generation of bacteriophage PP7CP and PP7-loop tagging plasmids was described previously ( Carroll et al . , 2011 ) . Plasmids for this study are described in Supplementary file 2A . Mutations in Dhh1 were generated using a QuikChange II site-directed mutagenesis kit ( Agilent Technologies , Santa Clara , CA ) using Pfu Ultra or Pfu Turbo . Mutagenic oligonucleotides were designed using the Agilent Technologies primer design platform . Construction of NOT1-TAP and not19X-Dhh1-TAP integration vectors were made using NOT1 and NOT1 ( F791A , N795A , K804A , E832R , N834A , Y835A , K962A , F967A , and E970A ) gene block fragments ordered from Integrated DNA Technologies ( IDT , Coralville , IA ) that were ligated into the single-integration vector pNH605 by Gibson Assembly ( New England Biolabs , Ipswich , MA ) . Primer sequences for strain construction are listed in Supplementary file 2B . Sample preparation was performed as previously described ( Carroll et al . , 2011 ) . Briefly , yeast cells were inoculated overnight in synthetic media containing 2% glucose and grown to saturation . The following morning , cultures were diluted and grown to exponential growth phase ( OD600 = 0 . 4–0 . 8 ) then collected by centrifugation and lysed in 1X phosphate-buffered saline ( PBS ) with 0 . 1% Tween-20 and protease inhibitors . Lysis was performed using a Mini-Beadbeater-96 ( BioSpec Products , Inc . , Bartlesville , OK ) with a 5-minute cycle . The extract was clarified by centrifugation , and RNA was isolated using the RNeasy RNA isolation kit ( Qiagen , Hilden , Germany ) . Resulting RNA samples were stored at −80°C . RNA was isolated as described above and quantified using a NanoDrop spectrophotometer ( Thermo Fischer , Waltham , MA ) . cDNA was generated by reverse transcription of 1 μg of RNA using a random hexamer oligonucleotide ( Invitrogen , Carlsbad , CA ) and Superscript II ( Invitrogen ) . Quantitative PCR was performed in real time using the StepOnePlus Real-Time PCR System ( Applied Biosystems , Foster City , CA ) and a SYBR-Green ROX qPCR Master Mix ( Thermo Fischer ) supplemented with gene-specific primers as reported in ( Carroll et al . , 2011 ) . The indicated strains were inoculated overnight in synthetic media containing 2% glucose and grown to saturation . The following morning , cultures were diluted and grown in synthetic complete media containing 2% glucose at 25°C to exponential growth phase ( OD600 = 0 . 6-0 . 8 ) , then shifted to synthetic complete media with or without glucose for 20 min and fixed for 15 min with 4% paraformaldehyde . Samples were processed for single molecule fluorescence in situ hybridization ( smFISH ) as described in ( Heinrich et al . , 2013 ) , with the exception of spheroplasting yeast cells for 20 min using 1% 20T zymolyase . Mixtures of DNA probes coupled to CAL Fluor Red 590 ( Stellaris , LGC Biosearch , Novato , CA; probes were synthesized by BioCat , Heidelberg , Germany ) were used for smFISH , targeting the FBA1 , GFA1 , PAT1 , or PGK1 open reading frame moiety ( Supplementary Supplementary file 3 ) . Microscopy was performed with an inverted epi-fluorescence microscope ( Nikon Ti ) equipped with a Spectra X LED light source and a Hamamatsu Flash 4 . 0 sCMOS camera using a PlanApo 100 x NA 1 . 4 oil-immersion objective and the NIS Elements software . Images were processed using FIJI software . Quantification of colocalization was performed on all planes of a 3D stack image using the Colocalization Threshold tool in Fiji . In brief , images were background-subtracted and thresholded with a defined minimum threshold set for each smFISH probe separately . Then , the Colocalization Threshold tool was applied , which highlighted the colocalization between PB and mRNA . The colocalization events and the total number of PBs were then counted manually . The percentage of colocalization was calculated by forming the ratio between the number of PBs colocalizing with mRNA and the total number of PBs . Immunoprecipitation experiments were performed as in ( Oeffinger et al . , 2007 ) . Yeast were inoculated in synthetic media containing 2% glucose and grown overnight to saturation , then diluted the following day in 1 L synthetic media and grown to exponential growth phase ( OD600 = 0 . 4–1 . 0 ) . Cells were harvested by centrifugation at 3000 x g for 10 min , then resuspended in resuspension buffer ( final concentration: 20 mM HEPES-KOH , pH 7 . 4 , 1 . 2% polyvinylpyrrolidone ( molecular weight = 40 , 000 ) , 1 mM DTT , 0 . 2 mM PMSF , 10 μg/mL Pepstatin A ) . Pellets were centrifuged at 2600 x g for 15 min at 4°C to remove extra buffer , then centrifuged again at 2600 x g for 15 min at 4°C and pellets were frozen in liquid nitrogen and stored at −80°C . Frozen yeast pellets were then lysed with a Retsch Planetary Ball Mill MM 301 ( Retsch , Newtown , PA ) for six cycles at 30 Hz for 3 min with cooling in liquid nitrogen between cycles . 0 . 5 g of lysate was then resuspended in 14 mL TBT buffer ( final concentration: 20 mM HEPES-KOH , pH 7 . 4 , 110 mM KOAc , 2 mM MgCl2 , 1 mM DTT , 0 . 5% Triton X-100 , 0 . 1% Tween-20 , 0 . 2 mM PMSF , 10 μg/mL Pepstatin A , 1:5000 SuperRNasin ( Ambion , Austin , TX ) , 1:5000 Antifoam B ( Sigma Aldrich , St . Louis , MO ) . Lysate was clarified through 2 . 7 µm and 1 . 6 µm GD/X Glass Microfiber syringe filters ( Whatman , Maidstone , UK ) , and then incubated with 8 mg rabbit IgG ( Sigma Aldrich ) -coupled magnetic beads ( Thermo Fischer , Waltham , MA ) – corresponding to 400 μL bead slurry at 20 mg/μL slurry – and rotated at 4°C for 30 min . The beads were collected using a magnetic rack , washed three times with 1 mL TBT buffer , and a final wash in 1 mL of 100 mM NH4OAc , ( pH = 7 . 4 , 0 . 1 mM MgCl2 , 0 . 2% Tween-20 ) for 5 min while rotating . Protein complexes were eluted from the beads directly in SDS-PAGE sample buffer and boiled at 95°C , and processed further for Western blot . For Western blot analysis , roughly 5 OD600 units of cells were harvested and treated with 5% trichloroacetic acid ( TCA ) and incubated at 4°C for 10 min . Acid was removed using an acetone wash , and the resulting pellet was dried 2–3 hr . Cell pellets were resuspended in 200 μL breakage buffer ( final concentration: 50 mM Tris-HCl pH = 7 . 5 , 1 mM EDTA , 2 . 75 mM DTT , and protease inhibitors ) and disrupted using glass beads and a Mini-Beadbeater-96 ( BioSpec Products , Inc . Bartlesville , OK ) . Samples were cooled on ice for 5 min and SDS sample buffer was added and homogenates were boiled . Proteins were resolved by 4–12% Bolt Bis-Tris SDS PAGE ( Thermo Fischer , Waltham , MA ) , then transferred to nitrocellulose membrane ( GE Life Sciences , Marlborough , MA ) . Membranes were blocked in PBS with 4% non-fat milk , followed by incubation with primary antibody overnight . Membranes were washed four times with PBS with 0 . 1% Tween-20 ( PBS-T ) and incubated with secondary antibody for 45 min . Membranes were then analyzed and quantified using an infrared imaging system ( Odyssey; LI-COR Biosciences , Lincoln , NE ) . The following primary antibodies were used for detection of tagged proteins at the indicated dilutions: rabbit-anti-Dhh1 ( 1:5000 ) as described in ( Fischer and Weis , 2002 ) , ( Weis Lab ETH Zurich Cat# Weis_001 , RRID:AB_2629458 ) , anti-FLAG-M2 ( 1:2500 ) ( Sigma-Aldrich Cat# F1804 , RRID:AB_262044 , St . Louis , MO ) , mouse-anti-HA . 11 ( 1:2000 ) ( Covance Research Products , Inc . Cat# MMS-101P-1000 , RRID:AB_291259 , Princeton , NJ ) mouse-anti-GFP ( 1:1000 ) ( Roche Cat# 11814460001 , RRID:AB_390913 ) , and rabbit-anti-Hxk1 ( 1:3000 ) ( US Biological Cat# H2035-01 , RRID:AB_2629457 , Salem , MA ) . IRdye 680RD goat-anti-rabbit ( LI-COR Biosciences Cat# 926–68071 , RRID:AB_10956166 ) and IRdye 800 donkey-anti-mouse ( LI-COR Biosciences Cat# 926–32212 , RRID:AB_621847 ) were used as secondary antibodies . Samples were grown overnight in synthetic media containing 2% glucose , diluted to OD600 = 0 . 05 or 0 . 1 the following day , and grown to mid-log phase ( OD600= 0 . 3–0 . 8 ) . Cells were harvested by centrifugation and washed in ¼ volume of fresh synthetic media +/− 2% glucose , then harvested again and resuspended in 1 volume of fresh synthetic media +/− 2% glucose and grown 15 min at 30°C . Cells were then transferred onto Concanavalin A-treated MatTek dishes ( MatTek Corp . , Ashland , MA ) and visualized at room temperature using the DeltaVision Elite Imaging System with softWoRx imaging software ( GE Life Sciences , Marlborough , MA ) . The system was based on an Olympus 1X71 inverted microscope ( Olympus , Japan ) , and cells were observed using a UPlanSApo 100 × 1 . 4 NA oil immersion objective . Single plane images were acquired using a DV Elite CMOS camera . Image processing for PB analysis was performed using Diatrack 3 . 5 particle tracking software as described below . Samples were grown as indicated in 'wide-field fluorescence microscopy' methods section and imaged using an Andor/Nikon Yokogawa spinning disk confocal microscope ( Belfast , United Kingdom ) with Metamorph Microscopy Automation & Image Analysis software ( Molecular Devices , Sunnyvale , CA ) . The system was based on a NikonTE2000 with inverted microscope , and cells were observed using a PlanApo100 × 1 . 4 NA oil immersion objective and single plane images were captured using a Clara Interline CCD camera ( Andor ) . Samples were grown overnight in synthetic media containing 2% glucose , diluted to OD600 = 0 . 05 or 0 . 1 the following day , and grown to mid-log phase ( OD600= 0 . 3–0 . 8 ) . Cells were harvested by centrifugation and washed in ¼ volume of fresh synthetic media +/− 2% glucose , then harvested again and resuspended in 1 volume of fresh synthetic media +/− 2% glucose and grown 15 min at 30°C . Cells were then transferred onto Concanavalin A-treated MatTek dishes ( MatTek Corp . , Ashland , MA ) and visualized at room temperature . Dhh1-GFP and Dhh1DQAD-GFP photobleaching experiments were performed on a Leica SP8 Laser Scanning Confocal Microscope ( Leica , Wetzlar , Germany ) using Leica LAS AF SP8 software ( version 3 . 3 ) . The system was based on a LeicaDMI6000B inverted microscope , and cells were observed using a PlanApo 63 × 1 . 4 NA oil immersion CS2 objective and a conventional photomultiplier tube ( PMT ) detector . Dcp1-GFP , Dcp2-GFP , Edc3-GFP , and Xrn1-GFP photobleaching experiments were performed on a Andor/Nikon Yokogawa spinning disk confocal microscope with acquisition parameters as described above . Using the Leica SP8 Laser Scanning Confocal Microscope , selected PBs were subjected to 5–10 pulses of an argon laser at 488 nm . Images were collected from a single plane with a 2 . 92 nm pinhole at 500 ms intervals for 50 s post-bleach . Using the Andor/Nikon spinning disk confocal microscope , selected PBs were pulsed once for 500 ms using a Mosaic 405 nm laser ( Andor ) and images were collected from a single plane at 3 s intervals for 3 min post-bleach . For all experiments , PB fluorescent intensity and total cellular fluorescence intensity were quantified in ImageJ/FIJI by manual tracing . The background was determined by determining the intensity of an ROI with the same size as either the PB or the total cell . PB intensity was normalized to the total fluorescent intensity of the cell using the equation:PBnormal=IntensityPB−IntensityPB backgroundIntensityTotalCell−IntensityTotalCellBackground Recovery curves were generated by normalization to the bleach point , and percent of fluorescent recovery values were determined by curve fitting using the equation:f ( t ) =A ( 1−e−rt ) To quantify PBs , we used Diatrack 3 . 5 particle tracking software ( Vallotton and Olivier , 2013; www . diatrack . org ) . A Matlab script transformed our foci images into a single long sequence that could be fed to Diatrack . This allowed the same image analysis parameters to be applied across all images and experiments . PBs display significant variations in appearance ( size and brightness ) . They were identified in Diatrack based on their intensity and contrast measure . Optimal parameters were selected interactively such that false negative and false positive rates were below 3% . Occasionally , yeast vacuoles pinch the cytoplasm against the cell wall . This tends to create narrow intensity ridges in some of our fluorescence images and can trigger the default Diatrack particle detector . Thus , we used an alternative contrast measure ( ‘blurred 360’ ) . This only retained particles around which the intensity decreases significantly in all directions ( rather than decreasing on average only ) . A sample movie showing detected PB for a variety of images is provided as supplementary information ( movieDetection . avi ) . For each image , a list of intensities corresponding to each PB in each image was exported from the software and further processed in Microsoft Excel ( Microsoft Corporation , Redmond , WA ) . The sum of particle intensities represents a suitable measure of overall PB abundance . Alternately , the number of PBs per image may also be used . This value was divided by the number of cells in each image to deliver per-cell PB abundances . Automated counting of cells was performed as described in ( Hadjidemetriou et al . , 2008 ) . Dhh1 ( wild-type , Dhh1DQAD , Dhh1F66R ) and Dbp5 were cloned into a pETMCN-based expression vector with a N-terminal 6xHis and V5 tag plus a C-terminal mCherry tag . The MIF4G domain of Not1 ( residues 754–1000 ) was cloned into a pETMCN-based expression vector with a N-terminal 6xHis and V5 tag . Recombinant proteins were expressed in E . coli BL21 DE3 cells grown in rich medium . Cells were grown at 37°C to an OD600 of 0 . 6 and induced with 300 µM IPTG . Cells were then grown overnight at 18°C , harvested and resuspended in 30 mL lysis buffer ( 500 mM NaCl , 25 mM Tris-HCl pH 7 . 5 , 10 mM imidazole , protease inhibitors ) per cell pellet from 2 L of culture . After cell lysis by sonication , the 6xHis tagged proteins were affinity extracted with Ni2+ sepharose and further purified by size exclusion with a Superdex 200 column ( Dhh1 and Dbp5 , in the final storage buffer 200 mM NaCl , 25 mM Tris-HCl pH 7 . 5 , 2 mM DTT ) or Superdex 75 column ( Not1MIF4G , in the final storage buffer 200 mM NaCl , 25 mM Tris-HCl pH 7 . 5 , 2 mM DTT , 10% glycerol ) ( GE Life Sciences , Marlborough , MA ) . Gel filtration fractions were analyzed by SDS-PAGE . Clean fractions were pooled , concentrated to about 500 µM and snap frozen in small aliquots . ATPase assays were performed according to ( Montpetit et al . , 2011 ) with the following modifications: 2 µM Dhh1 or Dbp5 was mixed with 2 µL 10x ATPase buffer ( 300 mM HEPES-KOH pH 7 . 5 , 1 M NaCl , 20 mM MgCl2 ) , Not1 or Gle1 as indicated , 4 µL 10 mg/ml polyU ( unless indicated otherwise ) , RNase inhibitors , 13 . 3 µL 60% glycerol , 2 . 7 µL 10 mg/mL BSA , and Not1 storage buffer to compensate for volume differences , in a final volume of 36 µL . Reactions were set up in triplicate in a 96-well NUNC plate . The assay was initiated by the addition of 40 µL of a master mix containing 1x ATPase buffer , 2 . 5 mM ATP ( from a 100 mM stock in 0 . 5 M HEPES-KOH pH 7 . 5 ) , 1 mM DTT , 6 mM phosphoenolpyruvate , 1 . 2 mM NADH ( from a 12 mM stock in 25 mM HEPES-KOH pH 7 . 5 ) and 125–250 units / mL PK/LDH . NADH absorption was monitored with a CLARIOstar plate reader at 340 nm in 30 s intervals for 400 cycles . Dhh1-mCherry ( wild-type , Dhh1DQAD , or Dhh1F66R ) was diluted at least tenfold to 50 µM with 1x ATPase buffer . From this solution , Dhh1 was added as indicated in Figure 6 to a 20 µL reaction ( and pre-incubated with Not1 , if applicable ) in a 384-well microscopy plate . Reactions were filled to 5 µL with 1x ATPase buffer . A master mix was prepared with 2 µL 10x ATP reconstitution system ( 40 mM ATP , 40 mM MgCl2 , 200 mM creatine phosphate , 70 U/mL Creatine Kinase ) , 1 µL HEPES-KOH pH 6 . 6 , 1 µL BSA ( 10 mg/mL ) , 1 . 5 µL 1 mg/mL polyU ( unless indicated otherwise ) , 0 . 2 µL RNase inhibitors and 10 µL buffer ( 150 mM KCl , 30 mM HEPES-KOH pH 7 . 4 , 2 mM MgCl2 ) and added to the protein solutions . For the reactions not containing the creatine kinase ATP regeneration system , ATP was supplemented to a final concentration of 5 mM together with 10 mM MgCl2 . Reactions were mixed , incubated at 4°C for the indicated length , and microscopy was performed with an inverted epi-fluorescence microscope ( Nikon Ti ) equipped with a Spectra X LED light source and a Hamamatsu Flash 4 . 0 sCMOS camera using a PlanApo 60 × NA 1 . 4 oil-immersion objective and the NIS Elements software .	Most cells and organisms live in changeable environments . Adapting to environmental changes means that organisms must quickly alter which of their genes they express . Varying which genes are switched on or off is not enough; cells must also degrade existing messenger RNAs ( or mRNAs for short ) , which contain the genetic instructions of the previously active genes . Therefore , cells must tightly regulate the machinery needed to degrade mRNAs . When Baker’s yeast ( also known as budding yeast ) cells experience certain stressful conditions , the proteins that break down mRNAs localize into specific structures inside the cell known as ‘processing bodies’ . These structures are found in many other organisms across evolution , from yeast to human . Processing bodies also form in a variety of biological contexts , such as in nerve cells and developing embryos . Still , why cells form processing bodies , and how their assembly is regulated , is not well understood . One essential component of processing bodies is an enzyme called Dhh1 . This enzyme has been conserved throughout evolution and is known to promote the decay of mRNAs as well as to repress their translation into proteins . Now , Mugler , Hondele et al . show that Dhh1’s must break down molecules of the energy carrier ATP ( referred to as its “ATPase activity” ) in order to regulate the dynamic nature of processing bodies . Mutant Dhh1 proteins that lack ATPase activity form permanent processing bodies in non-stressed yeast cells . This shows that that the breakdown of ATP by Dhh1 is required for the disassembly of processing bodies . Similar results were seen for mutant Dhh1 proteins that cannot interact with Not1 , a protein which enhances the ATPase activity of Dhh1 . Next Mugler , Hondele et al . mixed purified Dhh1 with ATP and RNA molecules and saw that the mixture underwent a “liquid-liquid phase separation” and formed observable granules , similar to oil droplets in water . These granules dissolved when Not1 was added to stimulate the Dhh1 enzyme to turnover ATP . This showed that several important biochemical and biophysical aspects of processing bodies seen within living cells could be recreated outside of a cell . Armed with a greater understanding of the rules that govern the formation of processing bodies , future work can now address how important processing bodies are for regulating gene expression . Another challenge for the future will be to examine the specific roles that processing bodies play in yeast and other cells , like human egg cells or nerve cells .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "biochemistry", "and", "chemical", "biology", "cell", "biology" ]	2016	ATPase activity of the DEAD-box protein Dhh1 controls processing body formation
Apical secretion from epithelial tubes of the Drosophila embryo is mediated by apical F-actin cables generated by the formin-family protein Diaphanous ( Dia ) . Apical localization and activity of Dia are at the core of restricting F-actin formation to the correct membrane domain . Here we identify the mechanisms that target Dia to the apical surface . PI ( 4 , 5 ) P2 levels at the apical membrane regulate Dia localization in both the MDCK cyst model and in Drosophila tubular epithelia . An N-terminal basic domain of Dia is crucial for apical localization , implying direct binding to PI ( 4 , 5 ) P2 . Dia apical targeting also depends on binding to Rho1 , which is critical for activation-induced conformational change , as well as physically anchoring Dia to the apical membrane . We demonstrate that binding to Rho1 facilitates interaction with PI ( 4 , 5 ) P2 at the plane of the membrane . Together these cues ensure efficient and distinct restriction of Dia to the apical membrane . Epithelial cells that comprise tubular organs are highly polarized , a feature that enables them to execute functions such as vectorial secretion and absorption of nutrients . Polarization is apparent in the distinct composition of membrane domains: the apical membrane—the surface facing the lumen , the basal membrane , which contacts the underlying extracellular matrix ( ECM ) , and the lateral surfaces , which contain specialized cellular junctions that adhere adjacent cells ( Bryant and Mostov , 2008 ) . The generation and maintenance of cell polarity is achieved by domain-specific proteins and lipids , which support the unique organization and function of each region . Among these , asymmetric distribution of phosphoinositides has been shown to be crucial for membrane identity and lumen formation in tubular systems . PI ( 4 , 5 ) P2 in the apical surface and PI ( 3 , 4 , 5 ) P3 in the basal membrane have been shown to tether specific polarity and cytoskeleton related proteins , which define their respective domains ( Martin-Belmonte and Mostov , 2007 ) . Cytoskeletal structures play key guidance roles underlying maintenance of epithelial cell polarity . They perform these functions by serving as membrane scaffolds , supporting adhesion , and enabling vesicle transport ( Nance and Zallen , 2011; Tepass , 2012 ) . One such structure , which is a common feature of tubular tissues , is a network of actin microfilaments lining the apical surface of the tube cells . In a previous study we found that in Drosophila tubular organs , this network mediates myosinV based transport of vesicles , promoting their secretion from the apical surface into the tube lumen . The actin-nucleator responsible for generating these structures was shown to be the formin-family protein Diaphanous ( Dia ) ( Massarwa et al . , 2009 ) . Apical restriction of Dia activity in this context is the consequence of tight apical localization of the Dia protein , which was shown to be a common feature of all epithelial cells generating the different Drosophila embryonic tubular organs . Thus , apical targeting of Dia is at the core of a cellular mechanism generating actin cables that emanate from the apical membrane , and enable apical secretion . While the delivery of apical and baso-lateral transmembrane proteins through specialized routes of the secretory pathway has been well studied ( Weisz and Rodriguez-Boulan , 2009 ) , much less is known about the targeting of cytoplasmic proteins such as Dia to distinct membrane domains . Dia belongs to the formin family of actin nucleators , which regulate the formation of linear actin cables . The Dia-related formins ( DRFs ) can be functionally divided into two major domains , each encompassing roughly one half of the protein sequence ( Goode and Eck , 2007 ) . The C-terminal portion of DRFs regulates actin polymer assembly by mediating microfilament nucleation , elongation and processive capping . Key functional sub-domains include the FH2 domain , which acts as a dimer , and moves processively with the growing barbed end , and the FH1 domain , which together with profilin acts to accelerate filament elongation by recruiting monomeric actin . The N-terminal portion of DRF nucleators is regulatory , governing the activation state of the molecule through interactions with various effectors . Importantly , this region has been shown to play significant roles in directing DRF localization in vivo ( reviewed in Higgs , 2005; Chesarone et al . , 2010 ) . DRFs are autoinhibited due to an intra-molecular interaction between the C-terminal DAD domain and the N-terminal DID domain , which maintains the molecule in a closed conformation . Upon binding of GTP-bound Rho1 to the N-terminal GTPase-binding domain , autoinhibition is relieved , allowing Dia to assume an open , active conformation and promote actin polymerization ( Lammers et al . , 2005; Otomo et al . , 2005; Rose et al . , 2005 ) . DRFs have been shown to play critical roles in the formation of diverse actin-based structures in many cellular contexts , including cytokinesis , cell adhesion , polarized cell growth and cell migration ( Faix and Grosse , 2006 ) . Accordingly , Dia-like proteins assume distinct subcellular localizations in a cell type specific manner . To prevent spatially unrestricted actin assembly in cells , localization of DRFs is highly regulated , and involves different factors and mechanisms . For example , the yeast formin Bni1p localizes during budding to the mother cell bud neck , by binding to the polarity factor Spa2 , via a mechanism that involves four different localizing domains ( Liu et al . , 2012 ) . Targeting of mDia2 , a mammalian homologue of Dia , to the cleavage furrow during cytokinesis , is achieved by binding to the scaffold protein anillin ( Watanabe et al . , 2010 ) . In Drosophila , Dia is recruited to lamellipodia by the actin regulator Ena/Vasp in a small subset of epidermal cells , during the process of embryonic dorsal closure ( Homem and Peifer , 2009 ) . In Drosophila tubular structures , dia mRNA is tightly concentrated at the apical side of the cells , by a mechanism that was shown to be distinct from the Dia protein apical targeting mechanism ( Massarwa et al . , 2009 ) . Thus , targeting of Dia-like proteins is regulated at multiple levels , to assure restriction of activity to the correct cellular domain . Here we examine the mechanisms that mediate apical targeting of the Dia protein in tubular organs of Drosophila . We find that efficient apical targeting of Dia requires the detection of both PI ( 4 , 5 ) P2 and Rho1 on the apical surface . We show , both in a mammalian cell culture model and by genetic manipulations in Drosophila tubular organs , that PI ( 4 , 5 ) P2 levels regulate Dia localization . An N-terminal basic domain of Dia is critical for this regulation , indicative of an electrostatic-based direct interaction between Dia and PI ( 4 , 5 ) P2 . We demonstrate that apical enrichment of PI ( 4 , 5 ) P2 is a common feature of tubular organs in the fly , which can be attributed in part to the apical restriction of the PIP5 kinase ( PIP5K ) Skittles . We further show that Rho1 facilitates Dia apical targeting , both by inducing the Dia open conformation , thereby exposing N-terminal domains that are critical for localization , as well as by physically anchoring Dia to the apical surface . This anchoring appears to promote the Dia–PI ( 4 , 5 ) P2 interaction , leading to synergistic effects of these two apical cues . We therefore propose that a multi-tiered targeting mechanism , utilizing both global and specific features of tubular organs , achieves a distinct localization pattern of Dia , thereby enabling execution of the cardinal process of apical secretion in these organs . Endogenous Dia exhibits a highly polarized distribution , accumulating at the apical surface of epithelial cells in diverse tubular tissues of the Drosophila embryo ( Figure 1A–C; Massarwa et al . , 2009 ) . To dissect and quantitatively assess the structural basis for this polarized distribution , we monitored the localization patterns of different GFP-tagged Dia constructs ( Figure 2A ) , following tissue-specific expression in several tubular organs . These constructs do not contain the 3′UTR sequences mediating RNA localization , and thus reflect targeting at the protein level . Upon expression at similar levels to endogenous Dia ( Figure 1—figure supplement 1 ) , a full-length Dia construct ( GFP-Dia-FL ) displayed an apically polarized distribution , albeit less dramatic than the endogenous protein ( Figure 1D–F ) . Measurements of GFP intensity in embryonic salivary gland cells further substantiated the enrichment of GFP-Dia-FL at the apical surface , relative to both the cell cytoplasm and to the septate junction ( SJ ) domain , representing the lateral membrane ( Figure 1J–L and Figure 1—figure supplement 1 ) . 10 . 7554/eLife . 00666 . 003Figure 1 . The open conformation enhances Dia apical localization . ( A–C′ ) Endogenous Dia staining ( red , shown separately [grey] in single-primed panels ) shows apical localization in the trachea ( A ) , hindgut ( HG , B ) and salivary gland ( SG , C ) of stage 15 embryos . Trh ( blue , A ) marks nuclei of tracheal cells . Crumbs ( blue , B and C ) marks the apical surface . Apical surfaces face the lumen ( L ) . ( D–I′ ) The localization of GFP-tagged Dia constructs ( Figure 2A , green and shown separately [grey] in single-primed panels ) was monitored following expression in stage 14–15 embryonic trachea , hindgut and salivary gland under btl-Gal4 ( D and G ) , drm-Gal4 ( E and H ) and fkh-Gal4 ( F and I ) , respectively . Trh ( blue , D and G ) marks nuclei of tracheal cells . Dlg and FasIII ( blue , E , F , H , I ) mark septate junctions . GFP-Dia-ΔDAD ( G–I′ ) is more apically localized than GFP-Dia-FL ( D–F′ ) . Scale bars , 10 μm . ( J–L ) Quantification of GFP fluorescence-intensity ratios . All measurements were carried out on confocal images of the different constructs expressed in the salivary gland . ( J and K ) Illustration of the different cellular domains measured in GFP-Dia-FL ( J ) and GFP-Dia-ΔDAD ( K ) expressing salivary gland cells . A–apical domain ( outlined in yellow ) , SJ–septate junctions ( outlined in blue ) , C–cytoplasm ( outlined in white ) , N–nucleus . ( L ) Apical enrichment quantification represented by the mean pixel intensity ratio between the apical domain and the cytoplasm ( A/C ) or the septate junctions region ( A/SJ ) . Error bars represent standard error . n = 20 . See also Figure 1—figure supplement 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 00666 . 00310 . 7554/eLife . 00666 . 004Figure 1—figure supplement 1 . Characteristics of Dia apical enrichment . ( A–A″ ) Dia protein levels are similar ( ∼two fold higher , quantification not shown ) in tracheal cells expressing the GFP-Dia-FL construct and in the adjacent posterior spiracle cells ( S ) . GFP-Dia-FL was expressed in stage15 embryonic tracheal cells under btl-Gal4 and localization of the endogenous protein ( in the S region ) as well as the expressed protein ( marked with GFP ) , was followed by staining with both anti-Dia ( red , A′ ) and anti-GFP ( green , A″ ) antibodies . Trh ( blue ) marks nuclei of tracheal cells . Scale bar , 10 µm . ( B ) Expression levels of GFP-Dia-FL and GFP-DiaΔDAD are similar . ( a and b ) Live imaging of whole embryos expressing the indicated constructs under drm-Gal4 showing the hindgut and Malpighian tubules . Scale bar , 20 μm . ( c ) Lysates of embryos expressing the indicated constructs were probed with an anti-GFP antibody and anti-Actin as a loading control . GFP-Dia-FL and GFP-DiaΔDAD are apparent at ≈150 kDa and ≈145 kDa , respectively ( arrowhead ) , but are absent from lysates of yw embryos . ( C ) Quantification of GFP intensity in salivary gland cells expressing GFP-Dia-FL ( blue bars ) and GFP-Dia-ΔDAD ( red bars ) ( Figure 1F , I and J , K ) . Left chart: percentage of intensity that is concentrated in the apical membrane and cytoplasmic domains from the total intensity in the cell . Right chart: percentage of intensity that is concentrated in the lateral membrane domain ( represented by the septate junctions [SJ] ) , from the total intensity in the cell . While a significant shift in GFP fluorescence levels from the cytoplasm to the apical domain is observed between GFP-Dia-FL and GFP-Dia-ΔDAD expressing cells , GFP fluorescence levels in the lateral membrane domain remain the same . Error bars represent standard error , n = 19–20 . DOI: http://dx . doi . org/10 . 7554/eLife . 00666 . 00410 . 7554/eLife . 00666 . 005Figure 2 . The N-terminal domain of Dia is necessary and sufficient for apical targeting . ( A ) Schemes of Dia constructs tagged with GFP at the N-terminus ( Homem and Peifer , 2009 and this study , see ‘Materials and methods’ ) . Shown for GFP-Dia-BD is a global pairwise amino acid sequence alignment of the basic domain ( BD ) between Drosophila Dia and the mammalian homologue mDia1 ( Needle , EMBOSS package , default parameters ) . Basic residues are marked in red . BD , Basic Domain; G , GTPase-binding domain; DID , Diaphanous Inhibitory Domain; DD , Dimerization Domain; FH1 and FH2 , Formin Homology domains 1 and 2; DAD , Diaphanous Autoregulatory Domain . ( B–D′ ) Stage 15 embryonic salivary glands expressing GFP-tagged Dia constructs ( A ) under drm-Gal4 ( green , shown separately ( grey ) in single-primed panels ) . Dlg and FasIII ( blue ) mark septate junctions . ( B and B′ ) The GFP-Dia-C’ construct displays cytoplasmic distribution . ( C and C′ ) The GFP-Dia-N’ construct is highly restricted to the apical domain . ( D and D′ ) The dimerization domain ( DD ) does not contribute to apical targeting of the Dia-C’ domain , as GFP-Dia-DD-C’ distribution is completely cytoplasmic ( compare with B ) . Some cells appear to contain two nuclei , which may reflect an effect of this constitutively-activated Dia construct on cytokinesis . Scale bars , 10 μm . ( E ) Quantification of GFP fluorescence intensity in ( B–D ) , represented by the mean pixel intensity ratio between the apical domain and the cytoplasm . Error bars represent standard error . n = 7–11 . See also Figure 2—figure supplement 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 00666 . 00510 . 7554/eLife . 00666 . 006Figure 2—figure supplement 1 . Dia constructs show similar localization patterns in all tubular organs . ( A–F′ ) Stage 14–15 embryonic trachea and hindgut expressing GFP tagged Dia constructs ( Figure 2A ) under btl-Gal4 and drm-Gal4 , respectively . Localization was followed by staining for GFP ( green , shown separately [grey] in single primed panels ) . Trh ( blue ) marks nuclei of tracheal cells . Dlg and FasIII ( blue ) mark septate junctions . ( A–B′ ) The GFP-Dia-C’ construct displays a cytoplasmic distribution . ( C–D′ ) The GFP-Dia-N’ construct is highly restricted to the apical domain . ( E–F′ ) The dimerization domain ( DD ) does not contribute to apical targeting of the Dia-C’ domain , as GFP-Dia-DD-C’ distribution is cytoplasmic . DOI: http://dx . doi . org/10 . 7554/eLife . 00666 . 006 Removal of the extreme C-terminus of Dia , which includes the DAD domain ( GFP-Dia-ΔDAD ) , generates an open form of Dia , which is relieved from auto-inhibition ( Li and Higgs , 2003 ) . While expression of this construct in tubular epithelia results in protein levels similar to the full-length form ( Figure 1—figure supplement 1 ) , GFP-Dia-ΔDAD displays a much tighter apical localization pattern ( Figure 1G–I ) , suggesting that the molecule is now more accessible to apical-targeting signals . Quantification demonstrates that apical enrichment of GFP-Dia-ΔDAD is enhanced up to fourfold relative to GFP-Dia-FL ( Figure 1J–L ) . Further measurements in GFP-Dia-ΔDAD expressing cells reveal that ∼45% of the total GFP intensity in the cell is concentrated at the apical surface , which encompasses only ∼5% of the total cell area . When compared with the GFP-Dia-FL expressing cells , a clear shift in the GFP distribution is observed from the cytoplasm to the apical domain . In contrast , a similar degree of localization to the lateral membrane SJ domain is exhibited by both constructs ( Figure 1—figure supplement 1 ) . Thus , enhanced apical localization of GFP-Dia-ΔDAD appears to result primarily from efficient apical recruitment of its cytoplasmic pool . The intensity differences between the apical and cytoplasmic domains were therefore used in this study as the primary measure for apical enrichment of Dia constructs . To identify the domains within Dia that mediate apical localization of the open form , we examined the localization patterns of different reporter constructs ( Figure 2A ) . While these assays focused on the embryonic salivary gland , qualitatively similar observations were made in other tubular organs ( Figure 2—figure supplement 1 ) . We initially examined separately the localizing activity of the two halves of the Dia-ΔDAD protein , each of which harbors distinct functional domains . A construct containing only the C-terminal half ( GFP-Dia-C′ ) was distributed within the cytoplasm upon expression in tubular epithelia ( Figure 2B and Figure 2—figure supplement 1 ) . In contrast , the reciprocal N-terminal construct ( GFP-Dia-N′ ) displayed striking apical enrichment , reminiscent of the localization pattern of the intact GFP-Dia-ΔDAD open molecule ( Figure 2C and Figure 2—figure supplement 1 ) . Indeed , quantification of relative fluorescence levels demonstrated strong apical enrichment of GFP-Dia-N′ , while no apical bias whatsoever was observed for GFP-Dia-C′ ( Figure 2E ) . These observations suggest that the N-terminal half of Dia is both necessary and sufficient for apical targeting . Since the Dia protein functions as a dimer , one concern is that the localization of the tested constructs will be influenced by dimerization to the endogenous , full-length protein . However , when the dimerization domain ( DD ) was added to the Dia-C′ construct ( GFP-Dia-DD-C′ ) , no apical localization was detected ( Figure 2D–E and Figure 2—figure supplement 1 ) , demonstrating that apical localization of Dia constructs is not biased by dimerization to endogenous Dia . Taken together , this structure–function analysis supports a model whereby the open conformation of Dia facilitates efficient apical localization by exposing critical domains to targeting cues . The Dia-targeting regions are restricted to the N-terminal half of the protein . The striking difference in the apical distribution of Dia between tubular organs and other non-polar tissues suggested the presence of an inherent apical cue , unique to tubular epithelial tissues . Phosphoinositides are general regulators of membrane identity in tubular tissues . In 3D cysts of cultured Madin–Darby canine kidney ( MDCK ) cells , a well-established system for recapitulation of epithelial tube morphogenesis , PI ( 4 , 5 ) P2 is a key determinant of the apical membrane , where it is highly enriched ( Martin-Belmonte et al . , 2007 ) . We therefore utilized the MDCK cyst system , in which phosphoinositide levels can be readily manipulated , to examine a possible role for PI ( 4 , 5 ) P2 in the regulation of Dia localization . Both the full-length and open forms of Drosophila Dia , tagged with GFP , were stably expressed in MDCK cyst culture , which were subsequently FACS-sorted , to ensure selection of cysts displaying similar GFP expression levels . The fluorescent GFP signal allowed us to monitor localization of these constructs following cyst maturation . Drosophila GFP-Dia-FL failed to localize to a specific subcellular domain , and was homogenously distributed in the cytoplasm of cyst cells ( Figure 3A ) . In marked contrast , Drosophila GFP-Dia-ΔDAD was targeted apically , co-localizing with the apical marker gp-135 ( Figure 3B ) . 10 . 7554/eLife . 00666 . 007Figure 3 . PI ( 4 , 5 ) P2 levels regulate Drosophila Dia localization in MDCK cysts . ( A–B′ ) The localization of GFP-tagged Dia constructs ( green , shown separately in single-primed panels ) was monitored following stable expression in MDCK cyst culture . GFP-Dia-FL ( A ) displays an entirely cytoplasmic distribution , while GFP-Dia-ΔDAD ( B ) is highly restricted apically . ( C–F′ ) GFP-Dia-ΔDAD expressing MDCK cysts were incubated with histone ( no lipid , C–D′ ) or PI ( 4 , 5 ) P2-histone complexes ( +PI ( 4 , 5 ) P2 , E–F′ ) for up to 10 min . Exogenous PI ( 4 , 5 ) P2 re-localized GFP-Dia-ΔDAD to the basal membrane ( E and magnified in F ) in comparison to control cysts ( C and magnified in D ) . Apical enrichment is slightly reduced in the control cysts ( compare C with B ) , probably due to pre-incubation with Trypsin and low-calcium levels . Nuclei are stained with Hoechst . Apical surfaces face the lumen ( L ) . Scale bars , 10 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 00666 . 007 The localization features of the open form of Drosophila Dia were therefore recapitulated in the heterologous MDCK system , consistent with the notion that apical targeting is mediated by a conserved element such as PI ( 4 , 5 ) P2 . To directly test the involvement of PI ( 4 , 5 ) P2 in Dia localization , we delivered exogenous PI ( 4 , 5 ) P2 to the basal surface of cysts expressing GFP-Dia-ΔDAD ( Gassama-Diagne et al . , 2006 ) , and monitored the effects on its localization pattern . GFP-Dia-ΔDAD localization was dramatically altered , even after very short treatments of 10 min , displaying redistribution from the apical to the basal surface , while remaining apical in mock treated cysts ( Figure 3C–F ) . The rapid shift in GFP-Dia-ΔDAD localization preceded the general shift in polarity induced by exogenous PI ( 4 , 5 ) P2 treatment in this system ( Martin-Belmonte et al . , 2007 ) . This can be readily appreciated by the partial basolateral localization of the apical membrane marker gp-135 , which no longer co-localized with GFP-Dia-ΔDAD following treatment ( Figure 3E , F ) . The rapid nature of Dia re-localization suggests that recruitment to the baso-lateral membranes is a direct result of exogenous PI ( 4 , 5 ) P2 treatment . These observations imply that PI ( 4 , 5 ) P2 could fulfill a prominent role in the regulation of Dia apical targeting in tubular structures . We next asked whether PI ( 4 , 5 ) P2 also serves as an apical cue for Dia localization in Drosophila tubular organs . To follow the endogenous distribution of PI ( 4 , 5 ) P2 , we expressed PH-PLCδ-GFP , a PI ( 4 , 5 ) P2 sensor in which the PH domain of PLCδ is tagged with GFP ( Downes et al . , 2005 ) , in different tubular organs . Imaging the GFP fluorescence in live tissues showed significant apical enrichment of the sensor in the embryonic hindgut and larval salivary gland cells ( Figure 4B , D ) , which was apparent but less pronounced in the embryonic trachea ( Figure 4C ) . Apical membranes of Drosophila tubular epithelia are therefore enriched in PI ( 4 , 5 ) P2 , as in mammalian systems . In general , imaging of these fine subcellular domains in live samples produced clearer and more consistent localization patterns than fixed preparations , leading us to use the live imaging approach whenever applicable , throughout the analysis . 10 . 7554/eLife . 00666 . 008Figure 4 . PI ( 4 , 5 ) P2 and Skittles , a PIP5 kinase , are apically enriched in Drosophila tubular organs . ( A ) Simplified scheme of the PI ( 4 , 5 ) P2 biosynthetic pathway . PI ( 4 ) P is phosphorylated by Sktl /PIP5K to generate PI ( 4 , 5 ) P2 . PI ( 4 , 5 ) P2 can be further phosphorylated by PI3K to produce PI ( 3 , 4 , 5 ) P3 , which can be dephosphorylated by PTEN phosphatase to regenerate PI ( 4 , 5 ) P2 . ( B–D ) Live imaging of PH-PLCδ-GFP following expression in stage 14 embryonic hindgut ( B ) and trachea ( C ) under drm-Gal4 and btl-Gal4 , respectively , and in third instar larval salivary gland FLP-out clones ( D ) . ( D ) Salivary gland FLP-out clones expressing PH-PLCδ-GFP under actin-Gal4 . Clones are marked by nuclear RFP ( red ) , and the salivary gland outline is visualized using transmitted light . Inset shows sensor distribution within a single magnified clone cell . PH-PLCδ-GFP is enriched at the apical surface of these tubular organs . ( E–J ) Live imaging of Sktl-RFP and Sktl-KID-RFP , following expression in the embryonic hindgut ( E and H ) and trachea ( F and I ) under drm-Gal4 and btl-Gal4 , respectively , and in the 2nd instar larval salivary gland under fkh-Gal4 ( G and J ) . Sktl is enriched at the apical domain , while a kinase-dead form ( Sktl-KID ) is localized throughout the cell cortex . Scale bars , 10 μm , and 20 μm ( D ) . DOI: http://dx . doi . org/10 . 7554/eLife . 00666 . 008 Local production of PI ( 4 , 5 ) P2 at the apical membrane by PIP5-kinases could serve as a means for enrichment at this site . Skittles ( Sktl ) , one of the two Drosophila PIP5-kinases which catalyze PI ( 4 , 5 ) P2 production from PI ( 4 ) P ( Figure 4A ) , was shown to be critical for PI ( 4 , 5 ) P2 production in various systems ( Gervais et al . , 2008; Fabian et al . , 2010 ) , and is endogenously expressed in different Drosophila embryonic tubular organs ( BDGP-Berkeley Drosophila Genome Project ) . Indeed , when we expressed an RFP-tagged version of Sktl in Drosophila tubular organs , it was targeted predominantly to the apical surface of cells ( Figure 4E–G ) . Thus , apical targeting of Sktl may account for PI ( 4 , 5 ) P2 accumulation at this membrane . It was proposed that PIP5-kinases are targeted to the membrane through electrostatic interactions between basic residues in their sequence and acidic lipids on the membrane , including PI ( 4 , 5 ) P2 itself ( Kwiatkowska , 2010 ) . Supporting this notion is the observation that a mutated form of PIP5-kinase , containing an inactive kinase domain , fails to localize to the same membrane domain as the wild-type protein ( Fairn et al . , 2009 ) , demonstrating that continuous PI ( 4 , 5 ) P2 generation is required to enhance the localization bias of PIP5-kinases . We find a similar relationship in Drosophila tubular organs between Sktl activity and apical localization: a mutated form of Sktl , where the kinase domain is inactive ( Sktl-KID ) , was no longer enriched apically , and instead assumed a general cortical distribution ( Figure 4H–J ) . We assume that the excess of Sktl-KID molecules may sequester endogenous levels of PI ( 4 , 5 ) P2 at the apical membrane , compromising localization of Sktl and other PI ( 4 , 5 ) P2-binding proteins , and leading to a disrupted morphology of tubular epithelia ( not shown ) . Taken together , these observations are consistent with a scenario where apically-biased Sktl generates PI ( 4 , 5 ) P2 at the apical membrane of tubular organs . Elevated apical PI ( 4 , 5 ) P2 then serves to reinforce apical targeting of Sktl , and provides an apical cue for additional proteins such as Dia . We next used Sktl overexpression in Drosophila tubular organs to manipulate PI ( 4 , 5 ) P2 levels and determine their capacity to influence Dia localization . The partial apical localization of GFP-Dia-FL in tracheal cells ( Figures 1D and 5A ) is barely detectable at late embryonic stages ( Figure 5C ) , but is significantly enhanced upon co-expression with Sktl-RFP ( Figure 5A–D ) . Corresponding measurements of GFP intensities show a clear shift in GFP-Dia-FL distribution from the cytoplasm to the apical domain in Sktl-expressing cells ( Figure 5G ) . Conversely , apical enrichment was not observed following co-expression of Sktl with the GFP-Dia-C′ construct , which does not contain apical targeting activity ( Figure 5E ) . Moreover , GFP-Dia-FL localization was not affected by co-expression with the catalytically inactive form of Sktl ( Figure 5F ) . These results suggest that Sktl-dependent increase in PI ( 4 , 5 ) P2 levels at the apical membrane leads to enhanced recruitment of Dia . 10 . 7554/eLife . 00666 . 009Figure 5 . PI ( 4 , 5 ) P2 levels regulate Dia apical localization in Drosophila tubular epithelia . ( A–F′ ) Localization of the indicated constructs was monitored by staining with antibodies against GFP and Ds-Red , following expression in the embryonic trachea under btl-Gal4 . Primed panels ( grey ) show the GFP channel alone . Trh ( blue ) marks tracheal cell nuclei . ( A–D′ ) The localization of GFP-Dia-FL is shifted from the cytoplasm to the apical surface following co-expression with Sktl-RFP in stage 13 ( B and B′ ) and stage 15 ( D and D′ ) tracheal cells . Compare with the control embryos expressing GFP-Dia-FL alone ( A , A′ and C , C′ ) . ( E , E′ ) When co-expressed with Sktl-RFP , the localization of GFP-Dia-C’ remained entirely cytoplasmic . ( F and F′ ) Apical enrichment of GFP-Dia-FL is not enhanced upon co-expression with Sktl-KID-RFP . Scale bars , 10 μm . ( G ) Quantification of GFP fluorescence intensity in stage 15 tracheal cells . Upper chart: mean pixel intensity ratio between the apical domain and the cytoplasm . Lower chart: percentage of intensity that is concentrated in the apical and cytoplasmic domains from the total intensity in the cell . Error bars represent standard error . n = 11–20 . ( H–K″ ) Localization of the indicated constructs was examined by live imaging in third instar larval salivary glands following expression under fkh-Gal4 . ( H ) In WT cells , the PI ( 4 , 5 ) P2 sensor PH-PLCδ-GFP is highly enriched in the apical membrane . ( I ) Upon co-expression with Pten-RNAi , PH-PLCδ-GFP is re-distributed between the cell membranes , and appears cortical . ( J–K″ ) Co-expression of GFP-Dia-N’ ( green ) with CD8-RFP ( red ) , which served as a general apical membrane marker . Separate channels ( grey ) are shown in primed panels . ( J–J″ ) In WT cells , both GFP-Dia-N’ and CD8-RFP are highly restricted apically . ( K–K″ ) In Pten-RNAi expressing cells , GFP-Dia-N’ is re-distributed between the cell membranes and appears cortical , while CD8-RFP remains restricted to the apical membrane . Scale bars , 20 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 00666 . 009 An additional regulator of PI ( 4 , 5 ) P2 levels in the membrane is the lipid phosphatase PTEN , which dephosphorylates PI ( 3 , 4 , 5 ) P3 to form PI ( 4 , 5 ) P2 , thus acting as a negative regulator of the PI3K pathway ( Figure 4A ) . PTEN has been shown to be associated with the regulation of cell polarity in both Drosophila and mammalian systems ( Martin-Belmonte et al . , 2007; Chartier et al . , 2011 ) . PTEN resides at adherens junctions ( von Stein et al . , 2005 ) , consistent with a role in the maintenance of PI ( 4 , 5 ) P2 levels at the apical membrane . We therefore used RNAi-mediated silencing of Pten to manipulate PI ( 4 , 5 ) P2 levels in vivo , and followed the effect on Dia localization . We chose to perform these studies by imaging of live third instar larval salivary glands . This system offers several favorable features , including large polarized cells with clearly distinguishable domains , avoidance of fixation artifacts , and a relatively long period for depletion of gene products through an RNAi-based approach . Monitoring of the GFP-Dia-N′ construct expressed in this tissue revealed a highly restricted localization to the apical surface of the cells lining the gland lumen ( Figure 5J ) , consistent with our observations in the embryonic salivary gland and other tubular organs . The PI ( 4 , 5 ) P2 sensor PH-PLCδ-GFP is localized apically in larval salivary glands , similar to other tubular organs ( Figure 5H ) . A pronounced redistribution of the sensor was observed following co-expression of Pten-RNAi , such that the sensor became evenly localized around the cortex of salivary-gland cells ( Figure 5I ) . This redistribution is consistent with studies in MDCK cysts that link the loss of PTEN activity with failure in segregation of PI ( 4 , 5 ) P2 from PI ( 3 , 4 , 5 ) P3 , leading to their homogenous , rather than polarized , membrane distributions ( Martin-Belmonte et al . , 2007 ) . Similar to the effect on the PI ( 4 , 5 ) P2 sensor localization pattern , expression of a Pten-RNAi construct caused a dramatic redistribution of the GFP-Dia-N′ construct , from a restricted apical localization to a homogenous distribution throughout the plasma membrane of third instar salivary gland cells ( Figure 5J–K ) . To rule out the possibility of secondary effects resulting from a general disruption of cell polarity , we co-expressed an RFP-tagged version of the membrane protein CD8 ( CD8-RFP ) , which resides exclusively in the apical membrane of salivary gland cells ( Xu et al . , 2002 ) . The localization of the CD8-RFP construct remained apically restricted following Pten knockdown , indicating that the loss of PTEN specifically affected Dia localization ( Figure 5J″–K″ ) . Distribution of myristoylated RFP , another apical marker in this tissue , was similarly unaffected ( not shown ) . In summary , manipulations of PI ( 4 , 5 ) P2 levels profoundly and specifically affect Dia localization in vivo , demonstrating the involvement of PI ( 4 , 5 ) P2 in Dia apical targeting . The construct containing the N-terminal half of Dia ( Dia-N′ ) faithfully recapitulated apical targeting ( Figure 2 ) , thus providing a reporter for the localization of Dia . Which domain within this construct mediates the interaction with PI ( 4 , 5 ) P2 ? Previous reports showed that the mammalian homologues of Dia , mDia1 and mDia2 , directly bind acidic phospholipids via electrostatic interactions , involving basic stretches near the N-terminus of the protein , termed the Basic Domain ( BD ) ( Ramalingam et al . , 2010; Gorelik et al . , 2011 ) . Indeed the Drosophila Dia N-terminal 60 amino acids are also enriched for basic residues ( Figure 2A ) . To examine a role for this region in apical targeting of Drosophila Dia , we generated transgenic flies that contain a variant GFP-Dia-N′ reporter lacking its N-terminal basic domain ( GFP-Dia-N′ΔBD ) . The same chromosomal insertion site as the intact GFP-Dia-N′ reporter was used , to maintain similar expression levels . Deletion of the BD profoundly reduced apical targeting of the GFP-Dia-N′ reporter in larval salivary glands , as well as in other tubular tissues ( Figure 6A–B , 7G and Figure 6—figure supplement 1 ) , demonstrating its significance for Dia localization . 10 . 7554/eLife . 00666 . 010Figure 6 . PI ( 4 , 5 ) P2-Dia interaction is mediated by direct binding through the BD domain . Live imaging of GFP-tagged Dia constructs ( Figure 2A ) , expressed in third instar larval salivary glands by fkh-Gal4 . Primed panels ( grey ) show an enlargement of the marked area . ( A–B′ ) Apical localization of GFP-Dia-N’ΔBD ( B and B′ ) is significantly weaker than GFP-Dia-N’ ( A and A′ ) . ( C–D′ ) GFP-Dia-BD displays a cytoplasmic and nuclear distribution ( C and C′ ) , which is comparable to that of GFP alone ( D and D′ ) . The apparent enrichment in the apical domain , resulting from absence of granules in this area , as well as nuclear localization , is observed with both constructs . Scale bars , 20 μm . See also Figure 6—figure supplement 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 00666 . 01010 . 7554/eLife . 00666 . 011Figure 6—figure supplement 1 . Dia constructs show similar localization patterns in all tubular organs . ( A–H′ ) Stage 14–15 embryonic trachea , hindgut and salivary glands expressing the indicated constructs ( Figure 2A ) under btl-Gal4 , drm-Gal4 and fkh-Gal4 , respectively . Localization was followed by live imaging ( A , B and D , E ) or staining with anti-GFP ( C , F and G–H , green and shown separately in primed panels ) . FasIII ( blue ) marks septate junctions . Apical localization of GFP-Dia-N’ΔBD ( D–F′ ) is significantly reduced compared to GFP-Dia-N’ ( A–C′ ) in embryonic tubular tissues . GFP-Dia-BD ( G and G′ ) displays a completely cytoplasmic and nuclear distribution , similar to GFP alone ( H and H′ ) . Scale bars , 10 μm . ( I ) Quantification of GFP intensity in larval ( left chart ) and embryonic ( right chart ) salivary gland cells , expressing the indicated constructs under fkh-Gal4 ( Figure 6 and this figure ) . Apical enrichment is represented by the mean pixel intensity ratio between the apical domain and the cytoplasm . Error bars represent standard error , n = 14–21 . DOI: http://dx . doi . org/10 . 7554/eLife . 00666 . 01110 . 7554/eLife . 00666 . 012Figure 7 . Direct binding to Rho1 mediates Dia apical localization . ( A–F′ ) Live imaging of GFP-tagged constructs ( Figure 2A ) expressed in third instar larval salivary glands under fkh-Gal4 . Primed panels ( grey ) show an enlargement of the marked area . ( A and A′ ) GFP-Rho1 is distributed between the apical domain and the cytoplasm . ( B–D′ ) Apical localization of Dia-N’ is Rho1 dependent . Co-expression with Rho1-RNAi dramatically alters GFP-Dia-N’ localization from apical restriction ( B and B′ ) to a complete cytoplasmic localization ( C and C′ ) . Correspondingly , GFP-Dia-N’-mutG , a variant lacking Rho-1 binding capacity , assumes an entirely cytoplasmic distribution ( D and D′ ) . Note the expanded lumen of Rho1-RNAi expressing salivary glands ( C and C′ ) , previously reported for Rho1 mutant alleles ( Xu et al . , 2011 ) , indicative of the loss of active Rho1 in this system . ( E–F′ ) Apical localization of Dia-N’ΔBD is Rho1 dependent . Co-expression with Rho1-RNAi dramatically alters GFP-Dia-N’ΔBD localization to a cytoplasmic localization ( E and E′ , compare with Figure 6B ) , while GFP-Dia- N’ΔBD -mutG , lacking Rho-1 binding capacity , assumes an entirely cytoplasmic distribution ( F and F′ ) . Scale bars , 20 μm . ( G ) Quantification of GFP intensity in larval ( left chart ) and embryonic ( right chart ) salivary gland cells , expressing the indicated constructs under fkh-Gal4 ( Figures 6 and 7 , Figure 6—figure supplement 1 and Figure 7—figure supplement 2 ) . Apical enrichment is represented by the mean pixel intensity ratio between the apical domain and the cytoplasm . For larval salivary gland cells , the values are normalized to measurements of cells expressing cytoplasmic GFP . Due to morphological changes resulting from Rho1–RNAi expression , Dia apical localization is even lower than the baseline . Error bars represent standard error . n = 10–24 glands . ( H ) A model describing the dynamics of Dia targeting to the apical membrane in Drosophila tubular organs . a ) Enrichment of both PI ( 4 , 5 ) P2 and Dia at the apical membrane is an inherent feature of Drosophila tubular epithelial cells . b ) In a closed conformation , Dia assumes a cytoplasmic distribution . Dia binding to Rho1-GTP generates an initial apical bias , and promotes PI ( 4 , 5 ) P2 binding via exposure of the N-terminal region and enrichment near the apical membrane . The Dia- PI ( 4 , 5 ) P2 interaction further stabilizes apical membrane association . Dia can shift dynamically between three states , being bound to each of the apical cues alone , or to both simultaneously , increasing the probability for re-binding and therefore the efficiency of apical targeting . See also Figure 7—figure supplement 1–3 . DOI: http://dx . doi . org/10 . 7554/eLife . 00666 . 01210 . 7554/eLife . 00666 . 013Figure 7—figure supplement 1 . Apical enrichment of Rho1 is not dependent on Dia . The indicated constructs were expressed in third instar larval salivary gland under fkh-Gal4 , and GFP was visualized by live imaging . GFP-Rho1 distribution remains the same in control cells ( A and A′ ) and upon co-expression with a dia-RNAi construct ( B and B′ ) . Primed panels show an enlargement of the marked area . Scale bars , 20 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 00666 . 01310 . 7554/eLife . 00666 . 014Figure 7—figure supplement 2 . Rho1 binding is critical for Dia localization in embryonic tubular organs . ( A–H′ ) The indicated constructs ( Figure 2A ) were expressed in the trachea ( A–B′ ) , the hindgut ( C–D′ ) and the salivary gland ( E–F′ and G–H′ ) of stage 15 embryos under btl-Gal4 , drm-Gal4 and fkh-Gal4 , respectively . ( A–F′ ) GFP-Dia-N’-mutG assumes a cytoplasmic distribution in embryonic tubular organs ( B , D and F ) , in contrast to GFP-Dia-N’ , which is apically restricted ( A , C and E ) . ( G–H′ ) GFP-Dia-N’ΔBD-mutG assumes a cytoplasmic distribution in embryonic salivary glands ( H and H′ ) , in contrast to GFP-Dia-N’ΔBD , which displays some apical enrichment in addition to cytoplasmic localization ( G and G′ ) . Localization was followed by staining with anti-GFP ( green , shown separately [grey] in primed panels ) . Trh ( blue ) marks nuclei of tracheal cells . FasIII ( blue ) marks septate junctions . Scale bars , 10 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 00666 . 01410 . 7554/eLife . 00666 . 015Figure 7—figure supplement 3 . Rho1 and PI ( 4 , 5 ) P2 do not regulate each other’s localization pattern . ( A–B′ ) Over-expression of Sktl does not effect Rho1 apical localization upon co-expression in the trachea . The indicated constructs were expressed by btl-Gal4 , and their localization was followed in stage 14–15 embryonic trachea by staining for GFP ( green , shown separately [grey] in primed panels ) and RFP ( DsRed , red ) . Trh ( blue ) marks the tracheal cell nuclei . ( A and A′ ) GFP-Rho1 displays a localization pattern that is distributed between the cytoplasm and the cell cortex when expressed in the trachea , which does not change upon co-expression with Sktl ( B and B′ ) . Scale bars , 10 µm . ( C ) Quantification of GFP intensity in ( A , -Sktl ) compared with ( B , +Sktl ) . Left chart: mean pixel intensity ratio between the apical domain and the cytoplasm ( A/C ) or the lateral membrane domain ( represented by the septate junctions , A/SJ ) . Right chart: percentage of intensity that is concentrated in the apical , cytoplasmic and septate junctions domains from the total intensity in the cell . Error bars represent standard error , n = 10–18 . ( D–E″ ) Pten silencing does not effect Rho1 localization in the salivary gland . The localization of the indicated constructs was examined by live imaging ( green , shown separately [grey] in single primed panels ) , following expression in the third instar larval salivary gland under fkh-Gal4 . Upon co-expression with a Pten-RNAi construct , GFP-Rho1 remained moderately enriched apically ( E and E′ ) , as in control cells ( D and D′ ) . CD8-RFP , which served as a general marker for apical membrane morphology , remained restricted to the apical membrane as well ( D″–E″ ) . ( F ) Quantification of GFP intensity ratios in ( D , WT ) compared with ( E , Pten-i ) , represented by the mean pixel intensity ratio between the apical and basal domains ( A/B ) . Both genotypes exhibit apical enrichment represented by a >1 ratio . Error bars represent standard error , n = 13–17 . ( G ) Rho1 silencing does not effect PI ( 4 , 5 ) P2 localization . PH-PLCδ-GFP remains apically enriched upon co-expression with a Rho1–RNAi construct . Inset shows an enlargement of the marked area . Scale bars , 20 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 00666 . 015 To assess the apical targeting capacity of the BD on its own , a GFP-Dia-BD construct ( Figure 2A ) was expressed in larval salivary glands and other tubular tissues . Remarkably , no apical localization bias was observed for this protein , as was also confirmed by comparison to the localization of GFP alone , and measurements of GFP apical enrichment ( Figure 6C–D and Figure 6—figure supplement 1 ) . Thus , despite being critical for Dia localization , the basic domain is not sufficient on its own to drive apical targeting . The inability of the BD to drive apical targeting on its own , suggested that PI ( 4 , 5 ) P2 may function in combination with additional factors to target Dia apically . The N-terminus of Dia is composed of several distinct domains that can serve as targets for apical recruitment ( Figure 2A ) . Prominent among these is the GTPase-binding domain , through which Rho1 binds and activates the actin nucleation and elongation capacities of Dia ( Otomo et al . , 2005; Rose et al . , 2005 ) . An initial indication that Rho1 can serve as an apical recruitment factor for Dia came from examination of Rho1 protein localization . A GFP-Rho1 construct specifically expressed in third instar larval salivary gland cells showed clear enrichment at the apical surface ( Figure 7A ) , as did a GFP fusion protein expressed from the endogenous Rho1 promoter ( not shown ) . Apical enrichment of GFP-Rho1 was not altered upon co-expression with a dia-RNAi construct , indicating that it is not mediated by Dia protein or by actin structures generated by Dia ( Figure 7—figure supplement 1 ) . We next examined the localization pattern of GFP-Dia-N′ following depletion of Rho1 . The ability to follow the inherently open Dia-N′ construct , allowed to critically examine the role of Rho1 as a physical anchor for Dia , independent of its role as an activator . GFP-Dia-N′ localization was dramatically altered upon co-expression with a Rho1–RNAi construct , and displayed a uniform distribution in the cytoplasm ( Figure 7B–C ) . Furthermore , a similar outcome was obtained following expression of GFP-Dia-N′-mutG ( Figure 2A ) , a GFP-Dia-N′ variant , which is mutated at residues that confer Rho1 binding ( Otomo et al . , 2005; Rose et al . , 2005; Lammers et al . , 2008 ) . The distribution of GFP-Dia-N′-mutG was essentially cytoplasmic upon expression in larval salivary glands , as well as in various embryonic tubular tissues ( Figure 7D and Figure 7—figure supplement 2 ) , indicating that direct interaction with Rho1 mediates Dia targeting . Thus , apically-localized Rho1 functions to physically anchor Dia to the apical surface , in addition to opening the protein . We assume that Rho1 exerts both activities only in the active , GTP-bound form ( Rose et al . , 2005 ) . We have shown that binding to both Rho1 and PI ( 4 , 5 ) P2 contribute to apical targeting of Dia . We next wanted to explore the possible relationship between these two cues . First , Rho1 localization was not altered following different manipulations designed to modify PI ( 4 , 5 ) P2 levels and affect Dia localization , including Sktl over-expression and PTEN depletion ( Figure 7—figure supplement 3 ) . Similarly , when we monitored the PI ( 4 , 5 ) P2 reporter PH-PLCδ-GFP , it remained apically enriched following co-expression with a Rho1–RNAi construct ( Figure 7—figure supplement 3 ) , indicating that Rho1 does not directly regulate PI ( 4 , 5 ) P2 distribution . Second , we analyzed the inter-relationship between the two cues with respect to Dia apical localization . Previous studies of the mDia1–Rho1 interaction showed that the BD is dispensable for the formation of an mDia1–Rho1 complex ( Otomo et al . , 2005; Rose et al . , 2005 ) . Co-expression of GFP-Dia-N′ΔBD with Rho1-RNAi or expression of a GFP-Dia-N′ΔBD mutG construct resulted in complete loss of apical localization ( Figure 7E–G ) . This indicates that the residual apical localization observed for the GFP-Dia-N′ΔBD construct ( Figure 6B ) can be attributed to Rho1 binding , consistent with the notion of two distinct functional modules for Dia binding . However , the inability of the BD to direct even a mild bias of apical localization on its own ( Figure 6C ) , suggests that the Dia–PI ( 4 , 5 ) P2 interaction requires simultaneous binding of an apical cue through an additional domain . Finally , quantification of Dia apical enrichment in larval salivary gland cells revealed that the ultimate apical distribution of Dia is strongly affected by the combined influence of both cues . While the capacity to bind PI ( 4 , 5 ) P2 does not promote apical localization in the absence of Rho1 , it strongly enhances apical targeting of Dia when Rho1 is present ( Figure 7G , compare Dia-N′ΔBD with Dia-N′ ) . Similar results were obtained in embryonic salivary glands , where a construct defective in PI ( 4 , 5 ) P2 binding ( Dia-N′ΔBD ) showed weak apical enrichment ( ≈2 ) , while a construct defective in Rho1 binding ( Dia-N′ΔBD-mutG ) showed no apical bias whatsoever ( Figure 7G ) . However , the Dia-N′ construct , in which the binding domains for both cues are intact , displayed apical enrichment values of ≈7 , clearly indicative of synergy between Rho1 and PI ( 4 , 5 ) P2 in promoting Dia apical localization . Taken together , we show that a partial apical localization bias of Dia by Rho1 binding is significantly enhanced by binding to PI ( 4 , 5 ) P2 . The enrichment of PI ( 4 , 5 ) P2 in apical membranes of tubular organs ( Martin-Belmonte et al . , 2007 ) suggested that PI ( 4 , 5 ) P2-binding could serve as a common localization cue for Dia , shared by different tubular organs . Indeed , a reporter for the endogenous levels of PI ( 4 , 5 ) P2 displayed a highly significant apical enrichment in Drosophila tubular organs ( Figure 4 ) . The causal role of PI ( 4 , 5 ) P2 in recruiting Dia was demonstrated in several ways . First , Drosophila Dia was apically localized in the heterologous MDCK cell culture system , and its localization was shifted basally within minutes of adding PI ( 4 , 5 ) P2 to the basal membranes ( Figure 3 ) . We wish to point out that while in this system Dia localization is regulated by PI ( 4 , 5 ) P2 under both excess and normal conditions , it is currently unknown whether an interaction between Dia and mammalian Rho GTPases also contributes to the apical localization . Second , elevation in Drosophila tubular organs in the levels of Sktl , a PIP5-kinase that generates PI ( 4 , 5 ) P2 , gave rise to a corresponding enhancement in the level of apical Dia . Conversely , knockdown of the Pten gene encoding the PI ( 3 , 4 , 5 ) P3 phosphatase in these tissues resulted in redistribution of PI ( 4 , 5 ) P2 , consistent with previous observations ( Martin-Belmonte et al . , 2007 ) . A corresponding alteration in Dia localization further supports the causal role of PI ( 4 , 5 ) P2 in its localization ( Figure 5 ) . It is interesting to note that generation of PI ( 4 , 5 ) P2 at the apical membrane by Sktl exhibits self-amplifying properties . Over-expression of tagged Sktl resulted in both a massive increase in the levels of apical PI ( 4 , 5 ) P2 ( not shown ) , and accumulation of Sktl at the same domain . In contrast , a Sktl protein that is catalytically inactive was distributed uniformly , when expressed in a similar manner ( Figure 4 ) . We assume that the signal targeting Sktl apically may be limited , and capable of providing only an initial bias . The resulting enrichment of PI ( 4 , 5 ) P2 following the activity of Sktl can lead to the recruitment of additional Sktl molecules and amplify this apical bias . The link between phospholipids and localization of formins extends to the plant kingdom . In the moss Physcomitrella patens , class-II formins bind PI ( 3 , 5 ) P2 , leading to their cortical localization , and allowing them to mediate polarized growth ( van Gisbergen et al . , 2012 ) . Taken together , phospholipids are emerging as important regulators of formin localization and activity at the tissue and organismal level . The mammalian formins mDia1 and mDia2 bind phospholipids through a basic domain ( BD ) , located at the extreme N-terminus . This BD was shown to regulate both activity and recruitment to membranes in cell culture models ( Ramalingam et al . , 2010; Gorelik et al . , 2011 ) . Similar to the mammalian homologues , Drosophila Dia also contains an N-terminal basic domain which is critical for apical localization ( Figure 6 ) , and is likely to mediate the association with PI ( 4 , 5 ) P2 through electrostatic interactions with the acidic phospholipid . These interactions are known to be weak , and are usually insufficient to stably associate proteins with the membrane under physiological conditions , requiring coupling to additional membrane-targeting motifs ( Mulgrew-Nesbitt et al . , 2006 ) . Indeed , the Dia BD proved insufficient to direct apical targeting on its own ( Figure 6 ) . In this respect , Dia differs from plant formins , which commonly contain a structured PTEN domain for phospholipid binding . This domain confers strong and more specific association , and is consequently sufficient for cortical localization , without requirement for additional targeting factors ( van Gisbergen et al . , 2012 ) . Insufficiency of the BD in mediating Dia apical targeting led us to search for additional targeting signals . The constitutively activated , open conformation of Dia was significantly more accessible to apical targeting ( Figure 1 ) , probably as a result of exposing the N-terminus more readily to apical-targeting factors . Interestingly , even high levels of the open conformation of Dia were efficiently targeted , indicating that the targeting machinery is in relative excess . Normally , the generation of the open Dia conformation is facilitated by binding to Rho1 ( Otomo et al . , 2005; Rose et al . , 2005 ) . In addition , for some formin-family proteins , phosphorylation of residues adjacent to the DAD domain by a Rho1 effector , ROCK-1 , is an essential element in the activation process ( Takeya et al . , 2008; Staus et al . , 2011 ) . It remains to be determined whether ROCK-1 also plays a role in activation of Dia in Drosophila tubular organs . Rho1 also contributes directly to apical-membrane anchoring of Dia . We were able to deplete Rho1 in larval salivary glands , and observed a complete loss of apical targeting ( Figure 7 ) . The ability to follow localization of an open form of Dia allowed us to separate the roles of Rho1 in Dia activation and localization . The open form of Dia still requires binding to Rho1 for its apical localization , as was shown by elimination of Rho1 and by mutating the Rho1-binding domain of Dia ( Figure 7 ) . There are indications for apical enrichment of Rho1 protein in Drosophila tubular organs ( Massarwa et al . , 2009 and Figure 7 ) . The distribution of the activated form of Rho1 may be even more apically biased . It appears , however , that Rho1 requires different signals than Dia for its apical localization . Alterations in the PI ( 4 , 5 ) P2 distribution , which resulted in dramatic effects on the localization of Dia , had only a marginal effect on the localization of Rho1 ( Figure 7—figure supplement 3 ) . While Dia localization depends on Rho1 , the localization of Rho1 is independent of Dia and the apical actin cables it forms ( Figure 7—figure supplement 1 ) , suggesting a hierarchy rather than a reinforcing feedback loop between the two proteins . One possibility is that apical Rho1 localization involves anchoring to Rho-GEFs , the specific guanine-exchange factors that activate Rho1 , and were also shown to be enriched in apical membranes of Drosophila tubular organs ( Massarwa et al . , 2009 ) . Our results demonstrate that the tight apical localization of Dia is achieved by combining functionally distinct apical-biasing cues . Neither Rho1 nor PI ( 4 , 5 ) P2 is sufficient to efficiently recruit Dia apically on their own . Nevertheless , combining these two relatively weak interactions gives rise to dramatic apical localization of Dia , demonstrating a synergistic contribution ( Figure 7 ) . The cooperative contribution of multiple domains within the same protein to drive membrane targeting is termed ‘Coincidence detection’ ( Lemmon , 2008 ) . This mechanism for localization of a multidomain protein can be used to generate enhanced targeting specificity . While both Rho1 and PI ( 4 , 5 ) P2 may separately reside in other subcellular domains , Dia will bind with high avidity exclusively to the apical membrane that encompasses both cues . The modular nature of these targeting mechanisms allows to examine the contribution of each cue , and the significance of their cooperation in Dia apical restriction . Binding to Rho1 alone was weak but nevertheless detectable . Conversely , binding to PI ( 4 , 5 ) P2 alone was not capable of generating any apical localization bias , despite its marked contribution to apical recruitment in the presence of Rho1 ( Figure 7 ) . This asymmetry between the two cues suggests a model for the dynamics of Dia-apical targeting ( Figure 7H ) . Binding to activated Rho1 may represent the more prevalent scenario for the initial interaction of Dia with the apical membrane . This interaction will also give rise to an open conformation of Dia , which is more amenable to targeting , and will provide physical proximity to the membrane . Membrane proximity can now allow the interaction of Dia with PI ( 4 , 5 ) P2 , which was not favored when the Dia protein was cytoplasmic , due to low affinity restrictions . The Dia-PI ( 4 , 5 ) P2 interaction stabilizes association with the apical membrane , resulting in synergy in the combined action of Rho1 and PI ( 4 , 5 ) P2 in Dia apical localization . Once in the vicinity of the apical membrane , Dia can shift dynamically among three states , being bound to each of the cues alone , or to both simultaneously . This feature can increase the effective levels of apical-targeting cues , since they do not necessarily have to be used simultaneously at all times . Mechanisms that similarly utilize the synergistic activity of two proteins may be used in the regulation of other actin regulators , which integrate between multiple signals . The activity of N-WASP , the activator of the ARP2/3 actin nucleation complex , was shown to be dependent on the synergistic binding of the Rho GTPase Cdc42 and of PI ( 4 , 5 ) P2 to distinct domains in the protein . However , this binding is not reported to lead to targeting of the protein to distinct subcellular domains , but rather to localized activation ( Prehoda et al . , 2000; Padrick and Rosen , 2010 ) . While the apical targeting of Dia protein is very prominent , additional regulation operating on other tiers of the pathway may ensure enhanced fidelity of the localization . Although these mechanisms seem redundant , they may provide higher accuracy to the system . As previously noted , dia mRNA is also apically localized by a mechanism that is distinct from the protein localization machinery , leading to localized translation of Dia protein at the apical membrane ( Massarwa et al . , 2009 ) . Indeed , we observed a more restricted apical distribution of endogenous Dia compared with the over-expressed protein , which is not localized at the RNA level ( Figure 1 ) . In addition , a bias in the apical localization of Dia activators ( Rho-GEFs and Rho1 ) may provide another level of refinement . Finally , the bias in Rho-GEFs may be influenced by apical PI ( 4 , 5 ) P2 levels , providing an additional layer of convergence between distinct signals ( Murray et al . , 2012; Viaud et al . , 2012 ) . It is likely that this general mechanism in Drosophila tubular organs extends to tubular tissues in the mammalian system . A collaboration between PI ( 4 , 5 ) P2 and a Rho-family GTPase in membrane recruitment of mammalian mDia2 was observed in a cell culture system ( Gorelik et al . , 2011 ) , and here we have demonstrated the capacity of PI ( 4 , 5 ) P2 to influence Dia targeting in mammalian MDCK cysts ( Figure 3 ) . Furthermore , the link between activation and localization , as well as the identity of the critical domain for localization of formin family proteins , appear to be conserved ( Carramusa et al . , 2007; Watanabe et al . , 2010 ) . In this context , we have recently examined the tubular epithelium of secretory acinar cells in the mouse pancreas , and shown that activated mDia1 is targeted to the apical membrane ( Geron et al . , 2013 ) . Thus , the Dia apical-targeting machinery appears to be common to a variety of tubular tissues , regardless of their origin or physiological function . In conclusion , this work uncovers a mechanism by which Dia is targeted to the apical membrane of tubular epithelia , thereby restricting actin cable formation , and consequently secretion , to a distinct membrane domain of these cells . This localization mechanism is likely to be universal to tubular epithelia in a broad range of organisms . For the generation of GFP-Dia-N′ , GFP-Dia-N′ΔBD and GFP-Dia-BD , eGFP was cloned into pUAST–attB at the BglII–EcoRI sites . The relevant segments from a Dia cDNA ( Figure 2A ) were then cloned using the NotI-XbaI sites . GFP-Dia-N′-mutG and GFP-Dia-N′ΔBD-mutG were created by site directed mutagenesis ( Phusion , NEB ) , replacing nucleotides TG ( 440–441 ) with AC and AGC ( 451–453 ) with GAG , using GFP-Dia-N′ and GFP-Dia-N′ΔBD , respectively , as templates . The resulting constructs were subsequently cloned into GFP-pUAST-attB at the NotI-XbaI sites . All constructs were sequenced , and injected into attP18 lines ( Markstein et al . , 2008 ) . For expression in MDCK cysts , Dia-FL and Dia-ΔDAD were cloned into pEGFP-C1 by inserting the appropriate segments from a Dia cDNA using the SalI and HindIII sites . The following lines were used: UAS-GFP-Dia-FL , UAS-Dia-ΔDAD , UAS-Dia-C′ , UAS-Dia-DD-C′ ( referred to as Dia , FH1-FH2 and DDFH1FH2 , respectively in Homem and Peifer , 2009 ) . UAS-GFP-Dia-N′ , UAS-GFP-Dia-N′ΔBD , UAS-GFP-Dia-BD , UAS-GFP-Dia-N′-mutG and UAS-GFP-Dia-N′ΔBD-mutG were generated by standard phi31 germline transformation procedures . Additional strains included UAS-PH-PLCδ-GFP ( von Stein et al . , 2005 ) , UAS-Sktl-RFP , UAS-Sktl-KID-RFP ( Raghu et al . , 2009 ) , UAS-GFP ( BM 5431 ) , UAS-Pten-RNAi ( VDRC 101475 ) , UAS-Rho1-RNAi ( VDRC 12734 ) , UAS-dia-RNAi ( VDRC 20518 ) , UAS-GFP-Rho1 ( BM 9392 ) , UAS-CD8-RFP ( BM 32218 ) , hsFLP ( BM 6 ) , P{Gal4-ActRc ( FRT . CD2 ) . P}S , P{UAS-RFP . W}3/TM3 ( BM30558 ) . Gal4 drivers were btl-Gal4 ( breathless , expresses Gal4 in tracheal cells ) , drm-Gal4 ( drumstick , expresses Gal4 in embryonic proventriculus , anterior midgut , posterior midgut , Malpighian tubules , hindgut and salivary glands ) and fkh-Gal4 ( fork head , expresses Gal4 in the salivary gland ) . Standard embryo fixation and staining procedures were followed ( Patel , 1994 ) . Detection of Dia was carried out according to ( Massarwa et al . , 2009 ) . Primary antibodies used were anti Dia ( rabbit 1:250 [Grosshans et al . , 2005] ) , Crumbs , Dlg ( mouse 1:100; DSHB , University of Iowa , USA ) , GFP ( chicken 1:500; Abcam ) , GFP ( rabbit 1:700; Life Technologies , for MDCK cysts ) , Trh ( rat 1:100 ) , FasIII ( mouse 1:20 ) , gp135 ( Podocalyxin , mouse 1:2000; from G Ojakian ) , DsRed ( Rabbit 1:500; Clontech ) , Actin ( mouse 1:1000 , Sigma ) . Anti-mouse , rabbit and chicken Alexa-conjugated secondary antibodies were obtained from Invitrogen and diluted 1:800 . Anti-rat Cy5-conjucated secondary antibody was purchased from Jackson ImmunoResearch and diluted 1:200 . Nuclei of MDCK cells were counterstained with Hoechst ( Molecular Probes ) . Live embryos and dissected third instar larval salivary glands were imaged following mounting in Halocarbon oil 700 ( Sigma ) . Images of immunofluorescent and live samples were acquired using a Zeiss LSM 710 confocal scanning system , using ×20 N . A 0 . 5 or ×63 N . A 1 . 4 objectives , and processed using Adobe Photoshop CS3 . For quantification of apical localization , mid-plane non-saturated images were selected . Mean intensities of GFP fluorescence were measured by ImageJ . The different subcellular domains were demarcated by hand according to specific stainings . 3D culture , immunofluorescence staining and microscopy of MDCK cysts , as well as lipid delivery to the cysts were performed as previously described ( Martin-Belmonte et al . , 2007 ) .	Many physiological processes are directional , which means that tissues and organs often need a sense of spatial orientation in order to function properly . In most tissues , this sense of direction relies on certain proteins and infrastructure components of the cell being located in specific subcellular regions , rather than being distributed in a more symmetrical fashion throughout the cell: the latter phenomenon is known as cell polarity . Exocrine tissues ( that is , glands ) are composed of tubular epithelial cells organized around a central lumen: the cells in the gland secrete various products ( such as enzymes ) into the lumen , so that they can be carried to the target organ elsewhere in the body . Epithelial cells in these tissues are therefore polarized to enable directional transport to the lumen . An example of cell polarity is a network of actin filaments that lines the apical surface of these cells ( the surface nearest the common lumen ) . This actin network helps to shuttle cargo to the lumen by assisting with directional , coordinated secretion , among other processes . In fruitflies , the construction of the apical actin network depends on the presence of a protein called Diaphanous . However , the signals that lead to the localization of this protein near the apical membrane of the cells are not well understood . Now Rousso et al . report that a modified lipid , called PI ( 4 , 5 ) P2 , is involved in this localization . However , they also show that this lipid does not govern the apical localization of Diaphanous on its own: rather , an enzyme called Rho1 must also be present to assist with the localization of Diaphanous and to ensure that actin is deposited in the correct place . Rousso et al . also demonstrate that PI ( 4 , 5 ) P2-mediated localization of Drosophila Diaphanous occurs in mammalian cells . Lipid-protein collaboration also targets other proteins to the apical membrane . A common mechanism may therefore underlie cell polarity in tubular organ tissues in flies and mammals .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "developmental", "biology", "cell", "biology" ]	2013	Apical targeting of the formin Diaphanous in Drosophila tubular epithelia
Group 1 CD1 molecules , CD1a , CD1b and CD1c , present lipid antigens from Mycobacterium tuberculosis ( Mtb ) to T cells . Mtb lipid-specific group 1 CD1-restricted T cells have been detected in Mtb-infected individuals . However , their role in protective immunity against Mtb remains unclear due to the absence of group 1 CD1 expression in mice . To overcome the challenge , we generated mice that expressed human group 1 CD1 molecules ( hCD1Tg ) and a CD1b-restricted , mycolic-acid specific TCR ( DN1Tg ) . Using DN1Tg/hCD1Tg mice , we found that activation of DN1 T cells was initiated in the mediastinal lymph nodes and showed faster kinetics compared to Mtb Ag85B-specific CD4+ T cells after aerosol infection with Mtb . Additionally , activated DN1 T cells exhibited polyfunctional characteristics , accumulated in lung granulomas , and protected against Mtb infection . Therefore , our findings highlight the vaccination potential of targeting group 1 CD1-restricted lipid-specific T cells against Mtb infection . The CD1 family of antigen presenting molecules presents self and microbial lipids to T cells ( Van Rhijn et al . , 2013; De Libero and Mori , 2014; Adams , 2014 ) . Two major groups of CD1 isoforms have been identified in humans: group 1 CD1 ( CD1a , CD1b , and CD1c ) and group 2 CD1 ( CD1d ) ( Adams , 2014 ) . While CD1d is broadly expressed , group 1 CD1 expression is limited to cortical thymocytes and professional antigen presenting cells ( Barral and Brenner , 2007 ) . Humans express all CD1 isoforms while mice only express CD1d . Due to the lack of a suitable small animal model to study group 1 CD1-restricted T cells , CD1d-restricted NKT cells have been better studied . A large proportion of CD1d-restricted NKT cells express an invariant TCR α chain and are known as iNKT cells ( Bendelac et al . , 2007 ) . Unlike conventional T cells , which are positively selected by thymic epithelial cells ( TECs ) , hematopoietic cells ( HCs ) select iNKT cells ( Bendelac et al . , 2007 ) . The unique developmental selection program is thought to drive their pre-activated phenotype , which allows for rapid effector function manifestations upon TCR stimulation ( Bendelac et al . , 2007 ) . Unlike iNKT cells , group 1 CD1-restricted T cells are known to have diverse TCR usage ( Grant et al . , 1999; Vincent et al . , 2005; Felio et al . , 2009 ) . Group 1 CD1-restricted T cell responses have mostly been characterized in the context of mycobacterial antigens , although recent studies have shown that humans have a significant proportion of autoreactive group 1 CD1-restricted T cells ( de Jong et al . , 2010; de Lalla et al . , 2011 ) . The Mtb cell wall is lipid rich ( 60% of its cell wall is composed of lipids ) and contains a plethora of lipid antigens that are presented by group 1 CD1 molecules ( Van Rhijn et al . , 2013; De Libero and Mori , 2014 ) . Among group 1 CD1 molecules , CD1b presents the largest pool of Mtb-derived lipids like mycolic acid ( MA ) , glucose monomycolate , glycerol monomycolate , diacylated sulfoglycolipids , lipoarabinomannan and phosphatidylinositol mannoside to cognate T cells ( Van Rhijn et al . , 2013; De Libero and Mori , 2014 ) . Of the Mtb lipids mentioned above , MAs are the major lipid constituents of the Mtb cell envelope and considered a potent Mtb virulence factor ( Barry et al . , 1998; Karakousis et al . , 2004 ) . Interestingly , MA-specific CD1b-restricted T cells have been detected in the blood as well as disease sites of Mtb-infected individuals ( Montamat-Sicotte et al . , 2011 ) . Tuberculosis ( TB ) , the disease caused by Mtb , is a global health burden , especially in developing countries and amongst HIV/AIDS patients . Additionally , due to the emergence of multidrug-resistant Mtb and the lack of an effective vaccine to prevent pulmonary TB in adults , it is important to decipher the role of various T cell subsets in Mtb infection for the development of better preventive or therapeutic vaccines ( Ottenhoff et al . , 2012 ) . The subunit vaccines currently under development for Mtb utilize peptide or protein antigens which target MHC-restricted conventional T cells ( Dorhoi and Kaufmann , 2014 ) , but the utility in targeting lipid antigens has not been explored . Since CD1 molecules are non-polymorphic , CD1-restricted Mtb lipid antigens are likely to be recognized by most individuals , making them attractive vaccine targets ( Barral and Brenner , 2007 ) . Several lines of evidence suggest that Mtb lipid-specific group 1 CD1-restricted T cells contribute to anti-mycobacterial immunity . Investigation of group 1 CD1-restricted T cell lines derived from healthy or mycobacteria-infected individuals has revealed that these T cells are cytotoxic and produce IFN-γ and TNF-α , cytokines critical for protective immunity to TB ( Van Rhijn et al . , 2013 ) . Moreover , group 1 CD1-restricted Mtb lipid-specific T cells are found in higher frequencies in individuals exposed to Mtb compared with control populations , suggesting that they are activated following infection with Mtb ( Moody , 2000; Ulrichs et al . , 2003; Gilleron , 2004; Layre et al . , 2009; Moody et al . , 2000 ) . Additionally , a robust Mtb lipid-specific group 1 CD1-restricted T cell response has been detected in Mtb-infected human group 1 CD1 transgenic mice ( Felio et al . , 2009 ) . However , it remains unclear whether this unique T cell subset plays a protective role during the course of infection . In this study , we generated transgenic mice expressing mycolic acid-specific CD1b-restricted TCR ( DN1Tg ) and human group 1 CD1 molecules ( hCD1Tg ) . Using this mouse model , we found that DN1 T cells were selected most efficiently by group 1 CD1-expressing HCs in the thymus . Upon adoptive transfer of DN1 T cells to Mtb-infected hCD1Tg mice , DN1 T cells were first activated in the mediastinal lymph nodes , exhibiting faster kinetics than Ag85B-specific CD4+ T cells . DN1 T cells were cytotoxic , polyfunctional and contributed to anti-mycobacterial immunity by reducing bacterial burdens in the lung , spleen and liver . Thus , this study provides the first direct demonstration that group 1 CD1-restricted Mtb lipid-specific T cells play a protective role during Mtb infection . We developed human CD1 transgenic mice , which expressed group 1 CD1 molecules in a similar pattern to that observed in humans . Using this model , we demonstrated the feasibility to study group 1 CD1-restricted T cell responses in aerosol infection with Mtb ( Felio et al . , 2009 ) . To facilitate the direct analysis of Mtb lipid-specific group 1 CD1-restricted T cells , we generated a novel transgenic mouse strain that expressed a human/mouse chimeric TCR , composed of variable region from human T cell clone DN1 ( Grant et al . , 1999 ) , specific for CD1b/mycolic acid ( MA ) , and mouse TCR constant region ( Figure 1A ) . DN1Tg founders and their progeny were screened for the presence of TRAV13-2-TRAJ57 gene fragment by PCR and for the surface expression of human Vβ5 . 1 ( TRBV5-1 ) by flow cytometry ( Figure 1B , C ) . Subsequently , DN1Tg mice were bred onto hCD1Tg/Rag-/- background to eliminate the expression of endogenous TCR . All DN1Tg mice used in this study were on a Rag-/- background . To examine whether the development of DN1 T cells was dependent on group 1 CD1 molecules , we compared DN1 T cells in WT and hCD1Tg backgrounds . We found that both frequency and absolute number of DN1 T cells were greatly reduced in DN1Tg mice compared with DN1Tg/hCD1Tg mice in all tested organs ( Figure 1D–F ) . This suggested that group 1 CD1 supported the development of DN1 T cells . Notably , unlike CD1d-restricted iNKT cells , DN1 T cells from the spleen and lymph nodes of DN1Tg/hCD1Tg mice exhibited a naïve phenotype ( characterized by low expression levels of T cell activation markers such as CD69 and CD44 ) similar to conventional CD8+ T cells and were either CD8αβ+ or CD4-CD8- ( DN ) . In addition , DN1 thymocytes from DN1Tg/hCD1Tg mice did not express PLZF , the master transcription factor for innate T cell lineages ( Figure 1G ) ( Kovalovsky et al . , 2008; Savage et al . , 2008 ) . 10 . 7554/eLife . 08525 . 003Figure 1 . Development of DN1 T cells is dependent on the presence of group 1 CD1 molecules . ( A ) Schematic diagram of DN1 TCR construct used to generate DN1Tg mice . ( B ) The presence of DN1 TCR in the genomic DNA of transgenic mice was examined by PCR using primers specific for TRAV13-2 and TRAJ57 . DN1 plasmid was used as a positive control ( Ctrl ) . ( C ) DN1 T cells in the spleen of DN1Tg+ and DN1Tg- mice ( in a B6 background ) were detected by FACS using anti-mouse TCRβ and anti-human Vβ5 . 1 mAbs . ( D ) Lymphocytes from the thymus , spleen and liver of DN1Tg/hCD1Tg and DN1Tg mice ( in the Rag-deficient background ) were analyzed for the presence of DN1 T cells ( TCRβ+hVβ5 . 1+ ) . ( E , F ) Bar graphs depict the mean and SEM of the percentages ( in the lymphocyte gate ) and absolute numbers of DN1 T cells from DN1Tg/hCD1Tg and DN1Tg mice ( n=3–8 per group ) . *p<0 . 001; p<0 . 01; p<0 . 05 . ( G ) Expression of indicated markers ( black line ) on DN1 T cells ( TCRβ+hVβ5 . 1+ ) from DN1Tg/hCD1Tg/Rag-/- mice , type I NKT cells ( CD1d/αGalCer tetramer+TCRβ+ ) from WT mice , and conventional CD8+ T cells ( TCRβ+CD8+ ) from WT mice , compared with isotype control ( gray filled ) . The expression of CD4 and CD8 on DN1 T cells and type I NKT cells were shown in the dot plots . Cells isolated from the thymus were used for PLZF staining . Results are representative of 3 experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 08525 . 003 Unlike conventional T cells , which are positively selected by TECs , iNKT cells are exclusively selected by CD1d-expressing thymocytes ( Bendelac , 1995; Coles and Raulet , 2000 ) . Several studies have demonstrated the correlation between positive selection on HCs and a pre-activated T cell phenotype of innate-like T cells ( Bendelac et al . , 2007; Cho et al . , 2011; Bediako et al . , 2012 ) . Given that DN1 T cells exhibited a naïve surface phenotype , one would expect DN1 T cells to be positively selected by TECs . To test this hypothesis , we adoptively transferred bone marrow from DN1Tg and DN1Tg/hCD1Tg mice ( in the Rag-deficient background ) into irradiated CD45 . 1 congenic WT and hCD1Tg recipients . 5 weeks after transfer , DN1 T cells were identified by CD45 . 2and hVβ5 . 1 surface expression in different groups ( Figure 2A ) . The percentage ( Figure 2B ) and absolute number ( Figure 2C ) of DN1 T cells were significantly higher in mice with group 1 CD1-expressing HCs compared to mice that only had group 1 CD1-expressing TECs . This suggested that HCs most efficiently mediate the positive selection of DN1 T cells . As a small number of DN1 T cells developed in mice that lack CD1b ( Figure 2A ) , it is possible that mouse CD1d is responsible for their selection . We compared the percentage of DN1 T cells in the spleen and thymus of DN1Tg/hCD1Tg ( CD1d+ ) , DN1Tg/hCD1Tg/CD1d-/- , DN1Tg ( CD1d+ ) , and DN1Tg/CD1d-/- mice ( all in the Rag-deficient background ) . We found that the percentage of DN1 T cells was comparable in DN1Tg/hCD1Tg and DN1Tg/hCD1Tg/CD1d-/- mice . In addition , DN1 T cells were barely detectable in the thymus and spleen of DN1Tg and DN1Tg/CD1d-/- mice . These data suggest that CD1d does not contribute to the thymic selection of DN1 T cells ( Figure 2—figure supplement 1 ) . 10 . 7554/eLife . 08525 . 004Figure 2 . CD1b-expressing hematopoietic cells are the major cell type that medicates the positive selection of DN1 T cells . Bone marrow from DN1Tg and DN1Tg/hCD1Tg mice ( in the Rag-deficient background ) were adoptively transferred into irradiated CD45 . 1 congenic WT and hCD1Tg recipients and the development of DN1 T cells were examined 5 weeks later . ( A ) Dot plots depict the proportion of CD45 . 2+ hVβ5 . 1+ cells in the lymphocyte gate . Data are representative of 2 experiments with 4–6 mice in each group . ( B , C ) Bar graphs depict the mean ± SEM of the percentage and absolute number of CD45 . 2+ hVβ5 . 1+ cells in each group . Statistical significance was evaluated by comparing HC , TEC and None group with HC+TEC group . p<0 . 001; p<0 . 01; p<0 . 05 . ( D ) CD1b expression on TEC ( CD45-MHCII+ ) and DP thymocytes from WT and hCD1Tg mice were shown as MFI values ( n=3 per group ) . ( E ) Percentage of CD8+ DN1 T cells in the spleen of HC+TEC , HC or TEC groups of mice . ( F ) Expression of indicated markers on DN1 thymocytes that developed in HC+TEC , HC or TEC groups . p<0 . 001; p<0 . 01; p<0 . 05 . Results are representative of 2 experiments with 3 mice per group . DOI: http://dx . doi . org/10 . 7554/eLife . 08525 . 00410 . 7554/eLife . 08525 . 005Figure 2—figure supplement 1 . CD1 expression does not significantly affect the development of DN1 T cells . Lymphocytes from the thymus and spleen of DN1Tg/hCD1Tg , DN1Tg/hCD1Tg/CD1d-/- , DN1Tg and DN1Tg/CD1d-/- mice ( all in the Rag-deficient background ) were analyzed for the presence of DN1 T cells ( TCRβ+hVβ5 . 1+ ) . Results are representative of 2 experiments with 3 mice per group . DOI: http://dx . doi . org/10 . 7554/eLife . 08525 . 005 Comparing CD1b expression on TECs and CD4+CD8+ ( DP ) thymocytes revealed that DP thymocytes express significantly higher levels of CD1b than TECs ( Figure 2D ) . Thus , CD1b-expressing thymocytes may be better suited to mediate the positive selection of DN1 T cells . Since DN1 T cells could also be selected by TECs , albeit with much lower efficiency compared to HCs , we compared the phenotype of DN1 T cells that developed in mice expressing CD1b on both HC and TEC , HC only and TEC only . DN1 T cells in the periphery of these three groups had a comparable proportion of CD8+/DN T cells ( Figure 2E ) . In addition , DN1 T cells in the thymus of these three groups expressed similar levels of PLZF and CD44 ( Figure 2F ) . However , DN1 T cells selected by HCs expressed higher levels of CD5 ( Figure 2F ) , a surrogate marker for the TCR signaling strength in developing thymocytes , suggesting they might receive stronger TCR signals . The human DN1 T cell line had been shown to secrete Th1 cytokines when stimulated with MA presented by CD1b+ APCs ( Beckman et al . , 1994 ) . To test whether DN1 T cells that developed in DN1Tg/hCD1Tg mice retained the same functional properties as the original human T cell line , we stimulated lymph node cells from DN1Tg/hCD1Tg mice with un-pulsed or MA-pulsed bone marrow derived dendritic cells ( BMDCs ) to detect IFN-γ production and antigen-specific cytotoxicity ( Figure 3A , B ) . DN1 T cells produced IFN-γ in response to MA-pulsed hCD1Tg DCs but not WT DCs or un-pulsed DC , suggesting the activation of DN1 T cells required both antigen and group 1 CD1 molecules . In addition , DN1 T cells showed cytotoxic activity against MA-pulsed hCD1Tg DCs . MA is located within the Mtb cell wall , either covalently attached via arabinogalactan to the cell wall peptidoglycan , or non-covalently associated in the form of trehalose dimycolate ( Barry et al . , 1998; Karakousis et al . , 2004 ) . To determine whether DN1 T cells can be activated by naturally processed MA , we set up co-culture of DN1 T cells with Mtb-infected BMDCs . As shown in Figure 3C , co-culture of DN1 T cells with Mtb-infected hCD1Tg DCs led to up-regulation of activation markers CD69 and CD44 on DN1 T cells . Furthermore , DN1 T cells produced multiple cytokines , of which the production of IFN-γ , TNF-α , IL-13 and IL-6 was dependent on the presence of group 1 CD1 molecules ( Figure 3D ) , with the exception of IL-17 . It is possible that co-stimulatory molecules and/or cytokines induced by Mtb-infected DCs can stimulate DN1 T cells to secrete this cytokine independent of TCR-CD1b interaction . Collectively , our data demonstrated that DN1 T cells became activated and exhibited effector functions in response to MA-pulsed or Mtb-infected DC in a group 1 CD1-dependent manner . 10 . 7554/eLife . 08525 . 006Figure 3 . DN1 T cells acquire effector functions in response to MA-pulsed DC and Mtb-infected DC . ( A ) DN1 T cells isolated from lymph nodes of DN1Tg/hCD1Tg/Rag-/- mice were co-cultured with un-pulsed or MA-pulsed WT or hCD1Tg DC and IFN-γ producing cells were determined by ELISPOT assays . ( B ) DN1 T cells were stimulated by hCD1Tg BMDCs pulsed with MA for 7 days and then tested for cytotoxic activity against hCD1Tg BMDCs with or without MA . Data are representative of 3 experiments ( mean ± SEM of triplicate cultures ) . ( C , D ) DN1 T cells isolated from lymph nodes of DN1Tg/hCD1Tg/Rag-/- mice were co-cultured with Mtb-infected BMDC for 48 hr . Activation markers on DN1 T cells were detected by flow cytometry and cytokines in the supernatant were detected by CBA flex set . p<0 . 001; p<0 . 01; p<0 . 05 . Results are representative of 2 experiments with 3 mice per experiment . DOI: http://dx . doi . org/10 . 7554/eLife . 08525 . 006 Macrophages are known as primary host cells for Mtb . Whereas BMDCs and a subset of myeloid DCs from hCD1Tg mice expressed high levels of CD1b , the expression of CD1b was almost undetectable on bone marrow derived macrophages ( BMDMs ) ( Figure 4A ) , similar to the observation in human monocyte derived macrophages . Accordingly , we detected only minimal DN1 T cell activation when stimulated with Mtb-infected BMDMs ( Figure 4B ) . While our data showed that Mtb-infected macrophages do not directly present antigen to DN1 T cells , apoptotic vesicles released from Mtb-infected macrophages have been shown to transfer mycobacterial antigens to uninfected APCs , such as DCs ( Ulrichs et al . , 2003; Schaible et al . , 2003 ) , which could in turn activate DN1 T cells . To explore whether and how DN1 T cells can control Mtb in infected macrophages , Mtb-infected BMDMs from WT or hCD1Tg mice were cultured together with DN1 T cells in the presence or absence of uninfected WT or hCD1Tg DCs . After 7 days , we determined the number of colony forming units ( CFU ) to investigate whether DN1 T cells inhibited intracellular bacterial growth within BMDMs . As expected , addition of DN1 T cells alone did not have a significant effect on bacterial burdens in Mtb-infected BMDMs . However , when DN1 T cells were added together with uninfected hCD1Tg DCs to infected BMDMs , the number of CFU decreased significantly compared with controls ( Figure 4C ) . For comparison , we also used Mtb-infected BMDCs from WT or hCD1Tg mice as targets . We found that DN1 T cells efficiently controlled Mtb growth within infected hCD1Tg DCs but not WT DCs . Similarly , if DN1 T cells and uninfected hCD1Tg DCs were added to Mtb-infected WT DCs , Mtb growth was significantly inhibited ( Figure 4D ) . These data indicated that group 1 CD1-expressing DCs mediated activation of DN1 T cells , which in turn controlled bacterial growth not only in the group 1 CD1-expressing DCs but also in macrophage and group 1 CD1-negative DCs . In addition , through a cytokine-blocking assay , we found that IFN-γ but not TNF-α was crucial for mediating anti-mycobacterial function of DN1 T cells ( Figure 4E ) . 10 . 7554/eLife . 08525 . 007Figure 4 . DN1 T cell-mediated control of Mtb is dependent on the antigen presentation by group 1 CD1-expressing DCs and IFN-γ production . ( A ) CD1b expression on BMDM and BMDC was detected using flow cytometry . ( B ) BMDMs were in vitro infected with Mtb ( MOI=5 ) and DN1 T cells were added 1 day after infection . After 48 hr co-culture , activation markers on DN1 T cells were detected by flow cytometry . ( C , D ) WT and hCD1Tg BMDMs and BMDCs were infected with Mtb . 1 day later , DN1 T cells with or without uninfected WT or hCD1Tg DCs were added into the culture for another 6 days . At day 7 post-infection , cells were lysed for CFU assay . Bar represents mean and SEM from replicate cultures ( n = 6 ) . ( E ) DN1 T cells were added into Mtb-infected BMDCs in the presence of control Ig ( Ig ) , anti-IFN-γ or anti-TNFα . At day 7 post-infection , cells were lysed for CFU assay . % reduction was calculated as 100x[ ( BMDCalone- BMDCwith DN1 ) / BMDCalone] . Results are representative of 2–3 experiments . p<0 . 001; p<0 . 01; p<0 . 05 . DOI: http://dx . doi . org/10 . 7554/eLife . 08525 . 007 To study when and where group 1 CD1-restricted Mtb lipid-specific T cells are first presented with antigens after aerosol infection with Mtb , we adoptively transferred naïve DN1 T cells into CD45 . 1 congenic hCD1Tg mice that had been infected with Mtb 7 days earlier . The expression of CD69 on DN1 T cells from various organs was monitored . The up-regulation of CD69 on DN1 T cells was first observed as early as day 11 post-infection in the lung-draining MLN , but not in other tissues examined . By day 15 after infection , the activation of DN1 T cells was also detected in the lung and spleen ( Figure 5A , B ) . Although bacterial burdens were much higher in the lung than in the MLN ( Figure 5C ) , DN1 T cell activation correlated with the first appearance of bacteria in MLN . Collectively , these data indicate that activation of DN1 T cells is initiated in MLN when Mtb disseminate from the site of primary infection ( lung ) to MLN . 10 . 7554/eLife . 08525 . 008Figure 5 . Activation of DN1 T cells is initiated in mediastinal lymph nodes after aerosol Mtb infection . ( A ) 3x106 naïve DN1 T cells were adoptively transferred into Mtb infected CD45 . 1 congenic hCD1Tg mice at day 7 post-infection . CD69 expression was detected on DN1 T cells ( hVβ5+TCRβ+ ) from the MLN , lung , spleen , and inguinal lymph nodes ( ILN ) of recipient mice at indicated time points . ( B ) Bar graphs depict the mean and SEM of the percentages of CD69hi population among DN1 T cells ( n=3 each time point ) . ( C ) Bacterial CFU in MLN and lung at indicated time points . Each symbol represents the bacteria burden in the MLN or lung of an individual mouse at the indicated time point . Horizontal bars represent the mean CFU counts ± SEM for each group . Results are representative of 2 experiments with 3 mice per time point . p<0 . 001; p<0 . 01; p<0 . 05 . DOI: http://dx . doi . org/10 . 7554/eLife . 08525 . 008 Several studies have shown that activation of Mtb-specific CD4+ conventional T cells is also initiated in the MLN ( Wolf et al . , 2008; Reiley et al . , 2008; Gallegos et al . , 2008 ) . To compare the kinetics of DN1 T cell priming with conventional T cells after aerosol Mtb infection , we co-transferred naïve DN1 T cells with CD4+ TCR transgenic T cells specific to a peptide from Mtb Ag85B ( P25 T cells ) to Mtb-infected mice to monitor their activation and proliferation . CellTrace Violet dye labeled P25 T cells and CFSE labeled DN1 T cells were co-transferred in equal numbers to CD45 . 1 congenic hCD1Tg mice that were infected 7 days earlier . Similar to the observation in Figure 5A , up-regulation of CD69 on DN1 T cells started at day 11 post infection while CD69 was up-regulated on a small percentage of P25 T cells in MLN 13 days after infection ( Figure 6A , C ) . Additionally , cell division was detected on DN1 T cells by day 13 and on P25 T cells by day 15 in the MLN ( Figure 6B , D ) . Also , compared to P25 T cells , a greater proportion of DN1 T cells in the lung and spleen expressed CD69 and underwent cell division at day 15 post-infection ( Figure 6E , F ) . Taken together , our data suggests that activation of MA-specific CD1b-restricted T cells occurs earlier than Ag85B-specific MHC II-restricted CD4+ T cells during Mtb infection . 10 . 7554/eLife . 08525 . 009Figure 6 . DN1 T cells are activated earlier than Ag85 specific CD4+ T cells after Mtb infection . ( A , B ) 3x106 CFSE-labeled DN1 T cells and 3x106 CellTrace Violet-labeled P25 T cells were co-transferred into Mtb infected CD45 . 1 congenic hCD1Tg mice . CD69 expression , CFSE and CellTrace Violet were detected on DN1 T cells and P25 T cells from MLN at indicated time points . ( C , D ) Bar graphs depict the mean and SEM of the percentages of CD69hi and CFSE/Violetlow populations among DN1 and P25 T cells . ( E , F ) CD69 expression , CFSE and CellTrace Violet were detected on DN1 and P25 T cells from lungs ( E ) and spleens ( F ) at day 13 and day 15 post-infection . Results are representative of 2 experiments with 3–4 mice per experiments . p<0 . 001; p<0 . 01; p<0 . 05 . DOI: http://dx . doi . org/10 . 7554/eLife . 08525 . 009 Group 1 CD1-restricted human T cell clones show effector functions including cytotoxic activity and cytokine production in response to Mtb-specific lipid antigens . However , whether group 1 CD1-restricted T cells confer protection against Mtb infection remains unknown . To address this question , DN1 effector T cells were adoptively transferred to hCD1Tg/Rag-/- mice and recipient mice were subsequently challenged with virulent Mtb via aerosol route . 4 weeks after infection , the number of bacteria in the lung , spleen and liver was determined . As shown in Figure 7A , DN1 T cells decreased the number of viable bacteria in hCD1Tg/Rag-/- recipient mice in all tested organs as compared to mice that received no DN1 T cell transfer . Moreover , adoptive transfer of DN1 T cells to Rag-/- recipient mice did not significantly reduce bacterial burdens suggesting that DN1 T cells confer protection in an hCD1Tg-dependent manner ( Figure 7A ) . We also compared the protective capacity of DN1 T cells with non-relevant Listeria LemA-specific H2-M3-restricted D7 T cells ( Figure 7—figure supplement 1 ) . D7 T cells did not significantly reduce bacterial burdens in the lung of hCD1Tg/Rag-/- recipient mice compared to mice that received no T cell transfer ( Figure 7—figure supplement 1 ) . 10 . 7554/eLife . 08525 . 010Figure 7 . DN1 T cells contribute to protective immunity against Mtb infection . Effector DN1 T cells ( 5-7x106 cells ) were transferred into hCD1Tg/Rag-/- or Rag-/- mice 1 day before infection . Mice were euthanized at 4 week post-infection . ( A ) The number of bacteria in the lung , spleen and liver of individual mouse in each group . Horizontal bars represent the mean CFU counts ± SEM for each group . ( B , C ) Cells harvested from lungs of hCD1Tg/Rag-/- or Rag-/- mice were stimulated with un-pulsed or MA-pulsed DC and intracellular stained for the indicated cytokines ( B ) and CD107a expression ( C ) . ( D ) Immunohistochemistry staining of anti-CD3 ( brown cells ) of the lung section from indicated groups . Pictures show granuloma area in infected lung tissues . ( E ) Bar graphs depict the mean and SEM of number of CD3+ cells per mm2 within granuloma areas ( n=3–6 mice each group ) . p<0 . 001; p<0 . 01; p<0 . 05 . DOI: http://dx . doi . org/10 . 7554/eLife . 08525 . 01010 . 7554/eLife . 08525 . 011Figure 7—figure supplement 1 . Adoptive transfer of listeria LemA-specific M3-restricted D7 T cells does not confer protection against Mtb infection . Effector D7 T cells ( 5x106 cells ) were transferred into hCD1Tg/Rag-/- mice 1 day before infection . Bacterial load in the lung was measured at 4 week post-infection . Each symbol represents the bacterial burden in the lung of an individual mouse in each group . Horizontal bars represent the mean CFU counts ± SEM . NS , not significant . DOI: http://dx . doi . org/10 . 7554/eLife . 08525 . 011 DN1 T cells isolated from the lung of infected hCD1Tg/Rag-/- mice produced multiple cytokines ( e . g . TNF-α , IFN-γ , and IL-2 , Figure 7B ) and expressed CD107a , a surrogate marker of cytotoxic activity ( Cho et al . , 2011 ) , after ex vivo MA stimulation ( Figure 7C ) . To further visualize the location and distribution of DN1 T cells , DN1 T cells in the lung were stained by immunohistochemistry using anti-CD3 antibody . A significantly higher number of DN1 T cells were seen within pulmonary granulomas of hCD1Tg/Rag-/- mice that received DN1 T cells compared to control groups ( Figure 7D , E ) . In summary , these data demonstrate that DN1 T cells accumulate in granulomas and contribute to protective immunity against Mtb by producing multiple Th1-related cytokines and exerting cytotoxicity . In this study we used a novel transgenic mouse model to examine the role of MA-specific CD1b-restricted T cells during Mtb infection . This model not only allowed for the in vivo tracking of Mtb lipid antigen-specific T cells but also for deciphering their developmental requirements and functional role in Mtb infection . Like CD1d-restricted NKT cells , DN1 T cells were most efficiently selected by HCs even though they did not exhibit a pre-activated phenotype . In both in vitro and in vivo Mtb infection systems , the presence of DN1 T cells resulted in lower bacterial burden , indicating their role in protective immunity against Mtb . Interestingly , DN1 T cells were first activated in the lung draining lymph node and seemed to have faster activation and proliferation kinetics compared to Ag85B-specific CD4+ T cells . The distinct kinetics of the priming of these two T cell populations may allow them to contribute to anti-mycobacterial activity at the different stages of infection . However , since only one TCR was investigated in this study , it remains unclear whether these findings could be extrapolated to all group 1 CD1-restricted mycolic acid specific T cells . While TECs select conventional T cells , innate-like T cells , characterized by an effector/memory phenotype in the naïve state , have been correlated with their selection by HCs ( Bendelac et al . , 2007; Cho et al . , 2011 ) . Interestingly , MHC Ib H2-M3-restricted T cells could be selected by both HCs and TECs , however , only HC-selected T cells exhibited a pre-activated phenotype ( Cho et al . , 2011 ) . Additionally , CD1d-restricted iNKT cells and CD1b-autoreactive T cells , which are also pre-activated , are selected by HCs ( Bendelac , 1995; Coles and Raulet , 2000; Li et al . , 2011 ) . Surprisingly , DN1 T cells , though phenotypically naïve , were most efficiently selected by HCs . This finding suggests that HC-mediated selection could lead to the development of two distinct phenotypes of group 1 CD1-restricted T cells . However , the mechanism behind this dichotomous selection process is unknown . Factors like TCR-CD1b avidity , the nature of the selecting lipid antigen ( s ) could all play a role in determining the phenotype of DN1 T cells . The fact that CD5 expression was significantly higher on DN1 T cells selected by HCs suggests that a stronger TCR-CD1b interaction ensued between HCs and DN1 T cells ( Azzam et al . , 2001 ) . The strength of the interaction could be a reflection of the high CD1b expression on DP thymocytes compared to TECs . Although a significant proportion of DN1 T cells in DN1Tg/hCD1Tg mice are CD8+ , we did not detect differences in the activation kinetics and effector functions between CD8+ or CD8- DN1 T cells . Also , HC-selected and TEC-selected DN1 T cells in bone marrow chimeric mice have similar proportions of CD8+ T cells , suggesting that CD8 may not play a critical role in the selection and function of group 1 CD1-restricted T cells . From a naïve phenotype , DN1 T cells transformed to an activated state upon encountering Mtb-derived MA . When CD1b-expressing DCs were infected with Mtb , DN1 T cells effectively lowered bacterial burden . This response was dependent on the presence of CD1b on DCs . Additionally , when Mtb-infected BMDMs ( which do not express CD1b and are known to be the natural reservoirs of Mtb ) were co-cultured with uninfected CD1b-expressing DCs , DN1 T cells retained their protective capacity . In this in vitro Mtb infection system , BMDMs were washed extensively after infection to remove extracellular Mtb before addition of the DCs and DN1 T cells . This suggests that CD1b-expressing DCs are most likely capable of cross-presenting Mtb-lipid antigen from infected macrophages to DN1 T cells . This is consistent with a previous study , which showed that upon death of mycobacteria infected-macrophages; apoptotic vesicles containing Mtb antigens are taken up by DCs , which present these antigens to T cells ( Schaible et al . , 2003 ) . When DN1 T cells were adoptively transferred to Mtb-infected mice , they were first activated in the lung draining MLN instead of the lungs . Recent studies have shown that , after aerosol infection of mice with Mtb , myeloid DCs become infected in the lung , and represent the predominant cells that contain Mtb in the MLN ( Wolf et al . , 2007 ) . Since peripheral CD1b expression is limited to a subset of myeloid DCs ( Felio et al . , 2009 ) , initial DN1 T cell activation could be mediated by infected CD1b-expressing DC or by uninfected DCs that cross-present MA from infected DCs migrating from the lungs . However , the events and processes leading to the presentation of MA by DCs in the MLN remain to be clearly elucidated . The protective capacity of DN1 T cells was also demonstrated in vivo upon transferring effector DN1 T cells to Mtb infected group 1 CD1 expressing Rag-/- mice . Bacterial burden reduction of approximately five ( lung ) to ten ( spleen and liver ) fold was observed in various organs . This demonstrated that DN1 T cells were capable of providing systemic immunity against Mtb . Numerous studies have shown that conventional CD4+ and CD8+ T cells are critical for resistance to Mtb infection , though it’s widely accepted that CD4+ T cells play a more dominant role ( Flynn and Chan , 2001; Woodworth and Behar , 2006 ) . Interestingly , studies using adoptive transfer of transgenic CD4+ T cells reactive to Mtb antigens have yielded mixed results . While Ag85B specific T cells did not confer protection against Mtb infection ( Reba et al . , 2014 ) , ESAT-6 reactive CD4+ T cells reduced bacterial burden by a hundred fold , which was dependent on the infectious dose ( Gallegos et al . , 2008 ) . On the other hand , immune conventional CD8+ T cells lowered Mtb CFU by ten fold in the lungs of infected mice ( Woodworth et al . , 2008 ) . Recently , the role of CD1d-restricted iNKT cells has been more appreciated in the context of in vivo Mtb infection . For example , mice that lack CD1d , the only CD1 isoform expressed in mice , did not show any differences in bacterial burden upon Mtb infection compared to wild type mice ( Behar et al . , 1999 ) . However , upon adoptive transfer of iNKT cells , bacterial burden was reduced five fold in the lungs , suggesting iNKT cells may play a protective role during infection ( Sada-Ovalle et al . , 2008 ) . Based on these studies , the protective capacity of DN1 T cells appears to be comparable to that of conventional CD8+ T cells and CD1d-restricted iNKT cells , suggesting an important role of CD1b-restricted T cells in anti-mycobacterial immunity . Traditionally , T effector responses against Mtb have been associated with production of IFN-γ by conventional T cells ( Behar , 2013 ) . Aside from cytokine-mediated protection , CD8+ T cells are also capable of contributing to anti-mycobacterial immunity through their cytolytic activities . Even though T cell derived IFN-γ has been associated with macrophage activation-induced control of Mtb , several studies have shown that IFN-γ alone does not absolutely correlate with protection ( Torrado et al . , 2011; Gallegos et al . , 2011 ) . However , IFN-γ , TNF-α and IL-2 polyfunctional cytokine producing T cells have been linked to protection against mycobacteria ( Darrah et al . , 2007; Aagaard et al . , 2009; Aagaard et al . , 2009 ) . The anti-mycobacterial immunity conferred by DN1 T cells was probably mediated in part by the concomitant secretion of cytokines ( IFN-γ , TNF-α and IL-2 ) and cytotoxicity . These data suggest that DN1 T cells employ a two-pronged mechanism for contributing to anti-mycobacterial immunity . A recent study suggests that CD1d-restricted iNKT cells mediate anti-mycobacterial immunity by producing GM-CSF ( Rothchild et al . , 2014 ) . However , activated DN1 T cells did not seem to produce significant amounts of GM-CSF , suggesting that the protective effects of DN1 T cells may not be mediated by this cytokine . Although DN1 T cells shared similar mechanisms of imparting protective immunity against Mtb with conventional T cells , DN1 T cells showed a faster activation and proliferation kinetics than Ag85B-specific CD4+ T cells when they were co-transferred to Mtb-infected mice . Several mechanisms could contribute to the distinct kinetics of DN1 T cells and CD4+ T cells during Mtb infection . Such mechanisms include: the selecting cell type in the thymus; availability of microbial antigen during infection; expression levels of antigen-presenting molecules and efficiency of antigen loading onto antigen presenting molecules . Interestingly , a previous study has shown that rapid DC maturation after Mtb infection induces MHC II trafficking to the plasma membrane without efficient antigen loading . On the other hand , CD1b , which continuously surveys the phagosome for antigens , is properly loaded with Mtb lipid antigen ( Hava et al . , 2008 ) . This leads to earlier antigen presentation by CD1b compared to MHC II molecules . This phenomenon could at least in part explain why priming of DN1 T cells occurs earlier than Ag85B-specific MHC class II-restricted CD4+ T cells during Mtb infection . Even though Mtb lipid-specific group 1 CD1-restricted T cells have been implicated to play a protective role during Mtb infection , this is the first study to directly demonstrate their anti-mycobacterial activity in an in vivo infection setting . The secretion of various cytokines , the cytotoxic capacity , and the ability to be recruited to the site of infection likely contribute to the protective effect of MA-specific T cells during Mtb infection . Additionally , these T cells are activated and proliferate earlier than conventional CD4+ T cells . Given these promising findings , it would be worthwhile to explore whether a multi-subunit vaccine , containing both Mtb protein and lipid antigens that activate both conventional and group 1 CD1-restricted T cells , is more effective against Mtb . Furthermore , like DN1 T cells , a substantial proportion of group 1 CD1-restricted T cells are CD4 negative ( Montamat-Sicotte et al . , 2011; Gong et al . , 1998 ) , and would not be affected by HIV infection . Thus , harnessing this subset of group 1 CD1-restricted T cells in vaccination strategies could prove to be particularly beneficial for HIV-infected patients . 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 Animal Care and Use Committee of the Northwestern University ( Protocol number: 2012–1736 ) . To generate DN1Tg mice , the variable regions of TCR genes were amplified from DN1 TCR plasmids ( Grant et al . , 1999 ) using the using the following primer pairs: TRAV13-2-for-5’-CCAAGATCTACCATGGCAGGCATTCG-3’ and TRAJ57-rev-5’-CACAGCAGGTTCTGGGTTCTGGATATATG-3 , and TRBV5-1-for-5’-CCTGGCCCAATGGGCTCCAG-3’ and TRBJ2∼ 7-rev-5’-GGAGTCACATTTCTCAGGTCCTCTGTGAC . The constant regions of mouse TCR were amplified from a HJ1 TCR plasmid ( Li et al . , 2011 ) , which encodes murine TCR α and b chain linked together by a 2A peptide , using the following primer pairs: Cα-for-5’-CATATATCCAGAACCCAGAACCTGCTGTG-3’ and 2Aend-rev-5’-CTGGAGCCCATT GGGCCAGG-3’ , and Cβ-for-5’GTCACAGAGGACCTGAGAAATGTGACTCC-3’ and Cβ-rev-5’-GCGTCGCTCGAGTCAGGAATTTTTTTTC-3’ . Recombinant PCR was performed to connect human VαJα , murine Cα-2A , human VβDβJβ and murine Cβ fragments . Amplified TCR fragment was cloned into the VA hCD2 cassette vector ( Zhumabekov et al . , 1995 ) . DNA fragment containing promoter and locus control regions of human CD2 and chimeric DN1 TCR was excised from the vector by NotI /SalI digestion and injected into fertilized B6 oocytes by the Northwestern Transgenic Core Facility . The presence of DN1 TCR in the genomic DNA of transgenic mice was examined by PCR using the TRAV13-2-for and TRAJ57-rev primers . DN1Tg mice were further crossed onto hCD1Tg ( Felio et al . , 2009 ) and Rag-/- backgrounds . P25Tg mice , expressing TCR specific for I-Ab/Mtb Ag85B peptide ( aa 240–254 ) , were purchased from Jackson Lab and further crossed onto Rag-/- background . D7Tg mice , expressing TCR specific for H2-M3/LemA peptide , were generated in our lab and were crossed onto the Rag-/- background ( Chiu et al . , 1999 ) . Single-cell suspensions from organs were prepared and stained with the appropriate combinations of mAbs as described previously ( Felio et al . , 2009 ) . All mAbs were purchased from BioLegend ( San Diego , CA ) and BD Biosciences ( San Jose , CA ) . PerCPCy5 . 5-conjugated anti-TCRβ , PerCP-conjugated anti-CD45 . 2 , APC-conjugated anti-hVβ5 . 1 , Pacific blue-conjugated anti-CD4 , IFN-γ and CD107a , FITC-conjugated anti-CD44 , CD122 and TNF-α , PE-conjugated anti-CD69 , PLZF , CD1b , CD5 and IL-2 , PeCy7-conjugated anti-CD62L and CD44 , and BV510 conjugated anti-CD8 mAbs were used . PLZF expression on thymocytes was analyzed via intracellular staining using the FoxP3 staining buffer set ( eBioscience , San Diego , CA ) . For CD107a staining , anti-CD107a mAb was added during in vitro stimulation . For intracellular cytokine staining , cells were fixed with 4% paraformaldehyde , permeabilized with 0 . 1% saponin , and then stained with anti-cytokine mAbs . Flow cytometry was performed with a FACSCanto II and analyzed using FlowJo software . Fluorescence Minus One ( FMO ) was used to identify gating boundaries . Thymic stromal cells were prepared by digesting thymus with collagenase IV ( 1 mg/ml ) and DNase ( 30 μg/ml ) . BMDCs and BMDMs were differentiated from mouse bone marrow progenitors as previously described ( Chun et al . , 2003 ) . To generate DN1 effector cells , naïve DN1 T cells were cultured with irradiated hCD1Tg BMDCs pulsed with 10 μg/ml of mycolic acids ( Sigma ) for 1–2 weeks . 1×107 BM cells from DN1Tg/Rag-/- and DN1Tg/hCD1Tg/Rag-/- mice were injected i . v . into recipients that were irradiated with 1000 rad one day before . Lymphocytes isolated from recipient mice were analyzed by flow cytometry 5–6 weeks after transfer . DN1 T cells were purified from the spleen and lymph nodes from DN1Tg/hCD1Tg/Rag-/- mice through depletion of CD11c+ , DX5+ , CD11b+ and MHC II+ cells using mAb and streptavidin magnetic beads ( Miltenyi Biotec ) . BMDCs from WT and hCD1Tg mice were pulsed with 10 μg/ml MA ( Sigma , St Louis , MO overnight . ELISPOT assay was performed as previously described ( Felio et al . , 2009 ) . IFN-γ producing cells were quantitated using an ImmunoSpot reader ( Cellular Technology Ltd . , Shaker Heights , OH ) . DN1 effectors were established by stimulating naïve DN1 T cells isolated from pooled peripheral lymph nodes of DN1Tg/hCD1Tg/Rag-/- mice with MA-pulsed hCD1Tg BMDCs for 7 days in supplemented Mischell Dutton medium with IL-2 ( 20 U/mL ) . Target cells were labeled with 50 μCi [51Cr] sodium chromate for 1 hr and cultured with effectors for 4 hr at 37°C . The percentage of specific lysis was calculated as ( experimental release-spontaneous release ) / ( maximum release-spontaneous release ) ×100 . Mtb H37Rv was grown and prepared as previously described ( Felio et al . , 2009 ) . Mtb were added to BMDCs and BMDMs at an effective multiplicity of infection ( MOI ) of 1 for CFU experiments ( or 5 for ELISA and FACS assays ) for 2 hr . Cultures were washed three times and treated with 20 ng/ml gentamycin for 2 hr to remove extracellular bacteria . DN1 T cells purified from DN1Tg/hCD1Tg/Rag-/- mice ( 3x105cells/well ) were added one day after infection . Culture supernatants were collected after 48 hr of co-culture and cytokines in the supernatant were detected using Cytometric Bead Array or by ELISA . For CFU measurement , cells were lysed at day 7 post infection with 1% Triton X-100 in PBS , lysate were plated in serial dilutions on Middlebrook 7H11 agar plates and cultured at 37°C for 2–3 weeks . Neutralizing mAb for IFN-γ and TNF-α were added on day 1 post infection at 10 μg/ml . For Mtb aerosol infection , mice were infected with 100–200 CFU using a nose-only aerosol exposure chamber ( In-Tox Products , Moriarty , NM ) . At indicated time-points after infection , bacterial burdens in lungs and spleens were determined by plating serial dilutions of homogenate on Middlebrook 7H11 agar plates . For in vivo protection experiments , 5–7×106 DN1 or D7 effector cells were injected i . v . into recipients 1 day before infection . For other adoptive transfer experiments , 3–5×106 CFSE-labeled naïve DN1 T cells ( from DN1Tg/hCD1Tg/Rag-/- mice ) or CellTraceTM Violet-labeled P25 T cells ( from P25Tg/Rag-/- mice ) were transferred into recipients that were infected 7 days before . Lungs from Mtb infected mice were fixed in 10% neutral buffered formalin and embedded in paraffin . 5 μm sections were acquired and deparaffinized . Sections were blocked with 5% normal donkey serum and incubated with primary antibody ( CD3 , Abcam-16669 , UK ) . Subsequently , sections were washed and stained with secondary antibody followed by incubation with Vectastain ( Vector Laboratories , Burlingame , CA ) . Lastly , sections were incubated with Biotinyl Tyramide working solution and streptavidin-HRP . The slides were scanned using the digital TissueFAXS imaging system equipped with a Zeiss Axio microscope ( TissueGnostics GmbH , Austria ) . The area of granulomas and number of CD3 positive cells in the tissue sections were analyzed using HistoQuest image analysis software . Statistical analysis was performed with the unpaired Student’s t test and one-way ANOVA followed by Bonferroni post-hoc test ( for comparison of in vivo CFU data between different groups ) . All statistical analyses were performed with Prism software . Statistically significant differences are noted ( p<0 . 001; p<0 . 01; *p<0 . 05 ) .	Most cases of tuberculosis are caused by a bacterium called Mycobacterium tuberculosis , which is believed to have infected one third of the world’s population . Most of these infections are dormant and don’t cause any symptoms . However , active infections can be deadly if left untreated and often require six months of treatment with multiple antibiotics . One reason why these infections are so difficult to treat is because the M . tuberculosis cell walls contain fatty molecules known as mycolic acids , which make the bacteria less susceptible to antibiotics . These molecules also help the bacteria to subvert and then hide from the immune system . The prevalence of the disease and the increasing problem of antibiotic resistance have spurred the search for an effective vaccine against tuberculosis . While most efforts have focused on using protein fragments in tuberculosis vaccines , some evidence suggests that human immune cells can recognize fatty molecules such as mycolic acids and that these cells could help manage and control M . tuberculosis infections . However , it has been difficult to determine whether these immune cells genuinely play a protective role against the disease because most vaccine research uses mouse models and mice do not have an equivalent of these immune cells . Now , Zhao et al . have engineered a “humanized” mouse model that produces the fatty molecule-specific immune cells , and show that these mice do respond to the presence of mycolic acids . Infecting the genetically engineered mice with M . tuberculosis revealed that the fatty molecule-specific immune cells were quickly activated within lymph nodes at the center of the chest . These cells later accumulated at sites in the lung where the bacteria reside , and ultimately protected against M . tuberculosis infection . The results show that these specific immune cells can counteract M . tuberculosis , and highlight the potential of using mycolic acids to generate an effective vaccine that provides protection against tuberculosis .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "immunology", "and", "inflammation" ]	2015	Mycolic acid-specific T cells protect against Mycobacterium tuberculosis infection in a humanized transgenic mouse model
Embryonic stem ( ES ) cells go though embryo-like cell cycles regulated by specialized molecular mechanisms . However , it is not known whether there are ES cell-specific mechanisms regulating mitotic fidelity . Here we showed that Autoimmune Regulator ( Aire ) , a transcription coordinator involved in immune tolerance processes , is a critical spindle-associated protein in mouse ES ( mES ) cells . BioID analysis showed that AIRE associates with spindle-associated proteins in mES cells . Loss of function analysis revealed that Aire was important for centrosome number regulation and spindle pole integrity specifically in mES cells . We also identified the c-terminal LESLL motif as a critical motif for AIRE’s mitotic function . Combined maternal and zygotic knockout further revealed Aire’s critical functions for spindle assembly in preimplantation embryos . These results uncovered a previously unappreciated function for Aire and provide new insights into the biology of stem cell proliferation and potential new angles to understand fertility defects in humans carrying Aire mutations . Self-renewal capability , defined as the ability of cells to proliferate while sustaining differentiation potential , is one of the defining features of stem cells ( Martello and Smith , 2014 ) . Robust proliferation ability , not only rapid cell division but also stable karyotype maintenance , ensures the maintenance and expansion of stem cell populations . Embryonic stem ( ES ) cells , the in vitro counterpart of preimplantation epiblast cells ( Boroviak et al . , 2014; Boroviak and Nichols , 2014 ) , possess particularly robust self-renewal capability ( Martello and Smith , 2014 ) . Unlike most somatic stem cell types that divide relatively infrequently , ES cells undergo constant proliferation while accumulating few karyotypic abnormalities ( Suda et al . , 1987; Weissbein et al . , 2014 ) . ES cells undergo specialized embryo-like cell cycles , characterized as rapid cell cycles with brief gap phases ( Kareta et al . , 2015; Coronado et al . , 2013 ) . Unlike in most somatic cells where coordinated fluctuation of Cyclin-CDK activities drive the cells through the cell cycle , ES cells possess higher and constant activity of most Cyclin-CDK pairs except for Cyclin B-CDK1 ( Stead et al . , 2002 ) . These cell cycle patterns have been proposed to limit the window in which ES cells are responsive to differentiation cues , therefore promoting self-renewal ( Dalton , 2015; Pauklin and Vallier , 2013 ) . However , it also subjects the mitosis process to additional stress since the fluctuating activity of Cyclin-CDKs is vital for aspects of mitotic fidelity , such as centrosome maturation/amplification and kinetochore microtubule stabilization ( Lacey et al . , 1999; Haase et al . , 2001; Chen et al . , 2002 ) . Given that ES cells actually show remarkable karyotypic stability , this might suggest that they possess additional specific molecular mechanisms to ensure mitosis fidelity . Autoimmune Regulator ( Aire ) is a central coordinator of immune tolerance . It is specifically expressed in medullary thymic epithelium cells ( mTECs ) and mediates negative selection against auto-reactive T cells by inducing promiscuous expression of tissue specific antigens ( Anderson et al . , 2002; Ramsey et al . , 2002 ) . However , it has long been known that Aire is also expressed in germ cell progenitors and that mutations of Aire could cause fertility defects in mouse and human ( Finnish-German APECED Consortium et al . , 1997; Hubert et al . , 2009; Schaller et al . , 2008 ) . Fertility defects were frequently attributed to aneuploidy in embryo cells , originating from meiotic or mitotic errors ( Tempest and Martin , 2009; Martin , 2008; Webster and Schuh , 2017; Daughtry and Chavez , 2016 ) , implying a possible role of Aire in these processes . We and others have reported the expression of Aire in mouse ES cells and early embryos ( Gu et al . , 2010; Nishikawa et al . , 2010; Bin et al . , 2012 ) . The expression of Aire was specific to undifferentiated mouse ES cells and gradually diminished with differentiation at both mRNA and protein levels ( Gu et al . , 2010 ) . Among embryo-derived stem cell lines , Aire expression was specific to pluripotent ES cells and not found in extraembryonic trophoblast stem cells ( the counterpart of postimplantation extraembryonic ectoderm ( EXE ) progenitor cells ) ( Roberts and Fisher , 2011 ) or eXtraembryonic Endoderm cells ( our unpublished data ) at both mRNA and protein levels . Interestingly , Aire has also been shown to be up-regulated during the final stage of induced pluripotent stem cell ( iPS ) formation when iPS clones become transgene-independent ( Hussein et al . , 2014 ) . The presence of Aire mRNA has been detected in mouse oocytes and all preimplantation stages and early postimplantation stages ( up till E6 . 5 ) embryos but the expression levels and patterns at the protein level are unknown ( Nishikawa et al . , 2010 ) . Moreover , it was recently shown that Aire mRNA is highly expressed in peri-implantation ( E4 . 5–5 . 5 ) mouse epiblast cells ( Boroviak et al . , 2014; Chen et al . , 2016 ) , the in vivo cell type that naïve ES cells most likely represent . We have previously shown that knocking down Aire in mouse ES cells caused self-renewal and karyotype defects , suggesting a role for Aire in mitosis ( Gu et al . , 2010 ) . However , the underlying mechanism remained elusive . AIRE is not a canonical transcription factor that recognizes specific DNA sequence motifs ( Mathis and Benoist , 2009 ) , and so there is considerable interest in understanding how AIRE functions in inducing promiscuous expression and other processes . Identifying its interaction partners is critical for revealing the molecular pathways in which AIRE is involved . Through different methods including GST pulldown , yeast two-hybrid and co-immunoprecipitation followed by mass spectrometry , a number of interacting partners , including DAXX , P-TEFb and ATF7ip-MBD1 , have been identified for AIRE ( Abramson et al . , 2010; Meloni et al . , 2010; Oven et al . , 2007; Waterfield et al . , 2014 ) . The general conclusion from these studies is that AIRE performs its transcriptional function by acting as a hub protein , coordinating chromatin remodeling , general transcription , RNA processing and nuclear transport , to induce promiscuous gene expression ( Abramson et al . , 2010 ) . However , most of these studies were performed in somatic cell lines lacking endogenous Aire expression , which may not reveal its natural interactome . Moreover , the co-immunoprecipitation methods used in most of these studies are more efficient in identifying soluble proteins , rather than proteins tightly bound to insoluble structures like condensed chromosomes or mitotic spindles ( Lambert et al . , 2015 ) . Other approaches are more suited for capturing partners of insoluble components . BioID is an in vivo proximity biotinylation based assay for identifying proximity partners of a protein of interest . In this assay the protein of interest ( bait protein ) is expressed as a fusion protein with an abortive biotin ligase ( BirA R118G or BirA* ) which effectively catalyzes the formation of locally-concentrated activated biotin that covalently labels proximal proteins , the labeled proteins can be recovered under harsh solubilizing conditions , purified by avidin affinity purification and then identified by mass spectrometry analysis . BioID has been employed to identify proximity interactions in relatively insoluble structures including chromatin , nuclear envelope and centrosomes ( Roux et al . , 2012; Lambert et al . , 2015; Gupta et al . , 2015; Kim et al . , 2014 ) . We reasoned that identification of AIRE’s partners with BioID methods in ES cells could help reveal its full function in stem cells and uncover new regulatory aspects of stem cell self-renewal . We characterized the proximal protein partners of AIRE in mouse ES ( mES ) cells using BioID technology and found that , besides proteins functioning in known AIRE-related processes such as general transcription and RNA processing , AIRE also interacts with a group of mitotic spindle-associated proteins . We present evidence that AIRE is a spindle-associated protein in mES cells where it is essential for spindle assembly , centrosome number and structural maintenance . We further showed that the last LxxLL ( LESLL ) motif , a multifunctional motif that has been implicated in mediating protein-protein interactions ( Plevin et al . , 2005 ) , is critical for the mitotic function of AIRE in mES cells and that a known human disease mutation specifically affecting this motif could induce mitotic defects . We also present data showing that maternal-zygotic Aire knockout embryos have mitosis defects at the blastocyst stage . These results provide a new insight into Aire’s non-immune functions in stem cells and suggest novel ES-specific mechanisms for regulating mitotic fidelity . To identify the interaction partners of AIRE in ES cells , we conducted BioID analysis ( Figure 1A ) ( Roux et al . , 2012 ) . ES cells expressing mCherry-BirA* were used as controls to exclude promiscuous biotinylation activities of the BirA* enzyme and Aire-BirA* expressing ES cell samples were used to identify interaction partners . As shown by streptavidin-FITC staining in Figure 1B , mCherry-BIRA* biotinylated proteins distributed evenly in cells , while AIRE- BIRA* biotinylated proteins showed an interesting mitotic spindle association ( Figure 1B arrowheads and magnified images ) . Western blotting analysis confirmed that the expression level of mCherry-BIRA/AIRE-BIRA and biotinylated protein amounts were similar between controls and samples ( Figure 1—figure supplement 1A ) . The biotinylated proteins were then subjected to LC-MS/MS analysis and a high confidence list of AIRE-interacting partners was generated using the SAINT algorithm with FDR ≤ 1% ( Figure 1C , Figure 1—source data 1 ) ( Choi et al . , 2011; Teo et al . , 2014 ) . Strikingly , Gene Ontology ( GO ) analysis of AIRE-interacting partners revealed that the two most enriched Biological Processes ( BP ) were ‘mitotic nuclear division’ and ‘cell division’ ( Figure 1—figure supplement 2 ) . As for Cellular Compartment ( CC ) , mitosis-related terms such as ‘cytoskeleton’ , ‘microtubule’ ‘centrosome’ , ‘spindle pole’ and ‘spindle’ were enriched ( Figure 1—figure supplement 2 ) . The interactors included proteins functioning in three crucial aspects of mitosis ( Figure 1C ) : ( i ) kinase cascade regulation of the mitotic process ( eg . AURKB , CDK2 ) ( Wieser and Pines , 2015 ) , ( ii ) centrosome/spindle pole maturation and integrity ( eg . SPICE1 , HAUS5 , HAUS8 ) ( Fu et al . , 2015; Comartin et al . , 2013; Lawo et al . , 2009 ) and ( iii ) regulation of microtubule dynamics in spindles ( eg . CKAP2 , CLASP1 , CLASP2 ) ( Bratman and Chang , 2008 ) . These data suggested spindle localization and mitosis-associated functions of AIRE in ES cells . Additionally , proteins functioning in general transcription processes and RNA processing were also identified , suggesting that Aire may also play a similar role in ES cells as in immune cells ( Figure 1C ) . Considering the proximity ligation nature of BioID ( ~10 nm [Kim et al . , 2014] ) , we performed Duolink Proximity Ligation Assay ( PLA ) ( Söderberg et al . , 2006 ) and validated a number of interaction partners of AIRE in ES cells . Duolink signal between anti-Flag M2 antibody and antibodies against respective candidate partners in mES cells before doxycycline induction of 3XFlag-AIRE were used to control background signal and Duolink signal between 3XFlag-AIRE and AURKA , a protein not identified as AIRE’s proximity partner but highly related to an identified partner AURKB , was used as negative control . As shown in Figure 1—figure supplement 1B , six AIRE proximity partners showed strong Duolink signal over background while AURKA only displayed background level signals , 10 . 7554/eLife . 28131 . 003Figure 1 . Proximity partners of AIRE in ES cells . ( A ) Flow chart of BioID experiments . mCherry-BirA-HA expressing ES cells were used as control to subtract the effect of promiscuous biotinylation activity of BirA . ( B ) Localization of bait proteins ( mCherry or HA ) and biotin-tagged proteins ( Streptavidin-FITC ) in mCherry-BirA-HA control ES cells and Aire-BirA-HA ES cells shown by immunofluorescence staining . Scale bars: 1 μm . Bottom panel for each group shows magnified views of a mitotic cell , marked by arrows in top panel . Scale bars: 4 μm for mCherry-BirA-HA , 2 μm for Aire-BirA-HA . ( C ) List of proximity partners of AIRE identified by SAINT analysis ( n = 3 , 1%FDR ) of BioID/mass spectrometry data and annotations of functional categories . DOI: http://dx . doi . org/10 . 7554/eLife . 28131 . 00310 . 7554/eLife . 28131 . 004Figure 1—source data 1 . List of proximity partners of AIRE . DOI: http://dx . doi . org/10 . 7554/eLife . 28131 . 00410 . 7554/eLife . 28131 . 005Figure 1—figure supplement 1 . Validations of BioID analysis . ( A ) Western Blotting analysis of mCherry-BirA-HA and Aire-BirA-HA proteins in mES cells ( left: anti-HA antibody ) and biotinylated cellular proteins ( right: streptavidin-HRP ) . Each lane corresponds to protein from 105 cells from one biological replicate . ( B ) Proximity Ligation Amplification validation of the interaction between AIRE and a panel of proximity partners in TetON-3XFlag-Aire mES cells . Red signals: Duolink Orange . DOI: http://dx . doi . org/10 . 7554/eLife . 28131 . 00510 . 7554/eLife . 28131 . 006Figure 1—figure supplement 2 . Gene ontology enrichment analysis of AIRE’s proximity partners with DAVID . Left panel: Biological Processes ( BP ) . Right panel: Cellular Compartment ( CC ) . X axis: -Log ( P ) . DOI: http://dx . doi . org/10 . 7554/eLife . 28131 . 006 The identification of mitotic interaction partners of AIRE in mES cells prompted us to examine the subcellular localization of AIRE in mES cells undergoing mitosis . Immunostaining with an antibody against mouse AIRE showed co-localization of AIRE with α-tubulin ( TUBA ) -marked mitotic spindles ( Figure 2A ) . To further corroborate this finding , we generated Aire-3XFlag mice ( C-terminal 3XFlag-Tag fusion to endogenous Aire ) using CRISPR-Cas9 technology and derived ES cells from these mice . Co-staining with the M2 anti-Flag antibody and β-tubulin ( TUBB ) antibody showed that AIRE co-localized with the mitotic spindle apparatus ( Figure 2—figure supplement 1 ) . To gain further insight into the precise localization of AIRE on the spindle , we performed Structured Illumination Microscope ( SIM ) analysis . We found AIRE presented as localized foci along the spindle microtubules during metaphase and anaphase ( Figure 2B ) . It also formed a cloud-like structure covering the spindle pole and its peripheral region , suggesting a role in spindle pole/centrosome functions . In mitotic telophase cells , AIRE localized to the base of the mid-spindle , sparing the central part of midbody ( Figure 2B ) , indicating a possible role in cytokinesis . These results identify AIRE as a bona fide spindle associated protein during mitosis in ES cells , in addition to its earlier reported nuclear foci localization in interphase cells ( Gu et al . , 2010 ) . 10 . 7554/eLife . 28131 . 007Figure 2 . AIRE localizes to mitotic spindles in ES cells . ( A ) Immunofluorescence staining of AIRE and α-tubulin ( TUBA ) in AmES8 cells . Scale bar: 10 μm . ( B ) Structural Illumination Microscope imaging of spindle localization of AIRE in AIRE-3XFlag expressing ES cells during different phases of mitosis . The Flag-tag was detected using the M2 mouse monoclonal antibody and spindle microtubules were marked by β-tubulin ( TUBB ) . Scale bar: 2 . 5 μm . ( C ) Localization of different truncated versions of Aire-3XFlag ( shown above each image ) in mitotic ES cells by immunofluorescence staining . Scale bar: 5 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 28131 . 00710 . 7554/eLife . 28131 . 008Figure 2—figure supplement 1 . 3XFlag-AIRE localized to mitotic spindles in Aire-3XFlag mES cell . Immunofluorescence image of homozygous Aire-3XFlag mES cell in mitosis ( upper panel ) and wild type mES control ( AmES8 ) ( bottom panel ) ( Blue: DAPI , Green: M2 ( AIRE-3XFlag ) , Red: TUBA ) , bar: 8 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 28131 . 00810 . 7554/eLife . 28131 . 009Figure 2—figure supplement 2 . AIREΔc70 functions as a dominant negative , disrupting spindle assembly and inhibiting proliferation in mES cells . ( A ) Left: immunofluorescence images of mitotic mES cells after 12 hr of Dox+ or control ( Dox- ) induction of AIREΔc70 expression . Scale bar: 8 μm . Right: relative ALP positive colony number ( Dox+/Dox- ) after 72 hr of Dox+ or control ( Dox- ) induction of AIREΔc70 expression in mES cells . Data presented as mean ± sd of 3 biological replicates . ( B ) Immunofluorescence images of ES cell markers OCT4 , NANOG and SOX2 in mES with ( Dox+ ) or without ( Dox- ) 12 hr induction of AIREΔc70 expression . Scale bar: 10 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 28131 . 009 We then performed domain mapping to identify the domains required for the spindle localization of AIRE . Wild type and six truncated forms of Flag-tagged AIRE ( Figure 2C ) were introduced into ES cells as doxycycline ( Dox ) inducible transgenes and the spindle localization was investigated following addition of Dox by immunofluorescence staining . We found that the removal of either the HSR/CARD domain or the SAND domain abrogated the spindle localization of AIRE , while deletions of any of the two PHD domains or the C-terminal 70 amino acid ( aa ) flexible tail region ( AIREΔc70 ) had no effect on localization to the spindle ( Figure 2C ) . However , defective spindle morphology and decreased colony formation ability were observed upon 12 hr overexpression of the AIREΔc70 truncated form , indicating a dominant negative effect on spindle assembly ( Figure 2—figure supplement 2A ) . The expression of core pluripotency factors OCT4 , NANOG and SOX2 was not changed at this time point ( Figure 2—figure supplement 2B ) . Therefore the two N-terminal domains HSR/CARD and SAND were critical for the spindle recruitment of AIRE while the last 70aa were essential for its spindle-associated function . We were not able to generate homozygous Aire null ES clones directly through CRISPR-Cas9 mediated knockout , likely due to an essential role of Aire in ES cells . In order to conduct Aire loss of function analysis , we generated an Aire floxed ( Fl ) mouse line using CRISPR-Cas9 technology ( Figure 3—figure supplement 1 ) . The mice were then bred with R26CreERT mice and ES cell lines were derived . A pair of Aire+/+;R26CreERT+/+ and AireFl/Fl;R26CreERT+/+ ( Pair 1 ) ( Figure 3 ) and a pair of AireFl/+;R26CreERT ± and AireFl/Fl;R26CreERT+/- ( Pair 2 ) ( Figure 3—figure supplement 1 ) ES cell clones were used for the experiments . 4OH-Tamoxifen ( TAM ) treatment induced recombination only in AireFl/Fl;R26CreERT+/+ cells ( from hereon referred to as Aire-/- cells ) ( Figure 3A ) . The knockout of the targeted region ( exons 8–9 ) was validated by Sanger sequencing . All control lines ( Aire+/+;R26CreERT+/+ , Aire+/+;R26CreERT+/+ + TAM and AireFL/FL;R26CreERT+/+ ) were normal and showed no changes in gene expression , morphology or colony formation . For simplicity we refer to all control lines as Aire+/+ cells in the main text from here on . qPCR analysis showed the loss of the expression of Aire mRNA containing Exon 8 and 9 in Aire-/- cells ( Figure 3B ) . A pair of primers against the 3’ end of Aire mRNA detected similar level of expression in both Aire+/+ and Aire-/- cells ( Figure 3B ) , indicating production of truncated mRNAs . However , the complete loss of AIRE protein expression was validated by immunostaining ( Figure 3C ) . Consistent with previous data from Aire knockdown ES cells ( Gu et al . , 2010 ) , the mRNA level of the pluripotency markers Pou5f1 ( Oct4 ) and Nanog were decreased in Aire-/- cells while no significant difference was detected for Sox2 . Correspondingly we also detected a decrease of Oct4 and Nanog protein levels in Aire-/- cells by immunofluorescence analysis ( Figure 3—figure supplement 3 ) . Both proliferation and colony formation capability of Aire-/- cells were impaired ( Figure 3D ) , suggesting a role of Aire in ES cell self-renewal . We investigated the spindle structures by immunofluorescence staining of α-tubulin ( TUBA ) and γ-tubulin ( TUBG ) and found defective spindles with multiple γ-tubulin positive foci in Aire-/- cells ( Figure 3E ) . Quantitative analysis showed a significant increase in the number of γ-tubulin foci in Aire-/- mitotic cells ( Figure 3F ) . We obtained similar phenotypes using the pair two conditional Aire knockout ES cell lines ( Figure 3—figure supplement 2 ) . 10 . 7554/eLife . 28131 . 010Figure 3 . Aire is essential for spindle assembly in ES cells . ( A ) Scheme of the AireFl allele and genotyping strategy before and after cre-induced deletion . Genotyping results from Aire+/+R26CreERT+/+ and AireFl/Fl R26CreERT+/+ ES cells without tamoxifen ( - ) and with 5 ng/ml tamoxifen for 24 hr ( + ) . ( B ) qPCR analysis of gene expression at passage one after tamoxifen treatment in Aire+/+R26CreERT+/+ and AireFl/Fl R26CreERT+/+ ES cells . Aire del: primers complementary to deleted region of Aire . Aire c-term: primers complementary to c-terminal region of Aire . Data presented as mean ± sd of 3 biological replicates . p-values were calculated with Wilson’s t-test . ( C ) Validation of Aire knockout by immunofluorescence imaging in Aire+/+R26CreERT+/+ and AireFl/Fl R26CreERT+/+ ES cells without tamoxifen ( - ) and with 5 ng/ml tamoxifen for 24 hr ( + ) . bar: 16 μm . ( D ) Proliferation over two passages and colony formation assay showing total cell number ( top panel ) and ALP positive colony number ( bottom panel ) in tamoxifen treated over non-treated Aire+/+R26CreERT+/+ and AireFl/Fl R26CreERT+/+ ES cells . Data in each panel presented as mean ± sd of 3 biological replicates . p-values were calculated with Wilson’s t-test . ( E ) Representative immunofluorescence images of γ-tubulin ( TUBG ) foci on mitotic spindles in passage 1 Aire+/+R26CreERT+/+ and AireFl/Fl R26CreERT +/+ ES cells without tamoxifen ( - ) and with 5 ng/ml tamoxifen for 24 hr ( + ) . Scale bar: 4 μm . Right column shows magnified views of spindle poles . Scale bar: 1 μm . ( F ) Quantitation of γ-tubulin ( TUBG ) foci on mitotic spindles in passage 1 Aire+/+R26CreERT+/+ and AireFl/Fl R26CreERT+/+ ES cells without tamoxifen ( Con ) and with 5 ng/ml tamoxifen for 24 hr ( TAM ) . γ-tubulin ( TUBG ) foci were counted from 30 mitotic cells from three biological replicates for each group . Each point represents one mitotic cell . Error bars: s . d . of mean . p-values were calculated with Mann–Whitney–Wilcoxon test . DOI: http://dx . doi . org/10 . 7554/eLife . 28131 . 01010 . 7554/eLife . 28131 . 011Figure 3—figure supplement 1 . Complete diagram of Aire floxed ( Fl ) allele and cre induced depletion of critical exons . DOI: http://dx . doi . org/10 . 7554/eLife . 28131 . 01110 . 7554/eLife . 28131 . 012Figure 3—figure supplement 2 . Aire knockout abrogated proliferation and spindle assembly in Pair 2 ES cells . ( A ) Genotyping results from Aire Fl/Fl R26CreERT ± and AireFl/+R26CreERT ± ES cells without tamoxifen ( - ) and with 20 ng/ml tamoxifen for 24 hr ( + ) . ( B ) qPCR analysis of gene expression at passage one after tamoxifen ( TAM ) treatment in Aire Fl/Fl R26CreERT ± and AireFl/+R26CreERT ± ES cells . Aire del: primers complementary to deleted region of Aire . Aire c-term: primers complementary to c-terminal region of Aire . Data presented as mean ± sd of 3 biological replicates . p-values were calculated with Wilson’s t-test . ( C ) Validation of Aire knockout by immunofluorescence imaging in Aire Fl/Fl R26CreERT ± and AireFl/+R26CreERT ± ES cells without tamoxifen ( - ) and with 20 ng/ml tamoxifen for 24 hr ( + ) . Scale bar: 16 μm . ( D ) Proliferation over two passages and colony formation assay showing total cell number ( top panel ) and ALP positive colony number ( bottom panel ) in tamoxifen treated over non-treated ( TAM/Con ) Aire Fl/Fl R26CreERT ± and AireFl/+R26CreERT ± ES cells . Data in each panel presented as mean ± sd of 3 biological replicates . p-values were calculated with Wilson’s t-test . ( E ) Representative immunofluorescence images of γ-tubulin ( TUBG ) foci on mitotic spindles in passage 1 Aire Fl/Fl R26CreERT ± and AireFl/+R26CreERT ± ES cells without tamoxifen ( - ) and with 20 ng/ml tamoxifen for 24 hr ( + ) . Scale bar: 10 μm . Right column shows magnified view of spindle poles . Scale bar: 0 . 5 μm . ( F ) Quantitation of γ-tubulin ( TUBG ) foci on mitotic spindles in passage 1 Aire Fl/Fl R26CreERT ± and AireFl/+R26CreERT ± ES cells without tamoxifen ( Con ) and with 20 ng/ml tamoxifen for 24 hr ( TAM ) . γ-tubulin ( TUBG ) foci were counted from 30 mitotic cells from three biological replicates for each group . Each point represents one mitotic cell . Error bars: s . d . of mean . p-values were calculated with Mann–Whitney–Wilcoxon test . DOI: http://dx . doi . org/10 . 7554/eLife . 28131 . 01210 . 7554/eLife . 28131 . 013Figure 3—figure supplement 3 . Immunofluorescence images of pluripotency markers in Aire Floxed;R26CreERT cells . Scale bar: 1 μm . ( A ) Aire+/+R26CreERT+/+: Upper: Con; Bottom: TAM . ( B ) AireFl/Fl R26CreERT +/+: Upper: Con; Bottom: TAM . ( C ) AireFl/+R26CreERT+/-: Upper: Con; Bottom: TAM . ( D ) Aire Fl/Fl R26CreERT+/-: Upper: Con; Bottom: TAM . DOI: http://dx . doi . org/10 . 7554/eLife . 28131 . 013 The LxxLL motif has been shown to mediate protein-protein interactions in many different contexts . There are four LxxLL motifs in AIRE protein ( Uniprot Q9Z0E3 ) , with the last one ( LESLL ) within the last 70aa of the protein ( Figure 4A ) . Since the c-terminal 70aa truncated form of AIRE produced a dominant negative effect on spindle assembly , we examined whether this motif was essential for the mitotic function of AIRE . Upon overexpression of full length AIRE protein lacking the LESLL motif ( AIRE ΔLESLL ) , most ES cells rounded up and underwent cell death within 48 hr ( Figure 4B ) . Similarly , overexpression of AIRE ΔLESLL almost completely abolished the colony-forming capability of ES cells ( Figure 4B ) . Immunofluorescence analysis demonstrated malformed spindle apparatus in these cells ( Figure 4C ) . Flow cytometry analysis using the mitotic marker pS10-H3 revealed a severe mitotic arrest in AIRE ΔLESLL overexpressing cells ( Figure 4D ) . Notably , although also localized to the spindle in mitotic NIH3T3 cells that do not express Aire endogenously , AIRE ΔLESLL did not induce spindle disruption or mitotic arrest in this cell type ( Figure 4C , D and Figure 4—figure supplement 1 ) . 10 . 7554/eLife . 28131 . 014Figure 4 . The last LxxLL ( LESLL ) motif is essential for the mitotic functions of Aire in ES cells . ( A ) Diagram showing Aire and Aire ΔLESLL transgene structures . ( B ) Characterization of proliferation and colony formation of ES cells upon doxycycline-induced overexpression of AIRE or AIREΔLESLL . Representative bright field images of ES cell cultures 48 hr after doxycycline induction from three biological replicates ( top panel ) . Representative images ( bottom left panel ) and quantification ( bottom right panel ) of ALP-stained colonies of AIRE or AIRE ΔLESLL-overexpressing ES cells 3 days after seeding . Quantification of ALP positive colonies was done from two independent clones for each transgene and normalized to Dox- controls . Data presented as mean ± sd of 3 biological replicates . p-values were calculated with Wilson’s t-test . ( C ) Immunofluorescence staining of spindles in control ( Dox- ) and AIRE or AIRE ΔLESLL-overexpressing ( Dox+ ) ES cells ( left panel ) and NIHT3T cells ( right panel ) . Scale bar: 7 μm . Magnified images of spindles from AIRE ΔLESLL-overexpressing groups . Scale bar: 4 μm . ( D ) Mitotic index ( number of cells in mitosis in Dox+/Dox- conditions ) of AIRE and AIRE ΔLESLL-overexpressing ES cells and NIH3T3 cells . Number of cells in mitosis was determined by flow analysis for PI-4N/pH3 markers . Data presented as mean ± sd of 3 replicates . p-values were calculated with Wilson’s t-test . ( E ) Diagram showing domain structure and partial amino acid sequence of human AIRE and AIRE505fs . ( F ) Immunofluorescence staining of spindles in control ( Dox- ) and AIRE505fs-overexperssing ( Dox+ ) ES cells . Scale bar: 5 μm . ( G ) Characterization of colony formation and mitotic index of ES cells upon doxycycline-induced overexpression of Aire505fs . Top panel: number of ALP-positive colonies per well analyzed for 3 AIRE505fs transgenic ES cell clones without ( Dox- ) and with ( Dox+ ) addition of doxycycline . Data presented as mean ± sd of 3 biological replicates for each clone , p-values were calculated with Wilson’s t-test . Bottom Panel: percent of cells in mitosis in control ( Dox- ) and AIRE505fs overexpressing ( Dox+ ) ES cells . Number of cells in mitosis was determined by flow analysis for PI-4N/pH3 markers ( bottom panel ) . Data presented as mean ± sd of 3 replicates . p-values were calculated with Wilson’s t-test . DOI: http://dx . doi . org/10 . 7554/eLife . 28131 . 01410 . 7554/eLife . 28131 . 015Figure 4—figure supplement 1 . Both AIRE and AIREΔLESLL localized to mitotic spindles in mES and NIH3T3 cells . Immunofluorescence images of spindle microtubules ( β-tubulin ( TUBB ) ) , spindle pole ( γ-tubulin ( TUBG ) ) in AIRE and AIREΔLESLL-overexperssing ( Flag-AIRE ) mES cells and NIH3T3 cells . DOI: http://dx . doi . org/10 . 7554/eLife . 28131 . 01510 . 7554/eLife . 28131 . 016Figure 4—figure supplement 2 . AIREΔLESLL activates the Spindle Assembly Checkpoint in ES cells . ( A ) Left: representative immunofluorescence images of BubR1 ( Green ) in mitotic ES cells without ( Dox- ) or with ( Dox+ ) induction of AIRE or AIREΔLESLL . Scale bar: 5 μm . Right: quantification of BUBR1 foci number/mitotic cell . 50 cells from two biological replicates were quantified for each group . Each point represents one mitotic cell . Error bars: s . d . of mean . p-values were calculated with Mann–Whitney–Wilcoxon test . ( B ) Top: scheme of MPS1 treatment experiment . Bottom: quantification of mitotic index ( the percentage of PI-4N/pH3 +cells ) by flow cytometry after respective treatment shown below the x-axis . Data presented as mean ± sd of 3 biological replicates . DOI: http://dx . doi . org/10 . 7554/eLife . 28131 . 016 The spindle assembly checkpoint ( SAC ) is the main mechanism that arrests cells with defective spindles and subsequently induces apoptosis . We examined the activation state of SAC in AIRE ΔLESLL overexpressing cells and found that the deposition of BubR1 on mitotic chromosomes was significantly increased ( Figure 4—figure supplement 2A ) . Inhibition of the major SAC kinase Mps1 rescued the mitotic arrest induced by AIRE ΔLESLL overexpression , further indicating that functional activation of SAC is the mechanism behind mitotic arrest and cell death in AIRE ΔLESLL overexpressing cells ( Figure 4—figure supplement 2B ) . In humans , mutations in AIRE cause Autoimmune Polyendocrinopathy-Candidiasis-Ectodermal Dystrophy ( APECED ) , which been associated with infertility in some patients through an elusive mechanism . We searched through known human AIRE mutations to see if any would result in production of AIRE truncated proteins mimicking AIRE ΔLESLL . In 2000 , a rare mutation , AIRE 33031delG , was reported from Japan ( Ishii et al . , 2000 ) . The mutation resulted in a frame-shift in the Ala505 codon , an amino acid sequence change thereafter and premature termination at amino acid 520 , which eliminated the LESLL motif ( aa 516–520 ( Figure 4E ) . Since the premature termination codon was located 7 bp upstream of the last Exon-Exon junction , the mRNA would likely escape nonsense-mediated decay , resulting in a truncated protein lacking the last LESLL motif . We investigated the function of this mutation in mouse ES cells by overexpressing a murine Aire version mimicking the AIRE 33031delG-induced truncation ) ( AIRE505fs ) . We found that the truncated protein caused similar spindle defects and mitotic arrest in mES cells ( Figure 4F and G ) . Therefore the last LESLL motif is essential for the mitotic function of Aire in mouse ES cells and a human mutant AIRE lacking the c-terminal residues containing LESLL motif could potentially lead to mitotic defects during human reproduction and development . The BioID data showed that Aire established proximal interactions with a number of proteins functioning in centrosome duplication and maturation , including SPICE1 , HAUS5 and HAUS8 . We also observed extra γ-tubulin positive foci of various sizes on defective spindles in both Aire-/- and AIRE ΔLESLL overexpressing cells ( Figures 3 and 4 ) . Therefore we investigated the centrosome and spindle pole structure under SIM with γ-tubulin/Nedd1 double staining . Interestingly , both Aire-/- and AIRE ΔLESLL overexpressing ES cells possessed extra centrosomes in S/G2 phases , while the size of the centrosome was not obviously different ( images in Figure 5A and C and quantifications in Figure 5B and D , Figure 5—figure supplement 1 ) . During mitosis , spindle poles with increased size consisting of multiple aggregated centrosomes were frequently observed in Aire-/- cells , while in AIRE ΔLESLL overexpressing ES cells , spindle poles were obviously fragmented ( images in Figure 5A and C and quantifications in Figure 5B and D , Figure 5—figure supplement 1 ) . Since AIRE ΔLESLL overexpression caused stronger proliferation and mitosis phenotypes , we further investigated the microtubule organization center ( MTOC ) functions of mitotic centrosomes in these cells . We found that instead of forming bipolar spindles , the spindle microtubules formed a disorganized meshwork with fragmented γ-tubulin rings upon AIRE ΔLESLL overexpression , suggesting the abrogation of MTOC functions ( Figure 5E ) . The spindle pole/centrosome defect observed here could be either due to malformed centrioles or to failed organization and maintenance of pericentriolar material ( PCM ) integrity . Using SIM , we imaged centriole and pericentriolar material by double staining mES cells for acetyl-tubulin and γ-tubulin . We did not observe any obvious morphological defects of centrioles in AIRE ΔLESLL overexpressing cells in both S/G2 and M phases ( Figure 5—figure supplement 2 ) . However , although the γ-tubulin-marked pericentriolar material compactly surrounded the mitotic centriole in all the control groups , it failed to cover the centriole in mitotic AIRE ΔLESLL overexpressing cells ( Figure 5—figure supplement 2 ) . These data suggested a crucial role of Aire in maintaining centrosome number and mitotic spindle pole integrity in ES cells . 10 . 7554/eLife . 28131 . 017Figure 5 . Aire is critical for centrosome number sustenance and the integrity of mitotic spindle poles . ( A ) Representative immunofluorescence SIM images of γ-tubulin ( TUBG ) and NEDD1 in control ( Aire+/+;R26CreERT+/+ tamoxifen-/+and AireFL/FL;R26CreERT+/+ tamoxifen- ) and Aire-/- ( AireFL/FL;R26CreERT+/+ tamoxifen + ) mES cells during S/G2 and M phases of the cell cycle . Scale bar: 0 . 7 μm . ( B ) Quantification of centrosome parameters in control and Aire-/- ES cells ( groups same as in panel ( A ) ) . Centrosome number in 30 S/G2 cells ( top panel ) from three biological replicates were counted for each group , mean ± sd is shown . Each point represents one mitotic cell . p-values were calculated with Mann–Whitney–Wilcoxon test . γ-tubulin ( TUBG ) ring diameter in S/G2 cells ( middle panel ) . Each point represents one centrosome . Mean ± sd was presented . p-values were calculated with Mann–Whitney–Wilcoxon test . γ-tubulin ( TUBG ) foci diameter in M phase cells ( bottom panel ) . 30 spindle poles from three biological replicates were quantified for each group . Each point represents one spindle pole . Mean ± sd is presented . p-values were calculated with Mann–Whitney–Wilcoxon test . ( C ) Representative immunofluorescence SIM images of γ-tubulin ( TUBG ) and NEDD1 in control ( Dox- ) and AIRE or AIRE ΔLESLL-overexpressing mES cells ( Dox+ ) during S/G2 and M phases of the cell cycle . Scale bar: S/G2 Figure 0 . 6 μm; M images in top three panels 0 . 5 μm , and bottom panel 0 . 9 μm . ( D ) Quantification of centrosome parameters in AIRE or AIRE ΔLESLL-overexpressing mES cells ( groups same as in panel ( C ) ) Centrosome number in S/G2 cells ( top panel ) . Each point represents one mitotic cell . Mean ± sd is presented . p-values were calculated with Mann–Whitney–Wilcoxon test . γ-tubulin ( TUBG ) ring diameter in S/G2 cells ( middle panel ) . Each point represents one centrosome . Mean ± sd was presented . p-values were calculated with Mann–Whitney–Wilcoxon test . γ-tubulin ( TUBG ) diameter in M phase cells ( bottom panel ) . Each point represents one spindle pole . Mean ± sd is presented . p-values were calculated with Mann–Whitney–Wilcoxon test . ( E ) Representative immunofluorescence SIM images of spindles and MOTCs ( γ -tubulin and α-tubulin ) in control ( Dox- ) and AIRE or AIRE ΔLESLL-overexpressing mES cells ( Dox+ ) during M phase . Scale bar: 1 . 3 μm . Magnified view of MTOC in AIRE ΔLESLL ( Dox- ) ( top scale bar: 0 . 6 μm ) and AIREΔLESLL ( Dox+ ) ( bottom scale bar: 0 . 4 μm ) . DOI: http://dx . doi . org/10 . 7554/eLife . 28131 . 01710 . 7554/eLife . 28131 . 018Figure 5—figure supplement 1 . Aire is critical for centrosome number sustenance and the integrity of mitotic spindle poles . ( A ) Representative immunofluorescence SIM images of γ-tubulin ( TUBG ) and NEDD1 in control ( AireFl/+;R26CreERT ± tamoxifen-/+and AireFL/FL;R26CreERT ± tamoxifen- ) and Aire-/- ( AireFl/Fl;R26CreERT+/- tamoxifen + ) mES cells during S/G2 and M phases of the cell cycle . Scale bar: 0 . 6 μm for S/G2 images , 0 . 5 μm for M images . ( B ) Quantification of centrosome parameters in control and Aire-/- ES cells ( groups same as in panel ( A ) ) . Centrosome number in 30 S/G2 cells ( top panel ) from three biological replicates were counted for each group , mean ± sd is shown . Each point represents one mitotic cell . p-values were calculated with Mann–Whitney–Wilcoxon test . γ-tubulin ( TUBG ) ring diameter in S/G2 cells ( middle panel ) . Each point represents one centrosome . Mean ± sd was presented . p-values were calculated with Mann–Whitney–Wilcoxon test . γ-tubulin ( TUBG ) foci diameter in M phase cells ( bottom panel ) . 20 spindle poles from three biological replicates were quantified for each group . Each point represents one spindle pole . Mean ± sd is presented . p-values were calculated with Mann–Whitney–Wilcoxon test . DOI: http://dx . doi . org/10 . 7554/eLife . 28131 . 01810 . 7554/eLife . 28131 . 019Figure 5—figure supplement 2 . AIREΔLESLL overexpression caused defects in the pericentiole matter ( PCM ) around centrioles . Representative SIM images of centrioles ( Acetyl-Tubulin ) and PCM ( γ-tubulin ( TUBG ) ) without ( - ) or with ( + ) 12 hr induction of the expression of AIRE or AIREΔLESLL . Left panels: Images of cells in S/G2 phase , scale bar: 0 . 47 μm . Right panels: Images of cells in M phase , scale bar: 0 . 5 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 28131 . 019 To understand whether the mitotic functions of Aire in ES cells reflect its role in germ cells and the early embryo , we generated paternal Aire knockout mice by breeding AireFl/Fl mice with Alpl-Cre transgenic mice and maternal Aire knockout mice by breeding AireFl/Fl mice with Zp3-Cre transgenic mice , respectively . Both male germ cell-specific Aire knockout mice ( AireFl/-;Alpl-Cre+ ) and female oocyte specific Aire knockout mice ( AireFl/Fl;Zp3-Cre+ ) developed normally and were fertile ( Figure 6—figure supplement 1 ) , suggesting that , at least in the CD1 background , germ cell-specific depletion of Aire was not sufficient to induce consistent fertility defects . We then produced maternal/zygotic depleted ( m-z- ) embryos by breeding AireFl/-;Alpl-Cre+ males and AireFl/Fl;Zp3-Cre+ females . Embryos were flushed at embryonic day ( E ) 3 . 5 , the time when centrosome-dependent mitosis is established in the embryo ( Courtois et al . , 2012 ) . As shown in Figure 6A , while control embryos were mostly at the blastocyst stage with a discernable blastocoel cavity , many of the m-z- embryos did not develop a blastocoel . Statistical analysis revealed that m-z- embryos showed a significantly lower blastocyst rate than controls ( Figure 6B ) . Immunostaining revealed Aire protein in both inner cell mass ( ICM ) ( Sox2+ ) and trophectoderm ( Cdx2+ ) cells in control blastocysts ( Figure 6C ) , but it was completely depleted in the m-z- blastocysts and morulae . M-z- embryos still expressed Sox2 and Cdx2 in a mutually exclusive pattern , indicating correct initiation of lineage specification ( Figure 6C and Figure 6—figure supplement 2 ) . However , we noted an overall decrease in total cell number in Aire m-z- embryos , indicative of a proliferation defect ( Figure 6B ) . We then investigated the spindle structure in mitotic cells in Aire m-z- embryos and found an increased number of γ-tubulin ( TUBG ) positive foci compared to control cells , which formed abnormal , multipolar spindles ( Figure 6D and E ) . These results suggested that , similar to ES cells , Aire also plays a critical role in sustaining centrosome number and spindle integrity in the cells of early embryos . 10 . 7554/eLife . 28131 . 020Figure 6 . Aire m-/z- preimplantation embryos showed proliferation delay and spindle defects . ( A ) Brightfield images of control ( Con ) and Aire m-/z- embryos at embryonic day ( E ) 3 . 5 . ( B ) Quantification of blastocyst-formation rate ( left panel ) and cell number at E3 . 5 ( right panel ) in control ( Con ) and Aire m-/z- embryos . Data is presented as mean ± sd . p-values were calculated with Wilson’s t-test . ( C ) Immunofluorescence staining for Aire in control ( Con ) ( upper ) and Aire m-/z- ( down ) E3 . 5 embryos . Scale bar: 22 μm . ( D ) Representative immunofluorescence image of spindles ( γ –tubulin ( TUBG ) , α-tubulin ( TUBA ) ) in E3 . 5 control ( Con ) ( top ) and Aire m-/z- ( bottom ) embryo cells . Scale bar: 2 μm . ( E ) Quantification of γ –tubulin ( TUBG ) positive foci per mitotic cell in control and Aire m-/z- embryos . Each point represents one mitotic cell . Mean ± sd is presented . p-values were calculated with Wilson’s t-test . DOI: http://dx . doi . org/10 . 7554/eLife . 28131 . 02010 . 7554/eLife . 28131 . 021Figure 6—figure supplement 1 . No obvious infertility phenotype was observed in paternal or maternal Aire knockout mice . Data presented as mean ± sd . ( A ) Litter numbers produced in 3 months in control and paternal Aire knockout mice . ( B ) Litter sizes produced in 3 months in control and paternal Aire knockout mice . ( C ) Litter numbers produced in 3 months in control and maternal Aire knockout mice . ( D ) Litter sizes produced in 3 months in control and maternal Aire knockout mice . DOI: http://dx . doi . org/10 . 7554/eLife . 28131 . 02110 . 7554/eLife . 28131 . 022Figure 6—figure supplement 2 . Aire m- z- embryos didn’t show obvious early lineage defects . Representative immunofluorescence images of control ( Con ) and Aire m- z- E3 . 5 embryos for the epiblast marker SOX2 , a trophectoderm marker CDX2 and AIRE . Legends for supplementary files . DOI: http://dx . doi . org/10 . 7554/eLife . 28131 . 022 In this study , we report a novel function of the Autoimmune Regulator , Aire , in the mitotic processes of embryonic stem cells and the early embryo . Using the BioID method , we identified proximity partners for AIRE in ES cells . In addition to proteins functioning in general transcription and RNA processing , processes in which Aire was already known to function in somatic cells such as mTECs ( Abramson et al . , 2010; Mathis and Benoist , 2009 ) , we identified many spindle associated proteins interacting with AIRE in ES cells . Indeed , spindle associated proteins were more enriched among AIRE-interacting proteins than were transcription-associated proteins . We then showed that AIRE localized to mitotic spindles in ES cells and determined that the HSR and SAND domains were responsible for spindle recruitment . The HSR domain has been proposed to be important for the self-dimerization of AIRE and for the localization of human AIRE to filamentous/microtubule cytoskeletal structures in somatic cell lines during interphase ( Rinderle et al . , 1999; Pitkänen et al . , 2001; Halonen et al . , 2004 ) . However , mitotic spindle localization was not investigated in previous studies . The SAND domain is generally considered a DNA binding domain; however , in AIRE the critical amino acid residues for DNA binding are not conserved ( Purohit et al . , 2005; Bottomley et al . , 2001; Perniola and Musco , 2014 ) . Rather , our data suggest that the SAND domain plays a specific role in spindle localization of AIRE in ES cells . Using conditional knockout strategies , we showed that Aire-/- ES cells and m-z- preimplantation embryo cells formed spindles with multiple centrosomes in mitosis , suggesting a critical function of Aire in centrosome sustenance and spindle assembly . To our knowledge this is a novel function for Aire and one that is likely to be ES cell/embryo specific since Aire is only expressed in a few restricted somatic cell types in adults and those cell types , such as mTECs , are generally not proliferative ( Gray et al . , 2007; Hubert et al . , 2008; Gardner et al . , 2008; Yamano et al . , 2015 ) . However , it is difficult at this point to exclude the possibility of Aire playing similar mitotic functions in rare populations of somatic stem cell types since the full range of expression of Aire is largely unknown . We have previously shown that knocking down Aire in ES cells using shRNA caused proliferation and karyotype defects ( Gu et al . , 2010 ) . Here we show that both proliferation and colony formation were impaired in Aire-/- ES cells . Aire m-z- preimplantation embryos also showed proliferation defects but early lineage segregation was not apparently disrupted . The fact that cell proliferation can still continue , albeit in an impaired form in Aire mutant ES cells and embryos , despite the clearly disturbed spindle and centrosome morphologies is not inconsistent with what is known about the roles of centrosomes in early embryonic cell cycles . It is known that cells can proliferate in the presence of multiple centrosomes by centrosome congregation ( Ring et al . , 1982; Quintyne et al . , 2005 ) , and in the absence of centrosomes , so long as a p53-mediated G1 checkpoint is somehow compromised ( Wong et al . , 2015; Lambrus et al . , 2015 ) . That checkpoint is likely compromised in ES cells and only begins to function at midgestation stages in mouse embryos ( Suvorova et al . , 2016; Aladjem et al . , 1998; Bazzi and Anderson , 2014 ) . In the preimplantation embryo , early mitotic divisions involve meiotic-like acentrosomal spindle assembly , while regular somatic-like mitosis where spindle poles are organized around centrosomes is only established around the blastocyst stage ( Courtois et al . , 2012 ) . Both germ cell-specific Aire knockout males and females were fully fertile , suggesting that at least in the genetic background investigated ( CD1 ) the depletion of Aire in germ cells was not sufficient to induce a consistent infertile phenotype , as had been reported before in Aire-/- conventional knockout mice ( Anderson et al . , 2002; Ramsey et al . , 2002 ) . It is plausible that the infertility defects reported in conventional Aire mutants may actually result from the combination of auto-immune and strain-dependent embryonic defects . Since human APECED patients also show high variability between individuals with regard to fertility ( Finnish-German APECED Consortium et al . , 1997; Perheentupa , 2006 ) , the relationship between autoimmune defects , germline and embryo defects and genetic strain/background factors still remain to be dissected . It is not clear whether the proliferation delay and spindle defects observed in Aire m-z- blastocysts would result in later problems in development , given that both human and mouse embryos can overcome a certain level of mitotic and karyotypic defects ( Bolton et al . , 2016; van Echten-Arends et al . , 2011 ) . We found the last LxxLL motif ( LESLL ) to be critical for the mitotic function of Aire in ES cells . Truncations involving this motif ( AIREΔc70 , AIRE ΔLESLL , AIRE505fs ) all perturbed centrosome numbers and spindle pole integrity when overexpressed in ES cells . Interestingly , the effect seemed to be specific to ES cells since the overexpression of the same AIRE ΔLESLL in NIH3T3 cells did not cause mitotic defects . These data support the idea that Aire plays a specific role in mitotic processes in ES cells relative to somatic cells Interestingly , not many spindle-associated proteins were identified as interaction partners of Aire in previous studies using somatic cell lines overexpressing Aire ( Abramson et al . , 2010; Meloni et al . , 2010; Waterfield et al . , 2014 ) . This could also support an ES cell-specific mitotic function for Aire but could also be related to technical differences between the BioID method used in this study and the co-immunoprecipitation or yeast two-hybrid methods used by previous studies ( Lambert et al . , 2015 ) . The mitosis-disrupting effect of overexpressing AIRE ΔLESLL was more severe than the phenotype following Aire depletion in ES cells . The reasons for this are not entirely clear . However , our BioID data suggest that Aire may act in a multi-protein complex , including SPICE1 ( Archinti et al . , 2010 ) , Human Augmin Complex ( HAUS ) complex proteins ( Haus5 and Haus8 ) and CLASP proteins ( CLASP1 and CLASP2 ) ( Lawo et al . , 2009; Bratman and Chang , 2008 ) , which have all been reported to function in centrosome maturation/duplication and mitotic spindle organization . Removal of AIRE from the complex by gene knockout could result in a partially functional complex , while replacement with a dominant negative form like AIRE ΔLESLL could disrupt the conformation of that complex and cause a more complete loss of function , as have been shown before in other experimental systems ( Veitia , 2007; Papp et al . , 2003 ) . Another possible explanation for the observation of more severe mitotic phenotype in AIRE ΔLESLL overexpressing cells than Aire depleted cells is the existence of redundant pathways . To this end , a particularly interesting gene is Deaf1 , another SAND domain containing transcription regulator structurally related to Aire that conducts similar functions to activate expression of auto-antigen genes in the peripheral lymphoid system ( Yip et al . , 2009 ) . Interestingly , Deaf1 has been shown to be expressed in embryos and ES cells and maternal zygotic depletion of Deaf1 homolog in Drosophila embryos caused early embryo arrest ( Park et al . , 2013; Gu et al . , 2010; Veraksa et al . , 2002 ) . It would be interesting to study the effect of depletion of Deaf1 or combined depletion of Aire and Deaf1 to understand the function of this group of genes in mES cells . It is also important to point out that a previously reported disease causing mutation ( AIRE505fs ) in humans ( Ishii et al . , 2000 ) results in the deletion of the LESLL motif . Mouse AIRE proteins with an analogous truncation produced a mitosis-disrupting phenotype similar to AIRE ΔLESLL in ES cells . This suggests that some patient mutations might have mitosis-disrupting effects in early embryos , which might be related to the reported fertility phenotypes . The mechanism by which Aire sustained the centrosome number stability and structure integrity in mES cells is not clear at this point . However , the acquisition of multiple centrosomes , especially in the situation of AIRE ΔLESLL overexpression , occurred within 12 hr of doxycycline induction , which is less than one cell cycle of mES cells . This suggests excess duplication of centrosomes . CDK2 has been identified as a proximity partner of AIRE in mES cells . Elevated centrosome associated CDK2 activity in a p53 deficient cellular environment has been shown to cause centrosome over-duplication ( Tarapore and Fukasawa , 2002; Adon et al . , 2010 ) . Interestingly , mES cells possess sustained elevated CDK2 activity and have inefficient p53 functions but normally manage to maintain normal centrosome numbers ( Stead et al . , 2002; Aladjem et al . , 1998 ) . We speculate that , by interacting with CDK2 , AIRE somehow limits the centrosome over-duplication that would be activated by the sustained elevated CDK2 activity . Furthermore AIRE also interacts with a group of centrosome associated proteins including SPICE1 ( Archinti et al . , 2010 ) and Human Augmin Complex ( HAUS ) complex proteins ( Haus5 and Haus8 ) ( Lawo et al . , 2009 ) which have been shown to function in maintenance of centrosome number and integrity . Disruption of this interaction could also contribute to the centrosome phenotypes of Aire depletion and AIRE ΔLESLL overexpression in mES cells . However , since AIRE also interacted with proteins in mES cells that function in processes such as transcription regulation and RNA processing , it is still possible that defects in those processes upon Aire depletion or AIRE ΔLESLL overexpression indirectly contributed to the mitotic phenotype . Further study is required to separate out the different functions of Aire in mES cells and embryos . Aside from the implications in mitosis regulation in ES cells and early embryos , this study also raises an interesting speculation concerning the evolution of the immune system . As a relatively newly evolved system , the immune system frequently co-opts proteins from basic processes like cell proliferation and embryonic development . For example , the Toll receptor plays a role in the formation of the dorsal-ventral axis in Drosophila ( Valanne et al . , 2011 ) and the Toll-like receptors not only form the basis of innate immunity , but have also been shown to play a role in proliferation in cell types including ES cells ( Taylor et al . , 2010 ) . Activation-induced cytidine deaminase ( AID ) , a critical enzyme for somatic hypermutation , gene conversion , and class-switch recombination in B cells ( Peled et al . , 2008 ) , has also been shown to function in epigenetic reprogramming of primordial germ cells and induced pluripotent stem cells ( Bhutani et al . , 2010 ) . The possibility of a similar , multi-functional role of Aire is especially intriguing because a central tolerance system coordinated by activating promiscuous gene expression is not required until the evolution of the adaptive immune system and its diversification of immune receptors ( Schatz and Swanson , 2011 ) . Organisms may have adapted molecules like Aire from existing biological processes to establish a tolerance mechanism and cope with the random immune receptor diversification . It would be an interesting question to trace the molecular evolution of Aire and determine its function in embryo development in lower organisms . In conclusion , this study uncovered a novel role of Aire in regulating mitosis in ES cells with implications for the biology of proliferation regulation in ES cells and early embryos . To generate the C-terminal Aire-3XFlag-tag expression construct , the full length coding sequence ( CDS ) of Aire transcription variant 1 ( longest variant ) was amplified from mES cDNA and cloned into the pCMV-3Xflag 14 vector ( ( Sigma , Oakville , Canada ) ) . To generate a doxycycline ( Dox ) -inducible Aire-3XFlag-tag expression construct , Aire-3XFlag-tag cDNA was cloned into the pDonor221 vector ( Invitrogen , Thermo Fisher Scientific , Waltham , MA ) and then a Dox-inducible Piggybac destination vector ( PB-T-RfA , kindly provided by Dr . Andras Nagy ) using a Gateway Cloning Kit ( Li et al . , 2013 ) ( Thermo Fisher Scientific ) . To generate a C-terminal Aire-BirA-HA expression construct , Aire CDS was cloned into the pcDNA3 . 1 MCS-BirA ( R118G ) -HA vector ( a kind gift from Dr . Kyle Roux ) ( Addgene Cambridge , MA ) ( Roux et al . , 2012 ) . To generate a Dox-inducible C-terminal Aire-BirA-HA expression construct , Aire-BirA-HA was cloned into a Dox-inducible Piggybac destination vector ( PB-T-RfA ) using Gateway cloning , as described before . A Dox-inducible mCherry-BirA-HA expression plasmid was constructed similarly by fusing mCherry CDS to BirA-HA . Mutations of Aire were introduced into the pDonor221 Aire-3XFlag-tag vector through PCR mediated mutagenesis using InFusion cloning kit ( Takara Bio , USA ) and then cloned into the PB-T-RfA destination vector . For CRISPR experiments , the Cas9 CDS was PCR amplified from the pX330-U6-Chimeric_BB-CBh-hSpCas9 ( pX330 ) plasmid ( a kind gift from Dr . Feng Zhang ) ( Addgene ) ( Cong et al . , 2013 ) and inserted into the pCS2 +plasmid using BamHI:EcoRI sites . Guide RNAs ( gRNAs ) were designed using the Optimized CRISPR design tool ( http://crispr . mit . edu ) . DNA oligos coding sgRNAs were synthesized ( Sigma ) and inserted into the pX330 plasmid in BbsI site ( Cong et al . , 2013 ) . All the primer sequences are shown in supplementary file 2 . Plasmids will be made available through Addgene . To produce Cas9 mRNA , the pCS2+-Cas9 plasmid was linearized with NotI restriction digestion and used as template to in vitro transcribe mRNAs using mMESSAGE mMACHINE SP6 Transcription Kit ( Thermo Fisher Scientific ) . For the production of sgRNAs , sgRNA coding sequences were PCR-amplified from the pX330 plasmids with primers containing T7 promoters ( supplementary file 2 ) and used as template to produce sgRNA using the MEGAshortscript T7 Transcription Kit ( Invitrogen ) . All RNA products were purified using RNeasy Mini Kit ( QIAGEN , Toronto . ON . Canada ) following the cleanup protocol . Female mice of 5–8 weeks age were injected with 5 IU each pregnant mare serum gonadotropin ( PMSG ) ( Sigma , Oakville , Canada ) and human chorionic gonadotropin ( hCG ) ( Sigma , Oakville , Canada ) , 48 hr apart . The females were then mated with males of 8–10 weeks age . Vaginal plugs were checked the following morning and plugged female were counted as 0 . 5dpc . Superovulated females were acquired from the Transgenic Core of The Centre for Phenogenomics and mated with CD1 males . Zygotes were isolated from the oviducts of 0 . 5dpc females and the cumulus cells were removed by incubating in 50 μg/ml Hyaluronidase ( EMD Millipore ) at room temperature for five mins and kept in KSOM . Pronuclear injections were performed using a Leica microscope and micromanipulators ( Leica Microsystems Inc . , Richmond Hill , Canada ) . The injection pressure was supplied by a FemtoJet ( Eppendorf ) . The injections were performed in a drop of EmbryoMax M2 Medium ( EMD Millipore ) covered with mineral oil ( Zenith Biotech , Guilford , CT ) . All animal work was carried out following Canadian Council on Animal Care Guidelines for Use of Animals in Research and Laboratory Animal Care under protocols approved by The Centre for Phenogenomics Animal Care Committee ( protocol number: 20–0026 hr ) . All mouse lines were generated on a CD1 ( breeding stock from Charles River , Montreal , Canada ) background . Aire-3xFlag-tag ( CD1 ( ICR ) -Airetm2 ( Flag-Aire ) Jrt ) and AireFl/Fl ( CD1 ( ICR ) -Aire<tm2Jrt> ) mice were generated using CRISPR-Cas9 genome editing ( Yang et al . , 2014 ) . Aire-3xFlag tag mice were generated by microinjecting the pronucleus of zygotes with a mixture of 20 ng/ul Cas9 mRNA , 10 ng/ul sgRNA ( targeting the terminal codon region of Aire ) and 10 ng/ul single strand ultramer oligo donor ( ssODN ) ( Integrated DNA Technologies ( IDT ) Coralville , Iowa , USA ) encoding 3XFlag-tag and 80 bp homology arm on both 5’ and 3’ sides . Injected zygotes were cultured overnight in KSOM ( EMD Millipore , Etobicoke ON . CA . ) with 50 uM SCR7 ( Xcess Biosciences , San Diego CA . USA . ) to facilitate homology directed repair ( HDR ) ( Maruyama et al . , 2015 ) and transferred to pseudopregnant females the following day . Founder mice were genotyped ( see supplementary file 2 for primer sequences ) and the PCR-amplified targeting region was Sanger sequenced to validate the insertion . Founder mice were out-crossed to CD1 mice for four generations before used to establish ES cell lines . The AireFl/Fl allele was designed as two loxp sites inserted into intron 7 and intron 9 of Aire , flanking critical exons 8 and 9 . The removal of exon 8 and 9 by Cre-mediated recombination caused a frame shift and premature termination of all known coding variants of Aire . The mouse line was derived by a two-step procedure . The pronucleus of zygotes was microinjected with Cas9 mRNA , sgRNA targeting intron seven and ssODN encoding a Loxp site with 80 bp homology arms on both 5’ and 3’ sides . The embryos were cultured and transferred same as before . Founder mice were genotyped and sequence-validated . Mice carrying one loxp site were bred to homozygosity , and a similar round of CRISPR-Cas9 editing was performed on zygotes from this line to target the second loxp into intron 9 . The AireFl/Fl ( ICR-Airetm2Jrt ) mice were then bred for three generations to CD1 , followed by crossing to R26CreERT mice ( B6;129-Gt ( ROSA ) 26Sortm1 ( cre/ERT ) Nat/J ) to make AireFl/Fl R26CreERT mice ( Badea et al . , 2003 ) . The AireFl/Fl mice were also crossed to Alpl-Cre ( 129-Alpltm1 ( cre ) Nagy/J ) mice to generate primordial germ cell-specific Aire knockout mice ( Lomelí et al . , 2000 ) . The AireFl/Fl mice were also crossed to Zp3-Cre ( C57BL6-Tg ( Zp3-cre ) 93Knw/J ) mice to generate oocyte-specific Aire knockout mice ( de Vries et al . , 2000 ) . The AmES8 cell line was a wildtype mES cell line derived from blastocysts acquired from breeding of heterozygotes of B6 . 129S2-Airetm1 . 1Doi/J mice ( Anderson et al . , 2002 ) . The cell line was derived and maintained in 2i-LIF + Serum conditions ( 1:1 mixing of LIF-Serum Medium and 2i-LIF medium ( Ying et al . , 2008 ) on mitomycin C ( MMC ) ( Sigma ) inactivated E15 . 5 mouse embryonic fibroblast ( MEF ) feeders . The pluripotency of the cell line has been tested by in vitro differentiation and chimera formation . Flag-Aire mES cell lines were derived from blastocysts acquired from homozygote breeding of Aire-3xFlag-tag mice and maintained in 2i-LIF-Serum condition . The Aire+/+;R26CreERT+/+ , AireFl/Fl;R26CreERT+/+ , AireFl/+;R26CreERT ± and AireFl/Fl;R26CreERT ± mES cell lines were derived from blastocysts acquired from breeding AireFl/+;R26CreERT ± mice and maintained in 2i-LIF + Serum conditions . For 4-Hydroxytamoxifen ( TAM ) ( Sigma ) treatment , TAM concentration was first titrated on Aire+/+;R26CreERT+/+ , and AireFl/+;R26CreERT ± mES cells respectively to exclude proliferation toxicity to Cre only cells . Based on the titration , R26CreERT+/+ and R26CreERT ± cells were treated with 5 ng/ml and 20 ng/ml TAM in 2i-LIF + Serum medium for 24 hr respectively and then maintained in 2i-LIF-Serum medium . The treated cells were then passaged and used for experiments . To derive mES cell lines carrying Dox inducible transgenes ( TetON cell lines ) , AmES8 cells were co-transfected with PBASE ( encoding piggyback transposase ) , PBCA-rtTA and respective destination vectors as previously described ( McDonald et al . , 2014 ) and selected for one week with 2 μg/ml Puromycin ( Sigma ) . The transfected cells were then either single cell-sorted into 96 well plates or plated sparsely on 10 cm dishes with feeders for clone picking . Clonal cell lines were acquired and tested for inducible expression by immunofluorescence . For the Dox inducible expression experiments , cells were cultured in 1 μ g/ml Dox ( Sigma ) for various times ( Figure legends ) and then cells were collected for downstream experiments . NIH3T3 cells ( Jainchill et al . , 1969 ) were acquired from ATCC and cultured in DMEM medium ( Thermo Fisher Scientific ) supplemented with 10% FBS ( Thermo Fisher Scientific ) . The TetON cell lines were established similarly to mES cells . All the cell lines were tested for Mycoplasma using the Universal Mycoplasma Detection Kit ( ATCC , Manassas , VA . USA ) and were negative and authentificated using STR method by IDEXX Bioresearch to be mouse origin . Transgenic AmES8 cells with Dox inducible mCherry-BirA-HA or Aire-BirA*-HA were grown on gelatin-coated tissue culture flasks in 2i-LIF-Serum . For BioID , the transgenes were first induced for 24 hr with 1 μg/ml Dox , then biotin was added to the medium to a final concentration of 50 μM for 24 hr . The streptavidin purification and protein digestion with trypsin were performed as previously described ( Lambert et al . , 2015 ) . Samples were collected from three independent batch of experiments and assigned as three biological replicates . A spray tip was formed on fused silica capillary column ( 0 . 75 μm ID , 350 μm OD ) using a laser puller ( program = 4; heat = 280 , FIL = 0 , VEL = 18 , DEL = 200 ) . 10 cm ( ±1 cm ) of C18 reversed-phase material ( Reprosil-Pur 120 C18-AQ , 3 μm ) was packed in the column by pressure bomb ( in MeOH ) . The column was pre-equilibrated in buffer A ( 6 μL ) before being connected in-line to a NanoLCUltra 2D plus HPLC system ( Eksigent ) coupled to a LTQ-Orbitrap Elite ( Thermo Fisher Scientific ) equipped with a nanoelectrospray ion source ( Proxeon Biosystems , Thermo Fisher Scientific ) . The LTQ-Orbitrap Elite instrument under Xcalibur 2 . 0 was operated in the data dependent mode to automatically switch between MS and up to 10 subsequent MS/MS acquisitions . Buffer A was 99 . 9% H2O , 0 . 1% formic acid; buffer B was 99 . 9% acetonitrile , 0 . 1% formic acid . The HPLC gradient program delivered an acetonitrile gradient over 125 min . For the first twenty minutes , the flow rate was 400 μL/min at 2% B . The flow rate was then reduced to 200 μL/min and the fraction of solvent B increased in a linear fashion to 35% until 95 . 5 min . Solvent B was then increased to 80% over 5 min and maintained at that level until 107 min . The mobile phase was then reduced to 2% B until the end of the run ( 125 min ) . The parameters for data dependent acquisition on the mass spectrometer were: one centroid MS ( mass range 400–2000 ) followed by MS/MS on the 10 most abundant ions . General parameters were: activation type = CID , isolation width = 1 m/z , normalized collision energy = 35 , activation Q = 0 . 25 , activation time = 10 msec . For data dependent acquisition , the minimum threshold was 500 , the repeat count = 1 , repeat duration = 30 s , exclusion size list = 500 , exclusion duration = 30 s , exclusion mass width ( by mass ) =low 0 . 03 , high 0 . 03 . RAW mass spectrometry files were converted to mzXML using ProteoWizard ( 3 . 0 . 4468; [Kessner et al . , 2008] ) and analyzed using the iProphet pipeline ( Shteynberg et al . , 2011 ) implemented within ProHits ( Liu et al . , 2016 ) as follows . The database consisted of the mouse RefSeq protein database ( version 53 ) supplemented with ‘common contaminants’ from the Max Planck Institute ( http://141 . 61 . 102 . 106:8080/share . cgi ? ssid = 0f2 gfuB ) and the Global Proteome Machine ( GPM; http://www . thegpm . org/crap/index . html ) . The search database consisted of forward and reversed sequences ( labeled ‘DECOY’ ) ; in total 58202 entries were searched . The search engines used were Mascot ( 2 . 3 . 02; Matrix Science ) and Comet ( 2012 . 01 rev . 3; [Eng et al . , 2015] ) , with trypsin specificity ( two missed cleavages were allowed ) and deamidation ( NQ ) and oxidation ( M ) as variable modifications . Charges + 2 , +3 and+4 were allowed and the parent mass tolerance was set at 15 ppm while the fragment bin tolerance was set at 0 . 6 amu . The resulting Comet and Mascot search results were individually processed by PeptideProphet ( Keller et al . , 2002 ) , and peptides were assembled into proteins using parsimony rules first described in ProteinProphet ( Nesvizhskii et al . , 2003 ) into a final iProphet protein output using the Trans-Proteomic Pipeline ( TPP; Linux version , v0 . 0 Development trunk rev 0 , Build 201303061711 ) . TPP options were as follows . General options are -p0 . 05 -x20 -PPM -d"DECOY’ , iProphet options are –ipPRIME and PeptideProphet options are –OpdP . All proteins with a minimal iProphet protein probability of 0 . 05 were parsed to the relational module of ProHits . Note that for analysis with SAINT , only proteins with iProphet protein probability ≥0 . 95 are considered . This corresponds to an estimated protein-level FDR of ~0 . 5% . SAINTexpress analysis ( Teo et al . , 2014 ) was performed on the iProphet results ( filtered at >0 . 95 protein probability ) as follows . SAINTexpress ( v 3 . 3 ) was used to estimate the probability of proximity interaction in the individual biological triplicates for Aire and mCherry control . Averaged probabilities ( AvgP ) were used to estimate the Bayesian false discovery rates . Hits with ≤1% FDR were deemed ‘high confidence’ . Downloadable files and all raw mass spectrometry files are deposited in the MassIVE repository housed at the Center for Computational Mass Spectrometry at UCSD ( http://proteomics . ucsd . edu/ProteoSAFe/datasets . jsp ) . The datasets has been assigned the MassIVE IDs MSV000080386 ( ftp://MSV000080386@massive . ucsd . edu ) . The dataset was assigned the ProteomeXchange Consortium ( http://proteomecentral . proteomexchange . org ) identifiers PXD005529 . The dataset is currently password protected until publication; the password is Aire . To further improve the specificity of the interactors detected by BioID following a SAINT analysis , we compared the results obtained here with our previous effort ( Lambert et al . , 2015 ) . We found that Pccb , Pcca , , Mccc2 , Acaca , Acacb and Mccc1 were detected at high level ( more than 20 spectral count in nine distinct control purifications ) and thus removed them from our final interactors list . Gene ontology analysis was performed with DAVID server with default parameters ( Huang et al . , 2009b , 2009a ) . Data were presented as p values for Biological Process ( BP_DIRECT ) and Cellular Compartment ( CC_DIRECT ) Western Blotting: Western blotting was performed as previously described . ( Gu et al . , 2010 ) . Samples were normalized by cell numbers . mES cells were seeded at a density of 105 cells/well on gelatin coated six well tissue culture plates ( Nunc ) in growth medium and cultured for 48 hr ( Passage 1 ( P1 ) ) . Cell numbers were counted at this point with a hemocytometer and recorded as P1 cell number . The P1 cells were subcultured at 105 cells/well on gelatin coated six well tissue culture plates ( Nunc ) in growth medium again and cultured for a further 48 hr ( Passage 2 ( P2 ) ) . Cell numbers were counted at this point with a hemocytometer and recorded as P2 cell number . Colony formation assays were performed as previously described ( Gu et al . , 2010 ) . Briefly , mES cells were seeded at a density of 500 cell/well in 24 well tissue culture plates on MMC inactivated E15 . 5 MEFs ( Nunc ) in growth medium . ALP staining was performed after 72 hr with the Leukocyte Alkaline Phosphatase ( ALP ) Kit ( Sigma ) and ALP positive colonies were counted manually under a dissection microscope . Cells for immunofluorescence were grown on gelatin-coated circular coverslips ( Thickness 1 . 5 ) ( Thermo Fisher Scientific ) . For all stainings , with the exception of staining for centriole acetyl-tubulin ( see below ) , cells were fixed with 4% paraformaldehyde ( PFA ) for 20 min at room temperature . They were then permeabilized with Dulbecco's phosphate-buffered saline ( DPBS ) ( Thermo Fisher Scientific ) containing 0 . 5% TritonX-100 for 15 min and then blocked in DPBS containing 0 . 1% TritonX-100% and 5% FBS ( Blocking Buffer ( BB ) for 1 hr at room temperature . They were then incubated with primary antibodies ( supplementary file 1 ) diluted in BB overnight at 4°C . After thorough washing , the cells were incubated with fluorophore-conjugated secondary antibodies ( supplementary file 1 ) for 45 min at room temperature . They were then washed and mounted on glass slides with the VECTASHIELD HardSet Antifade Mounting Medium with DAPI ( Vector ) and kept in 4°C before imaging . For cold treatment for analyzing acetyl-tubulins , cells were removed from the incubator and left on ice for 60 min before fixed with pre-chilled 100% methanol for 10 min at −20°C . The cells were then blocked with blocking buffer and stained and mounted similarly . The samples were then analyzed by either a spinning disk confocal microscopy ( Quorum ) or a Structured Illumination Microscope ( Zeiss ) . PLA assay is a immuno-PCR based assay to visualize proximity relationship of a pair of proteins in situ . ( Söderberg et al . , 2006 ) PLA assay was performed using Duolink In Situ Orange Starter Kit Mouse/Rabbit ( Sigma ) according to the product manual . All the PLAs were performed in TetON-3XFlag-Aire mES cells after 12 hr induction with 1μ g/ml Dox . Uninduced ( Dox- ) cells were used as negative control and the PLA between M2 anti-Flag mouse monoclonal antibody ( Sigma ) and corresponding rabbit antibodies against AIRE’s interacting partners were detected with PLA kit and visualized using a spinning-disc confocal microscope ( Quorum ) . Images were acquired using a Zeiss Axiovert 200 inverted microscope equipped with a Quorum spinning disk confocal scan head , a Hamamatsu C9100-13 EM-CCD camera , and Volocity ( version 6 . 3 . 1 ) aquisition software . Spindle images of both cultured cells and early embryos were acquired as Z-stacks ( at 0 . 25 μm intervals ) with a 63x oil ( NA = 1 . 35 ) objective . PLA images and images of pluripotency markers in mES cells were acquired as Z-stacks ( at 0 . 5 μm intervals ) with an 20x air ( NA = 0 . 75 ) objective . Images of lineage markers in early embryos were were acquired as Z-stacks ( at 2 μm intervals ) with an 20x air ( NA = 0 . 75 ) objective . Images were visualized and analysed using Volocity ( version 6 . 3 . 1 ) . Images were acquired using the structured illumination module of the Zeiss Elyra PS1 . A 63x/1 . 4 objective , in combination with a 1 . 6x optovar , was used for data acquisition . Images spanning the centrosomes or spindles were taken as Z-stacks with 0 . 110 μm optical sections . The final SIM images were reconstructed and aligned using the automatic processing toolbox of Zeiss Zen with 3D-SIM mode . The SIM Images were then visualized by Volocity 6 . 3 software and presented as extended focus views . For quantifications of mitotic γ-tubulin foci in confocal images , metaphase cells were identified by chromosome morphology under DAPI channel and then imaged for α-tubulin and γ-tubulin . The Images were then visualized with Volocity 6 . 3 software as extended focus views and γ-tubulin foci were manually counted . For quantification of S/G2 centrosomes in SIM Images , S/G2 centrosomes were identified as clusters of 2 or multiple γ-tubulin positive foci under wide field visualization and Z-Stack SIM Images acquired and reconstructed by 3D-SIM . The images were then visualized with Volocity 6 . 3 software as extended focus views and centrosome numbers were manually counted in the images . The diameter of the γ-tubulin rings was measured and calculated as the average of length of the long and short axis of the rings in SIM images measured with Volocity . For quantification of mitotic spindle poles , mitotic metaphase cells were identified by chromosome morphology under DAPI channel and Z-Stack SIM images were then acquired and reconstructed by 3D-SIM . The diameter of TUBG foci was measured and calculated as the average of length of the long and short axis of the γ-tubulin positive regions . Intracellular staining flow cytometry analyses were performed according to protocol from Cell Signaling Technology ( https://www . cellsignal . com/common/content/content . jsp ? id=flow ) . Briefly , cells were trypsinized to single cells and then fixed with 4% PFA for 10 min at 37°C . They were then permeabilized with pre-chilled ( −20°C ) 90% methanol for 1 hr on ice . After three washes in flow buffer ( DPBS containing 1% BSA ) , the cells were incubated in Pacific blue conjugated pS10-H3 antibodies ( Cell Signaling Technology ) diluted in flow buffer in the dark for 1 hr at room temperature . After three washes in flow buffer ( DPBS containing 1% BSA ) , the cells were incubated in Propidium Iodide ( PI ) /RNase Staining Solution ( Cell Signaling Technology ) in the dark for 30 min at 37°C and then stored at 4°C before analysis ( cells could be stored for up to 1 week ) . The cells were then analyzed on a BD LSRII-CFI BGRV and the data were analyzed by Flowjo software ( FlowJo , LLC ) . Total RNA was isolated with Trizol ( Invitrogen ) and cDNAs were synthesized with QuantiTect Reverse Transcription Kit ( Qiagen ) . Realtime PCR was performed using LightCycler 480 SYBR Green I Master Mix ( Roch ) with primers listed in ( Supplementary file 2 ) The expression levels were calculated with ΔΔCT methods . For the fertility test of paternal Aire knockout mice ( patΔ ) , 2 pairs of Aire+/-;Alpl-Cre + x CD1 breeding pairs were set up as control and 2 pairs of Aire+/-;Alpl-Cre+XCD1 breeding pairs were set up as AirepatΔ . For the fertility test of maternal Aire knockout mice ( matΔ ) , 3 pairs of CD1XAireFl/Fl;Zp3-Cre- breeding pairs were set up as control while 3 pairs of CD1XAireFl/Fl;Zp3-Cre+ breeding pairs were set up as AirematΔ . All the breedings were started at 8 weeks age and monitored for 3 months . The litter number and litter size were recorded . The pups were monitored till weaning ( 21 days ) and no obvious defects were observed . For the analysis of m-z- embryos , control embryos were generated by breeding AireFl/+;Alpl-Cre- males and superovulated AireFl/Fl;Zp3-Cre- females and m-z- embryos were generated by breeding AireFl/-;Alpl-Cre + males and superovulated AireFl/Fl;Zp3-Cre + females . Embryos were flushed from uterus at E3 . 5 and fixed in 4% PFA and imaged under a stereoscope for blastocyst rate analysis . Embryos derived from each female were considered as one experimental unit for blastocyst rate analysis . The embryos were then permeabilized with Dulbecco's phosphate-buffered saline ( DPBS ) containing 0 . 5% TritonX-100 for 15 min and blocked in DPBS containing 0 . 1% TritonX-100% and 5% FBS ( Blocking Buffer ( BB ) for 1 hr at room temperature . After removing the zona pellucida with EmbryoMax Acidic Tyrode's Solution ( Millipore ) , the embryos were incubated with primary antibodies overnight at 4°C as indicated in Figures and Figure legends . The next day , embryos were washed in BB three times and stained with fluorophore-conjugated secondary antibodies ( Supplementary file 1 ) for 45 min at room temperature . The embryos were then mounted in VECTASHIELD HardSet Antifade Mounting Medium with DAPI ( Vector ) and imaged under a spinning disk confocal microscope ( Quorum ) and analyzesd as stated in Confocal Analysis section . No statistical methods were used to pre-determine sample size . For cellular expriments , 2–3 biological replicates were performed . For 4-OH-tamoxyfen or Doxycycline treatment experiment , independent sets of control and treated cells were considered as biological replicates . For mouse fertility test , 2–3 breeding pairs of corresponding genotypes were analyzed . For blastocyst rate analysis , embryos isolated from 10 to 20 females from each groups were analyzed . For blastocyst cell number analysis , embryos collected from at least three females were analyzed in each group . For spindle analysis in blastocysts , 20–30 mitotic cells from at least 10 blastocysts of each genotypes were analyzed . The experiments were not blinded from investigators: the identities of the samples were known throughout the experiment . All statistics were performed with Prism software ( GraphPad Software , Inc . La Jolla , CA . USA ) . For cell growth , colony formation and qPCR , Wilson’s t-test was performed . For other data , data were first tested with D'Agostino and Pearson omnibus normality test , then Wilson’s t-test or Mann–Whitney–Wilcoxon test were used for data that passed or did not pass the normality test respectively .	Before the cells in our body separate to create copies of themselves , they need to duplicate their genetic material . To do so , they construct a machine called the spindle apparatus to divide their DNA evenly . In the embryo of mammals , embryonic stem cells – cells that can create all the cell types in adults – divide swiftly . This makes them prone to make mistakes that could affect the health of the organism they will develop into . To limit the number of errors , the embryonic stem cells divide more stringently and have more effective machineries to create a spindle apparatus . Together with a team of researchers , Gu previously showed that a gene and its protein called the ‘autoimmune regulator’ , or AIRE for short , are highly active in embryonic stem cells and embryos . The autoimmune regulator usually plays an important role in helping immune cells to distinguish the body's own proteins from those of foreign invaders . It was also shown that some people suffering from fertility problems carry mutations in the AIRE gene . However , until now , it was not known if AIRE also had a specific role in embryonic stem cells . Using mouse embryonic stem cells and early embryos that were modified to either completely lack the AIRE gene or to produce a defective AIRE protein , Gu et al . now discovered a new purpose for AIRE . The AIRE protein interacted with a group of proteins that are associated with the spindle apparatus and was needed so that the spindle could form properly . When cells lacked the AIRE gene , a faulty spindle apparatus was assembled . Moreover , when a defective AIRE protein was produced , the structure of spindle apparatus in embryonic stem cells was also disrupted . A next step will be to further investigate how AIRE mutations could affect stem cell maintenance and fertility , which could lead to better ways to detect fertility defects in the future .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "stem", "cells", "and", "regenerative", "medicine" ]	2017	AIRE is a critical spindle-associated protein in embryonic stem cells
Ebola is a deadly virus that causes frequent disease outbreaks in the human population . In this study , we analyse its rate of new introductions , case fatality ratio , and potential to spread from person to person . The analysis is performed for all completed outbreaks and for a scenario where these are augmented by a more severe outbreak of several thousand cases . The results show a fast rate of new outbreaks , a high case fatality ratio , and an effective reproductive ratio of just less than 1 . Ebola virus disease is an often fatal disease of humans that is not vaccine-preventable and has no specific treatment . A total of 25 outbreaks , believed to have arisen due to zoonotic transmission from wild mammals , have occurred since the first observed cases in humans in 1976 ( World Health Organisation , 2014a ) . The current epidemic is the largest to date ( World Health Organisation , 2014b ) . This gives particular urgency to quantitative estimation of epidemiological quantities relevant to Ebola , such as case fatality ratio , timing of new outbreaks , and the strength of human-to-human transmission . The most important epidemiological quantity to estimate for an infectious disease is typically the basic reproductive ratio , R0 , defined as the expected number of secondary cases produced per primary case early in the epidemic ( Diekmann et al . , 1990 ) . When R0 is greater than 1 , the expectation is that a new epidemic will eventually infect a significant percentage of the population if it is not stopped by interventions or chance extinction; conversely , when R0 is less than 1 , chance events may lead to a large number of cases , but these are always expected to be much less numerous than the total population size . Previous attempts to estimate R0 for Ebola have found values between 1 . 34 and 3 . 65 by fitting compartmental epidemic models to the incidence over time of the large outbreaks in the Democratic Republic of Congo in 1995 and Uganda in 2000 ( Chowell et al . , 2004; Ferrari et al . , 2005; Legrand et al . , 2007 ) , with similar results obtained for the ongoing outbreak ( Althaus , 2014 ) . This leads to the question of why all completed outbreaks numbered at most several hundreds , with the typical answer being that the medical and social response to an outbreak reduces transmission , leading to an effective reproductive ratio Rt<R0 ( Chowell et al . , 2004; Legrand et al . , 2007 ) , although it is also important to note that heterogeneity in transmission can lead to extremely high probabilities of an outbreak becoming extinct even if Rt is slightly greater than 1 ( Lloyd-Smith et al . , 2005 ) . Figure 1 shows the results of fitting to times between outbreaks , with Figure 1A showing the empirical distribution of times between outbreaks together with the fitted model distribution that has mean 1 . 49[1 . 02 , 2 . 24] years between outbreaks and Figure 1C showing the posterior for the rate parameter . Figure 1 also shows the results of fitting CFR to number of deaths and final size , with Figure 1B showing empirical CFRs for different outbreaks together with the fitted model distribution . Other plots in Figure 1D , E show the posteriors for the beta distribution parameters . 10 . 7554/eLife . 03908 . 003Figure 1 . Analysis of rate of new outbreaks and case fatality ratio . A shows empirical data and 95% CI ( black lines ) together with fitted distribution and 95% CI ( red lines ) for rate of new outbreaks . B shows empirical data and 95% CI ( black lines ) together with fitted distribution and 95% CI ( red lines ) for case fatality ratio . C shows the posterior density for rate of new outbreaks λ , while D and E show the posterior density for the beta distribution parameters α and β , respectively . DOI: http://dx . doi . org/10 . 7554/eLife . 03908 . 003 Figure 2 shows the results of fitting to completed outbreaks , with Figure 2A , B giving the fitted distribution against data , Figure 2C showing the posterior for the reproductive ratio , which is estimated to be Rt=0 . 88[0 . 64 , 0 . 96] . Figure 2D shows the posterior for the geometric parameter , which is estimated to be p=0 . 089[0 . 029 , 0 . 19] . 10 . 7554/eLife . 03908 . 004Figure 2 . Analysis of transmission dynamics for completed outbreaks . ( A and B ) Model ( solid red line ) and 95% CI ( dash-dot red line ) vs data ( black circles ) and 95% CI ( solid black lines ) for different axis scales . ( C ) Posterior for values of the reproductive ratio Rt . ( D ) Posterior for the geometric parameter p . DOI: http://dx . doi . org/10 . 7554/eLife . 03908 . 004 While the model is designed not to depend explicitly on the temporal dynamics of Ebola virus disease , Figure 3A shows a set of 24 outbreaks simulated from a continuous-time Markov chain with the same probability distribution for final size as the estimated model . These show behaviour that is typical of near-critical branching processes , which often becoming extinct early but also often grows to significant size before extinction . Figure 3B plots the likelihood surface for these simulated data showing parameter identifiability . 10 . 7554/eLife . 03908 . 005Figure 3 . Simulation study . ( A ) Real-time model simulations , with change in colour denoting a new outbreak . ( B ) Likelihood contours ( black lines and values multiplied by an unimportant constant ) together with parameters used to simulate ( red cross ) , showing that the parameters are identifiable from such data . DOI: http://dx . doi . org/10 . 7554/eLife . 03908 . 005 Figure 4 shows the results of fitting to completed outbreak final sizes augmented by an outbreak of uncertain size in the range 1000–5000 . In this study , Figure 4A gives the fitted distribution against data , and Figure 4B shows the posterior for the probability of the additional outbreak , which is estimated to be 0 . 023[0 . 0015 , 0 . 088] . Figure 4C shows the posterior for the reproductive ratio , which is estimated to be Rt=0 . 94[0 . 87 , 0 . 99] , and Figure 4D shows the posterior for the geometric parameter , which is estimated to be p=0 . 11[0 . 054 , 0 . 21] . 10 . 7554/eLife . 03908 . 006Figure 4 . Analysis of transmission dynamics for completed outbreaks plus one outbreak of size 1000–5000 . ( A ) Model ( solid red line ) and 95% CI ( dash-dot red line ) vs data ( black circles ) and 95% CI ( solid black lines ) . ( B ) Posterior for the probability of the large uncertain outbreak . ( C ) Posterior for values of the reproductive ratio Rt . ( D ) Posterior for the geometric parameter p . DOI: http://dx . doi . org/10 . 7554/eLife . 03908 . 006 The results obtained point to the following conclusions about Ebola transmission dynamics . ( i ) The rate of new epidemics and CFR are both high , but with significant variability from outbreak to outbreak . ( ii ) The effective reproductive ratio Rt for person-to-person transmission is just below 1 . ( iii ) There is extremely large variability in the final size of outbreaks . It is also important to consider the sensitivity of these conclusions . A larger final size for the current outbreak ( but still significantly less than the population size of a country ) as suggested by the analysis above will tend to lead to a narrower posterior about a value of Rt closer to 1; this can be understood from general properties of branching processes ( Athreya and Ney , 1992 ) . Such a finely tuned constant value of Rt would , however , become increasingly difficult to interpret as a fundamental property of the outbreak and a modelling approach in which Rt was allowed to vary in time—along with the public health and behavioural responses—would be preferred . Also , it is possible that a number of small outbreaks were not recorded by the WHO . This could be addressed through incorporation of additional variability into the model through introduction of explicit overdispersal parameters as in the study by Lloyd-Smith et al . ( Lloyd-Smith et al . , 2005 ) and Blumberg and Lloyd-Smith ( Blumberg and Lloyd-Smith , 2013 ) , although for the data currently available there was no strong evidence for overdispersal beyond that implied by the geometric distributions . All of these conclusions suggest no reason for complacency and give support to appeals for greater resources to respond to the ongoing epidemic ( Médecins Sans Frontières , 2014 ) . In this study , a different approach is taken based on using the time between outbreaks , number of deaths , and final number of cases , for all 24 completed Ebola outbreaks reported by the World Health Organisation ( World Health Organisation , 2014a ) . Full mathematical details of the approach are given below . First , we model the start of new outbreaks as a ‘memoryless’ Poisson process with a rate λ . Secondly , we assume that each new outbreak has a case fatality ratio ( CFR—the probability that a case will die ) picked from a beta distribution . Thirdly , the final size model involves two components: ( i ) a geometrically distributed number of cases , A , which includes cases arising from animal-to-human and pre-control transmission; ( ii ) a branching process model of human-to-human transmission ( Athreya and Ney , 1992; Ball and Donnelly , 1995 ) , whose offspring distribution has mean Rt , generating Z cases . The final size is then K=A+Z\|A . This quantity should be interpreted as arising from a combination of Rt , R0 , and timing of interventions . Bayesian MCMC with uninformative priors was used to fit all models ( Gilks et al . , 1995 ) . Since doubts have been raised in the literature about the use of final size data for emerging diseases ( Drake , 2005 ) , a simulation study was also performed to test identifiability , although a recent study by Blumberg and Lloyd-Smith ( Blumberg and Lloyd-Smith , 2013 ) of joint identifiability of two parameters in a related model is also highly relevant in this context . Finally , the final size data were augmented by an outbreak of unknown size in the range 1000–5000 ( with mathematical details given by Equation . ( 5 ) , below ) and the model was refitted . Due to the significant uncertainty in the severity of the current outbreak , this is not intended to be a real-time analysis , but rather to show how the modelling approach responds to such a scenario in general . We model the start of new outbreaks in the human population as a Poisson process of rate λ . If the time period over which N outbreaks is observed is T years , then the likelihood is ( 6 ) L ( D\|λ ) = ( λT ) Ne−λTN ! . We estimate λ=0 . 67[0 . 45 , 0 . 98] , with posterior distribution given in Figure 1C . The probability density function for t being the next outbreak time is ( 7 ) f ( t ) =λe−λt , which is shown in Figure 1A . We let Ci be a random variable for the probability of fatality for a given case in outbreak i . We assume a parametric model in which this is drawn from a beta distribution , meaning that the probability density function is ( 8 ) Beta ( c\|α , β ) =cα−1 ( 1−c ) β−1B ( α , β ) , B ( α , β ) :=∫01xα−1 ( 1−x ) β−1dx . Then if di≤ki is the number of fatalities in outbreak i , treating each fatality as independent , conditioned on infection , gives ( 9 ) Pr[di\|ki , α , β]= ( kidi ) B ( α+di , β+ki−di ) B ( α , β ) . Then the likelihood is ( 10 ) L ( D\|α , β ) =∏iPr[di\|ki , α , β] . We estimate α=6 . 1[2 . 8 , 11] and β=3 . 1[1 . 5 , 5 . 9] , with posterior distributions given in Figure 1D , E . The MCMC methodology used was Random-walk Metropolis–Hastings with thinning to produce 103 uncorrelated samples , with each posterior ultimately derived from one long chain . The parameter spaces involved are low-dimensional enough that large-scale sweeps can be performed to check for multimodality , which was not observed , and convergence of the chains was observed to be fast and independent of initial conditions . For the simulation study , the real-time incidence curves are produced by modelling the geometric distributions as arising from Poissonian transmission with exponentially distributed rates . The times between new introductions are not explicitly modelled or shown . MATLAB code to reproduce the analysis of this paper is available at: https://github . com/thomasallanhouse/elife-ebola-code .	The West Africa outbreak of Ebola virus disease is larger than any of the previous outbreaks over the last four decades . Most human outbreaks likely begin when a person is infected after contact with an infected wild animal—but during an outbreak the virus can spread from person-to-person via contact with blood or other bodily fluids . There is no vaccine against Ebola nor is there a specific treatment . The percentage of infected people who have been killed by the Ebola virus in the past outbreaks varies from 50% to 90% . However , predicting how an outbreak will progress once it has started remains difficult . For any infectious disease , it is important to estimate how many new people , on average , each person with the disease will go on to infect . When this value—called the ‘basic reproductive ratio’ ( or R0 ) —is greater than 1 , a significant percentage of the population is expected to eventually become infected if medical interventions are not introduced . Conversely , when R0 is less than 1 , chance events may lead to a large number of cases , but only a fraction of the total population will be affected . Previous estimates of the basic reproductive ratio for Ebola gave values greater than 1 , making it unclear why all the completed outbreaks of Ebola had infected at most several hundred people and had not caused global pandemics . Medical intervention and control measures were generally considered the most likely answer . However , it is important to note that these previous predictions were made using data from only two large outbreaks of Ebola in 1995 and 2000 . Now , House has used a different modelling approach to estimate Ebola's reproductive ratio . The new model is based on data from for all 24 completed Ebola outbreaks and includes the time between outbreaks , the number of deaths , and the final number of cases . The model also included ‘data’ from a hypothetical scenario of a more severe outbreak with several thousand cases . House revealed that new outbreaks tend to occur frequently and that often a large percentage of those infected with Ebola will die of the disease , although the exact values vary between different outbreaks . Furthermore , if there is no fundamental change compared to the past , the analysis predicts that the ‘effective reproductive ratio’ for person-to-person spread of Ebola ( which takes into account the effect of medical intervention ) is just less than 1 . It also predicts that the final number of cases can be very different for different outbreaks . House concluded that at first the current West African outbreak was unusual but still consistent with the pattern of previous outbreaks . However , as the number of people infected continued to grow , it makes this less plausible . It is now more likely that there is some fundamental difference , for example in the infectiousness of the Ebola strain , in the current outbreak compared to all the previous outbreaks; although further work would be needed to confirm this .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "short", "report", "epidemiology", "and", "global", "health", "microbiology", "and", "infectious", "disease" ]	2014	Epidemiological dynamics of Ebola outbreaks
Motoneurons developmentally acquire appropriate cellular architectures that ensure connections with postsynaptic muscles and presynaptic neurons . In Drosophila , leg motoneurons are organized as a myotopic map , where their dendritic domains represent the muscle field . Here , we investigate mechanisms underlying development of aspects of this myotopic map , required for walking . A behavioral screen identified roles for Semaphorins ( Sema ) and Plexins ( Plex ) in walking behavior . Deciphering this phenotype , we show that PlexA/Sema1a mediates motoneuron axon branching in ways that differ in the proximal femur and distal tibia , based on motoneuronal birth order . Importantly , we show a novel role for glia in positioning dendrites of specific motoneurons; PlexB/Sema2a is required for dendritic positioning of late-born motoneurons but not early-born motoneurons . These findings indicate that communication within motoneurons and between glia and motoneurons , mediated by the combined action of different Plexin/Semaphorin signaling systems , are required for the formation of a functional myotopic map . Motoneurons are key elements in the neural networks that generate behavior in all nervous systems . They represent the final common path for convergence of processed information from motor control circuitry in the central nervous system and from sensory feedback circuitry in the periphery ( Sherrington , 1906; Pearson , 1993 ) . Moreover , as final common output channels , they represent exclusive control elements of muscle effectors that mediate behavioral action . For their correct function , motoneurons must acquire specific cellular architectures during development such that their input domains , usually manifest as dendrites , receive connections from appropriate pre-motor neurons , and their axonal output domains make connections to correct target muscle cells ( Jessell et al . , 2011; Harris et al . , 2015; Arber , 2012 ) . How motoneuron development is orchestrated to establish the appropriate and specific dendritic input and axonal output connectivity needed for behavioral action is a central and still important question in neural development ( Tessier-Lavigne and Goodman , 1996; Jefferis , 2006; Dasen and Jessell , 2009 ) . Considerable insight into the mechanisms that control motoneuron structure has been obtained in the neurogenetic model system of Drosophila . In this system , the developmental processes involved in generating appropriate dendritic and axonal morphology of different types of larval ( Landgraf and Thor , 2006; Mauss et al . , 2009 ) and adult motoneurons have been studied extensively . For example , the specific architecture of the motoneurons that innervate the adult leg has been shown to depend on their lineage and birth order , in that the majority of the leg motoneurons are postembryonic lineal descendants of neuroblast 15 , and these motoneurons manifest a birth order-specific neuroanatomical organization ( Truman et al . , 2004; Brierley et al . , 2009; Brierley et al . , 2012; Baek and Mann , 2009 ) . In this lineage , early born motoneurons project their axons to proximal muscles in the leg segments and have dendritic arborization that extends toward the thoracic ganglion neuropile midline , whereas late born motoneurons innervate distal muscles in the leg segments and have dendritic arbors that are restricted to the lateral regions of the ganglionic neuropile . Thus , in terms of their central and peripheral nervous architecture , these leg motoneurons form a myotopic map ( Brierley et al . , 2009 ) . Although this myotopic map-specific targeting of leg motoneuron dendrites has been shown to require midline signaling through the Slit/Roundabout and Netrin/Frazzled signaling systems ( Brierley et al . , 2009 ) , the mechanism underlying compartmentalization in terms of birth order and distinct axonal and dendritic targeting is poorly understood . In order to understand the guidance cues responsible for leg motoneuron development , we performed a behavioral screen for locomotor defects caused by targeting RNAi in motoneurons and identified Plexin and Semaphorin as candidates . In this study , we have investigated the development of specific axonal projections and dendritic arbors in leg motoneurons required for walking in adult Drosophila mediated by Plexin/Semaphorin signaling system and , in doing so , discover novel cellular and molecular roles . The Semaphorins ( Sema ) are a large family of transmembrane and secreted glycoprotein ligands that together with their Plexin ( Plex ) receptors are known to be involved in control of cell migration , dendritic topography and axon guidance in vertebrates and invertebrates ( Kolodkin et al . , 1993; Yazdani and Terman , 2006; Pasterkamp , 2012 ) . Semaphorins and Plexins have evolutionarily conserved guidance function during nervous system development , and both transmembrane and secreted Semaphorin ligands can mediate a diverse set of repulsive and attractive guidance functions . In Drosophila , there are two Plexin receptors and five Semaphorin ligands . PlexinA ( PlexA ) strongly binds the Sema1a and Sema1b ligands while as Plexin B ( PlexB ) binds to Sema2a and Sema2b ( Bates and Whitington , 2007; Ayoob et al . , 2006; Lattemann et al . , 2007; Sweeney et al . , 2011 ) ( Sema5c is not expressed in the CNS ) . Both PlexA and PlexB have been shown to mediate a Sema-dependent repulsion of motor axons during embryonic development of body wall innervation ( Winberg et al . , 1998; Yu et al . , 1998; Jeong et al . , 2012 ) . Whether or not Plexin/Semaphorin signaling is also involved in the development of the post-embryonically generated motoneurons that innervate the leg musculature is currently unknown . We first carry out a behavioral screen for locomotor defects caused by targeted RNAi knockdown of Semaphorins and Plexins in leg motoneurons , which shows that knockdown of membrane-bound Plexins and Semaphorins results in abnormal walking gait . Based on this , we screen for corresponding neuroanatomical defects in the affected leg motoneurons . We find that motoneuron-specific knockdown of Sema1a or PlexA causes defective axonal defasciculation and targeting in motoneurons that innervate leg muscles and these phenotypes differ in proximal femur and distal tibia . In the femur , there occurs a reduction in axon branching and defasciculation of motoneurons , whereas an increase in axon branching of motoneurons occurs in the tibia . This suggests a compartment-specific activity of Plexin/ Semaphorin signaling system . Larval and adult motoneuron dendrites in Drosophila , organized as a myotopic map , have been shown to utilize Robo-Slit and Netrin-Frazzled signaling systems ( Brierley et al . , 2009; Mauss et al . , 2009 ) . The dendritic targeting of larval motoneurons is an active process that is independent of glial differentiation or target muscle formation ( Landgraf et al . , 2003; Landgraf and Thor , 2006 ) . In contrast , we find the role of glia in positioning dendrites of late born motoneurons that innervate distal muscles of tibia and occupy lateral regions of the thoracic neuropil in adult Drosophila . We demonstrate that motoneuron-specific knockdown of PlexB or knockdown of Sema2a secreted from glial cells results in a shift of dendritic arborization toward the ganglionic midline in late born leg motoneurons . This is an important and novel difference , which may have general implications for dendritic patterning mechanisms . These findings indicate that the integrative action of multiple Plexin/Semaphorin signaling systems mediate communication between glia and motoneurons and within motoneurons for the correct formation of axonal projections and dendritic arborization in leg motoneuron development . To determine which elements of the Plexin/Semaphorin signaling system might be required for correct leg locomotor activity , we first carried out a behavioral screen for walking defects caused by targeted RNAi knockdown of Semaphorins and Plexins in leg motoneurons . In these experiments , the OK371-Gal4 driver was used to target UAS-RNAi knockdown to leg motoneurons; this Gal4 driver targets reporter gene expression to all motoneurons during embryonic and postembryonic development and in the adult ( Mahr and Aberle , 2006; Brierley et al . , 2009; Brierley et al . , 2012 ) . In these experiments , UAS-RNAi knockdown constructs for Sema1a and its receptor PlexA as well as for Sema2a and its receptor PlexB were used . To monitor their locomotor activity , flies were allowed to walk freely over a soot plate ( Maqbool et al . , 2006 ) and their resulting footprint patterns were documented and analyzed in knockdown versus control flies . Locomotor defects were observed in RNAi-mediated knockdown of Sema1a , PlexA , and PlexB in leg motoneurons ( Figure 1 ) . In Sema1a knockdown experiments , flies showed uncoordinated walking activity with leg dragging . In PlexA knockdown experiments , flies showed uncoordinated walking , short steps or short steps with leg dragging . In PlexB knockdown experiments , flies walked with short steps and leg dragging . In Sema2a knockdown experiments , flies showed normal walking behavior . The frequency of occurrence of phenotypes is shown in Figure 1—source data 1 . Taken together , these results suggest that PlexA , PlexB , and Sema1a , but not Sema2a , are required in leg motoneurons during development and/or in the adult for the manifestation of correct locomotor activity . 10 . 7554/eLife . 11572 . 003Figure 1 . Targeted knockdown of Plexin/Semaphorin signaling in motoneurons results in walking defects . ( A ) Schematic of interactions between Sema1a , Sema1b , and PlexA as well as between Sema2a , Sema2b , and PlexB . ( B–I ) Walking patterns monitored by footprint tracking in control and Plexin/Semaphorin RNAi knockdowns targeted to motoneurons . ( B , C ) Wild-type walking pattern . 1 , 2 and 3 denote footprints of first , second , and third leg , respectively . Colored triangles denote the footprints of legs in the stance phase , where three legs are on the ground at any given time , representing tripod gait . The front leg and hind leg from one side move together with the middle leg on the opposite side . This alternates when the fly takes a step forward , represented by red and green triangles . Defective walking patterns are observed in ( E ) Sema1a , ( F ) Sema1b , ( H ) PlexA , ( I ) PlexB , but not in ( G ) Sema2a knockdowns . DOI: http://dx . doi . org/10 . 7554/eLife . 11572 . 00310 . 7554/eLife . 11572 . 004Figure 1—source data 1 . Summary of walking behavior upon Plexin/Semaphorin knockdown in motoneurons . DOI: http://dx . doi . org/10 . 7554/eLife . 11572 . 004 Given the previously demonstrated roles of Plexin/Semaphorin signaling in the development of appropriate dendritic and axonal morphology in other neuronal systems ( and during embryogenesis ) , we next investigated if these defects in locomotor activity might be predictive of knockdown-induced axonal and/or dendritic projection defects in the targeted leg motoneurons . We first focused on PlexA and its transmembrane ligand Sema1a . To determine if PlexA and Sema1a are required for correct targeting of motoneuron axons to specific muscles in the leg , we knocked down these signaling molecules in all the leg motoneurons using the OK371-Gal4 driver and examined fasciculation and targeting of their axons in the adult leg periphery . Effects on the motoneuronal innervation onto the proximal muscles of the femur and the distal muscles of the tibia were characterized separately ( Figure 2—figure supplement 1 ) . In the wild-type femur , the proximal axon branch defasciculates from the main motor nerve and innervates the ltm2 ( long tendon muscle 2 ) , while as another axon branch defasciculates from the main motor nerve more distally , projects further distally and innervates the tidm ( tibia depressor muscle ) ( Figure 2A–D ) . Following targeted knockdown of either PlexA or Sema1a , axon projection defects were seen in the proximal femur; for both PlexA and Sema1a knockdowns two similar kinds of projection phenotypes were observed either separately or in combination ( Figure 3—figure supplement 1A; Figure 3—source data 1 ) . In the first phenotype , the axon branch that normally innervates the tidm defasciculates from the main motor nerve correctly , but then stalls and does not project distally along the femur to innervate the tidm correctly ( Figure 2E–H ) . In the second phenotype , the axon branch that normally innervates ltm2 fails to defasciculate and exit from the main motor nerve and innervation of the ltm2 is lacking ( Figure 2I–L ) . 10 . 7554/eLife . 11572 . 005Figure 2 . PlexA and Sema1a are required for correct axonal projections of leg motoneurons in the proximal femur . Targeted knockdown , overexpression , and labeling mediated by motoneuron-specific OK371-Gal4 driver . ( A–D ) Control innervation of femur . Axon projection defects characterized by decreased innervation are observed in ( E–H ) Sema1a knockdown and ( I–L ) PlexA knockdown . Extensive defasciculation and ectopic branches exiting the main motor nerve are observed in ( M–P ) Sema1a overexpression and ( Q–T ) PlexA overexpression . ( A , E , I , M , Q ) show overview of proximal femur innervation . ( B , F , J , N , R ) show magnified views of innervation in ltm2 ( long tendon muscle 2 ) region . ( C , G , K , O , S ) show magnified views of innervation of proximal tidm ( tibia depressor muscle ) . ( D , H , L , P , T ) show schematic summaries of femur innervation of these two nerves . ltm2 muscle in schematic marked by green and tidm in red . The main nerve innervating ltm2 region is outlined by green arrowheads; main nerve innervating tidm outlined by red arrowheads . Green asterisk denotes absence of innervation . Multiple nerves innervating ltm2 and tidm in over-expression outlined by blue arrowheads . Scale bars = 20 microns . DOI: http://dx . doi . org/10 . 7554/eLife . 11572 . 00510 . 7554/eLife . 11572 . 006Figure 2—figure supplement 1 . Axonal innervation of motoneurons in the leg . DOI: http://dx . doi . org/10 . 7554/eLife . 11572 . 00610 . 7554/eLife . 11572 . 007Figure 2—figure supplement 2 . Plex A and Sema1a are required for correct axonal fasciculation of leg motoneurons in the proximal femur . Defective defasciculation in Sema1a and PlexA knockdown was accompanied by increased thickness of nerve’s main axon while as extensive defasciculation observed in Sema1a and PlexA overexpression was accompanied by correspondingly reduced thickness of the nerve’s main axon . DOI: http://dx . doi . org/10 . 7554/eLife . 11572 . 007 In the distal wild-type tibia , several axon branches defasciculate from the main motor nerve to innervate the tadm ( tarsus depressor muscle ) and two terminal branches defasciculate from the main motor nerve to innervate the tarm ( tarsus reductor muscle ) ( Figure 3A–D ) . After targeted knockdown of PlexA , axon projection defects characterized by a marked increase in the number of axon branches that exit the main nerve resulted ( Figure 3 E-H ) . This increase was seen both in the axon branches that innervate the tadm and in the axon branches that innervate the tarm . A comparable increase in the number of axon branches that exit the main nerve resulted after targeted knockdown of Sema1a; however , this was usually more pronounced in the axon branches that innervate tarm ( Figure 3I–L; Figure 3—figure supplement 1B ) . 10 . 7554/eLife . 11572 . 008Figure 3 . PlexA and Sema 1a are required for correct axonal projections of leg motoneurons in the distal tibia . Targeted knockdown , overexpression , and labeling mediated by motoneuron-specific OK371-Gal4 driver . ( A–D ) Control innervation of distal tibia . Axon projection defects characterized by increased innervation observed in ( E–H ) PlexA RNAi knockdown and ( I–L ) Sema1a RNAi knockdown . Decreased defasciculation of axonal branches exiting the main motor nerve are observed in ( M–P ) Sema1a overexpression and ( Q–T ) PlexA overexpression . ( A , E , I , M , Q ) show overviews of distal tibia innervation . ( B , F , J , N , R ) show magnified views of innervation of tadm ( tarsus depressor muscle ) . ( C , G , K , O , S ) show magnified views of innervation of tarm ( tarsus reductor muscle ) . ( D , H , L , P , T ) show schematic summaries of tibia innervation . Green arrowheads point toward innervations in tadm and red arrowheads point toward tarm . Scale bars = 20 microns . DOI: http://dx . doi . org/10 . 7554/eLife . 11572 . 00810 . 7554/eLife . 11572 . 009Figure 3—source data 1 . Summary of the axon defasciculation and targeting phenotypes in proximal femur and distal tibia . File contains underlying source data for Figure 3—figure supplement 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 11572 . 00910 . 7554/eLife . 11572 . 010Figure 3—figure supplement 1 . Summary of the axon defasciculation and targeting phenotypes mediated by Sema1a and PlexA , in ways that differ in the proximal femur and distal tibia . Bar graphs denote the percentage of legs showing defective motoneuron ( MN ) defasciculation and targeting phenotypes . Sema1a and PlexA knock down in motoneurons results in ( A ) decreased defasciculation and innervation of motoneuron axons in proximal femur , ( B ) increased defasciculation and innervation motoneuron axons in distal tibia . Sema1a and PlexA over-expression in motoneurons results in ( C ) increased defasciculation and innervation in proximal femur and ( D ) decreased defasciculation and innervation in distal tibia . DOI: http://dx . doi . org/10 . 7554/eLife . 11572 . 010 Taken together , these findings indicate that an abnormal reduction of PlexA/Sema1a in all leg motoneurons can lead to defective axonal branching phenotypes . The observed phenotypes differ in the proximal femur and the distal tibia in that a reduction in axon branching of motoneurons occurs in the femur and an increase in axon branching of motoneurons occurs in the tibia . Given the axon branching phenotypes observed following an abnormal reduction of PlexA/Sema1a in leg motoneurons in the knockdown experiments , we next investigated if an abnormal increase of PlexA/Sema1a in motoneurons might also result in aberrant axonal projection phenotypes . For this , we used the OK371-Gal4 driver to overexpress Sema1a and PlexA in all leg motoneurons . In the proximal femur , targeted overexpression of Sema1a in leg motoneurons resulted in an increase in axon branching that was characterized by excessive defasciculation and ectopic branches exiting the main motor nerve in all preparations analyzed ( Figure 2M–P ) . Moreover , this extensive defasciculation was accompanied by a correspondingly reduced thickness of the nerve’s main axon bundle ( Figure 2—figure supplement 2 ) . Comparable defasciculation phenotypes were observed in targeted PlexA overexpression experiments ( Figure 2Q–T; Figure 3—figure supplement 1C ) . However , in most cases , these defasciculation phenotypes were less extensive than in the Sema1a overexpression experiments , and a marked reduction of thickness in the main axon bundle of the nerve was usually not seen ( Figure 2—figure supplement 2 ) . In the distal tibia , targeted overexpression of PlexA in leg motoneurons resulted in a decrease in axon branching that was characterized by a striking reduction in the number of axon branches that exit the main motor nerve and innervate the tadm and tarm ( Figure 3Q–T ) . Targeted overexpression of Sema1a also resulted in reduced defasciculation and branching phenotypes in several cases; however , these were less extensive than in the PlexA overexpression experiments ( Figure 3M–P; Figure 3—figure supplement 1D; Figure 3—source data 1 ) . Thus , both reduction and increase of PlexA/Sema1a expression in motoneurons can result in aberrant axonal branch formation in the proximal muscles of the femur and in the distal muscles of the tibia . Moreover , reduction and increase in PlexA/Sema1a expression result in opposing branching phenotypes . Taken together , these experiments indicate that a normal level of PlexA/Sema1a-mediated signaling is required in leg motoneurons for the formation of correct axonal projections and targeted muscle innervation in the leg . The experiments reported above indicate that the targeted knockdown of PlexA/Sema1a in leg motoneurons result both in reduced axonal branching in the femur and increased axonal branching in the tibia . The coupled occurrence of these two phenotypes could arise , at least in part , if motoneurons that innervate the femur in the wild type mis-project into the tibia following PlexA/Sema1a knockdown . To investigate this possibility , we used the VGN6341-Gal4 driver which targets knockdowns to 3–4 motoneurons that innervate the femur as well as 1–2 motoneurons that innervate the tibia . In the wild type , femur motoneurons labeled by VGN6341-Gal4 always innervated the tidm and ltm2 muscles of the proximal femur and , similarly , tibia motoneurons always innervated the tadm muscles of the distal tibia . In contrast , following targeted knockdown of PlexA and Sema1a , motoneurons that normally innervated the femur manifested a significant number of mis-projection phenotypes characterized by lack of innervation of the tidm and ltm2 muscles of the femur and ectopic innervation of the ltm1 and tidm muscles of the tibia . Comparable mis-projection phenotypes were seen in these femur motoneurons in a Sema1a homozygous null mutant , Sema1a hetreroallelic combination and in animals trans-heterozygous for mutants in both Sema1a and PlexA of the genotype Sema1aP2/+;;PlexA09/+ . Figure 4 and Figure 4—figure supplement 1A; Figure 4—source data 1 show the range of phenotypes . We used another Gal4 driver , R6B011 which targets knockdowns to 3–4 motoneurons that innervate the femur and 2–3 motoneurons that innervate the tibia . In the wild type , femur motoneurons labeled by R6B011-Gal4 always innervated the proximal femur . In contrast , following targeted knockdown of PlexA , motoneurons that normally innervated the femur manifested a significant number of mis-projection phenotypes characterized by lack of innervation of the tidm and ltm2 muscles of the femur . Similar but weaker phenotype was observed following targeted knockdown of Sema 1a ( Figure 4—figure supplement 1B , Figure 4—figure supplement 2 ) . 10 . 7554/eLife . 11572 . 011Figure 4 . PlexA and Sema1a are required in femur motoneurons to maintain appropriate innervation of the femur and prevent inappropriate innervation of the tibia . VGN6341-Gal4 targeted knockdown of PlexA and Sema1a in a small subset of leg motoneurons . ( A , C , E , G , I , K ) innervation of proximal femur . ( B , D , F , H , J , L ) innervation of tibia . ( A , B ) Control innervation by targeted motoneurons . ( C , D ) Sema1a knockdown in targeted motoneurons . ( E , F ) PlexA knockdown in targeted motoneurons . ( G , H ) Sema1aP2 homozygous mutant . ( I , J ) Sema1aP1/Sema1aP2 heteroallelic combination . ( K , L ) Sema1ap2/+;;PlexA09/+ trans-heterozygous mutant combination . Decreased innervation occurs in femur ( C , E , G , I , K ) and ectopic innervation occurs in tibia ( D , F , H , J , L ) . Arrowheads point toward innervation and asterisk denote absence of normal innervation . Scale bars = 20 microns . DOI: http://dx . doi . org/10 . 7554/eLife . 11572 . 01110 . 7554/eLife . 11572 . 012Figure 4—source data 1 . Summary of the motoneuron axon defasciculation and targeting phenotypes in early and late-born motoneurons . File contains underlying source data for Figure 4—figure supplement 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 11572 . 01210 . 7554/eLife . 11572 . 013Figure 4—figure supplement 1 . Summary of the motoneuron axon defasciculation and targeting phenotypes mediated by Sema1a and PlexA in early and late-born motoneurons . Bar graphs denote the percentage of legs showing normal motoneuron ( MN ) defasciculation and targeting . ( A ) VGN6341-Gal4 labeled motoneurons show mis-projection phenotypes characterized by lack of innervation in ltm2 , tidm muscles of the femur and ectopic innervation in ltm1 , tadm muscles of the tibia in Sema1a knockdown , PlexA knockdown , Sema1a homozygous null mutant , Sema1a heteroallelic mutant combination and Sema1a/PlexA trans-heterozygous mutants . ( B ) R60B11-Gal4 mediated knockdown of Sema1a and PlexA results in decreased defasciculation and targeting of early-born motoneurons that innervate proximal femur . ( C ) VGN9281-Gal4 mediated knockdown of Sema1a and Plex A in one late-born motoneuron , does not affect the motoneuron axon targeting in distal tibia . DOI: http://dx . doi . org/10 . 7554/eLife . 11572 . 01310 . 7554/eLife . 11572 . 014Figure 4—figure supplement 2 . PlexA and Sema1a are required in femur motoneurons for axonal targeting and defasciculation . R6B011-Gal4 targeted knockdown of PlexA and Sema1a in a small subset of leg motoneurons . ( A , C , E ) innervation of proximal femur . ( B , D , F ) innervation of tibia . ( A , B ) Control innervation by targeted motoneurons . Decreased innervation occurs in femur following ( C , D ) Sema1a knockdown in targeted motoneurons and ( E , F ) PlexA knockdown in targeted motoneurons . Arrowheads point toward innervation and asterisk denote absence of normal innervation . Scale bars = 20 microns . DOI: http://dx . doi . org/10 . 7554/eLife . 11572 . 01410 . 7554/eLife . 11572 . 015Figure 4—figure supplement 3 . Plex A and Sema1a are not required in tibia late-born motoneurons for axonal targeting . VGN9281-Gal4 targeted knockdown of PlexA and Sema1a in a late-born leg motoneuron . ( A , B ) Control innervation by targeted motoneuron . ( C ) Sema1a knockdown and ( D ) PlexA knockdown in targeted motoneuron . DOI: http://dx . doi . org/10 . 7554/eLife . 11572 . 015 To investigate if tibial motoneurons mis-project processes into the femur following PlexA/Sema1a knockdown , we used the VGN9281-Gal4 driver which targets knockdowns to a single motor neuron that innervates the tarsus reductor muscle ( tarm2 ) . Interestingly , while the targeted motorneuron did not misproject into femur , it showed abnormally increased branching in the distal tibia ( Figure 4—figure supplement 1C , Figure 4—figure supplement 3 ) . The ltm2 muscle in proximal femur is innervated by early born motoneurons and the tadm/ tarm muscles in the distal tibia are innervated by late born motoneurons ( Brierley et al . , 2009; 2012 ) . Taken together , these findings imply that a reduction of PlexA/Sema1a in motoneurons that normally innervate the femur can result in a loss of appropriate femoral innervation and a gain of inappropriate tibial innervation . These specific mis-projection phenotypes of femur motoneurons are likely to contribute to the observed overall axonal projection phenotype in the leg , characterized by decreased axonal branching in the femur and increased axonal branching in the tibia . Previous work on the development of motoneurons during embryogenesis has shown that both the transmembrane Sema1a ligand acting through PlexA , and the secreted Sema2a ligand acting through PlexB are important for correct axonal outgrowth ( Matthes et al . , 1995; Yu et al . , 1998; Winberg et al . , 1998 ) . The absence of locomotor defects in our motoneuron-targeted Sema2a knockdown experiments ( see above ) suggests that Sema2a might not act in an autonomous manner on leg motoneuron development . Indeed , motoneuron-specific knockdown of Sema2a using the OK371-Gal4 driver did not result in peripheral axon projection defects in leg motoneurons ( data not shown ) . To investigate the possibility that Sema2a might act on leg motoneurons in a non-cell autonomous manner , we first compared the expression pattern of Sema1a and Sema2a in the thoracic ganglia during postembryonic development using immunolabeling . Throughout postembryonic development , both Sema1a and Sema2a are widely expressed throughout most of the developing thoracic ganglia ( Figure 5A–F ) . In this respect , their expression is comparable to that of their PlexA and PlexB receptors , which are also broadly expressed in the nervous system ( Winberg et al . , 1998 ) . However , Sema1a and Sema2a expression levels are significantly different at the ganglionic midline . Thus , whereas Sema1a expression decreases toward the midline and has its lowest level of expression there , Sema2a expression is highest at/and near midline where a peak in its expression level is manifest ( Figure 5G–H; Figure 5—figure supplement 1 ) . This high level of expression in and around midline cells , many of which are glial in nature , suggests that Sema2a ligand secreted from these cells could affect leg motoneuron development within the thoracic ganglia and hence , influence the pattern of dendrite outgrowth in leg motoneurons in a non-cell autonomous manner . 10 . 7554/eLife . 11572 . 016Figure 5 . Sema2a is highly expressed in the ganglionic midline and intermediate region . Immunocytochemical analysis of expression of Sema1a and Sema2a in prothoracic ganglion at 25h APF ( after puparium formation ) . ( A–C ) Ventral section . ( D–F ) Dorsal section . ( A , D ) Sema1a immunolabeling . ( B , E ) Sema2a immunolabeling . ( C , F ) Overlay of Sema1a and Sema2a immunolabeling . ( G ) Magnified view of neuropile . ( H ) Intensity profile of Sema1a and Sema2a immunolabeling taken along mediolateral axis of neuropile ( in Z-stack; overlay of all optical sections ) . Yellow asterisk is placed at the ganglionic midline ( ML ) . Sema2a level is highest at the midline and high in the intermediate regions in the neuropile and very low at the lateral edge . Scale bar = 20 microns . DOI: http://dx . doi . org/10 . 7554/eLife . 11572 . 01610 . 7554/eLife . 11572 . 017Figure 5—figure supplement 1 . Dorso-ventral view of Sema2a expression in the thoracic neuropil . ( A ) 3D- reconstruction of Sema1a and Sema2a expression along dorso-ventral axis reveals opposing gradients of Sema1a and Sema2a along dorso-ventral ( D-V ) and medio-lateral ( M-L ) axis in prothoracic ganglion at 25h APF . ( B ) Intensity profile of Sema1a and Sema2a immunolabeling taken along medio-lateral axis of neuropile representing dorsal and ventral optical sections . DOI: http://dx . doi . org/10 . 7554/eLife . 11572 . 017 To determine if Sema2a and PlexB are required for correct central dendritic arborization of leg motoneurons , we carried out RNAi knockdown experiments and examined the dendritic projection patterns of motoneurons in the thoracic ganglia . For this , we analyzed the dendrites of two sets of motoneurons labeled by dye backfilling techniques ( see 'Materials and methods' ) . Dye backfilling of leg nerve from proximal muscles of femur labeled early-born motoneurons while backfilling the leg nerve from distal muscles of the tibia labeled late-born motoneurons ( Figure 6A–D ) . In the wild type , early-born motoneurons have dendrites that arborize primarily in the ipsilateral neuropile , but also project one prominent branch to the ganglionic midline ( Figure 6C ) . In contrast , wild-type late-born motoneurons have dendrites whose arborizations are restricted to lateral part of the ipsilateral neuropile; none of their dendrites project medially ( Figure 6D ) . 10 . 7554/eLife . 11572 . 018Figure 6 . PlexB and Sema2a are required for correct dendritic targeting of leg motoneurons . Motoneuron dendrites in prothoracic hemiganglion labeled by dye backfilling from innervated muscles . ( A , C , E , G ) early born motoneurons that innervate ltm2 . ( B , D , F , H ) late born motoneurons that innervate tadm . ( A , B ) Schematic representation of dye backfilled leg muscles . ( C , D ) Dendrites of motoneurons in wild-type control . ( E , F ) Dendrites of motoneurons following motoneuron-specific targeted knockdown of PlexB . ( G , H ) Dendrites of motoneurons following glial-cell-specific targeted knockdown of Sema2a . A dendritic mis-projection phenotype occurs in late-born motoneurons ( F , H ) but not in early-born motoneurons ( E , G ) . White arrowhead indicates normal dendritic projection and red arrowhead indicates mis-projection . Red asterisk marks the region where projection is normally absent in late-born neurons . Scale bars =20 microns . DOI: http://dx . doi . org/10 . 7554/eLife . 11572 . 01810 . 7554/eLife . 11572 . 019Figure 6—figure supplement 1 . Sema2a secreted by midline cells is required for dendritic targeting of late-born motoneurons . Dendrites of late-born motoneurons mis-project toward the midline when Sema2a is knocked down in midline cells . Asterisk marks the region near the midline where dendritic projections of late-born motoneurons are normally absent . Mis-projected branch outlined by red arrowheads . DOI: http://dx . doi . org/10 . 7554/eLife . 11572 . 01910 . 7554/eLife . 11572 . 020Figure 6—figure supplement 2 . Plex B/Sema2a is required for correct dendritic targeting of late-born motoneurons . Dendrites of late-born motoneurons mis-project toward the midline . Asterisk marks the region near the midline where dendritic projections of late-born motoneurons are normally absent . Mis-projected branch outlined by red arrowheads . Scale bar = 20 microns . DOI: http://dx . doi . org/10 . 7554/eLife . 11572 . 020 RNAi knockdown of PlexB in motoneurons using the OK371-Gal4 driver resulted in a marked dendritic arborization phenotype in the late-born motoneurons but not in the early born motoneurons ( Figure 6E , F ) . This phenotype was characterized by a shift of the dendritic arborizations toward the midline as well as by the formation of a prominent ectopic dendritic branch , which projects to the ganglionic midline . In this respect , late-born neurons in the knockdown experiment resemble early-born neurons in the wild type . To confirm this dendritic arborization phenotype , we used the VGN9281-Gal4 driver line to target PlexB knockdown to specific late-born motoneurons ( see Brierley et al . , 2009 ) . This resulted in the expected dendritic arborization phenotypes in targeted motoneurons , in that these late-born motoneurons , which normally restrict their dendritic arbors to lateral domains , mis-projected major dendritic branches toward the ganglionic midline ( data not shown ) . Targeted knockdown of Sema2a in leg motoneurons using the OK371-Gal4 driver had no effect on dendritic arbors of late-born or early-born motoneurons ( data not shown ) . This corresponds to the lack of locomotor defects in behavioral experiments in which Sema2a RNAi was targeted to leg motoneurons ( see Figure 1 ) and supports the notion that Sema2a might not act cell autonomously in motoneurons . Since immunolabeling studies showed that Sema2a expression in the neuropile is most prominent in/and near midline ( glial ) cells ( see Figure 5 ) , we carried out a targeted knockdown of Sema2a in glial/midline cells using either the glial-specific Repo-Gal4 driver or the midline-specific Slit-Gal4 driver . In both cases , Sema2a knockdown resulted in clear dendritic arborization defects of leg motoneurons . Moreover , these defects were observed in late-born motoneurons but not in early-born motoneurons . Thus , following targeted knockdown of Sema2a in glial cells , late-born motoneurons , which normally restrict their dendritic arbors to lateral domains , shift their dendritic arbors medially and mis-project major dendritic branches toward the ganglionic midline ( Figure 6G , H; Figure 6—figure supplement 1 ) . Comparable mis-projection phenotypes were seen in dendrites of late born motoneurons in animals of the genotype Sema2a/+; PlexB/+ ( Figure 6—figure supplement 2 ) . Upon this dendritic mis-projection in late-born motoneurons , dendritic arborizations of these motoneurons acquired morphological characteristics of arborizations of normal early born neurons . To investigate if overexpressing Sema1a ( or Sema2a ) in the midline repels dendritic arbors of early born motoneurons from the midline , with these motor neurons acquiring morphological characterists of late-born neurons , we carried out a targeted over-expression of Sema1a ( or Sema2a ) using the midline-specific Slit-Gal4 driver . In this case , the dendritic arborization of early born neurons appeared to be normal ( data not shown ) . Taken together these experiments indicate that PlexB/Sema2a signaling is required for the formation of correct dendritic projections of late-born leg motoneurons . Moreover , they imply that the requirement in these motoneurons is cell autonomous for PlexB and non-cell autonomous ( i . e . glial/midline cell-derived ) for Sema2a . During development , motoneurons extend axons over relatively long distances from the thoracic ganglia to the numerous muscles located within their peripheral targets . During this process , multiple guidance cues act to ensure that the motoneurons establish their complex innervation patterns with high precision ( Tessier-Lavigne and Goodman , 1996; Raper and Mason , 2010; Kolodkin and Tessier-Lavigne , 2011 ) . As motoneurons navigate into the periphery , they often initially fasciculate with each other to form nerve branches and , once they reach their general target regions , defasciculate and exit their nerve branches to innervate specific muscles . Repulsive Semaphorin signaling through Plexin receptors has been implicated in this process in vertebrates and invertebrates ( Pasterkamp et al . , 2006; 2012 ) . In Drosophila , Plexin/Semaphorin signaling has been studied in the development of the embryonically generated motoneurons that innervate the larval body wall ( Winberg et al . , 1998; Yu et al . , 1998; Ayoob et al . , 2006 ) . Sema1a acting through PlexA has been shown to be important for the regulation of motoneuron axon defasciculation through axon-axon repulsion ( Winberg et al . , 1998; 2001; Yu et al . , 1998; 2000 ) . Similarly , Sema2a acting through PlexB is required for correct motoneuron axon defasciculation , but also has additional functions in motoneuron axon navigation ( Matthes et al . , 1995; Ayoob et al . , 2006 ) implying that PlexA/Sema1a and PlexB/Sema2a have both shared and distinct function in motor axon guidance . Our analysis of axonal targeting in leg motoneurons indicates that the PlexA/Sema1a signaling system is also required in postembryonic development of adult-specific motoneurons . Targeted knockdown of PlexA or Sema1a in leg motoneurons results in prominent defasciculation and targeting defects of their axons in the periphery . Interestingly , different types of defects are seen in the innervation of proximal versus distal leg muscles . In the innervation of the proximal femur muscles , PlexA/Sema1a knockdown results in phenotypes indicative of defective axon defasciculation and decreased branching . In contrast , in the innervation of the distal tibia muscles , PlexA/Sema1a knockdown results in phenotypes indicative of increased axonal branching . Thus , both decrease and increase of motor axon branching can result if PlexA/Sema1a action in leg motoneurons is impaired . At the cellular level , both these phenotypes may be due to defects in early born femur motoneurons which manifest a loss of appropriate proximal femoral innervation and gain tibial innervation . In addition , increased arborization of targeted late-born tibial motoneurons in the distal tibia is also likely to contribute to the observed phenotypes . These findings suggest that the same Plexin/Semaphorin signaling system might have multiple roles in motor axon guidance . Since these different effects appear to be motoneuron-specific , they may be linked to motoneuron birth-order . An interesting feature of all the PlexA/Sema1a phenotypes observed in our experiments is that they are the result of knockdowns limited to motoneurons , either in their ensemble or in specific small sets . In all cases , the motoneurons that manifest these peripheral axonal defasciculation and targeting defects are the only cells in the leg that have impaired levels of PlexA/Sema1a . In this respect , the requirement of both PlexA and Sema1 for correct axonal targeting is autonomous to leg motoneurons . This suggests that PlexA/Sema1a signaling might be at the level of axon-axon interactions . A role of Plexin/Semaphorin signaling in axon-axon interactions has been documented in the developing visual and olfactory systems of Drosophila ( Sweeney et al . , 2007; Hsieh et al . , 2014 ) . Alternatively , cis-interactions between PlexA and Sema1a within the affected motoneuron axons might also be involved . In murine models , cis- interactions between Sema6A and PlexA4 have been shown to modulate the repulsive response to Sema6A in sympathetic and sensory neurons ( Haklai-Topper et al . , 2010 ) . While our experiments document an autonomous requirement of PlexA/Sema1a within the ensemble of leg motoneurons , they do not rule out the existence of PlexA/Sema1a signaling between outgrowing motor axons and their target muscles in the leg . Thus , it will be important to study both the expression and function of Plexin/Semaphorin signaling in the developing leg muscles . During development , motoneurons must not only project their axons to the appropriate postsynaptic target muscles , they must also ensure that their dendrites , as major input sites , are positioned correctly in the neuropile in order to receive the appropriate presynaptic drive from pre-motor interneurons and sensory neurons ( Parrish et al . , 2007; Vrieseling and Arber , 2006; Harris et al . , 2015; Arber , 2012 ) . Although the mechanisms that control this type of targeted dendritogenesis are poorly understood , recent work on embryonic and adult-specific leg motoneurons in Drosophila indicate that their dendrites are organized topographically as myotopic maps that reflect their innervation pattern in the periphery ( Landgraf et al . , 2003; Brierley et al . , 2012; 2009; Baek and Mann , 2009 ) . For leg motoneurons , the majority of which are generated post-embryonically in a single neuroblast lineage , a birth-order dependent pattern of dendritic targeting and of peripheral innervation has been documented ( Brierley et al . , 2009; 2012 ) . Thus , early-born motoneurons , which innervate proximal muscle groups , position their dendrites in medial and lateral neuropile , while late-born motoneurons , which innervate distal muscle groups , elaborate their dendrites in the lateral neuropile only , and intermediate-born motoneurons target their dendrites to intermediate neuropile positions . Irrespective of their birth-order , all leg motoneurons initiate dendritogenesis synchronously and form their dendritic arbors in a targeted manner by growing into their appropriate area of innervation . Leg motoneuron dendrites attain their final architecture through targeted dendritogenesis . Apart from the intrinsic signaling cues , extrinsic cues could be implicated in this process . Although glia are known to secrete cues and neurotropic factors for axonal morphogenesis and survival ( Jacobs , 2000; Martin et al . , 2012; Zlatic et al . , 2009; Klämbt et al . , 1991 ) , their role in regulation of dendritic morphology is relatively unexplored . While recent findings in vertebrates and invertebrates suggest that glia could play a role in dendritogenesis , the mechanism underlying these effects is not known ( Procko and Shaham , 2010 ) . Midline glia extend gliopodia during axonogenesis , and reducing their numbers also leads to the shift in neuropil during embryonic nervous system development in Drosophila ( Vasenkova et al . , 2006 ) suggesting that these could also play a role in dendritogenesis . Investigation of the molecular mechanisms involved in dendritic development of leg motoneurons has focused on the role of Slit/Roundabout and the Netrin/Frazzled ( Brierley et al . , 2009 ) signaling systems in controlling the precise targeting of motoneuron dendrites along the medio-lateral axis of the ganglionic neuropile . While Slit/Netrin could be derived from midline glia , the role of other midline cells ( Kearney et al . , 2004 ) has not been ruled out . In this context , the substantially broader expression pattern of Sema2a , compared to that of Slit ( which is restricted to the midline ) is pertinent ( Figure 5B , E; Figure 5—figure supplement 1 ) . Dendritic targeting in larval motoneurons , which are of embryonic in origin is independent of glial differentiation ( Landgraf et al . , 2003; Landgraf and Thor , 2006 ) . Our investigation of dendritic targeting of leg motoneurons in adult Drosophila provides further insight into the role of glia in the control of dendritogenesis . Moreover , they identify PlexB/Sema2a as additional signaling system that mediates this control . Our data are in accordance with a model in which Sema2a is secreted from midline and other glial cells and acts through the PlexB receptor to restrict dendrites of late-born motoneurons to appropriate lateral neuropile domains . Loss of PlexB in these motoneurons as well as loss of Sema2a from glial cells , as seen by using both Repo-Gal4 and Slit-Gal4 drivers , results in a medial shift in dendritic arbors together with the mis-projection of dendrites toward the midline . As a result , the affected dendrites of late-born neurons acquire the topographic morphology of normal early-born neurons in the ganglionic neuropile . Since cells at/and near the ganglionic midline express high levels of Sema2a , and since targeted knockdown of Sema 2a specifically in glial cells using Repo-Gal4 results in the mis-positioning of late-born motoneuron dendrites , we postulate that glial cells provide the secreted Sema2a , which acts on dendrite targeting in late-born motoneurons . Thus , while the requirement of PlexB for dendrite targeting in late-born motoneurons is cell autonomous , the requirement of Sema2a is non-cell autonomous and presumably is provided through secretion from glial cells . We also observe a dorso-ventral gradient of Sema1a and Sema2a ( Figure 5—figure supplement 1 ) . This suggests a role in dorso-ventral patterning as well , an aspect which we have not examined . Taken together , our findings indicate that two different Plexin/Semaphorin signaling systems act to play a range of roles in establishing the specific leg motoneuron architecture needed for appropriate motor circuitry . PlexA/Sema1a is required for correct peripheral axon guidance , and for both PlexA and Sema1a this requirement is motoneuron-specific . PlexB/Sema2a is required for correct central dendrite targeting , and while this requirement is motoneuron-specific for PlexB , the requirement for Sema2a in motoneurons is non-cell autonomous and likely involves glial cells . Interestingly , while PlexA/Sema1a appears to be required for both early-born and late-born motoneuron axon guidance , PlexB/Sema2a appears to be required for late-born , but not for early-born motoneuron dendritic patterning . The relative independence of action of these two different Plexin/Semaphorin signaling systems in axon guidance and dendrite targeting of leg motoneurons suggests that both developmental processes could , in principle operate at separate times . This seems unlikely as timeline analysis indicates that both peripheral axonal outgrowth and central dendritic targeting of leg motoneurons occur at the same/overlapping times during postembryonic development ( Brierley et al . , 2009; 2012 ) . The molecular mechanisms that orchestrate this simultaneous formation of axonal and dendritic arbors are not known . Possible mechanisms include cell-intrinsic temporal modulation of transcription factors controlling guidance molecule expression as well as retrograde feedback from the innervated muscle targets as seen in the vertebrate spinal cord ( Vrieseling and Arber , 2006; Enriquez et al . , 2015 ) . Our behavioral experiments indicate that specific experimental perturbations in the expression levels of the Plexin/Semaphorin signaling systems result in abnormal walking behavior . Moreover , they show that the same molecular perturbations also result in defective axonal and dendritic branching and arborization patterns in leg motoneurons . Although other effects of the Plexin/Semaphorin signaling systems on the neural circuitry for walking control may exist , the striking correlation between perturbation in leg motoneuron architecture and defective walking behavior provides further support for the idea that peripheral and central architecture is likely to be important for the function of the motoneurons involved in the neural circuitry for walking . Although we did not observe any defective footprint pattern using the soot plate assay , following Sema2a knockdown in midline/glia using Repo-Gal4 or mid-line specific Slit-Gal4 ( data not shown ) , we do not rule out the possibility of other intricate walking parameters e . g . speed , stance , swing-duration or stance linearity etc . being affected in this case . Given the conservation of molecular cues involved in the development of neuronal morphology and connectivity , further insight into the Plexin/Semaphorin-dependent mechanisms that operate in motoneuron development in Drosophila is likely to be of general importance for understanding the basis of circuit formation in all parts of the nervous system ( Pasterkamp , 2012 ) . The following fly strains were used; OK371-Gal4 , UAS-mCD8GFP , UAS-dicer; OK371-Gal4 , UAS-mCD8GFP , Repo-Gal4 , Slit-Gal4 , VGN9281-Gal4 , UAS-mCD8GFP ( Brierley et al . , 2009 ) , VGN6341-Gal4 , UAS-mCD8GFP , R60B11-Gal4 ( 39238; Bloomington Drosophila Stock Center ) . RNAi lines and mutants used in these experiments have been described previously . UAS Sema1a RNAi ( VDRC , TRiP , Gift from Alex Kolodkin ( Sweeney et al . , 2007 ) , UAS PlexA RNAi ( VDRC; Gift from Liqun Luo ( Sweeney et al . , 2007; Pecot et al . , 2013 ) , UAS Sema2a RNAi ( VDRC ( Sweeney et al . , 2011 ) , UAS PlexB RNAi ( VDRC , TRiP ) , CA07125 ( Fly trap ) , Sema1a P1 and Sema1a P2 ( Yu et al . , 1998 ) , plexADf ( 4 ) C3 ( PlexA09 ) ( Winberg et al . , 1998 ) , plexBKG00878 ( Ayoob et al . , 2006; Bellen et al . , 2004 ) . A plasmid containing 505 base pair DVGlut gene enhancer that corresponds to the Drosophila melanogaster genome coordinates , BDGP6:2L:2394617:2395172 was created as follows: The nucleotide sequence was amplified from wild-type Drosophila melanogaster DNA to include restriction enzyme cleavage sites , BamH1 using the primers 5' EcoR1 ( + ) TTTTCGCCTTTTTGCAGTC and 3'BamH1 ( + ) GCTTCAGCAGCAAACAATGA . The pPT Gal-attB vector ( Brierley et al . , 2009 ) was modified to contain the previously amplified enhancer region cloned directionally into EcoR1 and BamH1 sites . This plasmid containing enhancer sequence was then injected into attP2 fly embryos ( Sharma et al . , 2002 ) to make VGN 6341-Gal4 transgenic line . Glass slides ( dimensions: 7 cm x 5 cm ) were coated with a thin layer of carbon soot using a candle . Care was taken so that the soot layer was uniform . As the fly walks across the soot plate , soot is dislodged from the points of contact . Thus we obtain footprints of the fly , which can be analyzed further ( as described in Maqbool et al . , 2006 ) . The soot plates were imaged under a trinocular transmission microscope using a digital camera ( Canon Powershot ) . The images were converted to gray-scale , and the brightness and contrast was adjusted , for easy comparison . Further binary dilation was performed to improve visibility of footprints . Thoracic ganglia were dissected at different developmental stages in pupa and adult in phosphate buffer saline ( pH 7 . 8 ) ( PBS ) and fixed in 4% buffered formaldehyde for 45 min at 4°C . Immunocytochemistry was performed according to Brierley et al . , 2009 . The following primary antibodies were used: Chicken pAb α-GFP ( 1:1000 , Abcam ) , Rabbit ( Rb ) α-GFP ( 1:1000 , Abcam ) , mouse ( ms ) α-Neuroglian ( BP104 , 1:40; DSHB ) , ms α-Sema2a ( 1:10; DSHB ) , Rb α -Sema-1a ( Yu et al . , 1998 ) ( 1:3000 ) , Rb α -PlexA ( Sweeney et al . , 2007 ) ( 1:500 ) . Secondary antibodies ( 1:400 ) from Invitrogen conjugated with Alexa fluor-488 , 568 and 647 were used in all staining procedures . Retrograde labeling of motoneurons was done by dye backfilling the specific leg muscles that are innervated by either early or late born motoneurons . A small crystal of the dye was inserted into the neuromuscular junction and it was allowed to diffuse for four hours at 4°C . To label the early born motoneuron , ltm2 muscle was backfilled . To label the late born motoneuron , distal muscle of the tibia ( tadm/ tarm ) was backfilled . Thoracic ganglia were dissected and fixed in 4% buffered formaldehyde for 45 min at 4°C . Following a wash with PTX , tissues were mounted on slides in Vectashield ( Vector labs ) . Alternatively after fixing the samples , immunolabeling was performed following the same protocol as mentioned above . The following dyes were used from Invitrogen: Fluorescently labeled thoracic ganglia and legs were imaged at 60X using Olympus FV 1000 confocal point scanning microscope . Optical sections were collected at 1 micron interval and imported into NIH Image J ( http://rsb . info . nih . gov . nih-image/ ) . For processing dendritic images , sensory axons arbors were removed using Lasso tool if required . The maximum z-projections were then imported into Photoshop ( Adobe , San Jose , CA ) and minor adjustments were made to the brightness and contrast when required . To view axonal arborization of the motoneurons , we directly imaged GFP labeled motoneurons through the body wall and leg of adult flies . For leg images , auto-fluorescence from the cuticle that marked the outline was removed from each optical section using Lasso tool . The 3D reconstructions were generated using Amira ( Visage Imaging , Berlin , Germany ) . To quantify the distribution of the Sema 1a and Sema2a along the medio-lateral axis of the neuropil , we used the plot profile tool in ImageJ . Analysis was performed as described in Brierley et al . , 2009 .	Nerve cells enable us to both sense the world around us and to move about it . The nerves responsible for movement are called motor neurons . While one end of a motor neuron stimulates the muscle it is connected to , the other end receives signals from nerves in the spinal cord that relay messages about movement from the brain . Motor neuron connections in the spinal cord , or its equivalent in insects , the ventral nerve cord , are organized into an arrangement known as a myotopic map , which reflects the anatomical arrangement of the muscles in the body . Much remains to be learnt about how these maps form . Syed et al . have investigated how the myotopic map develops for motor neurons in the legs of fruit flies by reducing the function of chosen genes in the ventral nerve cord and asking how this affects the myotopic map . The experiments disrupted a signaling system called the Semaphorin signaling pathway that guides motor neurons to the right target muscle and consists of different receptor-signaling molecule pairs . By looking for flies with an abnormal walk and with disrupted motor neuron organization , Syed et al . identified receptor-signal pairs that guide motor neurons to different leg muscles . Specific receptor-signal pairs also guide the organisation of motor neurons in the ventral nerve cord . This guidance depends on when neurons are ‘born’ . While a receptor-signal pair targets early born neurons to one leg muscle , the same receptor-signal pair regulates a different aspect of guidance in late-born neurons . Cells called glia , which are related to neurons , also help to position the connections of late-born motor neurons in the ventral nerve cord . Overall , the Semaphorin signaling system assists communication both within motor neurons and between glia cells and motor neurons during the formation of the myotopic map for leg motor neurons . These discoveries open new avenues of investigation into how else these cells communicate with each other to aid the development and organization of motor neurons .	[ "Abstract", "Introduction", "Results", "Discussion", "Material", "and", "methods" ]	[ "developmental", "biology", "neuroscience" ]	2016	Glial and neuronal Semaphorin signaling instruct the development of a functional myotopic map for Drosophila walking
Adiponectin-mediated pathways contribute to mammalian homeostasis; however , little is known about adiponectin and adiponectin receptor signaling in arthropods . In this study , we demonstrate that Ixodes scapularis ticks have an adiponectin receptor-like protein ( ISARL ) but lack adiponectin , suggesting activation by alternative pathways . ISARL expression is significantly upregulated in the tick gut after Borrelia burgdorferi infection , suggesting that ISARL signaling may be co-opted by the Lyme disease agent . Consistent with this , RNA interference ( RNAi ) -mediated silencing of ISARL significantly reduced the B . burgdorferi burden in the tick . RNA-seq-based transcriptomics and RNAi assays demonstrate that ISARL-mediated phospholipid metabolism by phosphatidylserine synthase I is associated with B . burgdorferi survival . Furthermore , the tick complement C1q-like protein 3 interacts with ISARL , and B . burgdorferi facilitates this process . This study identifies a new tick metabolic pathway that is connected to the life cycle of the Lyme disease spirochete . Adiponectin , adipocyte complement-related protein of 30 kDa ( or Acrp30 ) , plays important roles in the regulation of metabolism , insulin sensitivity , and inflammation across species ( Kadowaki et al . , 2006; Ouchi and Walsh , 2007; Yamauchi et al . , 2002 ) . Adiponectin mediates its actions mainly via binding adiponectin receptors with its globular C1q domain ( Buechler et al . , 2010; Yamauchi et al . , 2002 ) . Two adiponectin receptors , AdipoR1 and AdipoR2 , have been identified in mammals ( Yamauchi et al . , 2003 ) . AdipoR1 and R2 belong to a family of membrane receptors predicted to contain seven transmembrane ( TM ) domains with an internal N terminus and an external C terminus ( Yamauchi et al . , 2003 ) . AdipoR1 has a higher binding affinity for the globular form of adiponectin , whereas AdipoR2 has a greater affinity for full-length adiponectin ( Yamauchi et al . , 2003 ) . Interestingly , AdipoR1 and AdipoR2 double-knockout mice have increased triglyceride levels and exhibit insulin resistance , demonstrating that AdipoR1 and AdipoR2 regulate lipid and glucose homeostasis ( Kadowaki et al . , 2006; Yamauchi et al . , 2007 ) . In yeast , a homolog of mammalian adiponectin receptors , ORE20/PHO36 , is involved in lipid and phosphate metabolism ( Karpichev et al . , 2002 ) . PHO36 can also interact with a plant protein , osmotin , a homolog of mammalian adiponectin , thereby controlling apoptosis in yeast ( Narasimhan et al . , 2005 ) . Adiponectin and adiponectin receptors in disease-transmitting arthropods have not been characterized . By utilizing the amino acid sequence homology search in other model arthropods , adiponectin was not identified from Drosophila melanogaster; however , an adiponectin receptor that regulates insulin secretion and controls glucose and lipid metabolism was characterized ( Kwak et al . , 2013 ) . In addition , Zhu et al . , 2008 cloned an adiponectin-like receptor gene from the silk moth , Bombyx mori , and found that infection with B . mori nucleopolyhedrovirus significantly increased adiponectin receptor mRNA levels in the midgut of susceptible B . mori , suggesting an association with pathogen infectivity . Ixodes scapularis , the black-legged tick , is an important vector of the Lyme disease agent , Borrelia burgdorferi ( Estrada-Peña and Jongejan , 1999 ) , which causes approximately 300 , 000 cases annually in the United States ( Rosenberg et al . , 2018 ) . B . burgdorferi is acquired when larval or nymphal ticks feed on infected animals , and is transmitted by nymphs or adults to vertebrate hosts ( Kurokawa et al . , 2020 ) . Lyme disease in humans manifests as a multisystem disorder of the skin and other organs ( e . g . , joints , heart , and nervous system ) , resulting in patients experiencing cardiac , neurological , and arthritic complications ( Asch et al . , 1994; Singh and Girschick , 2004 ) . A human vaccine against Lyme disease was approved by the FDA but is not currently available ( Steere et al . , 1998 ) . Targeting tick proteins has the potential to disrupt tick feeding and alter B . burgdorferi colonization or transmission ( Kurokawa et al . , 2020 ) , thereby offering a new way to interfere with the life cycle of the Lyme disease spirochete . In the present study , we demonstrate that an I . scapularis adiponectin receptor-like ( ISARL ) protein facilitates B . burgdorferi colonization of the tick . ISARL-mediated stimulation of I . scapularis metabolic pathways are associated with spirochete colonization , and a tick complement C1q-like protein 3 contributes to ISARL activation . As tick metabolism changes during pathogen colonization , and adiponectin-associated pathways mediate diverse metabolic activities , we examined the I . scapularis database for two of the prominent genes linked to this pathway . The available I . scapularis database ( taxid:6945 ) in NCBI was searched with the genes for mammalian adiponectin and adiponectin receptors , and results with the human and mouse genes are shown . There were no tick genes with high homology to the genes for human and mouse adiponectin full-length sequences . Interestingly , there was an I . scapularis gene ( GenBank number: XM_029975213 ) with substantial homology to the human and murine adiponectin receptors , which we designated I . scapularis adiponectin receptor-like ( ISARL ) . The corresponding ISARL protein sequence ( GenBank number: XP_029831073 ) was also identified . The full-length ISARL mRNA encoded a protein with 384 amino acid residues and 71% amino acid sequence similarity to both the human and mouse adiponectin receptor proteins 1 and 2 . It also has high similarity ( 87% ) to homologs from insect species , including the D . melanogaster adiponectin receptor ( GenBank number: NP_732759 ) ( Figure 1—figure supplement 1 ) . Structure prediction and hydrophobicity analysis indicated that ISARL has seven TM domains ( Figure 1—figure supplement 2 ) . Comparison of the amino acid sequences between vertebrate and invertebrate species revealed that the predicted TM regions are highly conserved , especially in the TM3 domain ( Figure 1—figure supplement 1 ) . As I . scapularis lack an obvious adiponectin homolog , we examined whether expression of ISARL could be stimulated in the feeding vector by allowing ticks to engorge on mice , including uninfected and B . burgdorferi-infected animals . Interestingly , a blood meal containing B . burgdorferi resulted in significantly increased expression of ISARL in the nymphal tick guts ( p<0 . 0001 ) ( Figure 1A ) . This suggests that the presence of B . burgdorferi in the blood meal helps to stimulate tick metabolic activity and/or that ISARL may have an important role during B . burgdorferi colonization of the tick gut . Since ISARL expression was upregulated upon B . burgdorferi infection , we hypothesized that RNAi-mediated silencing of ISARL would affect B . burgdorferi colonization by nymphal I . scapularis . To this end , ISARL or GFP ( control ) dsRNA was injected into the guts of pathogen-free nymphs by anal pore injection . Then , the ticks were allowed to feed on B . burgdorferi-infected mice . Quantitative RT-PCR ( qPCR ) analysis showed a significant decrease of ISARL expression in the guts of ds ISARL-injected ticks ( p<0 . 01 ) when compared to that in control ds GFP-injected tick guts ( Figure 1B ) , indicating that the knockdown was successful . The engorgement weights of ds ISARL-injected nymphs and control ds GFP-injected nymphs were comparable ( p>0 . 05 ) ( Figure 1C ) , suggesting that silencing ISARL had no effect on tick feeding behavior . However , ISARL-silenced nymph guts showed a marked reduction of the B . burgdorferi burden ( p<0 . 001 ) when compared to that in control ticks ( Figure 1D ) , demonstrating that ISARL is associated with B . burgdorferi colonization in the nymphal tick gut . To determine whether ISARL might also play a role in the transmission of B . burgdorferi to the mammalian host , we silenced ISARL in B . burgdorferi-infected nymphs by microinjection of ds ISARL into the ticks . The results showed that B . burgdorferi burdens in the skin of mice ( ear skin distal from the tick bite site ) at 7 , 14 , and 21 days post tick detachment , and in heart and joint tissues 21 days post tick detachment were comparable ( p>0 . 05 ) in mice fed upon by ds GFP- or by ds ISARL-injected nymphs ( Figure 1E and F ) , suggesting that silencing ISARL had no observable effect on B . burgdorferi transmission by I . scapularis nymphs . To investigate the mechanisms underlying the association of ISARL with B . burgdorferi colonization by I . scapularis , we assessed the presence or absence of ISARL on tick physiology by comparing transcriptomes of ds ISARL and ds GFP ( control ) -injected ticks after engorgement on B . burgdorferi-infected or uninfected mice using RNA-seq . After feeding on uninfected mice , 18 genes were significantly differentially expressed in the guts of ds ISARL-injected nymphal ticks when compared to that in control ds GFP-injected tick guts ( Figure 2A; Supplementary file 1 ) , while 35 genes were differentially expressed after feeding on B . burgdorferi-infected mice ( Figure 2B; Supplementary file 1b ) . In particular , the ISARL gene was successfully silenced by RNAi as demonstrated by transcriptome analysis ( Supplementary file 1 ) and qPCR validation ( Figure 2C ) . No common genes except ISARL were observed between ticks feeding on uninfected or B . burgdorferi-infected mice ( Figure 2D ) , suggesting that the 34 genes ( Figure 2B; Supplementary file 1 ) were all altered by B . burgdorferi , or the influence of B . burgdorferi on the host blood components , rather than blood meal itself , in the absence of ISARL . In response to the blood meal , a significant change of the metabolic pathways in ticks was observed in the absence of ISARL . In particular , based on Gene Ontology ( GO ) functional classification and Kyoto Encyclopedia of Genes and Genomes ( KEGG ) pathways analyses , glutathione metabolism , including six gamma glutamyl transpeptidase genes ( Supplementary file 1 ) , was significantly altered in the absence of ISARL after engorgement of ticks on uninfected mice . Similarly , many metabolism-associated genes were significantly downregulated in the absence of ISARL after engorging on B . burgdorferi-infected mice ( Supplementary file 1 ) . GO functional classification and KEGG pathways also showed that the most downregulated genes were involved in fatty acid ( e . g . , 3-hydroxyacyl-CoA dehydrogenase ) , lipid and phospholipid ( e . g . , phosphatidylserine synthase I ) , and purine ( e . g . , GMP synthase ) metabolism pathways after silencing ISARL ( Figure 2E ) , suggesting that ISARL functions as a metabolic moderator in ticks . To further investigate the exact metabolism pathway ( s ) involved in B . burgdorferi colonization , we first selected 18 well-annotated and metabolism-related differentially expressed genes to validate the accuracy and reproducibility of the transcriptome bioinformatic analyses by qPCR . The samples for qPCR validation are independent of the sequencing samples . In general , the qPCR results indicated that all the tested genes showed concordant direction of change with the RNA-seq bioinformatic data except one gene , pyridoxine kinase ( PDXK ) ( Figure 2—figure supplement 1 ) , indicating the accuracy and reliability of our RNA-seq libraries . Of these 17 downregulated genes , 4 genes showed significant downregulation profiles ( p<0 . 05 ) . These four genes included phosphatidylserine synthase I ( PTDSS1 ) ( Figure 2F and G ) , N-CAM Ig domain-containing protein ( NCAM ) , vacuolar H+-ATPase V1 sector , subunit G ( V-ATPase ) , and sideroflexin 1 , 2 , 3 , putative ( SFXN ) ( Figure 2—figure supplement 2 ) . Then , we silenced these four genes individually and investigated their potential roles in B . burgdorferi colonization . We also silenced another four genes , whose p-values were very close to significant ( Figure 2—figure supplement 2 ) . These four genes included 3-hydroxyacyl-CoA dehydrogenase , putative ( 3HADH ) , adenylosuccinate synthetase ( ADSS ) , GMP synthase , putative ( GMPS ) , and alpha-actinin , putative ( ACTN ) . We did not observe a significant decrease of B . burgdorferi burden in nymphal tick guts after silencing NCAM , V-ATPase , SFXN , ADSS , GMPS , and ACTN compared to ds GFP-injected ticks ( Figure 2—figure supplement 3 ) . Instead , we found that PTDSS1-silenced nymphs showed a marked reduction in the B . burgdorferi burden in the guts when compared to that in control ticks ( p<0 . 05 ) ( Figure 2H ) . Furthermore , a blood meal containing B . burgdorferi resulted in significantly increased expression of PTDSS1 in the nymphal tick guts ( p<0 . 05 ) ( Figure 2I ) , suggesting that PTDSS1 indeed has a critical role during B . burgdorferi colonization of the tick gut . PTDSS1 is involved in phospholipid metabolism and mainly uses L-serine as the phosphatidyl acceptor to generate the anionic lipid phosphatidylserine ( PS ) , which serves as a precursor for phosphatidylethanolamine ( PE ) and phosphatidylcholine ( PC ) synthesis ( Figure 2J; Aktas et al . , 2014 ) . Importantly , PC is one of the main phospholipids on the cellular membrane of B . burgdorferi ( Kerstholt et al . , 2020 ) . However , B . burgdorferi lacks the central phospholipid metabolic enzymes . To further validate that the phospholipid metabolic pathway in tick is critical for B . burgdorferi , we silenced another enzyme ( ISARL-unrelated ) , phosphatidylserine decarboxylase ( PSD , ISCI003338 ) , which is an important enzyme in the synthesis of PE in both prokaryotes and eukaryotes ( Voelker , 1997 ) . Interestingly , we also found a significantly decreased B . burgdorferi burden in ds PSD-injected tick guts ( p<0 . 05 ) , and PSD and PTDSS1 elicit a similar degree of reduced B . burgdorferi levels ( Figure 2K ) . Taken together , ISARL-mediated phospholipid metabolic pathways associated with PTDSS1 have a critical role in B . burgdorferi colonization . We further explored how the ISARL signaling pathway is activated in ticks . Although the I . scapularis genome encodes an adiponectin receptor homolog , an adiponectin ligand is not present , at least in currently annotated Ixodes genome databases . This suggests that ticks may utilize vertebrate adiponectin to activate the adiponectin receptor during a blood meal , that tick have another ligand that stimulates the receptor , or both . Since ticks are habitually exposed to adiponectin present during a bloodmeal , we examined whether the tick adiponectin receptor could interact with incoming mammalian adiponectin during blood feeding . We injected recombinant mouse adiponectin into unfed ticks and investigated whether mammalian adiponectin could activate downstream signaling of tick adiponectin receptor by RNA-seq ( Figure 3A ) . The data showed that one classic downstream gene of mammalian adiponectin signaling , tick glucose-6-phosphatase ( G6p , ISCW017459 ) , was significantly downregulated in the presence of mammalian adiponectin ( Figure 3A; Supplementary file 1 ) . It has been demonstrated that in mammals the binding of adiponectin to its receptor suppresses G6p and phosphoenolpyruvate carboxykinase ( Pck ) expression through an AMP-activated protein kinase ( AMPK ) -dependent mechanism , which further inhibits glycogenolysis and gluconeogenesis ( Figure 3B; Kadowaki et al . , 2006 ) . We further searched G6p and Pck homologs in I . scapularis genome , and two G6p homologs ( ISCW017459 and ISCW018612 ) and three Pck homologs ( ISCW001902 , ISCW000524 , and ISCW001905 ) were identified . We designated them as G6pc1 , G6pc2 , Pck1 , Pck2 , and Pck3 , respectively . We evaluated gene expression of all these five genes after injection of recombinant adiponectin and GFP proteins . Interestingly , G6pc1 , G6pc2 , Pck2 , and Pck3 were significantly downregulated in the tick gut in the presence of adiponectin ( Figure 3C ) . To further validate the effects on tick glucose metabolism of interaction of mammalian adiponectin and tick ISARL , we fed ticks on C57BL/6J mice deficient in adiponectin ( Adipoq-/- ) and wild-type ( WT ) animals , and allowed them to feed to repletion ( Figure 3D ) . We then evaluated the expression of five G6p and Pck genes , and found that G6pc1 and G6pc2 also showed significant downregulation in the presence of adiponectin ( p<0 . 05 ) , while Pck gene expression was not altered ( p>0 . 05 ) ( Figure 3E ) . To investigate whether the interaction of adiponectin and the receptor in ticks influences B . burgdorferi colonization , pathogen-free nymphs were placed on B . burgdorferi-infected WT and Adipoq-/- mice and allowed to feed to repletion ( Figure 3F ) . No significant difference of the B . burgdorferi burden in ticks feeding on WT and Adipoq-/- mice was noted ( p>0 . 05 ) ( Figure 3G ) . We also silenced the G6pc1 and G6pc2 genes to determine whether G6p-mediated glucose metabolic changes affect B . burgdorferi colonization . Consistent with the previous observation , there was no significant difference in the B . burgdorferi burden between control and G6pc1-silenced ticks ( p>0 . 05 ) ( Figure 3—figure supplement 1 ) . G6pc2-silenced ticks also did not show altered B . burgdorferi levels ( p>0 . 05 ) ( Figure 3—figure supplement 1 ) . Furthermore , the expression of G6pc1 and G6pc2 in the nymphs was not influenced by B . burgdorferi infection ( p>0 . 05 ) ( Figure 3—figure supplement 1 ) , suggesting that G6pc1- or G6pc2-mediated changes do not affect B . burgdorferi colonization of the tick gut . To assess any changes in the adiponectin concentration in murine serum after B . burgdorferi infection , the mice were injected subcutaneously with 100 µL containing 1 × 104 or 1 × 107 B . burgdorferi , or PBS as a control . We found that B . burgdorferi does not influence the adiponectin concentration in murine blood ( Figure 3H ) . Taken together , these data suggest that mammalian adiponectin can regulate ISARL-mediated glucose metabolism pathway; however , it has no effect on B . burgdorferi colonization . We therefore examined whether I . scapularis protein ( s ) might interact with ISARL and whether B . burgdorferi could influence this process – for ISARL silencing diminished B . burgdorferi colonization . To this end , we performed a blastp search of the I . scapularis genome with the globular C1Q domain of human and mouse adiponectin , which is known to stimulate the adiponectin receptor ( Yamauchi et al . , 2002 ) . Two tick proteins had blastp hits with the human adiponectin C1Q domain ( Figure 4A ) and were annotated as complement C1q-like protein 3 ( C1QL3 ) ( GenBank number: XP_002415101 ) and conserved hypothetical protein ( GenBank number: EEC18766 ) , respectively . These are identical proteins except that C1QL3 has a signal peptide sequence , and we therefore focused on C1QL3 . We first examined whether expression of C1QL3 could be stimulated by B . burgdorferi infection . A blood meal containing B . burgdorferi resulted in significantly increased expression of C1QL3 in the nymphal tick guts ( p<0 . 01 ) ( Figure 4B ) . We then generated C1QL3-silenced nymphs and found that these ticks had a marked reduction of the B . burgdorferi burden in the guts when compared to that in control I . scapularis ( p<0 . 05 ) ( Figure 4C ) . This is the same observation as with silencing of ISARL , suggesting that B . burgdorferi activates the ISARL signaling pathway through the tick C1QL3 protein . Because C1QL3 C1Q domain has high similarity ( 64 . 0% ) with the human adiponectin C1Q domain ( Figure 4—figure supplement 1 ) , and C1Q proteins have been demonstrated to activate diverse pathways through the adiponectin receptor ( Zheng et al . , 2011 ) , we investigated whether tick C1QL3 could interact with ISARL . Human embryonic kidney HEK293T cells were transfected with the ISARL expression vector ( pEZT-ISARL ) . The results showed that tick ISARL can be successfully expressed , as validated by cell staining and western blot ( Figure 4D and E ) , on the HEK293T cell membrane ( Figure 4D ) . We then generated tick C1QL3 protein in a Drosophila expression system ( Figure 4F ) . The HEK293T cells were then incubated with the recombinant C1QL3 protein . After washing and staining , recombinant C1QL3 could be detected on the surface of ISARL-expressed rather than empty plasmid-transfected HEK293T cells ( Figure 4G ) . A pull-down assay also indicated that recombinant C1QL3 interacts with ISARL as demonstrated by the detection of C1QL3 only in ISARL-expressed cells pellet ( Figure 4H ) . In addition , co-immunolocalization demonstrated that the C1QL3 protein specifically binds to the ISARL-expressed cell membrane ( Figure 4I ) . Furthermore , C1QL3 also bound to tick ISE6 cells , a non-heterologous system ( Figure 4J ) . Since C1QL3 is a ligand of tick ISARL and also involved in Borrelia colonization , we further investigated whether C1QL3 has a role on the activation of ISARL by Borrelia . We first assessed if silencing of C1QL3 influenced ISARL expression after feeding on B . burgdorferi-infected mice ( Figure 4K ) . qPCR assessment showed that the ISARL transcript level following RNAi silencing of C1QL3 was significantly lower than that in control ds GFP-injected tick guts after feeding on B . burgdorferi-infected mice ( p<0 . 05 ) ( Figure 4L ) . More importantly , after silencing C1QL3 , a blood meal containing B . burgdorferi did not significantly increase expression of ISARL and PTDSS1 in the nymphal tick guts as compared to feeding on clean mice ( p>0 . 05 ) ( Figure 4M and N ) , further suggesting that C1QL3 plays a role in the ISARL signaling and phospholipid metabolism pathways . Adiponectin is a hormone , secreted mainly from adipocytes , that stimulates glucose utilization and fatty acid oxidation ( Berg et al . , 2001; Fruebis et al . , 2001 ) . The key roles of adiponectin in regulating energy homeostasis are mediated by adiponectin receptors across species including humans , yeast , nematodes , and flies ( Kwak et al . , 2013; Narasimhan et al . , 2005; Svensson et al . , 2011; Yamauchi et al . , 2003 ) . In this study , we have identified and characterized an adiponectin receptor homologue from I . scapularis , ISARL . ISARL shares significant sequence similarities with human , mouse , and Drosophila adiponectin receptors . In addition , ISARL contains the canonical features of adiponectin receptors , including conserved TM domains , a long internal N-terminal region , and a relatively short external C-terminal region . The highly conserved amino acids and the structures of ISARL and the receptor from D . melanogaster suggest that their ligands and signaling pathways may also be conserved in arthropods . However , homologs of adiponectin have not yet been identified in arthropods , suggesting that ligands for adiponectin receptors in arthropods may interact in different ways than in vertebrates . The Lyme disease agent , B . burgdorferi , engages in intimate interactions with I . scapularis during its acquisition and colonization of the tick gut ( Radolf et al . , 2012 ) . This is accompanied by dramatic changes in the expression profiles of Borrelia and tick gut genes , which are critical drivers for colonization , persistence , or transmission ( Kurokawa et al . , 2020; Narasimhan et al . , 2017 ) . In our study , expression of ISARL was significantly increased in the nymphal tick gut after B . burgdorferi infection . The upregulation of ISARL correlates with Borrelia infection in the gut . More interestingly , after silencing ISARL expression in the tick gut by anal pore injection , nymphal tick guts showed a marked reduction in the B . burgdorferi burden when compared to that in control ticks , demonstrating that ISARL facilitates B . burgdorferi colonization . We utilized RNA-seq to elucidate the pathways that are altered when ISARL is silenced in ticks that engorge on clean and B . burgdorferi-infected mice . Of note , ISARL can regulate a critical enzyme involved in phospholipid metabolism , PTDSS1 . Regulation of PTDSS1 by adiponectin receptors is also found in other organisms such as yeast , where the adiponectin receptor homolog Izh2 is connected to phospholipid metabolism through co-regulation of the expression of inositol-3-phosphate synthase ( INO1 ) and phosphatidylserine synthase ( CHO1 , homolog of PTDSS1 ) genes with zinc-responsive activator protein ( Zap1 ) ( Mattiazzi Ušaj et al . , 2015 ) . Silencing of I . scapularis PTDSS1 led to a reduced spirochete burden in the guts , thereby linking B . burgdorferi colonization with phospholipid metabolism . The B . burgdorferi genome is small and encodes a limited number of metabolic enzymes ( Casjens et al . , 2000; Fraser et al . , 1997 ) . The restricted biosynthetic capability forces B . burgdorferi to rely on its vertebrate and arthropod hosts for nutrients or enzymes that it cannot synthesize ( Tilly et al . , 2008 ) . Interestingly , we also found that silencing of I . scapularis 3HADH , which is involved in fatty acid metabolic processes , decreased the B . burgdorferi burden in tick gut ( Figure 2—figure supplement 3 ) . The markedly decreased B . burgdorferi burden in ticks after silencing of PTDSS1 , PSD , and 3HADH suggests that the spirochete may require the tick for selected metabolic needs . Indeed , previous studies have demonstrated that feeding ticks provide Lyme disease spirochetes with glycerol , an alternative carbohydrate energy source and essential building block for phospholipid biosynthesis ( Kerstholt et al . , 2020; Pappas et al . , 2011 ) . In addition , B . burgdorferi can also acquire lipids from the membranes of eukaryotic cells to which they are attached ( Crowley et al . , 2013 ) . We also found that B . burgdorferi can upregulate an adiponectin-related protein , C1QL3 , in ticks , which associates with ISARL and leads to phospholipid metabolism changes in ticks . We propose that C1QL3 in tick is mainly involved in metabolism , rather than complement activation , as demonstrated by the decreased B . burgdorferi level after silencing C1QL3 . Indeed , some of C1Q/TNF family proteins are associated with metabolism . In addition to adiponectin , proteins such as C1Q/TNF-related protein 3 ( CTRP3 ) , CTRP5 , CTRP9 , CTRP13 ( C1QL3 ) , and CTRP15 also belong to adipokine family and have been reported to be associated with the regulation of glucose , lipid , or other metabolisms ( Jiang et al . , 2018; Li et al . , 2017; Mi et al . , 2019; Wei et al . , 2011; Wolf et al . , 2016 ) . Importantly , C1Q proteins have been demonstrated to activate diverse pathways through adiponectin receptor ( Zheng et al . , 2011 ) . Additional efforts will investigate the mechanisms by which B . burgdorferi influence C1QL3 expression , and whether C1QL3 homologs in mammals such as adiponectin , CTRP13 , or other C1Q/TNF-related proteins may stimulate the tick C1QL3/ISARL pathway . Adiponectin receptors have diverse essential functions , and mutations in adiponectin receptors result in critical deficiencies . For instance , mutant of the AdipoR1 gene in retinal pigment epithelial cells results in the inability to take up and retain the essential fatty acid family member docosahexaenoic acid ( DHA , 22:6 , n-3 ) , further leading to photoreceptor cell death and retinal degeneration ( Rice et al . , 2015; Sluch et al . , 2018 ) . Adiponectin receptors are thought to have ceramidase activity ( Vasiliauskaité-Brooks et al . , 2017 ) , which is critical for cell survival through formation of antiapoptotic metabolite-sphingosine-1-phosphate ( S1P ) ( Holland et al . , 2011 ) . Whether targeting the tick adiponectin receptor signaling or the adiponectin pathway has the ability to influence human infection with B . burgdorferi remains to be determined . In summary , we demonstrate that ISARL plays a key role in metabolic pathways in I . scapularis . ISARL-mediated phospholipid metabolism by PTDSS1 contributes to B . burgdorferi colonization and an adiponectin-related protein , C1QL3 , is involved in ISARL signaling pathway . These studies elucidate a new pathway involved in tick metabolism and demonstrate that B . burgdorferi co-opts the activation of this pathway to facilitate colonization of I . scapularis . These processes are crucial to understanding the complex life cycle of the Lyme disease agent within ticks , and may be applicable to other arthropods and arthropod-borne infectious agents . C3H/HeJ mice , C57BL/6J mice WT , and C57BL/6J mice deficient in adiponectin ( Adipoq-/- ) were purchased from the Jackson Laboratory ( https://www . jax . org/strain/008195 ) . All mice were bred and maintained in a pathogen-free facility at Yale University . The spirochetes B . burgdorferi N40 were grown at 33°C in Barbour–Stoenner–Kelly H ( BSK-H ) complete medium ( Sigma-Aldrich , #B8291 ) with 6% rabbit serum . The live cell density was ~106–107 cells/mL as determined by dark field microscopy and hemocytometric analysis . To obtain B . burgdorferi-infected mice , the mice were injected subcutaneously with 100 µL of B . burgdorferi N40 ( 1 × 105 cells/mL ) . Two weeks after inoculation , the B . burgdorferi burden in mice was assayed by qPCR analysis of spirochete DNA in murine ear punch biopsies as described below . DNA was extracted from mouse skin-punch biopsies using the DNeasy tissue kit ( QIAGEN , #69506 ) according to the manufacturer’s protocol . The DNA was analyzed by qPCR using flagellinB ( flaB ) primers , and data were normalized to mouse actin . The primer sequences are shown in Supplementary file 1 . Pathogen-free I . scapularis larvae were acquired from the Centers for Disease Control and Prevention . The larval ticks were fed to repletion on pathogen-free C3H/HeJ mice and allowed to molt to nymphs . B . burgdorferi-infected nymphs were generated by placing larvae on B . burgdorferi-infected C3H/HeJ mice , and fed larvae were molted to nymphs . Nymphal ticks were maintained at 85% relative humidity with a 14 hr light and 10 hr dark period at 23°C . Human embryonic kidney HEK293T cells ( ATCC , #CRL-3216 ) and tick ISE6 cells ( ATCC , #CRL-11974 ) were used for vitro studies . The identity of the cells has been authenticated by ATCC , and no mycoplasma contamination . The human adiponectin receptor protein 1 ( GenBank number: NP_001277482 ) and 2 ( GenBank number: NP_001362293 ) sequences were used to conduct tblastn and blastp searches against the available black-legged tick database ( taxid:6945 ) using NCBI default parameters . Tick adiponectin receptor sequence was further validated by amplification with primers in Supplementary file 1 . Multiple alignment of protein sequences were performed using the Clustal Omega ( https://www . ebi . ac . uk/Tools/msa/clustalo/; Madeira et al . , 2019 ) or Uniprot ( https://www . uniprot . org/align/ ) . The similarities of adiponectin receptor protein sequences were measured in EMBOSS supermatcher ( https://www . bioinformatics . nl/cgi-bin/emboss/supermatcher ) . The protein structure of ISARL was predicted in SWISS-MODEL ( https://swissmodel . expasy . org/ Guex and Peitsch , 1997; Waterhouse et al . , 2018 ) . Hydrophobicity analysis was performed using ProtScale ( https://web . expasy . org/protscale/; Gasteiger et al . , 2005 ) . To evaluate gene expression of ISARL upon B . burgdorferi infection , pathogen-free I . scapularis nymphs were placed on B . burgdorferi-free and -infected mice ( C3H/HeJ ) . At least three mice were used in each experiment , and the ticks were allowed to feed to repletion . Both B . burgdorferi-free and -exposed tick guts were dissected under the dissecting microscope . The RNA from dissected guts was purified by TRIzol ( Invitrogen , #15596-018 ) following the manufacturer’s protocol , and cDNA was synthesized using the iScript cDNA Synthesis Kits ( Bio-Rad , #1708891 ) . qPCR was performed using iQ SYBR Green Supermix ( Bio-Rad , #1725124 ) on a Bio-Rad cycler with a program consisting of an initial denaturing step of 2 min at 95°C and 45 amplification cycles consisting of 20 s at 95°C followed by 15 s at 60°C , and 30 s at 72°C . The genes and corresponding primer sequences are shown in Supplementary file 1 . The specific target transcripts of ISARL and the reference gene tick actin were quantified by extrapolation from a standard curve derived from a series of known DNA dilutions of each target gene , and data were normalized to tick actin . Fed-nymph gut cDNA was prepared as described above and used as template to amplify segments of targeted genes . The PCR primers with T7 promoter sequences are shown in Supplementary file 1 . Double-stranded RNA ( dsRNA ) were synthesized using the MEGAscript RNAi kit ( Invitrogen , #AM1626M ) using PCR-generated DNA template that contained the T7 promoter sequence at both ends . The dsRNA quality was examined by agarose gel electrophoresis . DsRNA of the Aequorea victoria green fluorescent protein ( GFP ) was used as a control . Pathogen-free and -infected tick nymphs were injected in the anal pore with dsRNA ( 6 nL ) using glass capillary needles as described by Narasimhan et al . , 2004 . To examine the effects of silencing targeted genes on the colonization of B . burgdorferi in the tick gut , dsRNA microinjected pathogen-free I . scapularis nymphs were placed on B . burgdorferi-infected mice ( C3H/HeJ ) and allowed to feed to repletion . The ticks were then collected for gut dissection . The B . burgdorferi burden in the tick gut was quantified by amplifying flaB . FlaB was quantified by extrapolation from a standard curve derived from a series of known DNA dilutions of flaB gene , and data were normalized to tick actin . The knockdown efficiency of targeted genes was tested as described above . Specifically , the expression of targeted genes was estimated with the ΔΔCT method ( Schmittgen and Livak , 2008 ) using the reference gene actin . To test the effects of silencing ISARL on the transmission of B . burgdorferi , a group of 3–5 GFP or ISARL dsRNA-injected B . burgdorferi-infected nymphs were placed on each C3H/HeJ mouse ( at least five mice each in the GFP or ISARL dsRNA groups ) and allowed to feed to repletion . Ticks are placed on the mouse head/back between the ears . At 7 and 14 days post tick detachment , the mice were anesthetized , and skin was aseptically punch biopsied and assessed for spirochete burden by qPCR . Ticks feed in head area and skin punch biopsies are collected from the pinnae/ears . This site is considered distal as it is not at the site of tick bite . Twenty-one days post tick detachment , the mice were sacrificed , and ear skin , heart , and joints were aseptically collected and assessed for spirochete burden by qPCR . dsRNA ( ds ISARL and ds GFP ) microinjected pathogen-free I . scapularis nymphs were placed on clean and B . burgdorferi-infected mice ( C3H/HeJ ) , respectively , and allowed to feed to repletion . Then , the ticks were collected for gut dissection . Total RNA was purified as described above . In addition , to check the transcriptional alterations in the tick gut in the presence of mammalian adiponectin , pathogen-free tick nymphs were injected in the anal pore with approximately 12 ng recombinant mouse adiponectin ( MilliporeSigma , #SRP3297 ) and GFP proteins ( SinoBiological , #13105-S07E ) . The amount of injected protein was calculated based on the adiponectin concentration in mice blood ( ~3 μg/mL ) and nymphal tick engorgement ( ~4 mg ) . Then , the guts were dissected after 8 hr injection , and RNA was purified . The RNA samples were then submitted for library preparation using TruSeq ( Illumina , San Diego , CA ) and sequenced using Illumina HiSeq 2500 by paired-end sequencing at the Yale Centre for Genome Analysis ( YCGA ) . The I . scapularis transcript data were downloaded from the VectorBase ( https://vectorbase . org/vectorbase/app/ Giraldo-Calderón et al . , 2015 ) and indexed using the kallisto-index ( Bray et al . , 2016 ) . The reads from the sequencer were pseudo-aligned with the index reference transcriptome using kallisto ( Bray et al . , 2016 ) . The counts generated from three biological replicates each treatment were processed by DESeq2 ( Love et al . , 2014 ) in RStudio ( https://rstudio . com ) . The significant genes were then determined by the p-value and the adjusted p-value padj ( p<0 . 05 ) . The heatmaps of significant genes were also generated in RStudio . GO enrichment analysis and KEGG pathway enrichment analyses were conducted using the functional annotation tool DAVID 6 . 8 ( Huang et al . , 2009 ) . Tick ISARL gene was PCR amplified from nymph cDNA using the primer pair listed in Supplementary file 1 , then cloned into the XbaI and NotI sites of the pEZT-Dlux , a modified pEZT-BM vector ( Addgene , #74099 ) in-frame with a HA-tag sequence , by Gibson Assembly Cloning Kit ( NEB , #E5510S ) . The HEK293T cells were grown in Dulbecco’s Modified Eagle’s Medium ( DMEM , Thermo Fisher , #11965-118 ) media supplemented with 10% fetal bovine serum ( FBS , Sigma , #12306C-500 ) . HEK293T cells were transfected with the ISARL expression plasmid ( pEZT-ISARL ) using TransIT 2020 ( Mirus , #MIR5404 ) . After 40 hr post transfection , the cells were washed with 1× PBS and then incubated with 5 μg rC1QL3 protein ( 0 . 5 μg/μL ) with His/V5 tag , respectively . After 16 hr incubation with gentle agitation , the cells were washed with PBS and fixed in 4% PFA for 15 min at room temperature . Then , the cells were blocked in 1% BSA in PBS for 1 hr and subsequently immunolabeled with anti-HA antibody ( 1:100 , Cell Signaling Technology , #C29F4 ) for checking ISARL expression and V5 tag monoclonal antibody ( 1:100 , Invitrogen , # R960-25 ) for checking C1QL3 binding . Cells were washed with PBS three times and then immunolabeled with secondary antibodies of goat anti-rabbit IgG ( H + L ) Highly Cross-Adsorbed Secondary Antibody , Alexa Fluor 488 ( 1:100 , Invitrogen , #A-11034 ) and goat anti-mouse IgG ( H + L ) Cross-Adsorbed Secondary Antibody , Alexa Fluor 555 ( 1:100 , Invitrogen , #A-21422 ) for 1 hr at room temperature . Nuclei were stained with DAPI ( Invitrogen , #D9542 ) . After staining , the fluorescence signals were examined with an EVOS FL Auto Cell Imaging System ( Thermo Fisher Scientific ) . We also conducted plot profile to help analyze co-localization by ImageJ software . For checking ISARL expression by western blot , after 40 hr post transfection , the cells were washed with 1× PBS and then lysed with 4× Laemmli Sample Buffer ( Bio-Rad , #1610747 ) . After centrifuging at high speed , the supernatant was loaded to perform western blot as described below . HRP anti-His tag antibody ( 1:10 , 000 , Abcam , #ab3553 ) or anti-HA antibody ( 1:1000 , Cell Signaling Technology , #C29F4 ) was used to detect expression of ISARL . We conducted a pull-down assay to check the binding of ISARL and C1QL3 as described in Schuijt et al . , 2011a . Briefly , HEK293T cells were transfected as described above . After 40 hr post transfection , the cells were washed and suspended with 1× PBS and then incubated with rC1QL3 protein for 16 hr with gentle agitation , respectively . Then the cells were pelleted , and the pellet and supernatant were separated . The pellet was washed 5–8 times in 1 . 5 mL PBS/0 . 1% BSA and was resuspended in the same volume as the supernatant . The test of C1QL3 binding to tick ISE6 cells was conducted as described above . Equal volumes of supernatant and pellet were used to run western blot as described below . HRP V5-tag monoclonal antibody ( 1:1000 , Invitrogen , # R961-25 ) was used to detect protein . To assess the adiponectin concentration change in mice serum after B . burgdorferi infection , the C3H/HeJ mice were injected subcutaneously with 100 µL 1 × 104 and 1 × 107 cells/mL B . burgdorferi and PBS as a control ( five mice in each group ) . At 0 , 21 , and 28 days post inoculation , the blood was collected from mice . The sera were separated from mice blood samples by centrifugation at 1000× g for 10 min at 4°C . The adiponectin in mice serum was quantified by Mouse Adiponectin/Acrp30 Quantikine ELISA Kit ( R&D Systems , #MRP300 ) . Pathogen-free I . scapularis nymphs were placed on B . burgdorferi-infected WT and Adipoq-/- mice ( C57BL/6J ) and allowed to feed to repletion . The ticks were then collected for gut dissection . The B . burgdorferi burden in the tick gut was quantified as described above . The C1QL3 was PCR amplified from tick nymph cDNA using the primer pair listed in Supplementary file 1 , then cloned into the BglII and XhoI sites of the pMT/BiP/V5-His vector ( Invitrogen , #V413020 ) . The recombinant protein was expressed and purified using the Drosophila Expression System as described previously ( Schuijt et al . , 2011b ) . The protein was purified from the supernatant by TALON metal affinity resin ( Clontech , #635606 ) and eluted with 150 mM imidazole . The eluted samples were filtered through a 0 . 22 μm filter and concentrated with a 10 kDa concentrator ( MilliporeSigma , #Z740203 ) by centrifugation at 4°C . Recombinant protein purities were assessed by SDS-PAGE using 4–20% Mini-Protean TGX gels ( Bio-Rad , #4561094 ) and quantified using the BCA Protein Estimation kit ( Thermo Fisher Scientific , #23225 ) . Proteins were separated by SDS-PAGE at 160 V for 1 hr . Proteins were transferred onto a 0 . 45-m-pore-size polyvinylidene difluoride ( PVDF ) membrane ( Bio-Rad , #1620177 ) and processed for immunoblotting . The blots were blocked in 1% non-fat milk in PBS for 60 min . Primary antibodies of PTDSS1 rabbit pAb ( 1:1000 , Abclonal , #A13065 ) , anti-beta actin antibody ( 1:1000 , Abcam , #ab8224 ) , HRP anti-6X His tag antibody ( 1:10 , 000 , Abcam , #ab3553 ) , or HRP V5 tag monoclonal antibody ( 1:1000 , Invitrogen , # R961-25 ) were diluted in 0 . 05% PBST and incubated with the blots for 1 hr at room temperature or 4°C overnight . HRP-conjugated secondary antibody ( 1:2500 , Invitrogen , #62-6520 and #31466 ) was diluted in PBST and incubated for 1 hr at room temperature . After washing with PBST , the immunoblots were imaged and quantified with a LI-COR Odyssey imaging system . Statistical significance of differences observed in experimental and control groups was analyzed using GraphPad Prism version 8 . 0 ( GraphPad Software , Inc , San Diego , CA ) . Nonparametric Mann–Whitney test or unpaired t test were utilized to compare the mean values of control and tested groups , and p<0 . 05 was considered significant . The exact p-values are shown in the source data .	Many countries around the world are seeing an increase in the number of patients diagnosed with Lyme disease , with often serious joint , heart , and neurologic complications . This illness is caused by species of ‘spirochete’ bacteria that live and multiply inside black-legged ticks , and get injected into mammals upon a bite . Ticks are not simply ‘syringes’ however , and a complex relationship is established between spirochetes and their host . This is particularly true since Lyme disease-causing bacteria such as Borrelia burgdorferi rely on ticks to obtain energy and nutrients . Tang , Cao et al . delved into these complex interactions by focusing on the molecular cascades ( or pathways ) involving adiponectin , a hormone essential for regulating sugar levels and processing fats . Analyses of gene and protein databases highlighted that ticks carry a receptor-like protein for adiponectin but not the hormone itself , suggesting that an alternative pathway is at play . This may involve B . burgdorferi , which gets its fats and sugars from its host . And indeed , experiments showed that ticks produced more of the adiponectin receptor-like protein when they carried B . burgdorferi; conversely , silencing the receptor reduced the number of surviving spirochetes inside the tick . Further exploration showed that the receptor mediates molecular cascades that help to process fat molecules; these are associated with spirochete survival . In addition , the receptor-like protein was activated by C1QL3 , a ‘complement 1q domain-contained’ molecule which might be part of the tick energy-making or immune systems . Larger quantities of C1QL3 were found in ticks upon B . burgdorferi infection , suggesting that the spirochete facilitates an interaction that boosts activity of the adiponectin receptor-like protein . Overall , the work by Tang and Cao et al . revealed a new pathway which B . burgdorferi takes advantage of to infect their host and multiply . Targeting this molecular cascade could help to interfere with the life cycle of the spirochete , as well as fight Lyme disease and other insect-borne conditions .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "microbiology", "and", "infectious", "disease" ]	2021	The Lyme disease agent co-opts adiponectin receptor-mediated signaling in its arthropod vector
Ankyrin adaptors together with their spectrin partners coordinate diverse ion channels and cell adhesion molecules within plasma membrane domains and thereby promote physiological activities including fast signaling in the heart and nervous system . Ankyrins specifically bind to numerous membrane targets through their 24 ankyrin repeats ( ANK repeats ) , although the mechanism for the facile and independent evolution of these interactions has not been resolved . Here we report the structures of ANK repeats in complex with an inhibitory segment from the C-terminal regulatory domain and with a sodium channel Nav1 . 2 peptide , respectively , showing that the extended , extremely conserved inner groove spanning the entire ANK repeat solenoid contains multiple target binding sites capable of accommodating target proteins with very diverse sequences via combinatorial usage of these sites . These structures establish a framework for understanding the evolution of ankyrins' membrane targets , with implications for other proteins containing extended ANK repeat domains . Ankyrins are a family of scaffold proteins which play essential roles in connecting numerous ion channels , cell adhesion molecules , and receptors to the spectrin-based cytoskeleton beneath membranes and thereby provide mechanical support for plasma membranes and control the activities of excitable tissues including neurons and muscles ( Bennett and Chen , 2001; Bennett and Healy , 2009 ) . There are three members in the human ankyrin family: ankyrin-R/B/G ( AnkR/B/G ) encoded by ANK1/2/3 , respectively . They all consist of a highly similar N-terminal membrane binding domain composed of 24 ankyrin ( ANK ) repeats , a spectrin-binding domain comprised of two ZU5 domains , and a UPA domain followed by a death domain ( DD ) and a variable C-terminal regulatory domain ( Bennett and Lorenzo , 2013 ) ( Figure 1A ) . Although sharing similar domain organization , the three ankyrins have distinct and non-overlapping functions in specific membrane domains coordinated by ankyrin-spectrin networks ( Mohler et al . , 2002; Abdi et al . , 2006; He et al . , 2013 ) . As ankyrins are adaptor proteins linking membrane proteins to the underlying cytoskeleton , ankyrin dysfunction is closely related to serious human diseases . For example , loss-of-function mutations can cause hemolytic anemia ( Gallagher , 2005 ) , various cardiac diseases including several cardiac arrhythmia syndromes and sinus node dysfunction ( Mohler et al . , 2003 , 2007; Le Scouarnec et al . , 2008; Hashemi et al . , 2009 ) , bipolar disorder ( Ferreira et al . , 2008; Dedman et al . , 2012; Rueckert et al . , 2013 ) , and autism spectrum disorder ( Iqbal et al . , 2013; Shi et al . , 2013 ) . 10 . 7554/eLife . 04353 . 003Figure 1 . Identification of a 48-residue auto-inhibitory segment that binds to ANK repeats . ( A ) Schematic diagrams showing the domain organization of ankyrins . The AnkR-specific auto-inhibitory segment ( AS ) is indicated within the C-terminal regulatory domain . The same color codes ( 24 ANK repeats in rainbow and the AnkR_AS in magenta ) are used throughout the paper unless otherwise stated . ( B ) ITC-based mapping of the minimal AnkR_repeats binding region in the C-terminal regulatory domain . The minimal and complete AS identified is highlighted in magenta . ‘ND’ denotes that these constructs had no detectable binding to ANK repeats . ( C ) ITC-derived binding curve of AnkR_AS titrated to AnkR_repeats . ( D ) The binding affinities between AS and ANK repeats of the three ankyrin isoforms . DOI: http://dx . doi . org/10 . 7554/eLife . 04353 . 003 The wide-ranging physiological functions of ankyrins are the result of ankyrin's remarkable capacity for binding to and anchoring numerous membrane targets , via their N-terminal 24 ANK repeats , to specific membrane micro-domains in coordination with spectrin-based cytoskeletal structures ( Bennett and Chen , 2001 ) . One good example is the formation and maintenance of axon initial segments ( AIS ) in neurons . Interaction between AnkG and L1 cell adhesion molecules ( e . g . , the 186 kDa neurofascin , referred to as Nfasc in this study ) is required for the formation and stability of the AIS ( Hedstrom et al . , 2007; Zonta et al . , 2011 ) . AnkG is in turn responsible for clustering voltage-gated sodium channels at the AIS , which underlies the mechanistic basis of action potential generation and propagation ( Garrido et al . , 2003; Kole et al . , 2008 ) ( see review by Rasband , 2010 ) . Anchoring of L1 cell adhesion molecules also directs inhibitory GABAergic synapse innervation at the AIS of excitatory Purkinje neurons ( Ango et al . , 2004 ) , a critical step for balanced neuronal circuit formation . Depletion of AnkG both in cultured hippocampal neurons and in mice causes axons to lose axonal properties and acquire the molecular characteristics of dendrites , showing that AnkG is required for the maintenance of axonal polarity ( Hedstrom et al . , 2008; Sobotzik et al . , 2009 ) . Other examples of ankyrin function in organizing membrane signaling networks include , but are not limit to , AnkB/G-mediated coordination of voltage-gated sodium channels , Na/K ATPase , Na/Ca exchanger , and inositol 1 , 4 , 5-triphosphate receptors in cardiomyocytes ( Mohler et al . , 2004 , 2005; Hund et al . , 2008; Lowe et al . , 2008 ) and the dystrophin/dystroglycan complex in skeletal muscles ( Ayalon et al . , 2008 , 2011 ) . Finally , as originally discovered by Bennett and Stenbuck ( 1979a , 1979b ) , AnkR is well known to be essential for preserving erythrocyte membrane integrity . Although the critical functions of ankyrins in the specialized membrane domains have been recognized for decades , the underlying mechanistic basis governing ankyrin's coordination with such broad spectrum of membrane targets remains essentially unknown , largely due to challenges in characterizing the biochemical and structural properties of the elongated ANK repeats . Additionally , it is noted that the ANK repeats of ankyrins have been extremely conserved , whereas the membrane targets have continued to expand throughout evolution , presumably due to functional requirements for membrane microdomain-mediated fast signaling events in higher eukaryotes including mammals . In this study , we performed detailed biochemical characterizations of ANK repeats of ankyrins and their interactions with various binding partners . We solved the crystal structures of ANK repeats in complex with an auto-inhibitory segment from AnkR C-terminal domain and with a peptide from Nav1 . 2 , respectively . The 24 ANK repeats of ankyrins form a superhelical solenoid with an extremely conserved elongated inner groove , which contains multiple quasi-independent target binding sites . We further show that ankyrins can accommodate different membrane targets with diverse sequences by combinatorial usage of these binding sites . The ankyrin-Nav1 . 2 complex structure also provides a mechanistic explanation for the mutation found in Nav channels that causes cardiac disease in humans . Collectively , our findings provide a first glimpse into the mechanistic basis governing membrane target recognition by the highly conserved ANK repeats in ankyrins and establish a structural framework for future investigation of ankyrin's involvement in physiological functions and pathological conditions in diverse tissues . Our results also provide a molecular mechanism for the rapid expansion of ankyrin partners in vertebrate evolution . These insights also will be valuable for understanding the recognition mechanisms of other long ANK repeat proteins as well as many other long repeat-containing proteins in living organisms in general . To elucidate the mechanisms governing ANK repeat-mediated binding of ankyrins to diverse membrane targets , we attempted to determine the atomic structures of ANK repeats alone or in complex with their targets . However , extensive trials of crystallizing ANK repeat domains of AnkR/B/G were not successful , presumably because of the highly dynamic nature of the extended ANK repeat solenoid ( Howard and Bechstedt , 2004; Lee et al . , 2006 ) . Anticipating that ANK repeats binders may rigidify the conformation of ANK repeats , we turned our attention to the ANK repeat/target complexes . The C-terminal regulatory domains have been reported to bind to ANK repeats intra-molecularly and modulate the target binding properties of ankyrins ( Davis et al . , 1992; Abdi et al . , 2006 ) . We measured the interaction of AnkR_repeats with its entire C-terminal regulatory domain ( residues 1529–1907 ) using highly purified recombinant proteins , and found that they interact with each other with a Kd of around 1 µM ( Figure 1B ) . It is expected that the intra-molecular association between ANK repeats and its C-terminal tail of AnkR is very stable , and thus the full-length AnkR likely adopts an auto-inhibited conformation and ANK repeats-mediated binding to membrane targets requires release of the auto-inhibited conformation of AnkR . Using isothermal titration calorimetry ( ITC ) -based quantitative binding assays , we identified a 48-residue auto-inhibitory segment ( residues 1577–1624 , referred to as ‘AS’ ) as the complete ANK repeat-binding region ( Figure 1B , C ) . Further truncation at either end of this 48-residue AS fragment significantly decreased its binding to AnkR_repeats ( Figure 1B ) . The corresponding sequence does not exist in AnkB or AnkG , indicating the AS is specific to AnkR ( Figure 1A ) . AnkR_AS was found to bind to AnkR/B/G ANK repeats with comparable affinities ( Figure 1D ) , as expected since AnkR/B/G share extremely conserved ANK repeat sequences ( Figure 2B and see below ) . Thus , we tried the complexes of AnkR_AS with ANK repeats of all three isoforms to increase the chances of obtaining suitable crystals . Although crystals of various complexes were obtained , they all diffracted very poorly . After extensive trials of screening and optimization , we succeeded in obtaining good-diffraction crystals of AnkR_AS fused at its C-terminus with the AnkB_repeats and solved the structure of the fusion protein at 3 . 5 Å resolution ( Figure 2C and Table 1 ) . The NMR spectra of the 13CH3-Met selectively labeled fusion protein and the ANK repeats/AS complex produced by cleavage of the fusion protein at the fusion site are essentially identical ( Figure 2—figure supplement 1 ) , indicating that the fusion strategy used here facilitates crystallization but does not alter the structure of the ANK repeats/AS complex . There are three Met residues in AS ( Met1601 , Met1604 , and Met1607 ) and all three Met residues are in the binding interface between ANK repeats and AS ( Figure 2—figure supplement 2A ) . 10 . 7554/eLife . 04353 . 004Figure 2 . Vertebrate ANK repeats of ankyrins share the same architecture and target binding properties . ( A ) Sequence alignment of the 24 ANK repeats of human AnkB . Similar and identical residues are labeled gray and black , respectively . The helix formation residues are boxed with corresponding colors . The hydrophobic residues selected for mutation studies described in Figure 3 and onwards are labeled with corresponding colors . The last nine amino acids labeled in red from R24 are used as the C-terminal capping sequence for designed truncation mutants of various lengths of ANK repeats used in this study . ( B ) Sequence conservation map of the 24 ANK repeats of vertebrate ankyrins . The conservation score for each residue is calculated based on the sequences of vertebrate ankyrins aligned in Figure 2—figure supplement 3 through the Scorecons server ( http://www . ebi . ac . uk/thornton-srv/databases/cgi-bin/valdar/scorecons_server . pl ) . The position of each residue is the same as that shown in panel A . ( C ) Overall structure of the ANK repeats/AS complex viewed from the top ( left ) and side ( right ) . The three AS-binding surfaces on ANK repeats are circled with black dashed ovals . The sequences of AnkR_AS are listed below . ( D ) Surface conservation map of ANK repeats viewed from the side . The conservation map is derived from the ankyrins from worm to human as shown in Figure 2—figure supplement 3 with the same color coding scheme as in panel ( B ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04353 . 00410 . 7554/eLife . 04353 . 005Figure 2—figure supplement 1 . The fusion of AnkR_AS to the N-terminus AnkB_repeats does not alter the conformation of the ANK repeats/AS complex . ( A ) 1H-13C HSQC spectrum showing the 13CH3-Met-labeled , covalently linked AnkR_AS-AnkB_repeats fusion protein . The resolved Met methyl groups are labeled with blue asterisks . ( B ) Superposition plot of the 1H-13C HSQC spectra of the covalently linked ( red ) and thrombin-cleaved ANK repeats/AS complex , showing that the two spectra are essentially identical . The data also indicate that the fusion of AS to the N-terminal of ANK repeats , a step necessary for obtaining crystals of the complex , does not alter the conformation of the complex . DOI: http://dx . doi . org/10 . 7554/eLife . 04353 . 00510 . 7554/eLife . 04353 . 006Figure 2—figure supplement 2 . Crystallographic characterization of the ANK repeats/AS structure . ( A ) Electron density ( 2Fo-Fc ) map of ANK repeats/AS structure superimposed on the Cα model . The map is contoured at 1 . 5 σ . Insert , the Se-anomalous difference map contoured at 4 σ shows four Se peaks around R7–9 of the ANK repeats , indicating that three Met residues ( Met1601 , Met1604 , and Met1607 ) of AnkR_AS are located at site 2 of the ANK repeats , which also contains a Met ( Met338 ) . ( B ) The 2Fo-Fc map of AnkR_AS contoured at 1 σ with the final structural model superimposed . The densities of three selected side-chain interactions from the AS and ANK repeats are shown in the inserts . DOI: http://dx . doi . org/10 . 7554/eLife . 04353 . 00610 . 7554/eLife . 04353 . 007Figure 2—figure supplement 3 . Amino acid sequence alignment of ANK repeats of ankyrins . In this alignment , residues that are absolutely conserved and highly conserved are highlighted in red and yellow , respectively . The secondary structural elements are indicated above the alignment . The interface residues at sites 1 , 2 , and 3 for the binding of the ANK repeats to AnkR_AS are labeled with ‘1’ , ‘2’ , and ‘3’ , respectively . The interface residues involved in the ANK repeats/Nav1 . 2_ABD-C interaction are marked by triangles . The vertebrate ankyrin ANK repeats are aligned together , and used to derive the ANK repeats conservation plot shown in Figure 2B . DOI: http://dx . doi . org/10 . 7554/eLife . 04353 . 00710 . 7554/eLife . 04353 . 008Table 1 . Statistics of data collection and model refinementDOI: http://dx . doi . org/10 . 7554/eLife . 04353 . 008Native ANK repeats/ASSeMet-ANK repeats/ASR1-9/Nav1 . 2_ABD-CData collection Space groupR32R32P4222 Cell dimensions a , b , c ( Å ) 179 . 9 , 179 . 9 , 304 . 5179 . 7 , 179 . 7 , 304 . 9102 . 3 , 102 . 3 , 106 . 0 α , β , γ ( ° ) 90 , 90 , 12090 , 90 , 12090 , 90 , 90 Resolution range ( Å ) 50–4 . 0 ( 4 . 07–4 . 0 ) 50–3 . 5 ( 3 . 56–3 . 5 ) 50–2 . 5 ( 2 . 54–2 . 5 ) Rmerge ( % ) 8 . 7 ( 45 . 8 ) 12 . 1 ( 78 . 3 ) 7 . 7 ( 74 . 8 ) I/σI17 . 1 ( 3 . 4 ) 22 . 5 ( 2 . 2 ) 29 . 8 ( 3 . 5 ) Completeness ( % ) 98 . 9 ( 99 . 3 ) 96 . 0 ( 97 . 2 ) 99 . 4 ( 100 ) Redundancy4 . 3 ( 4 . 4 ) 10 . 2 ( 9 . 0 ) 9 . 5 ( 9 . 7 ) Structure refinement Resolution ( Å ) 50–3 . 5 ( 3 . 62–3 . 5 ) 50–2 . 5 ( 2 . 64–2 . 5 ) Rcryst/Rfree ( % ) †22 . 0 ( 35 . 0 ) /25 . 3 ( 36 . 6 ) 18 . 8 ( 22 . 7 ) /23 . 8 ( 24 . 5 ) r . m . s . d . bonds ( Å ) /angles ( ° ) 0 . 013/1 . 50 . 015/1 . 5 Average B factor113 . 563 . 5 No . of atoms Protein atoms62602243 Water molecules074 Other molecules4557 Ramachandran plot‡ Favored regions ( % ) 94 . 797 . 7 Allowed regions ( % ) 5 . 22 . 3 Outliers ( % ) 0 . 10 . 0Rmerge = Σ\|Ii − Im\|/ΣIi , where Ii is the intensity of the measured reflection and Im is the mean intensity of all symmetry related reflections . †Rcryst = Σ\|\|Fobs\| − \|Fcalc\|\|/Σ\|Fobs\| , where Fobs and Fcalc are observed and calculated structure factors , respectively . Rfree = ΣT\|\|Fobs\| − \|Fcalc\|\|/ΣT\|Fobs\| , where T is a test data set of about 5% of the total reflections randomly chosen and set aside prior to refinement . ‡Defined by MolProbity . Numbers in parentheses represent the value for the highest resolution shell . Except for a few connecting loops and termini of the chains , the rest of the ANK repeats and AS are properly defined ( Figure 2C and Figure 2—figure supplement 2 ) . The 24 ANK repeats form a left-handed helical solenoid with each repeat rotating anti-clockwise by ∼16° ( Figure 2C ) . Except for the capping helices in the first and last repeats ( i . e . , αA of R1 and αB of R24 ) , each repeat has the typical ANK repeat sequence pattern and forms a helix-turn-helix conformation ( Figure 2A , C ) . A well-defined finger-like hairpin loop ( finger loop ) connects two consecutive repeats . The inner αA helices and the finger loops of the 24 repeats line together to form an elongated concave inner groove , and the αB helices of the repeats form the solvent-exposed convex outer surface . The ANK repeats superhelix has outer and inner diameters of approximately 60 Å and 45 Å , respectively , and a total height of ∼150 Å ( Figure 2C ) . The size of the ANK repeats revealed here is consistent with the previous measurement by atomic force microscopy ( Lee et al . , 2006 ) . The C-terminal half of the ANK repeats structure aligns well with the apo-form structure of the last 12 ANK repeats of AnkR with an overall r . m . s . d . of 1 . 6 Å ( Michaely et al . , 2002 ) . We analyzed the amino acid residues at each position of vertebrate AnkR/B/G ANK repeats and found that conservation is above 80% at most of the positions ( Figure 2B and Figure 2—figure supplement 3 ) . Further analysis reveals that residues forming the target binding concave inner groove ( i . e . , residues of the finger loops and αA helices of the 24 repeats ) are essentially identical among vertebrate AnkR/B/G ( Figure 2B and Figure 2—figure supplement 3 ) , indicating that both the structure and the target binding properties of their ANK repeats are likely to be the same ( also see Figure 1D ) . Additionally , the residues in the entire inner groove of the ANK repeats superhelix are highly conserved for all ankyrins throughout evolution ( from worm to human ) ( Figure 2D and Video 1 ) , suggesting that the functions of ANK repeats in different species of ankyrins are highly conserved during evolution and that the inner groove of ANK repeats is the general binding site for membrane-associated targets of ankyrins . Consistent with this prediction , binding of AS to AnkG_repeats prevents voltage-gated sodium channel Nav1 . 2 and Nfasc from binding to AnkG ( Figure 3—figure supplement 1 ) . Therefore , we hypothesized that the ANK repeats/AS structure presented here serves as a general framework for understanding how ankyrins engage their membrane targets , and tested this hypothesis using mutations designed and tested as described below . 10 . 7554/eLife . 04353 . 009Video 1 . Surface conservation of 24 ANK repeats . This video shows the concave groove is highly conserved across various species from human to worm . DOI: http://dx . doi . org/10 . 7554/eLife . 04353 . 009 Before binding to ANK repeats , AS adopts a random coil structure as indicated by its NMR spectrum ( data not shown ) . In the complex , AS adopts a highly extended structure binding to part of the inner groove formed by the N-terminal 14 ANK repeats ( R1–14 ) with its chain orientation anti-parallel to that of ANK repeats ( Figure 2A , C ) . A 10-residue segment of AS ( residues 1592–1601 ) forms an α helix when bound to ANK repeats ( Figure 2C ) . The residues connecting AS and ANK repeats ( 10 residues in total , ‘GSLVPRGSGS’ ) are flexible , indicating that the fusion of the two chains together does not introduce obvious conformational restraints to the complex . The binding of AS to ANK repeats can be divided somewhat arbitrarily into three sites ( sites 1 , 2 , and 3 ) formed by the repeats 2–6 , 7–10 , and 11–14 , respectively ( Figure 2C and Figure 3A–C ) . Nonetheless , this division is supported by several lines of evidence . Structurally , there is a fairly clear boundary between each of the two binding sites in the ANK repeats/AS complex structure , whereas the interactions within each site are rather concentrated ( Figure 3 ) . The most direct evidence is from the interaction between ANK repeats and Nav1 . 2 ( see below ) . In the case of Nav1 . 2 binding , R1–6 of ANK repeats binds to the C-terminal half of the Nav1 . 2_ABD ( ankyrin binding domain ) and R11–14 binds to the N-terminal half of Nav1 . 2_ABD . R7–10 is not involved in the Nav1 . 2 binding . Thus , one can naturally divide ANK repeats R1–14 into three parts . Such division is further supported by the accepted concept that four to five ANK repeats can form a folded structural unit . In our case , sites 2 and 3 contain four repeats each , and site 1 contains five repeats if we do not count the repeat 1 which serves as a capping repeat . 10 . 7554/eLife . 04353 . 010Figure 3 . Structural and biochemical characterizations of target binding properties of ANK repeats . ( A–C ) Stereo views showing the detailed ANK repeats/AS interfaces of the three binding sites shown in Figure 1E . Hydrogen bonds and salt bridges are indicated by dashed lines . ( D ) Cartoon diagram of the first 14 repeats of the 24 ANK repeats . Different truncations used for the biochemical analyses are indicated below . Mutations of hydrophobic residues in the three AS binding sites are labeled . Red stars indicate the locations of the mutation sites . ( E ) Example ITC curves showing the bindings of Nav1 . 2_ABD or Nfasc_ABD to the wild-type or mutant ANK repeats . ( F ) The dissociation constants of the binding reactions of various mutants of ANK repeats to Nav1 . 2 and Nfasc derived from the ITC-based assays . DOI: http://dx . doi . org/10 . 7554/eLife . 04353 . 01010 . 7554/eLife . 04353 . 011Figure 3—figure supplement 1 . Analytical gel filtration analyses showing that binding of AS to AnkG_repeats prevents Nav1 . 2 and Nfasc ABDs from binding to AnkG_repeats . The AS was fused to the N-terminus of AnkG_repeats . No complex peaks formed between the fusion protein and Nav1 . 2 ABD ( A ) or Nfasc ABD ( B ) , as the mixture of the two proteins in each experiment did not change their elution volumes compared to that of each isolated protein . For clarity , the curve of the mixture in ( B ) was up-shifted by 10 mAU . DOI: http://dx . doi . org/10 . 7554/eLife . 04353 . 01110 . 7554/eLife . 04353 . 012Figure 3—figure supplement 2 . ITC-based analyses of the AnkG_repeats/Nfasc_ABD interaction . ( A ) Schematic diagram showing the domain organization of the L1-family cell adhesion molecules ( L1CAMs ) . The labeled amino acid numbers correspond with mouse Nfasc used in this study . ( B ) ITC titration of ABD ( 1187–1214 ) to AnkG_repeats . ( C ) ITC titration of the entire cytoplasmic tail ( 1132–1240 ) to AnkG_repeats , showing that the segment corresponding to 1187–1214 encompasses the complete binding domain for AnkG_repeats . DOI: http://dx . doi . org/10 . 7554/eLife . 04353 . 01210 . 7554/eLife . 04353 . 013Figure 3—figure supplement 3 . The ITC curves of the bindings of various ANK repeats to Nav1 . 2_ABD . The red asterisks in panels A , B , and C denotes AnkG_repeats with point mutations as indicated“ after ”The ITC curves of the bindings of various ANK repeats to Nav1 . 2_ABD . The ANK repeat protein assayed in each experiment is indicated in each panel . The curves were used to derive the dissociation constants shown in Figure 3F . DOI: http://dx . doi . org/10 . 7554/eLife . 04353 . 01310 . 7554/eLife . 04353 . 014Figure 3—figure supplement 4 . The ITC curves of the bindings of various ANK repeats to Nfasc_ABD . The red asterisks in panels A , B , and C denotes AnkG_repeats with point mutations as indicated . The ANK repeat protein assayed in each experiment is indicated in each panel . The curves were used to derive the dissociation constants shown in Figure 3F . DOI: http://dx . doi . org/10 . 7554/eLife . 04353 . 014 The interactions in site 1 are primarily charge–charge and hydrogen bonding in nature , although hydrophobic contacts also contribute to the binding ( Figure 3A ) . The interactions in site 2 are mediated both by hydrophobic and hydrogen bonding interactions , while interactions in site 3 are mainly hydrophobic ( Figure 3B , C ) . The structure of the ANK repeats/AS complex is consistent with the idea that ANK repeats bind to relatively short and unstructured peptide segments in ankyrins' membrane targets ( Bennett and Healy , 2009; Bennett and Lorenzo , 2013 ) . We next examined the interactions of AnkG_repeats with Nav1 . 2 and Nfasc using the structure of the ANK repeats/AS complex to design mutations specifically affecting each predicted site . The Kd of the binding of AnkG_repeats to the Nav1 . 2_ABD ( residues 1035–1129 , comprising the majority of the cytoplasmic loop connecting transmembrane helices II and III , see below for details ) and to the Nfasc_ABD ( a 28-residue fragment in the cytoplasmic tail; Figure 3—figure supplement 2 and see Garver et al . , 1997 ) is 0 . 17 and 0 . 21 μΜ , respectively ( Figure 3E , upper panels ) . To probe the binding sites of Nav1 . 2 and Nfasc on AnkG , we constructed AnkG_repeat mutants with the corresponding hydrophobic residues in binding site 1 ( Phe131 and Phe164 in R4 and R5 , termed ‘FF’ ) , site 2 ( Ile267 and Leu300 in R8 and R9; ‘IL’ ) , and site 3 ( Leu366 , Phe399 , and Leu432 in R11 , R12 , and R13; ‘LFL’ ) substituted with Gln ( Figure 3D ) , and examined their binding to the two targets . The mutations in site 1 significantly decreased ANK repeat binding to Nav1 . 2 , but had no impact on Nfasc binding . Conversely , the mutations in site 2 had minimal impact on Nav1 . 2 binding , but significantly weakened Nfasc binding . The mutations in site 3 weakened ANK repeat binding to both targets ( Figure 3F , Figure 3—figure supplement 3 and Figure 3—figure supplement 4 ) . The above results indicate that the two targets bind to ANK repeats with distinct modes , with Nav1 . 2 binding to sites 1 and 3 and Nfasc binding to sites 2 and 3 . This conclusion is further supported by the binding of the two targets to various AnkG_repeat truncation mutants ( Figure 3F , Figure 3—figure supplement 3 and Figure 3—figure supplement 4 ) . We have also assayed the impact of the mutations of the three sites on the binding of AnkR_AS to ANK repeats . The mutations in sites 1 and 2 led to ∼20-fold decrease in AnkR_AS binding , while the site 3 mutation only caused an approximately threefold decrease in AnkR_AS binding ( Figure 4A ) . Finally , we tested the binding of another two reported ankyrin targets , the KCNQ2 potassium channel ( Pan et al . , 2006 ) and the voltage-gated calcium channel Cav1 . 3 ( Cunha et al . , 2011 ) , to the ANK repeats and its mutants , and found that KCNQ2 mainly binds to sites 1 and 2 , and Cav1 . 3 primarily relies on site 2 of ANK repeats ( Figure 4B , C ) . Taken together , the above biochemical analysis plus the structure of the ANK repeats/AS complex reveals that through combinations of multiple binding sites on the extremely conserved and elongated inner groove formed by the 24 ANK repeats , ankyrins can bind to numerous targets with diverse amino acid sequences . It is likely that some ankyrin targets may bind to the groove formed by the rest of the repeats in addition to R1–14 . 10 . 7554/eLife . 04353 . 015Figure 4 . Fluorescence polarization-based measurement of the binding affinities of different targets to AnkB_repeats WT and its mutants . ( A ) Fluorescence polarization-based measurement of the binding affinities of AnkR_AS peptide to AnkB_repeats WT and its mutants . The insert shows the expanded view of the binding curves of the AnkR_AS peptides to WT and LFL of AnkB_repeats . The binding affinity between AnkR_AS and AnkB_repeats WT measured through this experiment is slightly different from the ITC assay ( 0 . 14 μΜ vs 0 . 40 μΜ ) . This may be because of the different measuring system , but the overall affinity range is quite similar . ( B ) Fluorescence polarization-based measurement of the binding affinities of the KCNQ2 peptide to AnkB_repeats WT and its various mutants . ( C ) Fluorescence polarization-based measurement of the binding affinities of the Cav1 . 3 peptide to AnkB_repeats and its various mutants . The fitted binding affinities are shown within the corresponding figures . DOI: http://dx . doi . org/10 . 7554/eLife . 04353 . 015 To further delineate the target binding mechanisms of ankyrins , we characterized the interaction between AnkG_repeats and Nav1 . 2 in detail . Previous studies have reported that the intracellular loop connecting the transmembrane helices II and III ( loop 2 ) is responsible for targeting Nav1 . 2 to the AIS via directly binding to AnkG , and identified a 27-residue motif within loop 2 ( ‘ABD-C’ , indicated in Figure 5A , D ) as the AnkG binding domain ( Garrido et al . , 2003; Lemaillet et al . , 2003 ) . First , we confirmed that a 95-residue fragment ( ABD , residues 1035–1129; Figure 5D ) is sufficient for binding to AnkG ( Figure 3E , upper left panel ) . Surprisingly , we found that the C-terminal part of the ABD ( ABD-C , the 27-residue motif identified previously for ANK repeats binding ) binds to ANK repeats with an affinity ∼15-fold weaker than the entire ABD , indicating that the ABD-C is not sufficient for binding to ANK repeats ( Figure 5B , C ) . Consistent with this observation , the N-terminal 68-residue fragment of loop 2 ( ABD-N , residues 1035–1102 ) also binds to ANK repeats , albeit with a relatively weak affinity ( Kd of ∼8 μΜ; Figure 5B , C ) . We further showed that the ABD-C fragment binds to repeats 1–6 ( R1–6 ) of ANK repeats , as ABD-C binds to R1–6 and the entire 24 ANK repeats with essentially the same affinities ( Figure 5B , C ) . These results also reveal that , like the AnkR_AS , the Nav1 . 2 peptide segment binds to ANK repeats in an anti-parallel manner . Taken together , the biochemical data shown in Figure 3E and Figure 5 indicate that two distinct fragments of Nav1 . 2 loop 2 , ABD-N and ABD-C , are responsible for binding to ANK repeats . The previously identified ABD-C binds to site 1 and ABD-N binds to site 3 of ANK repeats , and the interactions between the two sites are largely independent from each other energetically . 10 . 7554/eLife . 04353 . 016Figure 5 . Characterization of the interaction between Nav1 . 2 and AnkG_repeats . ( A ) Schematic diagram showing the domain organization of the Nav1 family ion channels . The ABD is located within loop 2 linking the transmembrane helices II and III and separated into N and C parts according to the data below . ( B ) Table summarizing the ITC-derived affinities of the bindings of various loop 2 fragments to AnkG_repeats . ( C ) ITC curves of the bindings of Nav1 . 2_ABD* ( upper left ) , ABD-N ( upper right ) , and ABD-C ( lower left ) to ANK repeats , and Nav1 . 2_ABD-C binding to ANK repeats R1–6 ( lower right ) , showing that ABD-C binds to site 1 of AnkG_repeats . ( D ) Amino acid sequence alignment of the ankyrin binding domains ( ABD ) of members of the voltage-gated sodium channel α-subunits ( Nav1 ) family . The mouse Nav1 . 2 used in this study was aligned with the human family members . Residues that are absolutely conserved and highly conserved are highlighted in red and yellow , respectively . The critical Glu1112 for the binding of Nav1 . 2 to the ANK repeats is indicated with a star . Other residues participating in the binding with the ANK repeats are indicated by triangles . The residues responsible for binding to site 1 of AnkG_repeats are completely conserved in all members of the Nav1 family , indicating that all sodium channels can bind to ankyrins following the mode revealed in this study . DOI: http://dx . doi . org/10 . 7554/eLife . 04353 . 016 We noted from the amino acid sequence alignment of the Nav1 members that the sequences of ABD-C ( the first half in particular ) are much more conserved than those of ABD-N ( Figure 5D ) . Further mapping experiments showed that the C-terminal less-conserved 10 residues of ABD-C are not essential for Nav1 . 2 to bind to ANK repeats ( Figure 5B , top two rows ) . Truncations at the either end of Nav1 . 2 ABD-N weakened its binding to ANK repeats ( data not shown ) , indicating that the entire ABD-N is required for the channel to bind to site 3 of ANK repeats . The diverse ABD-N sequences of Nav1 channels fit with the relatively non-specific hydrophobic-based interactions in site 3 observed in the structure of ANK repeats/AS complex ( Figure 3C ) . Although with very low amino acid sequence similarity , the Nav1 . 2_ABD-C ( as well as the corresponding sequences from Nav1 . 5 , KCNQ2/3 potassium channels , and β-dystroglycan [Mohler et al . , 2004; Pan et al . , 2006; Ayalon et al . , 2008] ) and the site 1 binding region of AnkR_AS share a common pattern with a stretch of hydrophobic residues in the first half followed by a number of negatively charged residues in the second half ( Figure 6C ) . Based on the structure of the ANK repeats/AS complex , we predicted that the Nav1 . 2_ABD-C may also bind to site 1 of AnkG_repeats with a pattern similar to the AS peptide . We verified this prediction by determining the structure of a fusion protein with the first nine ANK repeats of AnkB fused at the C-terminus of Nav1 . 2_ABD-C at 2 . 5 Å resolution ( Figure 6A , Figure 6—figure supplement 1 and Table 1; the ANK repeats/the entire ABD complex crystals diffracted very poorly , presumably because of the flexible nature of the interaction between Nav1 . 2_ABD-N and site 3 of ANK repeats ) . 10 . 7554/eLife . 04353 . 017Figure 6 . Site 1 of ANK repeats is a common binding site for Nav1 . 2 and other targets . ( A ) Ribbon representation of the binding of Nav1 . 2_ABD-C ( light green ) to site 1 of ANK repeats ( cyan ) . The interface residues are shown in the stick mode . Hydrogen bonds and salt bridges are indicated by dashed lines . The negatively charged Glu1112 , critical for interacting with a positively charged surface formed by ANK repeats R2 and R3 , is highlighted with a red box . ( B ) Charge potential surface of site 1 on the ANK repeats reveals a positively charged pocket important for anchoring of Glu1112 through charge complementation . The hydrophobic groove and the interacting residues from Nav1 . 2 are also shown . The surface diagram is drawn with the same orientation as in panel A . The electrostatic surface potentials were calculated by the APBS module embedded in PyMOL with the non-linear Poisson–Boltzmann equation and contoured at ±5 kT/e . ( C ) Amino acid sequence alignment of the site 1 binding sequences in various partners showing the similar sequence pattern , with the anchoring Glu boxed . The residues participating in site 1 binding are indicated by triangles . ( D ) Summary of the ITC-derived Kd values showing that Glu1112 is essential for the ANK repeats/Nav1 . 2 interaction . ( E ) Structural comparison of the ANK repeats site 1 bindings of AnkR_AS and Nav1 . 2_ABD-C showing that the two targets bind to ANK repeats with essentially the same mode . Subtle conformational differences in the finger loops R4 and R5 are indicated by arrows . DOI: http://dx . doi . org/10 . 7554/eLife . 04353 . 01710 . 7554/eLife . 04353 . 018Figure 6—figure supplement 1 . Crystallographic characterization of the ANK repeats/Nav1 . 2 structure . Electron density map of the AnkB/Nav1 . 2 complex contoured at 1 σ with the final structural model superimposed is shown . The Nav1 . 2 peptide is shown as the stick model and the AnkB_repeats is shown as the Cα model . DOI: http://dx . doi . org/10 . 7554/eLife . 04353 . 018 In the complex structure , the extended Nav1 . 2_ABD-C peptide interacts with the surface of the inner groove formed by the first five ANK repeats ( Figure 6A ) . In particular , the hydrophobic residues of Nav1 . 2_ABD-C and AS occupy very similar positions on the hydrophobic groove formed by residues from ANK repeats R4 and R5 , and subtle conformational differences in the finger loops of R4 and R5 can accommodate amino acid sequence differences between the two targets ( Figure 6E ) . This similar pattern and subtle accommodation illustrate that ANK repeats in general are incredibly adaptable and versatile as protein binding modules . Unique to Nav1 . 2 , the binding of ABD-C extends all the way to R1 via charge–charge and hydrogen-bond interactions ( Figure 6A , E ) . We also compared our ANK repeats complex structure with two recently determined peptide-bound ANK repeats structures , ANKRA2 and RFXANK in complex with HDAC4 and RFX5 peptides , respectively ( Xu et al . , 2012 ) . Although the HDAC4 and RFX5 peptides also bind to ANKRA2 and RFXANK ANK repeats in extended conformations , the key target binding residues are restricted to a small set of hydrophobic residues in the A helices of the five ANK repeats . Accordingly , a consensus sequence motif can be recognized to bind to the ANKRA2 and RFXANK ANK repeats . We noted that Glu1112 , which is completely conserved in both Na+ and K+ channels and mutation of which in Nav1 . 5 to Lys is known to cause Brugada syndrome in humans ( Mohler et al . , 2004 ) , occupies the identical position as Glu1622 of AS does in the ANK repeats/AS complex ( Figure 3A and Figure 6A , E ) . In contrast to the common expectation of directly interacting with positively charged residue ( s ) , Glu1112 of Nav1 . 2 is buried deeply in the groove and forms hydrogen bonds with the sidechains of Thr94 and Asn98 in the R2 and R3 finger loop ( Figure 6A ) . Charge potential calculation shows that the Glu1112 binding pocket formed by R2 and R3 is highly positive , and thus nicely accommodates the negatively charged carboxyl group of Glu1112 ( Figure 6B ) . As expected , the charge reversal mutation of Nav1 . 2 ( E1112K ) abolished the channel's binding to ANK repeats . Even mild substitutions ( E1112Q- and E1112A-Nav1 . 2 ) weakened the binding of Nav1 . 2 to ANK repeats by ∼30-fold ( Figure 6D ) . In agreement with our findings , E1112Q- and E1112A-Nav1 . 2 ( or E1100 in Nav1 . 6 ) failed to cluster at the AIS of hippocampal neurons ( Fache et al . , 2004; Gasser et al . , 2012 ) . Conversely , substitutions of Thr94 and Asn98 of ANK repeats with Ala and Glu , respectively , also weakened the ANK repeats/Nav1 . 2_ABD interaction ( Figure 6D ) . The above biochemical and structural data illustrate the importance of the absolutely conserved Glu in various ankyrin binding targets shown in Figure 5D and Figure 6C in anchoring these binding domains to site 1 of ANK repeats . The structures of ANK repeats in complex with the two different targets shown here also provide a framework for understanding the binding of KCNQ2/3 , β-dystroglycan , and potentially other ankyrin partners . We next evaluated the consequences of mutations of AnkG characterized in Figure 3 on its function in clustering Nav channels and Nfasc at the AIS in cultured hippocampal neurons . It is predicted that the ‘FF’ mutant in site 1 of AnkG_repeats disrupts its Nav1 . 2 binding but retains the Nfasc binding ( Figure 3F ) . As shown previously ( He et al . , 2012 ) , the defect in both AIS formation and Nav channels/Nfasc clustering at the AIS caused by knockdown of endogenous AnkG could be rescued by co-transfection of the shRNA-resistant , WT 270 kDa AnkG-GFP ( Figure 7 ) . The ‘FF’ mutant of 270 kDa AnkG-GFP was concentrated normally at the AIS , but failed to rescue clustering of endogenous Nav at the AIS ( Figure 7A , C , D ) , consistent with the significantly weakened binding of the mutant AnkG to Nav1 . 2 ( Figure 3E , F ) . This result confirms that the proper clustering of Nav at the AIS depends on AnkG ( Zhou et al . , 1998; Garrido et al . , 2003 ) . In contrast , Nfasc clustered properly at the AIS in neurons co-transfected with ‘FF’-AnkG ( Figure 7B , E ) , which was predicted since the ‘FF’ mutant had no impact on AnkG's binding to Nfasc . Interestingly , both the ‘IL’ ( site 2 ) and ‘LF’ ( part of site 3 ) mutants of AnkG-GFP failed to cluster at the AIS of hippocampal neurons ( Figure 7C and Figure 7—figure supplement 1 ) , suggesting that the L1-family members ( Nfasc and/or Nr-CAM ) or other potential ANK repeats site 2/3 binding targets may play a role in anchoring AnkG at the AIS . Not surprisingly , neither of these mutants can rescue the clustering defects of Nav or Nfasc caused by the knockdown of endogenous AnkG ( Figure 7D , E and Figure 7—figure supplement 1 ) . 10 . 7554/eLife . 04353 . 019Figure 7 . Mutations of residues at the target binding groove affect 270 kDa AnkG's function at the AIS in neurons . ( A ) WT 270 kDa AnkG-GFP effectively rescues AnkG self-clustering and clustering of sodium channels at the AIS . The FF mutant of AnkG is clustered at the AIS , but fails to rescue sodium channel clustering at the AIS . BFP marks the shRNA transfected neurons ( scale bars , 50 µm ) . White boxes mark the axon initial segment , which is shown at a higher magnification below each image ( scale bars , 10 µm ) . ( B ) Same as in panel A except that the red signals represent anti-neurofascin staining . ( C ) Quantification of anti-GFP fluorescence intensity ratio of axons to dendrites in cells depleted of endogenous 270/480 kDa AnkG and rescued with WT ( n = 34 ) , FF ( n = 30 ) , IL ( n = 24 ) , or LF ( n = 24 ) AnkG-GFP . p<0 . 05 . Error bars , S . E . ( D ) Quantification of the anti-endogenous pan-sodium channels fluorescence intensity ratio of axons to dendrites in cells depleted of endogenous 270/480 kDa AnkG and rescued with GFP alone ( n = 11 ) , WT ( n = 17 ) , FF ( n = 16 ) , IL ( n = 14 ) , and LF ( n = 10 ) AnkG-GFP . p<0 . 05 . Error bars , S . E . ( E ) Quantification of the anti-endogenous neurofascin fluorescence intensity ratio of axons to dendrites in cells depleted of endogenous 270/480 kDa AnkG and rescued with GFP alone ( n = 6 ) , WT ( n = 17 ) , FF ( n = 14 ) , IL ( n = 10 ) , and LF ( n = 10 ) AnkG-GFP . **p<0 . 05 . Error bars , S . E . DOI: http://dx . doi . org/10 . 7554/eLife . 04353 . 01910 . 7554/eLife . 04353 . 020Figure 7—figure supplement 1 . The IL and LF AnkG-GFP mutants do not cluster at the AIS and fail to rescue AnkG's functions in the AIS . GFP control , IL mutant of AnkG-GFP , and LF mutant of AnkG-GFP constructs were co-transfected with shRNA against mouse AnkG , respectively , to assay their ability to rescue AnkG's self-clustering and clustering of sodium channels ( A ) and neurofascin ( B ) at the AIS . The two AnkG-GFP mutants can neither cluster at the AIS nor rescue the clustering of sodium channels or Nfasc . The quantification data are shown in Figure 7C–E . DOI: http://dx . doi . org/10 . 7554/eLife . 04353 . 020 The 24 ANK repeats form an elongated , continuous solenoid structure with its extremely conserved target binding inner groove spanning a total length of ∼210 Å ( Figure 2C ) . We identified three distinct target binding sites in the first 14 repeats ( Figure 2 and Figure 3 ) . This is in agreement with earlier studies showing that three to five ANK repeats can form a stable structural unit capable of recognizing certain target sequences ( Li et al . , 2006; Tamaskovic et al . , 2012; Xu et al . , 2012 ) . Therefore , we predict that the last 10 ANK repeats of ankyrins can contain an additional two to three target binding sites . Importantly , the target binding sites on ANK repeats behave rather independently , as mutations/disruptions of interactions in each site do not lead to large perturbations in the interactions in the neighboring sites ( Figure 3 ) . Equal importantly , the ANK repeats targets bind to the inner groove with extended conformations , and the segments responsible for binding to each site do not seem to cooperate with each other ( i . e . , an alteration in one segment does not have a large impact on the neighboring segments ) ( Figure 3 and Figure 5 ) . Therefore , the multiple target binding sites on ANK repeats are quasi-independent . We further show that the AnkR_AS , the Nfasc , the Nav1 . 2 , the KCNQ2 , and the Cav1 . 3 peptides use different combinations of these sites that spread along the elongated and near completely conserved inner ANK repeat groove to form specific ankyrin/target complexes . One can envision that such combinatorial usage of multiple quasi-independent sites can in principle generate a large repertoire of binding targets with different sequences for ANK repeats . Although a number of ion channels use site 1 as the common binding site , the amino acid sequences of the corresponding site 1-binding peptide segments are rather diverse ( Figure 6C ) . One can expect that the sequences of target peptide segments responsible for binding to sites 2 and 3 will be even more diverse ( e . g . , the corresponding site 3 binding sequence of AnkR_AS and Nav1 . 2 ABD_N have no detectable sequence similarity ) , as the interactions in these two sites are more hydrophobic in nature ( Figure 3A–C ) . The combinatorial usage of the quasi-independent sites , together with the low sequence specificity of each binding site as well as the structural plasticity of the ANK repeat solenoid ( Lee et al . , 2006 ) , indicate that ANK repeats can have large capacities in binding to numerous membrane targets with diverse sequences . In light of the above points , unidentified ANK repeat binding proteins will likely be difficult to predict simply based on amino acid sequences , although a firm conclusion awaits detailed characterizations of more ankyrin binding targets . The combinatorial usage of multiple binding sites has also been observed in other long repeat-containing proteins including the Karyopherin family nuclear import/export scaffold proteins ( Conti et al . , 1998; Kobe , 1999; Chook and Blobel , 2001; Xu et al . , 2010 ) , the Wnt signaling regulatory scaffold β-catenin ( Graham et al . , 2000; Huber and Weis , 2001 ) , and tetratricopeptide repeats protein LGN/Pins ( Zhu et al . , 2011 ) . It is possible such a combinatorial target binding strategy may be a common feature for many other elongated repeat-containing proteins in diverse living organisms . The combinatorial multi-site interaction mode may also be advantageous for efficient regulation of ANK repeats/target interactions . By spreading a target binding to multiple sites along the ANK repeats inner groove that are not directly coupled , each binding site can be regulated independently and in a graded fashion . This might allow multiple regulatory signals to be integrated in a combinatorial manner to regulate ankyrin/membrane target interactions . Such a graded regulatory mechanism can be important for ankyrins to respond to various signal inputs when multiple membrane targets co-exist . For example , AnkG co-exists with Nfasc and sodium and potassium channels at the AIS ( Jenkins and Bennett , 2001; Pan et al . , 2006; Le Bras et al . , 2013 ) , and the components of the AnkG-mediated complex at the AIS can undergo distinct activity-dependent changes ( Hu et al . , 2009; Grubb and Burrone , 2010; Kuba et al . , 2010; reviewed in Kole and Stuart , 2012 ) and exhibit AIS plasticity during development ( Galiano et al . , 2012; Gutzmann et al . , 2014 ) . It has been reported that Nfasc and sodium channels can undergo activity-dependent phosphorylation in their ANK repeat binding domains ( Garver et al . , 1997; Whittard et al . , 2006; Brechet et al . , 2008 ) , which may underlie the distinct patterns of concentration gradients and their activity-dependent changes along the AIS . The target binding inner groove of ANK repeats of ankyrins essentially has not changed since the protein evolved over 500 million years ago . In contrast , most , if not all , currently identified ANK repeat-binding segments of ankyrin's membrane targets are either shown or predicted to be unstructured before binding to ankyrins ( Bennett and Lorenzo , 2013 ) . Such unstructured sequences are more tolerant of mutations as alterations are likely to have a minimal impact on the overall folding of proteins harboring them . Additionally , the conformational malleability is also advantageous for these unstructured peptide sequences to bind to ANK repeats with a molded groove . Since ANK repeats of ankyrins are responsible for binding to numerous targets with diverse sequences , it is likely that there is evolutionary pressure against random mutations in the ANK repeat sequences ( the residues in the inner groove in particular ) . The core function of the ankyrin/spectrin duo in patterning membrane micro-domains has remained unchanged throughout evolution , and thus the amino acid sequences of ANK repeats and the spectrin-binding domain of ankyrins are highly conserved ( Wang et al . , 2012 ) . New ankyrin binders ( e . g . , sodium and potassium channels at the AIS of neurons of higher mammals ) can evolve due to functional requirements . In summary , the structure of the 24 ANK repeats of ankyrins not only offers an explanation for the remarkable capacity of AnkR/B/G to bind to numerous membrane targets , but also provides a framework for guiding future studies of physiological functions and numerous pathological conditions directly associated with ankyrins . Since the three isoforms of ankyrins have distinct physiological functions , highly variable sequences outside the extremely conserved ANK repeats and spectrin-binding domain likely play critical roles in determining the cellular functions of each ankyrin isoform . Finally , proteins with extended ANK repeat domains ( e . g . , TRP channels , elongation factors , protein kinases and phosphatases , and various scaffold proteins ) may also interact with diverse partners via combinatorial uses of multiple , quasi-independent binding sites and thus are particularly suited as adaptors for assembling macromolecular complexes with broad cellular functions . The coding sequences of AnkB_repeats ( residues 28–873 ) were PCR amplified using the full-length human 220 kDa AnkB as the template . The coding sequences of the AnkR constructs , including AnkR_repeats ( residues 42–853 ) , and the full length AnkR C-terminal domain ( residues 1529–1907 ) , were PCR amplified from a mouse muscle cDNA library . The coding sequence of AnkG_repeats ( residues 38–855 ) were PCR amplified using the full-length rat 190 kDa AnkG as the template . The fusion construct of AnkR_AS and AnkB_repeats was made by standard two-step PCR with a coding sequence of thrombin recognition residues ‘GSLVPRGS’ as the linker . This construct was used to crystallize and determine the complex structure . The same strategy was used in making other fusion constructs , including the Nav1 . 2_ABD-C/AnkB_repeats_R1–9 fusion construct containing residues 1103–1129 from mouse Nav1 . 2 and human AnkB residues 28–318 followed by a capping sequence corresponding to the αB of R24 ( residues 814–822 ) and the AnkR_AS/AnkG_repeats fusion construct . For truncation mutations of ANK repeats constructs , the same capping sequence was added to the appropriate region of the C-terminus of each construct for protein stabilization . Mouse Nav1 . 2 ( NP_001092768 . 1 ) and mouse neurofascin ( CAD65849 . 1 ) were used here for studying their interaction with ankyrins . Peptides for mouse KCNQ2 ( NP_034741 . 2 , residues 826–845 ) and mouse Cav1 . 3 ( NP_083257 . 2 , residues 2134–2166 ) were commercially synthesized . For simplicity , we used human 220 kDa AnkB for the amino acid numbering throughout the manuscript . For the corresponding point mutations made on AnkG_repeats , each residue number should be increased by 10 . All point mutations were created using the Quick Change site-directed mutagenesis kit and confirmed by DNA sequencing . All of these coding sequences were cloned into a home-modified pET32a vector for protein expression . The N-terminal thioredoxin-His6-tagged proteins were expressed in Escherichia coli BL21 ( DE3 ) and purified as previously described ( Wang et al . , 2012 ) . The thioredoxin-His6 tag was removed by incubation with HRV 3C protease and separated by size exclusion columns when needed . Isothermal titration calorimetry ( ITC ) measurements were carried out on a VP-ITC MicroCal calorimeter ( MicroCal , Northampton , MA ) at 25°C . All proteins were dissolved in 50 mM Tris buffer containing 100 mM NaCl , 1 mM EDTA , and 1 mM DTT at pH 7 . 5 . High concentrations ( 200–300 µM ) of each binding partner assayed in this study , including AnkR_AS , different Nav1 . 2 ABD proteins and mutants , and neurofascin ABD , were loaded into the syringe , with the corresponding ANK repeats proteins of ankyrin-R/B/G ( 20–30 µM ) placed in the cell . Each titration point was obtained by injecting a 10 μl aliquot of syringe protein into various ankyrin protein samples in the cell at a time interval of 120 s to ensure that the titration peak returned to baseline . The titration data were analyzed using the program Origin 7 . 0 and fitted by the one-site binding model . Analytical gel filtration chromatography was carried out on an AKTA FPLC system ( GE Healthcare , Sweden ) . Proteins were loaded onto a Superose 12 10/300 GL column ( GE Healthcare ) equilibrated with a buffer containing 50 mM Tris , 100 mM NaCl , 1 mM EDTA , and 1 mM DTT at pH 7 . 5 . Fluorescence assays were performed on a PerkinElmer LS-55 fiuorimeter equipped with an automated polarizer at 25°C . In a typical assay , a FITC ( Molecular Probes ) -labeled peptide ( ∼1 μM ) was titrated with each binding partner in a 50 mM Tris pH 8 . 0 buffer containing 100 mM NaCl , 1 mM DTT , and 1 mM EDTA . The Kd values were obtained by fitting the titration curves with the classical one-site binding model . For the purpose of NMR analysis , AnkB_repeats fused with AnkR_AS was prepared by growing bacteria in M9 minimal medium supplemented with 13CH3-Met ( CIL , Cambridge , MA ) . The protein was expressed and purified using the same method as for the native proteins . Two identical NMR samples containing 0 . 35 mM of the fusion protein in 50 mM Tris buffer ( pH 7 . 0 , with 100 mM NaCl , 1 mM DTT , 1 mM EDTA ) were prepared , except that one of the samples contained 50 µg/ml of thrombin . The complete cleavage of the fusion protein was assessed by taking a small aliquot of the thrombin-added sample for SDS-PAGE analysis . NMR spectra were acquired at 35°C on a Varian Inova 750 MHz spectrometer equipped with an actively z-gradient shielded triple resonance probe . Crystallization of the native AnkR_AS/AnkB_repeats complex and its Se-Met derivative , and the Nav1 . 2_ABD-C/AnkB_repeats_R1–9 complex was performed using the hanging drop vapor diffusion method at 16°C . Crystals of the ANK repeats/AS complex were obtained from the crystallization buffer containing 0 . 5 M ammonium sulfate , 1 . 0 M lithium sulfate , and 0 . 1 M sodium citrate at pH 5 . 6 . Crystals of the Nav1 . 2_ABD-C/AnkB_repeats_R1–9 complex were harvested in the crystallization condition with 1 . 8 M ammonium sulfate , 6–8% dioxane , and 0 . 1 M MES pH 6 . 5 . Before diffraction experiments , crystals were soaked in crystallization solution containing 30% glycerol for cryoprotection . The diffraction data were collected at Shanghai Synchrotron Radiation Facility and processed and scaled using HKL2000 ( Otwinowski and Minor , 1997 ) ( Table 1 ) . By using the single isomorphous replacement with anomalous scattering method , the Se-Met sites were found and refined , and the initial phase was determined in AutoSHARP ( Vonrhein et al . , 2007 ) . The structure model of ANK repeats was built manually based on the experimental phase and the last 12 ANK repeats of AnkR_repeats ( PBD ID: 1N11 ) ( Michaely et al . , 2002 ) . Since the AS peptide contains three Met residues , the building of the AS structure was guided by the Se-anomalous difference map as the reference ( Figure 2—figure supplement 2 ) . Each asymmetric unit contains one ANK repeats/AS molecule . The model was refined against the Se-Met dataset of ANK repeats/AS in PHENIX ( Adams et al . , 2002 ) . COOT ( Emsley and Cowtan , 2004 ) was used for model rebuilding and adjustments . In the final stage , an additional TLS refinement was performed in PHENIX . The initial phase of the Nav1 . 2_ABD-C/AnkB_repeats_R1–9 complex was determined by molecular replacement using the different repeat regions of the ANK repeats structure as the search models . The Nav1 . 2_ABD-C peptide was further built into the model . The model was refined using the same strategy as that used for ANK repeats/AS . The model qualities were checked by MolProbity ( Davis et al . , 2007 ) . The final refinement statistics are listed in Table 1 . All structure figures were prepared by PyMOL ( http://www . pymol . org/ ) . The assay was performed as previously described ( He et al . , 2012 ) . The shRNA of AnkG was cloned into a BFP-pll3 . 7 vector with the sequence GCGTCTCCTATTAGATCTTTC , targeting a serine-rich region shared by both the 270 kDa and 480 kDa isoforms of mouse AnkG but not the rescuing rat AnkG . Hippocampal neurons were obtained from newborn C57bl/6 mice and cultured until day 4 before co-transfection with the shRNA and different forms of rescue vectors containing rat AnkG using Lipofectamine 2000 . On day 7 , the neurons were fixed and processed for immunostaining . All the antibodies used in the study were the same as those described in the previous study ( He et al . , 2012 ) . All the images in this study were captured using a Zeiss LSM 780 laser-scanning confocal microscope . The hippocampal neurons were captured using a 40 × 1 . 4 oil objective with 0 . 3 μm Z spacing and pinhole setting to 1 Airy unit . Fluorescence intensity analyses were processed using ImageJ software . The intensity ratios in neurons were quantified and analyzed using GraphPad Prism 5 . For the statistical analysis , the neuronal data were compared using one-way ANOVA followed by a Tukey post hoc test .	Proteins are made up of smaller building blocks called amino acids that are linked to form long chains that then fold into specific shapes . Each protein gets its unique identity from the number and order of the amino acids that it contains , but different proteins can contain similar arrangements of amino acids . These similar sequences , known as motifs , are usually short and typically mark the sites within proteins that bind to other molecules or proteins . A single protein can contain many motifs , including multiple repeats of the same motif . One common motif is called the ankyrin ( or ANK ) repeat , which is found in 100s of proteins in different species , including bacteria and humans . Ankyrin proteins perform a range of important functions , such as connecting proteins in the cell surface membrane to a scaffold-like structure underneath the membrane . Proteins containing ankyrin repeats are known to interact with a diverse range of other proteins ( or targets ) that are different in size and shape . The 24 repeats found in human ankyrin proteins appear to have essentially remained unchanged for the last 500 million years . As such , it remains unclear how the conserved ankyrin repeats can bind to such a wide variety of protein targets . Now , Wang , Wei et al . have uncovered the three-dimensional structure of ankyrin repeats from a human ankyrin protein while it was bound either to a regulatory fragment from another ankyrin protein or to a region of a target protein ( which transports sodium ions in and out of cells ) . The ankyrin repeats were shown to form an extended ‘left-handed helix’: a structure that has also been seen in other proteins with different repeating motifs . Wang , Wei et al . found that the ankyrin protein fragment bound to the inner surface of the part of the helix formed by the first 14 ankyrin repeats . The target protein region also bound to the helix's inner surface . Wang , Wei et al . show that this surface contains many binding sites that can be used , in different combinations , to allow ankyrins to interact with diverse proteins . Other proteins with long sequences of repeats are widespread in nature , but uncovering the structures of these proteins is technically challenging . Wang , Wei et al . 's findings might reveal new insights into the functions of many of such proteins in a wide range of living species . Furthermore , the new structures could help explain why specific mutations in the genes that encode ankyrins ( or their binding targets ) can cause various diseases in humans—including heart diseases and psychiatric disorders .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "biochemistry", "and", "chemical", "biology", "structural", "biology", "and", "molecular", "biophysics" ]	2014	Structural basis of diverse membrane target recognitions by ankyrins
Thrombocytopenic disorders have been treated with the Thrombopoietin-receptor agonist Eltrombopag . Patients with the same apparent form of thrombocytopenia may respond differently to the treatment . We describe a miniaturized bone marrow tissue model that provides a screening bioreactor for personalized , pre-treatment response prediction to Eltrombopag for individual patients . Using silk fibroin , a 3D bone marrow niche was developed that reproduces platelet biogenesis . Hematopoietic progenitors were isolated from a small amount of peripheral blood of patients with mutations in ANKRD26 and MYH9 genes , who had previously received Eltrombopag . The ex vivo response was strongly correlated with the in vivo platelet response . Induced Pluripotent Stem Cells ( iPSCs ) from one patient with mutated MYH9 differentiated into functional megakaryocytes that responded to Eltrombopag . Combining patient-derived cells and iPSCs with the 3D bone marrow model technology allows having a reproducible system for studying drug mechanisms and for individualized , pre-treatment selection of effective therapy in Inherited Thrombocytopenias . Bone marrow megakaryocytes are responsible for the continuous production of platelets in the blood , driven by Thrombopoietin ( TPO ) through interaction with its receptor MPL ( Hitchcock and Kaushansky , 2014; Kaushansky , 2015 ) . In vivo , megakaryocytes associate with bone marrow microvasculature , where they extend proplatelets that protrude through the vascular endothelium into the lumen and release platelets into the bloodstream ( Ito et al . , 2018; Junt et al . , 2007 ) . Countless human pathologies result in alterations in platelet production; yet , for many of these , pathogenesis , and thus optimal targeted therapies , remain unknown . Inherited Thrombocytopenias are a diverse group of disorders characterized by low platelet count , resulting in impaired hemostasis . While often stable , patients can experience hemorrhages and/or excessive bleeding provoked by hemostatic events such as trauma or surgery; in some cases , hemorrhages appear spontaneously ( Balduini et al . , 2018; Balduini et al . , 2017 ) . The treatment of Inherited Thrombocytopenias is still unsatisfactory . For patients affected with the severe forms , which are usually fatal at young ages , the treatment of choice is hematopoietic stem cell transplantation ( Balduini et al . , 2013; Locatelli et al . , 2003; Notarangelo et al . , 2008 ) . However , for most patients with Inherited Thrombocytopenias , transplantation is not recommended as the risks outweigh the benefits . The standard treatment protocols for these subjects were platelet transfusions to stop or prevent bleeding following trauma or during invasive procedures , anti-fibrinolytic agents , recombinant factor VIIa ( rVIIa ) , or local treatment . A significant advance in the treatment of thrombocytopenias is the use of drugs that stimulate platelet production by mimicking the effects of TPO . The TPO-receptor agonists Eltrombopag , Romiplostim , and very recently Avatrombopag , have been approved for the treatment of several forms of acquired thrombocytopenia ( Bussel , 2018; Cheng , 2011; Erickson-Miller et al . , 2009; Kuter , 2013; Santini and Fenaux , 2015 ) . TPO-receptor agonists were first explored in Inherited Thrombocytopenias in 2010 in a phase 2 trial of Eltrombopag in 12 patients with Myosin Heavy Chain 9 ( MYH9 ) mutations ( Pecci et al . , 2010 ) . In 2015 , Eltrombopag was tested in eight patients with Wiskott-Aldrich syndrome with platelet increases primarily in the X-Linked Thrombocytopenia ( XLT ) patients ( Gerrits et al . , 2015 ) . More recently , a follow on phase 2 trial showed that Eltrombopag was safe and effective in increasing platelet count and reducing bleeding symptoms in patients with different forms of Inherited Thrombocytopenia , including MYH9-Related Diseases ( MYH9-RD ) , Ankyrin Repeat Domain 26-Related Thrombocytopenia ( ANKRD26-RT ) , XLT/Wiskott-Aldrich syndrome , monoallelic Bernard-Soulier syndrome and Integrin beta 3 ( ITGB3 ) -Related Thrombocytopenia ( Zaninetti et al . , 2020 ) . Further , elective surgeries in MYH9-RD patients with severe thrombocytopenia have been performed safely after administration of Eltrombopag ( Zaninetti et al . , 2019 ) . Overall , these studies indicated that a sizeable proportion of patients with Inherited Thrombocytopenia respond to Eltrombopag , but that the extent of platelet response is highly variable not only among different forms of Inherited Thrombocytopenia but also among different patients affected by the same disease . Tools that recapitulate the function of specific tissues or organs are critical to test drug efficacy , reduce ineffective or suboptimal therapies , and personalize the choice of the best treatment for each specific patient as exemplified by organoids . Reproduction of the bone marrow has been very difficult because of its incompletely understood complexity . Current research is focused on duplicating characteristic features of the physiologic bone marrow microenvironment ex vivo using relevant biomaterials and bioreactors , along with appropriate human cell sources ( Chou et al . , 2020; Di Buduo et al . , 2018; Di Buduo et al . , 2021 ) . Silk is a naturally derived protein biomaterial with utility for studying platelet production since its fundamental features include non-thrombogenicity , low-immunogenicity , and non-toxicity ( Abbonante et al . , 2017; Di Buduo et al . , 2017; Di Buduo et al . , 2015; Omenetto and Kaplan , 2010 ) . A combination of modular flow chambers and vascular silk tubes and sponges was used to record platelet generation by primary human megakaryocytes , in response to variations in surface stiffness , functionalization with extracellular matrix components , and co-culture with endothelial cells ( Di Buduo et al . , 2017; Di Buduo et al . , 2015 ) . These systems were able to support efficient platelet formation and , upon perfusion , recovery of functional platelets , as assessed through multiple activation tests , including participation in clot formation and thrombus formation under flow conditions ( Di Buduo et al . , 2017; Di Buduo et al . , 2015 ) . We developed an ex vivo miniaturized 3D bone marrow tissue model that recapitulates ex vivo platelet biogenesis of patients with different forms of Inherited Thrombocytopenias . This device is a radical improvement of the previous model because it minimizes the number of cultured cells required in an unlimited number of simultaneous culture chambers . The results , starting from only 15 mL of peripheral blood , showed that the ex vivo tissue model could predict the in vivo clinical platelet response to Eltrombopag in individual patients . The number of platelets recovered in the ex vivo model under standardized conditions , including exposure to Eltrombopag , was significantly correlated with the increase in platelet count observed in vivo after Eltrombopag treatment in the same patients . Overall , our data suggest this tissue model will have substantial applicability for the evaluation of the effects of compounds to determine their impact on platelet production . In adults , hematopoietic bone marrow is located in the medullary cavity of flat and long bones ( Travlos , 2006 ) , served by blood vessels that branch out into millions of small thin-walled arterioles and sinusoids allowing mature blood cells to enter the bloodstream ( Figure 1A ) . To mimic such a structure , a device prototype of rectangular shape with 30 × 30 × 14 mm size and hollow cavities of 2 × 15 × 3 . 5 mm was developed . The device was connected to an outside peristaltic electronic pump ( Figure 1A ) through 0 . 9 mm diameter channels equipped with luer lock adaptors . We used devices with up to two reservoirs; however , the chamber can be designed to provide as many channels as required by the experimental conditions ( Figure 1—figure supplement 1 ) . Crosstalk between channels inside the device was eliminated by appropriate spatial separation and independent perfusion to allow assessment of patient-specific responses , following simultaneous exposure to TPO alone and TPO in combination with the tested drug . 3D printing technology is one emerging option for producing new devices in a customized , fast , and cost-effective manner . The printing process for the negative mold of our device is easily scalable . It can be created in less than 1 hr using a polylactic acid ( PLA High temperature , FormFutura Volcano , Figure 1—figure supplement 2 ) , which allows casting and curing of polydimethylsiloxane ( PDMS ) , a non-toxic polymeric organosilicon . The final shape of the system is optically clear ( Figure 1B–E ) . Importantly , the device is reusable and autoclavable to ensure overall sterility to the system . A silk fibroin structure functionalized with fibronectin was prepared with salt leaching method and inserted into the device to model a spongy scaffold that reproduces bone marrow architecture , composition , and microcirculation ( Figure 2A–C ) . A 2 days production process allowed us to obtain a sterile 3D silk-fibronectin scaffold that could be stored in water , at 4°C , up to 1 month after preparation and used upon experimental needs . The silk scaffold was connected to gas-permeable tubing allowing perfusion of the media with a peristaltic pump connected to inlet and outlet ports ( Figure 2A ) . A cover cap closes the system before starting perfusion . The 3D reconstruction of the silk scaffold revealed the presence of multiple spatially distinct niches ( Figure 2D and E ) and also demonstrated the homogeneous distribution of pores from top to bottom of the scaffold ( Figure 2F ) . This arrangement efficiently supported the diffusion of cells ( Figure 2G ) and media outflow without altering the shape and integrity of the silk . Importantly , the total volume collected after perfusion corresponded to that injected in the system by the pump . To ascertain the ability of the device to model physiological and pathological bone marrow , we took advantage of our expertise in culturing human hematopoietic stem and progenitor cells from peripheral blood of healthy controls and patients affected by two forms of Inherited Thrombocytopenia: ANKRD26-RT and MYH9-RD ( Bluteau et al . , 2014; Pecci et al . , 2009 ) . The bone marrow device was able to support efficient differentiation of mature megakaryocytes from both healthy controls and patients ( Figure 3A and B ) . However patient-derived megakaryocytes displayed a decreased percentage of proplatelet formation by about 80% , accompanied by less branching of proplatelet shafts due to a significantly lower number of bifurcations ( Healthy Control: 9 ± 2; ANKRD26-RT: 1 . 9 ± 0 . 7; MYH9-RD 1 . 8 ± 0 . 9 ) as compared to healthy controls ( Figure 3C–E ) . To validate the predictive value of the miniaturized bone marrow response to drugs specifically targeting hematopoiesis , we chose Eltrombopag as te model compound since Eltrombopag represents to date the only tested drug shown to increase platelet count of patients with Inherited Thrombocytopenias . First , we verified the ex vivo efficacy of Eltrombopag on human adult megakaryocytic progenitors from healthy controls and demonstrated the ability of the drug to increase megakaryocyte output and proplatelet formation with respect to the untreated control ( Figure 3—figure supplement 1 ) . Then , we tested the sensitivity of 24 pathological samples from ANKRD26-RT and MYH9-RD patients ( Table 1 ) . This cohort included 11 patients previously treated with Eltrombopag in a recent phase 2 clinical trial ( Zaninetti et al . , 2020 ) and two patients previously treated in preparation for elective surgery ( Zaninetti et al . , 2019 ) . Blood samples for this study were collected when patients were out of Eltrombopag therapy and had platelet count at their baseline levels . Equal numbers of megakaryocytic progenitors were divided between each channel for the ex vivo culture . Patients with Inherited Thrombocytopenias have normal or slightly increased serum levels of TPO ( Zaninetti et al . , 2020 ) , thus , in vivo hematopoietic progenitors are exposed to stimuli from endogenous TPO simultaneously with Eltrombopag treatment . To mimic this condition faithfully , all the samples were cultured in the presence of 10 ng/mL recombinant human TPO alone or in combination with 500 ng/mL Eltrombopag ( Figure 4A ) . Insights into the efficacy of Eltrombopag effects ex vivo were gained by simultaneously analyzing megakaryocyte differentiation at day 14 for each disorder . Specifically , cells were washed out of the device and analyzed . We observed comparable megakaryocyte maturation in terms of cell size ( Figure 4B ) , ploidy profile ( Figure 4C ) , and expression of lineage-specific markers ( Figure 4D and E ) , with and without Eltrombopag . However , the combination of TPO and Eltrombopag resulted in a significant two-fold increase in the output of mature megakaryocytes with respect to TPO alone for both ANKRD26-RT and MYH9-RD patients ( Figure 4F ) . Confocal microscopy analysis of 3D scaffolds revealed a homogeneous distribution of CD61+ megakaryocytes throughout the entire construct in both culture conditions , with more clusters in the presence of Eltrombopag , from both ANKRD26-RT and MYH9-RD ( Figure 5Ai-iv ) . Further , in the presence of Eltrombopag , megakaryocytes underwent characteristic cytoplasmic rearrangements typical of proplatelets ( Figure 5Aii , iv ) . β1-tubulin staining of megakaryocytes harvested from the device and seeded onto fibronectin-coated coverslips consistently highlighted that TPO in combination with Eltrombopag supported the extension of multiple branched shafts resembling nascent platelets at their terminal ends ( Figure 5Av-viii ) and a significant increase in the percentage of proplatelet-forming megakaryocytes in both ANKRD26-RT ( TPO: 3 ± 2 . 6%; TPO plus Eltrombopag: 7 . 7 ± 4 . 4% ) and MYH9-RD ( TPO: 1 . 5 ± 1%; TPO plus Eltrombopag: 4 . 4 ± 4 . 3% ) ( Figure 5B ) . Since the desired ultimate effect of Eltrombopag in patients with ANKRD26-RT or MYH9-RD is an increase in platelet count , platelet production was the most pertinent parameter evaluated in our ex vivo model . First we tested the possibility to harvest and count ex vivo produced platelets by perfusing the scaffolds cultured with megakaryocytes from healthy controls in the presence of TPO alone or TPO in combination with Eltrombopag . On day 15 of culture , each channel of the device was connected to a peristaltic pump at the inlet and a gas-permeable collection bag at the outlet . The number of ex vivo platelets produced was assessed and counted with a bead standard by flow cytometry after 4 hr of perfusion , at 37°C and 5% CO2 ( Figure 6A ) . The mean absolute number of collected platelets was 24 × 104/scaffold ( range 18−35 × 104 ) in the presence of TPO , with a significant 1 . 7-fold increase in the presence of TPO in combination with Eltrombopag ( p<0 . 05 ) . To test whether our device could predict the patient-specific response to Eltrombopag , we performed a systematic study comparing the extent of platelet production ex vivo to the platelet response observed in vivo in the same patients ( Zaninetti et al . , 2019; Zaninetti et al . , 2020 ) . Samples were perfused in the same standardized conditions used with healthy controls . After perfusion , ex vivo collected platelets exhibited the β1-tubulin coil at their periphery , typically present in peripheral blood platelets ( Figure 6B ) , further supporting the physiological relevance of the reproduced bone marrow environment for replicating in vivo thrombopoiesis . Ex vivo collected platelets were double-stained with anti-CD41 and anti-CD42b antibodies and counted by flow cytometry ( Figure 6C ) . The number of CD41+CD42b+ platelets collected per single channel increased significantly when treated with TPO in combination with Eltrombopag with respect to TPO alone , in both ANKRD26-RT and MYH9-RD groups ( Table 2 ) . However , while all samples from healthy controls responded to the treatment with Eltrombopag , in patients the platelet response was variable , with some samples demonstrating a slight or no increase in ex vivo platelet production . The same variability was present during the treatment in vivo ( Zaninetti et al . , 2019; Zaninetti et al . , 2020 ) . When the increase in platelet count obtained ex vivo in response to Eltrombopag was compared with the increase in platelet count observed in vivo after Eltrombopag administration in the same patients ( Table 2 ) , there was a statistically significant correlation ( R square = 0 . 78; p<0 . 0001 ) ( Figure 6D ) . The scientific relevance of this correlation was supported by evidence that the interpolation of the platelet count obtained in vivo after Eltrombopag administration with the megakaryocyte output calculated ex vivo did not register the same correlation ( R square = 0 . 35 ) ( Figure 6E ) , suggesting that ex vivo platelet count is the candidate parameter that is likely to predict the patients’ response better . Patient heterogeneity at the genetic and phenotypic levels is an increasingly important consideration in understanding the evolution of disease and resistance to treatment . Induced pluripotent stem cells ( iPSCs ) represent a useful tool to study disease mechanisms and testing drugs . Thus , iPSC clones were generated from one MYH9-RD patient and one healthy control . Clones were tested for pluripotency and found positive for OCT4 , NANOG , and SOX2 by qRT-PCR analysis ( Figure 7—figure supplement 1A ) . SOX2 , OCT4 , NANOG , TRA 1–81 , and SSEA4 marker expression was also confirmed by immunofluorescence analysis ( Figure 7—figure supplement 1B ) . Both control and patient clones displayed a normal diploid karyotype ( control: 46 , XX; patient: 46 , XY ) without noticeable abnormalities ( Figure 7—figure supplement 2 ) . Megakaryocyte differentiation of iPSC clones was confirmed in liquid culture conditions ( Figure 7A and Figure 7—figure supplement 3 ) and demonstrated that MYH9-RD iPSCs present a defect in proplatelet formation ( control: 5 . 6 ± 1 . 6%; MYH9-RD: 3 . 0 ± 1 . 1% ) and branching ( n° of bifurcations: control: 6 . 3 ± 2 . 8; MYH9-RD: 1 . 5 ± 1 ) with respect to control iPSCs ( Figure 7B–D ) . To validate their use within the bone marrow tissue bioreactor , megakaryocyte progenitors from iPSC clones were sorted at day 14 of differentiation based on the expression of CD61+ , an early progenitor lineage-specific marker , and cultured for an additional 5 days within the device in the presence of 50 ng/mL TPO supplemented or not with 500 ng/mL Eltrombopag ( Figure 8A ) . The disease clones showed comparable megakaryocyte maturation in terms of cell size ( Figure 8Bi–ii ) and expression of CD41 and CD42b ( Figure 8C and D ) in the presence of TPO or TPO plus Eltrombopag . 3D reconstruction of cell cultures from different MYH9-mutated clones revealed an increased number of megakaryocytes throughout the scaffold in the presence of TPO plus Eltrombopag ( Figure 8Biii-iv ) , paralleled by an increased percentage of proplatelets ( Figure 8Biii-iv ) having more branches with respect to TPO alone ( Figure 8Bv-viii ) . Statistical analysis demonstrated a significant increase in cell proliferation in the presence of TPO plus Eltrombopag with respect to TPO alone ( Figure 8E ) . Further , after perfusion , significantly increased platelet count was observed under treatment with Eltrombopag ( Figure 8F ) . Allogeneic platelet transfusions are widely used to treat acute bleeding in patients with thrombocytopenia of any origin and are also used to prevent bleeding in subjects who developed short-lasting , severe thrombocytopenia after chemotherapy or in those patients with more chronic thrombocytopenia in need of a procedure . However , platelet concentrates are not indicated for the prevention of hemorrhages in chronically thrombocytopenic patients for many reasons: they lose efficacy due to alloimmunization , acute reactions may occur , and transmission of infectious diseases is possible . Thus , platelet transfusions are not chronically administered to patients with Inherited Thrombocytopenia unless their platelet count is extremely low , and their risk of bleeding is relatively high . There is a need for alternative agents or approaches that could increase platelet count in these and chronic thrombocytopenic conditions . TPO-receptor agonists stimulate megakaryopoiesis and platelet production . Eltrombopag and/or Romiplostim and/or Avatrombopag are currently approved for the treatment of primary immune thrombocytopenia at various stages of ITP in adults and children ( Bussel , 2009 ) , thrombocytopenia related to liver disease if a procedure is needed , and severe acquired aplastic anemia ( Olnes et al . , 2012 ) . Small clinical trials and case reports have suggested that TPO receptor agonists are also effective in increasing platelet counts in patients with certain forms of Inherited Thrombocytopenia ( Rodeghiero et al . , 2018 ) and that at least Eltrombopag could be used to replace platelet transfusions to prepare patients to undergo hemostatic challenges ( Zaninetti et al . , 2019 ) . Indeed , a few patients have successfully received long-term treatment with TPO-receptor agonists , potentially paving the way for chronic treatment of these previously untreated forms of thrombocytopenia . However , platelet response to these drugs was variable among different patients , and sometimes the drugs were ineffective ( Gerrits et al . , 2015; Zaninetti et al . , 2019; Zaninetti et al . , 2020 ) . Here , we have developed a miniaturized 3D bone marrow tissue model that ex vivo reproduces in vivo platelet biogenesis in such a way that it allows us to predict response to drugs on a single patient basis . The major advantages of this device over the existing bone marrow models include rapid customization and manufacturing , handling ease , and the implementation of small silk-based three-dimensional scaffolds to allow the seeding of small amounts of adult megakaryocyte progenitors that are cultured for several days , perfused in parallel and simultaneously , to compare platelet production under different treatments ( Table 3 ) . As proof of principle , we applied our system to study thrombocytopenic patients affected by ANKRD26-RT and MYH9-RD who were treated with the TPO-receptor agonist Eltrombopag ( Zaninetti et al . , 2020 ) . According to our clinical data , the extent of platelet response to Eltrombopag in patients with ANKRD26-RT was lower than that in MYH9-RD ( Zaninetti et al . , 2019; Zaninetti et al . , 2020 ) . This range of variability was reflected ex vivo , clearly demonstrating that our tissue model can efficiently predict both the positive and negative response to Eltrombopag in individual patients and allow more personalized treatment , reducing the number of non-responders unnecessarily exposed to potential side effects of the treatment and ineffective preparation for procedures . In the future , patients might be able to create their own platelets and thus avoid most if not all of the complications discussed above . The study has several limitations . First , only patients affected by ANKRD26-RT and MYH9-RD were investigated; however , they are among the most frequent forms of Inherited Thrombocytopenia worldwide . For the many other forms of Inherited Thrombocytopenia , we do not know if our ex vivo predictive system will be similarly predictive . Second , since in vivo clinical trials for patients with Inherited Thrombocytopenia thus far have been limited to Eltrombopag; we , therefore , chose not to include either Romiplostim or Avatrombopag ( Bussel , 2018; Ghanima et al . , 2019 ) in our ex vivo testing . Third , the model was not tested to assess the effect of drugs that negatively impact platelet production , such as chemotherapy . Nonetheless , these limitations can readily be overcome by an additional study of the various permutations discussed . Viewed from this perspective , the studies reported here provide motivation and rationale for extending the model to allow identification of the impact of various molecules on platelet production for each patient . Furthermore , this ex vivo approach may be useful to study drugs not only in diseases characterized by thrombocytopenia but also in those with thrombocytosis . Inherited Thrombocytopenias each represent a prototype of thrombocytopenias deriving from defective platelet biogenesis within the bone marrow . For many Inherited Thrombocytopenias , the mechanisms of defective platelet production remain unknown . Understanding the cause of thrombocytopenia in these diseases could define the most suitable treatment for each disorder and identify both novel potential targets and either novel drugs or novel uses of existing drugs . Current 2D assays for functional assessment of megakaryocytes do not effectively monitor the final stage of maturation , in particular proplatelet spreading , platelet formation , and platelet release ( Balduini et al . , 2016 ) . By recreating megakaryocyte maturation from stem cells to platelet release , our miniaturized 3D bone marrow model demonstrated the ability to reproduce these key steps of thrombopoiesis , including alterations observed in diseased states . Megakaryocytes from patient-derived iPSCs reproduce the genetic background of peripheral-blood derived megakaryocytes and , thus , can be used to systematically study disease mechanisms and test candidate drugs . iPSCs represent a potentially unlimited source of megakaryocytes that can be frozen and made available on-demand without having to rely on frequent collection of patients’ blood . We hypothesized that combining the 3D bone marrow tissue model and iPSC technologies would be instrumental in addressing critical clinical needs for a more specific understanding of the mechanism of action of TPO-receptor agonists in patients . The possibility of generating iPSC clones from fibroblasts of a MYH9-RD patient has been previously shown ( Tangprasittipap et al . , 2019 ) , though , their differentiation potential was not proven . We derived iPSCs from hematopoietic stem and progenitor cells of one MYH9-RD patient and analyzed separately three different clones . The breakthrough of our approach includes an in-depth quality check of the iPSC clones to minimize heterogeneity and maximize replicability over-time . No defects in megakaryocyte maturation were observed from all the clones , while proplatelet formation was decreased as compared to healthy control clones . Besides its ability to stimulate megakaryopoiesis , Eltrombopag has also been well-demonstrated to promote multilineage hematopoiesis in patients with acquired bone marrow failure syndromes ( Olnes et al . , 2012 ) . Although the exact mechanisms of its effects on hematopoietic progenitor cells are not completely clear , Kao et al . recently demonstrated a stimulatory effect on stem cell self-renewal independently of the TPO receptor-mediated through iron chelation-dependent molecular reprogramming ( Kao et al . , 2018 ) . Our benchmark tests highlight that , besides platelet release , the 3D tissue model allowed us to track the effect of Eltrombopag on both progenitor cell and megakaryocyte functions , promising to provide a more comprehensive approach to study the effect of TPO-receptor agonists on hematopoietic stem and progenitor cells . In summary , we developed a proof-of-concept system that in two weeks measures the impact of TPO-receptor agonists on megakaryopoiesis and platelet production of individual patients starting from a small amount of their peripheral blood ( Figure 9 ) . This silk-based technology , which can be produced and customized in 2 days , reaches the expectation of cost efficiency , time-saving , convenience , and personalization of modern therapeutic approaches . The data demonstrated that the ex vivo system could predict in vivo clinical response to Eltrombopag . The increase in the number of platelets collected in the ex vivo model was comparable to the increase of platelet count in vivo upon treatment with Eltrombopag . The broader impact of this work is in the design of tools to mimic the bone marrow ex vivo that can uncover mechanisms of impaired platelet production and enable testing of candidate drug treatments on platelet production using patient-derived cells . In the future , our system may serve as the foundation for highly integrated approaches to generate solutions for the ex vivo production of all blood cells for transfusion . The system may also represent a benchmark for pre-clinical testing of new therapeutic applications for Inherited Thrombocytopenias or other hematologic diseases , and for testing the effects of potentially toxic agents for the whole hematopoietic niche . B . mori silkworm cocoons were supplied by Tajima Shoji Co . , Ltd . ( Yokohama , Japan ) . Pharmed tubing was from Cole-Parmer ( Vernon Hills , IL , USA ) . The immunomagnetic separation system was from Miltenyi Biotech ( Bergisch Gladbach , Germany and Bologna , Italy ) . Recombinant human TPO , interleukin-6 ( IL-6 ) , interleukin-11 ( IL-11 ) , human bone morphogenetic protein 4 ( BMP4 ) , human vascular endothelial growth factor ( VEGF ) , human fibroblast growth factor ( FGF ) , human Fms-related tyrosine kinase three ligand ( Flt3L ) , human stem cell factor ( SCF ) were from Peprotech ( London , UK ) . CHiR 99021 was from TOCRIS . TruCount tubes and human fibronectin were from Becton Dickinson ( S . Jose , CA , USA ) . The following antibodies were used: mouse monoclonal anti-CD61 , clone SZ21 , from Immunotech ( Marseille , France ) ; rabbit monoclonal anti-β1-tubulin was a kind gift of Prof . Joseph Italiano ( Brigham and Women's Hospital , Boston , USA ) . Alexa Fluor conjugated secondary antibodies and Hoechst 33258 were from Life Technologies ( Monza , Italy ) . Additional details can be found in the ‘Key resources table’ . The chamber was manufactured using 3D FDM printing technology and a biocompatible silicon molding approach . The modeling of the bioreactor was created using CAD software ( OnShape , Fusion360 , or Inventor2017 ) and used to generate 3D negative mold components exported as STL ( Standard Triangulation Language ) files , sliced with Slic3R PE , and export to the FDM 3D printer Prusa i3 MK3S ( Prusa Research , Czech Republic ) . The printing is done using a poly ( lactic acid ) ( PLA ) high-temperature filament of 1 , 75 mm ( FormFutura , Netherland ) deployed in layers of 100 µm by a 0 . 25 mm nozzle . After printing , the mold was cured in an oven at 100°C for 20 min to increase mechanical properties . To produce the perfusion channel , 21G needles were disposed in the dedicated holes and sealed with a gel of 25% Pluronic F-127 . The molding was performed using a polydimethylsiloxane ( PDMS ) ( Sylgard184 , Dow Corning ) , mixed in a 10:1 ratio of base material and curing agent . The selected material is stable both at low and high temperatures ( 45–200°C ) and it is resistant to UV , water , and solvents . The PDMS was poured into the 3D printed molds that were positioned into a vacuum chamber to remove all the air bubbles . The curing of the PDMS was performed in a dried oven at 70°C for 4 hr; the molds were then dissociated from the final silicon models sterilizable by autoclave . The chamber consisted of two wells of 10 × 22 mm , having a hollow cavity of 2 × 15 × 3 . 5 mm enclosed in a block of 30 × 30 × 14 mm and connected to the outside of the chamber through channels of 0 . 9 mm diameter . The luer adaptors for the inlet and outlet were mounted in the channel and sealed with biocompatible silicone adhesive MD7-4502 ( Dow Corning , USA ) . Then , the modular flow chamber was equipped with a silk fibroin sponge functionalized with fibronectin as described previously ( Di Buduo et al . , 2017 ) . Silk fibroin aqueous solution was obtained from B . mori silkworm cocoons according to previously published literature ( Di Buduo et al . , 2018 ) . Briefly , dewormed cocoons were boiled for 30 min in 0 . 02 M Na2CO3 solution at a weight to volume ratio of 10 g to 4 L . The fibers were rinsed for 20 min three times in ultrapure water and dried overnight . The dried fibers were solubilized for 4 hr at 60°C in 9 . 3 M LiBr at a weight to volume ratio of 3 g/12 mL . The solubilized silk solution was dialyzed against distilled water using a Slide-A-Lyzer cassette ( Thermo Scientific , Waltham , MA , USA ) with a 3500 MW cutoff for three days and changing the water a total of eight times . The silk solution was centrifuged at maximum speed for 15 min to remove large particulates and stored at 4°C . The concentration of the silk solution was determined by drying a known volume of the solution overnight at 60°C and massing the remaining solids . Silk solution ( 8% w/v ) ( Lovett et al . , 2007 ) was mixed with 25 µg/mL fibronectin and dispensed into the modular chamber . NaCl particles ( approximately 500 µm in diameter ) were then sifted into the solution in a ratio of 1 mL to 2 g of NaCl particles . The scaffolds were then placed at room temperature for 48 hr and then soaked in distilled water for 48 hr to leach out the NaCl particles . The scaffolds were sterilized in 70% ethanol and finally rinsed five times in PBS for over 24 hr . Silk scaffolds were characterized by confocal , as subsequently described . Perfusion of the silk scaffold has been tested at different flow rates ( 5–50 µL/min ) by using a peristaltic pump ( ShenChen Flow Rates Peristaltic Pump - LabV1 , China ) . The total volume collected after each test corresponded to that injected in the system by the pump . Human peripheral blood samples were obtained from healthy controls and thrombocytopenic patients after informed consent . All samples were processed following the ethical committee of the I . R . C . C . S . Policlinico San Matteo Foundation and the principles of the Helsinki Declaration . The main features of the 24 investigated samples from 20 different patients are reported in Table 1 . For four patients , the analysis was performed on two different occasions , with very similar results . Diagnosis of MYH9-RD or ANKRD26-RT had been confirmed by genetic analysis in all the cases . All patients provided written informed consent for this study , which was approved by the Institutional Review Board of the IRCCS Policlinico San Matteo Foundation , Pavia , Italy . A sample of 15 mL of peripheral venous blood anticoagulated with ACD was collected for the analysis in the 3D bone marrow system . Thirteen patients had previously received a short-term course of Eltrombopag ( 3–6 weeks ) either within a phase two clinical trial ( Zaninetti et al . , 2020 ) ( n = 11 ) or in preparation for elective surgery ( Zaninetti et al . , 2019 ) ( n = 2 ) . In any case , Eltrombopag was given at the dose of 50 or 75 mg/day for 3 or 6 weeks ( Zaninetti et al . , 2019; Zaninetti et al . , 2020 ) . The in vivo clinical response to the drug was expressed as the absolute increase in platelet count at the end of Eltrombopag treatment with respect to baseline . Blood samples for this study were collected when patients were out of Eltrombopag therapy ( minimum of 6 months to a maximum of 48 months washout ) . According to the fast pharmacokinetics of Eltrombopag ( plasma elimination half-life approximately 21–32 hr ) , all patients had platelet count at their baseline levels at one month of follow-up after the discontinuation of the treatment ( Pecci et al . , 2010; Zaninetti et al . , 2020 ) . Thus , we do not expect that previous exposure to Eltrombopag may have influenced haematopoietic stem and progenitor cell functions after months . CD45+ hematopoietic progenitor cells from peripheral blood samples were separated by an immunomagnetic bead selection kit ( Miltenyi Biotec , Bologna , Italy ) and cultured for 6 days in a flask in presence in Stem Span media ( StemCell Technologies , Canada ) supplemented with 1% penicillin-streptomycin , 1% L-glutamine , 10 ng/mL TPO , IL-6 , and IL-11 in the presence or not of 500 ng/mL Eltrombopag ( Novartis ) at 37°C in a 5% CO2 fully humidified atmosphere , as previously described ( Bluteau et al . , 2014; Pecci et al . , 2009 ) . On day 6 , CD61+ early megakaryocytic progenitors were sorted by immunomagnetic selection kit ( Miltenyi Biotec , Bologna , Italy ) and seeded for additional 8 days within the silk bone marrow model in presence of 10 ng/mL TPO supplemented or not with 500 ng/mL Eltrombopag . On day 14 of differentiation , the chamber was sealed , and the outlet ports were connected to the outlet needles . Culture media-filled tubes were connected to the inlet needles . The chamber was placed into the incubator ( 37°C and 5% CO2 ) , and transfer bags for platelet collection were secured to the outlet ports . The peristaltic pump ( ShenChen Flow Rates Peristaltic Pump - LabV1 , China ) was placed outside the incubator , and media was pumped for 4 hr at a flow rate of 10 µL/min , speed range: 0 . 18 rpm , perfusion pause: 120 s , perfusion run: 5 min , with a peristaltic pump . Induced pluripotent stem cells generation: iPSCs were derived from one MYH9-RD patient with heterozygous g . 103845T > A mutation and one healthy control providing their informed consent before the participation in this study and in accordance with the local ethical committee and the Declaration of Helsinki . At the time of the sampling , the patient was a 32-year-old male with a baseline platelet count of 50 × 109/L who had never been treated with Eltrombopag . CD34+ hematopoietic stem and progenitor cells were isolated from their peripheral blood using an immunomagnetic beads cell-sorting system ( AutoMacs; Miltenyi Biotec , Paris , France ) and amplified in serum-free media containing EPO ( 1 U/mL ) , FLT3L ( 10 ng/mL ) , G-CSF ( 20 ng/mL ) , IL-3 ( 10 ng/mL ) , IL-6 ( 10 ng/mL ) , SCF ( 25 ng/mL ) , TPO ( 10 ng/mL ) , and GM-CSF ( 10 ng/mL ) for 6 days . Cells were then transduced with the CytoTune iPS 2 . 0 Sendai Reprogramming Kit ( Thermo Fisher , Villebon-sur-Yvette , France ) and the reprogramming was performed according to the manufacturer's instructions . Colonies with an ES-like morphology were manually isolated , expanded for a small number of passages , and frozen . iPSCs were maintained on Essential eight or Essential 8 Flex media ( Gibco/Thermo Fisher ) , on plates coated with N-truncated human recombinant vitronectin ( Gibco ) . All derived clones were organized in colonies with defined edges and characterized by prominent nucleolus with a high nucleus-to-cytoplasm ratio ( Figure 7—figure supplement 2A ) . Cell passages were performed using a solution of EDTA 0 . 5 mM in PBS 1X , or TrypLE 1X ( Gibco ) . Supernatant samples from the generated iPSC lines were used for mycoplasma detection . A total of 100 µl supernatants were heated at 95°C for 10 min and centrifuged at 1000 g for 5 s to discard cellular debris . For the detection of mycoplasma species was used Venor GeM OneStep Mycoplasma Detection Kit for Conventional Polymerase chain reaction ( PCR ) ( Minerva Biolabs ) per manufacturer's instructions . PCRs were performed using a thermocycler . Amplified products were fractionated on 1 . 5% agarose and observed with Amersham Imager 680 ( GE ) . All derived clones resulted negative for Mycoplasma screening . Cell pellets from iPSC lines were used for DNA isolation . QIAamp DNA Mini Kit ( Qiagen ) was used following manufacturer instructions . DNA quantification was performed on Qubit . 2 . 0 Fluorometer ( Thermo Fisher ) . The genetic matching of the generated iPSC lines to the parental cells was confirmed by short tandem repeat ( STR ) analysis . The amplification was performed by PCR using ~1 ng/sample: nine autosomal STR molecular markers ( D21S11 , D7S820 , CSF1PO , TH01 , D13S317 , D16S539 , vWA , TPOX , D5S818 ) along with the gender determining marker Amelogenin with Promega GenePrint 10 Kit following manufacturer's recommended protocol . Appropriate positive and negative amplification controls were used as kit recommended guidelines . The amplified products were electrophoresed on an ABI Prism 3730xl Genetic Analyzer using an Internal Lane Standard 600 ( Promega ) . Data generated were analyzed using GeneMapper Software version 4 . 0 ( Applied Biosystems ) following the manufacturer's instructions . Results demonstrated that iPSC clones and the blood donor perfectly matched the 10 loci tested ( Table 4 ) . Metaphase-chromosome spreads were prepared from 80% confluent cultures according to standard procedures . Actively dividing cells were treated with 10 ng/ml colcemid ( Gibco KaryoMAX Colcemid solution in PBS , Thermo Fisher Scientific ) for 16 hr ( overnight ) at 37°C . Cells were combined in 0 , 56% KCl for 20 min at 37°C and were fixed with methanol/acetic acid ( 3:1 v/v ) . Chromosome analysis was carried out by applying Q-banding by fluorescence using quinacrine ( QFQ ) , according to routine procedures , following the guidelines of the International System for Chromosome Nomenclature 2009 ( ISCN 2009 ) ( Shaffer et al . , 2009 ) . Microscope observation was performed using the Fluorescence microscope Olympus BX63 , fully equipped with quinacrine mustard filter and CCD camera and the acquisition and analysis of ‘GenASIs’ Software , version 8 . 1 . 0 . 47741 ( Applied Spectral Imaging ) . On average , 25 metaphases were evaluated . No gross chromosomal alterations were observed by Q-banding ( Figure 7—figure supplement 2B ) . Comparative genomic hybridization array ( aCGH ) analysis was performed using Agilent Human Genome CGH Microarray 60K kit ( Agilent Technologies , Palo Alto , CA , USA ) , following the manufacturer’s instructions . A sex-matched commercial DNA sample ( Male , Promega , Milan , Italy ) was used as reference DNA . Hybridization signals were analyzed using Feature Extraction software ( v10 . 7 ) and DNA Analytics software ( v5 . 0 , Agilent Technologies , Palo Alto , CA , USA ) . Aberration Detection Method 2 ( ADM2 ) algorithm ( threshold 5 . 0 ) was used to identify DNA copy number aberrations . We applied a filtering option of a minimum of three aberrant consecutive probes ( Wu et al . , 2007 ) and a minimum absolute average log 2 ratio of 0 . 30 . University of California Santa Cruz ( UCSC ) human genome assembly hg18 was used as a reference and copy number variations ( CNVs ) were identified with a database integrated into the Agilent Genomic Workbench analytic software . Log two ratios lower than −0 . 30 were classified as losses , those greater than 0 . 30 as gains . The analysis revealed CNVs for the one iPSC clone , consisting of amplification of 36 kb and 175 Kb , on chromosomes 2 and 3 , respectively , involving genes with no significant associated clinical phenotype . The clone from healthy control did not reveal any CNVs ( Table 5 ) . Total RNA from each cell line was isolated using TRIzol reagent ( Sigma ) following the manufacture's protocol . For semi-quantitative PCR ( qPCR ) experiments , equal amounts of total RNA ( 1 . 3 μg ) were reverse transcribed by using the RevertAid First Strand cDNA Synthesis Kit ( Thermoscientific ) . The complementary DNA ( cDNA ) samples were used for validation of the self-renewal stem cell markers using RT-PCR analysis . qRT-PCR was assessed in triplicate on at least two independent biological replicates by the DDCt method on Rotor-Gene Q ( Qiagen ) using the Maxima SYBR Green qPCR Master Mix ( ThermoFisher Scientific ) . GAPDH was selected as a housekeeping gene and data were normalized to its expression . Statistical analysis was performed using REST ( relative expression software tool ) software . Primer sequences are described in Table 6 . Cells were seeded in four-well plates . Cells were fixed in 4% paraformaldehyde for 15 min , incubated in 0 . 1 M glycine for 10 min at room temperature ( RT ) , and then in blocking solution composed of 5% goat serum , 0 . 6% Triton in PBS for 30 min at RT . Cells were immunostained at 4°C overnight in blocking solution with primary antibodies anti-Sox2 ( 1:300; Millipore ) , anti-Oct4A ( 1:400; Cell Signaling ) , anti-Nanog ( 1:500; Abcam ) , anti-SSEA4 ( 1:50; Millipore ) , and anti-TRA-1–81 ( 1:50; Millipore ) . For the immunofluorescence characterization , samples were incubated with appropriate secondary antibodies Rhodamine-Red anti-mouse IgM and antirabbit IgG , Alexa Fluor 488 anti-mouse IgG , ( Jackson ImmunoResearch , distributed by Li StarFish , Milan , Italy ) for 1 hr at RT , and nuclei were counterstained with Hoechst 33258 . Cells were mounted with GelMount aqueous mounting media ( Sigma ) . The images were acquired using a Leica DMI4000B inverted microscope linked to a DFC360FX or a DFC280 camera ( Leica Microsystems ) . Pluripotency competence of all iPSC clones was assessed by embryoid body formation Presence of the three germ layer derivatives in the generated embryoid bodies was shown in vitro by immunofluorescence staining ( Figure 7—figure supplement 4 ) . Briefly , cells were thawed and seeded into 6-well plates ( 5 × 104 cells per well ) under appropriate culture conditions and incubated at 37°C . At confluency of 80% were detached as cell clumps , plated in six-well low-attachment plates , and cultured in Essential eight media ( Thermo Fisher Scientific ) supplemented with 4 mg/ml PVA ( polyvinyl alcohol , Sigma ) and 10 μg/ml ROCK inhibitor . Two days later , cell aggregates were nourished with Essential eight media ( Thermo Fisher Scientific ) and E6 media ( 1:1 mixture ) supplemented with 4 mg/ml PVA . On day 6 , Embryoid Bodies ( EBs ) were collected , plated on matrigel-coated wells , and allowed to differentiate for further 8 days with daily media changes . For immunofluorescence analysis , on day 14 , cells were seeded in four-well plates and fixed in 4% paraformaldehyde , incubated in 0 . 1 M glycine , and blocked in a solution consisting of 5% goat serum , 0 . 6% Triton in PBS . Cells were immunostained in blocking solution with primary antibodies: for ectoderm anti-βIII-Tubulin ( 1:100; Sigma ) , mesoderm anti- SMA ( 1:200; Sigma ) , and endoderm anti-AFP ( 1:50; R and D Systems ) . For the immunofluorescence characterization , samples were incubated with appropriate secondary antibodies Alexa Fluor 488 anti-mouse IgG and Rhodamine-Red anti-rabbit IgG ( Jackson ImmunoResearch , distributed by Li StarFish , Milan , Italy ) , and nuclei were counterstained with Hoechst 33258 . The images were acquired using a Leica DMI4000B inverted microscope linked to a DFC360FX or a DFC280 camera ( Leica Microsystems ) . At day −1 , colonies of pluripotent stem cells were seeded on Geltrex ( 12 ug/cm2 ) -coated 100 mm dish plates , in mTeSR1 ( STEMCELL Technologies ) with 10 µM ROCK Inhibitor ( Millipore ) . The starting cell concentration was adjusted for each cell line at 10–15% confluency range ( 5 × 105 cells/dish ) . After 4 hr , media was replaced with fresh mTeSR1 . At Day 0 , cells were transferred in a xeno-free media based on StemPro-34 SFM ( ThermoFischer Scientific ) , supplemented with Penicillin/Streptomycin 0 . 5% v/v ( Sigma ) , L-Glutamine 1% v/v ( Gibco ) , 1- Thioglycerol 0 . 04 mg/mL ( Sigma ) , and ascorbic acid 50 mg/mL ( Sigma ) . This media was retained for the entire experiment and supplemented with different cytokines , small molecules and growth factors , according to the following schedule ( Donada et al . , 2019 ) : days 0–2: BMP4 ( 10 ng/mL ) , VEGF ( 50 ng/mL ) and CHIR99021 ( 2 μM ) . Days 2–4: BMP4 ( 10 ng/mL ) , VEGF ( 50 ng/mL ) and FGF2 ( 20 ng/mL ) . Days 4–6: VEGF ( 15 ng/mL ) , and FGF2 ( 5 ng/mL ) . Day 6: VEGF ( 50 ng/mL ) , FGF2 ( 50 ng/mL ) , SCF ( 50 ng/mL ) , and FLT3L ( 5 ng/mL ) . Days 7–10: VEGF ( 50 ng/mL ) , FGF2 ( 50 ng/mL ) , SCF ( 50 ng/mL ) , FLT3L ( 5 ng/mL ) , TPO ( 50 ng/mL ) , and IL-6 ( 10 ng/mL ) . Days 10–14: SCF ( 50 ng/mL ) , FLT3L ( 5 ng/mL ) , TPO ( 50 ng/mL ) , and IL-6 ( 10 ng/mL ) . Starting from day 14 , CD61+ early megakaryocytic progenitors were sorted by immunomagnetic selection kit ( Miltenyi Biotech , Bologna , Italy ) and seeded for additional 5 days within the silk bone marrow model in presence of TPO ( 50 ng/mL ) supplemented or not with 500 ng/mL Eltrombopag . Megakaryocyte differentiation and proplatelet yields were evaluated by adhesion on fibronectin at the end of the culture ( 14th day ) , as previously described ( Di Buduo et al . , 2014; Pecci et al . , 2009 ) . Briefly , 12 mm glass cover-slips were coated with 25 µg/ml human fibronectin ( Merck-Millipore , Milan , Italy ) , for 24 hr at 4°C . Megakaryocytes were harvested from the silk bone marrow scaffold by extensive washing and seeded in a 24-well plate , at 37°C in a 5% CO2 fully humidified atmosphere . After 16 hr , adhering cells were fixed in 4% paraformaldehyde ( PFA ) , permeabilized with 0 . 1% Triton X-100 ( Sigma Aldrich , Milan , Italy ) , and stained for immunofluorescence evaluation with rabbit anti-β1-tubulin primary antibody ( 1:1000 ) or anti-mouse CD61 ( 1:100 ) and Alexa Fluor-conjugated secondary antibodies ( 1:500 ) ( Invitrogen , Milan , Italy ) . Nuclei were stained with Hoechst 33258 ( 1:10 , 000 ) ( Sigma Aldrich , Milan , Italy ) . The cover-slips were mounted onto glass slides with ProLong Gold antifade reagent ( Invitrogen , Milan , Italy ) and imaged by an Olympus BX51 microscope ( Olympus , Deutschland GmbH , Hamburg , Germany ) . Proplatelet-forming megakaryocytes were identified as cells displaying long filamentous structure ending with platelet-sized tips . The results were expressed as a percentage of the total number of cells analyzed . For immunofluorescence imaging of megakaryocyte cultures within the silk bone marrow tissue model , samples were fixed in 4% paraformaldehyde ( PFA ) for 20 min and then blocked with 5% bovine serum albumin ( BSA , Sigma ) for 30 min at room temperature . Samples were probed with anti-CD61 ( 1:100 ) overnight at 4°C and then immersed in Alexa Fluor secondary antibody ( 1:500 ) for 2 hr at room temperature . Nuclei were stained with Hoechst . Samples were imaged by a TCS SP8 confocal laser scanning microscope ( Leica , Heidelberg , Germany ) . For silk fibroin scaffolds imaging , we took advantage of silk auto-fluorescence in UV light . In some experiments , silk fluorescence was brightened by staining with Hoechst ( Talukdar et al . , 2011 ) . For all immunofluorescence imaging , the acquisition parameters were set on the negative controls . 3D reconstruction and image processing performed using Leica licensed software or Image J software . For analysis of peripheral blood and ex vivo collected platelet morphology , different approaches were used . First , megakaryocytes at the end of differentiation and platelets from peripheral blood or perfused media were visualized by light microscopy with an Olympus IX53 ( Olympus Deutschland GmbH , Hamburg , Germany ) . For analysis of cytoskeleton components , cells were stained as previously described ( Di Buduo et al . , 2016 ) . Briefly , collected platelets were fixed in 4% PFA and centrifuged onto poly-L-lysine coated coverslip while peripheral blood smears were air-dried and then fixed in 4% PFA , permeabilized with 0 . 1% Triton X-100 for 5 min , and blocked with 5% BSA for 30 min at room temperature . To visualize microtubule organization , samples were probed with anti-β1-tubulin ( 1:1000 ) for 1 hr at room temperature and then immersed in Alexa Fluor secondary antibody ( 1:500 ) for 2 hr at room temperature . Samples were mounted onto glass slides with ProLong Gold antifade reagent ( Invitrogen , Milan , Italy ) and then imaged by an Olympus BX51 fluorescence microscope ( Olympus , Deutschland GmbH , Hamburg , Germany ) . For all immunofluorescence imaging , the acquisition parameters were set on the negative controls , which were routinely performed by omitting the primary antibody . Flow cytometry settings for analysis of megakaryocytes and ex vivo generated platelets were established , as previously described ( Abbonante et al . , 2016; Cramer et al . , 1997; Fujimoto et al . , 2003; Nakamura et al . , 2014; Takayama et al . , 2008 ) . For analysis of the percentage of fully differentiated megakaryocytes at the end of the culture ( 14th day ) , 50 × 103 cells were suspended in phosphate buffer saline ( PBS ) and stained with a FITC-conjugated antibody against human CD41 and human CD42b ( PE ) ( eBioscience , Milan , Italy ) at room temperature in the dark for 30 min and then analyzed . Ex vivo collected platelets were analyzed using the same forward and side scatters as human peripheral blood and identified as CD41+CD42b+ events . Isotype controls were used as negative controls to exclude non-specific background signal . The platelet number was calculated using a TruCount bead standard . A minimum of 10 , 000 events was acquired . All samples were acquired with a Beckman Coulter Navios flow cytometer ( Indianapolis , IN , US ) . Off-line data analysis was performed using Beckman Coulter Navios software package . Values were expressed as mean plus or minus the standard deviation ( mean ± SD ) or mean plus or minus the standard error of the mean ( mean ± SEM ) . A two-tailed paired t-test was performed for statistical analysis of data from samples tested in parallel under different experimental conditions . A two-tailed unpaired t-test was performed for statistical analysis of data from different samples . Statistical analysis was performed with GraphPad Software . A p-value of less than 0 . 05 or 0 . 01 was considered statistically significant . All experiments were independently replicated at least three times .	Platelets are tiny cell fragments essential for blood to clot . They are created and released into the bloodstream by megakaryocytes , giant cells that live in the bone marrow . In certain genetic diseases , such as Inherited Thrombocytopenia , the bone marrow fails to produce enough platelets: this leaves patients extremely susceptible to bruising , bleeding , and poor clotting after an injury or surgery . Certain patients with Inherited Thrombocytopenia respond well to treatments designed to boost platelet production , but others do not . Why these differences exist could be investigated by designing new test systems that recreate the form and function of bone marrow in the laboratory . However , it is challenging to build the complex and poorly understood bone marrow environment outside of the body . Here , Di Buduo et al . have developed an artificial three-dimensional miniature organ bioreactor system that recreates the key features of bone marrow . In this system , megakaryocytes were grown from patient blood samples , and hooked up to a tissue scaffold made of silk . The cells were able to grow as if they were in their normal environment , and they could shed platelets into an artificial bloodstream . After treating megakaryocytes with drugs to stimulate platelet production , Di Buduo et al . found that the number of platelets recovered from the bioreactor could accurately predict which patients would respond to these drugs in the clinic . This new test system enables researchers to predict how a patient will respond to treatment , and to tailor therapy options to each individual . This technology could also be used to test new drugs for Inherited Thrombocytopenias and other blood-related diseases; if scaled-up , it could also , one day , generate large quantities of lab-grown blood cells for transfusion .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "medicine" ]	2021	Miniaturized 3D bone marrow tissue model to assess response to Thrombopoietin-receptor agonists in patients
Complex memory of personal events is thought to depend on coordinated reinstatement of cortical representations by the medial temporal lobes ( MTL ) . MTL-cortical theta and gamma coupling is believed to mediate such coordination , but which cortical structures are critical for retrieval and how they influence oscillatory coupling is unclear . We used magnetoencephalography ( MEG ) combined with continuous theta burst stimulation ( cTBS ) to ( i ) clarify the roles of theta and gamma oscillations in network-wide communication during naturalistic memory retrieval , and ( ii ) understand the causal relationship between cortical network nodes and oscillatory communication . Retrieval was associated with MTL-posterior neocortical theta phase coupling and theta-gamma phase-amplitude coupling relative to a rest period . Precuneus cTBS altered MTL-neocortical communication by modulating theta and gamma oscillatory coupling . These findings provide a mechanistic account for MTL-cortical communication and demonstrate that the precuneus is a critical cortical node of oscillatory activity , coordinating cross-regional interactions that drive remembering . Detailed , complex memories of personal events can last a lifetime and be brought to mind at will . Current models suggest that this ability depends on the hippocampus and surrounding MTL regions , as well as the coordinated reinstatement of retrieved information from neocortical regions ( McClelland et al . , 1995; Rolls , 2000 ) . The critical role of the MTL in retrieval of detailed personal memories has been extensively demonstrated ( Nadel and Moscovitch , 1997; Rosenbaum et al . , 2008; Scoville and Milner , 1957 ) . However , there is little empirical evidence indicating which neocortical structures are crucially involved in the reinstatement of detailed personal memories . Causal evidence is also lacking for a mechanistic account of how such coordinated reactivation occurs during complex personal memory retrieval . Interactions between MTL and medial parietal regions are thought to be particularly important for representing the spatial context of an event , a central contributor to the vivid recollection of memories ( Burgess et al . , 2001a; Burgess et al . , 2001b; Hassabis and Maguire , 2009; Robin et al . , 2015 ) . While previous studies have shown that MTL and medial parietal structures are associated with spatial aspects of personal memory retrieval ( Freton et al . , 2014; Hebscher et al . , 2018; St Jacques et al . , 2017 ) , they are unable to determine whether these regions are critical for retrieval , or how they may communicate with one another to allow for information transfer . Neural oscillations may be a mechanism by which such widespread regions communicate during memory retrieval . Theta ( 3–7 Hz ) oscillations are hypothesized to mediate MTL-neocortical orchestration during memory retrieval ( Battaglia et al . , 2011 ) , and theta phase coherence between regions has been identified in working ( Payne and Kounios , 2009 ) , spatial ( Kaplan et al . , 2014; Kaplan et al . , 2017 ) , and autobiographical memory ( Eckart et al . , 2014; Foster et al . , 2013 ) . Theta phase may also modulate spatially distributed local gamma oscillations through phase-amplitude coupling ( PAC ) , a form of cross-frequency coupling ( Sirota et al . , 2008 ) . Theta-gamma PAC is thought to be a mechanism for communication between distributed regions during cognitive processes ( Canolty et al . , 2007 ) and has been identified in the human MTL ( Axmacher et al . , 2010; Staudigl and Hanslmayr , 2013 ) and neocortex during memory ( Canolty et al . , 2007; Kaplan et al . , 2014; Sauseng et al . , 2009; van der Meij et al . , 2012 ) . In animals , theta oscillations reflect the organization of complex spatial memories as they unfold and are critical for accurate reinstatement of conceptually meaningful representations ( Wikenheiser and Redish , 2015 ) . Disruption of this oscillatory activity interferes with the integrated representation of complex , temporally extended memories ( Colgin , 2016 ) . Human studies of complex naturalistic memories have thus far only demonstrated the existence of theta phase synchronization as a correlate of memory retrieval ( Foster et al . , 2013; Fuentemilla et al . , 2014 ) . Theta-gamma coupling has only been demonstrated in relatively simple lab-based experiments ( Axmacher et al . , 2010; Canolty et al . , 2007; Kaplan et al . , 2014; Sauseng et al . , 2009 ) and no study has demonstrated its critical contribution to complex mnemonic functions . Importantly , memory for items presented during lab-based experiments recruits different neural substrates than memory for real-life events , which is considered to be more contextually rich , self-focused , and complex ( Chen et al . , 2017; Gilboa , 2004; McDermott et al . , 2009 ) . Here , we aim to ( i ) clarify the roles of oscillations within and across network nodes during retrieval of detailed complex personal memories and ( ii ) understand the causal relationship between these network nodes and oscillatory communication . The present study elucidates the communication between regions involved in autobiographical memory ( AM ) by measuring theta phase coupling and theta-gamma PAC during memory retrieval . Participants performed an AM task in which they were cued with familiar words and rated the subjective quality of memories ( Figure 1 ) . We further use continuous theta burst stimulation ( cTBS ) to examine whether the precuneus is causally involved in oscillatory communication between regions during complex memory retrieval . Previous studies have used parietal neurostimulation to alter autobiographical ( Bonnici et al . , 2018; Thakral et al . , 2017 ) and episodic memory ( Bonnì et al . , 2015; Nilakantan et al . , 2017; Wang et al . , 2014; Wang and Voss , 2015; Yazar et al . , 2014 ) , some of which have shown associated and sustained alterations of neural activity ( Nilakantan et al . , 2017; Wang et al . , 2014; Wang and Voss , 2015 ) . In the present study , we used cTBS to directly suppress neural activity in the precuneus in order to observe system-level changes in activity , specifically in the MTL and other structures that comprise the autobiographical memory network . The precuneus is a highly structurally and functionally connected association area ( Cavanna and Trimble , 2006 ) that demonstrates theta phase coupling during AM ( Fuentemilla et al . , 2014 ) , and has a causal behavioural role in episodic memory ( Bonnì et al . , 2015; Koch et al . , 2018 ) . We therefore predicted that precuneus stimulation would affect neural activity both within the precuneus and in regions functionally connected to the precuneus . Behaviorally , we predicted that stimulation would lead to differences in subjective aspects of complex memory retrieval . Specifically , we predicted that stimulation would alter the perspective from which memories were recollected based on the precuneus’ established role in spatial perspective representations during memory ( Freton et al . , 2014; Hebscher et al . , 2018; St Jacques et al . , 2017 ) . We first investigated whether precuneus stimulation would lead to differences in the quality of memory retrieval by comparing rating scales from the AM task between precuneus and vertex stimulation sessions . Rating scales included memory vividness , ease of recall , and perspective rating ( first- versus third-person perspective ) . We performed a series of analyses of variance ( ANOVAs ) using rating scales for precuneus and vertex sessions as within-subjects factors and session order ( i . e . counterbalancing order: precuneus stimulation or vertex stimulation first ) as the between-subjects factor , to account for potential effects of counterbalancing order on subjective memory . These analyses revealed a significant interaction between session order and session type on vividness ratings ( F ( 1 , 21 ) = 5 . 10 , p=0 . 035 , ηp2 = 0 . 195 ) but not on ease of recall ( F ( 1 , 21 ) = 0 . 84 , p=0 . 369 , ηp2 = 0 . 039 ) or perspective ratings ( F ( 1 , 21 ) = 0 . 10 , p=0 . 755 , ηp2 = 0 . 005 ) . There was a main effect of stimulation type on vividness ratings when accounting for counterbalancing order ( F ( 1 , 21 ) = 4 . 83 , p=0 . 039 , ηp2 = 0 . 187 ) , such that precuneus stimulation led to lower vividness ratings . There was no significant effect of stimulation on effort or perspective ratings ( all p’s > 0 . 28 ) . These findings suggest that the order of counterbalancing significantly interacted with the effects of stimulation on memory vividness . Based on the significant interaction between session order and session type , we performed a post-hoc exploratory analysis on vividness ratings using each participant’s first session only , such that we compared participants with precuneus stimulation first ( n = 12 ) to those with vertex stimulation first ( n = 11 ) . This analysis had a between-subjects design , which several previous parietal-TMS studies have used ( cf . Yazar et al . , 2014; Bonnì et al . , 2015; Wang and Voss , 2015 ) . Independent-samples t-tests revealed a significant difference in vividness ratings , such that precuneus stimulation led to less vivid memories compared to vertex stimulation ( t ( 21 ) = −2 . 34 , p=0 . 030 , CI [−0 . 98 −0 . 06] , d = −0 . 97 ) ( see Figure 2A ) . Although there was no significant interaction between session order and session type on effort rating , we also found a significant difference between stimulation sessions on ease of recall ratings , such that precuneus stimulation led to more effortful recall compared to vertex stimulation ( t ( 21 ) = −2 . 62 , p=0 . 016 , CI [−0 . 98 −0 . 11] , d = −1 . 09 ) ( Figure 2B ) . Comparison of each participant’s second session revealed no significant differences between stimulation sessions on vividness ( t ( 21 ) = 1 . 55 , p=0 . 136 , CI [−0 . 16 1 . 1] , d = 0 . 65 ) or ease of recall ( t ( 21 ) = 1 . 10 , p=0 . 287 , CI [−0 . 28 91] , d = 0 . 46 ) ( Figure 2—figure supplement 1 ) . To gain a better understanding of the oscillatory underpinnings of complex memory retrieval , we examined MEG data at vertex stimulation sessions . The subsequent sections describe results for theta power , phase coupling , and theta-gamma phase-amplitude coupling . We first sought to characterize the oscillatory correlates of complex personal memory retrieval by examining activity at vertex stimulation sessions . We found theta power increases in an extensive network of regions including occipital lobes , precuneus , inferior and superior parietal lobes , retrosplenial cortex , MTL , mPFC , and cerebellum . While previous studies have mainly focused on theta oscillations in a constrained network of regions during autobiographical memory retrieval ( Foster et al . , 2013; Fuentemilla et al . , 2014; Steinvorth et al . , 2010 ) , here we demonstrate that theta power across a widespread network of regions is associated with complex memory recollection . These findings converge with fMRI studies identifying a similar network of brain regions during autobiographical memory retrieval ( for review , see Svoboda et al . , 2006 ) . They also suggest that theta oscillations may be a means of communication between regions in this widespread network , although note that increased power does not imply coupling . To examine communication between these regions we subsequently examined theta phase coupling and theta-gamma PAC . Using a subject-specific seed in the right MTL , we identified a cluster in the occipital lobe that was theta phase synchronized with the seed during early memory elaboration . These findings indicate that the MTL communicates with posterior neocortical regions via theta phase coupling during memory recollection . Two previous studies have demonstrated network-level theta phase coupling as a correlate of autobiographical memory retrieval ( Foster et al . , 2013; Fuentemilla et al . , 2014 ) . Fuentemilla et al . ( 2014 ) identified theta phase coupling between an MTL seed and clusters in the precuneus and mPFC during AM recollection relative to a general semantic knowledge task . Differences in the specific cortical structures identified in our findings and theirs may be related to their use of continuous recordings of personal memories as they unfold ( direct cues ) , whereas we used personalized cues that require indirect retrieval processes . Different baselines ( semantic memory in Fuentemilla et al . , and rest in the present study ) are also likely contributors to the slight differences in the cortical structures identified across studies . Theta phase coupling may provide a scaffold for interregional coordinated activity ( Fell and Axmacher , 2011 ) , whereas local gamma is considered an index of stimulus-specific information processing . We therefore also investigated whether local cortical gamma amplitude is modulated by the phase of MTL theta oscillations . We identified phase-amplitude coupling between MTL theta and precuneus gamma during memory elaboration compared to rest which was sensitive to vividness ratings , but did not find PAC between the MTL and mPFC or TPJ . Interestingly , MTL theta modulated precuneus gamma but the reverse was not true , supporting the MTLs role in coordinating neocortical activity . A number of studies have identified theta-gamma coupling in human lab-based memory tasks ( Axmacher et al . , 2010; Staudigl and Hanslmayr , 2013; Canolty et al . , 2007 ) , but ours is the first to demonstrate this phenomenon in complex naturalistic memory retrieval . It is interesting that MTL theta was coupled with precuneus gamma but not with the other neocortical regions examined . Many previous studies have shown that medial prefrontal and lateral parietal regions are recruited during AM retrieval ( for review , see Svoboda et al . , 2006 ) , and we similarly found robust theta power increases in these regions during recollection . However , our findings do not indicate that local gamma activity in these regions is coordinated by MTL theta , which is unexpected given the hypothesized early role of mPFC during retrieval of self-related memories and its connectivity with the MTL ( Hebscher and Gilboa , 2016; McCormick et al . , 2018 ) . Further research is needed to elucidate the conditions under which communication between these nodes of the network become critical for retrieval . Our findings thus demonstrate that MTL-precuneus oscillatory coupling is important for memory recollection but leave open questions about communication across other nodes in the network . Together these results show that MTL and posterior neocortical regions interact via theta and gamma coupling during complex personal memory recollection , suggesting a specific means of information transfer between these distant regions . Having demonstrated that theta and gamma oscillations mediate the communication between widespread regions during recollection , we next examined whether the precuneus is critically involved in this communication . We show that precuneus stimulation disrupts theta phase coupling between the MTL and occipital lobe , supporting a causal role for this region in theta phase coupling between regions beyond the site of stimulation . These findings are consistent with the idea that theta coordination of network activity mediates integrated representations required for the reinstatement of complex memories . We further found that precuneus stimulation altered MTL-precuneus theta-gamma coupling during memory recollection relative to rest . Precuneus stimulation led to reduced 4 Hz theta modulation of gamma in a cluster that showed increased coupling during vertex stimulation sessions . Notably , MTL-precuneus communication within this cluster was initially correlated with vividness ratings , and stimulation-induced disruption of PAC within the cluster predicted disruption of vividness . These findings indicate that precuneus stimulation disrupted MTL-precuneus communication which was initially present during recollection , communication which was crucial for the subjective vividness of memories . Precuneus stimulation also increased 5 Hz theta modulation of gamma relative to vertex stimulation , an effect that corresponded with increased 5 Hz coupling during precuneus stimulation sessions . This stimulation-induced increase in 5 Hz coupling was not associated with change in subjective memory . One possible explanation for this stimulation-induced increase in coupling is that our cTBS protocol , consisting of high-frequency bursts applied at 5 Hz , entrained theta activity at 5 Hz . Indeed , previous studies have found that neurostimulation can synchronize oscillations to a specific frequency , although these studies used electrical stimulation ( Neuling et al . , 2012; Zaehle et al . , 2010 ) . Note , however , that this explanation is speculative and further research is needed to determine the feasibility of using cTBS to entrain neural oscillations . We found tentative evidence to suggest that precuneus stimulation alters subjective memory vividness and ease of recall . Interestingly , we found that the order of counterbalancing significantly interacted with the effects of stimulation on memory vividness . This order effect prompted us to perform an exploratory between-subjects analysis comparing participant’s first session , which revealed that inhibitory precuneus stimulation led to decreased memory vividness and more effortful recall compared to vertex stimulation . Comparison of participant’s second session revealed no significant effect of stimulation , although the pattern of results was opposite to that of the first session . One possible explanation for these findings is that , due to the subjective nature of the rating scales , memories recalled in the second session were rated relative to memories recalled in the first session . We also found that counterbalancing order did not influence the effects of stimulation on phase coupling or PAC , further suggesting that this order effect may have been specifically related to the subjective nature of the rating scales . Very few studies have used neurostimulation to examine the causal role of the precuneus in memory . One study found that high-frequency repetitive TMS to the precuneus modestly enhanced episodic memory in patients with Alzheimer’s disease ( Koch et al . , 2018 ) , while an earlier study from the same group found that precuneus cTBS enhanced source memory retrieval ( Bonnì et al . , 2015 ) . The only two TMS studies of AM to date both targeted the angular gyrus and reported reduced internal episodic details compared to vertex stimulation . Thakral et al . ( 2017 ) additionally found increased external semantic details , while Bonnici et al . ( 2018 ) found a reduction in the number of events recalled from a first-person perspective ( Thakral et al . , 2017 ) . Our findings add to this limited literature by demonstrating a causal role for the precuneus and precuneus-MTL interactions in subjective aspects of complex personal memory retrieval . Contrary to the proposed importance of MTL-precuneus communication in representing spatial information , precuneus stimulation did not affect the tendency to recall events from a first-person perspective . While we previously found that precuneus volume is positively associated with recalling autobiographical memories from a first-person perspective ( Hebscher et al . , 2018 ) , the present study does not support a causal role for the precuneus in this function . One interpretation of these results is that the precuneus is involved in egocentric processing during complex memory retrieval , but not causally so , perhaps , due to its interactions with other posterior parietal regions like the angular gyrus . As described above , one recent study found that angular gyrus stimulation reduced the number of autobiographical memories experienced from a first-person perspective , while also reducing the number of internal details recalled ( Bonnici et al . , 2018 ) . The authors interpret this result as implicating the angular gyrus in integrating memory features within an egocentric framework to enable the subjective experience of remembering . Other studies have implicated both the angular gyrus and precuneus in shifting visual perspectives during autobiographical memory ( Iriye and Jacques , 2018; St Jacques et al . , 2017 ) . Thus , it may be the case that interactions between these regions are important for representing events from an egocentric perspective , with the angular gyrus playing more of a critical role than the precuneus . Future TMS studies are needed to clarify the nature of the precuneus’ involvement in egocentric processing . We show that MTL and posterior neocortex interact via theta and gamma oscillatory activity during complex personal memory retrieval . Our results support the notion that theta phase coupling and theta-gamma phase-amplitude coupling mediate MTL-neocortical coordination during memory processes . We further show that precuneus stimulation alters oscillatory activity and subjective memory , demonstrating a causal role for this region . Our results indicate that continuous theta burst stimulation can be used to causally alter oscillatory activity , and that these effects are long-lasting . Together , these findings demonstrate the feasibility of using cTBS and MEG to study complex , naturalistic memory functions . Twenty-three healthy young participants ( 14 females , mean age = 26 . 3 , range = 19–36 ) were tested on a within-subjects combined TMS-MEG paradigm . Sample size was determined by an a priori power analysis based on a previous TMS-EEG study of episodic memory ( Nilakantan et al . , 2017 ) . Participants were recruited from the Rotman Research Institute’s healthy volunteer pool . Participants had completed an average of 16 . 4 ( range = 14–21 ) years of formal education , were all right-handed , native or fluent English speakers , had normal or corrected-to-normal vision , and were free from a history of neurological illness or injury , psychiatric condition , substance abuse , or serious medical conditions . Based on TMS safety guidelines ( Rossi et al . , 2009 ) , participants were excluded if they had a history of losing consciousness ( fainting ) , had a prior experience of a seizure , or had a diagnosis or family history of epilepsy . All participants provided informed consent prior to participating in the experiment in accordance with the Rotman Research Institute/Baycrest Hospital ethical guidelines . Participants received cTBS to their left precuneus and to a control region ( vertex ) on separate days , at least 24 hr apart ( mean = 5 . 4 days ) . Immediately following cTBS , participants completed the AM task inside the MEG scanner which was located nearby . All participants completed the MEG scan within an average of 27 . 4 ( SD = 3 . 8 ) min measured from the end of cTBS . Average time between the end of cTBS and start of MEG was 6 . 04 ( SD = 1 . 5 ) min . Anatomical MRIs for each participant were collected in a separate session . At least 48 hr prior to the study , participants provided the names of familiar places , objects and people in an online interview . These items were used as cues because they are elements that commonly make up an event ( Addis et al . , 2009; Burgess et al . , 2001b ) . Participants were instructed to name the first 20 items that came to mind and to limit items to those encountered within the past year . Based on the online interview , 60 cue words were created for each participant , 20 per category . These were randomly divided between the two stimulation sessions so that each session included 30 cues ( 10 of each cue ) , and each session was broken down into three runs to be used in the MEG scanner . E-Prime 1 . 2 software was used to display the items and collect response data . Items were presented in a randomized order . Participants were instructed to use the words as cues to recall personal specific events that had occurred within approximately the last year , not including the past week . Specific events were defined as ‘past events from a specific time and place for which you were personally involved . ’ Cue words were displayed for a maximum of 10 s and participants were instructed to retrieve a specific past event related to the cue as quickly as possible . Participants were asked to press a button on the response box corresponding to their right index finger as soon as a memory came to mind . Trials in which no memory was retrieved ( unsuccessful trials ) were discarded . The retrieval phase was terminated when a memory was retrieved , after which participants saw a slide asking ‘What was the very first thing that came to mind’ , and had to choose one of the following four options: person , object , place , other . An elaboration phase followed in which participants were prompted to imagine the event in as much detail as possible for 8 s . Next , participants rated the memory on four scales aimed at measuring different phenomenological characteristics of the memory . They were given a maximum of 5 s per rating scale . Participants were asked to rate the effort required to bring the event to mind ( 1 = very easy , 6 = very effortful ) , feelings of re-experiencing the event ( 1 = not at all , 6 = completely ) , recall of setting ( 1 = not at all , 6 = distinctly ) , and perspective ( 1 = saw event through my own eyes , 6 = saw myself from an external perspective ) ( Scales adapted from Addis et al . , 2007; Arnold et al . , 2011 ) . Participants were instructed to rate the experience of remembering and not the event itself . Response options on the screen appeared in square boxes representing the response box used in the MEG scanner , with each box representing one of the four buttons for each hand ( excluding the thumb of each finger ) . Participants completed practice trials outside of the MEG to familiarize themselves with the task before moving on to the test trials . See Figure 1 for a depiction of the autobiographical memory task . Participants received cTBS to their left precuneus ( MNI −14 , –66 , 56 ) and to a control region ( vertex; MNI 0 , –15 , 74 ) on separate days . The order of these sessions was counterbalanced . Participants were blind to the type of stimulation they received ( precuneus or vertex ) and later interviews indicated they could not distinguish between the two . The precuneus was chosen as a target region based on this regions’ involvement in AM and in representing spatial perspective during AM ( Freton et al . , 2014; Hebscher et al . , 2018 ) . We chose the left precuneus based on evidence showing that episodic/autobiographical memory is predominately associated with left-lateralized parietal activity ( Kim , 2011; Rugg and Vilberg , 2013; Shimamura et al . , 2011 ) . The left precuneus target region was selected based on a custom meta-analysis of 13 studies with the keyword ‘egocentric’ using NeuroSynth ( neurosynth . org ) . Within this map , the target region was selected by choosing the coordinates with the highest z-score that would be the most accessible with TMS ( the most superficial area ) . Prior to stimulation , both stimulation site coordinates were warped from MNI to individual space and the stimulation site was chosen based on individual anatomy from whole-brain anatomical MRIs as the most superficial region that was closest to these coordinates . At the beginning of the first stimulation session , resting motor threshold ( RMT ) was measured for each participant as the lowest intensity that produced motor evoked potentials ( MEPs ) above 50 µV in 5 out of 10 trials , recorded from the right first dorsal interosseous muscle . A Brainsight frameless stereotaxic neuronavigation system ( Rogue Research , Montreal , Quebec , Canada ) was used to target the selected stimulation sites . Three anatomical landmarks located on the face were used to co-register the anatomical MRI to the participant’s head . An infrared camera ( Polaris Vicra , Northern Digital ) recorded sensors attached to the participant and the TMS coil , allowing for real-time tracking of the TMS coil over the participant’s MRI . A biphasic Super-Rapid Stimulator with a 70 mm air-cooled Figure 8 coil ( Magstim Co . , Whitland , Dyfed , UK ) was used to deliver a modified continuous theta burst stimulation ( cTBS ) at 80% RMT , lasting for approximately 40 s . The cTBS protocol consisted of 600 pulses arranged into bursts delivered every 5 Hz ( 200 ms ) , with each burst containing three pulses delivered at 30 Hz . The coil was positioned perpendicular to the stimulation site . Although standard cTBS protocols are delivered at 50 Hz , we decided to lower the burst frequency to 30 Hz due to limitations of the coil circuitry , leading to overheating at high intensities . Reducing the frequency of stimulation allowed us to stimulate at a higher intensity than would be possible at 50 Hz . Similar protocols have previously been shown to induce stronger MEP suppression compared to 50 Hz cTBS at a reduced intensity ( Wu et al . , 2012 ) . MEG was recorded in a magnetically shielded room at the Rotman Research Institute with a 151-channel whole-head system with first order axial gradiometers ( CTF MEG , Coquitlam , BC , Canada ) ( VSM MedTech Inc ) , at a sampling rate of 625 Hz . Participants sat in an upright position and the behavioural task was projected onto a screen in front of them . For the first four participants , MEG data was recorded continuously during the 30 min behavioral task . We subsequently divided the behavioral task into three equal blocks approximately 10 min in length in order to reduce overall measures of head movement . To further minimize head movement , a towel was inserted to provide a tighter fit within the helmet . Head position was tracked at the beginning and end of each recording block by coils placed at three fiducial points on the head . Average head position across runs was used for source estimation and was co-registered with fiducial points marked on the anatomical MRI . After acquisition , continuous signals were divided into epochs corresponding to each trial . SAM and wPLI maps were thresholded at p<0 . 005 using a minimum cluster size to keep the family-wise error rate at p<0 . 05 . Cluster sizes were determined based on the spatial smoothness of the data using Monte Carlo simulations from the 3dClustSim tool in AFNI using FWHM values calculated based on noise maps produced by conducting a voxel-wise t-test on single-subject maps comparing the pre-trial rest period between two randomly selected cue-types ( locations and people ) . As different maps have different smoothness values , we obtained cluster sizes separately for each map . Cluster thresholds were 78 voxels ( FWHM 21 . 8 ) for SAM theta maps and 24 voxels ( FWHM 11 . 6 ) for wPLI maps .	When you recall an event from your past , such as a meal you ate last week , many regions of your brain become active . The coordinated activity of these regions enables you to recall the event in detail . This coordination depends on rhythmic waves of electrical activity called neural oscillations . These arise whenever large numbers of neurons synchronize when they fire . Electrodes on the scalp can be used to measure neural oscillations . Recordings show that the number of times each wave repeats per second ( also known as the frequency of the oscillation ) , varies from one brain region to the next . Two types of oscillations are particularly important for memory: theta waves and gamma waves . Theta waves repeat between three and seven times every second , and help coordinate activity between areas of the brain that are far apart . Gamma waves are faster , repeating 65 to 85 times per second , and help to support activity within individual regions of the brain . Importantly , theta and gamma waves also interact . Hebscher et al . set out to understand whether interactions between theta and gamma waves help us to recall personal memories . Healthy volunteers were asked to recall memories in response to cues such as ‘my kitchen’ , while sitting inside a brain scanner . As predicted , interactions between theta and gamma waves contributed to memory recall . Theta waves recorded from the medial temporal lobe , a region deep within the brain , altered gamma waves in another area called the precuneus . The latter forms part of the inner surface of the brain where the two hemispheres face one another , and is important for memory vividness and visual imagery . Hebscher et al . briefly disrupted the activity of the precuneus by applying harmless magnetic fields to the scalp above it . Doing so altered theta-gamma interactions across the brain , which was related to reduced vividness of the memories . Remembering events from our past is fundamental to our sense of self and our interactions with others . The results presented by Hebscher et al . show that reducing the activity of a single brain region , the precuneus , impairs memory recall . It does so by disrupting interactions between oscillations throughout the brain . This raises the possibility that stimulating the brain to enhance – rather than disrupt – oscillations could have the opposite effect and improve memory . Future studies could investigate whether enhancing oscillations could help to treat memory disorders .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "neuroscience" ]	2019	A causal role for the precuneus in network-wide theta and gamma oscillatory activity during complex memory retrieval
A central question to biology is how pathogenic bacteria initiate acute or chronic infections . Here we describe a genetic program for cell-fate decision in the opportunistic human pathogen Staphylococcus aureus , which generates the phenotypic bifurcation of the cells into two genetically identical but different cell types during the course of an infection . Whereas one cell type promotes the formation of biofilms that contribute to chronic infections , the second type is planktonic and produces the toxins that contribute to acute bacteremia . We identified a bimodal switch in the agr quorum sensing system that antagonistically regulates the differentiation of these two physiologically distinct cell types . We found that extracellular signals affect the behavior of the agr bimodal switch and modify the size of the specialized subpopulations in specific colonization niches . For instance , magnesium-enriched colonization niches causes magnesium binding to S . aureusteichoic acids and increases bacterial cell wall rigidity . This signal triggers a genetic program that ultimately downregulates the agr bimodal switch . Colonization niches with different magnesium concentrations influence the bimodal system activity , which defines a distinct ratio between these subpopulations; this in turn leads to distinct infection outcomes in vitro and in an in vivo murine infection model . Cell differentiation generates physiological heterogeneity in clonal bacterial infections and helps to determine the distinct infection types . Nosocomial pathogens often cause a broad range of diseases using diverse virulence factors , such as production of tissue-damaging toxins or production of adhesins during biofilm formation ( Bush et al . , 2011 ) . Staphylococcus aureus is one such pathogen that is able to cause different types of life-threatening infections in hospital settings , from acute bacteremia to endocarditis , pneumonia and chronic biofilm-associated infections in prosthetic devices ( Otto , 2012 ) . The underlying cellular processes that enable S . aureus to provoke these disparate types of infections is likely driven by host-microbe interactions ( Casadevall et al . , 2011 ) , in which specific , yet-to-be-described extracellular signals play a role to generate distinct , locally defined types of infections ( Veening et al . , 2008; López and Kolter , 2010 ) . Determining the cellular processes and the nature of the extracellular signals that define the different infection outcomes is crucial for understanding how difficult-to-treat bacterial infections develop and for improving strategies to overcome antimicrobial resistance . In S . aureus , infection outcome is controlled by the agr quorum sensing program , which is autoactivated in response to the self-produced extracellular signal AIP ( autoinducing peptide ) ( Recsei et al . , 1986 ) . AIP binds to the AgrC histidine kinase membrane receptor and activates its cognate regulator AgrA via phosphorylation ( Figure 1A ) . AgrA~P induces changes in cellular gene expression that results in rapid bacterial dispersion in the host and acute bacteremia ( Thoendel et al . , 2011 ) . Dispersion of S . aureus requires upregulation of surfactant phenol-soluble modulins ( psmα and psmβ ) , which are amphipathic small peptides that contribute to bacteria detachment ( Li et al . , 2009a; Peschel and Otto , 2013 ) and destabilize cell membranes , rendering them cytotoxic to host cells . Modulins are usually expressed during acute infections , as well as hemolytic toxins ( hla , hlb , hlg ) that facilitate tissue disruption during septicemia ( Recsei et al . , 1986 ) . In contrast , agr activation indirectly downregulates the icaADBC operon genes needed to synthesize the extracellular polysaccharide matrix that protects cells within a biofilm ( PNAG or PIA ) , as well as several adhesion proteins ( SpA and other MSCRAMM proteins ) responsible for cell aggregation/attachment during biofilm formation ( Recsei et al . , 1986; Boles and Horswill , 2008; Peng et al . , 1988 ) . Biofilms , which are associated with untreatable chronic infections , protect bacteria from antibiotics and host defenses ( Lewis , 2008; Lopez et al . , 2010; Nadell et al . , 2009; Parsek and Singh , 2003 ) . The S . aureus agr quorum sensing system antagonistically regulates the activation of planktonic and biofilm-associated lifestyles ( Recsei et al . , 1986; Boles and Horswill , 2008; Peng et al . , 1988 ) , which contribute to the development of acute and chronic infection outcomes , respectively . A large number of positive and negative regulators controls agr expression . Among those , the agr system is inhibited by the σB sigma factor ( Bischoff et al . , 2001 ) . Activation of σB occurs during early stationary phase ( Senn et al . , 2005 ) in response to distinct types of cellular stresses ( Geiger et al . , 2014; Geiger and Wolz , 2014; Kästle et al . , 2015 ) . σB triggers a general stress response that affects expression of a number of virulence and stress-response genes and indirectly represses agr ( Thoendel et al . , 2011 ) . Thus , σB antagonizes the influence of the agr system on virulence factor expression ( Senn et al . , 2005; Pané-Farré et al . , 2006 ) and biofilm formation ( Bischoff et al . , 2001; Kullik et al . , 1998 ) . This understanding of agr-mediated antagonistic regulation of chronic and acute S . aureus infection outcomes was built on comparative analyses of clinical isolates and characterization of infection related mutants . Through this approach , agr-defective isolates are frequently identified from chronic infections as these mutants usually show reduced hemolytic activity and develop robust biofilms ( Fischer et al . , 2014; Goerke and Wolz , 2010; Grundmeier et al . , 2010; Hirschhausen et al . , 2013; Savage et al . , 2013 ) . In addition , agr dysfunction is frequently correlated with chronic persistent S . aureus infections ( Fowler et al . , 2004 ) such as small colony variants ( SCV ) . SCV have exceptionally low agr expression levels ( Kahl et al . , 2016 ) and high expression of biofilm-related genes ( Tuchscherr et al . , 2010 ) . High σB expression is important for SCV phenotype acquisition ( Mitchell et al . , 2013 ) , because sigB mutants do not generate SCV ( Tuchscherr and Löffler , 2016; Tuchscherr et al . , 2015 ) . However , whether the capacity of nosocomial pathogens , such as S . aureus , to cause distinct types of infections is restricted to the emergence of genetic variants is still unclear . Staphylococcus aureus cells are exposed to a variety of local environmental signals during the course of an infection that can influence bacterial gene expression and thus , their infective potential in a given infection niche . These signals include , but are not limited to , changes in nutrient availability , temperature , pH , osmolarity or , oxygen concentration . Staphylococcus aureus might be able to respond collectively to these extracellular cues to adapt its behavior in a fluctuating environment ( Münzenmayer et al . , 2016 ) , allowing staphylococcal communities to generate distinct , locally defined infection types without modification of the bacterial genome ( Veening et al . , 2008; López and Kolter , 2010 ) . It has been hypothesized that changes in bacterial virulence potential are a response to local concentrations of tissue-specific signals , which have an important role in determining infection outcome ( Cheung et al . , 2004 ) . Yet how bacteria prepare for such unpredictable environmental changes is a question that remains unanswered . A fundamental feature of microbial cells is their ability to adapt to diverse environmental conditions by differentiating into specialized cell types ( Arnoldini et al . , 2014; López et al . , 2009; Veening et al . , 2005 ) . In most cases , the extracellular cues are responsible for defining coexisting cell fates in bacterial populations . Cell fates are genetically identical and phenotypically distinct bacterial subpopulations that express heterogeneously different sets of genes and have distinct functions within the microbial community ( George et al . , 2015 ) . A classical example of this is the bacterial response to antibiotics . Antibiotics kill most S . aureus cells , but it is frequent to observe a small subpopulation of genetically identical but antibiotic-persister cells that can cause recurrent infections in a post-antibiotic period ( Bigger , 1944; Lewis and cells , 2007 ) . The relative simplicity of agr-mediated antagonistic regulation of planktonic and biofilm-associated lifestyles provides a natural model to analyze how S . aureus cells collectively establish acute or chronic infection lifestyles and to identify extracellular factors that influence activation of the cellular program that leads to prevalence of one infection program over the other . Here , we report a bimodal behavior in the agr quorum sensing system that antagonistically regulates the differentiation of two genetically identical but physiologically distinct specialized cell types in S . aureus communities . One cell type contributes to the formation of biofilms responsible for chronic infections , whereas a second was constituted by dispersed cells that produced the toxins that contribute to an acute bacteremia . These subpopulations were present in S . aureus communities at different ratios depending on growth conditions , which contributed to determining the outcome of infection . We found that colonization niches with higher Mg2+ concentrations , which is inherent in tissues colonized preferentially by S . aureus ( Günther , 2011; Jahnen-Dechent and Ketteler , 2012 ) , influenced the bimodal switch and increased the size of the subpopulation of cells specialized in biofilm formation , as Mg2+ binding to teichoic acids increases cell wall rigidity and triggered a σB stress-induced genetic cascade that downregulates agr . In a mouse model , bacterial cell differentiation occurred during in vivo infections and the Mg2+ concentration in infected organs influenced collective bacterial behavior in simultaneous progress to a biofilm-associated chronic infection or a disperse bacteremia . This study shows that cell differentiation in S . aureus helps to diversify the types of infections that arise simultaneously from an infection caused by a clonal population of bacteria . We explored the role of agr-mediated antagonistic regulation of planktonic and biofilm-associated lifestyles in S . aureus aggregates growing on Mg2+-enriched TSB medium ( TSBMg ) , in which most S . aureus clinical isolates develop robust multicellular aggregates ( Koch et al . , 2014 ) . To study biofilm gene expression , we introduced transcriptional fusions of biofilm-associated ica/spa genes . The ica operon is responsible for production of the exopolysaccharide polymeric matrix ( PNAG or PIA ) that lends consistency to the biofilm . The spa gene encodes a cell wall-anchored adhesion protein , adhesin that is responsible for S . aureus cell aggregation and attachment to surfaces during biofilm formation ( Recsei et al . , 1986; Boles and Horswill , 2008; Peng et al . , 1988 ) . To monitor planktonic gene expression , we generated transcriptional fusions of psmα and psmβ genes . These genes code for small peptides , the phenol-soluble modulins , whose expression depends directly on agr . Due to their surfactant properties , PSMα and PSMβ act as cytolytic toxins that contribute to bacterial dispersion and play an important role in acute staphylococcal infections ( Li et al . , 2009a; Peschel and Otto , 2013 ) ( Figure 1A ) . These reporters were introduced into neutral loci in the S . aureus chromosome to ensure expression of a single reporter copy in each cell ( Yepes et al . , 2014 ) ; we monitored their expression at the single-cell level in S . aureus aggregates using quantitative analysis of fluorescence microscopy images ( Figure 1B and Figure 1—figure supplement 1A–B ) . All reporters showed bimodal expression and indicated the bifurcation of two cell subpopulations in S . aureus aggregates , one with lower and another with higher fluorescence levels . This bimodal expression pattern differed from the unimodal expression of the housekeeping and agr-independent gene dnaA , used as control reporter ( Figure 1B and Figure 1—figure supplement 1A ) . Cultures of different S . aureus isolates showed bimodal expression of these reporters ( Figure 1B and Figure 1—figure supplement 1C ) , which suggests that cell differentiation is a general phenomenon in S . aureus . Monitoring the temporal dynamics of the subpopulations that bifurcated during the development of the microbial aggregates from an initial inoculum in TSBMg revealed larger subpopulations of ica- and spa-expressing cells in early developmental stages ( ~72 hr; Figure 1B ) , compared to the size of these subpopulation at later growth stages ( ~120 hr ) . In contrast , the size of the subpopulations of psmα/β-expressing cells increased concomitantly with time , consistent with the reported antagonistic regulation of ica/spa and psmα/β by agr ( Recsei et al . , 1986; Boles and Horswill , 2008; Peng et al . , 1988 ) . We generated strains labeled with different pairwise combinations of these reporters , which were both introduced into neutral loci in the S . aureus chromosome; this allowed quantitative analysis of fluorescence microscopy images to examine simultaneous expression in S . aureus microbial communities ( Figure 1C and Figure 1—figure supplement 1D ) . This approach indicated coexpression of ica with spa and of psmα with psmβ in two distinct cell subpopulations , showing the bifurcation of two distinct subpopulation of cells specialized in expressing ica , spa and other biofilm-related genes ( BRcells ) and cells expressing psmα/β dispersion-related genes ( DRcells ) . The differential expression of agr-related genes in the distinct cell types led us to analyze the molecular mechanism of agr-mediated cell differentiation . The agr system is autoactivated once the extracellular AIP concentration reaches a given threshold ( 10–14 μM ) ( MDowell et al . , 2001 ) and is inhibited by σB induction ( Bischoff et al . , 2001 ) . Following agr activation , AgrA~P directly upregulates psmα/β expression ( Queck et al . , 2008 ) and binds to the two adjacent divergent promoters P2 and P3 , which trigger expression of RNAII and RNAIII transcripts , respectively ( Koenig et al . , 2004 ) . RNAII upregulates the agrBDCA operon , which encodes the agr signal transduction cascade , including the AIP signal , the AgrC sensor kinase and its AgrA cognate regulator . Therefore , AgrA~P binding to the P2 promoter constitutes a positive feedback loop in which AgrA~P regulator induces expression of the agrBDCA operon , which encodes the entire agr signal transduction cascade ( Thoendel et al . , 2011; Queck et al . , 2008 ) . Bimodal gene expression in microbial populations is usually triggered by a positive feedback loop in which a gene product induces its own expression . We hypothesize that , once a certain AgrA~P threshold is reached in a cell , AgrA~P induces its own expression and these cells maintain high AgrA~P levels . Thus , AgrA~P and AgrA-controlled genes will thus be activated in that cell , including upregulation of RNAIII via activation of the P3 promoter . RNAIII positively regulates a pool of agr-dependent genes that encode the cytotoxic toxins and virulence factors responsible for acute infection ( Koenig et al . , 2004 ) . Activation of the P3 promoter leads DRcells to specialize in dispersion and virulence . In contrast , cells that do not achieve the AgrA~P expression threshold needed to induce the positive feedback mechanism will not induce P3 promoter expression . In these cells , genes normally repressed by AgrA~P will be upregulated , including biofilm-related genes , which licenses cells to differentiate as biofilm-producing BRcell types . To determine whether activation of the agr positive feedback loop is sufficient to generate bimodality in a bacterial population , we genetically engineered an orthogonal agr system in B . subtilis in which the agr positive feedback loop was isolated from its native complex regulatory network ( Audretsch et al . , 2013 ) and thus exempt from interference from additional staphylococcal regulatory inputs . In this orthogonal system , B . subtilis harbored Ppsmα-yfp or Ppsmβ-yfp reporters and expressed the membrane kinase AgrC and its cognate regulator AgrA under the control of the AgrA-inducible P2 promoter ( Figure 2A ) . The orthogonal system does not express staphylococcal σB , and σB from B . subtilis did not interfere with the agr system , since we detected similar reporter expression in wild type ( WT ) and ΔsigB strains ( Figure 2—figure supplement 1A ) , ensuring the absence of the agr inhibitory input signal . We used this orthogonal system to identify the minimal components necessary for bimodal expression of agr-related genes , using psmα/β expression as readout for agr activity ( Figure 2B ) ( Zhang et al . , 2015 ) . Activation of the agr positive feedback loop in the orthogonal system requires addition of purified AIP ( 10 μM ) to B . subtilis cultures ( AIP contains a thiolactone ring thus it cannot be synthetically produced ) . AIP addition activates AgrC sensor kinase , which phosphorylates the AgrA regulator . The AgrA~P active form binds the P2 promoter to express high AgrA~P levels and activate the positive feedback loop; this resulted in bimodal expression of chromosomally integrated Ppsmα-yfp or Ppsmβ-yfp reporters , with a cell subpopulation in which reporter expression increased during a transition in which cells switched from off to on . In contrast , chromosomally integrated Pica-yfp and Pspa-yfp reporters , which are not controlled directly by the AgrA~P , did not activate in response to added AIP . In a similar manner , we did not detect fluorescence in control experiments with strains lacking the AgrC-AgrA system , or when no AIP was added ( Figure 2B and Figure 2—figure supplement 1A ) , demonstrating that stochastic expression of these reporter genes does not account for bifurcation of the cell populations . These results indicate that the minimal agr genetic program harbored in the orthogonal system acts as an autonomous program for cell differentiation in bacterial populations . Identification of the molecular mechanism that leads agr to act as an autonomous program to define cell fate bifurcation in S . aureus required analysis of the agr-signaling cascade in the presence of AIP . We therefore generated two additional orthogonal systems to monitor P2 ( PRNAII-yfp ) and P3 ( PRNAIII-yfp ) activation independently ( Figure 2C ) . In response to exogenous AIP ( 10 μM ) , these systems showed a transition period during which P2 and P3 cells switched on after which subpopulation sizes remained constant . The P2-expressing cell subpopulation differentiated earlier , however , and showed a more intense fluorescence signal in a larger cell subpopulation over time compared to the orthogonal system that differentiated P3-expressing cells . This result suggests that P2 promoter is more sensitive than P3 to agr activation , a characteristic feature of positive feedback loops in bimodal systems ( Siebring et al . , 2012 ) . Based on these results , we hypothesized that the bimodal behavior of agr and thus , cell differentiation in S . aureus , relies on the differential affinity of AgrA~P for P2 and P3 promoters . P2 thus activates the feedback loop at lower AgrA~P concentrations and only in a subpopulation of cells . These cells contain high AgrA~P levels , which licenses them to trigger the less-sensitive P3 promoter and induce the agr regulon , leading the cells to specialize in dispersion and virulence and become DRcells . Cells that express P2 below the threshold cannot activate the agr feedback loop and are thus unable to induce P3 promoter expression . In this subpopulation , agr-repressed genes are upregulated , including genes involved in biofilm formation , which licenses them to differentiate as BRcells . We tested this hypothesis by first analyzing the dynamics of the agr positive feedback loop using mathematical modeling coupled to computational simulations ( Figure 2—figure supplement 1B–E ) ( Chong et al . , 2014; Golding et al . , 2005 ) . The full network dynamics remained constant , since the orthogonal system was similar for all promoters tested . We optimized the promoter-specific rates; Kon ( association ) and Koff ( dissociation ) denote interaction with AgrA~P and Kt denotes reporter transcription ( Figure 2—figure supplement 1B ) ( Goñi-Moreno et al . , 2016 ) . The correct combination of these promoter rates sufficed to explain bimodal fluorescence distribution . Although the simulations considered AgrA~P saturation , we consistently detected bimodal reporter expression , with on and off cell subpopulations always present . Simulations favored a model in which P2-activated positive feedback loop induces P3-driven bimodal expression in response to AIP concentration or autoactivation time ( Figure 2—figure supplement 1C–E ) , suggesting that DRcells resulted from sequential P2 and P3 promoter activation . We tested this model experimentally in a dual orthogonal system harboring P2 ( PRNAII-cfp ) and P3 ( PRNAIII-yfp ) reporters expressed as two adjacent transcriptional units transcribed in opposite directions , similar to the chromosomal organization in the S . aureus genome ( Figure 2D and Figure 2—figure supplement 1F ) . We used flow cytometry analysis with simultaneous detection of CFP and YFP signals to determine quantitatively whether the P2-expressing subpopulation becomes P3-expressing cells over time after AIP addition ( 10 μM ) . At 4 hr post-AIP induction , we detected a cell subpopulation that expressed P2; a fraction of this subpopulation activated P3 at later times ( 6 hr ) . The subpopulation of P3-expressing cells increased over time until it expressed P2 and P3 promoters uniformly . Cells that expressed only the P3 promoter were not detected . This is consistent with our hypothesis that P2-mediated activation of the agr positive feedback loop is necessary to increase AgrA~P levels , which in turn induces expression of the less-sensitive P3 promoter in these cells . The molecular mechanism for bimodal gene expression thus relies on the differential AgrA~P affinity for P2 and P3 promoters . P2 is very sensitive and triggers the agr positive feedback loop , whereas P3 induces expression of virulence genes and is necessary for DRcell specialization . In the following section , we used this information to demonstrate that the self-regulatory activity of AgrA~P via binding to the P2 promoter is essential for triggering S . aureus cell differentiation while other additional cues that feed into the agr switch only modulate the activity of the system . Once the agr switch responsible for BRcell and DRcell differentiation is activated , distinct extracellular cues can arise from the niche to feed into the agr bimodal switch and modulate its activity . For instance , BRcell and DRcell subpopulations are detected at different ratios in TSB and TSBMg cultures . We hypothesized that variations in extracellular input signals would affect agr bimodal behavior and produce distinct outcomes in the bimodal system . This would define a distinct DRcell:BRcell ratio , which could have important clinical implications for the definition of infection outcomes . As the difference between TSB and TSBMg media resides in Mg2+ supplementation , we tested the effect of extracellular Mg2+ on the response of the agr-repressor σB , which is activated by environmental stresses . In TSBMg medium , qRT-PCR analysis showed increased expression of the σB-dependent stress gene asp23 ( alkaline shock protein ) ( Gertz et al . , 2000 ) ( Figure 3A ) and of staphyloxanthin , the pigment that gives S . aureus its typical yellow color and whose expression is regulated directly by σB ( Gertz et al . , 2000; Giachino et al . , 2001 ) ( Figure 3—figure supplement 1A ) . TSBMg medium also induced biofilm formation ( Figure 3B ) . Biofilm formation likely occurred via agr inhibition because the ΔsigB strain did not form biofilm and the biofilm formation phenotype was partially recovered in a ΔsigBΔagr double mutant ( Figure 3C ) . Thus , the Mg2+ signaling cascade acts on agr downregulation via σB activation to increase BRcell subpopulation size . This is consistent with the fact that biofilm-associated S . aureus colonization generally occurs in Mg2+-enriched niches such as bone and kidney , in which chronic staphylococcal infections often develop ( Günther , 2011; Jahnen-Dechent and Ketteler , 2012; Elin , 2010 ) . By contrast , tissues unintentionally depleted of Mg2+ are prone to acute staphylococcal infections , as Mg2+ sequestration from tissues due to tampon use was associated with an outbreak of staphylococcal toxic shock syndrome in women in the USA ( Parsonnet et al . , 1996; Schlievert , 1985 ) . We were prompted to analyze the molecular mechanism whereby extracellular Mg2+ regulates the agr bimodal switch and increases the BRcell subpopulation . Biofilms occur in TSB supplemented with Mg2+ but not with other cations ( Koch et al . , 2014 ) , which suggested that Mg2+ is a specific extracellular trigger for BRcell differentiation . Mg2+ has a function in stabilizing the Gram-positive bacterial cell wall , which is decorated with phosphate-rich teichoic acids ( TA ) that contribute to membrane integrity ( Heptinstall et al . , 1970 ) . To alleviate electrostatic repulsive interactions between neighboring phosphates , TA preferentially bind Mg2+ cations , to form a consolidated network that strengthens cell envelope rigidity ( Heckels et al . , 1977; Lambert et al . , 1975a; Swoboda et al . , 2010 ) . We therefore hypothesized that Mg2+ in TSBMg stabilizes S . aureus TA and increases cell wall rigidity , which cues σB activation . We tested this hypothesis using atomic force microscopy ( AFM ) to monitor S . aureus cell wall structural rigidity in vivo , comparing single cells grown in TSB and TSBMg media ( Figure 3D ) ( Saar-Dover et al . , 2012; Touhami et al . , 2004 ) . AFM detects forces acting between a sharp nanoscale cantilever and the bacterial cell wall; after pressure , the cantilever deflects and force can be quantified ( Dufrêne , 2014; Formosa-Dague et al . , 2016 ) . We detected greater rigidity in cells grown in TSBMg medium than those grown in TSB medium . The ΔdltA mutant was used as control , since the DltA-E machinery is responsible for D-alanylation of TA , which introduces positively charged amines and thus prevents repulsive interactions between neighboring TA ( Perego et al . , 1995 ) , similar to the effect of Mg2+ incorporation in the cell wall . AFM confirmed that the absence of positive charges reduces cell wall rigidity in the ΔdltA control in regular TSB , as reported ( Saar-Dover et al . , 2012 ) . In Mg2+-enriched growth conditions , extracellular Mg2+ binding complemented the cell wall rigidity defect in this mutant , as TA-coordinated Mg2+ provided cell wall rigidity in the absence of the Dlt machinery . Our AFM measurements showed greater cell wall rigidity in Mg2+-enriched growth conditions in the ΔdltA mutant ( Figure 3D ) , comparable to the wild type strain . These experiments indicate that extracellular Mg2+ is incorporated to cell wall TA to increase cell wall rigidity . Staphylococcus aureus cell wall TA are essential for a response to extracellular Mg2+ , which increases BRcell subpopulation size and thus induces biofilm formation in TSBMg . In these conditions , cells treated with sublethal doses of tunicamycin , which inhibits TarO and thus teichoic acid synthesis at low concentrations ( Swoboda et al . , 2010; Campbell et al . , 2011; Nunomura et al . , 2010; Swoboda et al . , 2009 ) , did not respond to Mg2+ and biofilm formation was inhibited ( Figure 3—figure supplement 1B ) . Based on these findings , we genetically engineered S . aureus strains that down- and upregulate genes related to TA biosynthesis , such as tagB ( Figure 3—figure supplement 1 ) , verified tagB down- and upregulation in these strains by qRT-PCR ( Figure 3—figure supplement 1C ) , confirmed that these strains show no significant defects in growth or peptidoglycan synthesis ( Figure 3—figure supplement 1D , E ) and tested their ability to form biofilms in TSBMg ( Figure 3—figure supplement 1F ) . Strains with reduced tagB expression did not respond to Mg2+ and thus did not develop biofilms ( low-tagB strain ) ( Figure 3E ) . In contrast , strains with upregulated tagB became hypersensitive to extracellular Mg2+ and produced more robust biofilms ( high-tagB ) even with Mg2+ traces that are present in regular TSB medium ( Figure 3F ) . We next tested whether the TA-mediated increase in cell wall rigidity downregulates agr bimodal behavior via σB activation . To study this , we used qRT-PCR analysis to quantify the relative expression of the σB target-gene asp23 and staphyloxanthin quantification to determine σB activation in low- and high-tagB strains ( Figure 3A and Figure 3—figure supplement 1G ) . The low-tagB strain responded more weakly to extracellular Mg2+ than the high-tagB strain , with limited σB activation in both TSB and TSBMg conditions . In contrast , the high-tagB strain was hypersensitive to extracellular Mg2+ , with higher σB activation than the other strains in TSB . These results are consistent with our hypothesis that extracellular Mg2+ stabilizes TA , increases cell wall rigidity and triggers the σB inhibitory signal responsible for downregulating the agr bimodal switch . Once the agr switch is activated , variations in the concentration of these types of input signals affect switch activity and modulate the size of the two subpopulations . For instance , Mg2+ in the colonization niche acts as a downregulatory signal , as it induces σB; activation of the agr switch becomes more difficult in these conditions and DRcell subpopulation size is reduced ( Figure 4A ) . However , since this cue neither generates nor abolishes the agr positive feedback loop , but only modulates its activity , its effect would be restricted to varying the BRcell:DRcell ratio . To substantiate this concept , we used quantitative analysis of fluorescence microscopy images and flow cytometry to monitor S . aureus cell differentiation in the presence of extracellular cues that influence the bimodal switch behavior ( AIP excess and σB activation ) . Purified AIP was added to Ppsmα-yfp or Ppsmβ-yfp reporter strain cultures at various concentrations above threshold concentration of ~10 μM usually found in cultures , which caused differentiation of a DRcell subpopulation that increased in parallel with AIP concentration but cell heterogeneity nonetheless remained detectable in cultures ( Figure 4B ) . When we analyzed downregulation of the bimodal switch , WT cultures in Mg2+-enriched growth conditions had a smaller DRcell subpopulation ( Figure 4A ) , whereas the ΔsigB mutant in TSB and TSBMg media differentiated a larger DRcell subpopulation than the WT strain ( Figure 4C ) . Nevertheless , in both cases BRcell and DRcell differentiation was detected in the ΔsigB mutant . Similarly , the low- and high-tagB strains , which are hypo- and hypersensitive to extracellular Mg2+ , showed larger and smaller DRcell subpopulations , respectively , in TSBMg ( Figure 4D ) although both subpopulations were detected . According to the information we obtained using the agr orthogonal system and to confirm that the only means by which to lead the system into unimodal gene expression is by disrupting the agr positive feedback loop , we monitored DRcell differentiation in the Δagr strain . In this strain , expression of Ppsmα-yfp or Ppsmβ-yfp labeled reporters was not detected thus cell differentiation was inhibited in both TSB and TSBMg media ( Figure 4E ) . Results from our orthogonal system pointed that the feedback loop activation mechanism relies on AgrA~P binding to the P2 promoter to turn on positive self-regulation of the agr operon , thus we monitored cell differentiation in a S . aureus strain in which P2 was replaced with a constitutive promoter ( Figure 4F ) . This strain does not have an active positive feedback loop , as the promoter that activates agr expression is no longer self-inducible . We monitored DRcell differentiation using Ppsmα-yfp or Ppsmβ-yfp reporter strains . In the absence of a functional agr feedback loop , we detected no cell differentiation and reporter expression was homogenous throughout the bacterial population . These results show that the agr positive feedback loop must be active to trigger cell differentiation , and that its activity is regulated by additional input cues that change the ratio of the specialized subpopulations . Given that local Mg2+ and AIP concentrations modulate agr switch activity , we explored spatial organization of cell types during colony development , as reported for other bacteria ( Yarwood et al . , 2007 ) . We developed a mathematical model that considers these factors in the context of nutrient availability and bacterial growth , and delineates growth of multicellular aggregates as a non-linear reaction-diffusion equation system ( Figure 5A and Figure 5—figure supplement 1A ) ( Hense et al . , 2012 ) . Based on the morphological traits of the multicellular aggregates in different genetic backgrounds ( Figure 5—figure supplement 1B–G ) , the model predicted that as an aggregate grows and expands , nutrients become limited in the older , central biofilm region , which has higher AIP levels and slow-dividing cells , which increases representation of the DRcell subpopulation . We sectioned mature aggregates into concentric , morphologically distinct regions and analyzed DRcell and BRcell subpopulation size by flow cytometry ( Figure 5B and Figure 5—figure supplement 2A ) . In accordance with the mathematical predictions , the most peripheral region had a larger proportion of BRcells and a smaller proportion of DRcells . DRcells were enriched in the aggregate center . We combined cryosectioning and confocal microscopy to determine subpopulation size and location within the inner zones of thick biofilms ( Figure 5C and Figure 5—figure supplement 2B ) . BRcells were highly represented in regions near the aggregate outer edge , where nutrient concentration is high ( Cramton et al . , 2001 ) , whereas DRcell representation was more prominent in the biofilm inner region , further from the nutrient source . These experiments showed enrichment of BRcells in newer and of DRcells in older biofilm regions , suggesting that the staphylococcal cells respond differently to local input signal concentrations , and differentiate distinct DRcell:BRcell ratios in different biofilm regions . To study the potential physiological specialization of BRcell and DRcell types beyond the differential expression of agr-regulated reporters , we determined their transcriptional profiles using Illumina RNA-sequencing after enrichment by fluorescence-activated cell sorting ( Figure 6 and Figure 6—figure supplement 1 ) . We grew separate mature aggregates of the strains labeled with the Ppsmα-yfp and Pica-yfp reporter fusions to identify the DRcell and BRcell subpopulations , respectively . Fluorescent cells from mature aggregates were sorted from the rest of the cell population; both fluorescent cells ( enriched ) and whole cell community ( non-enriched ) were collected simultaneously in separate samples . Genome-wide analysis showed similar genetic landscapes for DRcell and BRcell subpopulations , indicating that cell differentiation is not the result of accumulated mutations ( Figure 6—source data 1 ) . RNA was purified from each sample prior to Illumina Hi-seq RNA sequencing . The total number of reads allowed mapping of ~96% to the S . aureus genome ( Figure 6—figure supplement 2A–B ) . Comparison of the normalized gene expression profiles ( Figure 6A , Figure 6—figure supplement 2C and Figure 6—source datas 2–4 ) and qRT-PCR validation ( Figure 6—figure supplement 2D–E ) showed marked differences between the enriched subpopulation and its non-enriched counterpart , which suggested that a specific physiological state could be attributed to each particular cell type . BRcells had a large number of upregulated genes , including sigB sigma factor , and of many biofilm-related genes as well as genes related to peptidoglycan turnover , cell division and DNA replication . In addition , 49 tRNAs were upregulated , which indicates the higher metabolic activity of BRcells and their physiological predisposition to proliferation ( Figure 6B and Figure 6—figure supplement 2D ) . DRcells showed a smaller number of upregulated genes , which we attributed to lower gene expression activity potentially due to the lower physiological activity of this cell type . Among the few upregulated genes detected , we found a notable number related to toxin secretion and host invasion , such as the type-VII secretion system ( Burts et al . , 2005 ) , as well as genes related to protecting bacterial cells from the immune system , such as the hssRS-htrAB hemin detoxification system ( Stauff et al . , 2007 ) . We also detected upregulation of multi-drug efflux pumps that confer resistance to diverse antimicrobials , and of regulators such as graR and arsR , which positively control gene-related cell-wall antibiotic resistance and metal ion stress ( Figure 6B and Figure 6—figure supplement 2E ) . In contrast , the expression of selected housekeeping genes in aggregates growing in TSBMg and TSB showed no differences ( Figure 6—figure supplement 2F ) , suggesting that extracellular Mg2+ specifically influences the agr bimodal switch pathway rather than causing a global deregulation of gene expression . These results suggest that the DRcell subpopulation has lower metabolic activity than BRcell and is predisposed to resist different types of antimicrobials . An important question arises from the physiological differences detected between BRcells and DRcells in vitro , concerning the role and impact of these cell types during progression of S . aureus infections . We found that extracellular Mg2+ is an input signal that regulates agr activity and modulates the sizes of specialized cell subpopulations in vitro , and hypothesized that a similar correlation could be found in vivo , that is , that different tissue Mg2+ concentrations would lead to distinct subpopulation ratios and distinct infection outcomes . To address this question , we developed an infection model ( Koch et al . , 2014 ) in which mice were intravenously infected with 107colony forming units ( CFU ) of cells labeled with Pica-yfp ( BRcells ) or Ppsmα-yfp ( DRcells ) reporters . Infections were allowed to progress for four days , after which mice were sacrificed and the infected organs collected ( Figure 7 and Figure 7—figure supplement 1 ) . The in vitro results correlated with the in vivo experiments; bacteria proliferated more efficiently in Mg2+-rich organs such as kidney . Infected kidneys showed a bacterial load of 1010 CFU/g of tissue ( Figure 7—figure supplement 1A ) , and histological preparations of these organs showed large bacterial aggregates surrounded by immune cell infiltrates ( Figure 7A , Figure 7—figure supplement 1B and Figure 7—figure supplement 3 ) , indicative of long-term colonization during septicemia ( Prabhakara et al . , 2011 ) . Confocal microscopy analyses showed approximately three-fold more BRcells than DRcells in kidney aggregates ( Figure 7C Figure 7—figure supplement 2 ) , similar to levels detected in in vitro experiments; this was consistent with reports that kidneys are Mg2+ reservoirs in the body ( Günther , 2011; Jahnen-Dechent and Ketteler , 2012 ) , and that 82% of patients with urinary catheterization develop long-term S . aureus infections ( Muder et al . , 2006 ) . On the other hand , infected hearts showed a bacterial load of 107 CFU/g of tissue ( Figure 7—figure supplement 1A ) , which suggested that S . aureus cells that colonized heart tissues proliferated less actively than those in kidney . Infected hearts had a larger DRcell subpopulation , consistent with the lower metabolic activity , the lower proliferation rate of these cells in vitro and the lower Mg2+ concentration typically found in heart tissue ( Günther , 2011; Jahnen-Dechent and Ketteler , 2012 ) . Histological preparations of infected hearts revealed deposits of disperse cells with no immune cell infiltrates ( Figure 7B , Figure 7—figure supplement 2 and Figure 7—figure supplement 3 ) , which is indicative of acute bacteremia ( McAdow et al . , 2011 ) . Confocal microscopy analysis showed that as much as 60% of the total heart tissue-colonizing bacterial population consists of DRcells ( Figure 7D and Figure 7—figure supplement 2 ) , as observed in in vitro experiments . To further correlate the presence of extracellular Mg2+ with infection outcome , we performed infection studies using the suite of low- and high-tagB strains ( Figure 8A ) . Kidneys showed a reduction in bacterial load when infected with the Mg2+-hyposensitive low-tagB strain , although this strain colonized heart tissues more efficiently than S . aureus WT . qRT-PCR analyses verified that these differences were associated with upregulation of key genes whose expression is restricted to DRcells ( agrA , agrB and psmα/β ) , which suggests that infections with a low-tagB strain had marked representation of DRcells ( Figure 8B–C ) . In contrast , the Mg2+-hypersensitive high-tagB strain was able to infect kidneys more efficiently than the WT strain , concomitant with a reduced infection of heart tissues . qRT-PCR analyses showed higher expression of genes related to BRcells ( icaA , icaB and spa ) , which suggested that the high-tagB strain differentiated a larger subpopulation of BRcells . We generated a new strain derived from the high-tagB strain that also lacks σB . Higher TA content in this strain increases cell wall rigidity in response to Mg2+ , but the lack of σB should prevent activation of biofilm formation via downregulation of agr . The infection pattern of this ΔsigB high-tagB strain resembled the low-tagB pattern in all the organs analyzed . Kidneys infected with this strain thus showed a reduction in bacterial load , while heart tissues were colonized more efficiently . qPCR analyses demonstrated higher expression of genes related to DRcells in infected kidneys and heart tissues , pointing to a larger number of DRcells in the infection of these strains . In addition , infected livers , which had a moderate of Mg2+ concentration ( Günther , 2011; Jahnen-Dechent and Ketteler , 2012 ) , also showed comparable bacterial loads between low-tagB , high-tagB and ΔsigB high-tagB strains . The bimodal behavior of the agr system results in the differentiation of two genetically identical , cell types that specialize in biofilm- or dispersal-associated lifestyles in S . aureus communities . These cell types were detected in in vitro cultures and during in vivo infections . The expression of specific markers by BRcells and DRcells is due to activation of the agr bimodal switch , which requires the production of the activating signal AIP above a certain threshold , similar to other quorum sensing systems , such as the heterogeneous expression of natural competence in cultures of B . subtilis in response ComX signal ( Maamar and Dubnau , 2005; Smits et al . , 2005 ) or quorum sensing activation of bioluminescence in a subpopulation of cells in Vibrio harveyi ( Anetzberger et al . , 2009 ) . A growing number of bimodal switches are being found in QS pathways , and cell differentiation in response to a QS signal is becoming an established concept in microbiology . The molecular mechanism of agr bimodal behavior is based on a sequential activation of the two adjacent divergent promoters P2 and P3 . The P2 promoter triggers positive self-regulation of the agr operon ( by activating the agr positive feedback loop ) and the P3 promoter induces the agr regulon responsible for activation of virulence genes ( P3 promoter activation is necessary for DRcell specialization ) . The differential affinity of AgrA~P for P2 and P3 promoters is crucial for the sequential promoter activation and thus for agr bimodal switch activation . AgrA~P binds P2 with greater affinity than the P3 promoter . The P2 promoter thus activates and triggers the feedback loop at lower AgrA~P concentrations and only in a given subpopulation ( agr-on cells ) . Activation of this feedback loop produces high AgrA~P levels in this subpopulation , which licenses them to trigger the less-sensitive P3 promoter and induce the agr regulon , leading the agr-on cells to specialize in dispersion and virulence and become DRcells . In contrast , the cell subpopulation that expresses P2 below the threshold cannot activate the agr positive feedback loop ( agr-off cells ) , and thus do not produce sufficient AgrA~P to induce P3 promoter expression . In these cells , genes normally repressed by AgrA~P are upregulated , including biofilm-related genes , which licenses them to differentiate as biofilm-producing cells thus to become BRcells . Once the agr bimodal switch is activated and BRcells and DRcells differentiate , subpopulation size is modulated by other extracellular cues that affect bimodal switch activity . We report that extracellular Mg2+ is incorporated into the bacterial cell wall by binding teichoic acids and increases cell wall rigidity , provoking activation of sigB expression . As σB downregulates the agr bimodal switch , its activation above the threshold becomes more difficult , leading to differentiation of a smaller DRcell and a larger BRcell subpopulation , which in turn facilitates biofilm formation . In contrast , increasing AIP above the threshold concentration facilitates activation of the agr bimodal switch , leading to differentiation of a larger DRcell and a smaller BRcell subpopulation , which in turn facilitates dispersion and acute infection . The activity of these agr input cues neither generates nor abolishes the agr positive feedback loop; agr bimodal switch activity is only modulated , which regulates the size of the two subpopulations . DRcell and BRcell subpopulations can therefore be detected in the presence of input cues; in our in vitro and in vivo assays , only their ratio differed in the overall bacterial community , thus varying infection outcome in distinct scenarios . In mouse models , we found that in vivo infections developed a larger DRcell subpopulation in tissues with low Mg2+ levels ( cardiac tissue ) , in which the microbial population became dispersed . In contrast , BRcells were more prevalent in tissues with high Mg2+ concentrations ( bone , kidney ) and bacteria organized in microbial aggregates characteristic of biofilm-associated infections . The results recapitulate clinical studies in which a marked number of biofilm-associated persistent infections develop in urinary tract and bone , usually following surgery or catheterization ( Muder et al . , 2006; Brady et al . , 2008; Flores-Mireles et al . , 2015; Idelevich et al . , 2016 ) . These data help to understand how nosocomial pathogens such as S . aureus can simultaneously cause dissimilar types of infections in distinct organs , and show that cell differentiation in nosocomial pathogens is particularly relevant for adaptation to different host tissues . The molecular mechanism whereby extracellular Mg2+ downregulates the agr bimodal switch relies on the capacity of magnesium to bind TA to increase cell wall rigidity ( Heptinstall et al . , 1970; Hughes et al . , 1971 ) , in turn causing agr downregulation via activation of the repressor σB . Staphylococcus aureus cells with reduced cell wall TA adhere poorly , have poor biofilm formation ability ( Vergara-Irigaray et al . , 2008 ) , and do not colonize nasal ( Weidenmaier et al . , 2004 ) or kidney-derived endothelial tissues ( Weidenmaier et al . , 2005a ) . An effect similar to that of Mg2+ incorporation into the cell wall is caused by D-alanine esterification of TA ( Lambert et al . , 1975a; Lambert et al . , 1975b ) . The Dlt protein machinery decorates TA with D-alanine esters to reduce repulsive interactions between TA negative charges , which also increases cell wall rigidity ( Perego et al . , 1995 ) . Previous reports showed that Dlt activity is important for biofilm formation in S . aureus ( Gross et al . , 2001 ) and for developing biofilm-associated infections in vivo in animal models ( Weidenmaier et al . , 2005b ) . Our results are consistent with these findings and suggest that , given the intricacy of agr bimodal switch regulatory control , many extracellular cues probably contribute to the outcome of S . aureus infections . Extracellular Mg2+ and hence , increased cell wall rigidity are probably important cues for triggering biofilm-associated infections . Here we show that infections generated by clonal populations of bacteria can bifurcate into distinct , specialized cell types that localize physically in different colonization tissues in the course of an infection . The prevalence of physiologically distinct bacterial cells in different organs could offers bacteria an adaptive strategy that increases their chance of evading the immune system evasion during infection , or provides a bet-hedging strategy that increases the possibility of survival during antimicrobial therapy ( Beaumont et al . , 2009 ) . Understanding cell differentiation in bacteria is central to designing new anti-infective strategies that target specific cell subpopulations , particularly to pathogens like Staphylococcus aureus , considered endemic in hospitals and which has an approximately 20% mortality rate ( Klevens et al . , 2007 ) . A complete list of strains used in this study is shown in Supplementary file 1 . The laboratory S . aureus strain RN4220 ( Kornblum et al . , 1990 ) was used for cloning purposes . The strain Escherichia coli DH5α was used for propagating plasmids and genetic constructs in laboratory conditions . B . subtilis and E . coli strains were regularly grown in LB medium . When required , selective media were prepared in LB agar using antibiotics at the following final concentrations: ampicillin 100 μg/ml , kanamycin 10 μg/ml , chloramphenicol 5 μg/ml , tetracycline 10 μg/ml , and erythromycin 2 μg/ml . S . aureus strains were routinely propagated in liquid TSB medium incubated with shaking ( 220 rpm ) at 37°C for 16 hr . When required , selective media were prepared in TSB using antibiotics at the following final concentrations: kanamycin 10 μg/ml , chloramphenicol 10 μg/ml , tetracycline 10 μg/ml , erythromycin 2 μg/ml , neomycin at 75 μg/ml , Spectinomycin at 300 μg/ml and Lincomycin at 25 μg/ml . For S . aureus aggregates in TSBMg , 4 μl of an overnight liquid culture was spotted in TSBMg ( TSB medium supplemented with MgCl2100 mM ) ( Koch et al . , 2014 ) and dried in a sterile culture cabin . Plates were allowed to grow for five days at 37°C ( Koch et al . , 2014 ) . For S . aureus traditional biofilms in liquid TSBMg and TSB , we followed the protocol proposed by O'Toole and Kolter ( O'Toole and Kolter , 1998a ) . Briefly , an overnight liquid culture was inoculated 1:100 in fresh TSB media and after 6 hr at 37°C with shaking at 220 rpm , the OD600 nm was measured and normalized to a final OD600 0 . 05 . From the normalized cultures , 5 μl was inoculated in 995 μl of TSB or TSBMg in a 24-well microtiter plate ( Thermo ) ( RRID:SCR_008452 ) and incubated for 24 or 48 hr at 37°C without shaking . Biofilm formation was measured as follows: media was discarded by aspiration; wells were washed twice with PBS and allowed to dry for 45 min at 65°C . Then , 500 μl of a solution of Crystal Violet at 0 . 1% was added and to stain the organic material associated with the well for 5 min . After staining , wells were washed three times with deionized water . For quantitative analyses , Crystal Violet was solubilized using 500 μl of acetic acid at 33% . The solubilized dye was diluted 1:100 in deionized water and transferred to 96-well microtiter plates ( Thermo ) . The absorbance was determined at 595 nm using an InfiniteF200 Pro microtiter plate reader ( Tecan ) . Background was corrected by subtracting the absorbance values of non-bacteria inoculated wells . Specific growth conditions are presented in figure legends . For the experiments using the synthetic orthogonal agr model generated in B . subtilis wild type and ΔsigB mutant , cells were incubated in LB medium at 220 rpm at 37°C until cultures reached an OD600nm = 0 . 5 . After incubation , 50 μl of the culture was added to 50 ml of chemically-defined MSgg growing medium ( Branda et al . , 2001 ) and allowed to grow for 4 hr at 37°C at 220 rpm . After incubation , 50 μl of culture was used to inoculate 50 ml of fresh LB and allowed to grow for 12 hr at 37°C with constant shaking . Addition of AIP to the culture defined the initiation of the experiment ( time = 0 hr ) . Samples were taken at 0 hr , 2 hr , 4 hr , 6 hr , 8 hr , 10 and 12 hr . To generate the S . aureus strain Newman Δica , Δpsmα , Δpsmβ and ΔdltA-E mutants , 500 bp flanking each gene and the respective antibiotic cassettes were PCR amplified and the fragments were subsequently joined together using a long-flanking homology PCR ( LFH-PCR ) . The resulting fragments were cloned into pMAD plasmid ( Arnaud et al . , 2004 ) and transformed into the laboratory strain S . aureus RN4220 . To transfer the mutations from RN4220 to Newman and to generate the double mutant strains , φ11 phage lysates were generated from RN4220 mutants to infect Newman , NewHG and USA300 LAC* ( Rudin et al . , 1974 ) . Clones resistant to the respective antibiotic were further verified to carry the mutation using PCR . Positive clones were validated to carry the mutation using Sanger sequencing . To generate the S . aureus strains single-labeled with Pica-yfp , Pspa-yfp , Ppsmα-yfp , Ppsmα-mars , Ppsmβ-yfp , PdnaA-yfp and double-labeled with Pspa-yfp Pica-mars , Ppsmβ-yfp Ppsmα-mars , Ppsmα-yfp Pica-mars , Pspa-yfp Ppsmα-mars , Pica-yfp PclfA-mars , Pica-yfp PisdA-mars , Pspa-yfp PclfA-mars and Pspa-yfp PisdA-mars transcriptional fusions , the respective promoter region comprising 200 to 500 bp upstream of the start codon was fused to the yfp reporter-gene using the plasmid pKM003 or to the rfp ( mars ) reporter-gene using the plasmid pKMmars . These fusions were subcloned into the plasmids pAmy and pLac ( Yepes et al . , 2014 ) . The plasmids were integrated into the neutral amy and lac loci of S . aureus chromosome to ensure a uniform and chromosome-equivalent copy number of the reporters in all the cells within the microbial community . The integration of reporters in amy and lac neutral loci occurs in a two-step recombination process , as described in ( Yepes et al . , 2014 ) . Briefly , integration of the plasmid into the chromosome of S . aureus occurs via a single recombination event . This first recombination occurs by growing the plasmid-carrying strain overnight at 30°C , plating serial dilutions onto selective media ( erythromycin 2 μg/ml and X-Gal 100 μg/ml ) and incubating the plates at 44°C . This is a temperature-sensitive plasmid , which does not allow plasmid replication at higher temperatures ( Arnaud et al . , 2004 ) . Therefore , incubation at 44°C allows only the strains that incorporate the plasmid into the chromosome to grow . The genetic constructs obtained from the first recombination process in the strain RN4220 were transferred to strain Newman by φ11 phage transduction ( Rudin et al . , 1974 ) . Once the constructs were transferred to the recipient strain ( Newman and USA300 LAC* strains ) , we forced a second recombination process to leave only the reporter in the integration locus . To do this , a culture of a light-blue colony was incubated overnight at 30°C in the absence of erythromycin , plating serial dilutions onto selective ( erythromycin and X-Gal ) media and then incubating the plates at 44°C . After 48 hr of incubation at 44°C , the light-blue colonies still carrying the plasmid in the chromosome were discarded . A four-step process of screening , including fluorescence , antibiotic susceptibility , PCR and Sanger sequencing , was designed to validate whether the white colonies carried the corresponding insertion in the neutral loci . The S . aureus strains tagB-lower ( NWMN_0187 ) , tagG-lower ( NWMN_1763 ) and tagH-lower ( NWMN_1763 ) were obtained by phage transduction using as donor strain the respective mutants deposited in the transposon-mapped mutant collection ( Bae et al . , 2004 ) . Clones were verified using PCR and Sanger sequencing . The S . aureus strain that overexpresses the tagB gene ( tagB-higher ) ( NWMN_0187 ) or the agrBCDA operon ( NWMN_1943 to NWMN_1946 ) were obtained by cloning the complete ORF into the replicative plasmid pJL74 ( Klijn et al . , 2006 ) . The sarA P1 promoter and the RBS of sodA guarantee high expression and translation levels in S . aureus . To construct the agr synthetic orthologous model in B . subtilis ΔsigB , the two genes agrC and agrA , which are adjacent in the operon agrBDCA , were cloned as a chimeric version agrCA . The gene agrC encodes the histidine kinase and the gene agrA encodes the cognate regulator ( Recsei et al . , 1986; Boles and Horswill , 2008; Peng et al . , 1988 ) . The resultant construct was integrated into the amyE neutral chromosomal locus of B . subtilis ΔsigB ( strain 168 ) . Moreover , the Pica-yfp , Pspa-yfp , Ppsmα-yfp , Ppsmβ-yfp , PRNAII-yfp and PRNAIII-yfp transcriptional fusions were cloned into plasmid pDR183 and integrated into the neutral locus lacA . For transformation via double heterologous recombination , all plasmids were linearized and added to competence-induced liquid cultures of B . subtilis ΔsigB strain 168 ( Hardwood and Cutting , 1990 ) . Resultant colonies were verified that contained the reporter using Sanger sequencing . The synthetic model that recreates the divergent P2 and P3 promoters of S . aureus contains a DNA fragment of the construct of RNAII and RNAIII joined to the cfp and yfp genes divergently transcribed by the P2 ( RNAII ) and P3 ( RNAIII ) promoter , respectively . This fragment was cloned into the plasmid pDR183 and integrated into the neutral locus lacA . The integration of the fragment occurs by double heterologous recombination . The plasmids were linearized and added to competence-induced liquid cultures of B . subtilis ΔsigB strain 168 . The resultant colonies were verified to contain the construct using Sanger sequencing . For staphyloxanthin extraction , we used a protocol adapted from Pelz et al . ( 2005 ) . After 72 hr of growth , cells were harvested , washed once and resuspended in PBS buffer . The cell densities at OD600 nm were measured and the samples normalized . One ml of cells was centrifuged and the pellet resuspended in 200 μl of methanol and heated at 55°C for 3 min . Samples were centrifuged to eliminate debris . Then , 200 μl of the supernatant was taken and the methanol extraction repeated . A volume of 180 μl was recovered and added to 820 μl of methanol . Absorption spectra of the methanol extracts were measured using a spectrophotometer at a peak of 465 nm , normalized and reported as relative absorbance to express the total amount of staphyloxanthin pigment . Mechanical indentation via atomic force microscopy ( AFM ) was applied according to previous publications ( Dufrêne , 2014; Formosa-Dague et al . , 2016 ) . Overnight cultures were diluted into fresh TSB or TSBMg liquid media to a final OD600 of 0 . 05 . Cells were grown overnight at 37°C and 220 rpm . One ( 1 ) ml of the culture was washed with sterile PBS or PBS supplemented with MgCl2 final concentration 100 mM ( PBSMg ) , depending on culture conditions and were normalized to a final OD600 of 0 . 5 in PBS or PBSMg . Cells were fixed with 4% p-formaldehyde for 6 min and washed twice with 500 μl of PBS . One series of mild sonication was applied to produce a homogenous sample of single cells . Finally , 40 μl of a dilution of 1:5 in deionized water was immobilized on poly-lysine coated microscopy slides . Samples were washed twice with Milli-Q water and allowed to dry . Samples were processed immediately after immobilization , in air and at room temperature , using a CFM conical probe AFM ( Nanotec , Spain ) with nominal spring constant 3 N/m and resonant frequency 75 kHz . Optical lever calibration and sensitivity was obtained by tapping the probe cantilever onto the glass surface of the slide and measuring the force response to z-piezo extension ( z is vertical to the glass surface ) . For cell indentation , the AFM probe was placed above a cell and repeatedly pressed down onto the surface ( and retracted ) at 50 nm/s over distances of 100 nm , several times at several positions of the cell surface . Z position and speed of the AFM probe were controlled by a piezoelectric translator . The force response of the cell membrane was measured at three different positions for three individual cells . Young’s modulus was obtained by fitting the resulting force-indentation curves for forces <10 nN , resulting in indentations <~20 nm . Best fits were produced with a modified Hertz model assuming a conical punch probe geometry . For purification of AIP1 from S . aureus strain Newman , we used a protocol that is adapted from ( MDowell et al . , 2001 ) . To obtain an enriched fraction of AIP , a 500 ml culture of each strain was grown for 24 hr in TSB and , after the removal of bacterial cells by centrifugation , the supernatant was filtered through a 0 . 22 μm membrane filter and mixed 1:1 volumes with binding buffer ( 2% CH3CN and 1% trifluoroacetic acid ) . The filtered supernatant was loaded into a C18 Sep-Pak cartridge ( Waters ) previously stabilized with binding buffer . Elution of AIP was achieved with a 60% concentration range of CH3CN and subsequently concentrated using a SpeedVac system . Fresh AIP fractions were used in each experiment due to the instability of the preparation . Peptidoglycan from S . aureus was purified using a protocol adapted from ( Bera et al . , 2005; Peterson et al . , 1978 ) . Cells were grown in 2 L of TSB medium and incubated overnight at 37°C with vigorous shaking . Bacteria were harvested by centrifugation ( 5000 × g , 4°C , 10 min ) , washed with cold buffer 1 ( 20 mM ammonium acetate , pH 4 . 8 ) and resuspended in 30 ml of buffer 1 . Cell suspension was transferred to a falcon tube , centrifuged and determined the weight of the pellet . Cell pellet was resuspended in 2 ml of buffer and cells were disrupted with a bead beater ( Genogrinder , SPEXsamplePrep , USA ) . After centrifugation , the interphase between glass beads and the foam at the top of the tube was collected and treated with 40 U of DNase , 80 U of RNase and 5 mM of MgSO4 and were incubated 5 hr at 37°C . Cell walls were resuspended in 2% SDS in buffer 1 and incubated 1 hr at 65°C . The material was washed twice with distilled water and resuspended in 30 ml of buffer 1 . 5% trichloroacetic acid was added to remove WTA from peptidoglycan and incubated 4 hr at 60°C . PG was then washed four to six times with cold Milli-Q water , lyophilized , and weighed . Before use , PG was resuspended in PBS buffer and sonicated on ice . Quantification of peptidoglycan was performed using a protocol adapted from ( Nocadello et al . , 2016; Zhou et al . , 1988 ) . PG pellets were resuspended in 5 ml of cold buffer 1 and diluted 1:50 in a final volume of 2 ml . PG was labeled with Remazol Brilliant Blue ( RBB ) by incubating the samples with 20 mM RBB in 0 . 25 M NaOH ON at 37°C with constant shaking . The labeled samples were neutralized with HCl and pelleted by centrifugation at 14000 rpm for 20 min at room temperature . We performed intense washing using distilled water to eliminate the remaining RBB . After washing , the RBB-PG complexes were diluted 1:50 and the OD 595 nm was determined . OD 595 nm values were normalized to wet weight of each PG-isolated sample . Digital images of the development of S . aureus multicellular aggregates were captured with an AxioCAm-HR digital camera ( Carl Zeiss ) using AxioVision AC Release 4 . 3 software ( Carl Zeiss ) ( RRID:SCR_002677 ) . For fluorescence microscopy , cells from the multicellular communities or from the liquid cultures were washed in PBS and resuspended in 0 . 5 ml of 4% paraformaldehyde solution and incubated at room temperature for 6 min . After two washing steps with PBS buffer , samples were resuspended in 0 . 5 ml of PBS buffer and mildly sonicated to guarantee samples of dispersed single cells . Microscopy images were taken on a Leica DMI6000B microscope equipped with a Leica CRT6000 illumination system ( Leica ) . The microscope was equipped with a HCX PL APO oil immersion objective with 100 × 1 . 47 magnification and a color camera Leica DFC630FX . Linear image processing was done using Leica Application Suite Advance Fluorescence Software ( RRID:SCR_013673 ) . The YFP fluorescence signal was detected using an excitation filter 489 nm and an emission filter 508 nm ( excitation filter BP 470/40 and suppression filter BP 525/20 ) . The RFP-mars fluorescence signal was detected using an excitation filter 558 nm and an emission filter 582 nm ( excitation filter BP 546/12 and suppression filter BP 605/75 ) . Excitation times were 567 and 875 msec , respectively . Transmitted light images were taken with 21 msec of excitation time . To quantitatively measure cell fluorescence from microscopy images , we adapted an image protocol originally published by McCloy RA et al . , using ImageJ64 v1 . 48s ( NIH , USA ) ( RRID:SCR_003070 ) ( McCloy et al . , 2014 ) . Briefly , to quantify the number of fluorescent cells and determine their fluorescence level within a microscopy field , the overlapping image of the bright field and fluorescent channels was converted to RGB and inverted it to highlight fluorescent cells . An automatically adjusted threshold generated an image in which only fluorescent cells were represented ( image A ) . Using the same procedure without the inversion step generated an additional image in which only non-fluorescent cells were represented ( Image B ) . The sum between fluorescent and non-fluorescent cells from Images A and B , respectively , represented the total number of cells in the field . Quantification of fluorescence at the single cell level in the microscopy field was determined using the same software using the following commands: from the analyze menu , we selected set measurements , making sure that Area , Min and Max gray values and Mean gray value were selected . Then , we selected analyze particles and set: size in pixel unit from 20 to 200 pixels , circularity from 0 . 1 to 1 . 00 and show overlay outlines , making sure that the options display results , summarize and in situ show were selected . It is recommended to run a configuration test to set the analyze particles parameters that correctly cover all cells in the microscopy image . Analysis results were transferred to a Microsoft Excel sheet and to calculate the Total Cell Fluorescent ( TCF ) as TCF = Area of selected cell X Mean fluorescence . Results were used to generate a histogram that represents the number of particles for each mean fluorescence value . At least three independent images were analyzed for each experiment and the mean values were plotted . In average , each microscopy field comprised 500–765 cells . For analysis of overlapping signals using fluorescence microscopy , we considered signals to overlap when both signals were detected in a 3:1–1:3 range . This is the range in which green and red signals merge to yellow signal in microscopy and thus define green/red fluorescence overlap . For thin cryosectioning of S . aureus multicellular communities , bright field and fluorescence images were acquired using a TCS SP5 II confocal microscope ( Leica ) . The hardware settings included: Argon laser power at 25% and 496 nm laser intensity at 15% . Bright field images were collected using the PMT-1 Trans scan channel at 512 V with a gain offset of −0 . 15% . Fluorescent images were collected using the HyD-2 channel with a gain of 10 and an emission bandwidth of 500 nm for excitation and 550 nm for emission ( excitation filter BP 470/40 and suppression filter BP 525/20 ) . The acquisition mode included a xyz scan mode , with z-stacks in the z-wide mode from 4 to 8 μm . To determine the structural features of the thin sections and localize the fluorescence , a series of horizontal optical sections were collected using a z-step size of 0 . 3 μm . Width and height format in X and Y was set to 1024 × 1024 pixels at a scan speed of 200 Hz . Air 1 pinhole was set to automatic detection . A HCX PL APO CS 40 . 0 × 1 . 30 OIL UV objective was used for image acquisition . Digital images were captured using the Leica AF 6000 system software that is provided with the confocal microscope . All parameters were kept identical for the unlabeled control and the different labeled samples . To quantitatively measure fluorescence area from each thin cryosection image , we used ImageJ64 v1 . 48s and we adapted an image protocol as in ( McCloy et al . , 2014; Gavet and Pines , 2010; Potapova et al . , 2011 ) . Using this software , we quantified the bacterial aggregate area from each image of infected tissue . We quantify the proportion of fluorescent area from the total area occupied by S . aureus cells and referred it in percentage as relative fluorescence signal . We quantified fluorescence of three different thin cryosection samples obtained from three independent multicellular aggregates . The infected mice organs were aseptically extracted and immersed in a solution 1:1 of PBS and paraformaldehyde 4% and left at 4°C overnight . 4 μm-thick sections were obtained using a CM 3050 s cryostat set to −20°C ( Leica ) . These histological sections were placed on SuperFrost plus poly lysine-coated slides ( Thermo ) and immediately rinsed twice with PBS buffer precooled at 4°C . Then , fixed-samples where stained with Giemsa staining solution ( Sigma ) including a dehydration step before the staining and a rehydration step after staining using Xylol and ethanol at 96% , 70% and 50% ( Thammavongsa et al . , 2009 ) . Slides were immediately mounted with coverslips and processed by confocal microscopy . Histological digital images were obtained using the Diskus software ( Hilgers ) . For fluorescence imaging , a Leica TCS SP5 II confocal microscope equipped with A HCX PL APO CS 100 × 1 . 47 OIL objective was used . The hardware settings included: Argon laser power at 25% and 496 nm laser intensity at 10% . Bright field images were collected using the PMT-1 Trans scan channel at 512 V . Fluorescent images were collected using the HyD-2 channel with a gain of 5 and an emission bandwidth of 500 nm for excitation and 550 nm for emission ( excitation filter BP 470/40 and suppression filter BP 525/20 ) . The acquisition mode included a xyz scan mode , with z-stacks in the z-wide mode from 4 to 8 μm . To localize fluorescence , a series of horizontal optical sections were collected using a z-step size of 0 . 2 μm and with an optimized system . Width and height format in X and Y was set to 1024 × 1024 pixels at a scan speed of 200 Hz . Air one pinhole was set to automatic detection . Digital images were captured using the Leica AF 6000 system software provided with the confocal microscope . All parameters remained constant during the examination of the different labeled samples . To measure fluorescence signal in infected organs , we used ImageJ64 v1 . 48s and we adapted an image protocol from ( McCloy et al . , 2014; Gavet and Pines , 2010; Potapova et al . , 2011 ) . Using this software , we quantified the bacterial aggregate area that exists in each one of the infected tissue images . From the area that is occupied by S . aureus cells , we used the same software to quantify the proportion of fluorescent area and referred in percentage relative to the total bacterial aggregate area . We quantified fluorescence of three different histological sections obtained from independent organs from three different infected mice . For flow cytometry analysis , cells from the multicellular communities were fixed with a treatment of 4% paraformaldehyde as mentioned above , washed and resuspended in PBS buffer . After fixation , a sonication treatment was required to separate single cells in the sample . In this case , samples were subjected to series of 25 pulses ( power output 70% and cycle 0 . 7 s ) and kept on ice . Dilution of samples 1:500 was necessary prior flow cytometry analyses . For YFP fluorescence , a laser excitation of 488 nm coupled with 530/30 and 505LP sequential filter was used . The photomultiplier voltage was set at 777 V . To obtain samples enriched in BRcells or DRcells , we used single-labeled Staphylococcus aureus multicellular communities between day 4 and 5 of development . These strains were labeled to differentiate cells expressing the extracellular matrix-production reporter ( Pica-yfp ) or the detachment/virulence reporter ( Ppsmα-yfp ) . Multicellular communities were scraped from the TSBMg plates and immediately resuspended in RNAlater ( Qiagen ) in 1 . 5 ml RNAse-free Eppendorf tubes , in order to fix the cell fluorescence and at the same time preserve the RNA within the cells . Previous reports ( Rosenberg et al . , 2003 ) and fluorescence microscopy experiments performed in our laboratory ( data not shown ) showed that the fixing procedure of these multicellular communities in RNAlater had no effect in the conservation of the fluorescence when compared to cells fixed using 4% paraformaldehyde . Multicellular communities were disrupted in the RNAlater by extensive pipetting , followed by one series of mild sonication as mentioned above , and previously treating the sonicator with RNaseZap RNase Decontamination Solution ( Life Technologies ) ( RRID:SCR_008817 ) . All procedures were performed on ice . After sonication , samples were immediately processed using FACS . Cell fixation and subsequent mild sonication allowed cell separation without affecting cell integrity . For the sorting procedure , 50 μl of the cell suspension was resuspended in 10 ml of filtered and autoclaved PBS buffer prepared in DEPC-treated water . This cell suspension was sonicated , changing cycles from 70% to continuous ( 100% ) and performing 1 round of 20 s . Immediately , cells were FACS-sorted based on their fluorescence intensity in a FACS Aria III ( Becton Dickinson ) ( RRID:SCR_008418 ) using the following parameters: Nozzle size of 70 microns , FITC/Alexa Fluor 488 nm laser , a 530/30 nm filter for data collection and a 502 LP mirror . Flow cytometry parameters were set as: SSC 341 V with a threshold of 500 , FSC 308 V with a threshold of 500 , FITC 769 V and a variable Flow Rate to guarantee that the number of events per second never exceeded 1500; hence , the Sorting Efficiency never dropped from 97% . These data were analyzed using the BDIS FACS Diva software version 7 . 0 provided with the FACS Aria III . Sorting was performed in the Precision Mode set to ‘Single Cell’ in a first round , followed by a second sorting round set to ‘Purity’ . Using the sorting Precision Mode , we recovered approx . 25 million cells of each subpopulation ( fluorescent cells ) and their respective non-fluorescent counterparts , based on manually established Target Gates P1 for highly fluorescent cells and P2 for non-fluorescence cells . Once sorted ( approximately 5 million cells per 15 ml tube ) , cells were immediately quick-frozen by immersing the tubes in liquid nitrogen prior to ultra-freezing until sorting was completed . For the FACS-sorted bacterial cells , the ultra-frozen samples were thawed using a 37°C water bath . The volume was poured in a vacuum filter system provided with a 47 mm filter diameter and 0 . 45 μm pore-size . The equipment was previously sterilized using 75% ethanol in DEPC-treated water , followed by two DEPC-treated water rinses and finally UV light for 180 s and then precooled at −20°C . Filters containing each of the sorted samples were individually ground using liquid Nitrogen in an RNAse-free , sterile precooled mortar . The powder was carefully scraped out from the mortar and placed in a 2 ml RNAse-free Eppendorf tube and the RNA isolated as described below . To isolate RNA from S . aureus infected organs , 450 μl of the homogenized organs were incubated for 15 min on ice with 50 μl of RNAlater and 5 μl of Triton X100 briefly vortexing every 5 min to lyse the murine cells . After lysis , RNA was isolated as described below . To isolate RNA from S . aureus infected organs , 450 μl of the homogenized organs were incubated for 15 min on ice with 50 μl of RNAlater and 5 μl of Triton X100 briefly vortexing every 5 min to lyse the murine cells . After lysis , RNA was isolated as described below . One volume of TE lysis buffer ( Tris 20 mM pH 7 . 5 , EDTA 10 mM ) prepared in DEPC-prepared water and 25 μl of Lysostaphin ( 1 mg/ml ) were added and taken to a hybridization oven ( with rotation ) , pre-warmed at 37°C . Samples were incubated for 30 min at 37°C prior to transferring the whole material to a new tube for mechanical lysis . This was performed using FastPrep Lysing Matrix glass beads in a Fast Prep Shaker ( MP Biomedicals ) ( RRID:SCR_013308 ) set at 6500 rpm for 50 s . Samples were lysed using 2 cycles of 50 s . The lysate was transferred to a new 2 ml RNAse-free Eppendorf tube and centrifuged for 10 min at 14000 rpm and 4°C to remove the filter and beads . The supernatant ( 700 μl ) was recovered and used for RNA isolation using the standard hot phenol methodology with some modifications . The sample recovered from was mixed with 60 μl of 10% SDS and incubated at 64°C for 2 min . 66 μl of sodium acetate 3 M pH 5 . 2 and 750 μl of phenol Roti-Aqua ( Carl Roth ) were added and the mixture was incubated at 64°C for 6 min with mixing every 30 s . The sample was centrifuged for 10 min at 13000 rpm and 4°C and the upper aqueous layer was transferred to a 2 ml Phase Lock Gel Heavy ( PLGH ) tube ( 5Prime ) . Then , 750 μl of Chloroform was added and the sample was centrifuged for 12 min at 13000 and 15°C . The aqueous layer was transferred to a new tube and mixed with 2 volumes of a 30:1 ethanol and sodium acetate 3 M pH 6 . 5 mix . Additionally , 0 . 5 μl of GlycoBlue co-precipitant ( 15 mg/ml ) ( Life Technologies ) was added . This mix was left at −20°C overnight . The following day , the sample was removed from storage at −20°C , centrifuged for 30 min at 13000 rpm and 4°C and the pellet was washed with 300 μl of precooled 75% ethanol . After washing , the total RNA was resuspended in 42 μl of RNAse-free water ( Qiagen ) were added and the sample was incubated at 65°C at 1000 rpm for 5 min prior storage in ice . To remove any DNA traces , the isolated RNA was treated ( one to three times , depending on the sample ) with 4 Units of RNase-free DNase I ( Thermo ) , 10 Units of SUPERase In RNase Inhibitor ( Life Technologies ) and incubated for 45 min at 37°C . To remove the DNase I , 50 μl of RNAse-free water and 100 μl of Roti-Aqua-P/C/I ( Phenol , Chloroform , Isoamyl alcohol 25:24:1 pH 4 . 5–5 ) ( Carl Roth ) were added to the reaction tube , mixed , transferred to a 2 mL PLGH tube and centrifuged for 12 min at 13000 rpm and 15°C . The sample was mixed with 2 . 5 volumes of the 30:1 ethanol and sodium acetate with 0 . 5 μl of GlycoBlueTM , which led to RNA precipitation when stored at −20°C overnight . On the following day , samples were centrifuged for 30 min at 13000 rpm and 4°C , washed with 200 μl of 75% ethanol and the pellets were dried and resuspended in 50 μl of DEPC-water . To remove phenol residues , 50 μl of RNAse-free water and 100 μl of Chloroform ( Carl Roth ) were added to the reaction tube , mixed by inversion for 1 min , transferred to a 2 mL PLGH tube and centrifuged for 12 min at 13000 rpm and 15°C . The sample was mixed with 2 . 5 volumes of the 30:1 ethanol and sodium acetate with 0 . 5 μl of GlycoBlueTM , which led to RNA precipitation when stored at −20°C overnight . On the following day , samples were centrifuged for 1 hr at 13000 rpm and 4°C , washed with 200 μl of 75% ethanol and the pellets were dried . This RNA was resuspended in 22 to 44 μl of RNAse-free water and then incubated at 65°C at 1000 rpm for 5 min . To assess the concentration and purity of the total RNA , OD260 was measured using a Nanodrop ( Thermo ) and the OD260/OD280 ratio and the OD260/OD230 ratio determined . The cDNA prepared was strictly strand-specific , allowing transcriptome sequencing and expression profiling in both the forward and reverse strands . The combined-length of the flanking sequences was 100 bases . The cDNA is generated and size fractionated by preparative gel electrophoresis or by using the LabChip XT fractionation system from Caliper/PerkinElmer in order to obtain cDNA fractions , optimally suited for the different NGS systems . For this , the RNA samples were poly ( A ) -tailed using a poly ( A ) polymerase . The 5'-PPP were removed using tobacco acid pyrophosphatase ( TAP ) followed by the ligation of the RNA adapter to the 5'-monophosphate of the RNA . First-strand cDNA synthesis was performed with an oligo ( dT ) -adapter primer and the M-MLV reverse transcriptase . The resulting cDNA was PCR-amplified to reach a concentration of 20–30 ng/μl using a high fidelity DNA polymerase . The cDNA was purified using the Agencourt AMPure XP kit ( Beckman ) ( RRID:SCR_008940 ) and was analyzed by capillary electrophoresis . The primers used for PCR amplification were designed for TruSeq sequencing according to the instructions of Illumina . The following adapter sequences flank the cDNA inserts: TruSeq_Sense: 5’-AATGATACGGCGACCACCGAGATCTACACTCTTTCCCTACACGACGCTCTTCCGAT-T-3’ TruSeq-Antisense NNNNNN ( NNNNNN = Barcode ) 5’-CAAGCAGAAGACGGCAT ACGAGATNNNNNNGTGACTGG-AGTTCAGACGTGTGCTCTTCC-GATC ( dT25 ) −3’ . For quantification of gene expression , total RNA was reverse-transcribed using hexameric random primers followed by quantitative real-time PCR ( qRT–PCR ) using the SsoAdvanced SYBR Green Supermix ( Bio-Rad ) ( RRID:SCR_013553 ) , following manufacturer’s instructions . Primer pairs used are described in the Supplementary file 2 . Gene expression was normalized to gyrA/gapA expression and expression fold changes were calculated using the 2−ΔΔCt method . These qRT-PCR experiments were performed following the standard MIQE guidelines for publication of qRT-PCR experiments ( Bustin et al . , 2009 ) . The pooled sequence reads were de-multiplexed and the adapter sequences were removed . After that , the reads in Fastq format were quality trimmed using fastq_quality_trimmer ( from the FastX suite version 0 . 0 . 13 ) ( RRID:SCR_005534 ) with a cut-off Phred score of 20 and converted to Fasta format using Fastq_to_Fasta ( also from the FastX suite ) . The reads were processed , which included poly ( A ) removal , size filtering ( minimum read length of 12 nucleotides after clipping ) , statistics generation , coverage calculation and normalization , which were performed with the RNA-analysis pipeline READemption version 0 . 3 . 3 . READemption uses segemehl version 0 . 1 . 7 for the read alignment to the reference and DESeq 1 . 18 . 0 ( Anders and Huber , 2010 ) ( RRID:SCR_000154 ) for the differential gene expression analysis . The reference genome NC_009641 was taken from the NCBI database for the purpose of alignment and gene-quantification . DESeq calculates statistically significant expression fold change and their log2 values by computing the ratio of normalized read counts of each gene in two libraries . The genes with log2fold value higher than 1 . 5 and lower than −1 . 5 were selected as up- and down-regulated gene-sets , respectively . Scatter plots for the visualization of sample correlation were generated using matplotlib 1 . 4 . 2 . 2 . Datasets used for gene ontology for functional classification of genes differentially expressed , hierarchical clustering and the calculation of the hypergeometric probability for genes differentially expressed are presented in Figure 6—source data 2 to 4 . We estimated the difference between the reference genome and the sequences from the libraries ( total number of reads in the libraries range from ~6–11 million with an average of 100 nucleotides per read ) . Libraries were analyzed using SAMtools ( mpileup command ) and variant calling by BCFtools ( RRID:SCR_005227 ) . The variants were quantified for the reads mapped to the coding regions of the reference genome using the standard quality score >40 ( base call accuracy 99 . 99% ) . SAM ( Sequence Alignment/Map ) tools enables quality checking of reads , and automatic identification of genomic variants ( Li et al . , 2009b ) . A high quality score means higher number of sequences showing a particular kind of variation . Quality score is proportional to the number of reads mapped to a gene that account for the variation . The BCFtools were used for the variance calling to identify general variances in each library and compared with each other to determine the existence of library-specific variations ( Li , 2011 ) . Using this approach , we detected a minimum of 99 . 999934% genome similarities at the nucleotide level between distinct cell types . All statistical analyses were performed using the software Prism 6 ( version 6 . 0 f , GraphPad ) ( RRID:SCR_002798 ) . Graphs represent data from at least three independent experiments with at least three independent technical replicates for experiment . Error bars represent standard deviation ( mean ±SD ) . For the analysis of experiments with two groups , the parametric unpaired two-tailed Student’s t-test with Welch’s correction was done and , the non-parametric unpaired Mann-Whitney test were done . For the analysis of experiments with three or more groups , the parametric one-way ANOVA test was done . Post hoc analysis included multiple comparisons Tukey's test , Dunnett’s test or Dunn’s tests , depending on the data set . Differences were considered significant when p value was smaller than 0 . 05 . Statistical significance: ns = not statistically significant , p<0 . 05 , p<0 . 01 , *p<0 . 001 . All animal studies were approved by the local government of Lower Franconia , Germany ( license number 55 . 2-DMS-2532-2-57 and were performed in strict accordance with the guidelines for animal care and animal experimentation of the German animal protection law and directive 2010/63/EU of the European parliament on the protection of animals used for scientific purposes . Female BALB/c mice ( 16 to 19 g ) were purchased from Charles River ( Charles River Laboratories , Erkrath , Germany ) ( RRID:SCR_003792 ) , housed in polypropylene cages and supplied with food and water ad libitum . The different S . aureus strains used were cultured for 18 hr at 37°C on BHI medium . Subsequently , cells were collected and washed three times with PBS and diluted to reach an OD600 nm = 0 . 05 . Viable cell counts were determined by plating dilutions of the inoculum on TSB agar plates . To compare the representation of each cell type of S . aureus in the bacterial communities that colonize the distinct organs , three cohorts of 3 mice for the unlabeled strain and six mice for each labeled strains ( Pica-yfp and Ppsmα-yfp ) were infected with 150 µl of cultures of S . aureus , containing 1 × 107 cells via tail vein injection . Each strain was used to infect one cohort of mice . Infections were allowed to progress until severe infection signs occurred or to the endpoints of 24 hr , 48 hr , 72 hr and 96 hr after mice challenging . Animals were sacrificed when they met the following criteria: 1 ) loss of at least 20% of body weight; 2 ) loss of at least 15% of body weight and ruffled fur; 3 ) loss of at least 10% of body weight and hunched posture; or 4 ) 24 hr , 48 hr , 72 hr and 96 hr time points after infection . Infected kidneys and hearts were aseptically harvested and processed for thin sectioning , histology and confocal microscopy , as described above . Organs were also used to calculate bacterial burden calculated as CFU/g of organ , as proposed by Marincola et al . ( 2012 ) . For this purpose , kidneys and hearts were homogenized in 2 ml of sterile PBS using GentleMACSTM M Tubes ( Miltenyi Biotec ) ( RRID:SCR_008984 ) and serial dilutions from 10−2 to 10−8 of the organ homogenates were immediately plated on TSB plates and incubated at 37°C for 24 hr , To compare the representation of the S . aureus low-tagB and high-tagB strains colonizing the kidneys , heart and liver , four cohorts of 5 mice each , were infected with 150 µl of cultures of S . aureus and organs were processed as described above . Three days after bacterial challenge all mice were euthanized , organs were aseptically harvested and processed as described above , CFU determined and RNA isolated . To study the dynamic processes of the synthetic network , we mathematically described the reactions of the genetic circuitry of the orthologous system . In this circuit , the phosphorylation state is only reached inside the cell after the concentration of AIP is above the threshold . The complex AgrA~P is the transcription factor that upregulates the expression of P2 promoter ( also called PRNAII ) responsible for activation of the positive feedback loop . Moreover , AgrA~P upregulates the expression of the rest of the promoters that were used in this work; P3 ( Also called PRNAIII ) , Ppsmα , Ppsmβ . Pica and Pspa promoters are not directly regulated by AgrA~P and were therefore used as negative controls . The reactions that define the former genetic network are the following: ( 1 ) Activation/deactivation: AgrA P+PagrA k1⇌k−1 PagrAa ( 2 ) Fast transcription:PagrAa →k2PagrAa+mAgrA ( 3 ) Slow transcription:PagrA →k3PagrA+mAgrA ( 4 ) Translation:mAgrA →k4mAgrA+AgrA ( 5 ) Complex formation:AgrA +P →k5AgrA\ P ( 6 ) Creation ( appearance ) :ϕ →k6 P ( 7 ) Activation/deactivation:AgrA P+Px k7⇌k−7 Pxa ( 8 ) Fast transcription:Pxa →k8Pxa+mYFP ( 9 ) Slow transcription:Px →k9Px+mYFP ( 10 ) Translation:mYFP →k10 mYFP+YFP ( 11 ) Degradation ( disappearance ) :mAgrA→k11 ϕ ( 12 ) Degradation ( disappearance ) :AgrA→k12 ϕ ( 13 ) Degradation ( disappearance ) :AgrAP→k13 ϕ ( 14 ) Degradation ( disappearance ) :P→k14 ϕ ( 15 ) Degradation ( disappearance ) :mYFP→k15 ϕ ( 16 ) Degradation ( disappearance ) :YFP→k16 ϕ where k1 and k-1 are the binding and unbinding rates of AgrA~P to P2 , k2 is the transcription rate of P2 once AgrA~P binds the promoter , k3 is the basal transcription rate of P2 , k2 and k3 produce the mRNA of agrA , k4 is the translation rate of AgrA protein , k5 is the phosphorylation rate of AgrA , k6 is the availability rate of phosphate in the system , k7 and k-7 are the binding and unbinding rates of AgrA~P to the different promoters ( Px ) , k8 is the transcription rate of Px once AgrA~P binds to the promoter ( Pax ) , k9 is the basal transcription rate of Px , k10 is the translation rate of the YFP protein and k11 to k16 are the degradation rates of mRNAs and proteins involved in this system . Deterministic modeling using differential equations pointed to a quasi-steady state assumption ( Murray , 2002 ) , which we used to identify the elements responsible for the behavior of the system . The resulting equations are: ( 17 ) dAgrA Pdt= k6∙AgrAAgrA+ δ- k13 ∙ AgrAP ( 18 ) dAgrAdt= α1+ β1 ∙ AgrAPγ1+ AgrAP-k6 ∙AgrAAgrA+ δ- k12∙AgrA ( 19 ) dYFPdt= α2+ β2 ∙ AgrAPγ2+ AgrAP- k16∙YFP where α1 = k3k-1k4PtagrA/k1k11 , β1 = k2k4PtagrA/k11 , γ1 = k-1/k1 , δ = k14/k5 , α2 = k9k-7k10Ptx/k7k15 , β2 = k8k10Ptx/k15 , γ2 = k-7/k7 and Pti = Pi + Pai with i = [agrA , x] . The following values are used to run the Gillespie algorithm of the full model ( reactions 1–16 ) . We focused our first set of simulations on the dynamics that affect YFP directly , to further study the behavior of the system when considering variations in AgrA and AgrA~P . The first set of simulations allowed us to find parameters that show a fixed value among all of the experiments independently on the promoter that is under consideration . These parameters are shown here: ParameterMeaningValuek1Binding rate0 . 01 molecules/hk-1Unbinding rate2/hk2Transcription rates500/hk3Basal transcription rate50/hk9Basal transcription rate90/hk4 , k10Translation rates50/mink5Phosphorylation rate0 . 05/molecules/hourk6Entry rate40/molecules/hourk11 , k15Degradation rates10/hourk12Degradation rate0 . 05/hourk13Degradation rate0 . 1/hourk14 , k16Degradation rates1/hour Consequently , the values of these parameters did not change through the rest of the simulations that aimed to characterize the kinetics of each particular promoter . Initial simulations assumed the high stability of AgrA~P to test the response of the system to saturated levels of AgrA~P . The following simulations assumed unstable AgrA~P . The specific rates to simulate each promoter were defined as follows: ParameterMeaning ( rate ) PRNAIIIPpsmαPRNAIIPpsmβPRNAIII-dualPRNAII-dualk8Transcription300300450470300450/hourk7Binding36 . 510201240 × 10−4k-7Unbinding0 . 080 . 080 . 10 . 10 . 080 . 1/hour*molecules−1 hour−1 The simulations that involved an unstable AgrA~P established values of k5 = 0 . 2 × 10−4 and k6 = 5 . Using these values , we resolved a 3-mode decision-making model . Further simulations that involved changeable values of k6 ( phosphate availability ) revealed that the activation rate of P2 and P3 occurs within a range of 0 and a maximum value of saturation of 40 . The values for the rest of the parameters were selected from standard numbers obtained from previously reported studies ( Andersen et al . , 1998; Balagaddé et al . , 2008; Ben-Tabou de-Leon and Davidson , 2009; Dublanche et al . , 2006; Goñi-Moreno and Amos , 2012 ) . We used a so-called reaction diffusion system , which not only describes the changes of concentrations and density in time to any type of reaction but also their spread in space . In our case , we used a model with two spatial dimensions . Our system consisted of four equations , describing how the concentrations of nutrients , AIP and the density of replicative and non-replicative bacteria evolved in time . In the following , nx , tdenotes the nutrient concentration , b ( x , t ) the density of replicative bacteria , s ( x , t ) the density of non-replicative bacteria and q ( x , t ) the concentration of AIP . We assumed that nutrients and AIP underlie diffusion . Diffusion parameters are denoted by dn anddq , respectively . In the case of the replicative bacteria , the diffusion coefficient depends on nutrient concentration and the density of replicative bacteria . It is of the form db=σnbwhere σalso contains a stochastic fluctuation of random movement . This form for the diffusion of active bacteria cells was chosen according to the model developed by Kawasaki et al . ( 1997 ) . We assumed that non-replicative bacteria are not able to diffuse on their own and their diffusion is driven by movement of the replicative cells . The diffusion of non-replicative bacteria depends on the density of replicative bacteria . If there are more replicative bacteria , the non-replicative cells will be pushed in a type of diffusion . This effect is limited . The diffusion coefficient for the non-replicative bacteria is assumed to be of the form:ds=τbbs+b where bs and τare constant . The replicative bacteria proliferate by consuming nutrients . In our system of equations , the consumption rate of the nutrients was given by G1f ( n , b ) and the bacterial growth was described by the term G2 fn , b where G2/G1 was the conversion rate of consumed nutrients to bacterial growth . We assumed f ( n , b ) to be of the form:f ( n , b ) =bn1+γn ( 1+1qm+δq ) Here , we chose a Monod growth term ( Monod , 1949 ) to describe the increase in the concentration of replicative bacteria in relation to nutrient consumption . It reproduces the fact that a high nutrient concentration will cause a faster increase in the concentration of replicative bacteria but that these bacteria cannot reproduce infinitely fast . An additional factor accounts for the effect of the quorum sensing signal . If the concentration of active bacteria is already high , a high concentration of AIP slows down the conversion process . We furthermore assumed that there is only a transition from replicative to non-replicative cells . This process is described by a term of the form:εa ( b , n ) =εb ( 1+ba1 ) ( 1+na2 ) This choice is in agreement with ( Matsushita et al . , 1999 ) . The equation number four described the concentration of AIP . We considered that the diffusion and the increase in the concentration of AIP are related with the concentration of replicative bacteria but we also considered that there is degradation process of AIP . The degradation of AIP is described byµq , whereas the production of AIP typically has two levels in a Hill type function ( Gustafsson et al . , 2004 ) with Hill coefficient 2 to reflect bistability . This is caused by a positive feedback including nonlinearity in the underlying regulation system . The low production rate of AIP is denoted by p1while the increased production rate is denoted byp2 . The threshold between the low and the increased production is denoted byqthr . For more details see ( Müller et al . , 2006 ) . With the above-mentioned terms we achieved the following system of equations , which are able to represent the growth dynamics of S . aureus multicellular aggregates: ( 20 ) ∂n∂t=dn∇2d-G1bn1+γn1+1qm+δq ( 21 ) ∂b∂t=∇ ( σnb∇b ) +G2bn1+γn ( 1+1qm+δq ) −εb ( 1+ba1 ) ( 1+na2 ) ( 22 ) ∂s∂t= ∇τbbs+b∇s+εb1+ba11+na2 ( 23 ) ∂q∂t=dq∇2q+p1+p2q2qthr2+q2b-μqq It remains necessary to define the initial and boundary conditions . Since the bacteria grow on an agar plate , we chose no-flux boundary conditions , e . g . ∂n∂x\|∂Ω=0 , where Ω denotes the area of the agar plate . The agar has the same concentration of nutrients throughout and therefore we defined the initial condition for the nutrients as constant over the entire domain , e . g . nx , 0=n0 for x∈Ω . The replicative bacteria are set on the agar as a drop in the center . At this stage , neither non-replicative bacteria nor AIP exist and thus the initial conditions read b ( x , 0 ) =bM exp⁡ ( −x2+y26 . 25 ) with a compact support , sx , 0=0 and qx , 0=0 . The ( non-dimensionalized ) parameters chosen for this simulation are given by:σ=0 . 5;G1=G2=G=7;ε=5;τ=0 . 25;qm=0 . 3;dn=dq=1;z=1;ρ=1;γ=1;bs=2; δ=1;qthr=1;μ=1;n0=1 . 11;p1= p2=1;a1=2400; a2=120 . Our model was able to capture differences in texture on the surfaces of different S . aureus mutants . For the simulation of mutants , we only changed the parameter value , which corresponds to the modified gene and phenotype for the mutant , the other parameter values were maintained from the wild type , as follows: BackgroundσGp1p2Wild type0 . 5711spa0 . 5311ica0 . 75211psmα1811psmβ1411agr1500spa ica1211 The non-mentioned parameters were maintained at the same value as the wild type strain . In the case of the Δspa mutant , as influencing the biofilm , we modified the biofilm production rate constant G from 7 to 3 . The mutants Δpsmα and Δpsmβ differ from the wild type also by their ability to move on; therefore apart from Galso the parameter σ was modified . To obtain the agr mutant not only a simple change in the parameters is needed , but also a combination of different mutant behaviors since the agrsystem influences several components of biofilm production and properties . Especially the ability of the model to produce AIP is lost , expressed by setting p1 and p2 to 0 .	While in hospital , patients can be unwittingly exposed to bacteria that can cause disease . These hospital-associated bacteria can lead to potentially life-threatening infections that may also complicate the treatment of the patients’ existing medical conditions . Staphylococcus aureus is one such bacterium , and it can cause several types of infection including pneumonia , blood infections and long-term infections of prosthetic devices . It is thought that S . aureus is able to cause so many different types of infection because it is capable of colonizing distinct tissues and organs in various parts of the body . Understanding the biological processes that drive the different infections is crucial to improving how these infections are treated . S . aureus lives either as an independent , free-swimming cell or as part of a community known as a biofilm . These different lifestyles dictate the type of infection the bacterium can cause , with free-swimming cells producing toxins that contribute to intense , usually short-lived , infections and biofilms promoting longer-term infections that are difficult to eradicate . However , it is not clear how a population of S . aureus cells chooses to adopt a particular lifestyle and whether there are any environmental signals that influence this decision . Here , Garcia-Betancur et al . found that S . aureus populations contain small groups of cells that have already specialized into a particular lifestyle . These groups of cells collectively influence the choice made by other cells in the population . While both lifestyles will be represented in the population , environmental factors influence the numbers of cells that initially adopt each type of lifestyle , which ultimately affects the choice made by the rest of the population . For example , if the bacteria colonize a tissue or organ that contains high levels of magnesium ions , the population is more likely to form biofilms . In the future , the findings of Garcia-Betancur et al . may help us to predict how an infection may develop in a particular patient , which may help to diagnose the infection more quickly and allow it to be treated more effectively .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "microbiology", "and", "infectious", "disease" ]	2017	Cell differentiation defines acute and chronic infection cell types in Staphylococcus aureus
The small GTPase Rab7 is a key regulator of endosomal maturation in eukaryotic cells . Mutations in rab7 are thought to cause the dominant neuropathy Charcot-Marie-Tooth 2B ( CMT2B ) by a gain-of-function mechanism . Here we show that loss of rab7 , but not overexpression of rab7 CMT2B mutants , causes adult-onset neurodegeneration in a Drosophila model . All CMT2B mutant proteins retain 10–50% function based on quantitative imaging , electrophysiology , and rescue experiments in sensory and motor neurons in vivo . Consequently , expression of CMT2B mutants at levels between 0 . 5 and 10-fold their endogenous levels fully rescues the neuropathy-like phenotypes of the rab7 mutant . Live imaging reveals that CMT2B proteins are inefficiently recruited to endosomes , but do not impair endosomal maturation . These findings are not consistent with a gain-of-function mechanism . Instead , they indicate a dosage-dependent sensitivity of neurons to rab7-dependent degradation . Our results suggest a therapeutic approach opposite to the currently proposed reduction of mutant protein function . Several neuropathies , lysosomal storage diseases and neurodegenerative disorders primarily affect the nervous system , despite underlying defects in cellular processes that occur in all cells ( Schultz et al . , 2011; Wang et al . , 2012 ) . Charcot-Marie-Tooth 2B ( CMT2B ) is a sensory neuropathy that primarily affects some of the longest axon projections in the human body and is caused by mutations in the rab7 locus . rab7 encodes a GTPase that regulates endolysosomal degradation in all cells ( Elliott et al . , 1997 ) . All known mutations in CMT2B patients alter highly conserved amino acids in Rab7 and cause pathology in heterozygosity ( Kwon et al . , 1995; Verhoeven et al . , 2003; Houlden et al . , 2004; Meggouh et al . , 2006 ) . Hence , CMT2B is a genetically dominant disease . Several studies have proposed a neuron-specific gain-of-function mechanism of the CMT2B alleles to explain the dominant neuronal phenotype of this ubiquitous gene ( Spinosa et al . , 2008; Cogli et al . , 2010; McCray et al . , 2010; Cogli et al . , 2013; Zhang et al . , 2013 ) . In support of this hypothesis , several dominant functions of CMT2B mutant Rab7 have been described based on overexpression of the mutant proteins in neuronal or non-neuronal cultured cells . For example , CMT2B protein expression leads to altered EGF degradation in HeLa cells ( Spinosa et al . , 2008 ) , decreased upregulation of the growth-associated protein 43 in PC12 cells ( Cogli et al . , 2010 ) , increased interaction with the filament protein peripherin in Neura2A cells ( Cogli et al . , 2013 ) , modulatory effects on JNK signaling in N1E-115 cells ( Yamauchi et al . , 2010 ) , accumulation of the NGF receptor TrkA in cultured dorsal root ganglia cells ( Zhang et al . , 2013 ) , and altered EGF receptor signaling in HeLa , BHK-21 and A431 cells ( Basuray et al . , 2013 ) , amongst others . Furthermore , a recent report has suggested that overexpression of CMT2B mutants in HeLa and PC12 cells dominantly reduces rab7 function ( Basuray et al . , 2013 ) . It is unclear which ones of these effects are causally linked to the neuropathy in aging sensory and motor neurons in humans . Since Rab7 is a key protein required for endolysosomal function in all cells , its loss or gain-of-function is predicted to directly or indirectly affect many signaling pathways over time . In addition , it is currently unclear whether overexpression of the CMT2B mutant proteins actually causes axon terminal degeneration in a sensory or motor neuron . Indeed , overexpression in at least one cell culture system revealed no obvious toxic effects ( McCray et al . , 2010 ) . Consequently , the mechanism underlying the genetic dominance and the putative gain-of-function underlying the pathology of CMT2B remains unclear . Rab7 has a well understood and critical role in converting Rab5-positive early endosomes into late endosomal compartments and thereby represents a key step in endolysosomal maturation in all cells ( Bucci et al . , 2000; Rink et al . , 2005; Poteryaev et al . , 2010 ) . Rab7 GTPase function is biochemically well defined and GTP-locked ‘constitutively active’ and GDP-locked ‘dominant negative’ mutants have been tested and utilized in a plethora of systems , including human cell lines and Drosophila ( Mukhopadhyay et al . , 1997a , 1997b; Bucci et al . , 2000; Zhang et al . , 2007 ) . More recently , extensive biochemical characterizations of all four CMT2B proteins revealed that the majority of the overexpressed protein is in a GTP-bound form when compared to the GDP-bound form ( Spinosa et al . , 2008 ) . However , the same study showed that the overall binding to both GTP and GDP is drastically reduced for the CMT2B proteins , unlike a constitutively active form ( Spinosa et al . , 2008 ) . In addition , a 2 . 8 Å crystal structure of one of the CMT2B proteins revealed no intrinsic GTPase defect ( McCray et al . , 2010 ) . Furthermore , the CMT2B variants can at least partially rescue defects caused by reduced rab7 levels ( Spinosa et al . , 2008; McCray et al . , 2010 ) . In summary , neither overexpression studies in cell culture nor comprehensive biochemical analyses have so far pinpointed a molecular mechanism that directly causes adult-onset loss of synaptic function in sensory and motor neurons . In mouse ( Kawamura et al . , 2012 ) and Caenorhabditis elegans ( Kinchen et al . , 2008; Skorobogata and Rocheleau , 2012 ) rab7 null mutants cause embryonic lethality , which has so far precluded the analysis of neuronal phenotypes . In this study , we used the fruit fly Drosophila melanogaster to analyze the role and mechanism of CMT2B mutant Rab7 proteins in sensory and motor neurons in vivo . Our findings did not uncover a neuron-specific dominant gain- or loss-of-function for the CMT2B alleles . Instead , our findings indicate that the CMT2B alleles are partial loss of function alleles of rab7 that cause , in a dosage-dependent manner , first synaptic and subsequently neuronal degeneration . Our recent functional profiling of all rab GTPases in Drosophila suggested an increased neuronal demand for rab7 function compared to other cell types based on elevated neuronal expression of rab7 , but not rab5 or rab11 ( Chan et al . , 2011 ) . To investigate an enhanced or specialized role for rab7 in the nervous system , we generated a null mutant by replacing the rab7 open reading frame with a Gal4 knock-in cassette ( Figure 1A ) ; the Gal4 knock-in provides a means to express any gene of interest under the endogenous regulatory elements of rab7 in either heterozygous or homozygous rab7 mutants ( Brand and Perrimon , 1993; Chan et al . , 2011 , 2012 ) . Loss of rab7 leads to lethality between 50–80% of pupal development ( P+50%–P+80% ) with no gross morphological abnormalities . Loss of the maternal contribution causes lethality in fully developed embryos with no obvious developmental defects . Wholemount preparations of the brain in null mutant pupae at P+35% reveal a loss of Rab7 immunolabeling to background levels ( Figure 1B , C ) ( Chinchore et al . , 2009; Chan et al . , 2011 ) . By P+50% substantial accumulations of the endosomal marker Hrs become apparent , although the overall brain structure appears normal ( Figure 1D , E ) . In addition , immunolabeling of photoreceptor axons indicates that cell-type specification and axon pathfinding are normal , but the photoreceptor membrane protein Chaoptin ( 24B10 ) accumulates in the brain ( Figure 1F , G ) . Overexpressing UAS-YFP-Rab7 under control of the rab7Gal4-knock-in in the homozygous mutant rescues these phenotypes ( Figure 1F–H ) . 10 . 7554/eLife . 01064 . 003Figure 1 . Loss of rab7 in neurons causes adult-onset degeneration that begins with a loss of synaptic function . ( A ) Knock-out strategy: replacement of the complete rab7 open reading frame with a Gal4 knock-in cassette ( Chan et al . , 2011 , 2012 ) . ( B and C ) Pupal brains at P+35% for wild-type ( B ) and the rab7 mutant ( C ) . Red: Rab7 , Blue: synaptic vesicle marker CSP . Note that the red labeling in the center of ( C ) stems from 3xP3-RFP expression that marks the knock-in cassette . ( D and E ) Pupal brain at P+50% from wild-type ( D ) and the rab7 mutant ( E ) . Green: photoreceptor-specific mAb 24B10; magenta: the endosome marker HRS . ( F–H ) 3D visualization of photoreceptor axon projections in ctrl , rab7 homozygous mutant , and a rab7 homozygous mutant expressing UAS-YFP-Rab7 ( rescue ) ( Zhang et al . , 2007 ) . ( I and J ) Genetic mosaics with rab7 mutant photoreceptors in otherwise heterozygous flies exhibit no eye development defects . ( K–M ) Electroretinogram ( ERG ) responses from flies with rab7 mutant eyes . Light stimulation for 5 days leads to the almost complete loss of synaptic function ( ERG ‘on’ transient , M ) ; despite normal photoreceptor responses to light ( ERG Depolarization , L ) . ( K ) Sample ERG traces from 5-day old flies . ( N–P ) Electron microscopy of mutant eyes showing rhabdomere degeneration in rab7 mutant clones ( arrow ) ( N ) and synaptic terminals ( O and P ) . Note that the presence of pigment between ommatidia marks patches of wild-type ommatidia ( compare I and J and arrowheads in Figure 1—figure supplement 1C ) . ( O ) Light stimulation leads to vacuolarization and degeneration of rab7 synaptic terminals ( arrowheads ) . Scale bar in ( C ) for ( B and C ) and ( E ) for ( D and E ) : 50 µm; in ( F ) for ( F–H ) : 20 µm; in ( N ) : 10 µm; in ( O ) for ( O and P ) : 1 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 01064 . 00310 . 7554/eLife . 01064 . 004Figure 1—figure supplement 1 . Functional and morphological degeneration in rab7 mutant photoreceptors in 5-day light or 5-day dark-raised flies . ( A ) Representative ERG traces for indicated conditions and quantification in Figure 1L–M . ( B ) Phalloidin labeling of rhadomeres in the eye reveals structural degeneration in rab7 mutants as a function of light stimulation . ( C ) Electron micrograph of an eye cross-section in 5-day dark-raised flies . Arrowheads indicate pigment , which mark heterozygous or wild type control patches in the eye . Scale bars ( B ) and ( C ) : 10 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 01064 . 004 To study the effect of loss of rab7 on sensory neuronal survival , we generated genetic mosaics with mutant photoreceptor neurons in otherwise heterozygous chimeric flies ( Newsome et al . , 2000; Chotard et al . , 2005; Mehta et al . , 2005 ) . Photoreceptors are sensory neurons of the peripheral nervous system . Since blind flies are viable under laboratory conditions , photoreceptor neurons provide a model for the in vivo study of mutations that would be lethal if present in other tissues . Eyes mutant for rab7 look indistinguishable from wild-type , suggesting that rab7 mutant photoreceptors develop normally ( Figure 1I , J ) . Measurements of photoreceptor function using electroretinogram ( ERG ) recordings ( Coombe , 1986 ) revealed no defects in the amplitudes of light-evoked responses ( Figure 1K , L ) and synaptic function ( as indicated by the ‘on’ transient; Figure 1K , M ) in newly hatched adults . We conclude that rab7 mutant photoreceptor neurons develop and initially function without obvious defects . In contrast , 5-day-old rab7 mutants that were raised in constant ambient light ( ∼600 Lux , see ‘Materials and methods’ ) exhibit almost complete loss of synaptic function ( Figure 1M ) . Under the same conditions , wild-type photoreceptors exhibit no significant reduction in synaptic function . By contrast , the photoreceptor response amplitude does not significantly differ between wild-type and rab7 , suggesting that synaptic function is more sensitive than cell body function to loss of rab7 . Furthermore , the age-dependent synaptic defect can be fully rescued by minimizing their stimulation through raising the flies in the dark ( Figure 1M , Figure 1—figure supplement 1A ) . Similarly , electron microscopy of rab7 mutant photoreceptors reveals increased degeneration of cellular structure after 5 days in constant light compared with wild-type ( Figure 1N ) , while 5-day dark-reared rab7 mutants and controls exhibit no obvious defects ( Figure 1—figure supplement 1B , C ) . Synaptic terminals of light-exposed rab7 photoreceptors exhibit large-scale degeneration that is absent in unstimulated rab7 mutant terminals ( Figure 1O , P ) . These findings suggest that in the absence of neuronal stimulation rab7 in photoreceptor neurons is partially dispensable at least in young flies . Overall , our analysis of rab7-deficient sensory neurons revealed progressive defects that appear first as a loss of synaptic function but lead ultimately to degeneration of the entire neuron . Adult-onset synaptic degeneration is a hallmark of many human sensory neuropathies . Indeed , four independent mutations in rab7 that cause the sensory neuropathy CMT2B in patients have been characterized ( Verhoeven et al . , 2003; Houlden et al . , 2004; Meggouh et al . , 2006 ) . The four amino acids altered by these mutations ( L129 , K157 , N161 and V162 ) are 100% conserved in the single Drosophila rab7 ortholog , including their precise location in the primary sequence of identical length ( Figure 2A ) . All four mutations cause the CMT2B neuropathy independently and dominantly in heterozygote patients , which has led to a focus on identifying a putative gain-of-function effect of the disease mutants ( Spinosa et al . , 2008; McCray et al . , 2010; Cogli et al . , 2013; Zhang et al . , 2013 ) . Furthermore , it has been proposed that the disease mutations might mimic the constitutively active rab7Q67L , a known gain-of-function mutation ( Mukhopadhyay et al . , 1997b; De Luca et al . , 2008; Spinosa et al . , 2008 ) . To test the gain-of-function hypothesis in vivo , we generated transgenic flies for the expression of all four CMT2B mutants . rab7WT , rab7Q67L and the well-characterized dominant negative rab7T22N variants have previously been generated and successfully used in Drosophila , but their random genomic integrations preclude comparable levels of expression ( Zhang et al . , 2007 ) . To compare these variants quantitatively with the CMT2B mutations at identical expression levels , we generated new transgenic lines for rab7Q67L , rab7T22N and rab7WT using the same insertion site in the genome . Finally , we generated two transgenic fly lines for the expression of the human rab7 ortholog rab7A , one for wild-type hrab7A and one of the CMT2B mutation K157N . All nine fly and human rab7 variants ( WT , Q67L , T22N , K157N , L129F , N161T , V162M , hrab7A-WT , hrab7A-K157N ) were N-terminally tagged with the YFP variant Venus , similar to previous experiments showing that N-terminal tagging does not interfere with rab GTPase function ( Zhang et al . , 2007 ) . 10 . 7554/eLife . 01064 . 005Figure 2 . Overexpression of Venus-tagged Rab7 variants in motor neurons . ( A ) Protein alignment of fly and human Rab7 reveals 100% conservation of protein length and the precise locations of all CMT2B mutations ( red ) and the classically designated ‘dominant negative’ and ‘constitutively active’ mutations ( green ) . ( B–F ) Electrophysiological recordings from the larval neuromuscular junction for ctrl , the rab7 null mutant and overexpression of rab7N161T , rab7L129F and the human disease gene hrab7K157N . Spontaneous vesicle release ( minis ) exhibit normal frequency ( B ) and amplitude ( C ) . Similarly , evoked neurotransmission is indistinguishable for all genotypes ( E and F ) . ( G ) Synaptic boutons at the larval neuromuscular junction , immunolabeled for the presynaptic membrane ( HRP , red ) , the endosomal marker Hrs ( blue ) and six different Venus-Rab7 proteins . ( H ) Colocalization quantification for all nine Venus-Rab7 proteins at the neuromuscular junction with the endosomal markers Rab5 and Hrs . ( I and J ) Analysis of the subcellular protein localizations of all Venus-tagged Rab7 variants ( green ) in the larval ventral ganglion , as previously performed for all Rab GTPases ( Chan et al . , 2011 ) . Blue ( DNA labeled with Toto-3 ) indicates areas of cell bodies ( cb ) and synapses ( syn ) in two center stripes . ( J ) Quantification of Venus fluorescence signal in 3D datasets inside clearly discernable compartments ( arrows in I ) as ratio of total fluorescence signal . See ‘Materials and methods’ for details . Scale bar in ( I ) : 50 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 01064 . 00510 . 7554/eLife . 01064 . 006Figure 2—figure supplement 1 . Western blot analysis of total protein extract from fly eyes . Single copies of Venus-Rab7 transgenes of the indicated mutant were driven by rab7Gal4-knock-in . ( A ) An antibody against Drosophila Rab7 recognizes both endogenous Rab7 and the Venus-tagged Drosophila Rab7 variants , but not human Rab7 protein . An antibody again human Rab7 only recognizes the Venus-tagged human protein variants , but not th endogenous Drosophila Rab7 . ( B ) Actin Loading Control . ( C ) Quantification of relative band intensities over three separate experiments reveals that the Rab7-Gal4 expressed UAS-rab7 transgenes are expressed at 1 . 8–3-fold higher amounts than the endogenous Rab7 protein in heterozygosity . DOI: http://dx . doi . org/10 . 7554/eLife . 01064 . 006 The gain-of-function hypothesis predicts that overexpressing the constitutively active ( GTP-bound ) Rab7-Q67L or the CMT2B variants should cause a disease-related phenotype . We expressed all nine variants under the endogenous rab7 regulatory elements using the rab7Gal4-knock-in in heterozygotes . The rab7Gal4-knock-in -driven UAS-venus-rab7 expresses all variants in the endogenous spatiotemporal expression pattern of rab7 . However , potential caveats include ( 1 ) that the rab7Gal4-knock-in may exhibit small differences from the endogenous pattern not detected here , ( 2 ) that the UAS transgenes may exhibit small differences in expression levels even though they are inserted in the same landing sitesand in identical genetic background , and ( 3 ) that the Gal4/UAS amplification can increase overall protein levels ( Brand and Perrimon , 1993; Chan et al . , 2011 ) . Western Blot analysis of total protein extract from fly heads revealed that all Venus-Rab7 proteins are mildly overexpressed at ∼1 . 8 to 3-fold of the levels of the endogenous Rab7 protein ( Figure 2—figure supplement 1 ) . We established nine transgenic lines for each of the nine UAS-venus-rab7 transgenes in flies heterozygous for the rab7Gal4-knock-in null allele ( see “Generation of nine UAS-venus-rab7 constructs and transgenic lines” ) . Hence , each line stably expresses one of the nine mutant rab7 variants under the endogenous regulatory elements of rab7 with around twofold protein levels of the endogenous protein in a background of 50% of the endogenous protein . Remarkably , all nine lines are viable as adults without obvious behavioral defects or altered lifespan . To measure whether loss of rab7 or expression of the CMT2B variants causes functional defects in motor neurons , we performed electrophysiological recordings at the larval neuromuscular junction of the rab7 null mutant and for overexpression of three Drosophila and human CMT2B variants . If the CMT2B variants exert a dominant effect on synaptic function , these recordings should reveal an overexpression defect even if the null mutant does not . As shown in Figure 2B–D , both the frequency and amplitudes of spontaneous single vesicle release events are indistinguishable between control , rab7 null mutant and CMT2B protein overexpression . Similarly , evoked neurotransmission exhibits no defects for any genotype ( Figure 2E , F ) . The absence of synaptic defects in the rab7 null mutant is consistent with the electroretinogram recordings of photoreceptor function in young adults shown above ( Figure 1K–M ) . In addition , these findings indicate that mild overexpression of CMT2B mutant proteins in their endogenous expression pattern in an entire animal in vivo does not obviously affect synaptic function in motor neurons . To compare the behavior and subcellular distribution of all mutant Rab7 proteins , we investigated the subcellular location of the nine Drosophila and human Venus-tagged proteins . Venus-Rab7-WT-positive compartments are present in both presynaptic boutons and the surrounding muscle ( Figure 2G ) . Expressing the constitutively active Rab7-Q67L leads to increased compartment numbers mostly in the surrounding muscle , while Rab7-T22N exhibits only diffuse labeling ( Figure 2G ) . In contrast , the CMT2B variants accumulate in the center of synaptic boutons ( arrowheads in Figure 2G ) . Functional Rab7 is characterized by dynamic colocalization first with the early endosomal marker Rab5 and second with the early-to-late endosomal ESCRT protein Hrs ( Lloyd et al . , 2002 ) . Colocalization analysis with these markers therefore provides a means to assess the functional state of the nine Venus-tagged Rab7 variants . Co-labeling with Rab5 and Hrs reveals significant overlap with Venus-Rab7-WT that is further increased for Venus-Rab7-Q67L ( Figure 2H ) . In contrast , both Rab5 and Hrs colocalization are almost completely lost for Venus-Rab7-T22N as well as all CMT2B proteins ( Figure 2H ) . In particular , the synaptic accumulations of CMT2B variants exclude endosomal markers , suggesting that they are not functional . These findings are consistent with the absence of electrophysiological defects of these synapses . To further analyze the function of each of the nine Venus-Rab7 proteins , we analyzed their recruitment to distinct compartments vs diffuse localization in the ventral ganglion where the motor neuron cell bodies reside . The GTP-locked Rab7-Q67L has an increased activity that is quantitatively reflected in its localization to late endosomal compartments , whereas the GDP-locked Rab7-T22N is considered inactive , as reflected by a complete failure to be recruited to endosomal compartments . Hence , compartment localization is a quantitative readout for GTP-dependent Rab7 activity ( Mukhopadhyay et al . , 1997a; Zhang et al . , 2007; Chan et al . , 2011 ) . As shown in Figure 2I–J , the nine variants exhibit localization to distinct compartments ( arrows in Figure 2I ) and diffuse labeling in varying ratios . Quantification of the ratio of compartment fluorescence to total fluorescence in 3D datasets of individual cell bodies shows that 38% of Rab7-Q67L , compared to 21% of wild-type Rab7 and less than 4% of Rab7-T22N localize to distinct compartments ( Figure 2J ) . These data are consistent with a plethora of previous studies on these widely studied Rab7 variants ( Mukhopadhyay et al . , 1997a , 1997b; Bucci et al . , 2000; Zhang et al . , 2007 ) . Analysis of the Venus-tagged CMT2B variants reveals mostly a diffuse cytoplasmic location , similar to Rab7-T22N , with 2–11% of the total fluorescent protein localized to distinct compartments . Hence , the reduced compartment localization suggests that the CMT2B proteins retain ∼5–50% of the function of Rab7-WT . A similar difference is observed for the human proteins hRab7-WT and hRab7-K157N ( Figure 2J ) . These data do not support the hypothesis that the disease variants mimic the gain-of-function of the Q67L mutations . Instead , they suggest that CMT2B variants exhibit reduced function more similar to the T22N mutant , but may retain some wild-type protein function that , based on quantitative analysis of localization to distinct compartments , ranges between 5–50% of Rab7-WT . Our findings indicate that rab7 is not acutely required for synaptic function in motor neurons or photoreceptor sensory neurons . However , loss of rab7 leads to progressive , stimulation-dependent loss of synaptic function , which can be studied in photoreceptor neurons over prolonged time periods . We therefore asked whether overexpression of any of the nine Venus-Rab7 variants causes neuropathy-like phenotypes over time in vivo . Protein localization of the Venus-Rab7 variants in the photoreceptor synaptic layer in the brain revealed that all four CMT2B proteins accumulate under the exclusion of other endosomal markers , very similar to our findings in motor neurons ( Figure 3A–B , compare Figure 2G ) . In contrast , both Venus-Rab7-WT and Venus-Rab7-Q67L exhibit significant colocalization with both Rab5 and Hrs at synapses ( Figure 3B ) . We also analyzed cell bodies of central nervous system neurons in the same brains . Here , the constitutively active Rab7-Q67L exhibits a typical increase of endosomal compartments positive for Rab7-Q67L , Rab5 and Hrs ( Figure 3C–D ) . In contrast , the CMT2B variants exhibit strongly reduced Rab5 and Hrs colocalization . These findings resemble the findings in motor neurons and suggest that CMT2B proteins are mostly dissociated from functional endosomes . 10 . 7554/eLife . 01064 . 007Figure 3 . Overexpression of Venus-tagged Rab7 variants in photoreceptor sensory neurons . ( A–D ) Immuno-histochemical analyses of Venus-Rab7 protein localization and colocalization with the endosomal markers Rab5 ( blue ) and Hrs ( red ) . ( A ) Longitudinal sections through the adult lamina , where photoreceptor neurons R1-R6 terminate . ( C ) Cell bodies of neurons in the medulla cortex . Green: Venus-Rab7 proteins , red: Hrs , blue: Rab5 . ( B and D ) Quantification of Venus-Rab7 colocalization with Rab5 and Hrs for the indicated genotypes at synapses ( B ) and in cell bodies ( D ) . ( E and F ) Quantification of morphological and functional analyses of Rab7 overexpression for all nine variants reveals a loss of endosomal colocalization for all CMT2B mutants . ( genotype: UAS-rab7-X/+; rab7Gal4-knock-in/+ ) ( E ) The numbers of rhabdomeres per ommatidial cross section reveal that overexpression of none of the rab7 mutants leads to morphological disruption similar to the mutant . ( F and G ) Overexpression of none of the mutant rab7 variants causes defects in ERG depolarization or synaptic transmission ( ‘on’ transient ) . Control and overexpression experiments exhibit no statistically significant variance ( ANOVA ) . Picture and ERG traces for all genotypes are shown in Figure 3—figure supplement 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 01064 . 00710 . 7554/eLife . 01064 . 008Figure 3—figure supplement 1 . ERG recordings from overexpression experiments of the indicated rab7 mutant variants at identical levels using rab7Gal4-knock-in in heterozygosity . ( A ) Representative ERG traces . Quantification in Figure 2C , D . ( B ) Representative rhabdomere labelings in eye cross-sections . Quantification in Figure 2B . Scale bar in ( B ) : 10 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 01064 . 008 Next we asked whether the CMT2B proteins impair sensory neuronal function over longer periods of time or in a usage-dependent manner . To challenge the photoreceptor neurons , we exposed the flies to 10 days of constant light stimulation . In stark contrast to the null mutant , neither overexpression of rab7Q67L , rab7T22N nor any of the CMT2B variants in the Drosophila or human Rab7 protein causes any obvious morphological or functional defects ( Figure 3E–G , Figure 3—figure supplement 1 ) . In summary , our findings are not consistent with the gain-of-function hypothesis . None of the mutant Rab7 proteins exhibit obvious dominant negative characteristics at 2–3-fold overexpression in the presence of wild-type Rab7 in vivo . To measure the wild-type function or a putative dominant deleterious effects of each of the rab7 variants we designed a quantitative population experiment . We established nine stocks that are both heterozygous for the null mutant rab7Gal4-knock-in allele and heterozygous for each of the nine UAS-rab7 transgenes . Note that the rab7Gal4-knock-in is homozygous lethal , but can become homozygous viable in the presence of one or two copies of the rescuing UAS-rab7 transgenes ( compare Figure 1H ) . Hence , four genotypes are possible in each of the nine stocks ( Figure 4A ) : the heterozygotes with overexpression of one ( light blue ) or two ( dark blue ) copies of a UAS-rab7 variant and the null mutant homozygotes , rescued by one ( light green ) or two copies ( dark green ) of a UAS-rab7 variant . 10 . 7554/eLife . 01064 . 009Figure 4 . Rescue experiments using rab7Gal4-knock-in driven UAS-venus-rab7 variants . ( A ) A population experiment over four generations reveals the fittest and unhealthy genotypes . Light Blue shows the fraction of flies heterozygous for the null mutant and expressing 1 copy ( 2–3-fold overexpression ) for each of the nine transgenes ( genotype: UAS-rab7-X/+; rab7Gal4-knock-in/+ ) . Dark Blue shows the fraction of heterozygous rab7 flies expressing two copies of the respective transgenes ( genotype: UAS-rab7-X/UAS-rab7-X; rab7Gal4-knock-in/+ ) . Light green shows the fraction of homozygous null mutant flies rescued through expression of one copy ( 2–3-fold overexpression ) of the respective transgene ( genotype: UAS-rab7-X/+; rab7Gal4-knock-in/rab7Gal4-knock-in ) . Dark green shows the fraction of null mutant flies rescued by two copies ( 4–6-fold ) overexpression of one of the transgenes ( genotype: UAS-rab7-X/UAS-rab7-X; rab7Gal4-knock-in/rab7Gal4-knock-in ) . ( B–F ) Rescue experiments of synaptic and neuronal degeneration using each of the nine rab7 transgenes in null mutant photoreceptors after 10 days of constant light stimulation . ( B ) Representative ERG traces with measured components in ( D ) and ( E ) . ( C ) Representative eye cross sections showing the array and number of rhabdomeres per ommatidium , and corresponding counts in ( F ) . Variance for measurements in ( D–F ) is significantly different for rab7 null mutant and rab7T22N rescue ( ANOVA ) . Scale bar in ( C ) : 10 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 01064 . 009 We incubated the nine stocks for four generations and counted the relative populations . The ratio of genotypes that emerge after four generations is indicative of the fitness of each genotype relative to the others in the population . For instance , if overexpression of any of the nine rab7 variants has deleterious effects , two copies of the transgene will be less common than one . In the case of wild-type Rab7 , three quarters of the population had two copies of the Rab7-WT transgene , indicating little to no toxicity of the overexpressed Rab7-WT . Furthermore , 12% of the flies in the population were rescued homozygous rab7 null mutants ( light and dark green in Figure 4A ) . We note that this number reflects a remarkably healthy rescue , because a significant part of the population lives without the wild-type rab7 chromosome , but instead only due to Rab7-Gal4-driven UAS-rab7WT in direct competition with wild-type flies over four generations . Indeed , the ratio of rescued to wild-type progeny prior to the competition experiment obeyed Mendelian ratios and homozygous rab7 null mutant flies rescued with Rab7-WT expression ( UAS-rab7WT/UAS-rab7WT; rab7Gal4-knock-in/rab7Gal4-knock-in ) appear indistinguishable from wild type . In the case of Rab7-T22N 40% of the population had two copies of this widely used ‘dominant negative’ transgene , but no rescued homozygous rab7 null mutants were observed . Hence , Rab7-T22N exhibits little or no toxicity but is also not sufficiently functional to efficiently rescue the null mutant . Interestingly , close investigation of the fly stock revealed that the rab7T22N mutant actually yielded rare adult escapers that died shortly after emergence . In contrast , Rab7-Q67L never occurred in two copies , indicating that high levels of this constitutively active mutant are associated with reduced fitness ( Figure 4A ) . In contrast , two of the four CMT2B variants lead to populations consisting entirely of individuals expressing two copies of these disease-associated proteins . Remarkably , almost no rescue of the null mutant with a single copy of these CMT2B transgenes was observed , but between 12–35% of individuals were rescued by two copies of the CMT2B mutant transgenes . These findings suggest that higher levels of at least three of the four CMT2B mutant proteins increase the fitness and are indeed required to compensate for the loss of wild-type rab7 . In summary , all CMT2B disease mutants , including the Drosophila and human Rab7-K157N , exhibited significant rescue ( Figure 4A ) . Indeed , all UAS-rab7 variants except Rab7-T22N rescued rab7 lethality and yielded adult flies with no obvious defects . These findings reveal that the CMT2B proteins do not reduce fitness or neuronal health , but retain sufficient wild-type function to compensate for loss of rab7 if expressed at 2–6-fold endogenous levels . To measure the ability of all rab7 variants to rescue the progressive and usage-dependent synaptic and neuronal degeneration observed in the null mutant we measured ERGs in photoreceptors after 10 days constant light exposure . As shown in Figure 4B–F , all rab7 variants rescued the null mutant phenotypes . Only rab7T22N exhibited an only partial ability to rescue the response amplitudes ( Figure 4B , D ) , synaptic function ( Figure 4B , E ) and rhabdomere morphology ( Figure 4C , F ) . However , the observation that rab7T22N exhibits some rescue of the null mutant further indicates that it retains some wild-type function and does not obviously act as a genetically dominant negative mutant at these levels of overexpression in Drosophila . While both human Rab7 proteins exhibit a reduced ability to rescue synaptic functions under these stress conditions , hrab7K157N does not significantly differ from hrab7WT . The ability of all CMT2B mutant proteins to rescue the null mutant phenotype indicates that replacement of wild-type rab7 with mild overexpression in the correct spatiotemporal pattern is sufficient to compensate for the partial loss of function of the mutant variants . In summary , we conclude that all CMT2B mutants retain significant levels of wild-type function . What causes the reduced wild-type function of the CMT2B mutant proteins ? To measure the function of each of the Rab7 variants , we devised a live imaging method to assay the well-characterized role of Rab7 in the maturation of late endosomal compartments . We developed a culture system for high-resolution live imaging of endosomal dynamics in photoreceptor neurons in intact eye-brain complexes and made use of the lysosomal marker spinster/benchwarmer ( spin ) ( Sweeney and Davis , 2002; Dermaut et al . , 2005 ) . Co-expression of the Venus-tagged Rab7 proteins with spin-RFP reveals the dynamics of endolysosomal conversion both in cell bodies and at synapses ( Videos 1–10 ) . In cell bodies , Venus-Rab7-WT marks on average 15 clearly distinguishable , circular endosomal compartments of 0 . 5 µm diameter or more per 500 µm² tissue area . In contrast , Rab7-T22N marks no individual compartments and expression of the constitutively active Venus-Rab7-Q67L leads to a doubling of compartment numbers , many of which are significantly larger than those observed for wild type ( Figure 5A–B ) . As seen before , all four CMT2B mutant proteins exhibit mostly diffuse labeling similar to Rab7-T22N; however , in contrast to Rab7-T22N and Rab7-K157N , the other three CMT2B variants also reveal clear circular labeling of distinct compartments of 0 . 5 µm diameter or above ( Figure 5C ) , albeit at significantly reduced numbers ( Figure 5G , H ) . 3D time lapse imaging of the dynamics of these compartments reveals conversion of 8–9% of these compartments into spin-RFP positive compartments within a 5 min imaging interval . Conversion of wild type and CMT2B Venus-Rab7 marked compartments typically occurs quickly and within the 30 s between image acquisition time points ( Figure 5D–F ) . Remarkably , the conversion rate of 8–9% per 500 µm² tissue area per 5 min is identical between all genotypes ( Figure 5I ) . Based on these measurements , the CMT2B proteins exhibit 5–50% of the wild type levels of recruitment to endosomal compartments , but no defects in endosomal conversion . 10 . 7554/eLife . 01064 . 010Video 1 . Venus-Rab7-Q67L and spin-RFP in the eye . DOI: http://dx . doi . org/10 . 7554/eLife . 01064 . 01010 . 7554/eLife . 01064 . 011Video 2 . Venus-Rab7-Q67L and spin-RFP in photoreceptor terminals . DOI: http://dx . doi . org/10 . 7554/eLife . 01064 . 01110 . 7554/eLife . 01064 . 012Video 3 . Venus-Rab7-WT and spin-RFP in the eye . DOI: http://dx . doi . org/10 . 7554/eLife . 01064 . 01210 . 7554/eLife . 01064 . 013Video 4 . Venus-Rab7-WT and spin-RFP in photoreceptor terminals . DOI: http://dx . doi . org/10 . 7554/eLife . 01064 . 01310 . 7554/eLife . 01064 . 014Video 5 . Venus-Rab7-V162M and spin-RFP in the eye . DOI: http://dx . doi . org/10 . 7554/eLife . 01064 . 01410 . 7554/eLife . 01064 . 015Video 6 . Venus-Rab7-V162M and spin-RFP in photoreceptor terminals . DOI: http://dx . doi . org/10 . 7554/eLife . 01064 . 01510 . 7554/eLife . 01064 . 016Video 7 . Venus-Rab7-K157N and spin-RFP in the eye . DOI: http://dx . doi . org/10 . 7554/eLife . 01064 . 01610 . 7554/eLife . 01064 . 017Video 8 . Venus-Rab7-K157N and spin-RFP in photoreceptor terminals . DOI: http://dx . doi . org/10 . 7554/eLife . 01064 . 01710 . 7554/eLife . 01064 . 018Video 9 . Venus-Rab7-T22N and spin-RFP in the eye . DOI: http://dx . doi . org/10 . 7554/eLife . 01064 . 01810 . 7554/eLife . 01064 . 019Video 10 . Venus-Rab7-T22N and spin-RFP in photoreceptor terminals . DOI: http://dx . doi . org/10 . 7554/eLife . 01064 . 01910 . 7554/eLife . 01064 . 020Figure 5 . Live imaging reveals that CMT2B Rab7 proteins exhibit defective endosomal recruitment but do not affect endosomal maturation . ( A–F ) Snapshots from live imaging datasets for three genotypes with a time lapse of 30 s ( A and D ) Venus-Rab7-WT and spin-RFP; ( B and E ) Venus-Rab7-Q67L and spin-RFP; ( C and F ) Venus-Rab7-N161T and spin-RFP . Arrows mark individual Rab7-positive compartments >=0 . 5 µm diameter . The same compartments marked in ( A–C ) are followed over time in ( D–F ) . ( G ) Quantification of compartments as in ( A–F ) for all genotypes per 500 μm2 area . ( H ) Quantification of Venus-Rab7 compartments ( individual distinguishable green punctae ) per photoreceptor axon terminal in the brain for the indicated genotypes . ( I ) Quantification of the fraction of compartments that underwent green-to-red conversion over a 5 min period in same data as ( G ) . Note that Rab7-T22N and Rab7-K157N did not mark sufficient compartments for this analysis . See also: Videos 1–10 . Scale bar in ( A ) : 10 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 01064 . 020 Taken together , these live imaging experiments reveal that the CMT2B proteins are poorly functioning proteins; their inefficient recruitment to endosomes provides a mechanistic basis for their partial loss-of-function . However , the CMT2B proteins do not dominantly affect the conversion rate of endosomal compartments they are recruited to , corroborating the findings that the CMT2B proteins do not dominantly impair the function of wild-type Rab7 . Based on our findings we propose a partial loss-of-function model for CMT2B in which partial or complete loss of one copy of rab7 causes degeneration . This model predicts that sensory neurons in vivo are sensitive to the precise dosage of rab7-dependent endolysosomal degradation . We therefore analyzed viable heterozygous flies for increased sensitivity to stimulation using 10-days of light exposure in white-eyed flies . As shown in Figure 6A , B , constant stimulation leads to almost complete loss of synaptic function in heterozygous rab7 mutants similar to the null mutant . This defect is fully rescued by the absence of stimulation and does not significantly affect the photoreceptor response amplitude ( Figure 6A , B ) . Furthermore , rab7 heterozygous photoreceptors after stimulation exhibit a morphological haploinsufficiency phenotype with partial loss of rhabdomere structures ( Figure 6C–F ) . These findings indicate that partial loss of rab7 affects the function of these sensory neurons in a stimulation-dependent manner without any other obvious effects on development , viability or other tissues in Drosophila . 10 . 7554/eLife . 01064 . 021Figure 6 . Photoreceptor neurons are haploinsufficient for rab7 function as revealed by functional and morphological measurements 10 days after constant light stimulation . ( A and B ) . Quantification of ERG depolarization and synaptic transmission ( ‘on’ transients ) in null mutants and heterozygotes for rab7 . ( C–F ) Ommatidial rhabdomere composition from eye cross sections exposed to various experimental conditions . Scale bar in ( D ) for ( D–F ) : 10 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 01064 . 021 Our findings indicate that neurons are sensitive to the precise levels of Rab7 . If the CMT2B mutations are hypomorphic alleles and encode proteins with 5–50% function , as our data indicates , then the cause of the adult-onset loss of synaptic function in patients may be due to this partial loss of function . On the other hand , numerous overexpression experiments revealed clear dominant effects of the CMT2B proteins in cell culture ( Spinosa et al . , 2008; Cogli et al . , 2010; McCray et al . , 2010; Basuray et al . , 2013; Cogli et al . , 2013; Zhang et al . , 2013 ) . To determine the levels and range of CMT2B protein levels that are sufficient to rescue partial loss of rab7 or induce dominant effects , we generated heterozygous rab7 flies ( one copy of the wild-type chromosome ) and a second copy at half , identical and >10-fold levels of the endogenous copy of rab7 . To generate these flies , we introduced the temperature-sensitive Gal4 suppressor Gal80ts ( McGuire et al . , 2003 ) and determined temperature conditions to match the CMT2B overexpression levels to endogenous Rab7 levels and further reduce the CMT2B expression 50% below endogenous levels . As shown in Figure 7A , an antibody against Drosophila Rab7 ( Chinchore et al . , 2009 ) recognizes both endogenous Rab7 as well as our Venus-tagged transgenes . More than 10-fold overexpression of the CMT2B variants was achieved in the absence of Gal80ts at 25°C . Remarkably , all transgenes over this large range rescued the haploinsufficiency phenotype after a 5-day light stimulation protocol in the ERG depolarization ( Figure 7B ) , synaptic function ( Figure 7C ) and Rhabdomere structure of the eye ( Figure 7D–J ) . These findings quantitatively support two key findings reported in this study: First , Rab7 CMT2B proteins at 50% less than endogenous heterozygous levels ( corresponding to 25% of total Rab7 function ) retain sufficient wild-type function to rescue the rab7 phenotype . Second , Rab7 CMT2B proteins at >10-fold overexpression rescue , but exhibit no toxic or other obvious gain-of-function effects . Together with our quantitative colocalization and live imaging experiments that show quantitatively reduced wild-type function , these data corroborate our conclusion that CMT2B proteins represent partial loss-of-function alleles with no toxic gain-of-function in Drosophila . 10 . 7554/eLife . 01064 . 022Figure 7 . A wide range of CMT2B expression levels rescues partial loss of rab7 function without dominant toxic effects . ( A ) Western blots for Rab7 of adult fly eyes expressing different levels of the Venus-tagged mutant transgenes . ( B and C ) ERG depolarization and On transient measurements after a 5-day light stimulation protocol . WT and CMT2B mutant values are not statistically significantly different in an ANOVA test . ( D ) Morphological analyses of rhabdomere structure in fly eyes with a ratio of 0 . 5:1 expression of the mutant transgenes vs endogenous heterozygous protein amounts . ( E–J ) Representative images of the rhabdomer structure for the indicated genotypes . DOI: http://dx . doi . org/10 . 7554/eLife . 01064 . 022 Despite the identification of the CMT2B mutations as hypomorphic alleles , our findings are remarkably consistent with the majority of previous biochemical findings . For example , mammalian heterologous expression studies revealed that the GTP: GDP ratio is altered , which could suggest a dominant or constitutively active mutant . However , the same study revealed that , in contrast to the constitutively active variant , the total amount of both GTP and GDP bound Rab7 is dramatically decreased for the CMT2B mutant proteins ( De Luca et al . , 2008; Spinosa et al . , 2008 ) . Furthermore , our findings are largely consistent with the thorough biochemical characterization of the CMT2B Rab7 proteins by McCray et al . ( 2010 ) . Specifically , McCray et al . found that the CMT2B proteins have no GTPase activity defect , but augmented protein activity , based on increased guanine nucleotide dissociation , hydrolysis-independent inactivation , quantitative changes on effector interactions and decreased membrane cycling . However , McCray et al . also showed that the CMT2B proteins retain sufficient wild-type function to rescue reduced rab7 activity in HeLa cells without obvious toxic gain-of-function effects . All these findings are consistent with partial loss of function alleles . The key difference to the present study lies in the interpretation that misregulation of the disease proteins’ Rab7 activity may have a dominant effect , as opposed to our simpler interpretation as partial loss-of-function . Our findings do not provide an obvious explanation for recent reports on specific dominant effects of CMT2B protein overexpression in cell culture ( Spinosa et al . , 2008; Cogli et al . , 2010; McCray et al . , 2010; Basuray et al . , 2013; Cogli et al . , 2013; Zhang et al . , 2013 ) . These dominant phenotypes could be a result of overexpression in a heterologous cell line , or a bona fide property of the CMT2B proteins . In some cases , high levels of overexpression might at least partially explain the effects . Nonetheless , our data do not rule out the possibility of a dominant interaction with protein complexes or protein functions that are specific to vertebrate cells . However , we note that none of the studies to date has reported progressive neuronal or synaptic deterioration as a consequence of CMT2B protein overexpression in motor or sensory neurons ( Cogli et al . , 2013; Zhang et al . , 2013 ) . While we show here that such putative vertebrate-specific properties are not necessary to cause the neuropathy in a Drosophila model , they may contribute to specific neuronal changes over time in the different heterologous cell culture systems as well as in patients . A second limitation of the Drosophila model is revealed by our effort to precisely mimic the human patient genotypes that lead to identical levels of expression of one wild type copy of Rab7 and one CMT2B mutant variant . In patients , this genotype leads to slow motor- and sensory neuron degeneration with an onset between 12 to more than 40 years . In Drosophila , heterozygotes with an additional copy of a CMT2B chromosome are healthy and photoreceptor neurons are not sensitized to stimulation-dependent degeneration within the limits of our functional and morphological assays . The observation that this genotype is sufficient for normal human development and function for decades offers a possible explanation for this limitation of the fly model . Indeed , slow , adult-onset degeneration over a period of years is not easily mimicked in any model organism . It will be interesting to see if a rab7 heterozygous mouse model exhibits stimulation-dependent degeneration of motor- or sensory neurons over time , or if even the lifespan of a mouse is not sufficient to model this aspect of the human neuropathy . We propose that CMT2B in patients reflects a rab7 dosage-dependence , as described here in a fly model . Since no patients have been reported with a null mutant allele of rab7 , we speculate that complete loss of one copy of rab7 in humans may cause lethality ( Figure 8C ) . Adding a second rab7 allele with 5–50% function may be sufficient to retain normal function in most cells but slowly cause defects over a period of years only in the cells most sensitive to rab7-dependent endolysosomal degradation ( Figure 8C ) . Our model predicts that only mutations that reduce rab7 function within a certain range will lead to a neuropathy , explaining the rarity of these mutations and the variability of disease onset . We note that the CMT2B mutant with the least function in the fly ( Figure 8A ) , rab7K157N , is a sporadic new mutation in a patient with the earliest reported childhood onset ( 12 years ) of the four mutant variants ( Meggouh et al . , 2006 ) . Mild reductions in endolysosomal degradative capacity may be caused by numerous genetic polymorphisms as well as an increased degradative load in various degenerative disorders characterized by intracellular accumulations . Such accumulations and subsequent endomembrane degradative responses are hallmarks of most neurodegenerative disorders . An elevated neuronal demand for endolysosomal degradation is further highlighted by the recent discovery of a neuron-specific branch of the endolysosomal system ( Williamson et al . , 2010; Haberman et al . , 2012; Wang and Hiesinger , 2012 ) . In the case of CMT2B our findings suggest an increase of endolysosomal function as a therapeutic approach , which is opposite to the current proposal to reduce the function of the mutant Rab7 proteins . Neuronal sensitivity to rab7-dependent degradation may be a common factor contributing to neuronal pathology in numerous disorders with reduced degradation or increased degradative burden . Reduced endolysosomal capacity may thus contribute to pathology and increased endolysosomal function may represent a more general therapeutic opportunity . The generation of rab7 targeting vector by recombineering was performed as previously described ( Chan et al . , 2011 , 2012 ) with some modifications . We generated a new genomic fragment containing the rab7 locus with asymmetric flanking homology arms of 10 kb at 5′ and 5 kb at 3′ . The final targeting vector was injected into PBac{y[+]-attP-3B}VK00033 landing site by Rainbow Transgenic Services ( CA , USA ) using PhiC31-mediated integration . The knock-out screen was performed according to published protocols ( Chan et al . , 2011 , 2012 ) . The mobilization and reintegration by ends-out homologous recombination was initiated by heat shock . We screened approximately 80 , 000 F2 progeny for separation of the targeting cassette from the original landing site and identified 61 reintegrations in the genome . 58 of the 61 genomic integration events occurred on the third chromosome , the correct rab7 bearing chromosome . 12 lethal lines die at late pupal stage . To characterize the potential knock-outs molecularly , the genomic DNA of homozygous larvae was extracted and examined by PCR . The 46 viable lines showed no correct replacement . Of the 12 lethal lines , nine failed to complement two independent deficiencies uncovering the rab7 locus ( Df ( 3R ) ED10893 , 95C8;95E1 , 3R:19713027;19930781 and Df ( 3R ) Exel6196 95C12;95D8 , 3R:19747854-19747855;19857149 ) . Two knock-out lines were verified by rescue experiments as described in the manuscript . Mutagenesis of rab7 wild-type cDNA to generate the four CMT2B mutations ( L129F , K157N , N161T , and V162M ) was conducted by PCR SOEing . These four mutant cDNAs , the wild-type cDNA as well as the previously generated T22N and Q67L variants ( Zhang et al . , 2007 ) were cloned into the pTVW vector ( The Drosophila Gateway Vector Collection , Carnegie Institution ) to generate N-terminal Venus fusions . For human hrab7Awt and hrab7AK157N , Venus and hRab7 were PCR amplified separately and then fused using SOEing PCR to generate the N-terminal Venus fusions . All nine venus-rab7 variants were subcloned using PCR-generated Not1/Xho1 sites and ligated into the pUASt-attB ( Bischof et al . , 2007 ) for subsequent PhiC31-integrase mediated insertion in the PBac{y[+]-attP-3B}VK00002 landing site by Rainbow Transgenic Services ( CA , USA ) . The following nine stocks were generated:Wild type: yw; UAS-Venus-Rab7-WT/CyO; FRT82B rab7Gal4-knock-in/TM3 , SbGTP-bound ‘Constitutively active’: yw; UAS-Venus-Rab7-Q67L/CyO; FRT82B rab7Gal4-knock-in/TM3 , SbGDP-bound ‘Dominant Negative’: yw; UAS-Venus-Rab7-T22N/CyO; FRT82B rab7Gal4-knock-in/TM3 , SbCMT2B mutant: yw; UAS-Venus-Rab7-K157N/CyO; FRT82B rab7Gal4-knock-in/TM3 , SbCMT2B mutant: yw; UAS-Venus-Rab7-L129F/CyO; FRT82B rab7Gal4-knock-in/TM3 , SbCMT2B mutant: yw; UAS-Venus-Rab7-N161T/CyO; FRT82B rab7Gal4-knock-in/TM3 , SbCMT2B mutant: yw; UAS-Venus-Rab7-V162M/CyO; FRT82B rab7Gal4-knock-in/TM3 , SbHuman rab7A wild type: yw; UAS-Venus-hRab7A-WT/CyO; FRT82B rab7Gal4-knock-in/TM3 , SbHuman rab7A CMT2B mutant: yw; UAS-Venus-hRab7A-K157N/CyO; FRT82B rab7Gal4-knock-in/TM3 , Sb rab7-Fwd-ATG: ATG TCC GGA CGT AAG AAA TC; rab7-Rev-Stp: TTA GCA CTG ACA GTT GTC AG; L129F-Fwd: ATA AGG TGG ATT TCG ACA AC; L129F-Rev: TGG CGG TTG TCG AAA TCC AC; K157N-Fwd: AAA CGT CCG CCA ACG AGG GC; K157N-Rev: TTG ATG CCC TCG TTG GCG GA; N161T-Fwd: AGG AGG GCA TCA CCG TGG AGN161T-Rev: GCC ATC TCC ACG GTG ATG CC; V162M-Fwd: AGG GCA TCA ACA TGG AGA TG; V162M-Rev: AAC GCC ATC TCC ATG TTG AT . V-rab7-fwd-Not1: ATA AGA ATG CGG CCG CAC CAT GGT GAG CAA GGG CGA G; V-rab7-rev-Xho1: CCG CTC GAG TTA GCA CTG ACA GTT GTC . V-rab7-fwd-Not1: ATA AGA ATG CGG CCG CAC CAT GGT GAG CAA GGG CGA G; venus-hrab7 Rev: TAGAGGTCATCCGGTGCTTGTACAGCTCGTCCATG; venus-hrab7 Fwd: GCTGTACAAGCACCGGATGACCTCTAGGAAGAAAG; hrab7-Rev-XhoI: CCGCTCGAGTCAGCAACTGCAGCTTTCTG . The primers used for amplifying the 500 bp homology arms , left arm ( LA ) and right arm ( RA ) :LA Fwd ACAAGTTTGTACAAAAAAGCAGGCTTACAGTGTAGAAAGCAGCAA; LA Rev GGCAACGGATCCTCACGTTGGTTTCGGAACAC; RA Fwd ACGTGAGGATCCGTTGCCACCGCTTCCTGCAT; RA Rev ACCACTTTGTACAAGAAAGCTGGGTACACCGTCACTCACTAGACC . The replacement of the rab7 open reading frame with the Gal4knock-in cassette was performed by adding homology region to the Gal4knock-in cassette using the following primers: 5’-AAGAAACCATCACACCCCTACACTTCCTAATCGAATTAGAGGAAACCGCAATGGTATTTTTAACACACAATCAATAATATTCTGTGGTTTTCAGCACCAAATACCAAAAGAAATAACCCCGAGTAAGCCAACGCCACAAACTGCATCGAAATGAAGCTACTGTCTTCTAT-3’; 5’-TATGCTGTTTTGCTGAAATTGTTTTACTTAATCATATAACACCTTCCTCTATCGTCCTTTGTGTTGCTTGCTTCTCATGTTCATTATTATGTTGGAAATATTATTTAATAATATAGATTGTGTAATTATCCATTTTGCGTTGTTGTTACACCCACCCTTTGCTGCTGCGC -3’ . The following primers were used to screen for potential correct knock-outs by amplifying a region within rab7 or within the cassette:rab7-KOscrn-Fwd AGAGGAAACCGCAATGGTAT; rab7-KOscrn-Rev TTCCTCTATCGTCCTTTGTG; GL5 ( Cassette Fwd ) GAAATCACGGCTAGTAAAATTGAT; RR1 ( Cassette Rev ) CTTAGCGACGTGTTCACTTTGCT . PCR on the 12 lethal lines indicated presence of the knock-in cassette and absence of the rab7 open reading frame . Eye mosaics were generated using the ey3 . 5FLP insertion on the X chromosome generated by Iris Salecker ( Chotard et al . , 2005; Mehta et al . , 2005 ) . The rab7Gal4-knock-in chromosome was recombined on an FRT82B chromosome using G418 selection . For rescue experiments UAS-venus-rab7 variants were introduced on the second chromosome . Flies were kept at 25°C unless otherwise indicated . Dark-reared flies were kept in the same room next to light-raised flies . For constant light raising fly vials were kept in an aluminium foil covered box that provided even illumination provided by a Leica KL1500LCD cold light source . Light intensity was adjusted to 600 Lux using a hand-held digital lux meter at the level of the fly vials , 20 cm from the light source . To reduce expression levels of the UAS-rab7 transgenes we generated flies of the genotype UAS-rab7-X/+; rab7Gal4-knock-in/tub-Gal80ts . The tub-Gal80ts stock was obtained from the Bloomington Stock Center ( stock number 7018 ) . To reduce level to endogenous levels flies were kept at 22°C . To reduce levels to 50% of endogenous levels , flies were kept at 18°C . Larval and pupal brains and adult retinae were dissected and prepared for confocal microscopy as previously reported ( Williamson and Hiesinger , 2010; Williamson et al . , 2010 ) . The tissues were fixed in phosphate buffered saline ( PBS ) with 3 . 5% formaldehyde for 15 min and washed in PBS with 0 . 4% Triton X-100 . High-resolution light microscopy was performed using a Confocal Microscope ( Leica SP5 ) . Imaging data was processed and quantified using Amira 5 . 3 ( Indeed , Berlin , Germany ) and Adobe Photoshop CS6 as described ( Williamson et al . , 2010 ) . The following antibodies were used: Chaoptin ( at 1:100 ) , CSP ( 1:50 ) , Hrs ( at 1:300 ) , Rab7 ( at 1:500 ) . Secondary antibodies used were Cy3 and Cy5 ( Jackson ImmunoResearch Laboratories , Inc . , West Grove , PA ) raised against guinea pig , mouse or rabbit . Rhabdomeres were labeled with phalloidin ( at 1:500 ) as previously reported ( Haberman et al . , 2012 ) . Live imaging was performed according to published protocols ( Williamson and Hiesinger , 2010 ) with the following modifications: Pupal eye-brain complexes were dissected in ice-cold Schneider’s Drosophila Medium ( Gibco , Grand Island , NY ) at P+20–30% . Dissected tissues were pipetted into a drop of 0 . 4% dialyzed agarose ( in Schneider’s Medium ) on a Sylgard layer . The sample was covered with a coverslip . Imaging was performed with a Leica SP5 resonant scanner , using a 63X glycerol lens . A time interval of 30 s was used for live imaging . The voxel size in all scans was 0 . 096 × 0 . 096 × 0 . 500 μm . The compartment localization was quantified using Amira 5 . 3 ( Indeed , Berlin , Germany ) on high-resolution 3D confocal datasets . The confocal scans were obtained in 8 bit with no fluorescence signal outside the dynamics range . 10 cells were selected per genotype and individually cropped to yield separate datasets . Each dataset was separately binarized by threshold segmentation and visually controlled for only including clearly distinct compartments under exclusion of diffuse labeling . The cumulative fluorescence intensity of all voxels inside and outside the thresholded compartments was determined for the entire cell volume . Finally , the ratio of cumulative fluorescence inside/total fluorescence was calculated for each cell . Mean ratio , standard error ( SEM ) and p values were calculated for each genotype . To view the cell bodies and terminals of photoreceptors of rab7 mutant and control ommatidia by transmission electron microscopy , flies were fixed in a modified Karnovsky glutaraldehyde and formaldehyde fixative , followed by osmium , then embedded in Epon , sectioned at 50 nm , all as previously reported ( Meinertzhagen , 1996 ) . Images were obtained using a Philips Tecnai 12 at 80 kV . Images were captured with a Gatan 832 Orius SC1000 CCD camera using Gatan DigitalMicrograph software . ERGs were performed as described in Fabian-Fine et al . ( 2003 ) with the following modifications: flies were fixed using Elmer’s non-toxic Glue-All . We used 2M NaCl in the recording and reference electrodes . Electrode voltage was amplified by a Digidata 1440A , filtered through a Warner IE-210 , and recorded using Clampex 10 . 1 by Axon Instruments . A post-recording filter was also provided by the Clampex software . Light stimulus was provided in 1 s pulses by a computer-controlled white LED system ( Schott MC1500 ) . All ERG recordings in Figure 1 and Figure 6 were performed in matched white- genetic backgrounds , which are significantly more sensitive to light stimulation than pigmented eyes; recordings in Figures 3 and 4 were performed in flies with a single copy of mini-white and the same level of eye pigmentation . For quantification of depolarization and ‘on’ transients all experiment were carried out in duplicate or triplicate with at least 10 recording for each genotype and experimental condition . Third instar larvae were dissected and recorded from as described ( Imlach and McCabe , 2009 ) . The larvae were dissected in modified , ice-cold HL3 . 1 ( Stewart et al . , 1994 ) , containing ( in mM ) : Sucrose 115 , NaCl 70 , MgCl2 11 , NaHCO3 10 , KCl 5 , Trehalose 5 , HEPES 5 , CaCl2 0 . 5 , at pH 7 . 2 . To avoid damaging the microscope optics , the filets were adhered to a SYLGARD 184 ( Dow Corning , Midland , MI ) substrate with a cyanoacrylate glue that polymerizes in water; the dissection pins were removed after glue polymerization , and the motor neurons were severed . Current clamp recordings were performed at room temperature in HL3 . 1 . Recording microelectrodes were pulled to a sharp point with resistances between 20–40 MΩ . Only muscles 6 , in segments 2–4 , with membrane potentials between −65 and −60 mV were recorded . Voltage signals were amplified using a Multiclamp 700B amplifier ( Molecular Devices , LLC . , Sunnyvale , CA ) . Signals were digitized using a Digidata 1440 ( Molecular Devices , LLC . ) , filtered at 2 kHz , and recorded using Clampex 10 . 2 software ( Molecular Devices , LLC . ) . Miniature endplate potentials ( mEPPs ) were recorded for 60 s and analyzed using the ‘Template search’ feature of Clampfit 10 . 2 ( Molecular Devices , LLC . ) . Frequency was calculated by taking the number of events in a recording and diving by 60 s . For EPPs , stimulation was applied via the segmental motor nerve axons with a A365 stimulus isolator ( WPI ) , under digital control of the Clampex software . Stimulus electrodes had a ∼10 μm inner diameter after firepolishing . Superthreshold stimulation ( between 3–4 mA , for 100 ms ) was applied at 0 . 2 Hz for 60 s and recorded with Clampex . EPP amplitudes were measured using the Clampfit cursor function . Significance was calculated using Welch’s t test . Total proteins were extracted from fly heads in buffer containing 20 mM Tris , 150 mM NaCl , 1 mM PMSF , and 1x complete protease inhibitors ( Roche ) at pH 7 . 4 . The fly head extract was mixed well in 1% Triton X-100 ( BioRad , Hercules , CA ) and incubated for 1 hr at 4°C . Samples were centrifuged at 16 , 000 × g , 15 min at 4°C to remove cell debris . The resulting supernatants were loaded on 12% SDS-PAGE and transferred to PVDF membrane . Primary antibodies used were: mouse anti-human Rab7 ( 1:2500 ) ; rabbit anti-fly Rab7 ( 1:5000 ) ; muse ant-actin ( 1:5000 ) . Corresponding secondary antibodies were used at 1:10 , 000 . The signals were detected with the Pierce ECL Western blotting substrate ( Thermo Scientific , Rockford , IL ) . The quantification of Rab7 western blots was performed with ImageJ software ( NIH , Bethesda , MD ) . Data were analyzed with GraphPad Prism 4 .	Charcot-Marie-Tooth disease is an inherited disorder of the nervous system with symptoms that typically begin in adolescence or early adulthood . The sensory and motor nerves gradually degenerate , causing muscles to waste away and leading to the loss of touch sensation across the body . One subtype of the disease—Charcot-Marie-Tooth 2B—is caused by mutations in a gene called rab7 , which codes for a protein that helps to regulate the breakdown of waste proteins inside cells . Charcot-Marie-Tooth 2B is described as a genetically dominant disorder because all patients have one wild type copy and one mutant copy of the rab7 gene . Overexpression of the mutant gene in cells grown in culture alters many of the signaling pathways inside the cells , but it is unclear whether these alterations cause the pathology seen in the disease . Now , Cherry et al . have obtained new insights into the genetics of Charcot-Marie-Tooth 2B by creating the first animal model of the disorder . Fruit flies that did not have the rab7 gene in the light-sensitive sensory neurons in their eyes were used to compare normal and mutant cells . While the two cell types were initially similar , the mutant cells gradually degenerated in the adult animal . By contrast , cells that overexpressed a mutant form of the rab7 gene continued to function normally throughout adulthood . Moreover , when mutant Rab7 proteins were introduced into the cells that lacked the rab7 gene , the proteins restored the cells’ sensitivity to light . These results suggest that mutant Rab7 proteins do not cause degeneration; instead , it is the loss of normal Rab7 function that causes problems . At present , most research into treatment is aimed at finding ways to reduce the activity of mutant Rab7 proteins . However , the work of Cherry et al . suggests that increasing the activity of normal Rab7 proteins—or increasing the activity of alternative pathways that degrade waste proteins—may help to restore nerve function in this , and possibly other , neurodegenerative diseases .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "neuroscience" ]	2013	Charcot-Marie-Tooth 2B mutations in rab7 cause dosage-dependent neurodegeneration due to partial loss of function
Inhibitory interneurons target precise membrane regions on pyramidal cells , but differences in their functional effects on somata , dendrites and spines remain unclear . We analyzed inhibitory synaptic events induced by cortical , fast-spiking ( FS ) basket cells which innervate dendritic shafts and spines as well as pyramidal cell somata . Serial electron micrograph ( EMg ) reconstructions showed that somatic synapses were larger than dendritic contacts . Simulations with precise anatomical and physiological data reveal functional differences between different innervation styles . FS cell soma-targeting synapses initiate a strong , global inhibition , those on shafts inhibit more restricted dendritic zones , while synapses on spines may mediate a strictly local veto . Thus , FS cell synapses of different sizes and sites provide functionally diverse forms of pyramidal cell inhibition . Microcircuits of cerebral cortex are composed of excitatory pyramidal cells and different types of GABAergic interneurons . Inhibitory circuits regulate cortical activity ( Kubota et al . , 2011b; Lee et al . , 2012; Kubota , 2014 ) , development and plasticity ( Hensch , 2005; Donato et al . , 2013 ) . Perturbed inhibitory function is associated with pathologies including epilepsy , autism and schizophrenia ( Rubenstein and Merzenich , 2003; Gonzalez-Burgos et al . , 2010 ) . However , mechanisms controlling inhibitory synaptic actions are incompletely understood . For instance , inhibitory synapses target multiple membrane domains of pyramidal cells: soma , axon initial segment , dendritic shafts and spines ( Kisvarday et al . , 1985; Kawaguchi and Kubota , 1998; Szabadics et al . , 2006; Kubota et al . , 2007; Jiang et al . , 2013 ) . Contacts at these different sites produce inhibitory postsynaptic potentials ( IPSP ) with different properties ( Miles et al . , 1996; Xue et al . , 2014 ) . Recent data suggests IPSCs generated by FS basket cells may be matched to the level of synaptic excitation in cortical pyramidal cells ( Xue et al . , 2014 ) , and differ with target cell subtypes ( Lee et al . , 2014 ) . Unitary inhibitory postsynaptic currents ( uIPSCs ) are significantly smaller in neurons of Disc1 mice , a genetic model of depression , and may underlie reduced low-gamma oscillations in the frontal cortex ( Sauer et al . , 2015 ) . GABA receptors on spine heads are thought to control local synaptic excitation ( Chiu et al . , 2013 ) . However the structural basis for these effects remains unclear . Modeling studies assume that somatic , dendritic shaft and spine inhibition is mediated by pre-synaptic elements of identical size and strength ( Gidon and Segev , 2012 ) . In contrast , excitatory synaptic terminals vary in size and their strength is correlated with terminal size ( Holderith et al . , 2012 ) . We therefore examined this point for cortical inhibition by correlating structural and functional properties of synapses of FS basket cells on layer V ( L5 ) pyramidal cells of rat frontal cortex . Physiological and anatomical data from paired recordings let us simulate the dendro-somatic conduction of the effects of inhibitory synapses made on different membrane sites on pyramidal cells . We show that synapses made by FS basket cells on the soma and on dendritic shafts and spines have dramatically different functional effects . Crossed-corticostriatal ( CCS ) ‘slender untufted’ pyramidal cells ( Larkman and Mason , 1990 Morishima and Kawaguchi , 2006 ) are a discrete neuronal population in L5 . We investigated connections between FS basket cells and CCS pyramidal cells , identified by injecting a fluorescent retrograde tracer into the contralateral striatum ( Figure 1—figure supplement 1 ) . IPSCs were evoked in postsynaptic CCS pyramidal cell soma by single APs in FS basket cells ( Figure 1—figure supplement 2 ) . With pyramidal cell membrane potential maintained at −65 mV , IPSCs reversed on average at −52 . 5 mV ( Figure 1—figure supplement 2C ) , providing a mean driving force of 12 . 5 mV . After recording and biocytin-filling , axonal and dendritic morphology and the number and distribution of possible synaptic contacts from each coupled pair were analyzed ( n = 10 ) using Neurolucida software ( Figure 1B–E , G–I , Figure 2A–D , F–I ) . Paired recordings were made from neighboring cells ( Table 1 , inter-somatic distance: 44 . 5 ± 23 . 7 µm , 20 . 6–66 . 6 µm , n = 10 ) . There was typically a large overlap of the basal dendrites of postsynaptic pyramidal cells and the axonal arbor of presynaptic FS basket cells ( Figure 1B , G , Figure 2B , G , Figure 2—figure supplement 1 ) . In three cell pairs , FS basket cell axons established putative synaptic contacts on the soma and dendrites of a postsynaptic CCS pyramidal cell ( Figure 1J , upper three lines ) . In seven pairs , synaptic contacts were located exclusively on dendrites at various distances from the soma ( Figure 1J , lower 7 lines ) . The number of putative synaptic contacts was 5–14 ( 8 . 2 ± 4 . 8 , 10 pairs ) . Most light microscopic contacts were made where FS basket cell axons crossed basal pyramidal cell dendrites ( Figure 1D , E , H , I , Figure 2D , I , Figure 3B ) ( Marlin and Carter , 2014 ) . The distance from the soma to dendritic contacts was 5 . 8–208 . 4 µm with a mean value of 82 . 5 ± 50 . 0 µm . Peak IPSC amplitude was larger in pairs with putative somatic contacts than those when contacts were exclusively dendritic ( Figure 1J ) . Transmission never failed for pairs with somatic contacts but failures occurred with dendritic contacts ( Table 1 ) . Mean IPSC amplitude , from pairs with only dendritic contacts , was reduced at increasing distances from the soma to the nearest contact ( Figure 1J ) . IPSCs were not detected in two pairs , where light microscopy ( LM ) suggested 7 and 9 contacts were made at distances further than 33 µm from the soma ( Figure 1J , lower 2 lines ) . In each case the pyramidal cell elicited large EPSC in the interneurons ( Table 1 ) . 10 . 7554/eLife . 07919 . 003Figure 1 . Paired recording between FS basket cells and CCS pyramidal cells in L5 . ( A–E ) Structural and functional characteristics of pair CS28 . ( A ) The presynaptic FS basket cell shows a fast-spiking ( upper left ) and the postsynaptic pyramidal cell displayed a regular spiking behavior ( bottom left ) . Average IPSC response in the pyramidal cell ( bottom right ) to a single action potential ( AP ) elicited in the FS basket cell ( upper right ) . ( B ) Reconstruction of the neuron pair . The somatodendritic domain of the presynaptic FS basket cell is shown in blue , the axonal arborization in sky blue , and the somatodendritic domain of the postsynaptic pyramidal cell in gray . ( C ) Illustration showing the number and distribution of putative synaptic contacts ( red dots ) established by the FS basket cell axonal collaterals on the soma and proximal dendritic segments of the postsynaptic pyramidal cell . ( D ) LMg of the pyramidal cell soma with its inhibitory synaptic contacts ( arrows ) illustrated in ( C ) . ( E ) Dendrogram of the pyramidal cell basal dendrites with putative contact sites ( red bars ) . ( F–I ) Structural and functional characteristics of pair CS45 . ( F ) Averaged IPSC in the pyramidal cell in response to a single AP in the presynaptic FS basket cell . ( G ) Reconstruction of the cell pair . Same color code as in ( B ) with putative synaptic contacts ( red ) . Note that synaptic contacts were exclusively found on dendrites . ( H ) Low power LMg of the cell pair showing a putative contact site on the basal dendrite of the pyramidal cell ( red arrow ) by the FS basket cell axon at low ( left panel ) and high ( right panel ) magnification . ( I ) Dendrogram of the basal dendrites of the pyramidal cell with ten LM-identified contact sites ( red bars ) . ( J ) Summary diagram showing the number and distribution of putative contacts established on postsynaptic pyramidal cell somata and dendrites for all investigated pairs . The corresponding averaged IPSC peak amplitude is shown on the right . For the last two pairs , no IPSCs were detectable despite the presence of LM-identified contact sites . DOI: http://dx . doi . org/10 . 7554/eLife . 07919 . 00310 . 7554/eLife . 07919 . 004Figure 1—figure supplement 1 . The CCS pyramidal cell in layer V identified by retrograde fluorescent tracer . ( A ) Cholera toxin subunit B ( CTB ) conjugated with Alexa-555 fluorescent tracer was injected into contra-lateral striatum . ( B ) Subpopulation of layer V pyramidal cells were labeled with the fluorescent tracer in ipsilateral cortex . ( C ) Enlarged image of dotted square in ( B ) . Many labeled layer V CCS pyramidal cells are seen . DOI: http://dx . doi . org/10 . 7554/eLife . 07919 . 00410 . 7554/eLife . 07919 . 005Figure 1—figure supplement 2 . Physiological properties of IPSCs evoked in CCS pyramidal cells in L5 . ( A ) Presynaptic FS basket cell AP ( upper trace ) and the evoked IPSC in the postsynaptic pyramidal cell ( bottom trace ) . Fitting lines are shown for rise ( red ) and decay phase ( blue ) of the IPSC . ( B ) Bar histogram of the distribution of IPSC amplitudes and noise . ( C ) IPSC traces at several somatic holding membrane potential ( upper traces ) and corresponding diagram showing the relationship of the IPSC amplitude to membrane potential ( bottom ) . The reversal potential is −52 . 5 mV determined by a linear fit of the plot . ( D ) Bar histograms of the latency , ( E ) rise time , ( F ) decay time constant , and ( G ) success rates of IPSCs in the postsynaptic pyramidal cell . DOI: http://dx . doi . org/10 . 7554/eLife . 07919 . 00510 . 7554/eLife . 07919 . 006Figure 2 . Different unitary IPSCs induced by single FS basket cells in L5 CCS pyramidal cells . ( A ) Pre-synaptic basket cell and post-synaptic pyramidal cell . Light micrograph ( LMg ) of the CS56 pair . ( B ) Reconstruction of pyramidal cell soma-dendrites ( blue ) and axon ( sky blue ) , basket cell soma-dendrites ( red ) and axon ( pink ) . ( C ) Close-up of the pyramidal cell soma . Scale , 10 µm . ( D ) Putative synaptic contacts ( blue bars ) shown on dendrogram including basal ( gray ) and apical ( sky blue ) dendrites . ( E ) Maximum ( upper ) and averaged ( lower ) IPSCs evoked by single FS basket cell spikes . ( F ) Pre-synaptic basket cell and post-synaptic pyramidal cell . LMg of CS55 pair . Scale is as in A . ( G ) Reconstruction . Scale is as in B . ( H ) Close-up of the pyramidal cell soma . ( I ) Dendrogram with putative synaptic contact sites ( blue bar ) . ( J ) Maximum ( upper ) and average ( lower ) IPSCs evoked by single FS basket cell APs . Scale is as in E . DOI: http://dx . doi . org/10 . 7554/eLife . 07919 . 00610 . 7554/eLife . 07919 . 007Figure 2—figure supplement 1 . Drawings of the paired recording between FS basket cells and CCS pyramidal cells in L5 . Postsynaptic pyramidal cell soma-dendrites ( blue ) and axon ( sky blue ) , presynaptic FS basket cell soma-dendrites ( red ) and axon ( pink ) are shown . DOI: http://dx . doi . org/10 . 7554/eLife . 07919 . 00710 . 7554/eLife . 07919 . 008Figure 2—figure supplement 2 . Sholl analysis of presynaptic FS basket cell axon to postsynaptic CCS pyramidal cell soma center . ( A ) Sholl analysis showing entire FS basket cell axon fiber arborization . ( B ) The initial part up to 20 µm from soma is enlarged to see the difference in terms of the intersections between the cell pairs . DOI: http://dx . doi . org/10 . 7554/eLife . 07919 . 00810 . 7554/eLife . 07919 . 009Table 1 . Synapse properties of pair recordingsDOI: http://dx . doi . org/10 . 7554/eLife . 07919 . 009Amplitude ( pA ) Success rateNeurolucida analysisDistance from soma ( µm ) meansdmaxIPSCCS4−5 . 75 . 1−19 . 50 . 6CS8−8 . 64 . 0−17 . 60 . 5yes48 . 8CS20−10 . 95 . 3−27 . 40 . 9yes51 . 8CS21−7 . 63 . 2−17 . 80 . 6CS22−8 . 03 . 4−14 . 50 . 6CS28−76 . 920 . 9−107 . 31 . 0yes48 . 8CS36−6 . 52 . 6−12 . 50 . 4CS41−8 . 64 . 3−20 . 80 . 7yes41 . 3CS44−6 . 22 . 1−12 . 60 . 5yes66 . 6CS45−7 . 14 . 1−21 . 20 . 7yes53 . 2CS55−91 . 311 . 2−111 . 01 . 0yes35 . 8CS56−17 . 33 . 0−24 . 91 . 0yes20 . 6CS61−9 . 64 . 6−22 . 20 . 8CS62−36 . 414 . 0−69 . 51 . 0EPSCCS1067 . 522 . 2109 . 31 . 0yes26 . 5CS2118 . 68 . 744 . 50 . 9CS2270 . 938 . 3201 . 61 . 0CS2345 . 314 . 283 . 41 . 0yes51 . 2CS364 . 41 . 06 . 50 . 5CS4543 . 119 . 086 . 11 . 0yes53 . 2Reciprocal connection between FS and pyramidal cell was observed . 10 . 7554/eLife . 07919 . 010Figure 3 . 3D reconstruction from serial EMgs . ( A ) Neurolucida reconstruction of the postsynaptic pyramidal cell of the CS56 pair . A dendritic segment ( C1 ) is given in red and marked by red arrow . ( B ) Corresponding LMg of the dendritic segment C1 ( focus stack image ) . The FS basket cell axon terminal is indicated by arrow . ( C ) EMgs from three adjacent ultrathin sections of segment C1 . ( D ) 3D reconstruction of the dendritic segment C1 . The FS basket cell axon ( red ) did not establish a synaptic contact with the dendritic segment C1 ( red arrow ) . Scale bar in ( B ) is the same for ( D ) . DOI: http://dx . doi . org/10 . 7554/eLife . 07919 . 01010 . 7554/eLife . 07919 . 011Figure 3—figure supplement 1 . Focus step images for C1 dendritic segment with FS cell axonal fiber contact site shown in Figure 2B . ( A–F ) The dendritic segment images of every 0 . 5 µm focus step were captured . The focus stacking LMg of Figure 2B was composed of focused area of these images . DOI: http://dx . doi . org/10 . 7554/eLife . 07919 . 011 We found large differences in IPSC amplitude evoked by FS cells in L5 pyramidal cells ( Figure 1A , F , J , Figure 2E , J ) . Large IPSCs were found in two pairs with somatic synaptic contacts . The size of IPSCs in the other pair with somatic/dendritic contacts was smaller ( Figure 1J ) . Higher numbers of putative somatic terminals were correlated with larger synaptic events ( Figure 2C , D , H , I ) . Thus the number of intersections of the presynaptic FS cell axon fibers within 18 µm from somatic center were larger in the pair CS55 with an IPSC of amplitude −91 . 3 pA than in pair CS56 where IPSC amplitude was −17 . 3 pA ( Figure 2—figure supplement 2 ) . The number of synaptic terminals was verified and their size was measured using electron microscopy ( EM ) . Junctional size governs transmitter release probability ( Holderith et al . , 2012 ) and docking sites ( Pulido et al . , 2015 ) , with the number of postsynaptic receptors ( Nusser et al . , 1997; Tanaka et al . , 2005 ) which determines synaptic current amplitude . All putative synaptic contacts ( Figure 2D , I ) were completely reconstructed from serial EMgs ( Figure 3 , Figure 3—figure supplement 1 ) for measurement of synaptic junction and dendritic cross sectional areas . Similar data from sixty one dendritic segments ( mean length 16 . 8 ± 6 . 8 µm ) of the CS56 postsynaptic pyramidal cell and the entire soma of the pyramidal cell ( Figure 4 ) was also used in neuron simulations ( Kubota et al . , 2011a ) . 10 . 7554/eLife . 07919 . 016Figure 4 . Dendritic segments and the somatic region selected for further quantitative EM analysis . Dendrogram of the apical ( left ) and basal ( right ) dendrites of the postsynaptic pyramidal cell of pair CS56 . Dendritic segments indicated by red circles and numbers and the somatic region ( inset grey drawing ) were selected and analyzed in serial ultrathin sections at the EM level . In this pair seven synaptic contact sites were identified at the light microscopic level ( C1–C7 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 07919 . 016 EM analysis let us verify possible synaptic contacts from LM . For the pair CS56 , 3 of 7 possible contacts were verified by EM , but no synaptic contact was made at 4 other potential sites ( Figure 4 ) . One putative LM contact was resolved as three distinct en passant boutons ( S1–S3 in Figure 5A–E ) and another somatic contact was detected only by EM ( S4 , Figure 5—figure supplement 1 ) . The other two verified contacts terminated on spine heads ( Sp2 , Sp3 in Figure 6A , C ) . One with a thin dendrite ( D1 in Figure 5F , G , I , Figure 6A , C ) and nearby spine head ( Sp1 , Figure 5F–I , Figure 6A , C ) were detected only by EM . The junctional area of synapses made by single interneurons varied strikingly with the post-synaptic site that is innervated . For somatic synapses junctional area was 0 . 194–0 . 350 µm2 , it was 0 . 102 µm2 for synapses with dendritic shafts and 0 . 042–0 . 056 µm2 for synapses onto spine heads ( Figure 6F , Table 2 ) . Axonal bouton volume was linearly correlated with synaptic junction area ( Figure 6—figure supplement 1A ) . 10 . 7554/eLife . 07919 . 012Figure 5 . EM identification of synaptic contacts . ( A ) LMg of putative synaptic contacts ( white arrows ) established by a basket cell axon on the soma of a pyramidal cell of CS56 . ( B–D ) EMgs of three somatic synaptic contacts ( S1–S3 ) . Thick arrows indicate synaptic junctions , small arrows the extremities of the synaptic cleft . ( E ) The upper view is a 3D reconstruction of somatic synapses ( red ) on the soma ( green ) in the same plane as in ( A ) , the middle image , rotated by 90° , shows three boutons apposed to the pyramidal cell soma and the lower view shows their synaptic junctions . ( F ) LMg of putative synaptic contacts on a pyramidal cell dendrite . ( G ) EMg of synapses with a dendritic spine ( Sp1 , upper left arrow ) and dendritic shaft ( D1 , bottom right arrow ) 40° tilting angle . ( H ) EMg of the spine synapse in G ( arrow ) . ( I ) 3D reconstructions of the synapses in ( G ) . Lower left image shows the dendritic segment indicated by arrows in ( F ) . Middle view , rotated by ∼60° , shows the junction made with the spine ( red ) . Right image is rotated by ∼ −90° to visualize the junction on the dendrite . ( J ) Focus stack image of LMg of putative contacts ( arrows ) made by basket cell axonal terminals on a pyramidal cell soma and dendrites of CS55 . ( K , L ) EMgs of the S6 ( K ) and S7 ( L ) somatic junctions . ( M , N ) Two views of a 3D reconstruction of a FS cell axon ( red ) and pyramidal cell soma ( green ) showing all contacts . ( at , axon terminal; sp , spine; dend , dendrite ) . DOI: http://dx . doi . org/10 . 7554/eLife . 07919 . 01210 . 7554/eLife . 07919 . 013Figure 5—figure supplement 1 . Somatic synapse contact sites identified using electron microscopic observation . ( A ) Light micrograph showing FS cell axon terminal contacting to the postsynaptic pyramidal cell soma ( arrow ) . The axon terminal looks just a hump of the somatic surface under the light microscope . ( B ) Electron micrograph showing the somatic synapse of the FS cell axon terminal contacting to the pyramidal cell soma ( arrow ) . At: axon terminal . ( C ) 3D reconstruction images of the somatic synapse . Upper image is in the same angle as the light micrograph in ( A ) . Bottom left image is 90° rotated image showing soma surface ( green ) . Bottom right image shows synaptic junction . DOI: http://dx . doi . org/10 . 7554/eLife . 07919 . 01310 . 7554/eLife . 07919 . 014Figure 5—figure supplement 2 . Focus step images for CS55 pair neurons shown in Figure 3J . ( A–Y ) The somatic , dendritic segment and axonal fiber images of every 0 . 5 µm focus step were captured . The focus stacking LMg of Figure 3J was composed of focused area in these images . DOI: http://dx . doi . org/10 . 7554/eLife . 07919 . 01410 . 7554/eLife . 07919 . 015Figure 5—figure supplement 3 . The presynaptic FS basket cell axon terminal crosses the postsynaptic pyramidal cell CS55 dendrite . ( A ) Focus stack image of LMg of putative contacts made by the FS basket cell axonal terminal on the pyramidal cell dendrite ( C9 , arrow ) . This contact was not verified with EM . The other presynaptic fiber is accessing to the pyramidal cell soma ( white arrow ) . The presynaptic axon terminals contact on the other cell soma ( arrow heads ) . ( B ) Focus stack image of LMg of putative contacts made by the FS basket cell axonal terminal on a pyramidal cell dendrite ( C8 , arrow ) . This contact was not verified as synaptic contact with EM . DOI: http://dx . doi . org/10 . 7554/eLife . 07919 . 01510 . 7554/eLife . 07919 . 017Figure 6 . Synapse contact sites identified by EM observation of pairs CS56 and CS55 . ( A , B ) Synaptic contact sites are shown in drawings of CS56 pair neurons ( A ) and CS55 pair neurons ( B ) . Postsynaptic pyramidal cell soma and dendrites are in blue , presynaptic FS basket cell soma and dendrites are in red , and axon in pink . ( C , D ) The synapse contact sites are shown in dendrograms of the basal dendrites of postsynaptic pyramidal cell of CS56 pair ( C ) and CS55 pair ( D ) . ( E ) Distribution of putative synaptic contacts ( black bars ) made by single basket cells on somato-dendritic membrane of 10 pyramidal cells . Contacts confirmed by EM are shown in red . ( F , G ) Area of somatic synaptic junctions is significantly larger than those on dendritic shafts and spines of CS56 pair neurons ( F ) and CS55 pair neurons ( G ) . DOI: http://dx . doi . org/10 . 7554/eLife . 07919 . 01710 . 7554/eLife . 07919 . 018Figure 6—figure supplement 1 . Linear correlation of synapse junction area and bouton volume . ( A , B ) Diagram showing the positive linear correlation of synapse junction area and bouton volume of CS56 ( A ) and CS55 ( B ) . DOI: http://dx . doi . org/10 . 7554/eLife . 07919 . 01810 . 7554/eLife . 07919 . 019Table 2 . Synapse properties of pair CS56DOI: http://dx . doi . org/10 . 7554/eLife . 07919 . 019SynapseTargetjunction area ( µm2 ) Electric charge ( fC ) Conductance ( nS ) †Distance from soma ( µm ) S1Soma0 . 350120 . 10 . 710S2Soma0 . 17459 . 70 . 350S3Soma0 . 19466 . 60 . 390S4Soma0 . 23279 . 60 . 470Sub total0 . 950D1Dendrite0 . 10235 . 20 . 2134Sp1Spine0 . 05619 . 20 . 1134Sp2Spine0 . 05117 . 60 . 1083Sp3Spine0 . 04214 . 40 . 08106Estimated from junctional area . †Estimated from electric charge . Fourteen potential contacts , 3 at somatic and 11 at dendritic sites , were identified by LM for the pair CS55 ( Figure 2I ) . Complete EM reconstruction of the post-synaptic soma let us explore sites obscured in LM where axon crossed the soma ( Figure 5J–N , Figure 5—figure supplement 2 ) and revealed 13 synaptic contacts ( S1–S13 , Figure 5K-N , Figure 6B , E , G ) . Eight terminals made onto dendrites and spine heads less than 33 µm away from the soma presumably contributed to the somatic IPSC ( Figure 5J , Figure 6B , D , E , G ) . Three dendritic shaft synapses ( D5–D7 ) , were located further than 33 µm from soma . Two potential LM contacts showed 2 synaptic contact sites , each . Four potential LM contacts were discounted from EM data ( Figure 5—figure supplement 3 ) , 2 potential LM contacts were not analyzed by EM ( Figure 6E ) , and 4 synapses were only evident in EM . 3D EM reconstructions of all synapses ( CS55 and CS56 ) showed that synaptic area was larger for somatic than dendritic contacts ( Figure 6F , G , Table 3 ) and decreased continuously with distance from the soma . 10 . 7554/eLife . 07919 . 020Table 3 . Synapse properties of pair CS55DOI: http://dx . doi . org/10 . 7554/eLife . 07919 . 020SynapseTargetJunction area ( µm2 ) Electric charge ( fC ) Distance from soma ( µm ) S1Soma0 . 11640 . 90S2Soma0 . 22177 . 60S3Soma0 . 05218 . 40S4Soma0 . 12042 . 30S5Soma0 . 436153 . 00S6Soma0 . 19468 . 20S7Soma0 . 344121 . 00S8Soma0 . 15152 . 90S9Soma0 . 06823 . 80S10Soma0 . 13848 . 30S11Soma0 . 13246 . 20S12Soma0 . 21174 . 10S13Somatic spine0 . 09232 . 30Sub total2 . 274D1Dendrite0 . 04415 . 36D2Dendrite0 . 17661 . 88 . 6Sp1Spine0 . 18063 . 212 . 6D3Dendrite0 . 05820 . 322 . 6Sp2Spine0 . 05419 . 122 . 6D4Dendrite0 . 06021 . 124 . 7Sp3Spine0 . 09934 . 624 . 9Sp4Spine0 . 06723 . 424 . 9Sub total3 . 011D5Dendrite0 . 05519 . 344 . 8D6Dendrite0 . 06021 . 184 . 5D7Dendrite0 . 04616 . 2188 . 5Estimated from junctional area . Numbers of synaptic contacts were defined for two further neuron pairs , CS44 and CS23 , by serial EMgs ( Figure 1J ) . In the CS44 cell pair the closest confirmed synaptic contact was 32 µm distant from the soma , consistent with the inverse relation between synapse distance from the soma and the peak IPSC amplitude ( Figure 6E ) . In pair CS23 , EM verified five dendritic synaptic contacts with the nearest contact site 53 µm from the soma . Physiological analysis revealed the connection was nearly silent ( Figure 6E ) . IPSCs induced by single FS interneurons at dendritic shaft synapses at 32 µm from soma ( CS44 ) were detected with a somatic electrode , but with our recording configuration , IPSCs generated by terminals at 47 µm ( CS10 ) and 53 µm ( CS23 ) from the soma were not detected . Three types of FS basket cell innervation can then be distinguished . Multiple synapses made with the soma or proximal dendrites of L5 CCS pyramidal cell produce large IPSCs , weaker somatic and proximal dendritic innervation produce intermediate IPSCs , while IPSCs are small or absent when synapses terminate exclusively on dendrites . From all paired records , 28 . 4 ± 7 . 6% ( 17 . 2–43 . 1% ) of FS interneuron terminals contacted cell somata ( Figure 5 , Figure 5—figure supplement 3A , Table 4 ) , consistent with previous data ( Karube et al . , 2004 ) . We note that an FS cell that innervates only dendrites of one L5 pyramidal cell , may contact somatic sites of other postsynaptic neurons ( Figure 7 ) . 10 . 7554/eLife . 07919 . 021Table 4 . Proportion of basket terminalDOI: http://dx . doi . org/10 . 7554/eLife . 07919 . 021PairBasket terminalTotal boutonBasket terminal ( % ) CS5510628537 . 2CS28–––CS569121143 . 1CS205221724 . 0CS417324829 . 4CS85920129 . 4CS452615117 . 2CS446723328 . 8CS105922626 . 1CS236331520 . 0Total/average596208728 . 4 ± 7 . 610 . 7554/eLife . 07919 . 022Figure 7 . Schematic summary . Schematic drawing to summarize our main findings . DOI: http://dx . doi . org/10 . 7554/eLife . 07919 . 022 Excitatory synaptic currents are correlated with synaptic size ( Holderith et al . , 2012 ) . At larger synaptic junctions , Ca2+ entry into presynaptic terminals is greater , transmitter release probability is increased ( Holderith et al . , 2012 ) and the number of postsynaptic receptors is larger ( Nusser et al . , 1997 ) . We tested this relation for inhibitory transmission by comparing summed synaptic junction area with maximal IPSC amplitude for pairs CS56 and CS55 . Maximal IPSCs ( Table 5 ) were assumed to occur when all somatic and proximal dendritic terminals ( <33 µm ) ( Figure 5A–E ) released transmitter . The unit electrical charge was calculated as the maximum charge divided by the summed junction area of S1–S4: 326 . 1 fC/0 . 95 µm2 , or , 343 . 3 fC µm−2 for pair CS56 , and S1–S13 , D1–D4 , Sp1–Sp4: 1057 . 8 fC/3 . 011 µm2 , or 351 . 3 fC µm−2 for pair CS55 ( Table 6 ) . This parameter was similar for the two connections , suggesting that currents are well correlated with synaptic junction area . Thus at these inhibitory synapses , conductance can be calculated from junctional area based on the unit IPSC electric charge using morphologically realistic CS56 postsynaptic pyramidal model cell based on our measurement of the cell dimensions ( see ‘Materials and method’ , Table 2 ) . 10 . 7554/eLife . 07919 . 023Table 5 . IPSC properties of pair CS56 and CS55DOI: http://dx . doi . org/10 . 7554/eLife . 07919 . 023CS56CS55ElectricpeakElectricpeakCharge ( fC ) ( pA ) Charge ( fC ) ( pA ) Average193 . 1−17 . 3895 . 2−91 . 3SD56 . 23 . 096 . 211 . 2Max326 . 1−24 . 91057 . 8−111 . 0Min89 . 9−11 . 8766 . 0−74 . 0n60601010Average Trace217 . 5−14 . 2994 . 6−89 . 410 . 7554/eLife . 07919 . 024Table 6 . Unit IPSCDOI: http://dx . doi . org/10 . 7554/eLife . 07919 . 024PairElectric charge ( fC ) Junction area ( µm2 ) Unit IPSC ( fC/µm2 ) CS56326 . 10 . 950343 . 3CS551057 . 83 . 011351 . 3 Inhibitory synaptic connections made by FS basket cell axons terminate on the soma , dendritic shafts or spines of L5 CCS pyramidal cells ( Kubota et al . , 2007 ) . We asked how these differences in synaptic site and junctional size affect function in simulations based on our measurements of synaptic currents and dimensions . IPSC propagation was examined on an electrotonic simulation of the pyramidal cell from pair CS56 . Injecting a 0 . 11 nS current on the spine head of Sp1 ( Table 2 ) resulted in a strong 0 . 78 mV hyperpolarization of the spine , but only 0 . 12 mV was transmitted to the basal dendrite and 0 . 07 mV to the soma ( Figure 8A , C , K ) . The peak synaptic current was 1 . 27 pA at the spine head , and 0 . 81 pA at the soma ( Figure 8B ) . At noise levels of ∼10 pA ( Figure 1—figure supplement 2B ) , a spine-head IPSC would not be detected at the soma . The spine neck effectively isolated the spine head from the dendritic shaft ( neck length , 0 . 5 µm; diameter , 0 . 07 µm; volume , 0 . 043 µm3; resistance , 500 MΩ [Harnett et al . , 2012] ) . Thus spine inhibition did not change nearby dendritic shaft or somatic potential ( Araya et al . , 2006 ) . In contrast , injecting a 0 . 21 nS synaptic current on the dendritic shaft ( D1 ) ( Table 2 ) caused a hyperpolarization of 0 . 23 mV on the shaft and 0 . 13 mV at the soma ( Figure 8D , F , K ) . The spine head Sp1 was hyperpolarized without attenuation ( Harnett et al . , 2012 ) , while the D1 synapse reached only 30% of the Sp1 synapse peak membrane potential . The peak synaptic current was 2 . 45 pA at the spine head , and 1 . 55 pA at the soma ( Figure 8E ) . Injecting a synaptic waveform of 0 . 71 nS at the soma ( S1 ) ( Table 2 ) hyperpolarized that site by 0 . 48 mV ( Figure 8G , H ) resulting in an IPSC of 8 . 29 pA ( Figure 8H ) , in the range of background noise . Simultaneous activation of somatic contacts S1–S4 resulted in a hyperpolarization of 1 . 33 mV , corresponding to a somatic current of 22 . 67 pA , ( Figure 8I , J ) similar to IPSP amplitudes from paired recordings of FS basket cells to hippocampal pyramidal cells ( 0 . 5–3 mV ) ( Buhl et al . , 1994 ) and our own data ( Figure 2E , Table 5 ) . Thus for a similar driving force , proximal inhibitory synapses produce larger somatic hyperpolarizations than distal ones ( Figure 8K ) . 10 . 7554/eLife . 07919 . 025Figure 8 . Simulated conduction for dendritic spine , shaft and somatic IPSCs . ( A–C ) Dendro-somatic conduction of a spine synapse IPSC . ( A ) Peak membrane potential changes ( color-coded as in ( M ) ) over somato-dendritic membrane induced by an IPSC of 0 . 11 nS injected at Sp1 of the model pyramidal cell ( red arrow ) . Peak inhibitory potential of the spine in red . ( B ) IPSC waveform injected at Sp1 spine head is reduced to 64% at the soma . ( C ) Simulated IPSPs . Current flow indicated by arrows . IPSP attenuation was 15% at the basal dendrite and 9% at the soma . ( D–F ) Conduction of a dendritic shaft IPSC , D1 . ( D ) Peak somato-dendritic potential changes induced by an IPSC of amplitude 0 . 21 nS injected at a dendritic shaft ( red arrow ) . ( E ) IPSC waveform injected at D1 ( upper ) and simulated somatic IPSC ( lower trace ) with an attenuation of 63% . ( F ) IPSP wave form . Current flow indicated by arrows . IPSP attenuation at the soma is 57% , but no attenuation at the spine . ( G , H ) Conduction of a somatic IPSC , S1 . ( G ) Peak somato-dendritic potential changes induced by an IPSC of amplitude 0 . 7 nS injected at the S1 somatic site ( red arrow ) . ( H ) IPSC waveform injected at S1 ( upper ) resulting in a somatic IPSP ( lower ) . ( I ) Somato-dendritic conduction of the IPSC resulting from activating ( red arrow ) four somatic synapses S1 , S2 , S3 and S4 . ( J ) Summed IPSC waveform ( upper trace , S1–S4 ) and somatic IPSP ( lower ) . ( K ) Peak somatic IPSPs for eight different injected IPSCs . ( L ) Reduction ( green ) of the EPSP resulting from the injection of an EPSC waveform of 0 . 2 nS ( red ) at the spine head , Sp1 , by an IPSC ( blue ) injected at the same site and time . ( M ) Color-coded dendrogram and corresponding somatic synaptic contacts on the model cell . ( N ) Bar histogram showing the distribution of IPSC electric charge of the pair CS56 . ( O ) IPSC variance of the pair CS56 . ( P ) Bar histogram of the distribution of IPSC electric charge when simulated . Here , the IPSC electric charge also substantially varied from trial to trial and is not significantly different as in the paired recording ( Kolmogorov–Smirnov test , p = 0 . 41 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 07919 . 02510 . 7554/eLife . 07919 . 026Figure 8—figure supplement 1 . Relationship showing synapse conductance and release probability used for simulation analysis in Figure 6P . DOI: http://dx . doi . org/10 . 7554/eLife . 07919 . 026 Spines innervated by inhibitory synapses are typically excited by thalamic inputs ( Kubota et al . , 2007 ) . We modeled the Sp1 spine to ask how spine-head IPSCs affect these excitatory thalamic signals ( Gulledge et al . , 2012 ) . Excitatory synaptic events ( 0 . 2 nS ) were greatly reduced by a coincident spine-head IPSC ( Figure 8L ) . Excitation of the spine-head site depolarized the pyramidal cell soma by 0 . 12 mV . Simulated release from four somatic inhibitory synaptic sites hyperpolarized the soma by 1 . 33 mV . Thus inhibition from clustered somatic synapses of one FS basket cell effectively suppressed dendro-somatic conduction of inputs from ∼11 excitatory spine synapses . If release probability depends on terminal size ( Holderith et al . , 2012 ) , then GABA may be infrequently liberated from smaller inhibitory terminals made by FS basket cells at dendritic sites . Since inhibitory synapses from a single cell usually contact different , distant dendrites , resulting hyperpolarizations may sum poorly ( Figure 9 ) . Even so , summation of integrated dendritic signals during inhibitory cell firing at frequencies of 40–50 Hz ( Isomura et al . , 2009 ) together with GABAergic shunting effects ( Gidon and Segev , 2012 ) may permit FS cell synapses to suppress excitatory inputs on innervated dendritic branches ( Cossart et al . , 2001 ) . Diffusely located inhibitory terminals on dendritic shafts can therefore effectively control afferent excitatory signals . 10 . 7554/eLife . 07919 . 027Figure 9 . Dendrograms with contact sites of the post synaptic pyramidal cells . Individual dendrograms of all investigated postsynaptic pyramidal cells ( n = 10 ) . Apical dendrograms are shown in blue and basal dendrograms are in gray . DOI: http://dx . doi . org/10 . 7554/eLife . 07919 . 027 Variation in release from single synaptic boutons contributes to event-by-event fluctuations in post-synaptic currents ( Sasaki et al . , 2012 ) . IPSC amplitude varied substantially between trials in all dual recordings ( Figure 8N , O , Table 5 ) . Monte Carlo simulations were made on the model of pair CS56 to ask whether this variability might result from probabilistic IPSC generation at somatic terminals , S1–S4 ( Figure 8P ) . Mean IPSC charge transfer was 193 . 1 fC ± 56 . 2 ( 89 . 9–326 . 1 fC , n = 60 traces; Table 5 ) , with putative electric charge at somatic synapses calculated by multiplying junctional size by unit electrical charge , S1–S4 to give 120 . 1 , 59 . 7 , 66 . 6 and 79 . 6 fC respectively ( Table 2 ) . Release probability ( 0 . 59 ) was obtained by dividing the average electrical charge , 193 . 1 fC , by the maximum charge , 326 . 1 fC ( Table 5 ) . Somatic synapses were activated randomly with release probabilities correlated with junctional area ( S1: 0 . 8 . S2: 0 . 4 , S3: 0 . 45 , S4: 0 . 55 ) ( Figure 8—figure supplement 1 ) ( Holderith et al . , 2012 ) . IPSC charge distributions from paired recordings and simulations were statistically similar ( p = 0 . 41 Kolmogorov–Smirnov , Figure 8N , P ) , suggesting that IPSC amplitude variations result from an independent , stochastic activation of individual somatic and proximal synapses ( Sasaki et al . , 2012 ) . We suggest that FS cell inhibitory synaptic strength is progressively reduced from terminals contacting the soma to dendritic shafts and then spines of target pyramidal cells . We asked whether this represents a general principle for cortical inhibitory connections by comparing synapses made by different classes of cortical interneurons stained using the whole cell recording method ( Figure 10A ) ( Kubota et al . , 2007 ) . 3D reconstruction of serial EMgs let us calculate synaptic junction area and the cross sectional area of postsynaptic dendrite or spine volume , for 305 synapses made by 9 different types of interneuron . The junctional area of somatic inhibitory synapses was 0 . 40 ± 0 . 15 μm2 ( n = 23 ) , for dendritic shaft synapses it was 0 . 19 ± 0 . 12 μm2 ( n = 195 ) and for synapses terminating on spines it was 0 . 09 ± 0 . 05 μm2 ( n = 87 ) . Synaptic junctional area was therefore correlated with the size of the target structure ( Figure 10B–L ) with the possible exception of Martinotti cell terminals ( Figure 10J ) that contact distal pyramidal cell dendrites ( Silberberg and Markram , 2007 ) . Linear relations between synapse junction and post-synaptic target size ( Figure 10B–L ) may provide an effective impedance matching ( Kubota and Kawaguchi , 2000 ) and thus control the inhibitory efficacy at different sites . Thus the variation in effects of FS basket cell synapses targeting different membrane regions on L5 pyramidal cells may reflects a general principle for inhibitory cortical circuits . 10 . 7554/eLife . 07919 . 028Figure 10 . Linear correlation between synapse junction area and postsynaptic target size of non-pyramidal cells . ( A ) Different types of cortical GABAergic non-pyramidal cells . The somatodendritic domain of the neurons is given in black and their axons in red . Abbreviations: LS , late spiking cell; FS , fast spiking cell; BSNP , burst spiking non-pyramidal cell; RSNP , regular spiking non-pyramidal cell; CR , calretinin; CRF , corticotropin releasing factor . ( B , C ) 3D reconstructions of synaptic junctions ( red ) on target structures ( green ) of inhibitory axon terminals by cortical FS basket cell ( B ) and descending basket BSNP-CR cell ( C ) using 3D serial EMgs . The thickness of the target structure ( from left to right ) is positively correlated with the size of the junction area . ( D–L ) Line diagrams correlating synaptic junction area of the non-pyramidal neurons with spine head volume ( left panel ) , dendrite cross sectional area ( middle panel ) and plots with soma ( right panel ) . The synapse junction area on spines and dendrites is linearly correlated with the target size . The somatic synapse is larger when compared with dendritic and spine synapse . DOI: http://dx . doi . org/10 . 7554/eLife . 07919 . 028 These data show that FS basket cells mediate either a global somatic inhibition of variable strength , a local dendritic shaft inhibition or act as a local veto at single spines . These distinct effects depend on differences in junctional size . Local spine or shaft potential changes are small and locally restricted . In contrast , somatic inhibitory currents are large , and summation of events from several somatic terminals produces a global control of pyramidal cell excitation . Somatic junctions have large areas , suggesting high release probability ( Holderith et al . , 2012 ) and typically contact multiple sites ( Buhl et al . , 1994 ) . This enhances the likelihood of simultaneous release as FS cells fire repetitively at 30–50 Hz during motor behaviors in vivo ( Isomura et al . , 2009 ) . Some FS basket cell connections with pyramidal cells involved exclusively dendritic sites while others consisted of both peri-somatic and proximal dendritic contacts . Spines receiving inhibitory synapses are typically large ( Kubota et al . , 2007 ) and their thalamic excitatory inputs presumably express both NMDA and AMPA receptors ( Matsuzaki et al . , 2004; Kubota et al . , 2007 ) . Inhibitory synapses may then efficiently veto these thalamic inputs before activation of NMDA receptors ( Gulledge et al . , 2012 ) so reducing the probability of pyramidal cell firing . In paired recordings IPSCs were detected only for terminals that contacted proximal pyramidal cell dendrites . However , IPSCs initiated on distal dendrites have been recorded at the soma in some studies ( Silberberg and Markram , 2007; Jiang et al . , 2013 ) . Possibly differences in experimental paradigm are responsible . In this work post-synaptic potentials were more hyperpolarized ( −65 mV rather than −55/−57 mV ) and Cl− in the recording pipette was higher ( 19 rather than 10 mM ) than in other studies . Both differences would encourage somatic propagation of IPSPs initiated at distant dendritic sites . In our somatic recordings we did not detect IPSPs generated at synapses more distant than ∼40 µm . Possibly , the Cl− reversal potential was similar to the holding potential resulting in a small or null driving force at these sites . Indeed unperturbed Cl− reversal potentials may be 10–25 mV more hyperpolarized than in invasive whole-cell recordings ( Verheugen et al . , 1999; Bevan et al . , 2000 ) . Further work is needed to define unperturbed Cl− reversal potentials in the dendrites and soma of L5 pyramidal cells . Distinct numbers and sites of synaptic contacts made by FS interneurons with pyramidal cells may be regulated by network function ( Yoshimura et al . , 2005 ) and activity during different states ( Klausberger and Somogyi , 2008; Puig et al . , 2008 ) . The strength of inhibition mediated by hippocampal FS basket cells varies with different target pyramidal cells . Synaptic strength is greater at connections with CA1 pyramidal cells in deep rather than superficial layers of stratum pyramidale ( Lee et al . , 2014 ) and it is genetically coded ( Donato et al . , 2015 ) . The innervation patterns of cortical basket cells appear to be regulated by experience , environment or fear conditioning ( Donato et al . , 2013 ) , according to network properties ( Yoshimura et al . , 2005; Lee et al . , 2014 ) and the activity in specific target cells ( Xue et al . , 2014 ) , and activity level of them may be regulated by learning as well as genetics ( Donato et al . , 2015 ) . In contrast , the efficacy of synapses made by Martinotti cells seems to be independent of target pyramidal cell activity ( Xue et al . , 2014 ) . Thus different cortical interneurons respond in distinct ways to neuronal network state . The size , and thus efficacy , of synaptic terminals made by FS interneurons with the soma , dendritic shafts and spines of target pyramidal cells were measured from 3D EM reconstructions . Other GABAergic interneurons establish domain-specific contacts ( Kawaguchi and Kubota , 1998; Jiang et al . , 2013; Kubota , 2014; Marlin and Carter , 2014 ) . Paired recordings from other cortical interneurons and pyramidal cells followed by complete reconstruction of terminals will be needed to establish rules relating terminal size to efficacy . Nevertheless a somato-dendritic gradient of inhibitory terminal size may be a general principle . Our data suggests that relations between post-synaptic site , terminal properties including junctional area , and GABA release patterns may be maintained for other types of cortical interneurons . Inhibitory synapses terminating on spines form 25–50% of GABAergic contacts with cortical pyramidal cell ( Kubota et al . , 2007; Chen et al . , 2012 ) and so form a major part of inhibitory microcircuits . Spines contacted by an inhibitory synapse are typically co-innervated by an excitatory thalamic input ( Kubota et al . , 2007 ) . Our simulations show single inhibitory synapses can effectively veto synaptic excitation and intercept NMDA current ( Gulledge et al . , 2012; Harnett et al . , 2012; Chiu et al . , 2013 ) at the spine head . They could then prevent summation of thalamic excitatory inputs arriving within about 20 ms ( Marlin and Carter , 2014 ) , as pyramidal cell and FS basket cells are co-activated by thalamo-cortical afferents ( Kimura et al . , 2010 ) . Hence the FS basket cell acts as a feed forward inhibition to thalamic input . Excitatory synapses innervating cortical pyramidal cell spines can be modulated by visual experience ( Chen et al . , 2012 ) or by somatosensory stimulation ( Knott et al . , 2002 ) . The veto by inhibitory synapses terminating on spines may be especially important for such plastic changes ( Chen et al . , 2012 ) . Pyramidal cell dendritic spines are tuned to distinct modalities and spines with similar preferences may not cluster together on the same dendritic branch but averaged across a neuron biased towards the orientation tuning of the cell’s output ( Chen et al . , 2013 ) . Inhibitory synapses on dendritic shafts may then inhibit tuned/untuned excitatory inputs on the same but not different dendritic branches and so efficiently and specifically adjust pyramidal cell activity ( Liu , 2004; Marlin and Carter , 2014 ) . Our data shows dendritic IPSCs may exert strictly local effects . Cl− reversal potential at distal dendrite/spine synapses may normally be close to the local resting membrane potential . However this small driving force would be increased by depolarization due to dendritic EPSPs . IPSPs will then reduce EPSP amplitude at the soma even if they do not propagate somatically . FS cells can thus control excitation of L5 pyramidal cells by a specific , local veto of co-innervated spines , by reducing dendritic propagation of summed EPSPs as well as by a strong , global peri-somatic inhibition . We have estimated a peak amplitude of 5 . 7 ± 3 . 1 pA for EPSCs generated at single synaptic contacts with CCS pyramidal cell proximal dendrites ( Morishima et al . , 2011 ) . Here we found a peak IPSC amplitude of 2 . 4 pA at dendritic shaft synapses . Our simulations suggest that summation of single excitatory and inhibitory synaptic currents may reduce dendritic excitation and suppress calcium entry via NMDA receptors ( Larkum et al . , 2009 ) . GABAA receptor activation will also reduce EPSP amplitude by shunting ( Hao et al . , 2009; Gidon and Segev , 2012 ) . Thus , activation of a single dendritic inhibitory synapse should effectively suppress EPSCs at nearby excitatory synapses . This distal dendritic inhibition is functionally strong ( Cossart et al . , 2001; Gidon and Segev , 2012 ) . Inhibitory synapses on dendrites and spines act to reduce neuronal excitability by blocking local EPSCs and so decrease the amplitude of summed EPSPs . The synchronization of FS basket cell activity via gap junctions ( Gibson et al . , 1999 ) will further counter the summation of afferent EPSPs . It is generally accepted that synaptic contacts detected by LM must be confirmed with EM . We verified 14 synapses of 25 putative dendritic contacts with LM ( 56% ) in this study and 78% in our previous study ( Karube et al . , 2004 ) . In addition , we newly found 6 dendritic/spine synapses with EM ( 30%; 6/20 ) . Care must also be taken with somatic inhibitory terminals which are much smaller than the soma , so that terminals behind or in front of a soma may be impossible to resolve in LM . Indeed , we identified 14 somatic synapses with EM for CS55 and 4 somatic synapses with EM for CS56 , although our estimation of the contacts with LM was three for the CS55 and one for the CS56 pair . Our data shows the importance of EM data for quantitative measurements on the number and size of synaptic junctions . Passive cable properties and voltage-dependent resting conductances affect IPSP amplitude . Since postsynaptic target size is related to input resistance and synaptic junction area to the number of post-synaptic receptors ( Nusser et al . , 1997 ) , alterations in synaptic dimensions may govern the size of GABAergic currents . The dependence of synaptic terminal areas on postsynaptic dendritic cross sectional areas would tend to maintain a constant ratio of synaptic conductance to post-synaptic input resistance . Thus , presynaptic interneuron actions are efficiently regulated to provide an appropriate hyperpolarization of their post-synaptic target ( Kubota and Kawaguchi , 2000 ) . EPSC amplitude is correlated with synaptic junction area , release probability , calcium entry and receptor number ( Holderith et al . , 2012 ) . At inhibitory synapses , currents are also correlated with release probability , docking site number and receptor number ( Nusser et al . , 1997; Pulido et al . , 2015 ) . Synaptic junctional area should then govern IPSC amplitudes . Surprisingly unit IPSCs from recordings in this work were quite similar , suggesting that the inhibitory synaptic current is well correlated with synaptic junction area . Larger synapses may generate larger IPSCs , due to multiple release sites or higher numbers of post-synaptic receptors . The presence of multiple release sites at some synaptic junctions has been shown by anatomy ( Holderith et al . , 2012; Nakamura et al . , 2015 ) or estimated from neurophysiological data ( Nakamura et al . , 2015; Pulido et al . , 2015 ) . Clusters of the Cav2 . 1 Ca-channels in large synaptic junctions have been correlated with estimates of the number of vesicular docking sites . GABA release from multiple sites in a large synapse could saturate post-synaptic receptors and initiate large synaptic currents of similar amplitude , as at single-terminal synaptic connections made by molecular layer interneurons of the cerebellum . In contrast , the IPSCs examined here were mediated by multiple synaptic contacts of FS basket cells on L5 CCS pyramidal cells . IPSC amplitude fluctuations presumably reflected variations and failures in release from different terminals . Axons of cortical non-pyramidal cells project to distinct laminar and columnar zones ( Kubota , 2014 ) , enabling different subtypes of interneurons to form synapses with specific targets . Projecting to a specified zone , an axon could make contacts nonspecifically with any available target neuron ( Fino and Yuste , 2011; Packer and Yuste , 2011; Packer et al . , 2013 ) . Alternatively synaptic contacts may be established preferentially with specific neuronal subtypes or target domains , such as soma , axon or dendrites ( Jiang et al . , 2013 ) . Target preference may depend on an activity dependent control of excitatory and inhibitory synaptic input size in order to maintain E/I balance ( Xue et al . , 2014 ) . Our data show FS basket cells may form synaptic contacts with the perisomatic region of post-synaptic pyramidal cells or with their proximal dendritic shafts and spines . Inhibitory synaptic junctional area was matched to the synaptic site—it was larger at somatic than dendritic sites and larger at synapses made with shafts than at those made with dendritic spines . Molecular cues to recognize a somatic or dendritic innervation site may include chemoattractive and cell adhesion molecules . Such mechanisms are involved in a segregation of dendritic spine inhibitory inputs and distinct sources of afferent excitation . Spines innervated by FS basket cell terminals also receive excitatory synapses from thalamus , but never recurrent cortical pyramidal cell inputs ( Kubota et al . , 2007 ) . Both activity dependent chemoattractant factors ( Yee et al . , 1999 ) and cell adhesion molecules of the protocadherin family ( Meguro et al . , 2015; Yagi , 2015 ) have been linked to this specificity . Functionally it would permit FS cell inhibitory synapses to mediate an efficient and selective veto on excitatory inputs from the thalamus . A recent modeling paper ( Gidon and Segev , 2012 ) enhanced our understanding of dendritic inhibitory operations . It assumed that inhibitory synapses targeting pyramidal cell somata , dendritic shafts and dendritic spines possess a uniform size , and strength . Our data suggests the model could be refined to explore the effects of variation in synaptic size and strength from soma to dendrite spine . Quantitative 3-D EM reconstructions provide an exact basis to assign different weights to inhibitory synapses that contact different sites . This inhibitory synaptic machinery differs from that at excitatory synapses subject to both plasticity ( Matsuzaki et al . , 2004 ) and scaling functions ( Magee , 2000; Katz et al . , 2009 ) . Defects in these microcircuits may contribute to depression and other neuronal diseases ( Sauer et al . , 2015 ) . Our data thus provide novel insights into biophysical design principles for inhibitory synaptic operations in neural microcircuits . Retrograde labeling of CCS cells was performed as described previously ( Morishima and Kawaguchi , 2006 ) . Briefly , young Wistar rats ( between postnatal 19–23 days old; Charles River , Japan ) were anesthetized with ketamine ( 40 mg/kg body weight ) and xylazine ( 4 mg/kg body weight ) . Rats were placed in a stereotaxic frame and the skull on the injection hemisphere was partially removed and the cortex , hippocampus and fimbria caudal to the striatum were suctioned to prevent the spilling of dye into the cortex during injection . Cholera toxin subunit B conjugated with Alexa Fluor 555 ( CTB-555; C34776 , Invitrogen , NY ) was used as the retrograde tracer ( 0 . 2% dissolved in distilled water ) . Injection site was determined by using stereotaxic coordinates ( 0 . 8 mm posterior to bregma , 2 . 5 mm lateral to the midline , depth 4 mm ) and a glass pipette ( tip diameter is around 100 micron ) filled with CTB-555 was inserted to the striatum obliquely . Injection ( 80–100 nl ) was performed using positive pressure from a pneumatic pico-pump ( PV-820 , World Precision Instrument , Sarasota , FL ) . After injection , the aspirated brain space was filled with a gel sponge ( Spongel , Astellas Pharma Inc . , Tokyo , Japan ) immersed with saline and the skin was sutured . Rats recovered from surgery in the animal facility and were used for electrophysiological experiments at 2–3 days after the injection . Rats were deeply anesthetized with isoflurane and were decapitated after the loss of all responses to tactile stimuli , such as pinching legs . Slices of frontal cortex ( 300 μm thick ) were cut in ice-cold artificial cerebrospinal fluid ACSF with a vibratome ( VT1000S , Leica , Germany ) and kept at room temperature in ACSF until recordings . The ACSF consisted of ( in mM ) 124 NaCl , 3 KCl , 2 . 4 CaCl2 , 1 . 2 MgCl2 , 26 NaHCO3 , 1 NaH2PO4 , 20 glucose , 0 . 4 ascorbic acid , 2 pyruvic acid and 4 lactic acid and saturated with 95%O2/5%CO2 . Slices were transferred to a recording chamber and perfused at 1–2 ml/min with ACSF ( 25°C ) . Patch pipettes ( 3–5 MΩ ) were pulled from borosilicate glass and filled with 20 μl of internal solution containing ( in mM ) : 126 K-methylsulfate , 6 KCl , 2 MgCl2 , 0 . 2 EGTA , 4 ATP , 0 . 3 GTP , 10 phosphocreatine , 10 HEPES and 0 . 75% biocytin . The pH of the pipette solution was adjusted to 7 . 3 with KOH and the osmolality was set to 295 mOsm . Potassium-methylsulfate as internal solution provided a physiological space clamp ( Fleidervish and Libman , 2008 ) . Dual patch-clamp whole-cell recordings ( EPC9/dual , HEKA , Germany ) were made in the frontal cortex ( medial agranular and anterior cingulate cortex ) with the use of × 40 water-immersion objective ( Axioskop FS , Carl Zeiss , Germany ) . Series resistance was typically 6–15 MΩ and was not compensated . If it exceeded 20 MΩ , data were discarded . Liquid junction potential was not corrected . The data were recorded at 10 kHz and filtered at 2 kHz . For paired whole-cell recordings , retrogradely labeled pyramidal neurons were selected under fluorescence and differential interference contrast microscope ( DIC ) ( Stuart et al . , 1993 ) . FS basket cells were identified in acute slices by their appearance under DIC microscopy . FS cells were recorded using the above internal solution , while pyramidal cells were recorded using an internal solution with the KCl concentration raised to 15 mM and K-methylsulfate lowered to 117 mM to depolarize the reversal potential of Cl− ( −52 . 5 mV ) . IPSCs were recorded as inward currents at −65 mV holding potential . APs were initiated in the presynaptic neuron by 1 ms depolarizing pulses of 300 pA . Presynaptic APs and postsynaptic currents were recorded simultaneously . Recorded presynaptic potentials and postsynaptic IPSCs were analyzed off-line with IGOR software ( WaveMetrics , Lake Oswego , OR ) . For the calculation of kinetic parameters of postsynaptic currents , traces with spontaneous synaptic currents on the rising or decay phase were omitted . The onset of the postsynaptic current was estimated by fitting the rising phase with a parabola and extrapolating back to the baseline . Postsynaptic current amplitude was measured as the difference between the peak current , measured from a 1 . 5 ms window centered at the peak , and the average baseline current , measured in a 4 ms window preceding the presynaptic AP . The decay time constant was obtained by fitting the decay phase of postsynaptic current with a double exponential equation . Since synaptic responses systematically run-down during the time course of some experiments , the amplitudes of postsynaptic currents were plotted against time and only stable periods were selected for further analysis . On average 100 traces ( range 50–150 ) were analyzed for each experiment . Postsynaptic currents smaller than 2 times the noise level were discarded as failures , and the amplitudes of the remaining postsynaptic currents were analyzed . Cumulative histograms of postsynaptic current and noise were constructed and compared with a paired t-test and confirmed the separation between two ( Figure 1—figure supplement 2B ) . To average postsynaptic currents , the peaks of the postsynaptic currents were aligned . The electric charge of IPSC was analyzed using AxoGraph ( Molecular Devices , Sunnyvale , CA ) . Values are reported as mean ± standard deviation . After electrophysiological recordings , slices were immersion-fixed ( 1 . 25% glutaraldehyde , 4% paraformaldehyde , 0 . 2% picric acid in 0 . 1 M phosphate buffer ) and irradiated for 10 s using a microwave , and kept at room temperature for 2 hr . Slices were then cryoprotected with sucrose containing 0 . 1 M phosphate buffer ( 15% followed by 30% of sucrose solution ) and freeze-thawed in the liquid nitrogen . Slices were re-sliced at 50 µm thickness with the vibratome and reacted with avidin-biotin peroxidase complex solution ( ABC kit , Vector Laboratory , Burlingame , CA ) . Biocytin-filled cells were visualized with 3 , 3′-diaminobenzidine tetrahydrochloride ( 0 . 02% ) , nickel ammonium sulfate ( 0 . 3% ) , and H2O2 ( 0 . 004% ) . Slices were further post-fixed in 1% OsO4 with 7% glucose , dehydrated and embedded in plastic ( Epon 812 resin kit , TAAB , Aldermaston , UK ) between silicone ( Sigma coat , Sigma–Aldrich , St . Louis , MO ) coated glass slide and cover slip . Axons , dendrites , and somata of stained neurons were reconstructed using the Neurolucida software ( MBF Bioscience , Williston , VT ) attached to a NIKON ECLIPSE microscope equipped with a 60× objective lens ( NA 1 . 4 , NIKON , Tokyo , Japan ) . Inter point interval of drawing axons and dendrites was less than 2 micron . No correction was made for tissue shrinkage , which should be about 90% ( Karube et al . , 2004 ) . Putative synaptic contacts were identified and their location was marked on the traces of axons and dendrites . The software Neuroexplorer was used for morphometrical and quantitative analyses of reconstructed cells , including total dendritic length and distances between somata and putative synaptic contacts . The dendritic segment or soma images of every 0 . 5 µm focus step in the same image field were captured using the Neurolucida software ( MBF Bioscience , Williston , VT ) attached to a NIKON ECLIPSE microscope equipped with a 60× objective lens ( NA 1 . 4 , NIKON , Tokyo , Japan ) and CCD camera ( 1392 × 1040 pixels ) . The focus stack image was obtained using ‘auto-blend layers/stack images’ function of Photoshop ( Adobe , San Jose ) , which combine the best focused area of the multiple focus step images , to give a greater depth of field ( http://en . wikipedia . org/wiki/Focus_stacking ) . After reconstruction with Neurolucida , stained neurons were serially sectioned at a thickness of 50 nm with an ultramicrotome ( Reichert Ultracut S , Leica Microsystems , Germany ) . Ultrathin sections were mounted on Formvar-coated single-slot grids . EM images of labeled axon terminals and dendrites were captured with a CCD camera ( XR-41 , Advanced Microscopy Techniques ) in Hitachi H-7000 , and HT-7700 EMs ( Hitachi , Tokyo , Japan ) at magnification ×8 , 000 or ×15 , 000 . Structures of interest were reconstructed and quantified from the serial EM images , with the 3D reconstruction software , Reconstruct ( http://synapses . clm . utexas . edu/tools/index . stm ) ( Fiala , 2005 ) . The synaptic junctions were segmented at a typical cleft structure that was found between presynaptic vesicle aggregations and postsynaptic membrane density . Simulations were made with NEURON ( Hines and Carnevale , 1997 ) . The morphology of the model neuron was reconstructed from the EM imaging data . Pyramidal cell dendrites typically possessed elliptical cross sections , but NEURON is limited to circular morphologies . We circumvented this problem by first modeling the pyramidal neuron with circular dendritic cross sections , preserving the cross sectional area from EM . Then , leak conductance and membrane capacitance densities in each section in the circular model were adjusted to be equivalent to those predicted from EM imaging data . Our pyramidal model incorporates passive leak channels only . The passive leak conductance and membrane capacity before adjustment were 0 . 0001 S/cm2 and 1 µF/cm2 , respectively . The intracellular resistance for somata , basal and apical dendrites was 100 Ωcm , and for the spine head and spine neck 385 Ωcm , respectively . The equilibrium potential of the leak current was set to −65 mV . As above , the passive leak conductance and membrane capacitance density in each section in the NEURON model were modified in order to mimic the elliptical shape ( for further details , refer to our previous paper [Kubota , et al . , 2011a] ) . The relationship between cross sectional area ( S ) , circumference ( L ) and summed length of distal dendrites ( R ) we used here is ( S ) = 0 . 00033258 ( R ) + 0 . 048097 and ( L ) = 0 . 0012661 ( R ) + 1 . 3206 . The membrane potential was set to −65 mV ( Morishima and Kawaguchi , 2006 ) , and the GABAA reversal potential to −77 . 5 mV ( Gulledge and Stuart , 2003 ) to fit our measurements of driving force . The electrical charge of each synaptic contact was calculated by multiplying the synapse junction area by the unit electrical charge; in turn individual synaptic conductance was calculated from the electric charge ( Table 2 ) . The synaptic current was adjusted to the average current of pair CS56 ( Figure 2E , lower panel ) with a double exponential fit . It was injected at sites where the presynaptic FS basket cell axon established synaptic contacts with the pyramidal cell . A kinetic model was used for inhibitory synapses ( Destexhe et al . , 1994 ) . Parameters were estimated by fitting the model to the unitary max IPSC data ( Figure 2E upper panel ) . The estimated duration time , rise time constant , decay time constant and conductance are 2 . 3 ms , 0 . 45 ms , 14 . 17 ms and 1 . 92 nS , respectively . Individual synaptic conductance was estimated as multiplying 1 . 92 nS ( conductance of the unitary max IPSC ) by the ratio of synaptic junctional area of each synapse to the total area of the 4 somatic synaptic junctional area ( 0 . 950 µm2 ) . The values of synaptic conductance corresponding to contact sites , S1 , S2 , S3 , S4 , D1 , SP1 , SP2 , SP3 are given in Table 2 . The release probability for the simulation of IPSC variation was estimated with modified fitting line of Figure 4H in Holderith et al . , 2012 ) , y = 3 . 271 0 . 68 + 0 . 018 . We multiplied slope of the fitted line by 0 . 68 to get the similar release probability with pair cell recording result ( Figure 8—figure supplement 1 ) . Experiments were performed as described for the electrophysiological recording experiments previously ( Kubota et al . , 2007 ) . Briefly , whole-cell access was obtained in neurons using visual DIC optics and a 40x water immersion objective . The pipette solution consisted of ( in mM ) : potassium methylsulfate , 120; KCl , 5 . 0; EGTA , 0 . 5; MgCl2 , 1 . 7; Na2ATP , 4 . 0; NaGTP , 0 . 3; HEPES , 8 . 5; and biocytin , 17 . The recording was usually performed for 10–20 min . After re-slicing at 50 µm thickness , each slice ( a set of 50 µm sections after resectioning ) was further treated by one of the following two procedures . ( A ) Some slices were incubated with avidin-biotin peroxidase complex ( ABC ) solution ( Vector Laboratory , Burlingame , CA ) in Tris–HCl buffered saline ( TBS ) with or without 0 . 04% Triton X-100 ( TX ) , and reacted with 3 , 3-diaminobenzidine tetrahydrochloride ( DAB ) ( 0 . 05% ) and H2O2 ( 0 . 003% ) in 0 . 1 M phosphate buffer ( PB ) . ( B ) Other slices were processed for fluorescence immunohistochemistry to identify neurochemical markers , CRF and calretinin . The slices were incubated with the primary antibodies , CRF developed in rabbit ( 1:1000 , gift by Dr . Wylie Vale , #PBLrC70 ) and calretinin ( 1:1000 , Swant , Bellinzona , Switzerland , #6B3 ) in TBS containing 2% bovine serum albumin , 10% normal goat or horse serum and 0 . 04% TX . The slices were incubated in fluorescent secondary antibodies , followed by incubation with Alexa 350 streptavidin ( 1:200 , Molecular Probes , Eugene , OR , #S-11249 ) in TBS . After examination for fluorescence , the slices were incubated with ABC , and reacted with DAB and H2O2 . Slices were then post-fixed in 1% OsO4 in 0 . 1 M PB , dehydrated and flat embedded on silicon-coated glass slides in plastic ( Epon 812 resin kit , TAAB , Aldermaston , UK ) . Recovered neurons were drawn using a drawing tube , or 3D reconstructed using the Neurolucida software ( MBF Bioscience , Williston , VT ) attached to a NIKON ECLIPSE microscope equipped with a 60× objective lens ( NIKON , Tokyo , Japan ) . After light microscopic reconstruction , stained cells were serially sectioned into 90 nm thickness using an ultramicrotome ( Reichert Ultracut S ) . Ultrathin sections mounted on one-hole grids were stained with lead citrate . Electron micrographs were taken with a Hitachi H-7000 electron microscope ( EM ) , using tilting of up to 60° . EM images of the labeled terminals and associated structures were captured using a CCD camera and reconstructed three-dimensionally ( Visilog; Noesis , France ) . We used Mann Whitney U test ( non-parametric ) to compare the junctional area of somatic and dendritic/spine synapses ( Figure 6F , G ) and Kolmogorov–Smirnov test to compare electric charge distributions from paired recordings experiment and the simulation of Figure 8N , P . The datasets I can provide are Neurolucida reconstructed neuron to the “NeuroMorpho . Org” , http://neuromorpho . org/neuroMorpho/index . jsp ( Kubota , 2015a , 2015b , 2015c , 2015d ) , and authentic model cell for ‘Neuron’ simulator to the ‘ModelDB’ , https://senselab . med . yale . edu/modeldb ( Kubota , 2015e ) .	The brain contains millions of cells called neurons that communicate with one another as part of complex circuits . To send information around these circuits , neurons ‘fire’ electrical signals along their length . These trigger the release of chemicals across a structure—known as the synapse—that forms a connection with a neighboring cell . Different types of neurons affect their neighbors in different ways . For example , signals from a pyramidal cell make it more likely that the next cell in the circuit will fire , whereas a signal sent by an inhibitory interneuron has the opposite effect . Pyramidal cells and interneurons make up the circuits in the brain's outer layer—the cortex . Despite their opposing roles , these cells share the same basic structure . Each consists of a cable-like axon that can efficiently transmit electrical signals , and a cell body that contains the nucleus . The cell body bears numerous short branches called dendrites , which are in turn covered in bump-like protrusions called spines . Synapses typically form between the end of one cell's axon and a dendrite on another cell . However , synapses can also form between the end of an axon and an individual dendritic spine , or the end of an axon and a cell body . Models of inhibitory synapses—connections from interneurons that inhibit pyramidal cells—tend to assume that these three types of connection are equivalent . However , Kubota et al . have now combined electron microscopy with electrode recordings of the activity of pairs of connected cells to show that the size and ability of inhibitory synapses to inhibit signaling varies depending on their location . Specifically , inhibitory synapses that form with the cell bodies of pyramidal cells are larger and inhibit signaling more strongly than those that form with dendrites , which are in turn larger and more inhibitory than those on dendritic spines . Thus , depending on the point at which an interneuron contacts a pyramidal cell , it can inhibit signaling throughout the entire cell body , or only across a dendrite , or even just within a single dendritic spine . Incorporating this information into computer models of the brain will improve how accurately they simulate how the brain works . It will also help when modeling disorders in which inhibitory networks are disrupted , such as schizophrenia and depression .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "neuroscience" ]	2015	Functional effects of distinct innervation styles of pyramidal cells by fast spiking cortical interneurons
Many marine animals , ranging from corals to fishes , synchronise reproduction to lunar cycles . In the annelid Platynereis dumerilii , this timing is orchestrated by an endogenous monthly ( circalunar ) clock entrained by moonlight . Whereas daily ( circadian ) clocks cause extensive transcriptomic and proteomic changes , the quality and quantity of regulations by circalunar clocks have remained largely elusive . By establishing a combined transcriptomic and proteomic profiling approach , we provide first systematic insight into the molecular changes in Platynereis heads between circalunar phases , and across sexual differentiation and maturation . Whereas maturation elicits large transcriptomic and proteomic changes , the circalunar clock exhibits only minor transcriptomic , but strong proteomic regulation . Our study provides a versatile extraction technique and comprehensive resources . It corroborates that circadian and circalunar clock effects are likely distinct and identifies key molecular brain signatures for reproduction , sex and circalunar clock phase . Examples include prepro-whitnin/proctolin and ependymin-related proteins as circalunar clock targets . Appropriate timing is essential for reproductive success , especially in organisms that reproduce via external fertilisation . In order to maximise the chances of successful mating encounters , marine animals often rely on mass spawning events that require synchronisation of behaviour and gonadal maturation , both within individuals and on a population level ( Fischer , 1984; Guest , 2008; Hoeger et al . , 1999; Tessmar-Raible et al . , 2011 ) . Many species achieve this synchronisation through the interplay of multiple timing systems that operate on different time scales , ranging from months or days to hours or even minutes ( reviewed in Liedvogel et al . , 2011; Tessmar-Raible et al . , 2011 ) . One timing cue used for synchronised reproduction in organisms ranging from brown algae and corals to bristle worms , echinoderms and vertebrates , is provided by the changing phases of the moon ( Brady et al . , 2016; Coppard and Campbell , 2005; Grant et al . , 2009; Kaniewska et al . , 2015; Kennedy and Pearse , 1975; Oldach et al . , 2017; Rahman et al . , 2004; Raible et al . , 2017; Saavedra and Pousão-Ferreira , 2006; Saigusa , 1988; Skov et al . , 2005 ) . Depending on the species , the predominant physical cue interpreted by organisms is either gravitation or illumination . A study in reefs at two distinct lunar time points ( full moon vs . new moon ) has revealed a direct impact of moonlight illumination on the transcriptome of Acropora corals , but also found that light-induced transcriptome changes are most dramatic in mature animals , when light orchestrates the actual spawning event ( Kaniewska et al . , 2015 ) . While pointing at the transcriptome as a readout for organismal changes , this study therefore also illustrates that in natural samples , the molecular changes relating to illumination can be tightly interwoven with internal changes ( maturation and spawning ) , and might in addition mask endogenous monthly oscillations as they can be produced by endogenous timing systems . Laboratory research on animals such as the marine annelid Platynereis dumerilii or the marine midge Clunio marinus has indeed revealed evidence for such internal timers , showing that nocturnal light ( in nature provided by the moon ) does not only impact directly on transcription ( Zantke et al . , 2015 ) , but is both necessary and sufficient to entrain endogenous monthly ( circalunar ) oscillators ( Kaiser et al . , 2016; Zantke et al . , 2013 ) , which in turn orchestrate reproduction at distinct times of the lunar month ( Figure 1a ) . In Platynereis dumerilli , sexually mature animals reproduce by engaging in stereotypical swimming behaviours classically referred to as the ‘nuptial dance’ , the exact onset of which is signalled by sex pheromones and results in the coordinated release of gametes by male and female worms ( Beckmann et al . , 1995; Zeeck et al . , 1994; Zeeck et al . , 1998 ) . Sexual dimorphisms exist in the production and response of mature worms to spawning pheromones , as well as on the level of gonadal and parapodial morphology . These differences are not obvious during the early and juvenile developmental stages , but develop within a few weeks to months prior to spawning ( Figure 1a , b ) . The sex ratio of the worms remains constant within laboratory cultures , even under different temperature , feeding or light conditions and different population densities , consistent with genetic mechanisms underlying the sexual dimorphisms ( Beckmann et al . , 1995; Fischer and Dorresteijn , 2004; Schulz et al . , 1989; Zeeck et al . , 1994; Zeeck et al . , 1998 ) . Studies from established molecular model systems have shown that sexual behaviour involves neurons expressing sexually-dimorphic markers , including fruitless and doublesex in Drosophila , and mab genes in Caenorhabditis ( e . g . Kopp , 2012 ) . In Platynereis , differential expression of neuropeptides in whole mature males and females suggests that sexual dimorphism at the level of hormonal regulation coincides with sexual maturation ( Conzelmann et al . , 2013 ) . Given the roles of such neurohormones in other animals , it is likely that they act both locally within the brain , as well as in the periphery of the animals to orchestrate sexual dimorphism of the gonads and behaviour as the animals mature . However , no systematic information is available on sexual dimorphisms in the brain of Platynereis . Importantly , the bristleworm brain also plays a crucial role in circadian and circalunar timing ( Hauenschild , 1956a; Hauenschild , 1956b; Hauenschild , 1959; Hauenschild , 1960; Hauenschild , 1966; Hofmann , 1976; Hofmann and Schiedges , 1984; Schenk et al . , 2016; Zantke et al . , 2013 ) . Its importance in reproductive timing is highlighted by the fact that tails of decapitated animals undergo sexual maturation within two weeks , but are no longer synchronised to the lunar phase ( Hauenschild , 1966; Hauenschild , 1974; Schenk et al . , 2016 ) . The brain also possesses non-visual photoreceptors that likely receive light inputs for circalunar ( and circadian ) clock entrainment ( Arendt et al . , 2004; Tessmar-Raible et al . , 2007; Zantke et al . , 2013 ) . Taken together , these findings argue that the Platynereis brain integrates information on maturation , sexual differentiation , and circalunar phase in order to bring about lunar-synchronised reproduction . We therefore sought to identify the molecular signatures of these three major biological processes using an integrated molecular profiling approach , and to assess whether a circalunar component could be disentangled from the changes brought about by maturation and sexual differentiation . We carried out this analysis on two levels: First , we systematically assessed transcriptome signatures , building on our previous observation that the transcript levels of a subset of core circadian clock genes are influenced by the phase of the circalunar oscillator in Platynereis , a finding that has also been observed in corals and the reef fish Siganus guttatus ( Raible et al . , 2017; Altincicek and Vilcinskas , 2007 references cited therein ) . Second , we extended our analyses to the proteome level , reasoning that a combined quantitative transcriptomics and proteomics approach would allow us to obtain a better understanding of the major molecular changes that occur on different functional levels , and highlight molecules and pathways that are most prominent in the different processes . This systematic approach revealed that maturation induces the largest amount of changes on both transcriptome and proteome levels , whereas the circalunar clock has a much stronger impact on the proteome than the transcriptome . Our analyses identified molecular functions , pathways and candidate genes that are diagnostic for either sex and/or maturation stage , and show that representative genes associated with these processes localise in distinct brain regions . We further identify new factors that are subject to circalunar clock-dependent changes on mRNA and/or protein level , and map their locations in the adult brain . Specifically , we demonstrate that Ependymin-related proteins ( ERPs ) are strongly affected on both transcript and protein levels by maturation and circalunar clock phases . We further show that the neuropeptide Whitnin/Proctolin , which recently has been correlated with the spawning event in the semilunar spawner abalone Haliotis asinina ( York et al . , 2012 ) , is also regulated by the circalunar clock in Platynereis . Taken together , our work provides a comprehensive co-evaluation of transcriptomic and proteomic regulation of stage , sex , and circalunar phase-matched tissue samples , and provides the first systematic insight into the distinct genes and molecular processes influenced by maturation , sexual differentiation and the circalunar clock in adult Platynereis dumerilii . As the current Platynereis dumerilii genome assembly is not yet optimal for unique mapping of short reads for quantitative analyses , we generated a mapping resource based on transcript data . Whereas previous Platynereis transcriptomes were generated from whole animals during various adult life stages ( Conzelmann et al . , 2013 ) , or early developmental stages ( Chou et al . , 2016 ) , we reasoned that this might result in under-representation of head-specific transcripts . Moreover , when using whole animal samples , the increase in germ cells during maturation ( see e . g . Schenk and Hoeger , 2010; Schenk et al . , 2016 and references cited therein ) could obscure low abundance expression , which might be of regulatory importance . Therefore , we generated a head reference transcriptome assembled from RNA-Seq data combined from heads sampled under various circalunar conditions and maturation stages ( see Materials and methods ) . The FRFM condition refers to animals kept in a standard daily 16:8 hr light dark ( LD ) light cycle whose circalunar clock was previously entrained by nocturnal light for at least two months , but are not exposed to nocturnal light at the expected full moon time when sampling was performed . Such ‘free-running’ conditions can thus reveal effects solely caused by the endogenous circalunar clock ( Zantke et al . , 2013 and Figure 1c ) . Our primary assembly ( Pdu_HeadRef_TS_v1 ) was systematically curated to ( i ) remove redundancies , ( ii ) merge scaffold information provided by published Platynereis sequences , ( iii ) remove sequence contaminations ( e . g . prokaryotic ) , and ( iv ) flag additional sequences with high similarity to other species ( see Materials and methods section ) , resulting in our final reference sets ( Pdu_HeadRef_TS_v4 and v5 , see Figure 2 , Figure 2—source data 1–3 and Materials and methods section ) . To identify gene sets that were regulated by either one of the three sampled parameters ( circalunar clock phase , maturation , and sex ) , and assess to which extent these gene sets were distinct , we performed differential expression analyses using DESeq2 ( v1 . 10 . 0 , Love et al . , 2014 ) and EdgeR ( v3 . 12 . 0 , Robinson et al . , 2010 ) ( Figure 3a , b , Figure 3—figure supplements 1–9 , Figure 1—figure supplement 1 , Figure 1—source datas 1 and 2 , Figure 3—source datas 1 and 2 ) . Transcripts passing a Benjamin-Hochberg ( BH ) false discovery rate ( FDR ) of 10% were considered as significant ( FDR , Benjamini and Hochberg , 1995 ) . To compare the results of both methods , rank sum files were generated to combine the results of both DESeq2 and EdgeR differential expression testing , to yield the final lists of differentially expressed transcripts ( DETs ) for each condition ( Figure 3—source datas 3 , 4 , 5 , 9 , 10 and 11 ) . Of the three major comparison groups , that is maturation , sex biased and circalunar , maturation exhibits the largest number of DETs ( Altincicek and Vilcinskas , 2007 ) ; 16 . 52% of all transcripts , Figure 3a ) . Sexual differentiation was characterised by 640 DETs ( 1 . 23% ) , whereas 63 DETs ( 0 . 12% ) appear to distinguish the two sampled phases of the circalunar clock . Of the 640 sex-biased DETs , 149 ( 23 . 28% ) were also regulated by maturation and six ( 0 . 94% ) by the circalunar phase ( Figure 3—source data 3–5 ) ; of the 63 circalunar DETs 34 ( 53 . 97% ) were also regulated by maturation . Four transcripts were regulated in all three parameters ( Figure 3a ) . To further validate our quantitative sequencing results , we selected transcripts from the top 50 ranked DETs for additional qRT-PCR analyses , preferring those that also had evidence on the protein level ( see below ) . These experiments showed that the differences observed in our sequencing approach generally validated well when assessed by qRT-PCR on independent samples ( Figure 4a–d , Figure 4—figure supplement 1 ) . Within these experiments , the tested circalunar candidates appeared more variable ( Figure 4—figure supplement 1 ) . We noted that the majority of circalunar candidate DETs exhibited generally lower overall expression levels ( Figure 4—figure supplement 1 ) and showed additional susceptibility to strain specific polymorphisms or phase differences in their circalunar rhythm profiles , similar to previous findings ( Zantke et al . , 2014 ) . After confirming the overall fidelity of the sequencing experiment , we next investigated whether the specific differentially expressed gene sets fell into different regulatory subsets . For this , we analysed the expression profiles by soft clustering analysis using the Mfuzz R-package ( Futschik and Carlisle , 2005; Kumar and Futschik , 2007 ) . Mfuzz clustering on the mean expression values ( details see Materials and methods ) resulted in five clusters each for maturation- and sex-related DETs , and four for the circalunar DETs ( Figure 3b , gene lists Figure 3—source data 6 ) . The maturation-related DETs showed two clusters with transcripts peaking in the premature ( PM ) stage , two with a peak in the mature ( M ) stage , and one in which the expression values declined from immature ( IM ) over PM to M . The two most prominent clusters , with 2 , 687 DETs and 2 , 310 DETs , respectively , were those in which expression was reduced in mature animals ( clusters 3 and 5 in Figure 3b ) , while the single cluster showing an increase in transcript levels from IM over PM to M contained the least number of transcripts ( 649 DETs , cluster 3 ) . Clustering of the sex-related DETs set yielded four clusters with different profiles . Two of these displayed peaks of transcript levels either in mature females or mature males ( clusters 2 and 3 , respectively ) , whereas the other two clusters harboured transcripts peaking either in PM and M females , or in PM and M males ( clusters 4 and 1 , respectively , see Figure 3b ) . Cluster one was less clear , exhibiting a trough of expression in mature animals of either sex . Due to the comparatively low number of DETs in the circalunar comparison , the patterns of the circalunar-related DET clusters were less clear than for the sex and maturation related DETs . However , transcripts belonging to clusters two and three containing 22 and 18 DET of the 63DET , respectively , clearly exhibited differences between IM_FRFM compared to IM_NM , specifically a peak in IM_FRFM ( cluster three ) and a trough in clusters one and two ( Figure 3b ) . Taken together , these results suggest that the three assayed conditions – maturation , sex , and circalunar clock phase – affected largely different sets of genes and by this delineate specific brain ‘molecular signatures’ . To further analyse if the individual signatures also differed with respect to functional or molecular categories , Gene Ontology annotations were analysed using the GOStats R-package ( Falcon and Gentleman , 2007 ) . These analyses revealed that overall 239 out of a total 4 , 775 GO-terms were significantly enriched in our comparisons ( Figure 3c; Figure 3—figure supplements 1–9 , Figure 3—source data 7–11 ) . When analysing the ten most abundant GO-terms , these confirmed the above notion of distinct signature processes for maturation , sexual differences and circalunar phase . Specifically , in the Molecular Function category , ‘Protein binding’ is the most abundant , significantly-enriched single term for maturation , whereas ‘catalytic activity’ and ‘phosphorous-oxygen lyase activity’ are most enriched for the sex-specific and circalunar expression comparisons , respectively ( Figure 3c ) . The Biological Processes and Cellular Compartment ontologies also showed pronounced differences in significantly over- and under-represented GO-terms ( Figure 3—figure supplements 1–9 , Figure 3—source datas 6 and 7 ) . We found little overlap between all the 239 significantly enriched GO-terms when sorted by our three analysis categories: 131 out of 142 were uniquely present in maturation DETs , 83 of 97 were unique to sex differences and ten of fifteen were specific for the circalunar phase DETs ( see Figure 3—figure supplements 2–9 and Figure 3—source data 11 ) . No GO-terms were shared between maturation and circalunar DETs . Taken together , these results suggest that maturation , sexual differentiation and circalunar timing – while being tightly coordinated processes - can be clearly distinguished on a molecular level and are likely driven by different biological mechanisms . The view that these processes can be well separated in Platynereis is further supported by the exemplary whole mount in situ hybridisations ( ISH ) : Maturation candidates tested by in situ hybridisation generally showed broadly-distributed expression over the whole brain . One prominent region was found around the posterior eye pair , where two kidney shaped regions of expression were visible , and another in the medial brain between the eyes ( Figure 4a’ , b’ , a’’ , b’’ ) . Sex biased transcripts exhibited similarly broad expression domains , with noted dimorphism of expression domains , and/or staining intensity reflective of expression level differences between males and females ( Figure 4c’ , d’ , c’’ , d’’ , d’’’ , d’’’’ ) . While transcriptomic analyses provide broad-scale insights into molecular dynamics that occur on the level of mRNA regulation ( transcription itself and transcript stability regulation ) , most physiological processes are driven by protein function , but as has been pointed out , the knowledge about the proteome changes in lunar spawning cycles is missing ( Zoccola and Tambutté , 2015 ) We therefore complemented our transcriptomic study with a systematic proteomic analysis . This objective posed two challenges: ( i ) whereas general transcript analyses have been previously performed for Platynereis ( Chou et al . , 2016; Conzelmann et al . , 2013 ) , no comparable reference set exists yet at proteome level; ( ii ) we anticipated that maximal comparability between transcriptomic and proteomic data would require a method that was able to isolate both RNA and protein from the same specimen . To minimise sampling bias and to facilitate comparisons between the transcriptome and proteome data , we took advantage of the different chemical properties of ribonucleic acid and proteins , and the fact that most of the protein is expected to elute as the flow-through from RNA extractions that use SiO2-based columns under denaturing conditions . However , in these flow-through fractions , proteins are highly diluted and we found that the high concentrations of guanidin thiocyanate ( GuSCN ) impair downstream analyses ( data not shown ) . Thus , we first established an appropriate method for concentrating the protein fraction . We freeze-dried and then reconstituted the sample in the original volume , followed by protein precipitation . Due to the presence of thiocyanate in the samples acid based precipitations were not feasible . We tested different alcohol-based precipitation protocols ( details see Materials and methods ) , and found that a 5:1 dilution with acetone yielded the best recovery . The resulting pellets were washed thoroughly in ice-cold EtOH to remove any traces of GuSCN which may damage the filters ( FASP protocol , Wiśniewski et al . , 2009 ) . We found no evidence for non-precipitated proteins in the acetone precipitation supernatant ( for further details see Materials and methods ) encouraging us to continue with quantitative proteomic analyses . To enable quantitative proteomic analysis , open reading frames ( ORFs ) > 100 amino acids were predicted for all transcripts in the HeadRef_Tsv4 transcriptome for all six possible reading frames , resulting in 92 , 645 predicted protein sequences ( 45 , 065 transcripts; Pdu_HeadRef_prot_v4 ) . In total we identified 2 , 290 proteins with at least two unique peptides identified per protein from our experimental samples . The number of identified proteins with these stringent criteria matches well with analyses from other invertebrate head/brain proteomes; 2 , 987 proteins were identified in Apis mellifera ( Hernández et al . , 2012 ) and 3 , 004 proteins in the larval brain of Bombyx mori ( Li et al . , 2016 ) , despite the fact that both of these studies employed an unlabelled shotgun proteomics approach and required only one unique peptide match per protein . To further compare to existing datasets , we correlated the normalised log10-transformed transcriptome and proteome expression values to each other . This showed a significant , but relatively low correlation ( cor . test ( ) ) Pearson-correlation coefficient of 0 . 482 ( Figure 5—figure supplement 1 ) . This relatively low correlation between transcript and protein levels is highly consistent with previous transcriptome-proteome comparisons studies . Correlations with very similar R-values were found for the developmental transcriptome/proteome in Drosophila melanogaster ( Casas-Vila et al . , 2017 ) and sexual difference transcriptome/proteome comparisons in C . elegans ( Tops et al . , 2010 ) . A comparison of protein and mRNA expression levels for multiple stages of C . elegans development resulted in even much lower R values ( Grün et al . , 2014 ) . These ( and further ) studies imply that only limited conclusions about the level of protein can be drawn based on its transcript level in an organism , further emphasising the importance of proteomic comparisons . The consistency with other studies provides further evidence that our methodological approach did not negatively impact on the proteome data quality . As final quality estimation , we took advantage of another Platynereis head proteome that was independently generated using a conventional sample processing method , a label-free approach ( see Materials and methods ) and did not include large-scale comparisons between different maturation- or sex-specific stages . In total , 3 , 847 proteins were detectable in this conventionally-processed ( see Supplementary file 1 ) , label-free proteome set , out of which 2 , 105 proteins overlapped with the transcriptome/proteome co-processed proteome . To test for possible large-scale systematic compositional biases caused by the different sample processing , we performed a GO-term analysis with both sets comparing to all possible annotated GO-terms and relative frequencies present in the transcriptome dataset ( taking all three categories - molecular process , cellular compartment and function - into account ) . In the 2 , 290 proteins detected in the transcriptome/proteome co-processed set , 372 significantly over- and 244 under-represented GO-terms were present; whereas in the conventionally-processed set ( 3 , 847 proteins ) 444 G0-terms were over- and 248 under-represented ( see Figure 5—source datas 3 and 4 ) . Overall , the types of over- and under-represented GO-terms were similar and , in the majority , overlapped ( Figure 5—source data 3–5 ) , suggesting that the transcriptome/proteome co-processing of our samples did not induce any significant bias in terms of functional complexity . We next analysed the proteome with respect to the different sampling conditions . To enable robust comparison between all conditions , we defined a subset of ‘quantifiable protein’ comprising only those detected uniquely in at least 2 BR and in 3 out of the 5 TR per BR for all samples , yielding a total of 1 , 064 proteins; a number comparable to studies of the mosquito Aedes aegypti head proteome , applying similar stringency filters ( 1 , 139 proteins , Nunes et al . , 2016 ) . We tested for significantly different protein abundances using ROTS and LIMMA ( Elo et al . , 2008; Elo et al . , 2009; Ritchie et al . , 2015 ) statistics and combined the results from both tests using the same rank sum approach used previously for the transcriptome DET analyses ( Figure 5—source datas 6–8 and 19–21 ) . As for the transcriptome data , the largest proportion of proteins , that is 693 proteins were differentially expressed over the process of maturation , corresponding to 65 . 13% of quantifiable proteins ( Figure 5a ) . Contrary to the transcriptome profiling results , however , sexual differentiation accounted for only 17 differentially expressed proteins ( DEP , 1 . 60% ) . Conversely , circalunar phase affected 261 DEP ( 24 . 53% ) , significantly exceeding the amount observed in the transcriptomic analysis . Of the 17 sex biased proteins , 12 were regulated during maturation , as were 177 of the 261 ( 67 . 82% ) circalunar proteins . Six proteins were affected by all three processes ( Figure 5a ) . We are aware that transcriptomic versus proteomic comparisons are of course different in that RNA sequencing will allow to quantify most transcript types present in a given sample , whereas quantitative proteomics can at present only measure the most abundant proteins present in this sample . This likely explains the difference in the number of sex-specific DETs vs . DEPs , but not the difference observed in the numbers of circalunar DETs vs . DEPs . To further assimilate the proteomic data with the transcriptomic profiles , we performed the same type of soft cluster analysis of the significant DEPs using the Mfuzz algorithm ( Futschik and Carlisle , 2005; Kumar and Futschik , 2007 ) to assess if co-regulated clusters existed for the respective processes . Maturation as well as circalunar DEPs each clustered robustly into five protein expression clusters , whereas the sex-specific DEPs showed only weak associations to three clusters ( Figure 5b , gene lists Figure 5—source data 9 ) . A comparison between protein and RNA expression clusters for the maturation process revealed similar dynamic groups ( Figure 3b vs . 5b ) , however , these were not necessarily populated by corresponding transcripts and proteins , as only 40 . 12% ( 278 out of 693 ) of DETs showed a similar profile of regulation on the protein level . This finding suggests that transcript to protein changes are ‘offset’ in time for the majority of the transcripts . Comparisons between DET and DEP clusters were less robust for the sex-specific and circalunar categories , given that only few DEPs or DETs were identified , respectively . GO enrichment analysis of the DEPs showed that out of a total of 2 , 448 GO terms present in all 2 , 290 detectable proteins , 121 GO terms were enriched . Of these , 75 ( 62 . 5% ) GO terms were found to be enriched in maturation DEPs , 17 in sex biased DEPs , and 46 in circalunar DEPs . Like for the transcriptome , there was little overlap between the GO term enrichments across the three different analysed biological processes ( Figure 5c , Figure 5—figure supplements 2–6 , Figure 5—source data 10–13 ) . In our initial proteomic analysis for differences between NM and FRFM stages we discovered a much higher amount of differentially expressed proteins than we would have expected based on the transcriptome data ( 0 . 12% DET vs . 24 . 53% DEP , Figures 3a and 5a ) . This difference is specific for the circalunar comparison and not present or opposite for the maturation and sex-biased comparisons , respectively ( for further interpretation and discussion- see discussion section below ) . In order to verify that this result was not caused by sampling , extraction or measurement artefacts we repeated the proteomics analysis using independent samples . We limited this analysis to IM stages , which showed a large proportion of differences between circalunar phases and are less affected by sexual differentiation and maturation . This second round ( filtered as above ) led to a set of 1 , 671 quantifiable proteins ( Figure 5—source data 14 ) , out of which 1 , 017 were also present in the initial ( 1 , 064 quantifiable proteins ) dataset , thus 95 . 58% of the proteins initially passing our stringent quality filter criteria again passed these criteria , indicating a high reliability and reproducibility of protein detection for this sample type . Of the 1 , 671 proteins of the new dataset , 173 ( 10 . 35 % , Figure 5—source data 15 ) were significantly differentially expressed . Re-analysing the dataset by using only the 1 , 017 quantifiable proteins shared between the first and second dataset resulted in 64 DEPs ( testing was limited to IM animals , see above ) comparable to the 70 DEPs in IM animals in the first analysis ( see Figure 5—source datas 8 , 16 and 17 ) . To identify commonly regulated proteins in all tested proteome datasets , irrespective of frequently occurring sequence differences ( e . g . due to isoforms or alleles ) , which could obscure identical ID-calling , we ran BLASTx against the Uniprot DB for all 365 circalunar DEPs identified from both proteomic analyses ( i . e . the initial circalunar IM , PM_Female and PM_Male sets and the second IM set ) and grouped these according to their BLASTx annotations . We then manually inspected the DEPs for similar identities . Protein groups with similar identities or individual proteins that were either regulated in at least one dataset of the first proteomic analysis and the second proteomic analysis or in all three conditions ( IM , PM male , PM female ) of the first proteomic analysis ( only one –IM - condition was tested in the second proteomic analysis ) were marked as candidates . This resulted in 27 protein groups ( i . e . identified proteins with related , but not identical ID ) and 29 individual protein IDs , which were repeatedly detected in the different quantitative proteomic comparisons ( Figure 5—source data 18 ) . Finally , we tested if the high numbers of circalunar regulated proteins could be explained by the lower number of proteins compared to the high transcriptome number , which reduces the stringency of the Benjamini Hochberg multiple testing algorithm ( Benjamini and Hochberg , 1995 ) used to calculate the FDR . We tested how many transcripts are significantly regulated , if we focus the analyses on the top 5 , 000 or 1 , 000 expressed transcripts . For the 5 , 000 most highly expressed transcripts we found 20 circalunar DET ( 0 . 40% compared to 0 . 12% in the complete transcriptome set ) , a 3 . 3-fold enrichment; 262 sex different DET ( 5 . 24% compared to 1 . 22% in the complete transcriptome set ) , a 4 . 3-fold enrichment and 2 , 435 maturation regulated DET ( 48 . 70% compared to 16 . 52% in the complete transcriptome set ) , a 3 . 3-fold enrichment . For the , 1000 most highly expressed transcripts we found four circalunar DET ( 0 . 40% compared to 0 . 12% in the complete transcriptome set ) , a 3 . 3-fold enrichment; 52 sex different DET ( 5 . 20% compared to 1 . 22% in the complete transcriptome set ) , a 4 . 3-fold enrichment and 546 maturation regulated DET ( 54 . 60% compared to 16 . 52% in the complete transcriptome set ) , a 3 . 3-fold enrichment . This means that statistical differences ( due to overall lower numbers tested ) can explain the overall increase in the maturation regulated proteins , but cannot explain the about 245-fold increase we observe between circalunar transcriptome and proteome . We next started to characterise several circalunar DEP candidates and provide a showcase of five of them . To gain a better understanding as to where these proteins are synthesised and possibly active , we analysed their expression in adult heads . These expression analyses are based on whole mount in situ hybridisations , which we think is justified to do as we were interested in their place of origin . A larger group that caught our attention consisted of iron binding proteins . We selected two of these for validation: Myohaemerythrin ( myoHr ) , and Somaferritin . MyoHr is a representative of the Haemerythrins , small oxygen-binding non-haem iron proteins present in all phyla except for in Deuterostomes; myoHr in particular appears to be restricted to marine invertebrates ( Coates and Decker , 2017; Costa-Paiva et al . , 2017 ) . Ferritins are intracellular iron storage proteins , which additional to their role iron metabolism have been implicated in oxidative stress resistance and immunology ( Altincicek and Vilcinskas , 2007; Andrews et al . , 1992; Pham and Winzerling , 2010 ) . Another circalunar DEP that caught our attention was annotated as Whitnin precursor . Whitnin is the mollusc orthologue of the insect neuropeptide Proctolin ( Veenstra , 2010 , Figure 7—figure supplement 1 , Figure 7—source datas 1 and 2 ) . Of note , prepro-whitnin was shown to be differentially regulated on transcript level in the nervous system of abalone Haliotis asinina over the course of two weeks . Maximal expression occurred just prior to the semi-circalunar spawning events of this organism ( York et al . , 2012 ) . Whitnin is thus a potential candidate for relaying the circalunar signal to other cells . The last two DEPs we focused on both belong to the large family of Ependymin-related proteins . These are highly sequence-related and could either represent different alleles of the same gene or are the result of a recent gene duplication event ( Figure 6a , Figure 6—figure supplement 1 , Figure 6—source datas 1 and 2 ) . Ependymin-related proteins ( ERPs ) are a group of small extracellular glycoproteins originally identified as one of the most abundant proteins in teleost-fish cerebrospinal fluid ( Shashoua , 1991; Suárez-Castillo and García-Arrarás , 2007 ) . Later , ERPs were discovered in other vertebrates and due to their complete absence from all ecdysozoan genomes and transcriptomes analysed so far , were considered to be vertebrate specific ( Suárez-Castillo and García-Arrarás , 2007; Andrews et al . , 1992 ) references cited therein ) . However , recently ERPs have been identified in an increasing number of non-ecdysozoan invertebrates , including echinoderms , hemichordates , molluscs , annelids and even cnidarians ( Hall et al . , 2017; Suárez-Castillo and García-Arrarás , 2007 ) . A re-occurring feature of ERPs are their multiple group-specific gene duplications ( Hall et al . , 2017; Suárez-Castillo and García-Arrarás , 2007 ) . This can be on a higher level - such as the teleost-specific or echinoderm-specific multiplications , but also on the level of species ( groups ) , as pointed out for the Acanthaster planci species group ( Hall et al . , 2017 ) , or as we also observed in Platynereis dumerilii . We identified 15 ependymin-like sequences in the transcriptome , most of which cluster together in one specific subgroup ( Figure 6a , Figure 6—figure supplement 1 , Figure 6—source datas 1 and 2 ) . Seven Pdu-ERPs were detected in the proteome , all of which were significantly differentially abundant between the two circalunar phases ( Figure 5—source data 18 ) . Of those , two showed correlated changes between transcript and proteome level ( Figure 6b , b’ and Figure 6—figure supplement 2 ) and were subsequently investigated for their expression patterns . Both Pdu-erps exhibited identical patterns , covering sensory appendages of the head; palpae and peristomial cirri , a large expression domain on the ventral side of the head , as well as an anterior and posterior domain in the dorsal brain ( Figure 6c–e , Figure 6—figure supplement 2 ) . The other investigated circalunar candidates exhibited overall different expression domains ( Figures 6c–e and 7 ) : Pre-pro-whitnin is expressed in four main regions . Besides expression in single cells of the peristomial cirri , it is expressed in two areas located in the medial forebrain ( Figure 7a’ , a’’ ) , and one small paired cluster slightly anterior to and below the anterior pair of adult eyes ( Figure 7a’’’ ) . Similarly , myohr was present in the medial brain and to a lesser part around the posterior eyes ( Figure 7b’b’’ ) . Different to whitnin and myohr , ferritin showed a broad expression in the head ( Figure 7c’ , c’’ ) . The fact that the expression domains are already largely divergent among these few investigated circalunar candidates , suggests that the effect of the circalunar clock is not confined to specific regions of the head or brain . In this study , we pioneered a combined proteomic and transcriptomic profiling approach to identify coordinated-yet specific changes associated with maturation , sex and circalunar phase in the marine annelid Platynereis dumerilii . The analyses profited from our newly established and benchmarked protocol for the isolation and subsequent quantitative analysis of RNA and protein from the same tissue sample , which does not compromise the complexity of the detected proteins . We expect this method to be widely applicable and be particularly useful for studies that are restricted by the amount of samples that can be obtained and analysed . Using this protocol for Platynereis dumerilii , we find that ~ 17 percent of all analysed transcripts and ~66% of the analysed proteins undergo regulation during maturation . This indicates that the maturation process is the biggest driving force for changes on both transcriptome and proteome levels as worms approach the single circalunar-synchronised spawning event of their lifecycle . This finding is consistent with the fact that physiology , metabolism , body morphology and behaviour change dramatically during the maturation phase: Animals change from a benthic , solitary , and feeding life style to a sexually mature , anorexic state that prepares them for their ‘nuptial dance’ , during which germ cells are released . While this process is accompanied by major morphological changes in the trunk ( in particular , the generation of germ cells that amass to up to 40% of the body mass , accompanied by a deprivation of up to 60% of longitudinal muscle fibres ) , the head is known to be a major source of factors that orchestrate this transition ( Hauenschild , 1956a; Hauenschild , 1956b; Hauenschild , 1966; Hofmann , 1976; Schenk et al . , 2016 ) . The observed changes in the head transcriptome and proteome strongly support this notion , and the recent identification of a major brain hormone component that regulates maturation onset and progression ( Schenk et al . , 2016 ) provides an interesting entry point to understand how these significant changes are hormonally regulated and conveyed to the entire body . Maturation of Platynereis is also accompanied by a significant increase of overall eye volume ( Fischer and Brökelmann , 1966 ) . In line with the described expansion of the outer segments of the eye photoreceptors – that typically harbour the photopigments – we observed a significant increase in transcript abundance of the Platynereis r-opsin1 gene , a major opsin gene expressed in the adult eye photoreceptors ( Arendt et al . , 2002 ) . Comparing the observed transcriptome changes to the changes found in other species shows variable similarities , depending on brain regions and species . A similar study of the honeybee brain showed that ~40% of the assessed transcripts exhibited changes associated with maturation , the majority of which ( ~61% ) were associated with caste behaviour ( Whitfield et al . , 2003 ) . Other species show less pronounced changes , such as in the human neocortex , in which about 9% of the transcripts were reported to change during postnatal development ( Kang et al . , 2011 ) ; in zebrafish ~20% of the pituitary transcripts were differentially expressed during sexual maturation ( He et al . , 2014 ) , while ~11% of the murine hippocampus transcriptome were differentially expressed during maturation ( Bundy et al . , 2017 ) and the Mediterranean fruit fly Ceratitis capitata exhibited up to 7% of head transcriptome changes during maturation ( Gomulski et al . , 2012 ) . Overall proteome changes in the brain during development/maturation show a distribution range in vertebrates from 2–8% ( Ori et al . , 2015; Walther and Mann , 2011 ) , while 15–17% changed in the larval brain development of Bombyx mori and Apis mellifera , respectively ( Hernández et al . , 2012; Li et al . , 2016 ) . Sexual differences in the brain transcriptome are highly dependent on the species studied . In guppies ~ 13% and in the beetle Callosobruchus maculatus ~17% of the brain`s transcripts have been described to be sex biased ( Sharma et al . , 2014; Immonen et al . , 2017 ) ; while in birds around ~2% of the brain transcripts ( Naurin et al . , 2011 ) and human brain ~1 . 1% of the transcripts show a sex bias ( Kang et al . , 2011 ) . While the observed transcriptome and proteome changes in Platynereis dumerilii heads during development , growth and maturation and brain transcriptome changes between different sexes are thus within the range of expectations , we somewhat unexpectedly found that the circalunar phase has little impact on transcript expression changes , but more prominently affects protein level changes . Specifically , we find that only 0 . 12% of the transcripts we analysed are regulated . This is more than 20-fold less than what has been shown as an impact of the circadian clock on the regulation for brain/head transcriptome in several animals , for example Aedes aegypti ( 6 . 7% , Leming et al . , 2014 ) , Drosophila melanogaster ( 2 . 8% , Hughes et al . , 2012 ) and Mus musculus ( 3–4% , Zhang et al . , 2014 ) . Directly equivalent studies on head/brain proteomes have not been performed , instead specific central nervous system parts of mice and rats have been investigated . For the murine suprachiasmatic nucleus , 13–20% of the identified and quantifiable proteins ( 871 and 2 , 112 , respectively ) exhibited circadian abundance changes ( Chiang et al . , 2014; Deery et al . , 2009 ) , compared to about 5% of transcriptome changes in the same tissue ( Ueda et al . , 2002 ) . Thus , the proteomic changes we observe between two different circalunar time points ( with a similar number of quantifiable proteins ) are comparable to the circadian changes observed in the nervous system of other animals . While it is possible that we underestimate the amount of circalunar transcriptomic changes due to limited sampling resolution ( only sampling two circalunar time points and one circadian time point ) , as well as stringent multiple comparisons statistics ( see tests in result section ) , it is not obvious why such an underestimation effect should be different for the proteome under the same conditions . Alternatively , it is also possible that the level of variability is higher in the transcriptome than in the proteome , which – due to the higher experimental noise - would lead to a lower number of detected transcripts compared to the proteome . However , we do not see evidence for such a scenario in our data . We therefore suggest as the most parsimonious explanation that we underestimate the occurring circalunar changes in both transcriptome and proteome , but that the observed relative difference is representative . This implies that , differing from the known circadian clock studies , the circalunar oscillator has a significantly stronger impact on the proteome than on the transcriptome . Thus , the mechanisms responsible for circalunar reproductive timing possibly operate and might be coordinated more strongly at post transcriptional , translational/post-translational level ( s ) , than on transcript level ( s ) . Platynereis dumerilii were housed in plastic boxes and kept in a 1:1 mixture of Natural Sea water ( NSW ) and Artificial Sea water ( ASW , 30 ‰ Tropic Marine ) . Worms were either grown in open culture rooms , or in isolated shelf systems and exposed to a circadian light regime of 16 hr:8 hr light:dark regime ( Figure 1a ) , following classical culture conditions ( Hauenschild and Fischer , 1969 ) , with a circalunar light regime consisting of nocturnal illumination constituting full moon ( FM ) provided for eight nights , every 21 days ( see below for detailed light conditions ) . To allow for constant availability of mature worms and animals are housed in culture rooms with different circalunar entrainment regimes . Animals from the in-phase ( IP ) rooms receive the artificial nocturnal light stimulus ( ‘moon’ ) coinciding with the time of full moon in nature; while animals from the out-phase ( OP ) room receive the moon stimulus when IP room animals are under new moon conditions . For FRFM sampling , worms were transferred to culture rooms without nocturnal light ( i . e . from IP rooms to OP rooms or vice versa ) prior to the first night of the FM stimulus . Specifics are outlined below and summarised for all samples analysed in Figure 1—source data 1 . Animals were sourced from multiple boxes of VIO1 mix and VIO11 mix substrains , originating from the B32134 strain ( Zantke et al . , 2014 ) . Animals were staged based on the visibility of germ cells , and grouped into immature ( IM , no germ cells visible ) , premature ( PM ) , and epitokous non-swarming mature ( M ) males and females ( see Figure 1b ) . Mature animals were defined as having undergone metamorphosis , exhibiting typical yellow ( females ) or red/white ( male ) colouration , but were not yet actively swimming ( i . e . prior to spawning , Figure 1b ) . Large PM animals were selected for staging 3–5 days prior to sampling . Animals were first anesthetised for 10 min in a 1:1 mixture of 7 . 5% ( w/v ) MgCl2 and ASW . Sex and stage were checked by making a small incision in the trunk wall between two parapodia and gently squeezing out a sample of the coelomic contents . Coelomic cells ( eleocytes and oocytes/spermatogonia ) were collected in Nereis balanced salt solution ( Heacox et al . , 1983 ) in a 96-well plate , kept on ice , and imaged using an Axiovert 200M Microscope ( Carl Zeiss , Germany ) , with a Coolsnap HQ monochrome camera ( Photometrics , Vision Systems GmbH , Austria ) . Staging criteria were devised to obtain male and female animals in equivalent early stages of germ cell maturation . PM males were defined as those with germ cells at late spermatogonia ( Spg ) I stage as defined by Meisel , 1990: characterised by the presence of many large clusters with opaque , smooth , cloud-like appearance , that may also be beginning to dissociate into smaller kidney bean-shaped clusters . PM females were defined as those with a maximum oocyte diameter of 52–75 µm ( Figure 1b ) . Oocyte sizes were measured from images taken of oocytes extracted from individual females , using the straight line/measure tool in ImageJ ( v1 . 5g , http://imagej . nih . gov/ij/ ) . Measurements in pixels or inches were converted to µm according to the appropriate scale conversion for the obtained images . After staging , animals were allowed to recover in NSW/ASW for 24–48 hr . Two days prior to head sampling worms were transferred into sterile filtered NSW containing 0 . 125 mg/mL ampicilin and 0 . 500 mg/mL streptomycin-sulphate; after 24 hr , the antibiotics concentration was reduced to 50% for approximately 16 hr , until animals were sampled the following day . For all analyses , pools of 6 heads per biological replicate were sampled at Zeitgeber time 4 . For sampling , animals were anesthetised with MgCl2 as before and the head was cut directly behind the posterior eyes , and in front of the first parapodial segment ( red dotted line in Figure 1b ) . Cut heads were quickly rinsed in NBSS and snap frozen in liquid N2 in pools of six in 2 mL tubes containing metal beads . Heads of mature animals were frozen in pools of 2 or three to ensure more efficient RNA extraction . Samples for initial transcriptome assembly sequencing analysis were obtained from multiple samplings performed in January - March 2014 . Biological replicate samples for quantitative transcriptomics were obtained from multiple samplings between November 2014 and February 2015 . Samples with best RNA concentrations and quality were selected for sample preparation for RNA-Seq and matched protein fractions were analysed for quantitative proteomics . Light spectra and intensities in the culture room were measured with an ILT950 spectrometer ( International Light Technologies Inc Peabody , USA ) . Data were acquired over 300 ms , and 30 spectra averaged to yield one intensity spectrum , in total five measurement per condition were carried out , and the arithmetic mean of the five spectra was used . Animals grown in an ‘in-phase’ culture room were exposed to daylight stimulus of 0 . 0128 µWm−2 ( 3 . 4566 × 1010 photons s−2m−2 ) and 0 . 0001 µWm−2 ( 4 . 1480 × 108 photons s−2m−2 ) full-moon moonlight stimulus; animals grown in an ‘out-phase’ culture room were exposed to daylight stimulus of 0 . 0085 µWm−2 ( 2 . 3090 × 1010 photons s−2m−2 ) and 0 . 0002 µWm−2 ( 6 . 5192 × 108 photons s−2m−2 ) . Full-moon moonlight stimulus ( Figure 1—figure supplement 2a–d ) . Frozen heads were re-suspended in 350 µL RLT-buffer ( RNeasy-Kit , Qiagen ) and the head capsule was broken for 2 Min at 30 Hz in a Tissue Lyser II ( Retsch GmbH ) . RNA was then extracted following the manufacturer`s protocol with additional on-column DNaseI digest . Finally , the RNA was eluted in 30 µL nuclease-free H2O . RNA concentrations and integrity were assessed using the Agilent 2100 Bioanalyzer ( Agilent , Nanochip , Total RNA protocol ) . To recover proteins from the RNAeasy column flow-through we initially collected all flow-throughs , that is RLT + RW1+RPE . As this combination has a final EtOH content of 45% it readily precipitated proteins when incubated over night at −20°C . To optimise protein recovery we then collected different combinations off column flow-throughs , namely RLT , RLT + RW1 , and RLT + RW1+RPE serving as control . RLT and RLT-RW1 were then precipitated by making them either 50% in acetone , 80% in acetone , or 45% in EtOH to test for the effect of RPE on the precipitation . Analysis of the precipitates by Tricine-SDS-PAGE ( Schägger and von Jagow , 1987 ) showed that a four-fold excess of acetone ( 80% final concentration ) added to RLT + RW1 yielded the most quantitative result . Thus , finally the flow-throughs RLT and RW1 were collected and snap frozen in liquid N2 for protein extraction and subsequent LC-MS/MS analysis . After protein extraction via acetone precipitation ( see description in results ) , we controlled for the presence of non-precipitated proteins in the acetone precipitation supernatant on a monolithic C18 column and measuring the eluate at 214 nm . For this , the supernatant was dried down in vacuo , reconstituted in 0 . 1% trifluoro-acetic acid and concentrated on a C18 SPE column from which it was eluted with 90% acetonitrile , 0 . 1% trifluoro-acetic acid . The UV absorbing compounds eluted in a region typical for peptides and not for proteins . For the multiplexing strategy of RNA-Seq and proteomic samples see Figure 1—figure supplement 2e , f . The total RNA input used for sample preparation was 700 ng-1 , 500ng total RNA per sample . PolyA-enriched mRNA was purified from total RNA input using the Dynabeads mRNA-purification kit ( Invitrogen , #61006 ) according to the manufacturer`s protocol , and eluted in a final volume of 50 µL . Ribosomal RNA ( rRNA ) contamination was assessed by Agilent 2100 Bioanalyzer ( Total RNA protocol ) according to manufacturer`s instructions . Samples with rRNA content >25% were subjected to a second round of mRNA purification and re-evaluated by Bioanalyzer as previously . The total volume of purified mRNA was adjusted to 81 µL with nuclease-free water , 9 µL of fragmentation solution ( RNA fragmentation kit; Ambion , #AM8740 ) was added and samples were kept on ice . Fragmentation was carried out at 75°C for 3 min in a thermoblock , at 3 min samples were immediately cooled on ice and 11 µL of stop solution ( RNA fragmentation kit , Ambion ) was added . Cleanup was performed using the RNeasy Kit and RNA was eluted in 30 µL . Fragment size distribution and concentration was analysed by Bioanalyzer ( mRNA protocol ) . First strand cDNA synthesis was carried out with the SuperScript VILO Kit ( Invitrogen , #11754050 ) in a 50 µL reaction with an additional 7 . 5 µg random hexamers ( Invitrogen , #48190011 ) added to the reaction to increase cDNA yield . Reactions were incubated for 10 Min at 25°C ( priming ) , 60 Min at 42°C ( reverse transcription ) , and 5 Min at 85°C ( termination ) . Reaction clean-up was carried out by gel filtration using Sephadex G50-columns ( Mini Quick Spin DNA Columns; Roche #11814419001 ) according to the manufacturer’s protocol . Second-strand cDNA was synthesised in a final reaction containing the first strand cDNA reaction and final concentrations of: 550 µM of each dATP , cGTP , dCTP ( Thermo Scientific , #R0181 ) , and dUTP ( deoxy-UTP sodium salt; Roche , #11934554001 ) ; 0 . 300 U DNA Polymerase I ( E . coli; Invitrogen , #18010–025 ) , 0 . 073 U E . coli DNA ligase ( Invitrogen , #18052–019 ) , and 0 . 015 U RNase H ( Promega , #M4281 ) ; 20 µL 5X second strand buffer ( Invitrogen , #10812–014 ) , in a final volume of 100 µL . Reactions were incubated for 2 hr at 16°C then purified with by the MiniElute Reaction clean-up kit ( Qiagen , #28004 ) , eluted in 2 × 10 µL volumes in a single reaction tube . Samples were stored at −20°C prior to final library preparation for sequencing . Quality control for concentration , size selection , adapter ligation , pre-amplification PCR and dilution for sequencing was performed according to standard procedures and carried out by the VBCF Next-Generation Sequencing facility services ( VBCF-NGS , Vienna Biocenter Campus , https://www . vbcf . ac . at/facilities/next-generation-sequencing/ ) . Samples for initial transcriptome assembly were sequenced on the HiSeq2000 ( Illumina ) in single lane format ( one sample per lane for each of: IM-NM +FM combined , IM-FRFM , PM_NM_male , PM_NM_female , PM_FRFM_male , PM_FRFM_female , M_NM_male , M_NM_female ) . The IM_NM + FM sample was run in paired-end 100 bp format ( PE100 ) , all other samples were analysed as single-end 100 bp ( SE100 ) . Biological replicate samples for differential expression analysis were run using SE100 sequencing over seven lanes , in a 6 × 5 plex ( Figure 1—figure supplement 2e ) and 1 × 6 plex format on the HiSeq2500 instrument ( Illumina ) . The 6-plex lane was a technical replicate lane and contained a single technical replicate from IM samples of each of the other six 5-plex lanes . Read data was quality checked , de-multiplexed and converted to . bam format by NGS facility bioinformatics staff ( VBCF-NGS ) . Reads for individual sequencing libraries are deposited in the European Nucleotide Archive ( ENA ) under accession number PRJEB27496 ( Figure 2—source data 4 ) . Sequence reads from the eight libraries sequenced on single lanes for deep coverage were used for transcriptome assembly . Paired-end reads from the IM_NM + FM sample were used as a scaffold for the base and combined with the SE100 reads from the other seven libraries for the final assembly . The raw Illumina reads ( totalling 1 , 667 , 584 , 873 reads ) were quality checked using Fastqc . Cutadapt ( v1 . 9 . 1 , Martin , 2011 ) to remove the adapter sequences ( -b option ) of the paired end reads and to quality trim the reads ( -q 20 ) requiring that the remaining sequence was at least 13 bp long ( -m 13 ) . We used ngm-utils interleave ( v0 . 4 . 5 ) to reorder the fastq files and filter out reads not having a paired read anymore . The so processed reads were merged and used as input for the RNA-Seq assembly with Trinity ( v2 . 0 . 2 , Grabherr et al . , 2011 ) . Paired end as well as singleton reads were provided as input to Trinity which was run without taking the strand into account . This initial Pdu_HeadRef_TS_v1 assembly comprised a total of 407 , 172 . Of these multiple copies , only the longest sequence per cluster was selected and this Pdu_HeadRef_TS_v2 assembly was filtered by size to retain transcripts greater than 500 bp in size leading to Pdu_HeadRef_TS_v3 with 64 , 335 sequences . Finally , the completeness of our RNA-Seq assembly was tested using Cegma ( v2 . 5 , Parra et al . , 2007 ) , revealing that 85 . 1% of the genes predicted to be present were present full length , and 93 . 6% were present either full length or partially . The Pdu_HeadRef_TS_v3 assembly was then analysed for eukaryotic contaminations using MUMmer ( v3 . 23 , Delcher et al . , 2003 ) with a customised script ( https://github . com/fritzsedlazeck/sge_mummer ) to check for exact hits to common model organism genomes [human ( GRCh38 ) , mouse ( GRCm38 ) , zebrafish ( Zv9 ) , medaka ( MEDAKA1 ) , Arabidopsis ( TAIR10 ) , Drosophila ( BDGP6 ) , C . elegans ( WBcel235 ) ] , as well as dinoflagellates ( Oxyrrhis ) and algae ( Tetraselmis ) used as Platynereis food . Platynereis sequences which had 90% or higher identity and at least 90% of the RNA-Seq scaffold aligned with sequences from the above mentioned organisms were considered putative contaminations . In order to prevent the exclusion of highly conserved Platynereis orthologues , these identified putative eukaryotic contaminations were not excluded from the list , but marked with a tentative suffix ( ‘_contamination’ ) to enable them to be distinguished , investigated and post-filtered from downstream analyses , if necessary . Further curation was then applied to remove redundancy; and these sequences were assembled together with a previously published Platynereis dumerilii transcriptome ( herein refered to as Conz_TS: Conzelmann et al . , 2013 ) containing ~350 , 000 individual contigs . In addition we checked for the presence of all Platynereis transcripts present in GenBank and added/replaced shorter assembled contigs from our assembly with the NCBI sequences where necessary . These sequences were aligned with MUMmer and replaced the scaffolds in our transcriptome if the alignment showed at least 90% identity , at least 200 bp and more than 70% of the scaffold were aligned to the previously published scaffold and the published sequences was longer then our scaffold ( if the transcript was absent the sequence from NCBI was added; if the contig from our assembly was shorter than the ncbi sequence then the contig was replaced by the ncbi sequence ) . We then self-aligned the so obtained scaffolds using MUMmer and kept only one copy of the sequence if the sequences had over 98% identity and one of the copies is covered at least 99% of the length of the other and was longer , otherwise both sequences were kept . This resulted in Pdu_HeadRef_TS_v4 containing 57 , 869 total sequences . For this assembly we obtained the longest open reading frame ( ORF ) for all six frames ( Pdu_HeadRef_prot_v4 ) by using Transdecoder ( v2 . 0 . 1 , http://transdecoder . sourceforge . net ) and keeping the longest ORF per frame . To identify prokaryotic contaminations we used Kraken ( v0 . 10 . 6 , Wood and Salzberg , 2014 ) , which uses k-mer matching to identify hits between the query sequences ( Pdu_HeadRef_TS_v4 ) with the default database . To avoid falsely marking orthologues genes we only marked scaffolds that aligned to two different genomes . This approach identified 6 , 227 putative contigs of prokaryotic origin . We identified 500 of these to be verified Platynereis genes from GenBank , and thus constituting false positives , leaving a total of 5 , 727 contaminating sequences ( 9 . 9% of the total TS_v4 transcriptome , Figure 2—source data 1 ) . These , together with 83 transcripts spiking in one sample ( IM_NM ) discovered during data processing for differential expression analysis ( see below ) were then removed to generate the final Pdu_HeadRef_TS_v5 assembly containing 52 , 059 transcripts . Base annotation of the TS_v1 assembly was performed using Trinnotate ( v2 . 02 ) to identify blastx , blastp , HMMER , GO terms etc . for each contig ( Figure 2—source data 2 ) . The Pdu_HeadRef_TS_v5 assembly was aligned using blastx against the NCBI_nr_protein database ( v_April_2016 ) and NCBI_ensembl_mouse protein database ( v_April_2016 ) to identify putative protein orthologues with e-value cut-offs at 10−4 . Protein domains , GO term annotations , PANTHER IDs , and pathway analysis etc was additionally analysed and reported for all 57 , 869 transcripts using InterProScan ( v5 , using standard parameters ) . Nucleotide sequences were used as input , split into 18 subsets of ≤3 , 000 transcripts and first translated using EMBOSS getorf ( http://emboss . sourceforge . net/apps/cvs/emboss/apps/getorf . html ) and translated sequences were then subject to InterProScan analysis . The results were compiled into a single file merged in addition with the protein_nr and mouse_nr blastx results ( Figure 2—source data 3 ) . Further individual reciprocal blastx searches against ncbi_nr protein database were performed in addition to phylogenetic analyses to confirm the orthology of specific candidate genes of interest . The curated Pdu_HeadRef_TS_v4 assembly was used to map all sequencing reads ( adapter filtered and quality trimmed ) from the replicate head samples for differential expression analysis using NextGenMap ( v0 . 4 . 13 , Sedlazeck et al . , 2013 ) with default parameters . Subsequently , reads were quality filtered ( -q20 ) and converted to a bam file using samtools . The coverage per scaffold was assessed using bedtools ( v2 . 17 . 0 ) multicov . The obtained Pdu_HeadRef_TS_v4 count data ( generated by HTseq v0 . 6 . 1 ) were processed by first , collapsing the technical replicates ( see above and Figure 1—figure supplement 1e ) by sum , then by filtering out 83 transcripts with extremely uneven counts in IM_NM sample [counts IM_NM . 2>=10 x counts ( IM_NM . 1+IM_NM . 3 ) ] and finally by removing the 5 , 727 contaminations identified by KRAKEN leading to a 52 , 059 transcripts ( equivalent to Pdu_HeadRef_TS_v5 ) containing input table ( S5_Pdu_HeadRef_TS_v5_ReadCounts ) . Then differential expression analysis was performed for 14 pairwise comparisons ( Figure 1—figure supplement 1 , Figure 1—source data 2 ) and the DESeq derived normalised read counts were used to plot gene expression profiles ( Figure 3—source data 2 ) using the R packages DESeq2 ( v1 . 10 . 0 , Love et al . , 2014 ) and EdgeR ( v3 . 12 . 0 , Robinson et al . , 2010 ) . DESeq2 was run with standard parameters and EdgeR was run applying the generalised linear model ( Hughes et al . , 2012 ) algorithm . Output lists were cut off using a Benjamini-Hochberg FDR ( Benjamini and Hochberg , 1995 ) of 0 . 1 ( 10% ) . To filter candidate lists for validation , separate lists of differentially expressed transcripts ( DETs ) from DESeq2 and EdgeR were merged by transcript ID and rankings were applied separately to the DESeq2 and EdgeRGLM-derived lists , where transcripts were assigned a rank from one to n ( where n is the number of transcripts ) , based on ranking by lowest to highest adjusted p-value ( padj/FDR ) . The DESeq2 and EdgeRGLM rank values for each transcript were then added to give a rank sum value , and the table was re-sorted by this rank sum value from lowest to highest . For maturation comparisons , ‘unisex’ candidate lists were further defined as subsets of transcripts predicted as differentially-expressed between stages , for both male and female comparisons ( e . g . if transcript A is up-regulated in IM_NM vs PM_NM_male and up-regulated IM_NM vs PM_NM_female , it is classed as ‘unisex’ up-regulated for IM_NM vs PM_NM ) . A schematic overview of the comparison procedure can be found in Figure 1—figure supplement 1 , and see also Figure 1—source data 2 . Proteins for MS analysis were extracted from the same samples used for RNA sequencing , by collecting the first two flow-through fractions from the SiO2-based RNA extraction columns . The pooled flow-through samples were snap frozen in liquid nitrogen and stored at −80°C until further use . To recover proteins , the flow-through fractions were first lyophilised then re-suspended in H2O to a final volume of 1 . 0 mL and 4 . 0 mL of cold acetone was added and proteins were precipitated over-night at −20°C . The precipitate was then collected by centrifugation for 30 min by centrifugation at 3 , 500xg at 4°C . The resulting pellet and a quarter of the supernatant were transferred in to a fresh tube and the remaining supernatant split into three equal parts which were then centrifuged for 30 Min at 14 , 000xg and 4°C to collect the remaining proteins . Pellets were then washed with 100 µL −20°C-cold EtOH to remove the co-precipitated GuSCN from the RNA-extraction buffer and allowed to dry over-night , dissolved in 100 µL UA-buffer [8M urea , 50 mM triethylammonium bicarbonate ( TEAB ) , pH 8 . 5] by sonification , and finally re-pooled . Proteins were digested for MS/MS-analysis using the Filter Aided Sample Preparation method ( Wiśniewski et al . , 2009 ) . Briefly , the samples were reduced by 10 mM DTT for 30 Min at RT in UA , applied to an YM-30 filter membrane , centrifuged and washed . Thiols were alkylated for 30 Min with 50 mM iodoacetamide at RT in UA centrifuged and washed with 50 mM TEAB . Proteins were then digested with trypsin ( 1:50 trypsin:protein ratio ) overnight at 37°C in 50 mM TEAB . Digested peptides were eluted from filters by centrifugation and acidified by the addition of trifluoroacetic acid , and desalted by solid phase extraction ( SPE ) with Sep-Pak C18 columns ( Waters ) . Eluates were dried in a speed-vac and the resulting pellets were dissolved in 100 µM TEAB and labelled with TMT isobaric tags ( TMTsixplex Isobaric Label Reagent , Thermo Fisher Scientific ) according to manufacturer's protocol . Labelling efficiency was assessed by a LC-MS/MS run on Q Exactive plus Orbitrap mass spectrometer ( Thermo Fisher Scientific ) for each label separately and found to be complete when semi- or unlabelled peptides were less than 1% of total PSMs . Once maximal labelling efficiency was confirmed the reaction was quenched with 8 µL 5% hydroxylamine per reaction vial ( 15 min incubation time at room temperature ) . FASP-digested and labelled samples were then combined in a 5-plex multiplex format for Set1 and a 6-plex format for Set2 . Peptides were separated on an Ultimate 3000 RSLC nano-flow chromatography system using a pre-column for sample loading ( PepMapAcclaim C18 , 2 cm ×0 . 1 mm , 5 μm , ) and a C18 analytical column ( PepMapAcclaim C18 , 50 cm ×0 . 75 mm , 2 μm , all Dionex-Thermo Fisher Scientific ) , applying a linear gradient over for 4 hr from 2% to 35% solvent B ( 80% acetonitrile , 0 . 1% formic acid; solvent A 0 . 1% formic acid ) at a flow rate of 230 nL/min . Eluting peptides were analysed on a Q Exactive Plus Orbitrap mass spectrometer equipped with a Proxeon nanospray source ( all Thermo Fisher Scientific ) , operated in a data-dependent mode . Survey scans were obtained in a mass range of 380–1 , 650 m/z with lock mass off , at a resolution of 70 , 000 at 200 m/z and an AGC target value of 3 × 106 . The 12 most intense ions were selected with an isolation window width of 1 . 2 Da , fragmented in the HCD cell at 35% collision energy and the spectra recorded at a target value of 1 × 105 and a resolution of 17 , 500 . The peptide match and exclude isotope features were enabled and selected precursors were dynamically excluded from repeated sampling for 40 s . Five technical replicate runs were analysed for each set of multiplexed samples . Data analysis was performed with Proteome Discoverer 2 . 1 using Sequest to search against the Platynereis proteome database ( Pdu_HeadRef_prot_v4 , ProteomeXchange: identifier PXD010532 ) . Carbamidomethylation of cysteines , TMT-labelling on peptide N-termini and lysines were selected as fixed modifications , oxidation of methionine as variable modification . Precursor tolerance was set to 10ppm , fragment tolerance to 0 . 02 Da . Data were filtered for an FDR of 1% at PSM and protein level . Quantification was performed via the reporter ions quantifier mode in PD 2 . 1 based on the signal/noise values ( herein referred to as intensity ) . Spectra with isolation interference greater 30% were not used for quantification . Protein intensity values were calculated for each biological replicate as the average intensity of all peptides per protein across all five technical replicates ( herein referred to as Mean Intensities per protein ) . Mean intensities per protein were then normalised between isobaric ion channels to obtain a 1:1:1:1:1 ratio of reporter ion intensities based on the summed reporter intensities across all proteins per channel , within each set of 5 multiplexed samples ( herein these expression values are referred to as Norm . Mean intensities per protein ) . Quantifiable proteins were further defined as those with at least two uniquely mapping peptides and detected in at least 2/3 biological replicates per sample group . To normalise for variation in protein intensity distributions between individual samples , an additional channel median normalisation was applied , whereby the median protein intensity value was calculated for each sample ( ‘channel median value’ ) and all individual protein values per sample were then divided by this channel median value . We performed an additional inter-replicate median normalisation to account for inter-run variation between biological replicate sets within each of the NM and FRFM experiments , by: i ) calculating for each biological replicate set; the median intensity across all five samples per protein , ii ) calculating conversion factors to equalize the median intensity per protein between biological replicate sets . Conversion factors ( CF ) were calculated for the NM set relative to BR1 , for example: CF . BR2 = Median_NM_BR1/Median_NM_BR2; CF . BR3 = Median_NM_BR1/Median_NM_BR3 , and relative to BR3 for the FRFM set , for example: CF . BR1 = Median_FRFM_BR3/Median_FRFM_BR1; CF . BR2 = Median_FRFM_BR3/Median_FRFM_BR3 . All intensity values per protein per biological replicate were then multiplied by the respective conversion factor . To enable normalisation and comparison of the data from the NM and FRFM experiments we included a spiked-control sample from the NM_BR1 set ( TMT-128_PM_NM_female . 1 ) in each of the FRFM biological replicate sets . Therefore , as a final step the FRFM intensity values were normalised relative to those of the NM_BR1 set by multiplying the values for each protein in each BR set , by corresponding conversion factors ( CF ) calculated as the ratio of the NM_BR1 TMT-128_PM_NM_female1 intensity/intensity value recorded for this sample in FRFM BR sets 1 , 2 and 3 ( e . g . CF . FRFM1 = TMT-128_PM_NM_female1:NM/TMT-128_PM_NM_female1:FRFM_BR1; CF . FRFM2 = TMT-128_PM_NM_female1:NM/TMT-128_PM_NM_female1:FRFM_BR2 ) . The effect of normalisation resulted in median-centred data and decreased variability in the distribution of protein intensity values between biological replicates ( Figure 5—figure supplement 7 and Figure 5—source data 1 ) . TMT-based quantitative proteomics data were acquired using the VBCF instrument pool ( www . vbcf . ac . at ) . For benchmarking the protein set obtained with the described method , results were also compared with a set of proteins identified by a more conventional unlabelled proteomic analysis on separate head samples ( Bileck , Gerner et al . , unpublished ) . Briefly , for this analysis , tissue samples were incubated in 100 µl sample buffer ( 7 . 5M urea , 1 . 5M thiourea , 4% CHAPS , 0 . 05% SDS , 100 mM DTT ) for 4 hr and lysed using an ultrasonic probe . An adapted filter aided sample preparation protocol was used for protein digestion as described ( Slany et al . , 2016 ) . Cleaned peptide samples were reconstituted in 5 µl 30% formic acid containing 10 fmol each of 4 synthetic standard peptides and diluted with 40 µl mobile phase A ( 98% H2O , 2% acteonitrile , and 0 . 1% formic acid ) and analysed as described previously ( Slany et al . , 2016 ) . Identification of proteins was performed using the MaxQuant software ( v1 . 5 . 2 . 8 ) including the Andromeda search ( Cox and Mann , 2008 ) to map the identified peptides against the Pdu_HeadRef_prot_v4 database . A minimum of two peptide identifications , at least one of them unique , was required for positive protein identification . For peptides and proteins , a FDR of less than 0 . 01 was applied . Differential abundance analysis for the normalised proteomic data for the subset of 1064 quantifiable proteins was performed using both Linear Models for Microarray and RNA-Seq Data ( Li et al . , 2016 , http://bioinf . wehi . edu . au/limma ) and the Reproducibility-Optimised Test Statistic ( ROTS ) package for R ( Elo et al . , 2008; Elo et al . , 2009 , http://www . btk . fi/research/research-groups/elo/software/rots/ ) . Both ROTS and LIMMA are able to handle missing values , which is a common feature of proteomic datasets . ROTS and LIMMA were used to analyse differential expression using a BH FDR of 0 . 1 ( 10% ) , for the 14 pairwise comparisons ( Figure 1—figure supplement 1 , Figure 1—source data 2 ) defined previously for RNA-Seq differential expression analysis using DESeq2 and EdgeR , details with regard to the statistical testing are given below in the section Statistics . The same rank sum approach was taken to merge the results of LIMMA and ROTS predictions and rank candidates according to adjusted pvalues . Candidates were further classified into unisex groupings for maturation stage comparisons . To test whether the different input list sizes of transcripts and proteins , 52 , 059 and 1 , 064 respectively , might affect the relative amount of DE predictions , we also performed DESeq2 and EdgeR analysis on the 1000 and 5 , 000 most abundant transcripts . For this , we calculated the mean read counts over all comparison per transcript , sorted the transcript list and chose the 1 , 000 or 5 , 000 transcripts with the highest average number of read counts . These reduced lists were then subjected to DE analysis by DESeq2 and EdgeR as before , and rank sum lists of these analyses were then generated as outlined above . qRT-PCR , including primer design , primer testing for efficiency , cDNA synthesis protocol , reaction chemistry , cycle conditions and instrumentation , was performed as described previously ( Schenk et al . , 2016 ) , with the following modifications: Total RNA ( 200 ng per sample ) was reverse transcribed to cDNA using the Quantitect reverse transcription kit ( Qiagen , Cat# 205310 ) , with incubation at 42°C for 20 min for reverse transcription . Final cDNA reactions were diluted to a final volume of either 95 µL for low-expression level and lunar candidate qRT-PCR assays , or 350 µL for sex-biased and maturation candidate qPCR assays and all qRT-PCR reactions were performed in 15 µl final volume , using 3 µL of cDNA template . Primers for qPCR assays ( intron-spanning where possible ) were designed using the Universal Probe Library assay design tool ( Roche , Supplementary file 2 ) . Target genes and reference controls were analysed in duplicate reactions for all samples . Plate control cDNA and –RT controls for the target gene ( s ) were included on each plate as quality controls for inter-plate consistency . Cell cycle serine/threonine kinase ( cdc5 , Dray et al . , 2010; Zantke et al . , 2013 ) and S-adenosyl methionine synthetase ( sams , Schenk et al . , 2016 ) , were measured as reference genes , based on previous evaluation as stable qPCR reference controls ( Schenk et al . , 2016 ) . Expression levels for target genes were calculated using the Δ-Ct method , using the mean of cdc5 and sams Ct values for relative normalisation ( Figure 4—source data 1 ) . Calculated Δ-Ct values were exponentially transformed ( 2- ( Δ- Ct ) ) to give the final normalised relative expression values . Statistical analysis was performed as outlined below . CLC Main workbench ( version seven and higher ) were used for routine sequence analysis , small-scale BLASTx and BLASTp searches against NCBI , assemblies , basic multiple sequence alignments , and sequence annotations . Phylogenetic trees for ependymin-related proteins ( ERPs ) and Whitnin/Proctolin were generated from multiple sequence alignments ( MUSCLE: http://www . drive5 . com/muscle , Edgar , 2004 ) , see Figure 6—source datas 1 and 2; Figure 7—source datas 1 and 2 ) and IQ-TREE ( v1 . 6 . 3 , Minh et al . , 2013; Nguyen et al . , 2015 ) , ultra fast bootstrapping ( Hoang et al . , 2018 ) , and the integrated ModelFinder ( Kalyaanamoorthy et al . , 2017 ) . Simple Modular Architecture Research Tool ( SMART , http://smart . embl-heidelberg . de ) was used for analysis of protein domains in individual sequences . Functional descriptions of individual orthologue proteins were also sourced from the Uniprot database ( http://www . uniprot . org ) . Candidate genes of interest were cloned from cDNA reverse transcribed from total RNA of heads from appropriate sex , stage and circalunar condition , using the Transcriptor high fidelity cDNA synthesis kit ( Roche , #05081955001 ) . Final cDNA reactions were diluted to a final volume of 50 µL with nuclease free water . PCR: 20 µl final reaction volume , 1–5 µl of cDNA template , Phusion polymerase ( Thermo Scientific ) . PCR programs: 98°C denaturing cycles of 30 s , annealing temperatures ranging from 58–62°C with 20 s annealing times and extension at 72°C for 35–40 cycles ( primer details- Supplementary file 2 ) . PCR products were analysed by agarose gel electrophoresis ( GelGreen , Biotium ) , bands of expected size were isolated using the Qiagen gel extraction kit according to the manufacturer’s instructions . Extracted DNA was ligated into the pJET1 . 2 vector DNA ( CloneJET PCR cloning kit , Thermo Scientific ) . Plasmid DNA was purified after colony PCR using a standard alkaline lysis protocol with sequential precipitation using isopropanol and 70% ethanol ( Birnboim and Doly , 1979 ) . DNA pellets were resuspended in 30–50 µL of nuclease free water and verified by Sanger sequencing ( Microsynth , Austria ) . Plasmid cloning info- Supplementary file 2 . Animals for in situ hybridisation ( ISH ) were sourced from culture boxes , sexed and staged according to the aforementioned criteria and maintained in separate boxes in antibiotic-free ASW under respective circalunar conditions for at least 48 hr prior to sampling . ISH for head tissues was carried out as previously described ( Tessmar-Raible et al . , 2005; Zantke et al . , 2013 ) . DNA templates for in vitro transcription were generated by amplifying sense and antisense sequences from vector DNA templates using Sp6-conjugated pJET1 . 2 forward and reverse primer combinations ( see Supplementary file 2 ) . Probes were purified using the RNeasy kit ( Qiagen ) according to manufacturer’s instructions , eluted in 30 µL of RNase-free water , probe quality analysed by agarose gel electrophoresis and diluted to a concentration of 100 ng/µL or 50 ng/µL with hybridisation solution and stored at −20°C . For ISH 500 or 1000 ng of probe were used depending on the expression of the gene as judged from RNA-Seq . Stained tissues were stored in 80–100% glycerol/1X PTW at 4°C in 2 mL tubes . For all genes assessed sense controls were run in parallel in pools of 3–5 . In situs were imaged using an Axioplan Z2 Microscope ( Carl Zeiss , Germany ) , with AxioCam MRc5 colour CCD camera ( Carl Zeiss , Germany ) and captured using ZenPro Software ( v2 . 0 , Carl Zeiss , Germany ) . Images were saved in TIFF format . Where necessary brightness and contrast adjustments and cropping of images in TIFF format was carried out using ImageJ ( v1 . 50i ) or Adobe Photoshop CC ( v2015 ) . Figures were constructed using Adobe Illustrator CC ( v2015 ) . All statistical analyses were performed with R ( R Core Development Team , 2015 ) , and various statistical packages available for R . Analysis of differentially expressed genes was performed with EdgeR ( McCarthy et al . , 2012; Robinson et al . , 2010 ) and DESeq2 ( Love et al . , 2014 ) using custom made scripts , considering a BH FDR of 0 . 1 as statistically significant . Proteomic data were analysed using either LIMMA ( Ritchie et al . , 2015 ) and ROTS ( Elo et al . , 2008; Elo et al . , 2009 ) . LIMMA uses linear modelling to test the fit of obtained expression data to the null hypothesis based on the defined experimental design . The script used was modified based on a previously available application for labelled-proteomics data ( Kammers et al . , 2015 , http://www . biostat . jhsph . edu/~kkammers/software/eupa/R_guide . html ) , and used the more conservative BH FDR rather than q-values ( Storey and Tibshirani , 2003 ) originally in the script . ROTS is a modified form of the t-test , which uses bootstrapping to analyse differential expression between sample groups using random-subsets of samples . From the collective results a modified test statistic is calculated based on the reproducibility of the list of differentially expressed predictions ( Elo et al . , 2008 ) . GO-Term analysis was performed using the GOstats package ( v2 . 46 . 0 ) , which applies a hypergeometric test to test for significance . We used the conditional algorithm of hypergeometric test which takes into account the hierarchical structure of the GO-terms by testing the leaves ( GO-Terms with no child-terms ) first . The conditioning then leads to the elimination of those genes associated with a GO-Term which have been tested as significant in the GO-term`s children ( Falcon and Gentleman , 2007 ) . qRT-PCR data were first checked for normality by the Shapiro-Wilk test ( Shapiro and Wilk , 1965 ) using the generic shapiro . test ( ) function . Statistical significance was then assessed by an unpaired t-test with Welch`s correction for heteroskedasticity ( Welch , 1947 ) with at least five biological replicates per group . Corrections for multiple testing were done applying the BH algorithm ( Benjamini and Hochberg , 1995 ) .	Like many other sea creatures , the worm Platynereis dumerilii reproduces by dispersing eggs and sperm in the water . For these animals , timing is everything: if they fail coordinate their release , the precious reproductive cells will drift in the vastness of the ocean without ever meeting their male or female counterparts . Internal clocks are a set of mechanisms that allow organisms to tune their internal processes to their environment . For example , the circadian clock helps many creatures to adapt to the cycle of day and night . This involves switching genes on and off according to the time of day . When a gene is activated , its information is copied into a molecule of RNA , which is then read to create proteins that will go on performing specific roles . To produce their eggs and sperm at the right time , P . dumerilii worms rely on a poorly understood internal clock which is synchronized by the moon cycle . To investigate this ‘inner calendar’ , Schenk , Bannister et al . developed a new technique that allows them to extract both RNA and proteins from the miniscule heads of the worms . The results showed that the internal clock synchronized by the lunar phases influenced the levels of many more proteins than RNA molecules . In comparison , other life events such as the worms becoming sexually mature , had a more similar impact on both protein and RNA regulation . This might suggest that the inner calendar that coordinates the worms with the moon cycle could work by changing protein , rather than RNA levels . The analysis also highlighted several molecular actors that may be essential for the worm’s inner clock to work properly . In the future , the new technique will help to dissect more finely how P . dumerilii and many other marine creatures stay synchronized with the moon , and spawn at the right time .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "ecology", "tools", "and", "resources", "neuroscience" ]	2019	Combined transcriptome and proteome profiling reveals specific molecular brain signatures for sex, maturation and circalunar clock phase
The evolutionary mechanisms leading to duplicate gene retention are well understood , but the long-term impacts of paralog differentiation on the regulation of metabolism remain underappreciated . Here we experimentally dissect the functions of two pairs of ancient paralogs of the GALactose sugar utilization network in two yeast species . We show that the Saccharomyces uvarum network is more active , even as over-induction is prevented by a second co-repressor that the model yeast Saccharomyces cerevisiae lacks . Surprisingly , removal of this repression system leads to a strong growth arrest , likely due to overly rapid galactose catabolism and metabolic overload . Alternative sugars , such as fructose , circumvent metabolic control systems and exacerbate this phenotype . We further show that S . cerevisiae experiences homologous metabolic constraints that are subtler due to how the paralogs have diversified . These results show how the functional differentiation of paralogs continues to shape regulatory network architectures and metabolic strategies long after initial preservation . Gene duplication provides raw material for evolution to act upon . Even so , most duplicate genes are inactivated and become pseudogenes before fixation . The molecular mechanisms behind paralog retention and differentiation have attracted considerable attention , and several general models have been proposed , including neofunctionalization ( Ohno , 1970; Zhang et al . , 2002 ) , gene dosage selection ( Conant and Wolfe , 2007; Sandegren and Andersson , 2009; Conant et al . , 2014 ) , subfunctionalization by duplication-degeneration-complementation ( Force et al . , 1999 ) , and subfunctionalization by escape from adaptive conflict ( Hittinger and Carroll , 2007; Des Marais and Rausher , 2008 ) . Theoretical studies have proposed that the fates of duplicate genes are rapidly determined after gene duplication events ( Moore and Purugganan , 2003; Innan and Kondrashov , 2010 ) . These models generally treat the preservation of duplicate genes as a race to distinguish their functions prior to the complete inactivation of one of the redundant paralogs , either through neutral ( Force et al . , 1999; Lynch et al . , 2001 ) or adaptive changes ( Clark , 1994; Lynch et al . , 2001 ) . Regardless of the initial functional changes or dosage effects facilitating the fixation of paralogs , retention is not the end of their evolutionary paths ( Gordon et al . , 2009; Conant et al . , 2014 ) . Duplicate genes continue to diverge in different lineages , providing additional evolutionary opportunities for organisms to diversify . Previously fixed copies of duplicate genes can alter their expression timing and patterns ( Huminiecki and Wolfe , 2004; Tümpel et al . , 2006 ) , change substrate affinities ( Voordeckers et al . , 2012 ) , and switch between regulatory modules ( Thompson et al . , 2013 ) . In several cases , paralogs encoding enzymes have been recruited to perform regulatory functions , such as S . cerevisiae HXK2 , GAL3 , and ARG82 ( Gancedo and Flores , 2008; Conant et al . , 2014; Gancedo et al . , 2014 ) . Previously differentiated developmental roles can even be transferred from one paralog to another during evolution ( Ureña et al . , 2016 ) . Perhaps more significantly , long-preserved paralogs can be lost in lineage-specific manners , a common phenomenon observed across the tree of life , including in bacteria ( Gómez-Valero et al . , 2007 ) , yeasts ( Scannell et al . , 2007 ) , Paramecium ( Aury et al . , 2006; McGrath et al . , 2014 ) , plants ( De Smet et al . , 2013 ) , fishes ( Amores et al . , 2004 ) , and mammals ( Amores et al . 1998; Blomme et al . , 2006 ) . Although pervasive , the importance of ongoing paralog diversification to the evolution of organismal traits and phenotypes remains underappreciated . Duplicate gene differentiation has heavily impacted the evolution of regulatory and metabolic networks ( Reece-Hoyes et al . , 2013; Voordeckers et al . , 2015 ) . Paralogs have contributed to the expansion of regulatory networks ( Teichmann and Babu , 2004 ) , the derivation of novel networks ( Conant and Wolfe , 2006; Wapinski et al . , 2010; Pérez et al . , 2014; Pougach et al . , 2014 ) , the specialization of network regulation ( Lin and Li , 2011 ) , and the robustness of networks to perturbation ( Papp et al . , 2004; Deutscher et al . , 2006 ) . The WGD has even been proposed to have facilitated the evolution of an aerobic glucose fermentation strategy called Crabtree-Warburg Effect in the lineage of yeasts that includes Saccharomyces ( Conant and Wolfe , 2007; Jiang et al . , 2008 ) . Gene regulation and metabolism are heavily intertwined biological processes , but there are few eukaryotic models that allow for an integrated study of the ongoing differentiation of paralogous genes with regulatory and metabolic diversification ( Yamada and Bork , 2009; Conant et al . , 2014 ) . The Saccharomyces cerevisiae GALactose sugar utilization network is one of the most extensively studied eukaryotic regulatory and metabolic networks , and its homologous networks are evolutionarily dynamic in yeasts . In S . cerevisiae , it includes the three enzymes of the Leloir pathway ( Gal1 , Gal7 , and Gal10 ) that catabolize galactose , the galactose transporter Gal2 , and three regulators . In the absence of galactose , the transcription factor Gal4 is inhibited by the co-repressor Gal80 . When galactose is present , Gal80 is sequestered by the co-inducer Gal3 , allowing Gal4 to activate the expression of the GAL network ( Johnston , 1987; Bhat and Murthy , 2001; Egriboz et al . , 2013 ) . Numerous studies have shown that the GAL networks of various yeast lineages vary in gene content ( Hittinger et al . , 2004 , 2010; Wolfe et al . , 2015 ) and gene activity ( Peng et al . , 2015; Roop et al . , 2016 ) . Despite these findings , the impacts of variable network architectures on the evolution of gene regulation and metabolism are not well understood . As a model for how duplicate gene divergence creates variable network architectures , we functionally characterized the GAL network of Saccharomyces uvarum ( formerly known as Saccharomyces bayanus var . uvarum ) and compared it to S . cerevisiae . Here we show that two GAL network paralog pairs in S . uvarum have diverged to different degrees and states than their S . cerevisiae homologs . We further show that , unlike S . cerevisiae , S . uvarum deploys a second co-repressor that prevents over-induction of the network . S . uvarum mutants lacking both co-repressors revealed surprising constraints on the rapid utilization of galactose; specifically , they arrested their growth , and metabolomic investigations suggested that they experienced metabolic overload . We show that homologous constraints exist in a milder form in S . cerevisiae , and the degree of metabolic constraint is affected by how GAL network paralogs have diversified between the species . These results show how , after a hundred of million of years of preservation , two pairs of interacting duplicate genes have continued to diverge functionally in ways that broadly impact metabolism , regulatory network structures , and the future evolutionary trajectories available . S . uvarum has orthologs encoding all regulatory and structural genes that are present in S . cerevisiae , but it has duplicate copies of two additional genes . The first additional duplicate gene is GAL80B , which is a paralog of GAL80; this pair of paralogs was created by the whole genome duplication ( WGD ) event roughly 100 million years ago ( Wolfe and Shields , 1997; Marcet-Houben and Gabaldón , 2015 ) . GAL80B has been retained in the S . uvarum-Saccharomyces eubayanus clade , but it was lost in the S . cerevisiae-Saccharomyces arboricola clade ( Hittinger et al . , 2010 , 2004; Scannell et al . , 2011; Caudy et al . , 2013; Hittinger , 2013; Liti et al . , 2013; Baker et al . , 2015 ) . The second one is GAL2B , which was created by a recent tandem duplication in S . uvarum-S . eubayanus clade . Both S . cerevisiae and S . uvarum also contain a pair of specialized paralogs created by the WGD , GAL1 and GAL3 . By comparing amino acid sequences against the S . cerevisiae GAL network , we found that most GAL genes are diverged to a similar extent ( Figure 1 and Figure 1—source data 1 ) , except for GAL4 , which is primarily conserved in its DNA-binding and other functionally characterized domains . None of the S . uvarum GAL homologs exhibited significantly elevated rates of protein sequence evolution ( from previously calculated dN/dS ratios [Byrne and Wolfe , 2005] ) , which might have otherwise suggested extensive neofunctionalization . Thus , we focused on whether and how the key regulatory genes partitioned functions differently between the two species . 10 . 7554/eLife . 19027 . 003Figure 1 . The S . uvarum GAL network . ( A ) The GAL regulatory network . ( B ) The GAL or Leloir metabolic pathway . The colors show the amino acid identity of each component compared to their S . cerevisiae homologs ( full data in Figure 1—source data 1 ) . Proteins with two homologs in S . uvarum are split into two parts: Gal1/Gal3 and Gal80/Gal80b ( also known as Sbay_12 . 142 [Scannell et al . , 2011] or 670 . 20 [Caudy et al . , 2013] ) are two pairs of paralogs from a WGD event , while Gal2/Gal2b ( also known as Sbay_10 . 165 [Scannell et al . , 2011] or 672 . 62 [Caudy et al . , 2013] ) are paralogs from a recent tandem duplication event ( Hittinger et al . , 2004 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 19027 . 00310 . 7554/eLife . 19027 . 004Figure 1—source data 1 . Amino acid identity and GAL gene composition between S . uvarum and S . cerevisiae GAL network . Quantitative data underlying Figure 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 19027 . 004 In S . cerevisiae , the GAL1 and GAL3 paralogs are descended from an ancestral bi-functional protein that was both a co-inducer and a galactokinase ( Rubio-Texeira , 2005; Hittinger and Carroll , 2007 ) . They are almost completely subfunctionalized: ScerGAL3 lost its galactokinase activity and became a dedicated co-inducer , whereas ScerGAL1 lost most of its co-inducer activity but maintains galactokinase activity ( Platt and Reece , 1998; Platt et al . , 2000; Timson et al . , 2002; Hittinger and Carroll , 2007; Lavy et al . , 2016 ) . Unlike ScerGal3 , SuvaGal3 retains a -Ser-Ala- dipeptide in its active site that is sufficient to weakly restore galactokinase activity when added back to ScerGal3 ( Platt et al . , 2000 ) , so we hypothesized that SuvaGAL3 encodes a functional galactokinase . To test this hypothesis , we precisely replaced the coding sequence of ScerGAL1 , the gene encoding the sole galactokinase in S . cerevisiae ( Platt et al . , 2000 ) , with SuvaGAL3 in S . cerevisiae . As expected , SuvaGAL3 conferred robust growth in galactose when driven by the ScerGAL1 promoter , suggesting that SuvaGAL3 retains galactokinase activity ( Figure 2A ) . Nonetheless , the S . uvarum gal1 null mutant did not grow better in 2% galactose than it did without any carbon source , a phenotype similar to the S . cerevisiae gal1 null mutant ( Figure 2—figure supplement 1 ) , indicating that the native GAL3 promoter expression is insufficient to support robust metabolism . 10 . 7554/eLife . 19027 . 005Figure 2 . SuvaGAL1 and SuvaGAL3 are not as subfunctionalized as ScerGAL1 and ScerGAL3 . ( A ) S . uvarum GAL3 likely encodes a functional galactokinase . The error bars represent standard deviations of three biological replicates . A Wilcoxon rank sum test comparing the average times to first doubling between S . cerevisiae gal1 and S . cerevisiae gal1∆::SuvaGAL3 was significantly different ( p=5 . 2e-3 , n = 6 ) . Note that driving ScerGAL3 from the ScerGAL1 promoter was insufficient to support growth with galactose as the sole carbon source , but SuvaGAL3 was sufficient . ( B ) Unlike S . cerevisiae gal3∆ , S . uvarum gal3∆ does not show Long-Term Adaption ( LTA ) . Strains were cultured in SC + 2% galactose . Wilcoxon rank sum tests comparing the average times to first doubling between S . uvarum gal3∆ and S . uvarum wild-type strains were significantly different ( p=4 . 5e-5 , n = 12 ) . These experiments have been repeated independently at least twice with three biological replicates , but growth curves display only one representative replicate because LTA emergence is stochastic . ( C ) LTA was recapitulated in S . uvarum gal3∆ by replacing its GAL1 promoter with the S . cerevisiae GAL1 promoter ( left panel ) or , to a much lesser extent , by replacing the coding sequence ( right panel ) . The insets show the times to the first doubling for the strains for their respective panels . The bar colors in the inset are the same as the growth curves . To highlight strain comparisons that test discrete hypotheses , three genotypes are repeated in Figure 2B and in both panels of Figure 2C: S . uvarum gal3∆ , S . cerevisiae wild-type , and S . cerevisiae gal3∆ . Strains were cultured in SC + 2% galactose . Wilcoxon rank sum tests comparing the average times to first doubling between strains were as follows: ( 1 ) p=4 . 6e-4 and n = 9 for S . uvarum gal3∆ versus S . uvarum gal3∆ gal1∆::PSuvaGAL1-ScerGAL1 , ( 2 ) p=4 . 2e-5 and n = 12 for S . uvarum gal3∆ versus S . uvarum gal3∆ gal1∆::PScerGAL1-SuvaGAL1 , and ( 3 ) p=0 . 21 and n = 12 for S . uvarum gal3∆ gal1∆::PScerGAL1-SuvaGAL1 versus S . cerevisiae gal3 . DOI: http://dx . doi . org/10 . 7554/eLife . 19027 . 00510 . 7554/eLife . 19027 . 006Figure 2—figure supplement 1 . S . uvarum and S . cerevisiae have qualitatively similar gal1 null phenotypes . ‘+’ indicates growth after 7 days , while ‘−’ indicates no growth after 7 days when compared to the negative control ( minimal media without a carbon source ) ( Materials and methods ) . Note that driving SuvaGAL3 from the ScerGAL1 promoter was sufficient to support growth with galactose as the sole carbon source ( Figure 2A ) . DOI: http://dx . doi . org/10 . 7554/eLife . 19027 . 006 To further examine the functional divergence between SuvaGAL1 and SuvaGAL3 , we knocked out GAL3 in S . uvarum . Surprisingly , the S . uvarum gal3 null mutant did not show the classic Long-Term Adaptation ( LTA ) phenotype of the S . cerevisiae gal3 null mutant ( Tsuyumu and Adams , 1973 ) . Instead of a growth delay of multiple days , we observed a delay of only a few hours in S . uvarum gal3∆ relative to wild-type ( Figure 2B ) . These results suggest that other genes in S . uvarum may be able to partially compensate for the deletion of SuvaGAL3 , such as its paralog , SuvaGAL1 . To determine whether GAL1 differences between S . uvarum and S . cerevisiae might be responsible for the different gal3 null phenotypes , we replaced the SuvaGAL1 coding sequence or promoter sequence with their ScerGAL1 counterparts in the background of S . uvarum gal3∆ . The ScerGAL1 promoter swap in S . uvarum gal3∆ largely recapitulated LTA , while the ScerGAL1 coding sequence swap extended the delay to a lesser extent ( Figure 2C ) . Since the GAL1-GAL10 promoter is a divergent promoter , genetic modifications ( evolved or engineered ) inevitably impact both genes , as well as perhaps a lncRNA previously described in S . cerevisiae ( Cloutier et al . , 2016 ) . These results suggest that differences at the GAL1 locus , especially within this promoter , are primarily responsible for the lack of LTA in the S . uvarum gal3∆ mutant . Overall , the data suggest that SuvaGAL1 is functionally redundant with SuvaGAL3 to a much greater extent than are ScerGAL1 and ScerGAL3 . Thus , it is likely that the homologs in the common ancestor of S . uvarum and S . cerevisiae were more functionally redundant than in modern S . cerevisiae , and considerable subfunctionalization between ScerGAL1 and ScerGAL3 happened after the divergence of S . uvarum and S . cerevisiae . Next , we examined the functional divergence of the other pair of paralogous regulatory genes , SuvaGAL80 and SuvaGAL80B , which are homologous to the ScerGAL80 gene that encodes the sole GAL gene co-repressor in S . cerevisiae . We first examined the expression of these two genes in the presence or absence of galactose ( Figure 3A ) . RNA sequencing ( RNA-Seq ) showed that SuvaGAL80 was expressed at a higher level than SuvaGAL80B in the absence of galactose ( i . e . with glycerol or glucose as the sole carbon source ) . In contrast , in the presence of galactose , SuvaGAL80B was induced by 133-fold , much higher than the 6-fold induction observed for SuvaGAL80 ( Figure 3B ) . S . uvarum gal80 null mutants had a shorter lag time than wild-type in galactose , as seen in S . cerevisiae gal80 null mutants ( Torchia et al . , 1984; Segrè et al . , 2006; Hittinger et al . , 2010 ) , but gal80b null mutants did not ( Figure 3C ) . Deleting SuvaGAL80 also resulted in elevated GAL1 expression in the non-inducing condition ( i . e . 5% glycerol ) , while deleting SuvaGAL80B had no detectable effect ( Figure 3D ) . Therefore , we conclude that SuvaGAL80 is the main gene responsible for repressing the GAL network in the absence of galactose . 10 . 7554/eLife . 19027 . 007Figure 3 . SuvaGAL80 and SuvaGAL80B encode co-repressors with partially overlapping functions . ( A ) Expression divergence between SuvaGAL80 and SuvaGAL80B . The bar graph on the left shows the mRNA levels ( in log2 of Reads Per Kilobase of transcript per Million mapped reads or RPKM ) of SuvaGAL80 and SuvaGAL80B in SC + 2% galactose , SC + 5% glycerol , and SC + 2% glucose . Error bars represent the standard deviations of three biological replicates . ( B ) Divergent galactose induction between SuvaGAL80 and SuvaGAL80B . The bar graph shows the ratio of mRNA levels between galactose ( gal ) and glycerol ( gly ) , or between galactose and glucose ( glu ) from the data in Panel A . ( C ) Removing SuvaGAL80 conferred rapid initial growth in galactose . The bar graph shows the average time to first doubling of three biological replicates of each genotype in SC + 2% galactose from a representative experiment . S . uvarum gal80∆ grew significantly faster than wild-type ( p=1 . 8e-3 , n = 14 , Wilcoxon rank sum test ) , but S . uvarum gal80b∆ did not ( p=0 . 61 , n = 14 , Wilcoxon rank sum test ) . ( D ) Removing SuvaGAL80 resulted in constitutive GAL1 expression . The histogram shows the fluorescence levels of an EGFP reporter when driven by the S . uvarum GAL1 promoter in SC + 5% glycerol as determined by flow cytometry . ( E ) Removing SuvaGAL80B led to the elevated GAL1 expression in a mixture of glucose and galactose . Flow cytometry was conducted on strains cultured in SC + 5% galactose +2% glucose . ( F ) Removing SuvaGAL80B caused a fitness defect in a mixture of glucose and galactose . The specific growth rate of S . uvarum gal80b∆ was significantly lower than wild-type ( p=2 . 7e-4 , n = 18 , Wilcoxon rank sum test ) . ( G ) SuvaGAL80 and SuvaGAL80B were both able to partially compensate for the loss of the other in repressing conditions , but the double-knockout resulted in constitutive expression . The histogram reports flow cytometry data from strains cultured in SC + 2% glucose for 9 hr . DOI: http://dx . doi . org/10 . 7554/eLife . 19027 . 00710 . 7554/eLife . 19027 . 008Figure 3—figure supplement 1 . In SC + 2% galactose , S . uvarum gal80∆ and gal80b∆ had GAL1 expression levels similar to the wild-type at mid-log phase . Flow cytometry histogram of PGAL1-EGFP fluorescence . DOI: http://dx . doi . org/10 . 7554/eLife . 19027 . 008 Perhaps because of its dynamic expression , the deletion mutant phenotype of S . uvarum gal80b∆ proved condition dependent . Consistent with previous negative results ( Caudy et al . , 2013 ) , no apparent phenotypic differences were observed for the S . uvarum gal80b∆ strain when it was grown in galactose , nor were its GAL1 expression levels altered ( Figure 3C and Figure 3—figure supplement 1 ) . Nonetheless , in a mixture of galactose and glucose , we observed elevated GAL1 expression in S . uvarum gal80b∆ strains , beyond the levels observed in S . uvarum gal80∆ strains ( Figure 3E ) . Additionally , S . uvarum gal80b∆ grew significantly slower than wild-type after transfer from galactose to a mixture of galactose and glucose ( Figure 3F ) , suggesting that SuvaGAL80B plays a specific and biologically important repressive role in conditions where it is required to prevent network over-induction . We also observed strong negative epistasis when both co-repressors were removed: the co-repressor double knockout had substantially higher GAL1 expression than either single knockout strain or the S . uvarum wild-type strain in the absence of galactose ( Figure 3G ) . Thus , SuvaGAL80 and SuvaGAL80B encode partially redundant GAL gene co-repressors that can each partially compensate for the loss of the other . We conclude that SuvaGAL80B may play a minor role in the absence of galactose , but it provides important modulation in induced conditions . Surprisingly , knocking out both GAL80 and GAL80B in S . uvarum resulted in a strong Temporary Growth Arrest ( TGA ) phenotype in galactose ( Figure 4A ) . This result stands in sharp contrast to the observation that S . cerevisiae gal80 null mutant strains from multiple genetic backgrounds ( the lab strains S288c , W303 , and R21 , as well as the vineyard strain RM11-1a examined here ) grew faster in galactose , a phenotype shared with Saccharomyces kudriavzevii gal80 null mutants and attributed to the constitutive GAL expression ( Torchia et al . , 1984; Segrè et al . , 2006; Hittinger et al . , 2010 ) . This growth arrest was not a genetic engineering artifact; reintroducing SuvaGAL80 completely rescued the growth arrest , and knocking out these two genes with different markers produced the same mutant phenotype ( Figure 4—figure supplement 1 ) . More importantly , introducing ScerGAL80 completely rescued the growth arrest ( Figure 4—figure supplement 1 ) , suggesting that the TGA phenotype was not due to novel molecular functions specific to SuvaGAL80 or SuvaGAL80B . Instead , the dramatically varied phenotypes imply that these two species have different regulatory or metabolic wiring for galactose metabolism . 10 . 7554/eLife . 19027 . 009Figure 4 . The galactose-dependent temporary growth arrest phenotype of S . uvarum gal80∆ gal80b∆ . ( A ) The Temporary Growth Arrest ( TGA ) phenotype in SC + 2% galactose . The averages of the log2 of the ratios between absorbances at each time point ( ODt ) and initial absorbances ( OD0 ) for three biological replicates are shown . The error bars represent standard deviations . The inset shows the first six hours for three biological replicates each of S . cerevisiae wild-type and gal80∆ ( in the background of S . cerevisiae RM11-1a , a vineyard strain ) . ( B ) The degree of the TGA phenotype was concentration dependent . A representative experiment with three biological replicates is shown; the experiment has been repeated three times . ( C ) Excessive reactive oxygen species ( ROS ) were accumulated in S . uvarum gal80∆ gal80b∆ during the TGA phase . ROS levels are reported as relative fluorescence and were measured 6 . 5 hr after inoculation into SC + 2% galactose ( p=8 . 6e-6 , n = 11 , Wilcoxon rank sum test ) . The bar graph on the right shows a positive control using S . uvarum wild-type in YPD and YPD + 10 mM H2O2 . DOI: http://dx . doi . org/10 . 7554/eLife . 19027 . 00910 . 7554/eLife . 19027 . 010Figure 4—figure supplement 1 . The TGA phenotype of S . uvarum gal80∆ gal80b∆ can be rescued by S . cerevisiae GAL80 or by re-introducing SuvaGAL80 . The bar graphs show the average times to first doubling time of three biological replicates . Strains were cultured in SC + 2% galactose . DOI: http://dx . doi . org/10 . 7554/eLife . 19027 . 01010 . 7554/eLife . 19027 . 011Figure 4—figure supplement 2 . Galactose-specific global differential expression of S . uvarum gal80∆ gal80b∆ . ( A ) S . uvarum GAL network comprises similar targets as the S . cerevisiae GAL network . The bar graph shows the log2 of the RPKM ratio between S . uvarum gal80∆ gal80b∆ and wild-type in SC + 5% glycerol . Note that GAL80 and GAL80B are not in the list because they were knocked out in the double mutant , but both genes contain putative Gal4 binding sites in their promoters . GAL3 was considered differentially expressed by edgeR ( p-value = 2 . 95e-8 in the condition of glycerol at 11 . 2-fold and p=2 . 83e-8 in glucose at 7 . 3-fold , both at FDR < 1 . 4e-5 ) , although it was not by EBSeq ( posterior probability of being equally expressed was 0 . 13 in glycerol and 0 . 28 in glucose ) . The two other experimentally verified S . cerevisiae Gal4 target genes ( MTH1 and PCL10 ) were not considered up-regulated by either edgeR ( p=0 . 5 at 0 . 9-fold for MTH1 and p=0 . 8 at 1-fold for PCL10 in glycerol , p=0 . 1 at 1 . 5-fold for MTH1 and p=0 . 1 at 1 . 5-fold for PCL10 in glucose , all at FDR = 1 ) or EBSeq ( posterior probabilities of being equally expressed for MTH1 and PCL10 were 0 . 7 and 0 . 8 in glycerol , respectively , and 1 . 0 and 1 . 0 in glucose , respectively ) , despite having conserved putative Gal4 binding sites in their upstream sequences in S . uvarum . The three genes described as down-regulated in S . uvarum gal80b∆ strains by Caudy et al . ( Caudy et al . , 2013 ) were not differentially expressed in S . uvarum gal80∆ gal80b∆ strains in our growth conditions . ( B-C ) Venn diagrams of differential expression of S . uvarum gal80∆ gal80b∆ harvested at ( B ) the TGA phase , and ( C ) mid-log phase in SC + 2% glucose , +5% glycerol , or +2% galactose . Note that most gene expression changes were galactose-specific , suggesting that they were caused by metabolic defects , rather than direct regulation by Gal4 . Note that , relative to wild-type , there were still hundreds of differentially expressed genes at the mid-log phase , but most ( 78% , 783 of 1006 ) genes that were differentially expressed during the TGA phase had returned to normal expression . DOI: http://dx . doi . org/10 . 7554/eLife . 19027 . 01110 . 7554/eLife . 19027 . 012Figure 4—figure supplement 3 . High performance liquid chromatography measurements of key metabolites in SC + 2% galactose in S . uvarum gal80∆ gal80b∆ and wild-type during the TGA phase and after the growth resumed . Statistically significant data points are marked by asterisks ( p<0 . 05 , *p<0 . 01 , one-tailed Student’s t-test ) . Red corresponds to ethanol , and blue corresponds to galactose; ethanol was produced by galactose catabolism , but ethanol production provided a more sensitive readout than galactose consumption in early-stage cultures . Note that S . uvarum gal80∆ gal80b∆ produced significantly more ethanol by the 1 hr time point ( before the TGA phase ) , but the S . uvarum wild-type strain produced significantly more ethanol at the 3 hr and later TGA time points . DOI: http://dx . doi . org/10 . 7554/eLife . 19027 . 01210 . 7554/eLife . 19027 . 013Figure 4—figure supplement 4 . GAL1 expression was higher at the early stages of growth in SC + 2% galactose in the S . uvarum gal80∆ gal80b∆ background but gradually decreased . Fluorescence levels were obtained by flow cytometry , normalized to forward scatter , and plotted as histograms . 4 hr was during the TGA phase , whereas 8 hr was approaching the end of the TGA phase . DOI: http://dx . doi . org/10 . 7554/eLife . 19027 . 01310 . 7554/eLife . 19027 . 014Figure 4—figure supplement 5 . Fructose , mannose , or glucose alone did not lead to a TGA phenotype or other growth defects . All experiments were performed in SC media with the carbon sources indicated . DOI: http://dx . doi . org/10 . 7554/eLife . 19027 . 01410 . 7554/eLife . 19027 . 015Figure 4—figure supplement 6 . The regulation of PGM1 by galactose was inferred as the ancestral state . ( A ) mRNA levels of S . uvarum PGM1 and PGM2 during mid-log phase in SC + 2% galactose , SC + 5% glycerol , and SC + 2% glucose . Note that PGM2 , which encodes the major isoform of phosphoglucomutase , has long been known to be transcriptionally induced by ~three–four fold in galactose , but it lacks a Gal4 binding site and does not appear to be a direct target in S . cerevisiae ( Oh and Hopper , 1990; Rubio-Texeira , 2005 ) . These features are broadly shared with S . uvarum PGM2 , which is transcriptionally induced two-fold by galactose relative to glycerol but is not transcriptionally up-regulated in the gal80∆ gal80b∆ mutant; nor does it have a consensus Gal4 site . ( B ) Conservation of putative Gal4 binding sites upstream of PGM1 in S . uvarum , S . eubayanus , S . arboricola , and two outgroup species . The orange dot indicates the inferred loss of direct regulation of PGM1 by Gal4 based on the presence or absence of putative Gal4 binding sites ( CGGN11CCG ) . The distances upstream from the start codon are shown at the right . The putative sites are shown as red boxes at the relative position of the upstream sequences of PGM1 . Note that , in Kazachstania nagansihii , the upstream intergenic region of PGM1 ortholog is 1958 bp , an unusually long intergenic region for yeasts , and contains a divergent promoter that also drives expression of the PMU1 ortholog . DOI: http://dx . doi . org/10 . 7554/eLife . 19027 . 015 To test whether the TGA phenotype was associated with S . uvarum-specific GAL network members , we performed RNA-Seq on S . uvarum gal80∆ gal80b∆ in 2% glucose or 5% glycerol , conditions where the complete GAL network is expected to be constitutively expressed ( Torchia et al . , 1984; Segrè et al . , 2006; Hittinger et al . , 2010 ) . We identified genes as GAL network members if and only if they were: ( 1 ) significantly up-regulated in S . uvarum gal80∆ gal80b∆ compared to the wild-type at FDR = 0 . 05 ( 35 genes ) ; ( 2 ) up-regulated by at least two-fold ( 19 genes ) ; ( 3 ) up-regulated in both glucose and glycerol ( nine genes ) ; and ( 4 ) predicted to contain Gal4 consensus binding sites ( CGGN11CCG ) upstream of their coding sequences . Using these stringent criteria , we found eight potential GAL network members in S . uvarum , seven of which were shared with S . cerevisiae based on previous chromatin immunoprecipitation and gene expression data ( GAL1 , GAL2 , GAL2B , GAL7 , GAL10 , MEL1 , and GCY1 ) ( Torchia et al . , 1984; Ren et al . , 2000 ) ( Figure 4—figure supplement 2A ) . GAL3 , a well-established Gal4 target in S . cerevisiae , was considered differentially expressed using less stringent criteria , but orthologs of two other known targets were not ( MTH1 and PCL10 ) . The sole novel GAL network member in S . uvarum was the PGM1 gene , which was up-regulated 26-fold in 5% glycerol in S . uvarum gal80∆ gal80b∆ relative to wild-type . In S . cerevisiae , PGM1 encodes the minor isoform of phosphoglucomutase , which , along with Pgm2 , connects the Leloir pathway to glycolysis ( Figure 1 ) . Notwithstanding the PGM1 gene , we conclude that the S . uvarum and S . cerevisiae GAL networks have similar compositions , and the handful of differences do not seem to readily explain the remarkably strong and unexpected TGA phenotype seen in S . uvarum strains lacking their co-repressors . In contrast to the constitutive expression of a fairly small network of direct Gal4 targets seen during growth in glucose and glycerol , S . uvarum gal80∆ gal80b∆ double mutants experienced global changes in gene expression that were specific to growth in galactose ( Figure 4—figure supplement 2B , C ) . During the TGA phase , 1006 genes were differentially expressed in S . uvarum gal80∆ gal80b∆ relative to wild-type ( 620 genes up-regulated and 386 genes down-regulated by at least two-fold with FDR = 0 . 05 [Figure 5—source data 1] ) . After the mutant resumed growth in galactose , the vast majority ( 78% , 783 of 1006 genes ) of these genes returned to expression levels indistinguishable from wild-type , and Gene Ontology ( GO ) term analysis showed that most of the biological processes affected during the TGA phase returned to normal ( Supplementary file 1 ) . The TGA phase gene expression profile was not consistent with a global environmental stress response ( e . g . nuclear ribosome biogenesis and rRNA processing were up-regulated ) but instead suggested a complex and incoherent integration of the regulatory signals that govern metabolism ( Figure 5—source data 1 and Supplementary file 1 ) . Several lines of evidence suggested that this mis-regulation might be caused by overly rapid galactose catabolism immediately prior to the TGA phase . First , the optical density of the co-repressor double mutant initially increased faster than the wild-type in galactose and only plateaued after about 1 . 5 hr ( Figure 4A ) . Second , during this early growth in galactose , the co-repressor double mutant produced more ethanol than the wild-type ( Figure 4—figure supplement 3 ) . Third , GAL1 was also strongly overexpressed in the mutant early during growth in galactose , but GAL1 expression gradually converged with the wild-type strain as the cells transitioned into the TGA phase ( Figure 4—figure supplement 4 ) . Finally , the severity of the TGA phenotype depended strongly on galactose concentration ( Figure 4B ) , and growth defects were not seen in other carbon sources ( Figure 4—figure supplement 5 ) . To further characterize how overly rapid galactose catabolism might lead to the TGA phenotype , we performed metabolomic analyses using mass spectrometry on co-repressor double mutant and wild-type strains cultured in 2% galactose . Prior to the TGA phase , the co-repressor double mutant accumulated galactose-1-phosphate , a known toxic intermediate of galactose metabolism ( Petry and Reichardt , 1998; de Jongh et al . , 2008 ) , but this two-fold accumulation ( relative to wild-type ) seemed unlikely to be sufficient to explain the TGA phenotype . The level of galactose-1-phosphate in S . uvarum gal80∆ gal80b∆ returned to normal during the TGA phase ( Figure 5 and Figure 5—source data 2 ) and was not nearly as strong as in S . cerevisiae gal7∆ or gal10∆ controls ( seven- to 11-fold relative to S . cerevisiae wild-type ) ( Figure 5—figure supplement 1 ) . Moreover , we did not observe gene expression signatures consistent with the previously described responses to galactose-1-phosphate toxicity ( e . g . , environmental stress response , unfolded protein response ) ( Slepak et al . , 2005; De-Souza et al . , 2014 ) . Instead , both transcriptomic and metabolomic analyses revealed broad metabolic defects as bottlenecks developed downstream of the Leloir pathway . During the growth arrest , we observed increased expression of genes that encode glycolytic enzymes ( Figure 5 and Figure 5—source data 1 ) . Key metabolic intermediates also accumulated in S . uvarum gal80∆ gal80b∆ strains before and during growth arrest , especially in upper glycolysis and interacting pathways ( Figure 5 and Figure 5—source data 2 ) . In particular , fructose-1 , 6-biphosphate accumulated significantly prior to the TGA phase ( 12 . 6-fold of wild-type levels ) ( Figure 5 and Figure 5—source data 2 ) , a bottleneck that frequently occurs when upper glycolysis outpaces lower glycolysis ( van Heerden et al . , 2014 ) . Under these conditions , inorganic phosphate becomes a limiting factor for growth as the ‘investment’ steps in upper glycolysis deplete the cells of ATP and phosphate to form sugar phosphates ( Teusink et al . , 1998; van Heerden et al . , 2014 ) . Indeed , S . uvarum gal80∆ gal80b∆ strains had one-fifth of the ATP as wild-type prior to the TGA phase ( Figure 5—source data 2 ) and had significantly up-regulated ( 25-fold ) expression of PHO84 , which encodes a high-affinity phosphate transporter ( Figure 5—source data 1 ) . 10 . 7554/eLife . 19027 . 016Figure 5 . Overly rapid galactose catabolism leads to metabolic overload and bottlenecks . The graph shows the metabolite levels and transcript expression for the Leloir pathway , glycolysis , trehalose cycle , glycerol biosynthesis , TCA cycle , and electron transport chain . Purple steps cost ATP or inorganic phosphate ( Pi ) , while green steps generate ATP or Pi . Strains were cultured in SC + 2% galactose . The arrows are color-coded to represent the RNA-Seq gene expression differences of S . uvarum gal80∆ gal80b∆ relative to wild-type at 4 hr ( red , up in the mutant; blue , down in the mutant; black , similar to wild-type; mixed colors ( e . g . black and blue ) indicate that the expression of genes involved in this step differs ) . The boxes next to each metabolite represent the log2 of the metabolite concentration differences relative to wild-type over time ( 0 . 5 , 1 . 5 , 3 , 5 , and 16 hr , respectively ) . The statistical significance for metabolite levels was assessed using Student’s t-tests ( n = 3 , p<0 . 05 with gray reported as not significant ) . The 1 . 5 hr to 5 hr time points correspond to the TGA phase , whereas the 16 hr time point corresponds to mid-log phase after recovery from the TGA phase . 1 , the sum of the metabolite concentrations of glycerol-3-phosphate and glycerol-2-phosphate , the latter of which is not known to be a major metabolite in Saccharomyces; 2 , the sum of the metabolite levels of 3-phosphoglycerate and 2-phosphoglycerate . DOI: http://dx . doi . org/10 . 7554/eLife . 19027 . 01610 . 7554/eLife . 19027 . 017Figure 5—source data 1 . Differential gene expression between S . uvarum gal80∆ gal80b∆ and wild-type during the TGA phase and at mid-log phase . DOI: http://dx . doi . org/10 . 7554/eLife . 19027 . 01710 . 7554/eLife . 19027 . 018Figure 5—source data 2 . Mass spectrometry metabolomic results comparing S . uvarum gal80∆ gal80b∆ to wild-type during the TGA phase and mid-log phase . DOI: http://dx . doi . org/10 . 7554/eLife . 19027 . 01810 . 7554/eLife . 19027 . 019Figure 5—figure supplement 1 . Galactose-1-phosphate accumulation of S . cerevisiae gal7∆ and gal10∆ . Galactose-1-phosphate levels were quantified by mass spectrometry . Samples were harvested after 4 . 5 hr of growth in 2% galactose . ‘LOQ’ stands for “Limit of Quantification” . DOI: http://dx . doi . org/10 . 7554/eLife . 19027 . 019 S . cerevisiae combats metabolic overload in upper glycolysis by using two main pathways to restore phosphate pools . The trehalose cycle temporarily reroutes upper glycolysis to store sugars as trehalose ( van Heerden et al . , 2014 ) , while glycerol biosynthesis offers an early exit from glycolysis ( Luyten et al . , 1995 ) . Disrupting the S . cerevisiae trehalose cycle leads to the accumulation of fructose-1 , 6-biphosphate , decreased ATP levels , and ultimately growth arrest due to a metabolically unbalanced state ( van Heerden et al . , 2014; Gibney et al . , 2015 ) , metabolic changes similar to the S . uvarum TGA phenotype . Strikingly , both pathways experienced dramatic bottlenecks in S . uvarum gal80∆ gal80b∆ strains before and during the TGA phase . Specifically , S . uvarum gal80∆ gal80b∆ cells accumulated 79- to 231-fold more trehalose-6-phosphate before and during the TGA phase , while they accumulated 225-fold more glycerol-3-phosphate before the TGA phase , the latter of which lessened to some extent during the TGA phase ( 3- to 16-fold ) ( Figure 5 and Figure 5—source data 2 ) . These data are consistent with the hypothesis that the trehalose cycle and the glycerol biosynthesis pathway are unable to handle the metabolic overload when galactose is catabolized too rapidly in S . uvarum strains lacking the GAL network repression system . The metabolic effects of the TGA phenotype also reverberated downstream , leading to the transcriptional down-regulation of the lower part of the TCA cycle and the electron transport chain ( Figure 5 ) . Reduced respiratory activity has been shown to increase the formation of reactive oxygen species ( ROS ) ( Barros et al . , 2004 ) , and the co-repressor double mutant had strong signatures of mitochondrial dysfunction . GO terms related to mitochondrial structural components , mitochondrial translation , and respiration were among the most strongly down-regulated ( Supplementary file 1 ) . Indeed , we observed significantly higher accumulation of ROS in S . uvarum gal80∆ gal80b∆ during the TGA phase by using the general ROS indicator dichlorodihydrofluorescein diacetate ( H2DCF-DA ) ( Figure 4C ) . We conclude that disconnecting S . uvarum galactose metabolism from the negative feedback loops normally provided by the co-repressors Gal80 and Gal80b likely allows galactose to enter the Leloir pathway and glycolysis too rapidly , leading to metabolic defects far beyond the mild accumulation of galactose-1-phosphate and deep into central metabolism . To determine whether the TGA phenotype reflected a more general metabolic constraint imposed by the interplay between glycolysis and interacting metabolic pathways , we grew S . uvarum gal80∆ gal80b∆ in mixtures of galactose with fructose , mannose , or glucose . Fructose , mannose , and glucose are all primarily catabolized through glycolysis , but only glucose generates glycolytic intermediates that are upstream of the trehalose cycle ( Figure 5 ) . Thus , fructose and mannose are expected to contribute directly to metabolic overload with minimal offsetting effects from the trehalose cycle . If the interaction between glycolytic load and the trehalose cycle were important to the TGA phenotype , growing the double mutant in mixtures of galactose with fructose or mannose would exacerbate the growth arrest . In contrast , if the TGA phenotype were caused by galactose-specific metabolism , the addition of these more preferred sugars would have no effect , or perhaps mitigate the TGA phenotype . Consistent with the TGA phenotype being caused by a general overloading of upper glycolysis , both fructose and mannose strongly exacerbated the TGA phenotype in S . uvarum gal80∆ gal80b∆ , while glucose partially rescued the TGA phenotype ( Figure 6A ) . Importantly , mixing fructose or mannose with galactose had much stronger defects than the identical amounts of galactose alone ( Figure 4B and Figure 6A ) . Co-culturing wild-type S . uvarum strains in galactose with these sugars was not inherently toxic ( Figure 6A ) , so the presence of the co-repressors allows cells to cope with this challenge . Growing S . uvarum gal80∆ gal80b∆ in fructose , mannose , or glucose alone also did not cause growth defects ( Figure 4—figure supplement 5 ) . Moreover , deleting GAL1 completely rescued the TGA phenotype in the co-repressor double mutant ( Figure 6B ) , while mixtures of mannose and galactose dramatically increased the levels of ROS in S . uvarum gal80∆ gal80b∆ ( Figure 6C ) , implying that the phenotypic enhancement caused by this sugar mixture acts through the same mechanism observed in galactose alone . Collectively , these results suggest that overly rapid catabolism of sugars can lead to general metabolic and growth defects when the appropriate futile metabolic cycles and negative feedback regulatory loops are not able to slow down catabolism . 10 . 7554/eLife . 19027 . 020Figure 6 . The addition of sugars downstream of the trehalose cycle exacerbated metabolic overload . ( A ) Fructose and mannose exacerbated the TGA phenotype in the S . uvarum gal80∆ gal80b∆ background , whereas glucose partially rescued the TGA phenotype . ( B ) The S . uvarum TGA phenotype in galactose and fructose or mannose can be rescued by the deletion of GAL1 . The average times to first doubling are shown for three biological replicates . The error bars represent standard deviations . S . uvarum gal80∆ gal80b∆ gal1∆ was significantly different than S . uvarum gal80∆ gal80b∆ in both SC + 1% galactose +1% fructose ( p=4 . 8e-3 , n = 6 , Wilcoxon rank sum test ) and SC + 1% galactose + 1% mannose ( p=2 . 9e-3 , n = 6 , Wilcoxon rank sum test ) . S . uvarum gal80∆ gal80b∆ gal1∆ was not significantly different from S . uvarum gal1∆ in SC + 1% galactose +1% fructose ( p=0 . 43 , n = 6 , Wilcoxon rank sum test ) but was marginally different from S . uvarum gal1∆ in SC + 1% galactose +1% mannose ( p=0 . 03 , n = 6 , Wilcoxon rank sum test ) . ( C ) Elevated accumulation of ROS in S . uvarum gal80∆ gal80b∆ in SC + 1% galactose +1% mannose . S . uvarum gal80∆ gal80b∆ had significantly higher ROS levels than the wild-type ( p=8 . 6e-6 , n = 11 , Wilcoxon rank sum test ) . ROS levels are reported as relative fluorescence levels . DOI: http://dx . doi . org/10 . 7554/eLife . 19027 . 020 We next considered whether the differences between the GAL networks of S . cerevisiae and S . uvarum might explain why a similar phenotype had not been reported for S . cerevisiae co-repressor mutants . Recent work has convincingly shown that the S . uvarum GAL network is more transcriptionally active than the S . cerevisiae GAL network , especially in non-inducing and mixed sugar conditions ( Caudy et al . , 2013; Roop et al . , 2016 ) . Thus , we wondered whether S . cerevisiae and S . uvarum galactose catabolism might be under qualitatively similar constraints , even as the more poised and active state of the S . uvarum GAL network might render it more vulnerable to metabolic overload . First , we examined S . cerevisiae gal80 null mutants more closely and found a similar but less-pronounced early rapid increase in optical density , followed by a brief but reproducible TGA phenotype ( Figure 4A , inset ) . This observation was missed by earlier studies , which were focused on later time points , because S . cerevisiae gal80 null mutants eventually grow much faster on galactose ( Torchia et al . , 1984; Segrè et al . , 2006; Hittinger et al . , 2010 ) . To test whether the weak TGA phenotype seen in S . cerevisiae was due to mechanistically similar metabolic constraints , we sought to exacerbate the phenotype of a S . cerevisiae gal80∆ strain in a mixture of mannose and galactose . Indeed , the co-repressor mutant produced significantly more ROS than wild-type under these conditions ( Figure 7A ) and grew slightly more slowly ( Figure 7B ) . 10 . 7554/eLife . 19027 . 021Figure 7 . The less active S . cerevisiae GAL1 gene is partially responsible for a subtle temporary growth arrest . ( A ) Elevated accumulation of ROS in S . cerevisiae gal80∆ in SC + 1% galactose +1% mannose . S . cerevisiae gal80∆ had significantly higher ROS than wild-type ( p=2 . 3e-6 , n = 12 , Wilcoxon rank sum test ) . ROS levels are reported as relative fluorescence levels . ( B ) S . cerevisiae gal80∆ grew more slowly when galactose was mixed with mannose . The average of three biological replicates from a representative experiment is shown , and the error bars represent standard deviations ( p=0 . 028 , n = 6 , Wilcoxon rank sum test ) . ( C ) Both the ScerGAL1 coding sequence and promoter are able to partially rescue the TGA phenotype . The error bars show the standard deviation of three biological replicates . ( D ) Both the ScerGAL1 coding sequence and promoter reduced the growth rate of an otherwise wild-type strain of S . uvarum in SC + 2% galactose , while the reciprocal swap of the GAL1 promoter in S . cerevisiae increased its growth rate . Wilcoxon rank sum tests comparing the specific growth rates of each subpanel were all significant: ( 1 ) p=2 . 3e-6 and n = 12 for S . uvarum gal1∆::PSuvaGAL1-ScerGAL1 versus S . uvarum wild-type , ( 2 ) p=2 . 5e-4 and n = 9 for S . uvarum gal1∆::PScerGAL1-SuvaGAL1 versus S . uvarum wild-type , and ( 3 ) p=8 . 8e-3 and n = 9 for S . cerevisiae gal1∆::PSuvaGAL1-ScerGAL1 versus S . cerevisiae wild-type . DOI: http://dx . doi . org/10 . 7554/eLife . 19027 . 021 Given the interspecific functional differences described above for GAL1 ( Figure 2C ) and its role as the gatekeeper of the Leloir pathway , we hypothesized that the varied strengths of the TGA phenotype might be due to genetic differences in the GAL1 locus . Thus , we precisely replaced the S . uvarum GAL1 promoter or the GAL1 coding sequence with their S . cerevisiae counterparts in S . uvarum gal80∆ gal80b∆ . The S . cerevisiae GAL1 promoter rescued the TGA phenotype to some extent , but the GAL1 coding sequence swap was able to rescue the TGA phenotype to an even greater extent ( Figure 7C ) . To confirm that ScerGAL1 was less active than SuvaGAL1 and not less toxic for other reasons , we examined the same precise allele replacements in an otherwise wild-type S . uvarum strain ( i . e . containing functional copies of both co-repressors ) , as well as a precise reciprocal swap in S . cerevisiae replacing the ScerGAL1 promoter with the SuvaGAL1 promoter . Swapping the ScerGAL1 promoter and coding sequence into S . uvarum both led to lower growth rates in galactose , while swapping the SuvaGAL1 promoter into S . cerevisiae led to faster growth ( Figure 7D ) . We conclude that the S . uvarum GAL1 promoter and coding sequences both encode higher activity than their S . cerevisiae counterparts . Thus , differences in their GAL network activities at least partly explain the relative strengths of their TGA phenotypes and the constraints placed on their galactose metabolisms . The deep conservation of metabolism and many molecular processes contrasts sharply with the rapid turnover in the regulatory networks that sculpt organismal and phenotypic diversity . Here we have shown how numerous genetic differences between the S . cerevisiae and S . uvarum GAL networks , especially in the functions of paralogous regulatory genes , contribute to a more poised and active state in S . uvarum that is coupled to more robust repression system . When genes encoding the co-repressors were deleted , S . uvarum displayed a strong and unexpected growth arrest in galactose , likely due to metabolic overload . Even though S . cerevisiae produced qualitatively similar results , decades of previous research on this iconic metabolic and regulatory network overlooked their relatively mild presentation . Just as exaggerated manifestations facilitated the discoveries of transposons in maize , RNAi in Caenorhabditis elegans , and telomeres in Tretrahymena ( Blackburn et al . , 2006 ) , the striking phenotype observed in the non-traditional model organism S . uvarum allowed us to more fully characterize the defect caused by overly rapid galactose catabolism , while demonstrating metabolic constraints conserved across sugars and organisms . In contrast to glucose , fructose and mannose each had strikingly deleterious effects on cells that were already consuming galactose too rapidly . In Saccharomyces , these differences can be explained both by their effects on signaling pathways and by their entry points into glycolysis . Several glucose signaling pathways directly repress GAL gene transcription ( Johnston et al . , 1994 ) and increase the degradation rate of Gal2 protein ( Horak and Wolf , 2001 ) , both of which would serve to reduce glycolytic load . In S . cerevisiae , fructose and mannose do not trigger glucose repression as strongly as glucose ( Dynesen et al . , 1998; Meijer et al . , 1998 ) . Perhaps as importantly , fructose and mannose bypass the trehalose cycle , a futile cycle recently shown to detour more than a quarter of early-stage glycolytic flux to prevent an unbalanced metabolic state and growth arrest ( van Heerden et al . , 2014 ) . The challenges of the catabolism of sugars other than glucose may be widespread . For example , in humans , bypassing glucose-responsive regulatory mechanisms with fructose has been associated with diabetes ( Lê et al . , 2009; Kolderup and Svihus , 2015 ) and cancer ( Port et al . , 2012; Jiang et al . , 2016 ) . The intrinsic constraints imposed by galactose metabolism may have led to the evolution of regulatory mechanisms that protect against the risks of metabolic overload . Many of the differences between the S . uvarum and S . cerevisiae GAL networks can be explained as offering alternative protective strategies , while affording varied catabolic capabilities . For instance , the direct regulation of the PGM1 gene by Gal4 would enhance the connection between the Leloir pathway and glycolysis in S . uvarum relative to S . cerevisiae ( Fu et al . , 2000; Ostergaard et al . , 2000; Garcia Sanchez et al . , 2010 ) . S . uvarum PGM1 is highly induced by galactose ( Figure 4—figure supplement 6 ) , but this likely ancestral regulatory connection was lost in the S . cerevisiae-S . kudriavzevii clade ( Figure 4—figure supplement 6 ) . Nearly all of the known differences between the S . cerevisiae and S . uvarum GAL networks make S . uvarum more active , including ( 1 ) apparent regulation of PGM1 by Gal4; ( 2 ) the presence of genes encoding two galactose transporters ( Figure 1 ) ; ( 3 ) the galactokinase activity of SuvaGal3 ( Figure 2A ) ; ( 4 ) the higher activity of both the GAL1 coding and cis-regulatory sequences ( Figure 7D ) ; and ( 5 ) higher background gene expression across the network ( Caudy et al . , 2013; Roop et al . , 2016 ) . Indeed , the possession of a gene encoding a second co-repressor appears to be one of the few features of the S . uvarum GAL network that would serve to counteract its higher activity . Thus , the dramatic up-regulation of GAL80B during induction may offer a robust negative feedback loop that helps prevent over-induction and metabolic overload . The retention of GAL80B may have allowed S . uvarum to maintain a more active GAL network , while the S . cerevisiae GAL network evolved lower activity . Comparison of yeast genomes beyond the Saccharomyces suggests that galactose metabolism may impose similar constraints across the yeast phylogeny . The genes encoding the Leloir enzymes occur in one of the few broadly conserved yeast gene clusters ( Wong and Wolfe , 2005; Slot and Rokas , 2010; Wolfe et al . , 2015; Riley et al . , 2016 ) , which has been suggested could promote enzyme co-regulation to prevent the accumulation of toxic intermediates ( Price et al . , 2005; Lang and Botstein , 2011 ) or ensure that only complete networks are co-inherited ( Lawrence and Roth , 1996; Hittinger et al . , 2010 ) . In addition to S . uvarum , many yeast species that underwent the WGD retain GAL80B ( Hittinger et al . , 2004 ) . Perhaps due to these intrinsic metabolic challenges and the limited benefits of maintaining a dedicated GAL network , the ability to consume galactose has been lost many times across diverse yeast lineages ( Hittinger et al . , 2004 , 2015; Slot and Rokas , 2010; Wolfe et al . , 2015; Riley et al . , 2016 ) . In contrast to more commonly studied processes of the rapid neofunctionalization and subfunctionalization of paralogs ( Moore and Purugganan , 2003; Innan and Kondrashov , 2010 ) , we have shown how duplicate GAL genes continued to diverge functionally in ways that dramatically influenced the metabolic and regulatory states of extant Saccharomyces species . Based on the redundancy observed between GAL1 and GAL3 and between GAL80 and GAL80B in S . uvarum , we infer that the functions of these two paralog pairs overlapped more at the origin of the genus Saccharomyces than in S . cerevisiae ( Figure 8 ) . After the S . uvarum-S . eubayanus clade diverged from the S . arboricola-S . cerevisiae clade , these genes met distinct fates in different lineages ( Figure 8 ) . GAL80B was lost in the S . cerevisiae-S . arboricola clade , while it was retained in S . uvarum and S . eubayanus ( Hittinger et al . , 2010 , 2004; Scannell et al . , 2011; Caudy et al . , 2013; Hittinger , 2013; Liti et al . , 2013; Baker et al . , 2015 ) . The fates of GAL1 and GAL3 were still more varied . GAL3 was lost in a European population of S . kudriavzevii , resulting in an induction defect , while the entire GAL network was lost in an East Asian population of this species that cannot consume galactose ( Hittinger et al . , 2010 ) . GAL1 and GAL3 were nearly completely subfunctionalized in S . cerevisiae ( Hittinger and Carroll , 2007 ) , but we have shown here that they maintain considerable redundancy in S . uvarum . 10 . 7554/eLife . 19027 . 022Figure 8 . Ongoing diversification of the functions of the GAL1-GAL3 and GAL80-GAL80B duplicate gene pairs in Saccharomyces . Important evolutionary events are shown on the cladogram . WGD , the whole genome duplication that created the two pairs of paralogs . The inferred duplicate divergence fates are shown at the bottom of the tree . The inferred timeline is depicted by the dashed line . Roughly ~100 million years ago , these two pairs of duplicate genes were fixed in the ancestral genome following a WGD event . Considerable partial redundancy was maintained in the lineage leading to the origin of the genus Saccharomyces . In the last ~10 million years , the fates of the duplicate genes have functionally diverged along different evolutionary trajectories . The bifunctionality of the GAL1/GAL3 genes is represented by green for the enzymatic function and blue for the co-induction function . The color shading represents approximate functionality for experimentally characterized genes: a darker color indicates a stronger function , whereas a lighter color indicates a weaker function . The dashed circle with a cross indicates the loss of the indicated gene . Note that the S . kudriavzevii Asian population lost its entire GAL network , while the European population retained most of its GAL network but lost GAL80B and GAL3 . The additional co-repressor in S . uvarum may minimize the risk of metabolic overload due to an otherwise highly active GAL network . DOI: http://dx . doi . org/10 . 7554/eLife . 19027 . 022 For both paralog pairs , the ongoing functional diversification has been asymmetric . Deleting GAL80B and GAL3 produced less striking phenotypes than deletion of their paralogs in S . uvarum , and some lineages have experienced inactivation or loss of these genes naturally . In the lineage leading to S . cerevisiae , Gal3 completely lost enzymatic activity , while a decrease in the promoter activity of its paralog GAL1 reduced , but did not eliminate , its ability to induce the network rapidly . Other GAL genes also experienced an adaptive decrease in promoter activities in the lineage leading to S . cerevisiae ( Roop et al . , 2016 ) , which may have been enabled or necessitated by the loss of the secondary co-repressor encoded by GAL80B . Remarkably , the disparate resolutions of the functions of these paralogs did not happen soon after the WGD that created them . Instead , the diversification described here occurred within the last 10 million years of a 100 million year history , demonstrating that the echoes of duplication events continue to resonate through gene networks much longer than is generally appreciated ( Gordon et al . , 2009; Conant et al . , 2014 ) . The ongoing functional diversification of ancient paralogs likely has an even greater impact on the evolution of plants and vertebrates , where nearly all extant species are the products of multiple rounds of WGD , and differential paralog retention is widespread ( Amores et al . , 1998 , 2004; Aury et al . , 2006; Blomme et al . , 2006; Gómez-Valero et al . , 2007; Scannell et al . , 2007; De Smet et al . , 2013; McGrath et al . , 2014 ) . Molecular and genetic dissection is much more challenging in these systems , but there are hints that the diversification of ancient paralogs continues to have functional consequences for the evolution of metabolism ( Steinke et al . , 2006; Conant et al . , 2014 ) and development ( Kassahn et al . , 2009; Cortesi et al . , 2015 ) . Paralog diversification is often asymmetric as one paralog acquires a more specialized or auxiliary role ( Force et al . , 1999; Moore and Purugganan , 2003; Hittinger and Carroll , 2007; Des Marais and Rausher , 2008; Innan and Kondrashov , 2010; Conant et al . , 2014 ) . Even if this specialization is conditionally adaptive , the auxiliary paralog can become more susceptible to gene loss when conditions change . Paralog loss ends the saga of duplicate gene diversification , possibly forcing partially redundant functions back onto the remaining paralog , relieving paralog interference ( Baker et al . , 2013 ) , or leading to compensatory changes elsewhere in the network . Perhaps more interestingly , paralog loss eliminates redundancy and limits the long-term potential for adaptation . The ongoing evolutionary processes affecting the GAL paralogs show how gene duplication facilitates phenotypic change and network diversification in ways that continue to reverberate . To construct GAL gene knockouts , we used MX cassettes ( hphMX , natMX , or kanMX ) ( Wach et al . , 1994; Goldstein and McCusker , 1999 ) to precisely replace the coding sequence from start codon to stop codon . Transformations were based on the standard lithium acetate/PEG method optimized for S . uvarum ( room temperature incubation , followed by a 37˚C heat shock ) ( Gietz et al . , 1995; Caudy et al . , 2013 ) . To perform allele swaps , the coding sequence or promoter was first replaced by a selectable and counter-selectable TK-hphMX cassette , which does not require the prior introduction of an auxotrophy ( Alexander et al . , 2014 ) . The coding sequence or promoter of the desired replacement sequence was amplified by PCR primers with overhangs homologous to the targeted genomic flanking region . In some cases , extended homology ( 100–300 bp ) was then introduced through PCR sewing . For each GAL1 promoter swap , we swapped the entire upstream intergenic region . Note that the S . cerevisiae and S . uvarum GAL1 promoters are both divergent promoters that also regulate GAL10 and may also impact a lncRNA previously described in S . cerevisiae ( Cloutier et al . , 2016 ) . Successful replacement strains were isolated by selecting for the loss of thymidine kinase activity by resistance to 5-fluorodeoxyuridine ( FUdR ) , as well as the loss of resistance to hygromycin by replica plating ( Alexander et al . , 2014 ) . GFP reporters were constructed in three parts: the hphMX cassette was placed upstream as the selection marker , the S . uvarum GAL1 promoter was used to drive the expression of the reporter , and the reporter was a yEGFP ( yeast Enhanced Green Fluorescence Protein ) construct with a S . cerevisiae CYC1 terminator that was amplified from FM1282 ( Hittinger and Carroll , 2007; Hittinger et al . , 2010 ) . GFP reporters were introduced to replace S . uvarum gto1 , an inactive pseudogene ( chr7: 767 , 328–766 , 478 ) orthologous to S . cerevisiae GTO1 ( Scannell et al . , 2011 ) . The modified loci of all transformants were verified by Sanger sequencing . S . cerevisiae is NCBITaxon:4932 , S . uvarum is NCBITaxon:230603 , and the strains used in this study are listed in Supplementary file 2 . Strains were first streaked on YPD ( 10 g/L yeast extract , 20 g/L peptone , 20 g/L glucose , 18 g/L agar ) plates from frozen glycerol stocks . Next , a single colony of each strain was cultured in synthetic complete ( SC ) medium plus 0 . 2% glucose ( 1 . 72 g/L yeast nitrogen base without amino acids , 5 g/L ammonium sulfate , 2 g/L complete dropout mix , 2 g/L glucose ) for 2–3 days , a condition that does not induce and only minimally represses the GAL network . There were at least two biological replicates for each genotype , generally from independent transformants . These pre-cultures were washed with water and inoculated into the desired growth media in a 96 well plate . No explicit power analyses were performed to determine sample sizes or the number of replicates . Instead , each experiment was independently performed at least twice on separate days; details can be found in each legend . Biological replicates were defined as independent isogenic colonies on agar plates , which were used for subsequent precultures and growth assays; technical replicates were defined as independent growth assays from the same preculture . The absorbance of each well was read by an unshaken BMG FLUOstar Omega plate reader every 10 min at 595 nm . The number of cell divisions for each time point was calculated as log2[ ( ODstrain−ODmedia ) / ( ODstart−ODmedia ) , an equation that normalized each optical density time point to its starting optical density and the optical density of the medium . The times to first doubling were calculated as the times for the optical densities to double from their normalized starting points . Specific growth rates were calculated using the Growth Curve Analysis Tool ( GCAT ) ( Bukhman et al . , 2015 ) . Replicates that failed to grow as precultures or during growth assays were considered as outliers and were excluded from subsequent analyses; no other data were excluded . For S . cerevisiae and S . uvarum gal1 mutant growth assays ( Figure 2—figure supplement 1 ) , strains were pre-cultured in SC plus 0 . 67% fructose for 2 days and inoculated at a 1:1000 ratio into supplemented minimal medium ( 1 . 72 g/L yeast nitrogen base without amino acids , 5 g/L ammonium sulfate , 85 . 6 mg/L uracil , 85 . 6 mg/L lysine , 20 g/L galactose ) plus 2% galactose or no carbon source . The growth properties of these strains were determined by subtracting the optical densities of cultures in media without a carbon source from media with galactose; differences less than 0 . 05 were considered as 'no growth . ' In each case , S . cerevisiae strains were cultured at 30˚C , while S . uvarum strains were cultured at 24˚C , except when they were cultured in the same 96 well plate . In these cases ( Figure 2C , Figure 2B and Figure 4A ) , strains were grown at 26˚C , and the results were summarized in one graph . The pre-culture and growth conditions were identical to those described above for the 96-well growth assays . At the indicated time points , 1–30 μL cultures were transferred from the 96-well plate to fresh medium of the same type in a new 96-well plate to obtain a concentration of 200–500 cells/μl for flow cytometry . There were at least three biological replicates for each genotype . The flow cytometry was conducted using a Guava EasyCyte Plus flow cytometer . Each experiment was independently conducted at least twice on separate days . The data were extracted from FCS 2 . 0 formatted files using FlowCore ( Hahne et al . , 2009 ) ( RRID:SCR_002205 ) . The fluorescence levels were normalized by forward scatter to control for cell size . For each genotype , histograms of normalized fluorescence levels of 6000 cells were smoothed by Kernel density estimation and plotted using the R statistical package . Strains were pre-cultured in SC plus 0 . 2% glucose for 2 days and inoculated into SC plus 2% galactose , 2% glucose , or 5% glycerol . Samples were harvested at the indicated time points and frozen using a dry ice/ethanol bath . Total RNA was extracted using the standard acidic phenol protocol ( Hittinger and Carroll , 2007 ) , and residual DNA was removed through DNase I treatment . Poly-A enrichment was performed with the NEBNext Poly ( A ) mRNA Magnetic Isolation Module ( NEB #E7490 , in the experiment to examine S . uvarum GAL network membership ) or with the NEB Magnetic mRNA Isolation kit ( NEB #S1550 , in the experiment sampled during the TGA phase and at mid-log phase in galactose ) . Illumina libraries were constructed using the NEB Ultra Directional RNA Library Prep Kit for Illumina ( NEB #E7420 ) and sequenced using the Illumina HiSeq 2500 platform . Reads were mapped onto the S . uvarum reference genome ( CBS7001 ) ( Scannell et al . , 2011 ) using Bowtie version 2 . 2 . 2 with local read alignment and otherwise default settings ( Langmead et al . , 2009 ) . Read counts were quantified by HTSeq version 0 . 6 . 0 ( Anders et al . , 2015 ) ( RRID:SCR_005514 ) . Differential expression was determined using EBseq version 1 . 1 . 5 with a false discovery rate ( FDR ) of 0 . 05 ( Leng et al . , 2013 ) ( RRID:SCR_003526 ) . Analysis with edgeR ( RRID:SCR_012802 ) using the default settings was performed in parallel to examine known S . cerevisiae Gal4 targets that were not scored as differentially expressed in S . uvarum ( Robinson et al . , 2010 ) . Differentially expressed genes were further analyzed by GO term analysis ( Ashburner et al . , 2000; Cherry et al . , 2012 ) ( Generic GO Term Mapper , RRID:SCR_005806; SGD Gene Ontology Slim Mapper , RRID:SCR_005784 ) . The RNA-Seq data are available at NCBI's SRA under accession number SRP077015 . The pre-culture conditions were identical to those described above for the growth assays . The ROS measurement protocol was adapted from a previous study ( Dudgeon et al . , 2008 ) . A 10 mM stock of H2DCFDA ( 2' , 7'-dichlorodihydrofluorescein diacetate ) was freshly prepared in ethanol before each use . Cells were washed once and inoculated into the stated growth medium with 10 µM H2DCFDA . Cultures were harvested at the indicated time points . Fluorescence levels and optical densities were measured using a BMG FLUOstar Omega plate reader , which can read both fluorescence and optical density . To establish standard curves , a 2-fold serial dilution for each strain was measured . Since the standard curves suggested a linear relationship between fluorescence levels and cell number , fluorescence levels were normalized to optical densities . The S . uvarum wild-type strain was inoculated into YPD plus 10 mM H2O2 and into YPD only as positive and negative controls , respectively . Each experiment was independently conducted at least twice on separate days with at least three biological replicates in each experiment . The 13C yeast metabolome extract ( Bennett et al . , 2008 ) was prepared by growing Y22-3 ( McIlwain et al . , 2016 ) aerobically on YNB ( -AA ) + 1% 13C glucose . Yeast cultures were inoculated at an OD of 0 . 05 into 13C medium . Samples were harvested from each culture by centrifugation and frozen in liquid N2 . Frozen pellets were first extracted with 750 µL of 40:40:20 ACN/MeOH/H2O , followed by a second extraction with 500 µL of the same extraction solvent . Extracts were pooled , centrifuged , and the supernatant was collected for later use as an internal standard for absolute metabolite quantification ( Bennett et al . , 2008 ) . Lyophilized cell culture metabolites were extracted from mutant and wild-type strains with 5 mL ice-cold 7:2:1 MeOH/CHCl3/H2O , and 100 µL of the extract was mixed with 10 µL 13C-labelled yeast metabolome extract . Three biological replicates were included for the S . uvarum strains ( Figure 5 ) , while two were included for the S . cerevisiae strains ( Figure 5—figure supplement 1 ) . Chromatographic separations based on a previously described method ( van Dam et al . , 2002; Long et al . , 2012 ) were carried out on an Agilent 1200 series HPLC comprising a vacuum degasser , binary pump , heated column compartment , and thermostated autosampler set to maintain 6˚C . Mobile phase A ( MPA ) was 0 . 5 mM NaOH , and mobile phase B ( MPB ) was 100 mM NaOH . 20 μL of intracellular extract or calibrant standard mixture was separated on a Dionex IonPac AS11-HC IC column ( 2 . 0 mm x 250 mm , 9 . 0 µm ) held at 40°C using a flow rate of 0 . 35 mL/min . Metabolite elution was achieved by first holding at 5% MPB for 22 . 5 min to separate isobaric phosphosugar species . MPB was then linearly increased from 5% to 100% over 27 . 5 min to elute the remaining metabolites . MPB was held at 100% for 7 min for column cleaning followed by an 8 min re-equilibration step at 5% MPB . The LC system was coupled to a Dionex ERS 500 suppressor controlled by a Dionex Reagent-Free Controller ( model RFC-10 ) and an Agilent 6460 A Triple Quadrupole MS . The MS was operated in negative mode , acquiring MRM scans for each metabolite . Quantification based off external standard calibration curves and correction with the 13C-labelled yeast standard was performed with Agilent MassHunter Quantitative Analysis software ( version B . 06 . 00 ) . The pre-culture conditions were identical to those described above for the growth assays . At indicated time points , 1 mL of cells were centrifuged , and 500 μL supernatant was harvested and frozen at −80˚C . HPLC was conducted at the GLBRC Metabolomics Lab using an HPLC-RID system with an Aminex HPX-87H ( BioRad , Inc . Hercules , CA ) following previously described protocols ( Moore and Johnson , 1967; Ehrman and Himmel , 1994 ) . Instrument control , data collection and analyses were conducted using ChemStation B . 04 . 03 software ( Agilent Technologies , Inc . , Palo Alto , CA ) . All p-values , except for the RNA-Seq , metabolomics ( two-sided student’s t-test ) , and HPLC analyses ( two-sided student’s t-test ) , were calculated using a conservative two-sided nonparametric test . Specifically , we used a Wilcoxon rank sum test that allows the rank data from multiple independent experiments to be pooled to account for day-to-day variation without making assumptions about the distribution of the variance . These tests were performed using Mstat software version 6 . 1 . 4 ( http://mcardle . oncology . wisc . edu/mstat/ ) .	Genetic information is organized into units called genes , which encode sets of instructions needed to make proteins and other molecules in cells . When an organism reproduces , it passes on some or all of its genes to its offspring . Over many generations , individual genes may acquire changes known as mutations . Some mutations may improve the ability of the individual to survive and reproduce , but most are harmful and may even lead to death . Organisms can bypass these constraints by creating extra copies of genes so that if one copy acquires a harmful mutation the other copy still works normally . Duplicate genes are crucial to evolution because they offer opportunities to evolve new characteristics , but most duplicated genes are quickly inactivated , in part because the organism does not need them . Decades of research have focused on how some duplicate genes manage to remain active , often by specializing or acquiring new roles . However , relatively little attention has been paid to what happens to duplicates after they have been retained . Baker’s yeast – also known as Saccharomyces cerevisiae – is a single-celled organism that is often used to study genetics . Kuang et al . investigated whether yeast genes that were duplicated millions of years ago are still actively evolving and whether they have achieved different evolutionary outcomes in baker’s yeast and a closely related yeast called Saccharomyces uvarum . The experiments examined two pairs of genes that help to break down a sugar called galactose . Even though these genes were duplicated around 100 million years ago , they have continued to evolve in these yeasts . In the last ten million years , these duplicate genes have taken on different roles so that baker’s yeast and S . uvarum have evolved different strategies for consuming galactose . S . uvarum breaks down galactose more aggressively than baker's yeast , but it also has a better system in place to prevent it from breaking down more galactose than it needs . The findings of Kuang et al . suggest that the new roles that duplicated genes adopt in organisms might have a bigger effect on the evolution of organisms in the long run than is currently appreciated . Future work will test this idea by studying other duplicated genes in different species .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "evolutionary", "biology" ]	2016	Ongoing resolution of duplicate gene functions shapes the diversification of a metabolic network
Most cellular features have a range of states , but understanding the mechanisms responsible for interspecific divergence is a challenge for evolutionary cell biology . Models are developed for the distribution of mean phenotypes likely to evolve under the joint forces of mutation and genetic drift in the face of constant selection pressures . Mean phenotypes will deviate from optimal states to a degree depending on the effective population size , potentially leading to substantial divergence in the absence of diversifying selection . The steady-state distribution for the mean can even be bimodal , with one domain being largely driven by selection and the other by mutation pressure , leading to the illusion of phenotypic shifts being induced by movement among alternative adaptive domains . These results raise questions as to whether lineage-specific selective pressures are necessary to account for interspecific divergence , providing a possible platform for the establishment of null models for the evolution of cell-biological traits . As with nearly all biological traits , most cellular features vary among individuals within populations in a nearly continuous fashion , owing to genetic differences among individuals and the myriad of stochastic factors experienced by all organisms ( ranging from intrinsic cellular noise to external environmental forces; Lynch and Walsh , 1998 ) . This is true , for example , for catalytic rates , rates of gene expression and intracellular transport , numbers and sizes of organelles , etc . Ultimately , some fraction of within-species genetic variation is transformed into among-species divergence as alternative alleles arise by mutation and in some cases proceed to fixation ( Wright , 1969; Walsh and Lynch , 2018 ) . The magnitude of such divergence is dictated by three major evolutionary factors: the pattern of selection ( the phenotypic fitness function ) , which imposes a directional and/or stabilizing force on the mean phenotype; the rate of origin and distribution of mutational effects , which define the raw materials upon which natural selection operates; and the power of random genetic drift , which imposes noise on the selective process . Although considerable effort has been devoted to understanding the divergence of mean phenotypes among lineages ( Walsh and Lynch , 2018 ) , most of this work is focused on the evolution of morphological phenotypes in response to external pressures , which can vary greatly depending on the ecological setting . In contrast , owing to homeostatic effects , the internal environment of cells remains largely constant over long time scales and broad geographic locations , raising the possibility of establishing general evolutionary principles that transcend the imposition of transient ecological changes . ( The same might be true for the internal organs of multicellular species ) . The goal here is to derive general expressions for the divergence of mean phenotypes among species under scenarios that are likely to hold for a wide variety of cellular traits . The specific focus will be on the magnitude of divergence expected among lineages in the face of identical evolutionary forces , as this helps clarify the degree to which phenotypic diversification can proceed in the absence of lineage-specific selection pressures . Such a perspective is essential to establishing the degree to which adaptive explanations need to be sought to explain patterns of variation among populations . The general approach will draw from well-established constructs employed in the field of quantitative genetics ( the study of continuously distributed traits with a multifactorial genetic basis; Lynch and Walsh , 1998; Walsh and Lynch , 2018 ) . The traditional focus of this field has been on complex traits in multicellular species , but these same methods can be profitably applied to intracellular morphological and molecular features , such as those involved in the cytoskeleton , gene expression , binding energy , and metabolic rates ( Nourmohammad et al . , 2013; Farhadifar et al . , 2015; Phillips and Bowerman , 2015 ) . Indeed , although most work in phenotypic evolution proceeds as though cellular details are irrelevant , the models employed may be equally if not more relevant to cell-biological traits , owing to their potentially less temporally variable fitness effects . All genetically encoded traits are subject to the recurrent forces of mutation and random genetic drift , and potentially to selection . Selection favors some genotypes over others , while mutation modifies existing genotypes independent of the selective process , and random genetic drift causes stochastic variation in gene transmission across generations . Owing to this latter factor , even if the forces of selection and mutation remain constant , the population mean phenotype of a trait will wander within a certain range over evolutionary time , with the frequency of occurrence of alternative mean phenotypes depending on patterns and strengths of selective and mutational effects ( Figure 1 ) . The focus of this study , the stationary distribution of mean phenotypes , can be viewed as a summary distribution of: ( 1 ) phenotypic means across a large number of replicate populations exposed to identical conditions for a very long period; or ( 2 ) a historical survey of mean phenotypes in a single population over a long time period , again under constant environmental and population-genetic conditions . Among many other applications , such an approach has long been exploited in attempts to understand the steady-state distribution of allele frequencies expected under a constant regime of selection , mutation , and random genetic drift ( e . g . Wright , 1969 ) . From an empirical perspective , this steady-state view of evolution implicitly assumes that enough time has elapsed between observed taxa that the dynamics of the evolutionary process are of negligible significance ( which would not be the case for closely related species ) . The approach taken here relies on the Kolmogorov forward equation for a diffusion process ( Appendix 1 , Walsh and Lynch , 2018 ) , the assumption being that the trait of interest is continuously distributed , with z denoting the phenotypic value of an individual . The population mean , z¯ , moves in arbitrarily small increments each generation via the deterministic forces of selection and mutation and the stochastic process of drift . Under most reasonable biological conditions , independent of the starting conditions , a stationary distribution of mean phenotypes ( among hypothetical replicate populations ) is eventually converged upon , at which point there is an exact balance between opposing forces . The probability that a population’s mean phenotype will reside at any particular point is defined by this distribution , which has the general form ( 1a ) Φ ( z¯ ) =C⋅exp⁡ ( 2∫z¯[M ( x ) /V ( x ) ]dx ) , where M ( x ) defines the rate of directional change ( resulting from selection and/or mutation ) for a population with mean phenotype x , and V ( x ) is the variance in change ( resulting from drift ) . C is the normalization constant ( containing only terms that are independent of z¯ ) that ensures that the entire probability density sums to 1 . 0 . For a quantitative trait , the directional term can be subdivided into independent selection and mutation components , Ms ( x ) and Mm ( x ) , both of which will be discussed in detail below . Under the assumption of negligible genotype × environment interaction and epistasis , the variance of the change in means , which results from the sampling of heritable genotypic values of individuals , is equal to the underlying additive genetic variance for the trait , σA2 , divided by the effective population size , Ne , in the case of haploidy ( assumed here; and 2Ne in the case of diploidy ) . The latter is typically far below the number of reproductive individuals in the population , and defined by various demographic features and interference imposed by chromosomal linkage , with values ranging between ∼105 for multicellular eukaryotes to ∼109 for bacteria ( Charlesworth , 2009; Lynch et al . , 2016; Walsh and Lynch , 2018 ) . Individual phenotypes are comprised of the sum of a heritable additive genetic component ( A ) and a nonheritable residual deviation ( e , which includes environmental and nonadditive genetic effects ) , such that z=A+e , with the within-population phenotypic variance being partitioned as σz2=σA2+σe2 . For cellular features , a large fraction of σe2 may be a consequence of stochastic gene expression , imprecise placement of cell-division septa , etc . Assuming that both σA2 and Ne remain constant , which is the model adhered to here , Equation ( 1a ) can be rewritten as ( 1b ) Φ ( z¯ ) =C⋅exp ( ( 2Ne/σA2 ) ∫z¯[Ms ( x ) +Mm ( x ) ]dx ) , showing that the stationary distribution of mean phenotypes ( conditional on a particular level of genetic variance , a point that will be returned to below ) is proportional to the product of the distributions expected under selection alone and under mutation alone . With extremely weak selection , Ms ( x ) would be essentially a flat function , with the overall distribution reflecting the biases due to mutation alone . Conversely , with a flat mutation function , an unlikely scenario , the distribution will follow that expected under selection alone . The influence of selection on the mean phenotype ( the response to selection ) is embodied in the breeder’s equation , ( 2 ) Ms ( z¯ ) =z¯ ( t+1 ) −z¯ ( t ) =h2[ z¯s ( t ) −z¯ ( t ) ] , a general statement about the connection between directional selection within generations and the transmission of such change across generations ( Walsh and Lynch , 2018 ) . Here , z¯ ( t ) and z¯s ( t ) denote the mean phenotypes before and after selection in generation t , the difference being the selection differential . The heritability of the trait , h2=σA2/σz2 , which equals the proportion of the total phenotypic variance , σz2 , associated with additive genetic variation , σA2 , constitutes the fraction of the within-generation change in the mean transmitted to the next generation . Critical to everything that follows , the selection differential can be described in terms of the within-population phenotype distribution , p ( z , t ) , and the function relating individual fitness to phenotype , W ( z ) . The mean fitness in generation t is ( 3 ) W¯=∫p ( z , t ) ⋅W ( z ) ⋅dz . The mean phenotype after selection ( but before inheritance ) is then obtained by weighting the pre-selection phenotypes by their relative fitnesses , ( 4 ) z¯s ( t ) =1W¯∫z⋅p ( z , t ) ⋅W ( z ) ⋅dz . We will make use of the fact that most quantitative traits have an approximately normal phenotype distribution on some scale of measurement , which follows from the central limit theorem ( Lynch and Walsh , 1998 ) . The distribution of individual measures is therefore described completely by the phenotypic mean and variance , ( 5 ) p ( z , t ) =12πσz2⋅exp⁡ ( −[ z−z¯ ( t ) ]22σz2 ) . Substituting Equation ( 5 ) into ( 3 ) and differentiating , the change in mean fitness with respect to mean phenotype is ( 6 ) ∂W¯∂z¯ ( t ) =∫∂p ( z , t ) ∂z¯ ( t ) ⋅W ( z ) ⋅dz=1σz2∫[ z−z¯ ( t ) ]⋅p ( z , t ) ⋅W ( z ) ⋅dz ( Lande , 1976 ) . From Equation ( 4 ) , the first term to the right of the integral is equal to z¯s ( t ) ⋅W¯ , and the second term is z¯ ( t ) ⋅W¯ . This provides a direct link to Equation ( 2 ) , which upon rearrangement becomes ( 7 ) Ms ( z¯ ) =σA2⋅∂W¯W¯⋅∂z¯ ( t ) . This expression states that , provided the phenotype distribution is normal , the change in mean phenotype caused by selection is equal to the product of the genetic variance for the trait and the gradient in the logarithm of mean fitness with respect to mean phenotype . Evolution by natural selection comes to a standstill when there is no genetic variance for the trait or the phenotypic mean resides at a point where the slope of the function of mean fitness with respect to mean phenotype is zero . To endow this expression with practical utility , specific expressions for the fitness function , W ( z ) , will be considered below . Most attempts to consider the long-term evolutionary features of quantitative traits have assumed one of two mutation models: ( 1 ) a distribution of mutational effects always having a mean equal to zero and a constant variance , independent of the starting genotype ( Kimura , 1965; Lande , 1975; Lynch and Hill , 1986 ) ; or ( 2 ) a rate of appearance of each type of mutant allele being independent of the ancestral type ( Cockerham , 1984; Turelli , 1984 ) . Under the first scenario , mutation has no directional effect on the mean phenotype , and there are no bounds on the possible mutational effects or the physical limits to which the trait can evolve . Under the second scenario , there is a physical limit to phenotypic divergence , and because the directional effect of mutations depends on the current location , more extreme alleles generate mutations with effects biased back toward the center of the distribution . Neither of these mutational schemes captures the features of a wide variety of cell biological traits , which often have finite numbers of possible states and state-dependent spectra of mutational effects . A few examples will suffice to make this point . Protein-protein interactions ( e . g . the interfaces between dimeric molecules ) typically depend on no more than a few dozen amino-acid sites . The same is true for intramolecular interactions such as the constellation of backbone residues that assemble during protein folding . In both cases , the underlying residues operate in an approximately binary manner , for example , hydrophobic vs . hydrophilic , or hydrogen-bonding vs . non-hydrogen bonding . Likewise , the catalytic sites of enzymes often consist of a small-to-moderate numbers of residues that either facilitate or inhibit catalytic rates , and the sizes of intracellular organelles and cytoskeletal components are constrained by cell size . Many other examples could be cited , including those involved in RNA-RNA and DNA-protein interactions . The approximate structure of a mutation function with a bounded range can be arrived at by considering a trait determined by n binary factors ( or sites ) , each with state b having effect 0 , and state B having effect m . For a trait with an additive genetic basis , the mean phenotype in a haploid population can then be represented as ( 8 ) z¯=z0+nmq¯ , where z0 is an arbitrary baseline value for the trait , and q¯ is the mean frequency of B-type alleles averaged over all n factors in the population ( Lynch and Walsh , 1998 ) . Letting u be the mutation rate from B to b alleles , and v be the reciprocal rate , the per-generation change in the mean phenotype resulting from mutation is ( 9 ) Mm ( z¯ ) =nm[v ( 1−q¯ ) −uq¯] . With q^=v/ ( u+v ) being the equilibrium frequency of B alleles under mutation pressure alone , and θm=z0+nmq^ being the expected mean phenotype under neutrality , Equation ( 9 ) further reduces to ( 10 ) Mm ( z¯ ) =− ( u+v ) ( z¯−θm ) . This expression is quite general in that ( z¯−θm ) is simply the distance of the mean phenotype from that expected under mutation equilibrium , and ( u+v ) is a measure of the mutational restoring force per locus . The essential feature of Equation ( 10 ) is that mutation acts to reduce the distance between the mean phenotype and θm to a degree that depends on the magnitude of this deviation . Charlesworth ( 2013 ) implemented a similar mutation model in an investigation of genomic features . Application of Equations ( 7 ) and ( 10 ) to ( 1b ) yields a useful simplification of the stationary distribution that will be adhered to below , ( 11 ) Φ ( z¯ ) =C⋅[ W¯ ( z¯ ) ]2Ne⋅exp⁡ ( − ( z¯−θm ) 22σN2 ) , with σN2=σA2/[2Ne ( u+v ) ] . As will be discussed below , under neutrality , the genetic variance σA2 often scales directly with Ne , and population size would have no influence on the distribution in this limiting case , as σN2 would be independent of Ne . More generally , σA2 is also a function of the intensity of selection , but the bulk of the steady-state distribution will be represented by mean phenotypes that are in the range of effective neutrality with respect to each other , so the scaling relationship of σA2 under neutrality is expected to be a reasonable first-order approximation . Equation ( 11 ) shows that , provided the genetic variance remains roughly constant , the stationary distribution is equal to the product of the expectation under neutrality ( where mutation and drift are the only operable evolutionary forces ) and the mean fitness function exponentiated by 2Ne , that is , the stationary distribution is equivalent to a transformation of the neutral expectation by a function of the fitness landscape . Thus , to obtain the overall distribution in the following applications , we require an expression for mean population fitness in terms of the trait mean . In what follows , insight into the approximate magnitude of σN2 will be useful . This can be achieved by noting that 2Ne ( u+v ) will have values of the order of magnitude of 4Neμ , where μ is the mutation rate per nucleotide site . This composite parameter is equivalent to the amount of standing heterozygosity at neutral nucleotide sites in natural populations under mutation-drift equilibrium , and generally ranges from 0 . 001 to 0 . 1 , with the lower and higher ends of the range being typical in vertebrates and microbes , respectively ( Lynch , 2007 ) . Thus , because heritabilities ( σA2/σz2 ) of traits are typically on the order of 0 . 1 to 0 . 5 ( Lynch and Walsh , 1998 ) , σN2 is expected to be in the range of 1× to 100× the average within-population phenotypic variance for the trait . The preceding models are meant to provide heuristic guidance into the evolutionary mechanisms responsible for the dispersion of mean phenotypes of a diversity of subcellular and molecular features . Although such traits may sometimes be under selection for an intermediate optimum , selection may often operate in a continuous directional fashion . In either case , there are two reasons why mean phenotypes are unlikely to commonly achieve states that endow a population with maximum fitness . First , if mutation bias conflicts with the directional effects of selection , the optimum phenotype will not coincide with the mean phenotype . Second , even in the absence of mutation bias and regardless of the form of the fitness function , a drift barrier exists beyond which the gradient of the selection function is not steep enough to overcome the vagaries of genetic drift , thereby preventing further adaptive progress . Within the confines of the drift barrier , the mean phenotype will wander to a degree that depends on the strength of local patterns of mutation and selection . These points have implications for the degree to which the ‘adaptive paradigm’ should be embraced as an explanatory framework for diversification at the cellular level . For example , with mutation bias encouraging the mean phenotype to deviate from the optimum , the result will be a population under persistent directional selection despite the existence of an attainable ( but not sustainable ) phenotype with maximum fitness . Even without mutation pressure and in the face of intrinsic directional selection , for example , a hyperbolic or mesa fitness function , the most common mean phenotype will not be equivalent to the optimum phenotype , and the drift barrier will ensure variation in mean phenotypes among populations exposed to identical selection pressures . An attempt has been made to couch the stationary forms of mean-phenotype distributions in terms of underlying parameters that are at least in principle observable empirically . Consider , for example , the model for stabilizing selection for a specific optimum . From Equation ( 14a ) , the expected deviation of the mean phenotype from the optimum resulting from mutation bias is θm/κ , which expands to θm[ 2 ( u+v ) ( ω2+σz2 ) /σA2 ] , a somewhat complex function that may not be immediately transparent . However , a wide variety of models suggest that σA2 scales directly with Neμ provided selection is weak ( Bürger et al . , 1989; Zeng and Cockerham , 1993; Charlesworth , 2013 ) , and because u and v ( the forward and reverse mutation rates ) are both proportional to μ ( the total mutation rate per site ) , this implies that the average deviation of the mean from the optimum scales as θm ( ω2+σz2 ) /Ne , or approximately as θmω2/Ne assuming weak selection . Thus , the deviations of phenotypic means from the selective optimum are expected to be inversely proportional to Ne , a point also made by Charlesworth ( 2013 ) in a somewhat different analysis . Note , however , that this is only the expected pattern , as the mean phenotype is still expected to drift above and below the expectation to a degree depending on the effective strength of selection . As noted in Equation ( 14b ) , and previously pointed out by Lande ( 1975 ) and Lande ( 1976 ) , the magnitude of this drift variance is also inversely proportional to Ne , which implies that the standard deviation with respect to the expected mean scales as ∼1/Ne . Of course , θm ( the mean phenotype expected under neutrality ) may differ among lineages and the within-population genetic variance σA2 is sensitive to the strength of selection , in which case the power to detect such relationships may be challenging . In addition , the linear scaling of σA2 with Ne is unlikely to continue indefinitely , unless Ne in natural populations rarely attains levels where all constituent loci are saturated with segregating mutations . The salient issue is that the preceding expressions provide qualitative insight into the behavior of mean phenotypes in alternative population-genetic environments , while also revealing the types of measurements that need to be made if we are to understand such behavior . For example , we know essentially nothing about the key mutational ( θm ) and selection ( ω2 ) parameters for cell biological features and how these might vary among species . This is not a trivial issue , as the influence of both parameters in determining the most likely locations of mean phenotypes are just as central as the role played by Ne . Applying the same logic to results for plateaued fitness functions leads to the prediction that the expected mode of mean phenotypes will scale fairly strongly with the effective population size , in the limit approaching proportionality to Ne , that is , a 10-fold increase in the mean phenotype with a 100-fold increase in Ne . As shown in Figure 3 , a simple change in the mutational variance σM2 ( with no associated change in mutational bias ) can also cause a substantial shift in the position of the mean phenotype . These sorts of observations raise the significant possibility that species with substantially different population-genetic environments may commonly exhibit measurable differences in trait means despite experiencing identical forms of directional selection , again raising challenging issues for those who wish to interpret phenotypic differences as reflections of different underlying processes of selection . Although the data are not extensive , several lines of evidence support the idea that the mean phenotypes of cellular attributes are indeed modulated by the power of random genetic drift . The most compelling example derives from observations on the mutation rate ( per nucleotide site per generation ) , which scales approximately inversely with the 1000-fold range of variation in Ne across the Tree of Life ( Lynch et al . , 2016 ) . Such a scaling is qualitatively consistent with the drift-barrier hypothesis for mutation-rate evolution ( Lynch , 2010; Lynch , 2011 ) , which postulates that because most mutations are deleterious , selection will typically operate to improve replication fidelity , with refinements in molecular performance eventually being thwarted by random genetic drift – as the mutation rate is progressively lowered , there is less room for improvement and hence a narrower range of selectively advantageous replication-fidelity variants accessible by selection . Enzyme efficiency provides a second broad category of traits with evolutionary behavior seemingly in accordance with the theory outline above . For example , Bar-Even et al . ( 2011 ) have found that enzymes involved in secondary metabolism are on average ∼30× less efficient than those involved in central metabolism , suggesting that selection operates less effectively on enzymes further removed from core energetic determinants . More directly relevant to the points made above , Bar-Even et al . ( 2011 ) also found that prokaryotic enzymes have slightly better kinetics than those from eukaryotes , as expected for species with higher effective population sizes and consistent with the prediction that improvement of enzyme efficiencies will stall once the gradient of the fitness surface is on the order of 1/Ne ( Hartl et al . , 1985 ) . The fact that bacteria utilize transcription-factor binding-site motifs with stronger affinity to their cognate transcription factors than is the case in eukaryotes is also plausibly related to a higher efficiency of selection in the former ( Lynch and Hagner , 2015 ) . Finally , proteins typically evolve to the ‘margin of stability , ’ such that only one or two mutations are usually enough to destabilize the folding process ( Taverna and Goldstein , 2002; Tokuriki and Tawfik , 2009 ) . Protein stability is deemed to be positively associated with fitness because destabilized proteins are prone to loss of function , aggregation , and/or direct toxicity . Strikingly , however , it is relatively easy to obtain more stable proteins by mutagenesis ( Matsuura et al . , 1999; Bershtein et al . , 2013; Sullivan et al . , 2012 ) , with the contributing residues typically interacting in an additive fashion ( Wells , 1990; Serrano et al . , 1993; Zhang et al . , 1995 ) . Moreover , although it is commonly argued that marginal stability is required for proper protein function , with excess stability somehow reducing protein performance , this has not held up to close scrutiny . Many examples exist in which increased stability has been achieved in laboratory modifications of proteins with few if any consequences for enzyme efficiency ( e . g . Giver et al . , 1998; Zhang et al . , 1995; Taverna and Goldstein , 2002; Borgo and Havranek , 2012; Moon et al . , 2014 ) . These observations suggest that despite persistent selection for high folding stability , the plateau-like nature of the fitness landscape results in diminishing fitness advantages of increasing stability . A hyperbolic relationship between fitness and the binding energy involving protein stability follows from biophysical principles ( Govindarajan and Goldstein , 1997; Taverna and Goldstein , 2002; Bloom et al . , 2005; Zeldovich and Shakhnovich , 2008; Wylie and Shakhnovich , 2011; Serohijos and Shakhnovich , 2014 ) , and under this model , proteins are expected to be pushed by natural selection to more stable configurations until reaching the point where any further fitness improvement is small enough to be offset by the vagaries of random genetic drift and/or mutation pressure towards less stable states . Notably , proteins of equivalent length fold at least ten times more rapidly in bacteria than in eukaryotes ( Galzitskaya et al . , 2011 ) . Moreover , an in vitro evaluation of the folding stability of the dihydrofolate reductase enzyme from 36 species of mesophilic bacteria illustrates the existence of a substantial range of variation among species , with the standard deviation being roughly 10% of the mean ( Bershtein et al . , 2015 ) . In principle , such a distribution may reflect the dispersion in mean phenotypes associated with drift around the drift barrier . Although the mutation function employed here likely comes closer to approximating the situation for cellular features than do previous functions relied on in quantitative genetics , in reality we do not know the exact form of this function for any cellular feature . Thus , the mathematical theory developed here is best viewed as a guide to approaching the problem at hand rather than as an indelible platform for quantitative analysis . Despite such uncertainties , however , the central feature of the theory presented above is that , regardless of the form of the underlying mutation and selection functions , the stationary distribution of mean phenotypes can generally be viewed as the product of the pattern expected under neutrality alone and the associated function for mean population fitness taken to the 2Ne power , as described by Equations ( 1a , b ) and ( 11 ) . Similar behavior was previously pointed out for the stationary distribution of allele frequencies ( Wright , 1969 ) . Thus , once the key underlying functions have been elucidated , the precise details of the theory can be readily modified with alternative mathematical functions . Finally , a key issue that is not formally evaluated here , but is arguably relevant to a number of cellular features , concerns the matter of peak shifts across the stationary distribution . Questions regarding this matter are typically inspired by Wright’s ( 1932 ) metaphor of an adaptive topography , with multiple fitness peaks and valleys of various depths over the phenotypic landscape . However , unless the distribution of mutational effects is completely flat , the relevant topography is not simply defined by the fitness landscape but by the joint action of both selection and mutation . Although the stationary distribution was unimodal in all of the cases examined above , plausible cases exist in which the stationary distribution exhibits two peaks , one largely driven by selection and the other by mutation pressure . For this to occur , the gradient of mutation pressure in one direction has to be of a form such that its product with the selection gradient has an internal minimum ( Figure 4 ) . In principle , this can happen when at the intersection of intermediate phenotypes the two functions are sufficiently upwardly concave that their product reaches a local minimum . Under such a scenario , the population mean phenotype is expected to reside in two alternative semi-stable domains for extended periods of time , with the rates of transitions between domains depending on the relative heights of the two peaks , the depth of the distributional valley , and the curvatures of the stationary distribution at the inflection points ( Lande , 1985; Barton and Rouhani , 1987 ) . Over long evolutionary time periods , such a system will exhibit detailed balance – the net fluxes will be equal in both directions , with the ratio of the occupancy of the two alternative domains being inversely related to the ratio of the transition rates between them , that is , with the less frequent domain having a higher conditional rate of transition to the more frequent domain . Although the frequency of stationary distributions with multimodal forms is unknown , they have been predicted to arise in some situations involving transcription ( Lynch and Hagner , 2015; Tuğrul et al . , 2015 ) . Should they exist , the picture from comparative analyses would be one of qualitative changes in mean phenotypes in adjacent lineages . Tempting as it might be to invoke shifting ecological pressures to explain such patterns , they would be occurring in the absence of any underlying changes in selection , being a simple consequence of the multiplicity of mutational opportunities in one direction balanced by selective pressures in the other . Such ideas may be helpful in attempts to decipher the substantial and seemingly disorganized diversity of certain cellular features such as open vs . closed mitosis ( Sazer et al . , 2014 ) , the structure of the centrosome ( Carvalho-Santos et al . , 2011 ) , and the variable multimeric states of proteins ( Dayhoff et al . , 2010; Lynch , 2013; Ahnert et al . , 2015 ) across the Tree of Life .	When most people think about evolution , they commonly think of natural selection: the evolutionary force that helps populations to develop toward an optimum state for their environment . The observable traits and features of a cell or organism are known as its phenotype . Under natural selection , genes that produce phenotypes that help a cell or organism to thrive and reproduce are more likely to be passed on to future generations . This means that over several generations the population becomes – on average – better adapted to its environment . Other ‘non-adaptive’ evolutionary forces also influence phenotype . For example , damage to DNA can introduce mutations into the genes that a cell or organism passes on to their offspring . Some mutations are more likely to produce working variants of a gene than others; this is known as a mutation bias . In addition , even in the absence of natural selection , the proportion of particular gene variants in a population changes over the generations because genes are randomly transmitted and not all individuals reproduce . This is known as genetic drift . Together , mutation bias and genetic drift could prevent a population’s average phenotype from reaching an optimal state . Lynch has now developed mathematical models that describe how certain biological features of cells – such as the structure of the proteins they produce – are likely to evolve due to mutation bias and genetic drift . These models show that these evolutionary processes can cause the features of the cells in a population to diversify , which often leads to a suboptimal average phenotype . Lynch calculated that two alternative phenotypes could even emerge in isolated populations in cases where there is only one optimum phenotype . For example , a mutation bias could drive some cells in one population to evolve one phenotype , while natural selection drives another population towards the other phenotype . Overall , the model emphasizes that natural selection is not the only force that drives diversity in cells . Future research into cell biology needs to take a broad view of the joint roles played by natural selection , mutation bias and genetic drift .	[ "Abstract", "Introduction", "Theory", "Discussion" ]	[ "evolutionary", "biology", "cell", "biology" ]	2018	Phylogenetic divergence of cell biological features
Most transgenic crops are produced through tissue culture . The impact of utilizing such methods on the plant epigenome is poorly understood . Here we generated whole-genome , single-nucleotide resolution maps of DNA methylation in several regenerated rice lines . We found that all tested regenerated plants had significant losses of methylation compared to non-regenerated plants . Loss of methylation was largely stable across generations , and certain sites in the genome were particularly susceptible to loss of methylation . Loss of methylation at promoters was associated with deregulated expression of protein-coding genes . Analyses of callus and untransformed plants regenerated from callus indicated that loss of methylation is stochastically induced at the tissue culture step . These changes in methylation may explain a component of somaclonal variation , a phenomenon in which plants derived from tissue culture manifest phenotypic variability . Rice is one of the world's most important food crops , and genetic modifications are extensively used for various purposes such as to increase yield and tolerate harsh environments . Tissue culture has been heavily used for decades for transformation procedures to generate transgenic crops such as rice and maize ( Rao et al . , 2009 ) . A previous study has reported that Arabidopsis cell suspension culture has a different epigenomic profile compared to wild-type plants , such that certain transposable elements ( TEs ) become hypomethylated and certain genes become hypermethylated ( Tanurdzic et al . , 2008 ) . This raised the question of how tissue culture processes affect the epigenome of regenerated plants derived from tissue culture . Changes in the epigenome have been proposed to be a source of somaclonal variation ( i . e . , phenotypic variation among regenerated plants ) for decades ( Kaeppler and Phillips , 1993; Kaeppler et al . , 2000; Thorpe , 2006; Rhee et al . , 2010; Miguel and Marum , 2011; Neelakandan and Wang , 2012 ) . Indeed , some evidence suggesting changes in the epigenome of regenerated plants have been reported at several specific loci or by methods such as methylation sensitive restrictive enzyme digestion ( Neelakandan and Wang , 2012 ) . However , the extent of methylation changes on a genome-wide level has not been previously assessed . Because , unlike most crops , Arabidopsis is almost exclusively transformed via Agrobacterium-mediated floral dip methods that do not utilize tissue culture ( Clough and Bent , 1998 ) , Arabidopsis is not a good model for the study of the effect of plant regeneration on the epigenome . The study of the model plant rice , however , may have practical implications for other crop species that are transformed using similar tissue culture methods . The rice genome is DNA methylated in all three cytosine contexts ( CG , CHG , CHH , where H=A , T , or C ) , with high levels of CG and CHG methylation and very low levels of CHH methylation ( Feng et al . , 2010; Zemach et al . , 2010 ) . Whole genome bisulfite sequencing ( BS-seq ) enables measurement of DNA methylation at single nucleotide resolution and thus allows one to distinguish DNA methylation in different cytosine contexts ( Cokus et al . , 2008; Lister et al . , 2008 ) . To investigate the effect that tissue culture processes have on regenerated rice epigenomes , we generated genome-wide , single-nucleotide maps of DNA methylation in several regenerated rice lines that had been transformed with various transgenes , callus , and rice regenerated from tissue culture without transformation . We observed that the tissue culture procedure induced stable changes in DNA methylation in regenerated plants , such that all regenerated lines had ectopic losses of DNA methylation . We found that loss of DNA methylation occurred stochastically , affecting individual plants somewhat differently , was associated with loss of small RNAs , and changes were enriched at promoters of genes . Loss of DNA methylation at promoters was associated with altered expression of particular genes . We performed deep BS-seq to map DNA methylation in nine regenerated rice lines in the Nipponbare ecotype background that were transformed by various transgenes and were at various stages of inbreeding after transformation: rice blast resistance lines PiZ-t , PiZ-t-839 ( a non-functional PiZ-t ) , Pi9 , and an RNAi line for flowering time regulator Spin1 ( Zhou et al . , 2006; Vega-Sanchez et al . , 2008; Table 1 ) . For the PiZ-t line , both transgenic and non-transgenic T2 and T4 plants were available by genetic segregation of the PiZ-t transgene ( Table 1 ) . For comparison , we profiled an untransformed wild-type line , which was used to generate all the regenerated lines , WT2003 ( sample 1 ) . WT2003 was also inbred 5–7 generations to produce WT2007 ( sample 2 ) , and WT2007 was inbred 5–7 additional generations to produce WT2011 ( sample 3 ) . We obtained an average genome coverage of 15× and error rates were low at 1 . 5% , 1 . 2% , 0 . 8% , for CG , CHG , CHH methylation , respectively , indicating high quality data ( Table 1 ) . 10 . 7554/eLife . 00354 . 003Table 1 . BS-Seq samples analyzed in this studyDOI: http://dx . doi . org/10 . 7554/eLife . 00354 . 003SampleDescriptionRaw readsUniquely mapping readsCoverage ( X ) CG error rateCHG error rateCHH error rate1WT200323156890210057278013 . 51780 . 01760 . 01220 . 00992WT200720354135710437698814 . 02920 . 01070 . 00870 . 00823WT20111878031098430190411 . 33090 . 01580 . 00950 . 00654T2-PiZt-11-R22965025911871009415 . 95570 . 01390 . 00990 . 00695T2-PiZt-11-S26332960213647141118 . 34290 . 01010 . 00960 . 00766T4-PiZt-11-R27067087113105670017 . 61510 . 01170 . 01000 . 00747T4-PiZt-11-S25215029812846772117 . 26720 . 00960 . 00760 . 00748T6-PiZt-11-R23728013712196674516 . 39340 . 01050 . 00960 . 00649T6-Pi9-R2047526998699574211 . 69300 . 01060 . 00930 . 005010T6-Spin1i-1-R2154510229046823612 . 15970 . 01130 . 00880 . 006111T2-PiZt-839-8-R ( non functional PiZt ) 23873028111747133215 . 78920 . 01290 . 00790 . 005612T2-PiZt-839-8-S ( non functional PiZt ) 21100611910617287214 . 27050 . 01780 . 01290 . 009513WT Callus 12171215229614527912 . 92280 . 01850 . 01780 . 007014WT Callus 21992614938261764311 . 10450 . 02320 . 02220 . 008415WT regenerated from tissue culture 121800883511636762615 . 64080 . 01700 . 01550 . 007816WT regenerated from tissue culture 22252021139790514213 . 15930 . 02620 . 02060 . 009317WT regenerated from tissue culture 325230642810654473514 . 32050 . 01940 . 01600 . 007318WT2011 ( replicate ) 25397182711814006215 . 87900 . 01720 . 01480 . 0086Number of raw sequencing reads , number of uniquely mapping reads ( post-removal of identical reads ) , genome coverage ( rice genome size = 372 Mb ) , and error rates are listed . DNA methylation levels of the chloroplast genome were used to estimate error rates . Samples 1–12 and samples 13–18 were prepared separately . “R” and “S” correspond to plants that either contain the transgene ( R ) or plants in which the transgene was segregated away ( S ) . We observed strong losses of DNA methylation at certain sites in the genome in the regenerated plants but not in wild-type plants ( Figure 1A ) . To further characterize these sites , we defined differentially methylated regions ( DMRs ) in CG contexts by applying stringent thresholds ( see ‘Materials and methods' ) . We found that all regenerated plants tested were significantly enriched with CG hypomethylation DMRs ( Figure 1B ) . On average , we identified 1344 CG hypomethylation DMRs in the regenerated plants , whose sizes ranged from 100 to 3200 bp ( Figure 1C ) , whereas on average we identified only eight CG hypomethylation DMRs in the inbred wild-type lines ( Figure 1—source data 1 ) . Importantly , we observed hypomethylation even in the T2/T4 non-transgenic plants in which the transgenes had been segregated away ( samples 5 , 7 and 12 ) , suggesting that loss of DNA methylation is likely due to the tissue culture or transformation process , but not due to the fact that the plants contain transgenes . While loss of DNA methylation in different regenerated lines did not always occur at the same sites ( Figure 1D ) , there were significant overlaps of hypomethylation DMRs among regenerated lines ( Figure 1E ) . This suggests that certain sites in the genome are susceptible to loss of DNA methylation in regenerated plants . 10 . 7554/eLife . 00354 . 004Figure 1 . Aberrant loss of DNA methylation in regenerated rice . ( A ) Genome browser views of fractional CG methylation levels . Sample numbers correspond to those listed in Table 1 . Regenerated samples of the same line are grouped together in red boxes . ( B ) Genome coverage of identified CG hypermethylation and hypomethylation DMRs . DMRs were defined relative to sample 1 ( wild type ) . ( C ) Distribution of sizes of CG hypomethylation DMRs in regenerated plants . ( D ) Heat map representation of hierarchical clustering based on CG methylation levels within DMRs . Rows represent all 3610 CG-DMRs identified and columns represent the samples . ( E ) Overlap of CG-DMRs between samples . The bottom triangle represents the percent overlap of elements listed in the x-axis with those listed in the y-axis . The upper triangle on the other hand represents the percent overlap of elements listed in the y-axis with those listed in the x-axis . DOI: http://dx . doi . org/10 . 7554/eLife . 00354 . 00410 . 7554/eLife . 00354 . 005Figure 1—source data 1 . List of CG , CHG , CHH DMRs identified in this study . Defined hypomethylation DMRs for each sample are listed . DOI: http://dx . doi . org/10 . 7554/eLife . 00354 . 005 We next investigated the stability of DNA methylation losses across generations . To test this , we analyzed a line for which we had plants in T2 , T4 , and T6 generations ( samples 4 , 6 , 8 ) . 84% of sites that lost CG methylation in the T2 did not recover methylation in the T4 and T6 generations ( Figure 2 ) . This suggests that most sites do not regain DNA methylation over several subsequent generations during the process of inbreeding . Approximately 10% of sites recovered methylation in T4 , and this methylation was maintained in T6 . In addition , 4 . 4% of sites recovered methylation in T6 but not in T4 . This suggests that certain sites are able to regain methylation over generations . Approximately 2% of sites regained methylation in T4 , but methylation was lost again in T6 , suggesting that a small fraction of sites are epigenetically unstable and continue to switch states . Our results suggest that most of the DNA hypomethylation in regenerated plants was stable over generations . 10 . 7554/eLife . 00354 . 006Figure 2 . Stability of loss of DNA methylation over generations . Methylation status of sample 4 ( T2 ) DMRs in T4 and T6 generations are indicated . Loss: less than half of respective wild-type CG methylation levels . Gain: more than half of respective wild-type CG methylation levels . DOI: http://dx . doi . org/10 . 7554/eLife . 00354 . 006 Loss of DNA methylation in regenerated plants also occurred in non-CG contexts ( Figure 1—source data 1 ) . Loss of CG methylation was generally associated with loss of CHG methylation and to a lesser extent with loss of CHH methylation ( Figure 3A , B ) . Small interfering RNAs of 24-nt in length ( 24-nt siRNAs ) are associated with DNA methylation , and are required to guide CHH methylation to particular sites ( Law and Jacobsen , 2010 ) . We performed small RNA sequencing ( smRNA-seq ) on seven randomly chosen regenerated plants along with wild type ( Table 2 ) . We examined the distribution of 24-nt siRNAs over CHH hypomethylation DMRs and found that siRNAs are enriched over these sites in wild type , but eliminated in regenerated plants ( Figure 3C ) . Hence loss of DNA methylation is associated with loss of 24 nt siRNAs . Moreover , these siRNA alterations independently confirm our findings showing loss of epigenetic marks at these loci . 10 . 7554/eLife . 00354 . 007Figure 3 . Loss of DNA methylation occurs in all three cytosine contexts . ( A ) Average distributions of DNA methylation in wild type ( faded ) and regenerated plants ( solid ) were plotted over defined CG hypomethylation DMRs in the indicated samples . Flanking regions are the same lengths as the middle region . ( B ) Heat map of DNA methylation levels within all defined hypomethylation DMRs ( CG + CHG + CHH ) . ( C ) Average distribution of smRNA-seq reads in wild type ( black ) and regenerated plants ( red ) over defined CHH hypomethylation DMRs in indicated samples . Flanking regions are the same lengths as the middle region . DOI: http://dx . doi . org/10 . 7554/eLife . 00354 . 00710 . 7554/eLife . 00354 . 008Table 2 . smRNA-seq samples analyzed in this studyDOI: http://dx . doi . org/10 . 7554/eLife . 00354 . 008SampleDescriptionRaw readsUniquely mapping reads1WT20032203066331866662WT20071706949825987803WT20111486076723997134T2-PiZt-11-R2202488139653175T2-PiZt-11-S1764162331279386T4-PiZt-11-R1899941530909337T4-PiZt-11-S2211507442587528T6-PiZt-11-R1299519320446159T6-Pi9-R16700524311492310T6-Spin1i-1-R172758132973100Number of raw sequencing reads and number of uniquely mapping reads are listed . We next examined the genomic characteristics of sites that lost DNA methylation in regenerated plants . We tested the extent of overlap between 3597 CG DMRs , 1875 CHG DMRs , and 2298 CHH DMRs defined in the regenerated lines within gene bodies , gene promoters , downstream regions of genes , gene coding sequences , gene introns , and TE genes . Although loss of DNA methylation occurred at a variety of sites , we found the most significant enrichments of DMRs at the promoters of genes ( Figure 4A ) . These genes were not significantly associated with any particular biological processes ( data not shown ) . Rather , they appeared to be a random set of genes involved in different processes ( Figure 4—source data 1 ) . Recent studies in Arabidopsis have shown that spontaneous changes in methylation over generations predominantly occurred in gene bodies ( Becker et al . , 2011; Schmitz et al . , 2011 ) . It is possible that hypomethylation observed in regenerated plants occurs through an accelerated process of whatever mechanism causes spontaneous methylation changes over generations . Alternatively , since the DNA methylation changes we observed in regenerated plants was enriched in gene promoters , and was primarily in the direction of methylation loss , it could be a distinct phenomenon from the spontaneously occurring methylation changes in wild type . 10 . 7554/eLife . 00354 . 009Figure 4 . Loss of DNA methylation at promoters may impact gene expression . ( A ) Overlap of hypomethylation DMRs with indicated genomic elements . Observed overlap ( dark bars ) is compared to randomized regions of similar number and size distribution as the DMRs ( light bars ) . Gene body: transcribed region of protein coding genes . Gene promoter: TSS to 2 kb upstream of TSS . 3' downstream of gene TTS ( transcription termination site ) : TTS to 2 kb downstream of TTS . CDS: Coding sequence . TE: Transposable element . Error bars represent standard deviation . *Significant enrichment , p<0 . 01 . ( B ) Percentages of genes with CG hypomethylation DMRs near TSSs that have significantly altered expression levels ( fourfold up/down regulation , FDR<0 . 01 ) . Genes with zero mRNA-seq reads in both wild type and regenerated samples were removed from the analyses . An average of 11 . 3 genes were deregulated . ( C ) Genome browser views of DNA methylation and gene expression levels . DOI: http://dx . doi . org/10 . 7554/eLife . 00354 . 00910 . 7554/eLife . 00354 . 010Figure 4—source data 1 . List of genes with CG hypomethylation DMRs at promoters and their expression levels . Genes with CG hypomethylation DMRs at the promoter regions ( TSS minus 2 kb to TSS ) in samples 4–10 along with their normalized expression levels are listed . Also indicated are whether they were significantly up- or down-regulated based on fourfold and FDR < 0 . 01 cutoffs . Descriptions of genes were directly taken from the rice genome annotation project website ( http://rice . plantbiology . msu . edu/ ) . DOI: http://dx . doi . org/10 . 7554/eLife . 00354 . 01010 . 7554/eLife . 00354 . 011Figure 4—figure supplement 1 . Impact of loss of DNA methylation at promoters on gene expression . Relative expression levels of genes with CG hypomethylation DMRs near TSS . Log2 ratios between RPKM values of indicated regenerated lines and wild type ( sample 2 ) were calculated , and data is represented as boxplots . Genes with zero mRNA-seq reads in both wild type and regenerated samples were removed from the analyses . Red lines , median; edges of boxes , 25th ( bottom ) and 75th ( top ) percentiles; error bars , minimum and maximum points within 1 . 5 × IQR ( Interquartile range ) ; red dots , outliers . DOI: http://dx . doi . org/10 . 7554/eLife . 00354 . 01110 . 7554/eLife . 00354 . 012Figure 4—figure supplement 2 . Genome browser views of DNA methylation and gene expression levels . DOI: http://dx . doi . org/10 . 7554/eLife . 00354 . 01210 . 7554/eLife . 00354 . 013Figure 4—figure supplement 3 . Significantly up-regulated genes are largely different across different lines . Defined significantly up-regulated genes with CG hypomethylation DMRs at promoters were categorized based on the number of lines ( out of seven tested ) in which they were up-regulated . Gene identifiers are listed below . DOI: http://dx . doi . org/10 . 7554/eLife . 00354 . 01310 . 7554/eLife . 00354 . 014Figure 4—figure supplement 4 . DNA methylation levels over upregulated TE genes in regenerated samples . Average distributions of DNA methylation in wild type ( faded lines ) and regenerated lines ( solid lines ) over defined up-regulated TE genes in the indicated regenerated samples . DOI: http://dx . doi . org/10 . 7554/eLife . 00354 . 014 While the losses of DNA methylation in regenerated plants occurred within a relatively small proportion of the rice genome , they were concentrated near protein-coding gene promoters and therefore in regions of the genome that are more prone to alter gene expression . We therefore examined the impact of hypomethylation on gene expression by performing mRNA-seq on the same seven randomly chosen regenerated plants as well as on wild-type plants ( Table 3 ) . We found that loss of DNA methylation at promoters was associated with higher expression levels of certain genes ( Figure 4B , C , Figure 4—source data 1 , Figure 4—figure supplement 1 , 2 ) . Notably , the closer the hypomethylation was to the gene transcription start site , the more likely the gene tended to be misregulated ( Figure 4B ) . Furthermore , the expression of these genes was much more frequently increased , rather than decreased , suggesting that the misexpression of these genes is likely a direct consequence of losses of DNA methylation ( Figure 4B ) . Hence loss of DNA methylation in regenerated plants is associated with deregulated transcription of certain protein-coding genes . 10 . 7554/eLife . 00354 . 015Table 3 . mRNA-seq samples analyzed in this studyDOI: http://dx . doi . org/10 . 7554/eLife . 00354 . 015SampleDescriptionRaw readsUniquely mapping reads2WT200744029089294611623WT201133997755226570984T2-PiZt-11-R42550136278395985T2-PiZt-11-S43173764286883816T4-PiZt-11-R46624891358268617T4-PiZt-11-S31729173226676338T6-PiZt-11-R46624532353356279T6-Pi9-R389785413062363310T6-Spin1i-1-R4228023532485204Number of raw sequencing reads and number of uniquely mapping reads are listed . We further sought to determine whether it was the tissue culture process or the transformation process that induced loss of DNA methylation in regenerated plants . To test this , we performed BS-seq on callus and three individual plants regenerated from untransformed callus , all of which were derived from a single parent plant ( WT2011; Table 1 ) . We were not able to perform BS-seq on individual calli because calli at the stage of transformation did not yield enough genomic DNA . Instead , we pooled multiple calli , and sequenced two separate batches . We found a strong loss of DNA methylation in plants regenerated from untransformed callus ( Figure 5A ) . Loss of DNA methylation in callus was much more modest , though significant ( Figure 5A ) . This relatively weak loss of DNA methylation may be because individual calli lose DNA methylation at different sites ( despite being derived from the same parent plant ) , and pooling multiple calli diluted the loss of DNA methylation . Consistent with this notion , individual plants regenerated from untransformed callus showed differences in sites that lost DNA methylation ( Figure 5B ) . Furthermore , when examining methylation levels of these samples at CG hypomethylation DMRs that were common in all regenerated plants , we found significant losses of DNA methylation at these sites in callus ( Figure 5C , D ) , indicating that the methylation losses observed in callus were at largely the same sites as those observed in regenerated plants . Like in the regenerated lines , the losses of DNA methylation in the non-transformed regenerated plants occurred stochastically , affecting DNA methylation in each plant somewhat differently ( Figure 5A–D ) . In summary , the loss of DNA methylation in regenerated plants is likely caused by the tissue culture step , and not due to the transformation process . 10 . 7554/eLife . 00354 . 016Figure 5 . Tissue culture step induces loss of DNA methylation . ( A ) Genome coverage of identified CG hypermethylation and hypomethylation DMRs . DMRs were defined relative to sample 18 ( wild type ) . ( B ) Heat map of CG methylation levels within all 1074 CG hypomethylation DMRs identified in samples 13 to 17 ( callus samples and wild-type plants regenerated from callus ) . ( C ) Heat map of CG methylation levels within 241 CG hypomethylation DMRs that were observed in all tested regenerated plants . ( D ) Boxplot representations of ( C ) . Red lines , median; edges of boxes , 25th ( bottom ) and 75th ( top ) percentiles; error bars , minimum and maximum points within 1 . 5 × IQR ( Interquartile range ) ; red dots , outliers . DOI: http://dx . doi . org/10 . 7554/eLife . 00354 . 016 Previous reports have indicated that certain genes are hypermethylated in Arabidopsis cell suspension culture and callus ( Berdasco et al . , 2008; Tanurdzic et al . , 2008 ) . Consistent with those data we found that rice callus showed hypermethylation throughout the genome ( Figure 6A ) . Interestingly we found that the hypermethylation occurred specifically in CHH contexts ( Figure 6A , B , Figure 6—figure supplement 1A ) , and showed high coincidence between the two callus samples ( 13 and 14 ) ( Figure 6—figure supplement 1B ) . These CHH hypermethylated regions mostly corresponded to promoter regions ( Figure 6C , Figure 6—figure supplement 1A ) . Hence in callus , certain promoters are CHH hypermethylated , while others are hypomethylated in all cytosine contexts . Interestingly , CHH hypermethylation observed in callus was completely lost in regenerated plants ( Figure 6A , B , Figure 6—figure supplement 1A ) . This suggests that unlike tissue culture-induced DNA hypomethylation that is largely stable after regeneration , CHH hypermethylation is eliminated after regeneration . 10 . 7554/eLife . 00354 . 017Figure 6 . Tissue culture-induced CHH hypermethylation is eliminated upon regeneration . ( A ) Genome browser views of DNA methylation . ( B ) Genome coverage of identified CHH hypermethylation and hypomethylation DMRs . Regenerated samples of the same line are grouped together in red boxes . ( C ) Overlap of callus CHH hypermethylation DMRs with indicated genomic elements . Observed overlap ( dark bars ) is compared to randomized regions of a similar number and size distribution as the DMRs ( light bars ) . Error bars represent standard deviation . DOI: http://dx . doi . org/10 . 7554/eLife . 00354 . 01710 . 7554/eLife . 00354 . 018Figure 6—figure supplement 1 . Callus induced CHH hypermethylation . ( A ) Average distribution of DNA methylation over defined CHH hypermethylated regions in callus , genes , and TE genes . Flanking regions are the same lengths as the middle region . ( B ) Overlap between defined CHH hypermethylation DMRs of the two callus samples in this study ( 13 and 14 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 00354 . 018 In this report , we have investigated the effect that tissue culture processes have on the epigenome of regenerated plants by generating high-resolution maps of DNA methylation . Consistent with a previous study in Arabidopsis cell culture using microarray hybridization on chromosome 4 ( Tanurdzic et al . , 2008 ) , we observed hypermethylation at certain genes in rice callus . We extend this observation by showing that hypermethylation predominantly occurs in CHH sequence contexts , most notably occurring at the promoters of genes . Interestingly , we found that this CHH hypermethylation was completely eliminated upon regeneration , suggesting that CHH hypermethylation may be linked specifically to the dedifferentiated state . In contrast to Arabidopsis cell culture , we did not observe global hypomethylation at TEs in rice callus . Instead , we found that DNA methylation was specifically lost at certain sites in the genome , appearing to affect individual plants somewhat differently despite coming from the same parent plant . We found that loss of DNA methylation was maintained upon plant regeneration , and was largely stable over subsequent generations . It is possible that some of the DMRs affected only one homologous chromosome and were segregating . However , because we required DMRs to have at least 70% reduction in DNA methylation compared to wild-type , the sites we analyzed in Figure 2 are likely homozygous for loss of DNA methylation , consistent with their stability across generations . Loss of DNA methylation occurred in all sequence contexts , and was associated with loss of 24-nt siRNAs . Notably , these sites were frequently associated with promoters of genes , and loss of DNA methylation was associated with misregulation of expression of proximal protein-coding genes , indicating a biological importance of this phenomenon . Interestingly , genes significantly up-regulated ( fourfold upregulated compared to wild type , p<0 . 01 ) in each regenerated line were somewhat different ( Figure 4—figure supplement 3 ) . For this reason , it is difficult to assess the severity of impact of misregulated gene expression for any particular regenerated line , since some lines may have more biologically important genes affected than others . This would correlate with the observation that somoclonal variation affects only a proportion of plants that arise from regeneration experiments ( Kaeppler and Phillips , 1993; Kaeppler et al . , 2000; Thorpe , 2006; Rhee et al . , 2010; Miguel and Marum , 2011; Neelakandan and Wang , 2012 ) . Previous studies have shown that certain TEs such as Tos17 and mPing are reactivated in tissue culture , and are associated with changes in DNA methylation ( Neelakandan and Wang , 2012 ) . While our results suggest that most DNA hypomethylation occurs near genes and are relatively depleted at TE related sequences ( Figure 4A ) , some of the hypomethylation did occur proximal to TE genes ( average of 62 . 1 TE genes per line ) . The association of loss of methylation with TE gene reactivation was not clear ( data not shown ) , however very subtle depletion of DNA methylation was observed over reactivated TE genes ( Figure 4—figure supplement 4 ) , suggesting that loss of methylation may in part be responsible for reactivation of TEs . Our results suggest that each regenerated plant has distinct DNA methylation profiles despite coming from the same parent ( Figure 5A–D ) . It therefore appears that the tissue culture step induces DNA hypomethylation in a rather stochastic manner affecting individual plants differently . We further show that descendants of regenerated plants stably maintain most hypomethylation across plant generations ( Figure 2 ) . Indeed , lines derived from the same original regenerated plant show very similar methylation profiles ( Figure 1E; samples 4–8 and 11–12 ) . It has long been proposed that changes in the epigenome may be a source of somaclonal variation ( Thorpe , 2006; Rhee et al . , 2010; Miguel and Marum , 2011; Neelakandan and Wang , 2012 ) . Our genome-wide data support this notion since we show that stochastic hypomethylation in individual regenerants is associated with misregulated expression of certain genes . These epigenetic changes likely explain a component of somaclonal variation that has been observed for decades in a number of plant species . In summary , our results suggest that use of tissue culture leaves behind an epigenetic footprint in regenerated plants that is stable over multiple generations and may partially explain somaclonal variation . Whereas the material used in this study were self-fertilized plants , a common practice in the development of agricultural biotechnology traits is to introgress new transgene loci into commercial genetic backgrounds , meaning that the plants used in agriculture are many generations removed from the initial regenerated plants ( Bregitzer et al . , 2008; Bennetzen and Hake , 2009; Johnson , 2009; Yang et al . , 2012 ) . The crosses utilized in these introgression schemes are likely to correct the vast majority of tissue culture-induced epigenetic changes . Wild-type rice ( Oryza sativa ssp japonica cv Nipponbare ) and regenerated rice lines ( in Nipponbare background ) were used in this study ( Zhou et al . , 2006; Vega-Sanchez et al . , 2008 ) . Hygromycin was used as the selection marker in rice transformation . All the resistant plants were selfed for indicated generations ( Table 1 ) . Homozygosity was confirmed by PCR analysis of the transgene . Rice seeds were surface-sterilized and transferred to 1/2 MS medium . After germination , rice seedlings were transplanted into soil and kept in a growth chamber at 26/20°C under a 14-hr light/10-hr dark cycle . The rice plants regenerated from untransformed rice callus induced from Nipponbare seeds ( WT2011 ) were prepared as previously described ( Zhou et al . , 2006; Vega-Sanchez et al . , 2008 ) . Rice leaf samples were collected at 3 weeks after transplanted into soil and the rice callus were harvested from the callus inducing media . BS-seq libraries were generated as previously described using premethylated adapters ( Feng et al . , 2011 ) using 1 μg of genomic DNA isolated using DNeasy Plant Maxi Kit ( Qiagen , Hilden , Germany ) . Libraries were single-end sequenced on a HiSeq 2000 , and reads were base-called using the standard Illumina software . The read counts for these libraries are listed in Table 1 . Reads ( 50 nt ) were mapped to the MSU 6 . 1 version genome using BS-seeker ( Chen et al . , 2010 ) allowing up to two mismatches . Identical reads were collapsed into one read . Fractional methylation levels were calculated by #C/ ( #C+#T ) . DMRs for each sample were defined by comparing methylation levels to wild type in 100 bp bins across the genome . Fischer's exact test was used to identify bins that were significantly differentially methylated by comparing #C and #T ( Benjamini-Hochberg corrected FDR < 0 . 01 ) . In addition , we required an absolute methylation difference of 0 . 7 , 0 . 5 , 0 . 1 , for CG , CHG , CHH methylation , respectively . Bins that were within 100 bp were merged . Finally , only bins that contained 10 informative cytosines ( i . e . , covered by ≥4 reads ) in both the sample and wild type were considered as DMRs . Sample 1 was used for the wild type control for samples 2–12 , whereas sample 18 was used for the wild type control for samples 13–17 . This was because sample preparation ( i . e . , growth of plants and library constructions ) were performed in two batches: 1∼12 and 13∼18 . All heat maps in this study were generated by complete linkage and using Euclidean distance as a distance measure . Rows with missing values were omitted for presentation purposes but did not affect the conclusions in the paper . For determining overlap of DMRs with different genomic elements , we considered 1 bp overlap as an overlap . To assess significance , we generated 100 sets of ‘randomized DMRs' which mimicked both the number and size distributions as the observed DMRs , and examined their overlaps with the different genomic elements . RNA-seq libraries were constructed from total RNA isolated using TRIzol reagent ( Invitrogen , Life Technologies , Carlsbad , CA ) from leaf tissues of samples 2∼10 . Total RNA ( 10 μg ) for each sample was used to purify poly-A mRNA; this mRNA was used for synthesis and amplification of cDNA . The RNA-seq libraries were prepared using the TruSeq RNA Sample Preparation Kit from Illumina ( San Diego , CA ) . Libraries were sequenced on an Illumina HiSeq 2000 . The read counts for these libraries are listed in Table 3 . smRNA-seq libraries were constructed from total RNA isolated from the same tissues as described for the mRNA libraries , using the TruSeq Small RNA Sample Prep Kit from Illumina ( San Diego , CA ) . The libraries were sequenced on the same Illumina HiSeq 2000 as the mRNA-seq libraries . The read counts for these libraries are listed in Table 2 . Gene annotations ( MSU 6 . 1 ) were obtained from the Rice Genome Annotation Project website ( http://rice . plantbiology . msu . edu/ ) . mRNA-seq reads were mapped and processed as previously described ( Stroud et al . , 2012 ) . Briefly , reads were uniquely mapped to the genome using Bowtie ( Langmead et al . , 2009 ) allowing two mismatches , and differentially expressed genes were defined by applying fourfold and FDR < 0 . 01 cutoffs . smRNA-seq reads were uniquely mapped to the genome using Bowtie allowing no mismatches , and reads were categorized by their lengths for analyses . Reads per kilobase per million mapped reads ( RPKM ) was used to quantify RNA-seq datasets . All sequencing data have been deposited in the NCBI Gene Expression Omnibus with accession number GSE42410 .	Rice is one of the most important food crops and is estimated to provide more than a fifth of the calories consumed by the world's population . For several decades , rice has been modified by conventional breeding methods to produce plants with increased yields and greater resistance to pests and harsh weather conditions . Efforts are also being made to create rice plants with superior yield traits and resistance to biotic and abiotic stresses using genetic engineering techniques . Genetically modified plants are usually produced using tissue culture . New genes are introduced into plant cells that are growing in a dish , and each cell then replicates to form a mass of genetically identical cells . The application of plant hormones triggers the tissue to produce roots and shoots , giving rise to plantlet clones . In addition to the genes that comprise its genome , the genetic make-up of an organism also includes its epigenome—a collection of chemical modifications that influence whether or not a given gene is expressed as a protein . The addition of methyl groups to specific sequences within the DNA , for example , acts as an epigenetic signal to reduce the transcription , and thus expression , of the genes concerned . Now , Stroud et al . reveal that the techniques used to modify a plant's genome—in particular , the process of tissue culture—also affect its epigenome . They prepared high-resolution maps of DNA methylation in several regenerated rice lines , and found that regenerated plants produced in culture showed less methylation than control plants . The changes were relatively over-represented around the promoter sequences of genes—regions of DNA that act as binding sites for the enzymes that transcribe DNA into RNA—and were accompanied by changes in gene expression . Crucially , the plants' descendants frequently also inherited the changes in methylation status . These results are likely part of the explanation for a phenomenon called somaclonal variation , first observed before the era of modern biotechnology , in which plants regenerated from tissue culture sometimes show heritable alterations in the phenotype of the plant .	[ "Abstract", "Introduction", "Results", "Discussion", "Material", "and", "methods" ]	[ "chromosomes", "and", "gene", "expression", "plant", "biology" ]	2013	Plants regenerated from tissue culture contain stable epigenome changes in rice
Synaptic inputs from different brain areas are often targeted to distinct regions of neuronal dendritic arbors . Inputs to proximal dendrites usually produce large somatic EPSPs that efficiently trigger action potential ( AP ) output , whereas inputs to distal dendrites are greatly attenuated and may largely modulate AP output . In contrast to most other cortical and hippocampal neurons , hippocampal CA2 pyramidal neurons show unusually strong excitation by their distal dendritic inputs from entorhinal cortex ( EC ) . In this study , we demonstrate that the ability of these EC inputs to drive CA2 AP output requires the firing of local dendritic Na+ spikes . Furthermore , we find that CA2 dendritic geometry contributes to the efficient coupling of dendritic Na+ spikes to AP output . These results provide a striking example of how dendritic spikes enable direct cortical inputs to overcome unfavorable distal synaptic locale to trigger axonal AP output and thereby enable efficient cortico-hippocampal information flow . The active properties of neuronal dendrites are important for integrating and processing excitatory and inhibitory synaptic inputs ( Johnston et al . , 1996; London and Hausser , 2005; Johnston and Narayanan , 2008; Major et al . , 2013 ) . Over the past few decades , dendritically generated Na+ , Ca2+ , and NMDA spikes have been identified in many types of neurons , both in vitro and in vivo ( Llinas et al . , 1968; Wong et al . , 1979; Stuart and Sakmann , 1994; Chen et al . , 1997; Schiller et al . , 1997 , 2000; Stuart et al . , 1997a; Kamondi et al . , 1998; Larkum et al . , 1999; Waters et al . , 2003; Larkum et al . , 2007 , 2009; Kim et al . , 2012; Smith et al . , 2013 ) . One proposed function of dendritic Na+ spikes is to amplify synaptic potentials and facilitate somatic AP initiation ( Hausser et al . , 2000; London and Hausser , 2005 ) . However , in most instances , dendritic Na+ spikes propagate poorly to the soma and so fail to act as reliable triggers of somatic APs ( Stuart and Sakmann , 1994; Stuart et al . , 1997a; Golding and Spruston , 1998 ) . Indeed , under physiological conditions , the APs in most principal neurons , including neocortical layer 5 and hippocampal CA1 pyramidal neurons ( PNs ) , are usually initiated at the axonal initial segment ( AIS ) before back-propagating to the dendrites ( Stuart and Sakmann , 1994; Stuart et al . , 1997a , 1997b; Golding and Spruston , 1998 ) . Thus , whereas dendritic Na+ spikes can fine-tune neuronal output and regulate synaptic plasticity ( Golding et al . , 2002; Ariav et al . , 2003; Jarsky et al . , 2005; Remy and Spruston , 2007 ) , it is less certain whether these spikes may serve as necessary events to allow synaptic input to trigger AP output ( Hausser et al . , 2000; Spruston , 2008 ) . In this study , we report that dendritic Na+ spikes play an important role in the ability of hippocampal CA2 PNs to generate axonal AP output in response to synaptic input from the direct entorhinal cortical ( EC ) projections that terminate on CA2 PN distal dendrites . Hippocampal CA2 PNs represent a relatively small population of cells interspersed between CA3 and CA1 . Nonetheless , these neurons have recently been shown to be crucial for social memory ( Hitti and Siegelbaum , 2014; Stevenson and Caldwell , 2014 ) and aggression ( Pagani et al . , 2014 ) . CA2 PNs also have unique synaptic properties that distinguish them from their CA1 and CA3 neighbors ( Zhao et al . , 2007; DeVito et al . , 2009; Simons et al . , 2009; Chevaleyre and Siegelbaum , 2010; Lee et al . , 2010; Jones and McHugh , 2011; Caruana et al . , 2012; Hitti and Siegelbaum , 2014 ) . Thus , whereas the perforant path ( PP ) inputs from the EC form weak excitatory synaptic connections at the distal dendrites of CA1 PNs located in stratum lacunosum-moleculare ( SLM ) , PP inputs to CA2 PNs provide a much stronger excitatory drive ( Chevaleyre and Siegelbaum , 2010 ) . In contrast the Schaffer collateral ( SC ) inputs to CA2 are relatively weak and dominated by powerful feed-forward inhibition ( Chevaleyre and Siegelbaum , 2010 ) . Finally , individual CA2 PNs provide stronger excitatory drive to CA1 compared to the weaker influence of single CA3 SC inputs ( Chevaleyre and Siegelbaum , 2010 ) . Such properties enable the CA2 region to function as the nexus of a powerful disynaptic circuit ( EC → CA2 → CA1 ) , directly linking EC input to hippocampal CA1 output ( Jones and McHugh , 2011 ) . How do the EC inputs trigger CA2 AP output , given that synaptic responses at distal dendrites are normally severely attenuated by the cable properties of the dendrites ? In this study , we report the PP inputs to distal dendrites in CA2 reliably initiate dendritic Na+ spikes that are necessary to trigger axonal AP output in response to a single or a burst of PP stimuli . Furthermore , these spikes can overcome strong inhibition to elicit AP output . In contrast , activation of PP inputs to distal dendrites of CA1 PNs with a single stimulus fail to elicit dendritic spikes or somatic APs . Through computational modeling based on morphological reconstructions of CA2 and CA1 PNs , we find that the distinct dendritic geometry of CA2 PNs contributes to the ability of CA2 neurons to efficiently couple dendritic Na+ spikes to AP output . Thus our data provide a striking example of how dendritic structure and functional properties control the ability of dendritic Na+ spikes to couple synaptic input at distal dendrites to axonal AP initiation . In this manner , CA2 PN dendritic Na+ spikes ensure the efficient propagation of cortical information through the EC → CA2 → CA1 disynaptic pathway . Our laboratory found that CA2 PNs receive strong excitatory inputs from EC and fire APs with high probability in response to a brief burst of stimuli delivered to the EC PP axons ( 5 pulses at 100 Hz ) . In contrast , the same PP stimuli generate a smaller synaptic response in CA1 PNs that is usually insufficient to elicit spike output ( Chevaleyre and Siegelbaum , 2010 ) . In this study , we first re-investigated the input–output relation between distal synaptic stimulation strength and sub-threshold EPSP size recorded in the soma of CA2 PNs . In these experiments the stimulation electrode was placed in SLM of the CA1 region as before but was closer to the CA1/CA2 border ( ∼50 µm ) than in our previous study ( ∼200 µm ) . Consistent with our previous findings ( Chevaleyre and Siegelbaum , 2010 ) , the sub-threshold EPSP amplitude in CA2 PNs evoked by PP stimulation was 5–6 times larger than that observed in CA1 ( Figure 1A–C ) . Moreover , the EPSP in CA2 PNs was slightly larger than that seen in our previous study , most likely due to the closer proximity to CA2 of the stimulation electrode . 10 . 7554/eLife . 04551 . 003Figure 1 . A single stimulus delivered to the perforant path ( PP ) evokes APs in CA2 PNs . ( A ) Diagram illustrating the configuration for the experiment . SLM: stratum lacunosum-moleculare , SR: stratum radiatum , SP: stratum pyramidale . S: stimulating electrode , R: recording electrode . ( B ) Sample traces of EPSPs evoked by PP stimuli using an electrode placed in SLM of CA1 . ( C ) Mean input–output curves of somatic EPSPs in CA1 ( n = 5 ) and CA2 ( n = 6-12 ) PNs . ( D ) Simultaneous whole-cell recording from a CA2 PN ( top ) and extracellular field potential recording from the CA2 cell body layer ( bottom ) in the absence ( black traces ) or the presence ( red traces ) of 20-μM NBQX and 50-μM D-APV . Top: somatic AP evoked by a single PP stimulus . Bottom: extracellular population spike ( PS ) in CA2 cell body layer . ( E ) Histogram ( bars ) and cumulative plot ( circles ) of PP stimulus threshold required to evoke APs in different CA2 PNs ( n = 42 cells ) . ( F ) Mean input–output curves of PS in CA1 ( n = 5 ) , CA2 ( n = 12 ) , and CA3b ( n = 6 ) cell body layers evoked by single PP stimuli . Inset: sample traces of PS in response to a single PP stimulus recorded in CA1 , CA2 , and CA3b cell body layers . The noise level was measured using a section of baseline that did not exhibit a PS . ( G ) Sample traces of sub-threshold and AP waveforms in response to somatic current injection ( I ) and PP stimulation ( PP ) . The dashed lines indicate mean minimal somatic depolarization required to fire APs in response to somatic current injection ( black ) and PP stimulation ( red ) , respectively . ( H ) The pooled data from CA2 PNs which fire APs in response to PP stimulation ( n = 27 ) . Peak somatic voltage amplitude plotted against stimulating intensity . ( I ) Expanded view of sub-threshold EPSP response in ( H ) . Note , the dashed lines in ( H ) and ( I ) indicate mean minimal somatic depolarization required to fire APs in response to somatic current injection ( black , 27 . 2 ± 1 . 2 mV , n = 5 ) and PP stimulation ( red , 15 . 6 ± 0 . 6 mV , n = 27 ) , respectively . DOI: http://dx . doi . org/10 . 7554/eLife . 04551 . 00310 . 7554/eLife . 04551 . 004Figure 1—figure supplement 1 . Axons from EC layer II project to CA1 SLM . ( A ) The configuration of the experiment shown in ( A–E ) . ( B ) Expanded view of the recorded GC filled with Biocytin shown in ( A ) . ( C ) The firing pattern of the GC shown in ( A and B ) in response to current injections . ( D ) Input–output relation of EPSPs from an individual GC ( same cell as in A–C ) . Inset: sample traces of EPSPs . ( E ) Mean input–output relation ( n = 6 ) . ( F ) The configuration of the experiment shown in ( G ) . ( G ) Sample EPSPs recorded in the GC shown in ( F ) . Note , there is no EPSP response once the stimulating electrode moved slightly to CA1 SR ( Stim . 1 ) . SR: stratum radiatum , SLM: stratum lacunosum moleculare , HF: hippocampal fissure . DOI: http://dx . doi . org/10 . 7554/eLife . 04551 . 004 We previously surmised that CA2 synaptic responses elicited by a focal stimulating electrode placed in SLM of CA1 were generated by activation of EC inputs from LIII pyramidal neurons , because their axons form dense projections throughout the CA1 SLM layer ( Chevaleyre and Siegelbaum , 2010 ) . However , recent studies indicate that CA2 PNs receive input largely from EC LII neurons ( Cui et al . , 2013; Hitti and Siegelbaum , 2014; Kohara et al . , 2014 ) . We investigated whether our focal stimulating electrode in SLM of CA1 might also activate EC LII axons by recording from dentate gyrus granule cells , which receive exclusive innervation from EC LII ( Witter , 2007 ) . Indeed , we find that stimulation in SLM of CA1 evokes EPSPs in the granule cells ( Figure 1—figure supplement 1 ) , consistent with a tracing study that reported the presence of EC LII axons in SLM of CA1 ( Tamamaki and Nojyo , 1993 ) . Thus we conclude that the EPSPs recorded in CA2 PNs in response to distal stimulation in SLM of CA1 likely results from the activation of inputs from EC LII . In the vast majority of CA2 PNs tested ( 35/42 cells ) , a single electrical stimulus ( up to 48 V ) delivered to the PP was sufficient to elicit a large , fast AP that was >80 mV in amplitude when recorded in the CA2 soma ( Figure 1D , E ) . The cumulative distribution of spike threshold was well fit by a sigmoidal function ( Figure 1E ) . A moderate stimulus strength ( 28 V ) was sufficient to elicit spiking in about half of all CA2 PNs studied , with a small fraction ( ∼5% ) firing APs with a very weak stimulus ( 12 V ) ( Figure 1E ) . By contrast , CA1 PNs never fired a spike in response to a single PP stimulus , even at strengths up to 60 V ( with the stimulating electrode located within 50 μm of the CA1 PN as in our CA2 recordings ) , but required a burst of high-frequency PP stimuli to fire APs ( e . g . Figure 2D , Figure 3B , Figure 4B ) , consistent with a previous report ( Jarsky et al . , 2005 ) . 10 . 7554/eLife . 04551 . 005Figure 2 . Voltage threshold of APs evoked by PP stimulation in CA2 , but not CA1 , PNs is lower than threshold of APs evoked by somatic current injection . ( A ) Sample traces of somatic voltage responses in response to somatic current injection ( I ) and a single PP stimulus ( PP ) from a CA2 PN . Inset shows the dV/dt of the AP waveforms . Arrow indicates the dendrite spike with PP stimulation . ( B ) Phase-plane plots of dV/dt vs instantaneous voltage from data shown in ( A ) . Note , the arrow indicates a dendrite spike preceding a full-blown AP . ( C ) Pooled data of AP threshold induced by somatic current injections ( I ) vs PP stimulation ( PP ) ( n = 17 ) . Filled circle: constant current injection . Open circle: EPSC-like current injection . AP threshold defined as the somatic voltage at which dV/dt exceeds 10 V/s ( left ) or 50 V/s ( right ) . *p < 0 . 001 . ( D ) Sample traces of somatic voltage responses in response to somatic current injection ( I ) and high-frequency ( 50 Hz , 5 pulses ) burst PP stimulation ( PP ) from a CA1 PN . Inset shows the dV/dt of AP waveforms . ( E ) Phase-plane plots of dV/dt vs instantaneous voltage from data shown in ( D ) . Note , the phase plot from somatic current injection ( I ) is identical with that from PP stimulation ( PP ) . ( F ) Pooled data of AP threshold induced by somatic current injections ( I ) vs PP stimulation ( PP ) ( n = 16 ) . Filled circle: constant current injection . Open circle: EPSC-like current injection . AP threshold defined as the somatic voltage at which dV/dt exceeds 10 V/s ( left ) or 50 V/s ( right ) . n . s . , not significant . DOI: http://dx . doi . org/10 . 7554/eLife . 04551 . 00510 . 7554/eLife . 04551 . 006Figure 2—figure supplement 1 . A single stimulus to the PP triggers dendritic spikes ( D spikes ) in CA2 PNs . ( A ) Sample traces of sub-threshold EPSPs ( black ) , D spikes ( green ) , and APs ( orange ) in a CA2 PN in response to PP stimuli of increasing strength . Inset: the expanded view from the dashed square . ( B ) dV/dt plotted against peak EPSP amplitude ( APs are not shown ) . Traces and data points color-coded as in ( A ) . Inset: dV/dt of corresponding traces from ( A ) . APs are truncated . ( C and D ) Pooled data showing dV/dt ( C ) and the 20–80% EPSP rise time ( D ) plotted against peak EPSP amplitude ( n = 9 cells ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04551 . 00610 . 7554/eLife . 04551 . 007Figure 3 . Hyperpolarization of membrane potential reveals prominent spikelets at the soma of CA2 , but not CA1 , PNs in response to PP stimulation . ( A ) Top: a PP stimulus in a CA2 PN reliably elicited a somatic AP at the normal resting potential ( trace 1 ) . Hyperpolarization of the resting membrane ( traces 2–4 ) reveals that a PP stimulus with constant strength induced spikelets that variably succeeded ( trace 2 ) or failed ( trace 3 ) in triggering a somatic AP from the hyperpolarized potential . Trace 4 shows a very weak spikelet response to the PP stimulus . Bottom: dV/dt for corresponding voltage responses on top . dV/dt of APs is truncated . Right: overlay of traces 2–4 . ( B ) High-frequency burst PP stimulation ( 50 Hz , 5 pulses ) in a CA1 PN triggers somatic APs at the resting potential ( left ) . Hyperpolarization of the resting membrane fails to reveal spikelets ( right ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04551 . 00710 . 7554/eLife . 04551 . 008Figure 4 . Somatic TTX application reveals prominent spikelets at the soma of CA2 , but not CA1 , PNs in response to PP stimulation . ( A ) Diagram illustrating the configuration of CA1 PN experiment as shown in ( B ) . ( B ) Sample traces of somatic voltage response in a CA1 PN to high-frequency burst stimulation ( 100 Hz , 10 pulses ) in the absence ( control ) or the presence of somatic TTX ( somatic TTX ) . Note the absence of spikelets at the soma . ( C ) Diagram illustrating the configuration of CA2 PN experiment shown in ( D–I ) . ( D ) Sample traces of somatic voltage response in a CA2 PN to a single PP stimulus in the absence ( control ) or the presence of somatic TTX ( TTX ) . Note the presence of a prominent spikelet during somatic TTX application at the soma of CA2 PN . ( E ) A phase-plane plot from the traces shown in ( D ) . Note the overlap of the initial rising phase in control and TTX ( arrow ) . ( F ) Superimposed traces of somatic voltage response of a CA2 PN to suprathreshold and subthreshold PP stimuli in the absence or the presence of somatic TTX . ( G ) Input–output of somatic voltage response from the CA2 PN shown in ( F ) . Inset shows an expanded plot of the sub-threshold somatic voltage response . Note , the spikelet amplitude reaches >35 mV , providing ∼20 mV extra somatic depolarization on top of the EPSP . ( H ) Sample traces of dV/dt from ( F ) . ( I ) Input–output curve of dV/dt from the CA2 PN shown in ( F–H ) . Inset shows an expanded plot of sub-threshold dV/dt . Note , dV/dt of the spikelets reaches ∼50 V/s . DOI: http://dx . doi . org/10 . 7554/eLife . 04551 . 008 To rule out the possibility that CA2 spike firing was an artifact of washout caused by whole-cell recording and to measure the collective activity of a population of neurons , we measured the extracellular population spike ( PS ) in response to PP stimulation using an extracellular field recording electrode placed in stratum pyramidale ( SP ) of the CA1 , CA2 , or CA3b region ( Figure 1F ) . Consistent with the whole-cell recordings , a prominent PS was elicited by a single stimulus in the CA2 region . Moreover the CA2 PS input–output curve was fit by a sigmoidal function that closely matched that seen with whole-cell recordings ( Figure 1F ) . By contrast , we failed to detect a measurable PS in the cell body layer of the CA1 or CA3b region ( Figure 1F ) . The electrically evoked APs were eliminated by blockade of fast glutamatergic synaptic transmission , using bath application of NBQX ( 20 μM ) and D-APV ( 50 μM ) ( Figure 1D , n = 4 ) . Thus , the spikes were driven by synaptic excitation , not by direct electrical stimulation of CA2 dendrites . The efficient triggering of action potential output from CA2 PNs in response to PP stimulation is surprising given that the largest EPSP observed in the CA2 PN soma was ∼11 mV below the voltage threshold of AP output in response to somatic current injection ( Figure 1G–I , see also Figure 2C ) . Thus , even though the PP EPSP is fivefold larger in CA2 than CA1 PNs , the mechanism coupling synaptic input to AP output is unclear . In the remainder of this study , we address the hypothesis that dendritic spikes are critical for enabling the distal synaptic inputs to trigger this output . Initial evidence for the triggering of dendritic spikes came from inspection of CA2 PN voltage responses to PP stimuli that were just sub-threshold for eliciting somatic action potentials . Nearly all CA2 PNs fired APs in response to a PP stimulus that was strong enough to induce a somatic EPSP greater than 15 . 6 ± 0 . 6 mV ( ranging from 7 . 7–20 . 3 mV , n = 27 , Figure 1G–I ) . In a subset of CA2 PNs ( n = 9/42 cells ) , PP stimuli that were just below threshold for eliciting an AP evoked a somatic voltage response with a rapidly rising phase of depolarization not seen in voltage responses evoked by weaker PP stimuli ( Figure 2—figure supplement 1A ) . In cells that displayed this rapid voltage response , a small increase in stimulus strength usually evoked a full-blown AP ( Figure 2—figure supplement 1A ) . This rapidly rising depolarization is similar to the somatic spikelets associated with dendritic Na+ spikes reported in other studies ( Losonczy and Magee , 2006; Losonczy et al . , 2008; Remy et al . , 2009; Muller et al . , 2012 ) . To further characterize the components of the voltage response , we analyzed their maximal rate-of-rise ( dV/dt ) ( Figure 2—figure supplement 1B , C ) . Although dV/dt initially increased linearly with peak EPSP amplitude , as the EPSP reached values around 15–20 mV , dV/dt increased sharply in a non-linear manner . This suggests that large EPSPs are sufficient to trigger a non-linear membrane response that likely reflects the firing of a dendritic spike ( Figure 2—figure supplement 1B , C ) . Consistent with this view , the 20–80% rise time of the EPSP also decreased non-linearly as the peak EPSP amplitude reached values above ∼15 mV ( Figure 2—figure supplement 1D ) . As the rapidly rising spikelets were observed in only a subset of our recordings , we wondered whether they were a consistent feature of CA2 responses to PP stimuli but were normally masked by the much larger somatic AP ( e . g . Figure 1D , G , Figure 2A ) . In agreement with this idea , a phase-plane plot of dV/dt vs membrane voltage showed that APs induced by PP stimulation in CA2 PNs consistently exhibited an initial rapid dV/dt signal at potentials sub-threshold to the main spike ( Figure 2A , B ) . Such an early rapid phase of depolarization was not observed when APs were elicited by somatic current injection , suggesting that they do indeed represent dendritic spikes ( Figure 2A , B ) . Furthermore , we compared the threshold to fire APs induced by PP stimulation with that induced by somatic current injection . The AP threshold ( defined as the somatic voltage at which dV/dt reached 50 V/s ) was significantly lower in response to PP stimulation ( −51 . 6 ± 1 . 0 mV ) compared to somatic current injection ( −44 . 3 ± 0 . 8 mV , n = 17 , p < 0 . 001; Figure 2C ) , indicating that dendritic spikes may indeed be required for PP input to trigger somatic APs . We next asked whether dendritic spikes were also a feature of PP-evoked action potentials in CA1 PNs . As a single PP stimulus was ineffective in triggering a somatic action potential in CA1 PNs , we used a 50-Hz burst of 5 strong PP stimuli to evoke somatic spiking . In marked contrast with our results with CA2 PNs , neither our somatic voltage recordings nor phase-plane plots showed evidence of a spikelet ( Figure 2D , E ) . Moreover , individual spikes , phase-plane plots , and voltage threshold of somatic APs induced by PP stimulation were identical with those induced by somatic current injections ( Figure 2D–F ) . These results suggest that in CA1 PNs , somatic depolarization resulting from temporal summation of PP-evoked EPSPs , rather than dendritic spikes , drives AP output . To examine more directly whether dendritic spikes consistently underlie the CA2 PN somatic AP , we prevented AP firing by injecting negative current to hyperpolarize the somatic membrane to −82 . 5 ± 2 . 8 mV ( n = 10 ) ( Figure 3A ) . Strikingly , this revealed that PP stimulation consistently evoked a rapidly rising spikelets ( dV/dt of 20 . 6 ± 2 . 6 V/s; n = 10 ) in response to PP stimulation ( Figure 3A ) . When we applied repeated single PP stimuli of a constant strength near the threshold for eliciting somatic APs , the CA2 PN membrane response often fluctuated between a full-blown AP response and a spikelet ( Figure 3A ) . Overlaying these responses showed that the spikelets preceded the full-blown APs ( Figure 3A ) , suggesting that the dendritic spikes triggered the full-blown APs . This was particularly evident in plots of dV/dt , where the spikelet waveform observed in isolation could be identified in traces associated with full-blown APs immediately prior to the AP response ( Figure 3A ) . In marked contrast , we failed to observe spikelets in CA1 soma upon membrane hyperpolarization ( Figure 3B , n = 9 ) . As a second means of recording spikelets in the absence of somatic APs , we locally applied tetrodotoxin ( TTX , 1 µM ) to the soma of CA1 or CA2 PNs while maintaining the neurons at their initial resting potential ( Figure 4 ) . In CA2 PNs , this manipulation revealed the consistent presence of spikelets ( dV/dt = 36 . 4 ± 3 . 0 V/s , n = 8 ) in response to PP stimuli ( Figure 4C–I ) . These spikelets provide a substantial somatic depolarization ( 33 . 6 ± 2 . 8 mV , n = 8 ) that would normally trigger a somatic action potential in the absence of TTX . Overlaying dV/dt in phase-plane plots in the absence and the presence of TTX again showed that the spikelets preceded the full-blown APs ( Figure 4E ) . In contrast , we failed to observe spikelets with TTX applied to the soma of CA1 PNs ( Figure 4A , B , n = 4 ) . Taken together , we conclude that single PP stimuli are sufficient to evoke prominent dendritic spikes , which result in high-amplitude somatic spikelets in CA2 but not CA1 PNs . Next , we asked whether PP-driven dendritic spikes are necessary to generate AP output in CA2 PNs . As discussed above , one indication for the necessity of dendritic spikes is our finding that the threshold to fire a somatic AP with PP stimulation is negative to the threshold with somatic current injection ( Figure 5—figure supplement 1 ) . To directly determine whether dendritic spikes are required for eliciting a CA2 PN somatic action potential under the conditions of our experiments , we applied TTX locally to the proximal dendrites of CA2 PNs , which should block dendritic Na+ spikes . Indeed , this manipulation fully blocked the ability of single PP stimuli to elicit somatic APs ( Figure 5A–D , n = 6 ) . After blockade of dendritic Na+ spikes , the EPSP evoked by strong PP stimulation maximally depolarized the soma by ∼20 mV positive to the resting potential ( −70 to −73 mV ) ( Figure 5D ) , below the threshold for driving AP output by somatic current injection . Importantly , the local TTX application exerted a selective effect on dendritic excitability and did not alter PP synaptic transmission or somatic excitability . Thus , there was no change in the sub-threshold PP EPSP or AP firing in response to somatic current injection ( Figure 5E–H ) . These results strongly suggest that dendritic spikes in CA2 PNs are mediated by TTX-sensitive voltage-gated Na+ channels ( see below for additional evidence ) and are necessary for somatic AP initiation with a single PP stimulus . 10 . 7554/eLife . 04551 . 009Figure 5 . Cortically driven dendritic Na+ spikes are necessary to fire APs in CA2 PNs in response to a single PP stimulus . ( A ) Diagram illustrating the configuration of the experiment . ( B ) Superimposed voltage responses of CA2 PN in response to suprathreshold and subthreshold PP stimuli . Responses obtained in the absence ( Control ) or the presence of TTX applied to dendrites in SR ( Dendritic TTX ) . Note dendritic TTX blocks AP in response to strong stimulus but does not alter subthreshold EPSP in response to weaker stimulus . ( C ) Phase-plane plots of dV/dt vs instantaneous voltage from data shown in ( B ) . Arrow indicates a dendritic spike preceding a full-blown AP . ( D ) Input–output curve of somatic voltage response of an individual CA2 PN to PP stimulation in the absence of TTX ( Control ) or during local TTX application in SR ( Dendritic TTX ) . ( E ) Sample traces of AP waveforms in response to somatic EPSC-like current injection . Responses obtained in the absence ( Control ) or the presence of TTX applied to dendrites in SR ( Dendritic TTX ) . ( F ) Phase-plane plots of dV/dt vs instantaneous voltage from data shown in ( E ) . Note lack of the rising phase preceding a full-blown AP seen in ( C ) . ( G and H ) Sub-threshold EPSP evoked by PP stimulation ( n = 6 ) and AP voltage threshold evoked by somatic EPSC-like current injection ( n = 5 ) in the absence ( Control ) and the presence of TTX applied to dendrites in SR ( TTX ) for individual experiments ( circles ) and mean ( squares ) . Error bars show SEM . SEM was smaller than symbol . n . s . , not significant . DOI: http://dx . doi . org/10 . 7554/eLife . 04551 . 00910 . 7554/eLife . 04551 . 010Figure 5—figure supplement 1 . Dendritic Na+ spikes are necessary to fire APs in CA2 PNs . Sample traces of somatic voltage response in a CA2 neuron to EPSC-like current injection ( left ) or a PP stimulus ( middle and right , the same stimulation strength ) . Bottom: dV/dt of the corresponding traces on top . Constant negative current injection was applied to hyperpolarize the membrane potential to ∼−100 mV . DOI: http://dx . doi . org/10 . 7554/eLife . 04551 . 010 In awake-behaving animals , EC neurons often fire in high-frequency bursts rather than isolated single APs ( Burgalossi et al . , 2011 ) . It is thus possible that temporal summation of somatic depolarization driven by EC bursting may be sufficient to elicit AP output in CA2 PNs without the need for dendritic spikes . However , a phase-plane plot revealed the presence of an early rapid rising voltage response preceding the full-blown somatic spikes in response to a 50 Hz burst of 5 PP stimuli ( Figure 6A , B ) . Moreover , application of somatic TTX revealed the consistent presence of somatic spikelets ( Figure 6C , D , n = 6 ) that precede AP initiation in the absence of TTX ( Figure 6D ) . Importantly , local application of TTX to the proximal dendrites fully blocked the ability of PP burst stimuli to drive AP output ( Figure 6E , F , n = 3 ) . Thus , under the conditions of our experiments , cortically-driven dendritic Na+ spikes were necessary for AP output in CA2 PNs with either single or bursts of PP input . 10 . 7554/eLife . 04551 . 011Figure 6 . Dendritic Na+ spikes evoked by high-frequency burst stimulation to PP are necessary to fire APs in CA2 PNs . ( A ) Sample somatic voltage response to high-frequency PP burst stimulation ( 50 Hz , 5 pulses ) . ( B ) Phase-plane plot from the traces shown in ( A ) . The numbers ( 1–3 ) correspond to AP waveforms shown in ( A ) . Note , the arrows indicate the rapid rising phase ( dendritic Na+ spikes ) preceding the full-blown APs . ( C ) Diagram illustrating the configuration of the experiment shown in ( D ) . ( D ) Sample traces of somatic spikelets revealed by somatic TTX application in response to 48 V ( top ) or 32 V ( bottom ) high-frequency PP burst stimulation ( 50 Hz , 5 pulses ) . Left: control , middle: somatic TTX , right: overlay . ( E ) Diagram illustrating the configuration of the experiment shown in ( F ) . ( F ) Sample traces of somatic voltage response to high-frequency PP burst stimulation ( 50 Hz , 5 pulses ) . Left: control , middle: dendritic TTX , right: overlay . Note , APs are blocked by dendritic TTX application . DOI: http://dx . doi . org/10 . 7554/eLife . 04551 . 01110 . 7554/eLife . 04551 . 012Figure 6—figure supplement 1 . Latency to fire APs in response to PP burst stimulation is shorter in CA2 than CA1 . ( A ) Sample traces of population spike ( PS ) recorded in CA1 and CA2 cell body layers in response to a burst of 5 PP stimuli ( at 50 Hz ) . Arrows indicate PS . ( B ) Mean PS amplitude as function of stimulus number . CA1 , n = 8; CA2 , n = 9 . Error bars show SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 04551 . 012 Dendritic Na+ spikes can influence the timing of AP output and thus have been proposed to contribute to temporal coding ( Ariav et al . , 2003 ) . To address whether dendritic Na+ spikes affect the timing of AP output , we measured population spikes in the CA1 and CA2 stratum pyramidale ( SP ) cell body layers in response to a burst of PP stimuli ( 5 pulses at 50 Hz ) . In CA1 , a measurable PS was never observed following the first PP stimulus and was only detected during the later stimuli in a burst , resulting in a latency to firing of 20 ms or more ( Figure 6—figure supplement 1 ) . In contrast , the PS amplitude in CA2 was largest in response to the first stimulus of a burst and gradually diminished with successive stimuli ( Figure 6—figure supplement 1 ) . The decrease in PS response was not due to synaptic depression because the paired-pulse ratio ( PPR ) of PP synapses in CA2 PNs exhibited facilitation rather than depression ( PPR = 1 . 39 ± 0 . 06 , n = 17 ) . Thus , CA2 PNs fired precisely and immediately in response to PP stimulation ( latency of the first PS = 4 . 4 ± 0 . 1 ms , n = 5 ) , suggesting that dendritic Na+ spikes enable CA2 PNs to respond rapidly to EC burst firing with a high degree of temporal fidelity . To examine dendritic spike firing in response to PP stimulation more directly , we measured the local field responses simultaneously from two extracellular recording electrodes placed near the middle of the axial axis of the apical dendrites in the stratum radiatum layer ( SR ) and in the SP layer of CA1 , CA2 , or CA3 regions ( Figure 7A ) . CA2 dendrites generated an active excitatory current response to PP stimulation , manifested as a negative field voltage response in recordings from SR ( Figure 7A , B ) . In contrast , PP stimuli never evoked an active response in SR of CA1 or CA3 . As the excitatory synaptic response is local to the site of PP input in SLM , the negative field response in SR must be generated by voltage-gated excitatory conductances . Importantly , the amplitude of the active response in SR was correlated with PS size in SP ( somatic spike; R = 0 . 91 , Figure 7C ) . Furthermore , the SR potential preceded the PS in SP ( Figure 7D ) , suggesting that dendritic Na+ spikes precede somatic APs . Moreover , the difference in latency between SR and SP active responses was correlated with the distance between the dendritic and somatic recording sites ( Figure 7D ) . These results suggest that the dendrites of CA2 PNs , but not CA1 or CA3 PNs , fire spikes that propagate to the soma of CA2 PNs to trigger AP output . 10 . 7554/eLife . 04551 . 013Figure 7 . CA2 , but not CA1 or CA3 , dendrites are active in response to PP stimulation . ( A ) Left: diagram illustrating the configuration for extracellular field recording . Right: sample traces of field EPSP ( fEPSP ) responses in SLM , SR , and SP of CA1 and CA2 regions . Arrow indicates active dendritic response ( negative field potential ) in SR of CA2 . ( B ) Mean input–output curves of PS amplitude in SR of CA1 ( n = 4 ) , CA2 ( n = 7 ) , and CA3 ( n = 6 ) regions . ( C ) PS amplitude in SR plotted against PS amplitude in SP from the simultaneous field recordings in SR and SP in CA2 ( correlation coefficient = 0 . 91 , p < 0 . 001 , n = 17 ) . ( D ) Left: scaled simultaneously recorded field responses in SR and SP in CA2 ( red circle in the right panel; distance from SP = 105 μm ) . Right: the time difference of the response latency between SR and SP plotted against the distance between the two recording electrodes ( correlation coefficient = 0 . 92 , p < 0 . 001 , n = 13 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04551 . 013 To characterize further the ionic mechanisms of the dendritic spikes and the active dendritic properties of CA2 PNs , we recorded voltage responses directly from CA2 dendrites under whole-cell current clamp conditions . CA2 dendrites had distinct electrophysiological properties from CA1 dendrites , including lack of voltage sag , a late depolarizing response to hyperpolarizing current injection characteristic of Ih , and lack of AP adaptation ( CA2 recordings of Figure 8B compared to CA1 recordings in Figure 8—figure supplement 1 ) . The identity of all recorded CA2 PNs was subsequently verified by biocytin staining based on morphology , including a lack of large thorny excrescences ( a characteristic of CA3 PNs ) , the presence of a large cell body , and distinct dendritic branching patterns ( Figure 8A; Ishizuka et al . , 1995 ) . We found that dendritic spikes could be reliably triggered by brief depolarizing current pulses in all CA2 dendrites examined , with a current threshold of 1 . 7 ± 0 . 2 nA ( n = 10 ) ( Figure 8C ) . Furthermore , the dendritic spikes were generated by voltage-gated Na+ channels as they were blocked by bath application of TTX ( 0 . 5 µM , n = 3; Figure 8C ) . Finally , the dendritic recordings directly demonstrated that a single PP synaptic stimulus was able to evoke a dendritic spike ( Figure 8D , E ) . 10 . 7554/eLife . 04551 . 014Figure 8 . Local dendritic Na+ spikes observed with dendritic whole-cell recordings in CA2 PNs . ( A ) Top: a typical CA2 PN filled with biocytin following dendritic whole-cell recording ( recording distance: ∼125 μm from soma ) . Bottom: an expanded view of the box shown on top . Arrows show the lack of thorny excrescences , the postsynaptic spines of mossy fiber synapses seen in CA3 PNs . ( B ) The voltage response of a CA2 PN dendrite to local current injection ( same neuron as in A ) . ( C ) A dendritic spike ( recording distance: ∼150 μm ) evoked by a 5-ms current pulse in the absence or the presence of TTX ( 0 . 5 μM ) . ( D ) Diagram illustrating the configuration for simultaneous dendritic whole-cell recording and extracellular field recording in CA2 cell body layer shown in ( E ) . ( E ) Dendritic spike ( top ) , dV/dt ( middle ) , and PS ( bottom ) in response to a single PP stimulus ( same neuron as in C ) . Note , dendritic spike precedes PS in CA2 cell body layer . DOI: http://dx . doi . org/10 . 7554/eLife . 04551 . 01410 . 7554/eLife . 04551 . 015Figure 8—figure supplement 1 . Dendritic whole-cell recordings in CA1 dendrites . ( A ) The experimental configuration . ( B ) Sample traces of the voltage response of a CA1 dendrite ( 195 μm ) to negative or positive constant current injections . Note a larger Sag ( characteristic of Ih ) than that from CA2 dendritic recordings ( Figure 8B ) . ( C ) Sample traces of the EPSPs recorded in a CA1 dendrite ( 217 μm ) in response to single PP stimuli . Note the lack of dendritic spikes . ( D ) An isolated dendritic spike elicited by a high-frequency burst of PP stimuli ( 50 Hz , 5 pulses ) in a CA1 dendrite ( 196 μm ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04551 . 01510 . 7554/eLife . 04551 . 016Figure 8—figure supplement 2 . NMDAR activation is not required for dendritic spikes in CA2 PNs . ( A ) Left: mean input–output curves of EPSPs in response to a single PP stimulus in the absence vs the presence of 50-μM D-APV . Inset: sample EPSPs in the absence ( black ) and the presence ( red ) of D-APV . Right: EPSP amplitude of individual neurons in response to a single 12 V PP stimulus . ( B ) Left: mean input–output curves of PS in CA2 cell body layer in response to a single PP stimulus in the absence vs the presence of D-APV . Inset: samples traces of PS . Right: PS amplitude of individual neurons in response to a single 12 V PP stimulus in the absence or the presence of D-APV . Square symbols show mean . Error bars show SEM . ( C ) Left: sample whole-cell voltage responses from a CA2 PN to PP stimuli of fixed strength in the absence ( black ) vs the presence of D-APV ( red ) . APs are truncated . Right: dV/dt of sample traces shown in the left . Note the presence of D spikes in the presence of D-APV . DOI: http://dx . doi . org/10 . 7554/eLife . 04551 . 016 In addition to Na+ spikes , apical tuft dendrites can generate local NMDA spikes to enhance synaptic input ( Schiller et al . , 2000; Nevian et al . , 2007; Larkum et al . , 2009; Lavzin et al . , 2012 ) . However , we found that bath applied D-APV ( 50 μM ) did not block dendritic spikes elicited by a single PP stimulus ( Figure 8—figure supplement 2 ) and failed to inhibit the PS in the CA2 PN layer elicited with strong current stimulating pulses ( >24 V; Figure 8—figure supplement 2 ) . However , D-APV did produce a small decrease ( 20% ) in EPSP amplitude ( Figure 8—figure supplement 2 ) . Thus NMDAR activation is not required for dendritic spike initiation or AP output in response to PP activation in CA2 PNs . The experiments so far described were performed in the presence of GABA receptor antagonists ( see ‘Materials and methods’ ) to facilitate dendritic excitation . However , under physiological conditions , the distal dendrites of CA2 PNs receive powerful inhibition that can significantly influence their excitation by the EC inputs . We thus asked whether dendritic Na+ spikes can enable CA2 PNs to overcome inhibition and allow the PP inputs to trigger AP output ( Muller et al . , 2012 ) . We first compared the input–output relation for the CA2 PN somatic postsynaptic potential ( PSP ) amplitude as a function of PP stimulus strength in the presence or the absence of GABAR antagonists . For stimulus strengths above 12 V , the PP PSP was significantly enhanced by the GABAR antagonists , indicating that PP stimulation did indeed elicit strong inhibition ( Figure 9A ) . At an intermediate stimulus strength of 24 V , the PP-evoked PSP depolarized the membrane by only 6 . 6 ± 0 . 9 mV with inhibition intact , whereas the PSP size increased to 11 . 7 ± 1 . 3 mV when inhibition was blocked ( p < 0 . 001 , n = 10 ) . Even at maximum strength ( 52 V ) , the PSP amplitude only reached 8 . 4 ± 1 . 1 mV with inhibition intact ( ranging from 4 . 3 mV to 13 . 6 mV , n = 9 , excluding responses with dendritic Na+ spikes or APs ) . Given that the resting potential of CA2 PNs was −75 . 6 ± 0 . 7 mV ( n = 23 ) and the AP threshold with somatic current injection was −44 . 3 ± 0 . 8 mV ( n = 17 ) , this level of depolarization is very far from threshold for eliciting somatic APs . Surprisingly , however , extracellular field recordings showed that a single PP stimulus was , in fact , able to evoke a PS in the CA2 cell body layer even with inhibition present ( Figure 9B ) . As expected , the stimulation threshold for eliciting a PS was higher and the PS amplitude was reduced when inhibition was intact compared to when inhibition was blocked ( Figure 9B ) . 10 . 7554/eLife . 04551 . 017Figure 9 . Dendritic Na+ spikes in CA2 PNs overcome inhibition to trigger APs . ( A ) Mean input–output curves of sub-threshold postsynaptic potentials ( PSP ) in response to a single PP stimulus in the absence ( red ) or the presence ( black ) of GABAA and GABAB receptor antagonists , 2 μM SR 95531 , and 1 μM CGP 55845 , respectively ( SR/CGP; n = 8–10 ) . Inset: sample traces of PSPs in the absence and the presence of GABAR antagonists . Trials in which stimulus elicited an AP were not included . ( B ) Mean input–output curves of PS in response to a single PP stimulus in the absence or the presence of GABAR antagonists ( n = 13 ) . Inset shows sample PS . ( C ) Sample traces of PSPs and APs obtained from somatic whole-cell recordings in response to a PP stimulus with increasing strength in the absence or the presence of GABAR antagonists . ( D ) Bars show population frequency histogram of AP threshold in the absence ( red ) and the presence ( black ) of GABAR antagonists . Circles show cumulative distribution of firing probability as function of stimulus voltage ( control: red , n = 34 cells; SR/CGP: black , n = 42 cells ) . ( E ) Peak somatic voltage amplitude plotted against stimulating intensity from individual CA2 PNs which fire APs in responses to a PP stimulus in the absence of GABAR antagonists ( n = 6 ) . ( F ) Expanded view of sub-threshold PSP response in ( E ) . Note , the red dashed line in ( E ) and ( F ) indicates mean PSP amplitude ( 10 . 99 ± 0 . 59 mV , n = 6 ) right before CA2 PNs fire APs . ( G ) Left: sample traces of sub-threshold PSPs , dendritic Na+ spikes , and an AP in a CA2 PN in response to single PP stimuli of constant strength near the threshold for AP firing ( 60 V , 5 trials ) in the absence of GABAR antagonists . Note , PSPs are only able to depolarize membrane to −60 . 4 mV ( black line ) . Right: dV/dt of the corresponding traces shown at left . DOI: http://dx . doi . org/10 . 7554/eLife . 04551 . 01710 . 7554/eLife . 04551 . 018Figure 9—figure supplement 1 . Strong PP stimulation with constant strength variably triggers APs , D spikes , or PSPs in the presence of inhibition . ( A ) Sample traces of PSPs , D spikes , and an AP in response to PP stimulation with increasing strengths . The AP is truncated . ( B ) Quantification of the experiment shown in ( A ) . Note , 10 trials with 48 V stimulation variably trigger D spikes , PSPs , or APs ( not shown ) . ( C ) dV/dt corresponding to traces shown in ( A ) . The AP is truncated . ( D ) Quantification of the experiment shown in ( C ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04551 . 018 Somatic whole-cell recordings confirmed that a single PP stimulus was able to elicit AP firing when inhibition was present , although with a reduced probability compared to when inhibition was blocked ( Figure 9C , D ) . With inhibition intact , CA2 PNs fired APs following somatic PSPs whose peak amplitude , on average , reached a threshold of 10 . 99 ± 0 . 59 mV ( Figure 9E , F , ranging from 8 . 6–13 . 6 mV , n = 6 ) . In the presence of inhibition , repeated trials of PP stimulation using a constant stimulus strength near threshold elicited variable responses , with full-blown APs , spikelets ( dendritic Na+ spikes ) , or sub-threshold PSPs ( Figure 9G , Figure 9—figure supplement 1 , n = 8 ) . A close inspection of traces with APs or dendritic spikes and an analysis of dV/dt demonstrated that dendritic Na+ spikes consistently preceded full-blown APs ( Figure 9G , Figure 9—figure supplement 1 ) . Thus , dendritic Na+ spikes driven by EC inputs substantially boost somatic depolarization to overcome inhibition and enable CA2 PNs to generate AP output . In CA1 PNs , the propagation of dendritic Na+ spikes to the soma is severely attenuated by the dendritic cable properties . As a result , these spikes appear at the soma as small ( <5 mV ) , slowly rising ( dV/dt < 10 V/s ) spikelets ( Spruston , 2008 ) . In contrast , the spikelets at CA2 PN soma have a much larger amplitude ( 25 . 5 ± 2 . 6 mV ) and rate-of-rise ( dV/dt = 36 . 4 ± 3 . 0 V/s , n = 8 ) and so are capable of driving action potential output . Why do CA2 PNs produce such large Na+ spikelets at the soma compared to CA1 PNs ? To explore the factors that contribute to these differences , we constructed morphologically realistic computational models of CA2 and CA1 PNs , based on Neurolucida reconstructions of biocytin-filled cells ( Hines and Carnevale , 1997 ) . CA2 PN apical dendrites have a number of distinct morphological features compared with CA1 dendrites that might influence spike propagation ( e . g . Figures 8A and Figure 10 , Figure 10—figure supplements 1 and 2 ) . CA2 neurons extend a single apical dendrite from the soma that , within 50–100 μm , splits into multiple secondary branches that project into SLM . Thus each branch provides an independent direct route for voltage to propagate to the soma . In contrast , CA1 neurons send a single apical dendrite to the border of SR and SLM , where the branch splits into a number of fine secondary and tertiary tuft dendrites . 10 . 7554/eLife . 04551 . 019Figure 10 . Modeling the differential coupling of dendritic Na+ spikes to AP output in CA1 vs CA2 dendrites . ( A , B ) Weak PP stimulation ( ∼75 synapses ) triggered local spikes at distal apical dendrites in CA1 ( A ) , but not in CA2 ( B ) , PNs . In both models , dendritic spikes failed to propagate to the soma . ( C ) Strong PP stimulation ( ∼1000 synapses ) triggered local Na+ spikes in the apical tuft of the CA1 PN that failed to propagate to the soma . ( D ) Strong PP stimulation ( ∼1000 synapses ) triggered local Na+ spikes in the apical tuft of CA2 PN that propagated effectively to the soma and triggered an AP . Note the presence of a prominent dendritic Na+ spike at the CA2 PN primary apical dendrite branch point . APs are truncated . Color maps in ( A–D ) represent voltage snapshots . Left: snapshot taken at the time of peak voltage response in distal dendrites of CA1 and CA2 PNs . Right: snapshot taken at the time of peak voltage response at main apical dendritic trunk for CA1 PN or primary dendritic branch point for CA2 PN . Scale: 100 µm . Traces show voltage response at indicated positions . ( E ) Increasing numbers of secondary or tertiary CA2 apical branches ( ∼300 μm from the soma ) were activated by ∼150 synapses per branch to trigger dendritic spikes . Offset traces at bottom show voltage responses in individual branches . Simultaneous activation of six out of twelve branches triggered a spike at the soma . Note a prominent dendritic Na+ spike at the branch point of primary apical dendrite of the CA2 PN . Color maps represent snapshots captured at time point of peak voltage observed at the main branch point of CA2 apical dendrites . Scale: 100 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 04551 . 01910 . 7554/eLife . 04551 . 020Figure 10—figure supplement 1 . Quantification of dendritic morphology of CA1 vs CA2 using Sholl analysis . Note , the number of intersections ( crossings of the dendritic branches ) in the middle and distal apical dendrites in CA2 is significantly larger than that in CA1 . p < 0 . 01 . DOI: http://dx . doi . org/10 . 7554/eLife . 04551 . 02010 . 7554/eLife . 04551 . 021Figure 10—figure supplement 2 . Modeling the differential coupling of dendritic Na+ spikes to AP output in CA1 vs CA2 dendrites . Models based on morphological reconstructions from a second set of CA1 and CA2 PNs . Strong PP stimulation ( ∼1000 synapses ) evoked local Na+ spikes in the apical tuft of both CA1 ( top ) and CA2 ( bottom ) PNs . These spikes propagated effectively to the soma and triggered an AP at the CA2 soma ( bottom ) , but not at the CA1 soma ( top ) . Note a prominent dendritic Na+ spike at the main branch point of the primary apical dendrite of the CA2 PN ( cyan trace ) . The AP ( red ) is truncated . Color maps represent the snap shots captured at time points of peak voltage immediately after PP stimulation ( left ) or peak voltage observed at main apical trunk of CA1 and the main branch point of CA2 apical dendrites ( right ) . Note the large depolarization ( warm color ) at the main branch point ( ‘hot spot’ zone ) in CA2 model . DOI: http://dx . doi . org/10 . 7554/eLife . 04551 . 021 As there are few quantitative measurements of voltage-gated conductances in the very thin dendrites of mouse CA1 and CA2 PNs , we used the same conductance parameters previously used to model rat CA1 PN dendrites , consisting of a voltage-gated Na+ conductance ( GNa ) , delayed rectifier and A-type K+ conductances ( GKdr and GKA ) , and a hyperpolarization-activated cation conductance , Ih ( Jarsky et al . , 2005; see ‘Materials and methods’ ) . We lowered the Ih conductance in CA2 PNs relative to that of CA1 PNs to match our experimental measures of voltage sag , a slow depolarizing response that follows the hyperpolarizing response to inward current steps that is characteristic of the activation of Ih . We then slightly adjusted the values of GNa , GKdr , and GKA so that the excitability of our models matched our experimental results . The final values of these conductances were identical in the CA1 and CA2 models ( see ‘Materials and methods’ ) . We first asked whether simulated PP synaptic input onto distal dendrites of CA1 and CA2 PNs is capable of generating dendritic spikes that are sufficient to drive AP output . Consistent with our experimental results , strong synaptic stimulation onto CA2 distal dendrites did indeed produce dendritic Na+ spikes that propagated to the soma to trigger AP output . By contrast , in the CA1 PN model , although a similar level of distal synaptic stimulation was able to trigger local Na+ spikes at the apical tufts , these spikes propagated poorly to the soma and failed to initiate AP output ( Figure 10A–D ) . Why do the CA2 dendrites propagate dendritic spikes to the soma more efficiently than the CA1 PN dendrites ? One important clue comes from the presence of a large local spike at the main branch point of the apical dendrites . Moreover this spike precedes the somatic AP in response to PP stimulation ( Figure 10D , Figure 10—figure supplement 2 ) . Such findings suggest that the main branch point in CA2 PNs acts as a ‘hot spot’ that generates large dendritic spikes that trigger AP output . Similar results were obtained in models from a second set of reconstructed CA1 and CA2 PNs ( Figure 10—figure supplement 2 ) . To explore further the influence of dendritic architecture on spike firing , we examined the influence of activating an increasing number of secondary or tertiary apical branches in the CA2 model . Interestingly , the simultaneous firing of dendritic spikes in six out of twelve branches ( ∼300 μm from the soma ) was necessary to evoke a somatic AP ( Figure 10E ) . The multiple independent spikes in the dendritic branches were each subject to considerable attenuation as they propagated from the distal region of the dendrites to the primary branch point . However , at primary dendritic branch point , the spikes from each branch summated to produce a very large dendritic spike , which then propagated with little decrement over the short remaining distance to the soma to generate a supra-threshold spikelet ( Figure 10E ) . Taken together , these simulations have identified CA2 PN dendritic morphology as a key factor that helps enable the efficient coupling of dendritic Na+ spikes to AP output . The importance of dendritic Na+ spikes in the generation of CA2 PN action potential output represents one end of a continuum of results on the role of these spikes in different classes of neurons . Thus , in both CA1 PNs and neocortical layer 5 neurons , dendritic Na+ spikes normally fail to propagate to the soma and are only weak triggers of somatic APs ( Stuart et al . , 1997a; Golding and Spruston , 1998; Jarsky et al . , 2005; Larkum et al . , 2007 ) . Nonetheless , dendritic Na+ spikes do sometimes precede somatic spikes and may trigger AP output with strong synaptic stimulation ( Turner et al . , 1991; Stuart et al . , 1997a; Golding and Spruston , 1998 ) or direct current injection ( Williams and Stuart , 2002; Gasparini et al . , 2004 ) . However , in most neurons , the stimulating intensity required to initiate dendritic Na+ spikes is significantly higher than that required for axonal AP initiation ( Turner et al . , 1991; Stuart and Sakmann , 1994; Stuart et al . , 1997a ) . Therefore , although dendritic Na+ spikes do have the capability of transforming synaptic inputs into neuronal outputs under certain condition; in most pyramidal neurons , they are neither sufficient nor necessary for axonal AP initiation ( Hausser et al . , 2000; Spruston , 2008 ) . Consistent with this notion , we did not detect spikelets ( a hallmark of dendritic spikes ) in CA1 PN soma in response to a burst of PP stimuli . Instead , our data suggest that temporal summation of somatic depolarization in response to short bursts of PP stimuli , rather than dendritic Na+ spikes , is what drives AP output in CA1 PNs . A previous study reported that stimulation of PP inputs onto CA1 PNs in rats can evoke dendritic Na+ spikes that sometimes appear as somatic spikelets that may help drive somatic AP output ( Jarsky et al . , 2005 ) . However , such somatic spikelets were only observed in a small minority of cells ( ∼5% ) in response to high-frequency bursts of PP stimuli ( Jarsky et al . , 2005 ) , indicating that the vast majority of local spikes generated at the distal apical dendrites failed to propagate to CA1 soma , consistent with our CA1 results . Thus , compared to CA2 , the influence of dendritic spikes on AP output in CA1 PNs is rather limited . In contrast to the results on CA1 PNs , dendritic spikes play a more important role in CA1 oriens-alveus interneurons , whose high density of dendritic voltage-gated Na+ channels ensures active spike propagation to the soma , which triggers axonal AP output under some conditions ( Martina et al . , 2000 ) . Using a computational model based on reconstructed CA2 and CA1 PNs , we found that the morphology of the CA2 apical dendritic arbor contributes to the efficient coupling of dendritic Na+ spikes to AP output in CA2 PNs . Specifically , the main branch point of the apical dendrites in CA2 acts as a ‘hot spot’ that integrates spikes from multiple secondary dendrites to generate a large amplitude spike in the short primary dendrite that triggers AP output . This provides a striking example of how dendritic morphology critically influences the propagation of dendritic spikes , as suggested previously ( Vetter et al . , 2001 ) . Our results do not rule out the possibility that additional factors , such as distinct distribution patterns and/or biophysical properties of voltage-gated ion conductances along the CA2 dendrites , may also contribute to the efficient coupling of dendritic spikes to AP output . To overcome the unfavorable geography of their cortical inputs , CA2 PNs utilize a number of mechanisms that boost the magnitude of the somatic response to the distal synaptic inputs from EC . One set of mechanisms , which remains to be identified , increases the magnitude of the sub-threshold somatic EPSP ( Chevaleyre and Siegelbaum , 2010 ) . However , despite its larger amplitude , the EPSP generated by the EC inputs is still below the threshold for action potential firing ( using somatic current pulses ) . Given a mean CA2 PN resting potential of −75 mV and an AP threshold of −44 mV , a >30 mV somatic depolarization is required for CA2 PNs to reach the threshold to fire an AP . Yet we find that , in the absence of dendritic Na+ spikes , the EPSP reaches a peak value of around 15–20 mV with strong PP stimulation , far negative to the threshold for eliciting a somatic spike . This is consistent with our finding that dendritic spikes are necessary to trigger an action potential output in response to EC input in the CA2 PNs . Recent results show that CA2 PNs also receive direct input from DG granule cells through the mossy fiber pathway , although the CA2 PNs lack the thorny excrescences characteristic of CA3 mossy fiber synapses ( Kohara et al . , 2014 ) . However , the DG inputs provide relatively weak synaptic drive onto CA2 PN apical dendrites , evoking small PSPs whose peak amplitude of 5–10 mV is far below the threshold for eliciting spikes with somatic current injection . Nonetheless , the DG inputs can drive CA2 spike output ( Kohara et al . , 2014 ) . To explain the discrepancy between EPSP size and CA2 PN threshold , we suggest that dendritic Na+ spikes may also enable AP output through this additional route of information transfer . Dendritic Na+ spikes have been suggested to contribute to temporal coding in neural networks ( Ariav et al . , 2003; Gasparini and Magee , 2006 ) . Consistent with this idea , we find that CA2 neurons fire precisely and immediately in response to a single PP stimulus . This is in contrast with CA1 , where temporal summation of synaptic potentials is required for generating APs . With a burst of PP stimuli , CA2 fires with the highest probability in response to the first PP stimulus . In contrast , the probability of CA1 firing increases during successive stimuli in the burst . We speculate that this mechanism may exert an influence on the temporal structure of information flow through the cortico-hippocampal circuit . Consistent with our in vitro findings , in vivo extracellular recording demonstrated that CA2 neurons fire APs earlier than CA1 or CA3 neurons in response to EC stimulation ( Bartesaghi and Gessi , 2004; Bartesaghi et al . , 2006 ) . Dendritic Na+ spikes have been observed in vivo in some types of neurons ( Kamondi et al . , 1998; Waters et al . , 2003; Smith et al . , 2013 ) . Although it is not known whether dendritic Na+ spikes occur in vivo in CA2 PNs , an active response in the CA2 dendritic field has been observed in response to an electrical stimulus to the PP using extracellular field recording in anesthetized guinea pigs ( Bartesaghi and Gessi , 2004 ) . Importantly , this active dendritic response precedes that in the cell body layer ( Bartesaghi and Gessi , 2004 ) , which is consistent with our observations in acute hippocampal slices ( Figure 7 ) . This suggests that dendritic Na+ spikes in CA2 PNs may similarly influence AP initiation in vivo . Whether cortically driven dendritic Na+ spikes in CA2 PNs occur in awake-behaving animals and whether they are important for specific behaviors , including social behaviors ( DeVito et al . , 2009; Hitti and Siegelbaum , 2014; Pagani et al . , 2014 ) , remain open questions . Transverse hippocampal slices were prepared from 5- to 8-week old C57BL/6J male mice from the Jackson Laboratory , as described previously ( Chevaleyre and Siegelbaum , 2010 ) . In brief , animals were anesthetized and killed by decapitation in accordance with institutional regulations . Hippocampi were dissected out , and transverse slices ( 400 µm thickness ) from the dorsal hippocampus were cut on a vibratome ( Leica VT1200S , Germany ) in ice-cold dissection solution containing ( in mM ) : 10 NaCl , 195 sucrose , 2 . 5 KCl , 10 glucose , 25 NaHCO3 , 1 . 25 NaH2PO4 , 2 Na Pyruvate , 0 . 5 CaCl2 , and 7 MgCl2 . The slices were then incubated in 33°C ACSF ( in mM: 125 NaCl , 2 . 5 KCl , 20 glucose , 25 NaHCO3 , 1 . 25 NaH2PO4 , 2 Na Pyruvate , 2 CaCl2 , and 1 MgCl2 ) for 20–30 min and then kept at room temperature for at least 1 . 5 hr before transfer to the recording chamber . Cutting and recording solutions were both saturated with 95% O2 and 5% CO2 ( pH 7 . 4 ) . All electrophysiological recording experiments were performed at 31–32°C . For some experiments , a cut was made between CA2 and CA3 regions . Whole-cell recordings were obtained from PNs ‘blindly’ in current clamp mode with a patch pipette ( 4–6 MΩ for somatic recording; 7–10 MΩ for dendritic recording ) containing ( in mM ) : 135 K gluconate , 5 KCl , 0 . 1 EGTA-Na , 10 HEPES , 2 NaCl , 5 Mg ATP , 0 . 4 Na2GTP , 10 Na2 phosphocreatine ( pH 7 . 2; 280–290 mOsm ) . Series resistance and resting membrane potential were monitored throughout each experiment . Neurons with series resistance >25 MΩ ( somatic ) or >50 MΩ ( dendritic ) were excluded from analysis . Neurons with resting potential more positive than −60 mV were also rejected from analysis . Synaptic potentials , dendritic spikes , and AP outputs were evoked by monopolar stimulation with a patch pipette filled with 1 M NaCl and located in SLM of the CA1 region ( ∼50 μm from CA2 region ) . CA2 PNs were identified based on a number of electrophysiological properties as described previously ( Chevaleyre and Siegelbaum , 2010 ) , including resting membrane potential , input resistance , and firing properties . The paired-pulse ratio was calculated as the ratio of the second to the first EPSP response using two PP stimuli with 50 ms interpulse interval . In some somatic whole-cell recording and all dendritic recording experiments , neurons were filled with biocytin ( 0 . 2–1 % , Sigma , St . Louis , MO ) during recording and morphological reconstruction was subsequently performed for further verification . Neurons were held at −70 to −73 mV for input–output curves and for examining the effect of PP stimulation on dendritic spikes and AP output . Resting membrane potential was measured immediately upon break-in . Except for the experiment shown in Figure 9 and Figure 9—figure supplement 1 , all EPSPs , dendritic spikes , APs , and population spikes were recorded in the presence of GABAA and GABAB antagonists ( 2 μM SR 95531 and 1 μM CGP 55845 , Tocris , Bristol , UK ) . Extracellular field potentials were recorded with glass patch pipettes containing 1 M NaCl . The recording pipettes were placed in the pyramidal layer or various locations along SR in CA1 , CA2 , or CA3 fields . Field responses were evoked using a stimulating electrode placed in SLM of the CA1 field ( ∼50 μm from the border with CA2 ) . Except for the experiments shown in Figure 9 and Figure 9—figure supplement 1 , all experiments were performed in the presence of GABAR antagonists . In some experiments , to prevent the contamination of polysynaptic activation from CA2 and CA3 neurons in response to PP stimulation , both CA2 and CA3 regions were cut-off for assessing PS response in CA1 cell body layer . Neurons were filled with biocytin using whole-cell patch recordings that were held for >15 min to allow for diffusion of biocytin . The slices were fixed and kept overnight in 4% paraformaldehyde in 0 . 1 M phosphate buffer ( PB ) at 4°C . The slices were then rinsed five times for five minutes per rinse in 0 . 1 M PB and were treated with 0 . 3–1% hydrogen peroxide in 0 . 1 M PB for 30–40 min . After three rinses , slices were treated with 2% Avidin–Biotin-Peroxidase Complex ( ABC , Vector Laboratories , Burlingame , CA ) for 1–2 days . Each slice was then developed with 0 . 05–0 . 07% 3 , 3′-diaminobenzidine tetrahydrochloride ( DAB ) and 0 . 005% hydrogen peroxide until the slice turned light brown . Subsequently , slices were rinsed in PB several times and processed through increasing concentrations of glycerol and then embedded in mounting media ( Fino and Yuste , 2011 ) . Neurons with robust staining of the dendritic tree were reconstructed using Neurolucida software ( MBF Bioscience , Williston , VT ) . The neurons were viewed with a 63x oil objective on a Zeiss upright light microscope . Whole-cell reconstructions included the soma and dendritic branches and shafts , but not dendritic spines . Three dimensional whole neuron reconstructions , including dendritic diameters and lengths , were imported into the NEURON simulation environment ( Hines and Carnevale , 1997 ) . To build active models of CA1 and CA2 PNs , we used a similar approach as described previously ( Jarsky et al . , 2005 ) . To the best of our knowledge , there are no available experimental data regarding distributions or biophysical properties of dendritic voltage-gated conductances from mouse hippocampal CA1 or CA2 PNs . Thus the parameters used in our models were derived from the experimental data obtained from rat hippocampal CA1 PNs . The models incorporated passive membrane properties ( Rm = 40 , 000 Ω cm2 , Cm = 0 . 75 μF/cm2 , Ri = 150 Ω cm ) . To account for spines , Cm of the dendritic compartments was multiplied by a spine scale factor and their Rm was divided by the same factor . In the CA1 PN model , we used a spine scale factor of 2 in compartments 150 μm beyond the soma , whereas in the CA2 PN model we used spine scale factors of 2 or 3 for compartments below or beyond 150 μm from the soma , respectively . These spine scale factor values were chosen to match the membrane time constant and input resistance values of the models to our experimental values . The models also included four active conductances: a Na+ conductance ( GNa ) , a delayed rectifier K+ conductance ( GKdr ) , an A-type K+ conductance ( GKA ) , and a hyperpolarization-activated cation conductance ( Gh ) . The biophysical parameters of GNa , GKdr , GKA , and Gh were implemented , as described previously ( Magee , 1998; Jarsky et al . , 2005 ) . These conductances were inserted in all compartments of the models . The distribution of GNa and GKdr is uniform throughout the somato-dendritic axis in both CA1 and CA2 models with a conductance value of 0 . 022 S/cm2 and 0 . 035 S/cm2 respectively . GKA was modeled with sixfold increase in conductance along the somato-dendrtic axis as described previously ( Hoffman et al . , 1997; Jarsky et al . , 2005 ) , with conductance values of 0 . 035 S/cm2 at the soma in both models . Gh was modeled with a sevenfold increase in conductance along the somato-dendritic axis as described previously for CA1 PNs ( Magee , 1998 ) , whereas in CA2 PNs its distribution was uniform based on our inspection of immunocytochemistry results ( Santoro et al . , 2004 ) . All simulations were performed at a resting potential of −70 mV . All excitatory synapses were modeled using two exponential functions to describe the conductance time course ( τrise of 0 . 2 ms and τdecay of 2 ms ) with a reversal potential of 0 mV with specific synaptic conductance values described below . For distal synaptic activation , excitatory synapses ( 0 . 0002 µS per synapse ) were distributed randomly onto the distal dendritic arbor ( >400 µm from the soma ) . Weak synaptic stimulation was performed by activating ∼75 synapses randomly , whereas in strong synaptic stimulation ∼1000 synapses were activated randomly in both models . For the branch model ( Figure 10E ) , each branch ( ∼300 µm from the soma ) was activated by ∼150 synapses . Data were digitized with a Digidata 1440A interface ( Molecular Devices , Sunnyvale , CA ) and were acquired using AxoGraph X software ( AxoGraph , Berkeley , CA ) . Data analysis was performed using Igor Pro ( Wavemetrics , Lake Oswego , OR ) , AxoGraph X , and Excel ( Microsoft , Redmond , WA ) . Phase-plane plot and dV/dt values were obtained with the build-in programs in AxoGraph X . AP threshold was defined as the somatic voltage at which dV/dt exceeded 10 V/s or 50 V/s . To determine the EPSP amplitude when dendritic spikes were present ( e . g . Figure 2—figure supplement 1A , B ) , EPSP amplitude was determined at the peak depolarization of the EPSP waveform ( the latency of peak EPSP is typically >8 ms ) following the dendritic spike ( dendritic spike latency is usually <5 ms ) . For Figure 1C , EPSP data were excluded if a cell started to fire somatic APs in response to increasing stimulation intensity . In a subset of CA2 PNs , PP stimulation with constant high-intensity stimuli variably triggered full-blown APs , dendritic spikes without APs , or PSPs in the presence of inhibition ( e . g . Figure 9G , Figure 9—figure supplement 1 ) . In those cases , the PSP values were used to generate the input–output relation of PSP and stimulating intensity ( Figure 9A ) . Statistical comparisons were performed using Student's t test or ANOVA . Results are expressed as mean ± SEM .	Cells called neurons carry information—in the form of electrical signals—around the brain . These cells connect to each other in complex networks and each neuron is able to form junctions , or synapses , with many neighbors . In a neuron , small electrical signals start from synapses at the tips of branched structures called dendrites . From there , these signals travel to the cell body of the neuron to activate a larger electrical signal—called an action potential—that travels along a long tail-like extension , called the axon , to reach synapses with other neurons . In the dendrites , the small electrical signals can be amplified by rapid changes in the concentration of sodium ions , known as Na+ spikes . Although they were first recorded over 40 years ago , it is not clear how important the Na+ spikes are for triggering action potentials . In this study , Sun et al . studied a type of neuron in the hippocampus called CA2 pyramidal neurons , which are involved in social memory and aggression . Unlike most other neurons in this region , CA2 neurons are strongly activated by signals from a neighboring region of the brain called the entorhinal cortex . The experiments show that Na+ spikes are able to travel from the dendrites to the cell body of these neurons , where they are required to trigger action potentials . However , this is not the case for other neurons in the hippocampus , where the Na+ spikes are very weak by the time they reach the cell body . Sun et al . used a computational modeling technique to compare the different types of neurons in the hippocampus . The dendrites of these cells have different branching patterns and shapes , and the model suggests that this may explain the differences in how well the Na+ spikes travel to the cell body . The next major challenge is to understand the role of the Na+ spikes in social memory and other complex behaviors that are controlled by CA2 neurons .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "neuroscience" ]	2014	Dendritic Na+ spikes enable cortical input to drive action potential output from hippocampal CA2 pyramidal neurons
Aberrant display of the truncated core1 O-glycan T-antigen is a common feature of human cancer cells that correlates with metastasis . Here we show that T-antigen in Drosophila melanogaster macrophages is involved in their developmentally programmed tissue invasion . Higher macrophage T-antigen levels require an atypical major facilitator superfamily ( MFS ) member that we named Minerva which enables macrophage dissemination and invasion . We characterize for the first time the T and Tn glycoform O-glycoproteome of the Drosophila melanogaster embryo , and determine that Minerva increases the presence of T-antigen on proteins in pathways previously linked to cancer , most strongly on the sulfhydryl oxidase Qsox1 which we show is required for macrophage tissue entry . Minerva’s vertebrate ortholog , MFSD1 , rescues the minerva mutant’s migration and T-antigen glycosylation defects . We thus identify a key conserved regulator that orchestrates O-glycosylation on a protein subset to activate a program governing migration steps important for both development and cancer metastasis . The set of proteins expressed by a cell defines much of its potential capacities . However , a diverse set of modifications can occur after the protein is produced to alter its function and thus determine the cell’s final behavior . One of the most frequent and variable of such alterations is glycosylation , in which sugars are added onto the oxygen ( O ) of a serine or threonine or onto the nitrogen ( N ) of an asparagine ( Kornfeld and Kornfeld , 1985; Marshall , 1972; Ohtsubo and Marth , 2006 ) . O-linked addition can occur on cytoplasmic and nuclear proteins in eukaryotes ( Comer and Hart , 2000; Hart et al . , 2011 ) , but the most extensive N- and O- linked glycosylation occurs during the transit of a protein through the secretory pathway . A series of sugar molecules are added starting in the endoplasmic reticulum ( ER ) or cis-Golgi and continuing to be incorporated and removed until passage through the trans Golgi network is complete ( Aebi , 2013; Stanley et al . , 2009 ) . N-linked glycosylation is initiated in the ER at consensus NxS/T X≠P site , whereas the most common GalNAc-type O-linked glycosylation is initiated in the early Golgi and glycosites display no clear sequence motifs , apart from a prevalence of neighboring prolines ( Bennett et al . , 2012; Thanka Christlet and Veluraja , 2001 ) . Glycosylation can affect protein folding , stability and localization as well as serve specific roles in fine-tuning protein processing and functions such as protein adhesion and signaling ( Goth et al . , 2018; Varki , 2017 ) . The basic process by which such glycosylation occurs has been well studied . However our understanding of how specific glycan structures participate in modulating particular cellular functions is still at its beginning . The need to understand the regulation of O-glycosylation is particularly relevant for cancer ( Fu et al . , 2016; Häuselmann and Borsig , 2014 ) . The truncated O-glycans called T and Tn antigen are not normally found on most mature human cells ( Cao et al . , 1996 ) but up to 95% of cells from many cancer types display these at high levels ( Boland et al . , 1982; Cao et al . , 1996; Howard and Taylor , 1980; Limas and Lange , 1986; Orntoft et al . , 1985; Springer , 1984; Springer et al . , 1975 ) . The T O-glycan structure ( Galβ1-3GalNAcα1-O-Ser/Thr ) is synthesized by the large family of polypeptide GalNAc-transferases ( GalNAc-Ts ) that initiate protein O-glycosylation by adding GalNAc to form Tn antigen and the core1 synthase C1GalT1 that adds Gal to the initial GalNAc residues ( Tian and Ten Hagen , 2009 ) to form T antigen ( Figure 1A ) . The human C1GalT1 synthase requires a dedicated chaperone , COSMC , for folding and ER exit ( Ju and Cummings , 2005 ) . In adult humans these O-glycans are normally capped by sialic acids and/or elongated and branched into complex structures ( Tarp and Clausen , 2008 ) . However , in cancer this elongation and branching is reduced or absent and the appearance of these truncated T and Tn O-glycans correlates positively with cancer aggressiveness and negatively with long-term prognoses for many cancers in patients ( Baldus et al . , 2000; Carrasco et al . , 2013; Ferguson et al . , 2014; MacLean and Longenecker , 1991; Schindlbeck et al . , 2005; Springer , 1997; Springer , 1989; Summers et al . , 1983; Yu et al . , 2007 ) . The molecular basis for the enhanced appearance of T antigen in cancers is not clear ( Chia et al . , 2016 ) , although higher Golgi pH in cancer cells correlates with increases in T antigen ( Kellokumpu et al . , 2002 ) . Interestingly , T antigen is also observed as a transient fetal modification ( Barr et al . , 1989 ) and cancer cells frequently recapitulate processes that happened earlier in development ( Cofre and Abdelhay , 2017; Pierce , 1974 ) . Identifying new mechanisms that regulate T antigen modifications developmentally has the potential to lead to insights into cancer biology . Drosophila as a classic genetic model system is an excellent organism in which to investigate these questions . Drosophila displays T antigen as the predominant form of GalNAc- , or mucin-type , O-glycosylation in the embryo with 18% of the T glycans being further elaborated , predominantly by the addition of GlcA ( Aoki et al . , 2008 ) . As in vertebrates , the GalNAc-T isoenzymes directing the initial step of GalNAc addition to serines and threonines are numerous in Drosophila , with several already known to display conserved substrate specificity in vitro with their vertebrate orthologs ( Müller et al . , 2005; Schwientek et al . , 2002; Ten Hagen et al . , 2003 ) . The Drosophila GalNAc-Ts affect extracellular matrix ( ECM ) secretion , gut acidification and the formation of the respiratory system ( Tian and Ten Hagen , 2006; Tran et al . , 2012; Zhang et al . , 2010 ) . In flies the main enzyme adding Gal to form T antigen is C1GalTA ( Müller et al . , 2005 ) whose absence causes defects in ventral nerve cord ( vnc ) condensation during Stage 17 , hematopoetic stem cell maintenance , and neuromuscular junction formation ( Fuwa et al . , 2015; Itoh et al . , 2016; Lin et al . , 2008; Yoshida et al . , 2008 ) . While orthologous to the vertebrate Core1 synthases , the Drosophila C1GALTs differ in not requiring a specific chaperone ( Müller et al . , 2005 ) . Most interestingly , T antigen is found on embryonic macrophages ( Yoshida et al . , 2008 ) , a cell type which can penetrate into tissues in a manner akin to metastatic cancer ( Ratheesh et al . , 2018; Siekhaus et al . , 2010 ) . Macrophage invasion of the germband ( Figure 1B , arrow in Figure 1C ) occurs between the closely apposed ectoderm and mesoderm ( Ratheesh et al . , 2018; Siekhaus et al . , 2010 ) from late Stage 11 through Stage 12 . This invasion occurs as part of the dispersal of macrophages throughout the embryo ( Figure 1C ) along other routes that are mostly noninvasive , such as along the inner ventral nerve cord ( vnc ) ( arrowhead in Figure 1C ) ( Campos-Ortega and Hartenstein , 1997; Evans et al . , 2010 ) . Given these potentially related but previously unconsolidated observations , we sought to determine the relationship between the appearance of T antigen and macrophage invasion and to use the genetic power of Drosophila to find new pathways by which this glycophenotype is regulated . To identify glycan structures present on fly embryonic macrophages during invasion we performed a screen examining FITC-labelled lectins ( see Materials and methods for abbreviations ) . Only two lectins had higher staining on macrophages than on surrounding tissues ( labeled enriched ) : PNA , which primarily binds to the core1 T O-glycan , and UEA-I , which can recognize Fucα1-2Galβ1-4GlcNAc ( Molin et al . , 1986; Natchiar et al . , 2007 ) ( Figure 1D , Figure 1—figure supplement 1A–B ) . Both glycans are associated with the invasive migration of mammalian cancer cells ( Agrawal et al . , 2017; Hung et al . , 2014 ) . SBA , WGA , GS-II , GS-I , ConA , MPA and BPA bound at similar or lower levels on Drosophila macrophages compared to flanking tissues ( Figure 1D , Figure 1—figure supplement 1C–I ) . We saw no staining with the sialic acid-recognizing lectin LPA , and none with DBA and HPA , that both recognize α-GalNAc ( Piller et al . , 1990 ) ( Figure 1D , Figure 1—figure supplement 1J–L ) . Thus PNA and UEA-I display enriched macrophage binding during their embryonic invasive migration . To confirm T antigen as the source of the upregulated PNA signal in embryonic macrophages during invasion and to characterize its temporal and spatial enrichment , we used a monoclonal antibody ( mAb 3C9 ) to the T O-glycan structure ( Steentoft et al . , 2011 ) . Through Stage 10 , macrophages displayed very little T antigen staining , similar to other tissues ( Figure 1E , F ) . However , at late Stage 11 ( Figure 1—figure supplement 1A ) and early Stage 12 , when macrophages start to invade the extended germband , T antigen staining began to be enriched on macrophages moving towards and into the germband ( Figure 1E–H ) . Our results are consistent with findings showing T antigen expression in a macrophage-like pattern in late Stage 12 embryos , and on a subset of macrophages at Stage 16 ( Yoshida et al . , 2008 ) . We knocked down the core1 synthase C1GalTA required for the final step of T antigen synthesis ( Figure 1A ) ( Lin et al . , 2008; Müller et al . , 2005 ) using RNAi expression only in macrophages and observed strongly reduced staining ( Figure 1I , Figure 1—figure supplement 1M ) . We conclude that the antibody staining is the result of T antigen produced by macrophages themselves . To determine if these T O-glycans on macrophages are important for facilitating their germband invasion , we knocked down C1GalTA in macrophages with the RNAi line utilized above as well as one other and used the P element excision allele C1GalTA[2 . 1] which removes conserved sequence motifs required for activity ( Lin et al . , 2008 ) . We visualized macrophages through specific expression of fluorescent markers and observed a 25 and a 33% decrease in their number in the germband for the RNAis ( Figure 1J , K ) , and a 44% decrease in the C1GalTA[2 . 1] mutant ( Figure 1L ) . When we counted the number of macrophages sitting on the yolk next to the germband in the strongest RNAi we observed an increase ( Figure 1—figure supplement 1N ) that we also observed in the C1GalT mutant ( Figure 1—figure supplement 1O ) . The sum of the macrophages in the yolk and germband is the same in the control , RNAi knockdown ( control 136 . 5 ± 6 . 4 , RNAi 142 . 3 ± 6 . 6 , p=0 . 7 ) and mutant ( control 138 . 5 ± 4 . 9 , mutant , 142 . 3 ± 7 . 4 , p=0 . 87 ) arguing that macrophages in which C1GalTA levels are reduced cannot enter the germband but are retained on the yolk . We observed no effect on the migration of macrophages on the vnc , a route that does not require tissue invasion ( Figure 1—figure supplement 1P ) ( Campos-Ortega and Hartenstein , 1997; Evans et al . , 2010 ) . 18% of T antigen in the embryo has been found to be further modified , predominantly by glucuronic acid ( GlcA ) ( Aoki et al . , 2008 ) . Of the three GlcA transferases found in Drosophila only GlcAT-P is robustly capable of adding GlcA onto the T O-glycan structure in cells ( Breloy et al . , 2016; Itoh et al . , 2018; Kim et al . , 2003 ) . To examine if the specific defect in germband invasion that we observed by blocking the formation of T antigen is due to the need for a further elaboration by GlcA , we utilized a lethal MI{MIC} transposon insertion mutant in the GlcAT-P gene . We observed no change in the numbers of macrophages within the germband in the GlcAT-PMI05251 mutant ( Figure 1M ) and a 20% increase in the number of macrophages on the yolk ( Figure 1—figure supplement 1Q ) . Therefore , our results strongly suggest that the T antigen we observe being upregulated in macrophages as they move towards and into the germband is itself needed for efficient tissue invasion . We sought to determine which proteins could temporally regulate the increase in the appearance of T O-glycans in invading macrophages . We first considered proteins required for synthesizing the core1 structure , namely the T synthase , C1GalTA , and the UDP-Gal sugar transporter , Ugalt ( Aumiller and Jarvis , 2002 ) ( Figure 1A ) . However , q-PCR analysis of FACS sorted macrophages from Stage 9–10 , Stage 12 , and Stage 13–17 show that though both are enriched in macrophages , neither is transcriptionally upregulated before or during Stage 12 ( Figure 2A , B ) . We therefore examined the Bloomington Drosophila Genome Project ( BDGP ) in situ database looking for predicted sugar binding proteins expressed in macrophages with similar timing to the observed T antigen increase ( Tomancak et al . , 2007; Tomancak et al . , 2002 ) . We identified CG8602 , a predicted member of the Major Facilitator Superfamily ( MFS ) , a protein group defined by shared structural features , whose members are known to transport a diverse set of molecules across membranes ( Yan , 2015 ) . CG8602 contains regions of homology to known sugar responsive proteins and predicted sugar or neurotransmitter transporters ( Figure 2C ) and in a phylogenetic analysis is on a branch neighboring the SLC29 group shown to be involved in nucleoside transport ( Baldwin et al . , 2004; Perland et al . , 2017 ) . BDGP in situ hybridizations ( Tomancak et al . , 2007; Tomancak et al . , 2002 ) ( http://insitu . fruitfly . org/cgi-bin/ex/report . pl ? ftype=10&ftext=FBgn0035763 ) indicate that CG8602 RNA is maternally deposited , with expression throughout the embryo through Stage 4 after which its levels decrease , with weak ubiquitous expression continuing through Stage 9–10 . This is followed by strong enrichment in macrophages from Stage 11–12 , with apparently equivalent levels of expression in macrophages entering the germband as in those migrating along other routes such as the ventral nerve cord . We confirmed this by q-PCR analysis of FACS sorted macrophages , which detected seven-fold higher levels of CG8602 RNA in macrophages than in the rest of the embryo by Stage 9–10 and 12-fold by Stage 12 ( Figure 2D ) . These data show that RNA expression of CG8602 , an MFS protein with homology to sugar transporters , increases in macrophages preceding and during the period of invasion . To determine if CG8602 could affect T antigen levels , we examined a viable P-element insertion mutant in the 5’UTR , CG8602EP3102 ( Figure 2—figure supplement 1A ) . This insertion displays strongly reduced CG8602 expression in FACS-sorted macrophages to 15% of wild type levels , as assessed by q-PCR ( Figure 2E ) . We also created an excision allele , ∆33 , removing the 5’UTR flanking the P-element , the start methionine , and 914 bp of the ORF ( Figure 2—figure supplement 1A ) . This is a lethal allele , and the line carrying it over a balancer is very weak; exceedingly few embryos are laid and the embryos homozygous for the mutation do not develop past Stage 12 . Therefore , we did not continue experiments with this allele , and instead utilized the insertion mutant . This CG8602EP3102 P-element mutant displays decreased T antigen staining on macrophages moving toward and entering the germband ( Figure 2F ) in Stage 11 through late Stage 12 . q-PCR analysis on FACS sorted macrophages show that the reduction in T antigen levels in the mutant is not caused by changes in the RNA levels of the T synthase C1GalTA or the Ugalt Gal and GalNAc transporter ( Aumiller and Jarvis , 2002; Segawa et al . , 2002 ) ( Figure 2G ) . These results argue that CG8602 is required for enriched T antigen levels on macrophages . To assess if CG8602 could directly regulate T antigen addition , we examined if it is found in the Golgi where O-glycosylation is initiated . We first utilized the macrophage-like S2R+ cell line , transfecting a FLAG::HA or 3xmCherry labeled form of CG8602 under the control of srpHemo or the copper inducible MT promoter . We detected significant colocalization with markers for the cis-Golgi marker GMAP , the Trans Golgi Network marker Golgin 245 and the endosome markers Rab7 , Rab11 and Hrs ( Riedel et al . , 2016 ) ( Figure 2H , Figure 2—figure supplement 1C–G ) . We detected no colocalization with markers for the nucleus , ER , peroxisomes , mitochondria or lysosomes ( Figure 2H , Figure 2—figure supplement 1B , H–J ) . We confirmed the presence of CG8602 in the Golgi and endosomes in macrophages from late Stage 11 embryos through colocalization with Golgin 84 and Hrs , using cells extracted from positions in the head adjacent to the germband ( Figure 2I ) . We conclude that the T antigen enrichment on macrophages migrating towards and into the germband requires a previously uncharacterized atypical MFS with homology to sugar binding proteins that is localized predominantly to the Golgi and endosomes . We examined if CG8602 affects macrophage invasive migration . The CG8602EP3102 mutant displayed a 35% reduction in macrophages within the germband at early Stage 12 compared to the control ( Figure 3A–B , D , Figure 3—figure supplement 1A ) . The same decrease is observed when the mutant is placed over the deficiency Df ( 3L ) BSC117 that removes the gene entirely ( Figure 3D ) , arguing that CG8602EP3102 is a genetic null for macrophage germband invasion . The P element transposon insertion itself causes the migration defect because its precise excision restored the number of macrophages in the germband to wild type levels ( Figure 3D ) . Expression of the CG8602 gene in macrophages can rescue the CG8602EP3102 P element mutant ( Figure 3C–D , Figure 3—figure supplement 1A ) , and RNAi knockdown of CG8602 in macrophages can recapitulate the mutant phenotype ( Figure 3E , Figure 3—figure supplement 1B ) . Our data thus argue that CG8602 is required in macrophages themselves for germband invasion . Decreased numbers of macrophages in the extended germband could be caused by specific problems entering this region , or by general migratory defects or a decreased total number of macrophages . To examine the migratory step that precedes germband entry , we counted the number of macrophages sitting on the yolk next to the germband in fixed embryos in the CG8602EP3102 mutant . We observed a 30% decrease compared to the control ( Figure 3F ) , suggesting a defect in early dissemination . Entry into the germband by macrophages occurs between the closely apposed DE-Cadherin expressing ectoderm and the mesoderm and is accompanied by deformation of the ectodermal cells ( Ratheesh et al . , 2018 ) . We tested if reductions in DE-Cadherin could ameliorate the germband phenotype . Indeed , combining the CG8602EP3102 mutation with shgP34 which reduces DE-Cadherin expression ( Pacquelet and Rørth , 2005; Tepass et al . , 1996 ) produced a partial rescue ( Figure 3G ) , consistent with CG8602 playing a role in germband entry as well as in an earlier migratory step . There was no significant difference in the number of macrophages migrating along the vnc in late Stage 12 compared to the control in fixed embryos ( Figure 3H ) from the CG8602EP3102 mutant or from a knockdown in macrophages of CG8602 by RNAi ( Figure 3—figure supplement 1C ) , arguing against a general migratory defect . There was also no significant difference in the total number of macrophages in either case ( Figure 3—figure supplement 1D–E ) . From analyzing the CG8602 mutant phenotype in fixed embryos we conclude that CG8602 does not affect later vnc migration but is important for the early steps of dissemination and germband invasion . To examine the effect of CG8602 on macrophage speed and dynamics , we performed live imaging of macrophages labeled with the nuclear marker srpHemo-H2A::3xmCherry in control and CG8602EP3102 mutant embryos ( Figure 3—video 1 and 2 ) . We first imaged macrophages migrating from their initial position in the delaminated mesoderm up to the germband and detected a 33% decrease in speed ( 2 . 46 ± 0 . 07 μm/min in the control , 1 . 66 ± 0 . 08 μm/min in the CG8602EP3102 mutant , p=0 . 002 ) ( Figure 3I , J ) and no significant decrease in persistence ( 0 . 43 ± 0 . 02 in the control , 0 . 40 ± 0 . 01 in the mutant , p=0 . 22 ) ( Figure 3—figure supplement 1F ) . We then examined the initial migration of macrophages into the germband at late Stage 11 . We observed a range of phenotypes in the six movies we made of the mutant , with macrophages pausing at the germband edge from twice to six times as long as in the control before invading into the tissue ( Figure 3K shows average time for entry , control = 22 . 00 ± 1 . 53 min , CG8602EP3102 mutant = 102 . 0 ± 20 . 35 min ) . As we observed no change in the timing of the initiation of germband retraction ( 269 . 6 ± 9 min in control and 267 . 1 ± 3 min in mutant , p=0 . 75 ) but did observe a decreased speed of its completion in the mutant ( 107 ± 12 min from start to end of retraction in control and 133 ± 6 min for mutant p=0 . 05 ) , we only analyzed macrophages within the germband before its retraction begins . We observed a 43% reduction in macrophage speed within the germband ( 2 . 72 ± 0 . 32 μm/min in the control and 1 . 55 ± 0 . 04 μm/min in the mutant , p=0 . 02 ) ( Figure 3L , M ) . To assess this phenotype’s specificity for invasion , we used live imaging of macrophage migration along the inner vnc that occurs during the same time period as germband entry; we observed no significant change in speed ( 2 . 41 ± 0 . 06 μm/min in the control and 2 . 23 ± 0 . 01 μm/min in the mutant , p=0 . 11 ) or directionality ( 0 . 43 ± 0 . 03 in the control and 0 . 43 ± 0 . 02 in the mutant , p=0 . 9742 ) ( Figure 3—figure supplement 1G , Figure 3—video 3 ) . We conclude from the sum of our experiments in fixed and live embryos that CG8602 is important for the initial disseminatory migration out of the head and for invasive migration into and within the germband , but does not alter general migration . We name the gene minerva ( mrva ) , for the Roman goddess who was initially trapped in the head of her father , Jupiter , after he swallowed her pregnant mother who had turned herself into a fly . To assess if Minerva only affects macrophage invasion or also other types of tissue penetration in Drosophila , we examined the migration of germ cells and border cells . Germ cells move in an Integrin-independent fashion through gaps in the midgut created by ingressing formerly epithelial cells ( Devenport and Brown , 2004; Seifert and Lehmann , 2012 ) . We found no defect in germ cell migration when examining control and mrva3102 mutant embryos stained with the Vasa Ab ( Figure 3—figure supplement 1H–I ) . Border cells are born in the epithelia surrounding the ovary and then delaminate to move invasively between the nurse cells towards the oocyte ( Montell , 2003 ) , guided by the same receptor that macrophages use during their embryonic dispersal , PVR ( Duchek et al . , 2001 ) . They migrate as a tumbling collective , using invadopodia and Cadherin-based adhesion to progress ( Cai et al . , 2014; Niewiadomska et al . , 1999 ) . mrva is expressed in dissected control ovaries and the mrva3102 mutant reduces the levels of mrva RNA in the ovary by 70% , similar to the reduction observed in macrophages ( Figure 3—figure supplement 1J ) . We identified border cells by staining with DAPI to detect their clustered nuclei . We observed no significant change in border cell migration towards the oocyte in the mrva3102 mutant compared to the control ( Figure 3—figure supplement 1K–L ) . These results support the conclusion that Mrva is not generally required for all migratory cells that move confined through tissues during development , but specifically for the invasion of macrophages , which is an Integrin-dependent process ( Siekhaus et al . , 2010 ) . We set out to determine if Minerva induces T glycoforms on particular proteins . We first conducted a Western Blot with a mAb to T antigen on whole embryo extracts . We used the whole embryo because we were unable to obtain enough protein from FACS sorted macrophages or to isolate CRISPR-induced full knockouts of minerva in the S2R+ macrophage like cell line . We observed that several bands detected with the anti-T mAb were absent or reduced in the minerva mutant ( Figure 4A ) , indicating an effect on the T antigen modification of a subset of proteins . We wished to obtain a more comprehensive view of the proteins affected by Minerva . Since there is little information about Drosophila O-glycoproteins and O-glycosites ( Schwientek et al . , 2007; Aoki and Tiemeyer , 2010 ) , we used lectin-enriched O-glycoproteomics to identify proteins displaying T and Tn glycoforms in Stage 11/12 embryos from wild type and mrva3102 mutants ( Figure 4—figure supplement 1A ) . We labeled tryptic digests of embryonic protein extracts from control or mutant embryos with stable dimethyl groups carrying medium ( C2H2D4 ) or light ( C2H6 ) isotopes respectively to allow each genotype to be identified in mixed samples ( Boersema et al . , 2009; Schjoldager et al . , 2012; Schjoldager et al . , 2015 ) . The pooled extracts were passed over a Jacalin column to enrich for T and Tn O-glycopeptides; the eluate was analyzed by mass spectrometry to identify and quantify T and Tn modified glycopeptides in the wild type and the mutant sample through a comparison of the ratio of the light and medium isotope labeling channels for each glycopeptide ( see Figure 4—figure supplement 1B–C for example spectra ) . In the wild type we identified T and Tn glycopeptides at 936 glycosites derived from 270 proteins ( Supplementary file 1 and Figure 4B ) . 62% of the identified O-glycoproteins and 77% of identified glycosites contained only Tn O-glycans . 33% of the identified O-glycoproteins and 23% of glycosites displayed a mixture of T or Tn O-glycans , and 5% of identified O-glycoproteins and 4% of glycosites had solely T O-glycans ( Figure 4C ) . In agreement with previous studies ( Steentoft et al . , 2013 ) , only one glycosite was found in most of the identified O-glycoproteins ( 44% ) ( Figure 4D ) . In 20% we found two sites , and some glycoproteins had up to 27 glycosites . The identified O-glycosites were mainly on threonine residues , ( 78 . 5% ) with some on serines ( 21 . 2% ) and very few on tyrosines ( 0 . 3% ) ( Figure 4—figure supplement 1D ) . Metabolism , cuticle development , and receptors were the most common functional assignments for the glycoproteins ( Figure 4—figure supplement 1E ) . We sought to assess the changes in glycosylation in the mrva mutant . A majority of the quantifiable Tn and T O-glycoproteome was unaltered between the wild type and the mrva3102 mutant , with only 63 proteins ( 23% ) showing more than a three-fold change and 18 ( 6% ) a ten-fold shift ( Figure 4F ) . We observed both increases and decreases in the levels of T and Tn modification on proteins in the mutant ( Figure 4F–G , Supplementary file 1 and 2 ) , but a greater number of proteins showed decreased rather than increased T antigen levels . 67% of the vertebrate orthologs of Drosophila proteins displaying shifts in this O-glycosylation have previously been linked to cancer ( Figure 4H , Supplementary file 2 ) . These proteins were affected at specific sites , with 40% of glycosites on these proteins changed more than three fold and only 14% more than ten fold . The glycosite shifts in T antigen occurred either without significant alterations in Tn ( 33% of glycosites had only decreased T antigen , 17% of glycosites had only increased T antigen ) or with changes in T antigen occurring in the same direction as the changes in Tn ( 22% of glycosites both Tn and T antigen increased , 22% of glycosites both Tn and T decreased ) ( Supplementary file 2 ) . Only 1% of glycosites displayed decreased T antigen with a significant increase in Tn . Interestingly , a higher proportion of the glycoproteins with altered O-glycosylation in the mrva3102 mutant had multiple glycosites than the general glycoproteome ( Figure 4D ) ( p value=0 . 005 for ten-fold changes ) . We conclude that Minerva affects O-glycosylation occupancy on a small subset of O-glycoproteins , many of whose vertebrate orthologs have been linked to cancer , with both T and Tn O-glycopeptides being affected . Given that blocking Tn to T conversion through the knockdown of the C1GalTA enzyme resulted in a germband invasion defect , we examined the known functions of the 18 proteins with lower T antigen in the absence of Minerva to distinguish which processes Minerva could influence to facilitate invasion ( Figure 4H ) . We excluded two proteins involved in eggshell and cuticle production . To spot proteins whose reduced T antigen-containing glycopeptides are caused directly by alterations in glycosylation rather than indirectly by decreased protein expression in the mrva mutant , we checked if glycosylation at other identified glycosites was unchanged or increased . We identified ten such proteins , several of which were in pathways that had been previously linked to invasion in vertebrates . Qsox1 , a predicted sulfhydryl oxidase required for the secretion and thus potential folding of EGF repeats ( Tien et al . , 2008 ) showed the strongest alterations of any protein , with a 50-fold decrease in T antigen levels in the mrva mutant ( Figure 4I ) . The mammalian ortholog QSOX1 has been shown to affect disulfide bond formation , is overexpressed in some cancers , promotes Matrigel invasion , and can serve as a negative prognostic indicator in human cancer patients ( Chakravarthi et al . , 2007; Katchman et al . , 2011; Lake and Faigel , 2014 ) . Dtg , with a 13-fold reduction in T antigen ( Hodar et al . , 2014 ) , and Put with a five-fold reduction ( Letsou et al . , 1995 ) respond to signaling by the BMP-like ligand , Dpp . Dpp signaling directs histoblast invasion in the fly ( Ninov et al . , 2010 ) . Gp150 shows a four fold decrease in T antigen and modulates Notch signaling ( Fetchko et al . , 2002; Li et al . , 2003 ) . Notch and BMP promote invasion and metastasis in mice ( Bach et al . , 2018; Garcia and Kandel , 2012; Owens et al . , 2015; Pickup et al . , 2015; Sahlgren et al . , 2008; Sonoshita et al . , 2011 ) . We conclude that Mrva is required to increase T O-glycans on a subset of the glycosites of selected glycoproteins involved in protein folding , glycosylation and signaling in pathways frequently linked to promoting cancer metastasis . Its strongest effect is on a predicted sulfhydryl oxidase , the Drosophila ortholog of the mammalian cancer protein , QSOX1 . We wished to determine how Qsox1 might affect Drosophila macrophage germband invasion . Embryos from the KG04615 P element insertion in the 5’UTR of the qsox1 gene displayed 42% fewer macrophages in the germband compared to the control ( Figure 5A , B ) with an increase in macrophages remaining on the yolk ( Figure 5—figure supplement 1A ) . We observed a small decrease in migration along the vnc ( Figure 5—figure supplement 1B ) and no change in total macrophage numbers in these embryos ( Figure 5—figure supplement 1C ) . These migration phenotypes were also observed in embryos in which RNAi line v108288 knocked down qsox1 only in macrophages ( Figure 5C , Figure 5—figure supplement 1D–E ) . We then conducted live imaging ( Figure 5D , Figure 5—video 1 ) ( compare to control shown in Figure 3—video 1 ) to examine how the qsox1KG04615 mutant affected the dynamics of macrophage migration . During the movement of macrophages labeled with the nuclear marker srpHemo-H2A::3xmCherry from their initial position up to the germband we detected an 18% decrease in speed ( Figure 5E ) ( 2 . 46 ± 0 . 07 μm/min in the control , 2 . 02 ± 0 . 03 μm/min in the qsox1KG046152 mutant , p=0 . 006 , n = 3 ) and no significant decrease in persistence ( Figure 5—figure supplement 1F ) ( 0 . 43 ± 0 . 02 in the control , 0 . 39 ± 0 . 01 in the mutant , p=0 . 13 ) . Macrophages in the qsox1 mutant were delayed twice as long at the germband edge before entering ( Figure 5F ) ( time to entry 22 . 00 ± 1 . 53 min in the control and 49 . 67 ± 9 . 33 min in the qsox1KG046152 mutant , n = 3 ) . Once in , they moved within the germband with a 17% slower speed , a reduction that was not statistically significant ( Figure 5G ) ( 2 . 72 ± 0 . 32 μm/min in the control , 2 . 27 ± 0 . 20 μm/min in the qsox1KG046152 mutant , p=0 . 30 , n = 3 ) . We conclude that Qsox1 aids the disseminatory migration of macrophages but is most strongly required for their initial invasion into the germband tissues . We wished to examine how Qsox1 could be exerting this effect on macrophage tissue entry . Vertebrate QSOX1 has been shown to localize to the Golgi and act as a sulfhydryl oxidase , catalyzing disulfide bond formation and protein folding ( Alon et al . , 2012; Chakravarthi et al . , 2007; Heckler et al . , 2008; Hoober et al . , 1996 ) . The Drosophila protein has been shown to be required for the secretion of multimerized EGF domains and was hypothesized to act redundantly with ER oxidoreductin-like-1 to form disulfide bonds ( Tien et al . , 2008 ) . We found that an HA-tagged form of Qsox1 transfected into the Drosophila macrophage like cell line , S2R+ , colocalizes little with markers for the ER , and considerably with those for Golgi and endosomes ( Figure 5H , Figure 5—figure supplement 1G–I ) . We also observed significant colocalization with 3xmCherry-tagged Mrva ( Figure 5H , Figure 5—figure supplement 1J ) . Vertebrate QSOX1 can be cleaved from its transmembrane domain to allow secretion ( Rudolf et al . , 2013 ) , and has been shown in vitro to be required extracellularly for the incorporation of laminin produced by fibroblasts into the extracellular matrix ( ECM ) , thereby supporting efficient cancer cell migration ( Ilani et al . , 2013 ) . Drosophila Qsox1 also has a transmembrane domain , yet we detected an HA-tagged form in the media after transfection into S2R+ cells ( Figure 5I ) , indicating that it can be secreted . To examine if Drosophila Qsox1 might also affect Laminin , we stained mrva3102 and qsox1KG046152 mutant embryos with an antibody against Laminin A ( LanA ) ( Figure 5—figure supplement 1K ) . In both mutants we observed increased amounts of LanA inside and somewhat higher levels adjacent to the macrophages , but no significant alteration at the cell edges compared to the control ( Figure 5J , Figure 5—figure supplement 1L–N ) . We conclude that Drosophila Qsox1 can be secreted but is also found in the Golgi and endosomes like Mrva , and that both proteins affect LanA , a component of the ECM . To determine if our studies could ultimately be relevant for mammalian biology and therefore also cancer research , we searched for a mammalian ortholog . MFSD1 from mus musculus shows strong sequence similarity with Mrva , with 50% of amino acids displaying identity and 68% conservation ( Figure 6A , Figure 6—figure supplement 1A ) . A transfected C-terminally GFP-tagged form ( Figure 6—figure supplement 1B ) showed localization to the secretory pathway , colocalizing with the Golgi marker GRASP65 in murine MC-38 colon carcinoma , 4T1 breast cancer cells and LLC1 . 1 lung cancer ( Figure 6B–C , Figure 6—figure supplement 1C–E ) and with Golgi and endosomal markers in B16-BL6 melanoma cells ( Figure 6C , Figure 6—figure supplement 1F ) . mmMFSD1 expression in macrophages in mrva3102 mutant embryos can completely rescue the germband invasion defect ( Figure 6D–E ) . This macrophage-specific expression of MFSD1 also resulted in higher levels of T antigen on macrophages when compared to those in mrva3102 mutants ( Figure 6F–G ) . Thus MFSD1 not only displays localization in the Golgi apparatus in multiple types of mammalian cancer but can also rescue O-glycosylation and migration defects when expressed in Drosophila , arguing that the functions Mrva carries out to promote invasion into the germband are conserved up to mammals . O-glycosylation is one of the most common posttranslational modifications , yet the intrinsic technical challenges involved in identifying O-glycosites and altered O-glycosylation on a proteome-wide level has hampered the discovery of biological functions ( Levery et al . , 2015 ) . Here we provide two important new advances for the field . First , we identify a key regulator of this O-glycosylation , Minerva , with an unexpected role for a member of the major facilitator superfamily . Our demonstration that this conserved protein affects invasion and the appearance of the cancer-associated core1 T glycoform on a set of proteins connected to invasion provides a new perspective on T glycoform regulation and may have implications for cancer . Second , we define the GalNAc-type O-glycoproteome of Drosophila embryos . As O-glycosites cannot as yet be reliably predicted , our proteomic characterization in a highly genetically accessible organism will permit future studies on how glycosylation affects cell behavior; we highlight T and Tn O-glycosylated receptors in Supplementary file 3 to further this goal . Our identification of a MFS family member as a regulator of O-glycosylation is surprising . MFS family members can serve as transporters and shuttle a wide variety of substrates ( Quistgaard et al . , 2016; Reddy et al . , 2012 ) . Minerva displays homology to sugar transporters and is localized to the Golgi and endosomes . Minerva could thus affect O-glycosylation in the Golgi through substrate availability . However , the lower and higher levels of glycosylation in the mrva3102 mutant we observe are hard to reconcile with this hypothesis . Given that the changes in T antigen on individual glycosites in the mrva mutant are found either with no significant change in Tn or with a change in the same direction ( Supplementary file 1 and 2 ) , regulation appears to occur at the initial GalNAc addition on the protein subset as well as on further T antigen elaboration . 95% of the proteins with 10-fold altered glycosylation in the mrva mutant had multiple O-glycosylation sugar modifications compared to 56% of the general O-glycoproteome . Greatly enhanced glycosylation of protein sequences containing an existing glycan modification is observed for some GalNAc-Ts due to a lectin domain ( Hassan et al . , 2000; Kubota et al . , 2006; Revoredo et al . , 2016 ) and Minerva could theoretically affect such a GalNAc-T in Drosophila . Alternatively , Minerva , while in the ‘outward open’ conformation identified for MFS structures ( Quistgaard et al . , 2016 ) , may itself have a lectin-like interaction with Tn and T glycoforms that have already been added on a loop of particular proteins . Minerva’s binding could open up the target protein’s conformation to increase or block access to other potential glycosites and thus affect the final glycosylation state on select glycoproteins . The changes we see in O-glycosylation are also likely due to a combination of Minerva’s direct and indirect effects . O-GalNAc modification of vertebrate Notch can affect Notch signaling during development ( Boskovski et al . , 2013 ) ; the Drosophila ortholog of the responsible GalNAc transferase is also essential for embryogenesis ( Bennett et al . , 2010; Schwientek et al . , 2002 ) . A GalNAcT in Xenopus can glycosylate a peptide corresponding to the ActR IIB receptor and inhibit Activin and BMP type signaling ( Herr et al . , 2008; Voglmeir et al . , 2015 ) . Thus the changed glycosylation we observe on components of the Notch and Dpp pathways could alter transcription ( Hamaratoglu et al . , 2014; Ntziachristos et al . , 2014 ) , shifting protein levels and thereby changing the ratio of some glycopeptides in the mrva mutant relative to the wild type . Proteins in which glycosylation at other sites is unchanged or changed in the opposite direction are those most likely to be directly affected by Minerva . Such proteins include ones involved in protein folding and O-glycan addition and removal ( Figure 4H ) ( Tien et al . , 2008 ) . If changes in the glycosylation of these proteins alters their specificity or activity , some of the shifts we observe in our glycoproteomic analysis could be indirect in a different way; an initial effect of Minerva on the glycosylation of regulators of protein folding and glycosylation could change how these primary Minerva targets affect the glycosylation of a second wave of proteins . The truncated immature core1 T and Tn O-glycans are not usually present in normal human tissues but exposure of these uncapped glycans has been found on the majority of cancers and serves as a negative indicator of patient outcome ( Fu et al . , 2016; Springer , 1984 ) . Increases in Tn antigen due to a shift in GalNAcT localization to the ER promote invasion and metastasis ( Gill et al . , 2013 ) . An antibody against T antigen has decreased the metastatic spread of cancer cells in mice ( Heimburg et al . , 2006 ) . Here we further strengthen the case for a causative relationship between T antigen modification and the invasive migration that underlies metastasis . The transient appearance of T antigen in human fetuses ( Barr et al . , 1989 ) and the conserved function of Minerva lead us to propose that the change in O-glycosylation in cancer represents the reactivation of an ancient developmental program for invasion . Our embryonic glycoproteome analysis identifies 106 T antigen modified proteins , a very large set to investigate . However , the absence of Mrva causes invasion defects and deficits in T antigen modification on only 10–20 proteins; these include components involved in protein folding , glycosylation modification , and the signaling pathways triggered by Notch and the BMP family member , Dpp . Our working model is that the defect in germband tissue invasion seen in the mrva mutant is caused by the absence of T antigen on this group of proteins that act coordinately ( Figure 6H ) . 56% of these have vertebrate orthologs , and 55% of those have already been linked to cancer and metastasis . The vertebrate ortholog of Qsox1 , the protein with the largest changes in T antigen in the mrva mutant , can enhance cancer cell invasion in in vitro assays and higher levels of the protein have been associated with poor patient outcomes ( Katchman et al . , 2013; Katchman et al . , 2011 ) . We find that the strongest effect of Drosophila Qsox1 on macrophage migration is to reduce the time by two fold that macrophages take sitting at the germband edge before they successfully begin to invade into the germband tissues . We also observe in qsox1 and mrva mutants that LanA levels are higher within the macrophages and somewhat elevated near but not at the macrophage cell edges . This could be due to some combination of the following shifts in cellular processes: an increase in LanA production , a decrease in its degradation , a slowing of its secretion or a speeding of its diffusion . We base our model on the functions that have been previously defined for the Qsox1 sulfhydryl oxidase family , in integrating laminin into the ECM ( Ilani et al . , 2013 ) and aiding secretion of EGF domains ( Tien et al . , 2008 ) which are found in Drosophila Laminins . If Qsox1 is needed for the efficient secretion and integration of LanA into the ECM , its absence could result in a less robustly cross-linked matrix . ECM crosslinking has been shown to enhance Integrin signaling , focal adhesion formation , and invasion of mammalian tumor cells ( Levental et al . , 2009 ) . In its absence Drosophila macrophages which utilize Integrin during invasion ( Siekhaus et al . , 2010 ) and whose invasive migration is accompanied by deformation of the flanking tissue ( Ratheesh et al . , 2018 ) , could be unable to generate sufficient traction forces to enter . Indeed , mutating another subunit of the Drosophila Laminin trimer , LanB1 , reduces both normal LanA deposition and germband invasion by macrophages ( Matsubayashi et al . , 2017; Sánchez-Sánchez et al . , 2017 ) . A determination of the effect of Minerva’s regulation awaits a characterization of Qsox1 mutated such that it is incapable of being modified by T antigen on the Mrva-dependent sites . Nonetheless , the similarity of the changes in LanA we observe in the mrva3102 and qsox1KG046152 mutant supports the conclusion that Mrva dependent T-antigen modification of Qsox1 is necessary for its activity on some substrates . Given that mrva3102 mutants take even longer than qsox1KG04615 to enter germband tissue and display much stronger defects thereafter , we propose that T antigen modifications on other proteins are also crucial for tissue entry , and underlie the defect in invasive migration within the germband . Minerva’s vertebrate ortholog , MFSD1 , can rescue macrophage migration defects and restores higher T antigen levels . Tagged versions of Minerva’s vertebrate ortholog , MFSD1 , detected the protein in lysosomes in HeLa and rat liver cells ( Chapel et al . , 2013; Palmieri et al . , 2011 ) . In four metastasizing mouse tumor cell lines we find MFSD1 mainly in the Golgi , where O-glycosylation is known to occur ( Bennett et al . , 2012 ) . We do not yet know if invasion and metastasis is altered by the absence of MFSD1 but will be testing this in future work . Akin to how kinases add phospho-groups to affect a set of proteins and orchestrate a particular cellular response , we propose that Minerva in Drosophila macrophages and its vertebrate ortholog MFSD1 in cancer trigger changes in O-glycosylation that coordinately modulate , activate and inhibit a protein group to affect cellular dissemination and tissue invasion . Flies were raised on food bought from IMBA ( Vienna , Austria ) which contained the standard recipe of agar , cornmeal , and molasses with the addition of 1 . 5% Nipagin . Adults were placed in cages in a Percival DR36VL incubator maintained at 29°C and 65% humidity; embryos were collected on standard plates prepared in house from apple juice , sugar , agar and Nipagin supplemented with yeast from Lesaffre ( Marcq , France ) on the plate surface . Embryo collections for fixation ( 7 hr collection ) as well as live imaging ( 4 . 5 hr collection ) were conducted at 29°C . Figure 1D-H: w-; +; srpHemo-3xmCherry . Figure 1I-K: Control: w- P ( w+ ) UAS-dicer/w-; P{attP , y[+] , w[3`]/+; srpHemo-Gal4 UAS-GFP UAS-H2A::RFP/+ . C1GalTA RNAi: w- P ( w+ ) UAS-dicer2/w-; RNAi C1GalTA ( v110406 ) /+; srpHemo-Gal4 UAS-GFP UAS-H2A:RFP/+ . Figure 1L: Control: w-; +; srpHemo-H2A::3xmCherry . C1GalLTA mutant: w-; C1GalTA2 . 1; srpHemo-H2A::3xmCherry . Figure 1M: Control: w-; srpHemo-H2A::3xmCherry . GlcAT-P mutant: w-; srpHemo-H2A::3xmCherry , Mi{MIC}GlcAT-PMI05251 . Figure 1—figure supplement 1A–L: w-; +; srpHemo-3xmCherry . Figure 1—figure supplement 1M , N , P: Control: w- UAS-Dicer2/w-; P{attP , y[+]w[3`]/+; srpHemo-Gal4 UAS-GFP UAS-H2A::RFP/+ . C1GalTA RNAi: w-UAS-Dicer2/w-; RNAi C1GalTA ( v110406 ) /+; srpHemo-Gal4 UAS-GFP UAS-H2A::RFP/+ . Figure 1—figure supplement 1O: Control: w-; +; srpHemo-H2A::3xmCherry . C1GalTA mutant: w-; C1GalTA2 . 1; srpHemo-H2A::3xmCherry . Figure 1—figure supplement 1P: Control: w- UAS-Dicer2/w-; P{attP , y[+]w[3`]/+; srpHemo-Gal4 UAS-GFP UAS-H2A::RFP/+ . C1GalTA RNAi: w-UAS-Dicer2/w-; RNAi C1GalTA ( v2826 ) /+; srpHemo-Gal4 UAS-GFP UAS-H2A::RFP/+ . Figure 1—figure supplement 1Q: Control: w-; srpHemo-H2A::3xmCherry . GlcAT-P mutant: w-; srpHemo-H2A::3xmCherry; Mi{MIC}GlcAT-PMI05251 . Figure 2A , B , D: w-; +; srpHemo-3xmCherry . Figure 2E , F , G: Control: w-; +; srpHemo-3xmCherry . CG8602 mutant: w-; +; srpHemo-3xmCherry , P{EP}CG86023102 . Figure 2I: w-; srpHemo-Gal4; UAS-CG8602::FLAG::HA . Figure 3A: w-; +; srpHemo-H2A::3xmCherry . Figure 3B: w-; +; srpHemo-H2A::3xmCherry , P{EP}CG86023102 . Figure 3C: w-; srpHemo-CG8602; srpHemo-H2A::3xmCherry P{EP}CG86023102 . Figure 3D: Control: w-; srpHemo-Gal4 UAS-mCherry::nls; + . CG8602 ( Mrva ) mutant: w-; srpHemo-Gal4 UAS-mCherry::nls; P{EP}CG86023102 , Df cross: w-; srpHemo-Gal4 UAS-mCherry:nls; P{EP}CG86023102/Df ( 3L ) BSC117 . Rescue: w-; srpHemo-Gal4 UAS-mCherry:nls; UAS-CG8602::FLAG::HA P{EP}CG86023102 . Precise excision: srpHemo-Gal4 UAS-mCherry:nls; P{EP}CG86023102Δ32 . Figure 3E: Control: w- P ( w+ ) UAS-dicer/+; +; srpHemo-Gal4 UAS-GFP UAS-H2A:RFP/+ . Mrva RNAi: w- UAS-dicer2/w-; RNAi CG8602 ( v101575 ) /+; srpHemo-Gal4 UAS-GFP UAS-H2A:RFP/+ . Figure 3F: Control: w-; srpHemo-Gal4 UAS-mCherry::nls; + . Mrva mutant: w-; srpHemo-Gal4 UAS-mCherry::nls; P{EP}CG86023102 . Figure 3G: Control: w-; +; srpHemo-3xmCherry . Mrva mutant: w-; +; srpHemo-3xmCherry P{EP}CG86023102 . Cadherin Mrva double mutant: w-; shgP34; srpHemo-3xmCherry P{EP}CG86023102 . Figure 3H: Control: w-; +; srpHemo-3xmCherry . Mrva mutant: w-; +; srpHemo-3xmCherry P{EP}CG86023102 . Figure 3I-M: Control: w-; +; srpHemo-H2A::3xmCherry . Mrva mutant: w-; +; srpHemo-H2A::3xmCherry P{EP}CG86023102 . Figure 3-figure supplement 1A: Control: w-; +; srpHemo-H2A::3xmCherry . Mrva mutant: w-; +; srpHemo-H2A::3xmCherry P{EP}CG86023102 . Rescue: w-; srp-CG8602; srpHemo-H2A::3xmCherry P{EP}CG86023102 . Figure 3-figure supplement 1B , C , E: Control: w-; +; srpHemo-Gal4 UAS-GFP UAS-H2A:RFP/+ . Mrva RNAi: w-; RNAi CG8602 ( v101575 ) /+; srpHemo-Gal4 UAS-GFP UAS-H2A::RFP/+ . Figure 3-figure supplement 1D , F-G: Control: w-; +; srpHemo-H2A::3xmCherry . Mrva mutant: w-; +; srpHemo-H2A::3xmCherry P{EP}CG86023102 . Figure 3-figure supplement 1H-L: Control: w-; +; srpHemo-3xmCherry . Mrva mutant: w-; +; srpHemo-3xmCherry P{EP}CG86023102 Figure 4A-I: Control: w-; + , srpHemo-3xmCherry . Mrva mutant: w-; + , srpHemo-3xmCherry P{EP}CG86023102 . Figure 5A-B: Control: w-; +; srpHemo-3xmCherry . Qsox1 mutant: w-;P{SUPor-P}Qsox1KG04615; srpHemo-3xmCherry . Figure 5C: w/y , w[1118]; P{attP , y[+] , w[3`]}/srpHemo-Gal4; srpHemo-H2A::3xmCherry/+ . Qsox1 RNAI: w-/y , w[1118]; v108288/srpHemo-Gal4; srpHemo-H2A::3xmCherry/+ . Figure 5D-G: Control: w-; +; srpHemo-H2A::3xmCherry . Qsox1 mutant: w-;P{SUPor-P}Qsox1KG04615; srpHemo-H2A::3xmCherry . Figure 5J: Control: w-; +; srpHemo-3xmCherry . Mrva mutant: w-; +; srpHemo-3xmCherry P{EP}CG86023102 . Qsox1 mutant: w-; P{SUPor-P}Qsox1KG04615;srpHemo-3xmCherry . Figure 5-figure supplement 1A-B: Control: w-; +; srpHemo-3xmCherry . Qsox1 mutant: w-;P{SUPor-P}Qsox1KG04615; srpHemo-3xmCherry . Figure 5-figure supplement 1C , F: w-; +; srpHemo-H2A::3xmCherry , w-; P{SUPor-P}Qsox1KG04615; srpHemo-H2A::3xmCherry . Figure 5-figure supplement 1D-E: Control: w-/y , w[1118]; P{attP , y[+] , w[3`]}/srpHemo-Gal4; srpHemo-H2A::3xmCherry/+ . Qsox1 RNAi: w-/y , w[1118]; v108288/srpHemo-Gal4; srpHemo-H2A::3xmCherry/+ . Figure 5-figure supplement 1K-N: Control: w-; +; srpHemo-3xmCherry . Mrva mutant: w-; +; srpHemo-3xmCherry P{EP}CG86023102 , Qsox1 mutant: w-; P{SUPor-P}Qsox1KG04615; srpHemo-3xmCherry . Figure 6D: w-; srpHemo-MFSD1; srpHemo-H2A::3xmCherry P{EP}CG86023102 . Figure 6E: Control: w-; +; srpHemo-H2A::3xmCherry . Mrva mutant: w-; +; srpHemo-H2A::3xmCherry P{EP}CG86023102 . MFSD1 rescue: w-; srpHemo-MFSD1; srpHemo-H2A::3xmCherry P{EP}CG86023102 . Figure 6F , G: Control: w-; +; srpHemo-3xmCherry . Mrva mutant: w-; +; srpHemo-3xmCherry P{EP}CG86023102 . MFSD1 rescue: w-; srpHemo-MFSD1; srpHemo-3xmCherry P{EP}CG86023102 . Embryos were collected on apple juice plates from between 6 and 8 . 5 hr at 29°C . Embryos were incubated in 50% Chlorox ( DanClorix ) for 5 min and washed . Embryos were fixed with 17% formaldehyde/heptane for 20 min followed by methanol or ethanol devitellinization except for T antigen analysis , when embryos were fixed in 4% paraformaldehyde/heptane . Fixed embryos were blocked in BBT ( 0 . 1M PBS + 0 , 1% TritonX-100 +0 , 1% BSA ) for 2 hr at RT . Antibodies were used at the following dilutions: α-T antigen ( Steentoft et al . , 2011 ) 1:5 , α-GFP ( Aves Labs Inc . , Tigard , Oregon ) 1:500; α-LanA ( Kumagai et al . , 1997 ) ( a gift from Stefan Baumgartner ) 1:500; α-Vasa ( Aruna et al . , 2009 ) ( DSHB , deposited by A . Sprading/D . Williams ) 1:25; and incubated overnight at 4°C ( GFP ) or room temperature ( T antigen , LanA ) . Afterwards , embryos were washed in BBT for 2 hr , incubated with secondary antibodies ( Thermo Fisher Scientific , Waltham , Massachusetts , USA ) at RT for 2 hr , and washed again for 2 hr . Vectashield ( Vector Laboratories , Burlingame , USA ) was then added . After overnight incubation in Vectashield at 4°C , embryos were mounted on a slide and imaged with a Zeiss Inverted LSM700 Confocal Microscope using a Plan-Apochromat 20X/0 . 8 Air Objective or a Plan-Apochromat 63X/1 . 4 Oil Objective . 3–5 day old females were fed with yeast for 2 days at 25°C . For ovary dissection , females were anesthetized using the FlyNap anesthetic kit ( Carolina , Burlington , NC , USA ) and further transferred to ice cold PBS in which ovaries were extracted with pre-cleaned forceps . Individual ovaries were fixed in 4% Paraformaldehyde/PBS at room temperature ( RT ) for 20 min with agitation . Three wash steps with PBS at RT for 10 min were performed and individual ovaries were incubated in PBS supplemented with 0 . 1% of Triton X-100 ( PBT ) for 10 min at RT to allow permeabilization of the tissue . Ovaries were incubated in phalloidin-A488 ( Thermo Fisher ) diluted in PBT ( 1:300 ) overnight at 4°C . After being washed with PBT and PBS , ovaries were mounted in Vectashield + DAPI ( LifeTechnologies , Carlsbad , USA ) . Ovaries were imaged as a Z-series ( 1 µm apart ) with a Plan-Apochromat 20X/0 . 8 Air Objective on a Zeiss LSM700 inverted microscope . Images were acquired from stage 10 oocytes and maximum-intensity projections were created using ImageJ ( NHI , USA ) . Border cells were identified by the clustered nuclei and their enriched actin staining . Border cell migration was quantified in the DAPI images as the percentage observed relative to the expected migration to the edge of the oocyte for these cells in stage 10 oocytes . Measurements were performed using ImageJ software ( NIH , USA ) . Embryos were fixed with 10% formaldehyde/heptane and devitellinized with Ethanol . Blocking was conducted in BBT for 2 hr at room temperature . A FITC-labeled lectin kit #2 ( EY laboratories , San Mateo , CA , USA ) was utilized ( table below summarizes abbreviations of used lectins ) . Each lectin was diluted to 1:25 and incubated with fixed embryos overnight at room temperature ( RT ) . Embryos were washed in BBT for 2 hr at RT and Vectashield was added . After overnight incubation at 4°C , embryos were mounted on a slide and imaged with a Zeiss Inverted LSM700 Confocal Microscope using a Plan-Apochromat 63X/1 . 4 Oil Objective . Macrophages in late Stage 11 embryos were imaged at germband entry and evaluated by eye for enriched staining on macrophages compared to other tissues . LectinPeanut agglutininUlex europaeus agglutininWheat germ agglutininGriffonia simplicifolia agglutinin IMaclura pomifera agglutininGriffonia simplicifolia agglutinin IIAbbreviationPNAUEA-IWGAGS-IMPAGS-IILectinSoybean agglutininDolichos biflorus agglutininConcanavalin AHelix pomatia agglutininLimulus poly-phenus agglutininBauhinia purpurea agglutininAbbreviationSBADBAConAHPALPABPA Embryos were bleached in 50% Chlorox in water for 5 min at RT . Stage late 11/early 12 embryos were lined up and then glued to 50 mm Dish No . 0 Coverslip , 14 mm Glass Diameter , Uncoated dish ( Zeiss , Germany ) . Cells from the germband margin were extracted using a ES Blastocyte Injection Pipet ( spiked , 20 μm inner diameter , 55 mm length; BioMedical Instruments , Germany ) . Extracted cells were placed in Schneider’s medium ( Gibco , Dublin , Ireland ) supplemented with 20% FBS ( Sigma-Aldrich , Saint Louis , Missouri , USA ) . Extracted macrophages were collected by centrifugation at 500 g for 5 min at room temperature . The cell pellet was resuspended in a small volume of Phospho-buffered saline ( PBS ) and smeared on a cover slip . The cell suspension was left to dry before cells were fixed with 4% paraformaldehyde in 0 . 1M Phosphate Buffer for 20 min at room temperature . Cells were washed 3 times in 0 . 1M PBS and permeabilized in 0 . 5% Triton-X 100 in PBS . Cells were blocked for 1 hr at room temperature in 20% Fetal Bovine Serum +0 . 25% Triton X-100 in PBS . Primary antibodies were diluted in blocking buffer: anti-HA ( Roche , Basel , Switzerland ) 1:50 , anti-Golgin 84 , 1:25 , anti-Calnexin 99a 1:25 , anti-Hrs . 8 . 2 1:25 or anti-Rab7 1:25 all from DSHB ( Riedel et al . , 2016 ) , and incubated for 1 hr at room temperature . Cells were then washed 5 times in blocking buffer . Secondary antibodies were diluted in blocking buffer: anti-rat 633 1:300 , anti-mouse 488 1:300 ( both from ThermoFisher Scientific , Waltham , Massachusetts , USA ) . Secondary antibodies were incubated for 1 hr at room temperature . Cells were washed 5 times in PBS + 0 . 1% Triton X-100 and mounted in VectaShield + DAPI ( LifeTechnologies , Carlsbad , USA ) utilized at 1:75 . S2R+ cells ( a gift from Frederico Mauri of the Knoblich laboratory at IMBA , Vienna ) were grown in Schneider’s medium ( Gibco ) supplemented with 10% FBS ( Gibco ) and transfected with PTS1-GFP ( a gift from Dr . McNew ) and/or the srpHemo-CG8602::3xmCherry construct using Effectene Tranfection Reagent ( Qiagen , Hilden , Germany ) following the manufacturer’s protocol . Transfected S2R+ cells were grown on Poly-L-Lysine coated coverslips ( ThermoFisher Scientific , Waltham , Massachusetts , USA ) in complete Schneider’s medium ( Gibco ) supplemented with 10% FBS ( Sigma-Aldrich , Saint Louis , Missouri , USA ) and 1% Pen/Strep ( Gibco ) to a confluency of 60% . To visualize lysosomes , cells were incubated with Lysotracker 75 nM Green DND-26 ( Invitrogen ) in complete Schneider’s medium for 30 min at 25°C . Cells were washed in complete Schneider’s medium 3 times before imaging on an inverted LSM-700 ( Zeiss ) . To visualize mitochondria , mitotracker Green FM ( Invitrogen , Carlsbad , CA , USA ) was diluted in prewarmed Schneider’s medium supplemented with 1% Pen/Strep to a concentration of 250 nM . Cells were incubated in the Mitotracker solution for 45 min at 25°C . Cells were then washed 3 times in complete Schneider’s medium before imaging . To visualize Golgi , ER , early and late endosomes as well as the nucleus , S2R+ cells were transfected with MT-CG8602::FLAG::HA ( DGRC: FMO06045 ) or MT-Qsox1::FLAG::HA ( DGRC: FMO06379 ) with Effectene Tranfection Reagent ( Qiagen ) following the manufacturer’s protocol . 24 hr after transfection gene expression was induced by addition of 1 mM Cu2SO4 ( Sigma ) and cells were incubated for an additional 24 hr . Cells were then fixed in 4% PFA ( Sigma ) in 0 . 1M PB for 20 min at room temperature , permeabilized in 0 . 5% Triton X-100 ( Sigma ) in PBS for 15 min and blocked for 2 hr in 20% FBS ( Sigma ) , 0 . 25% Triton X-100 in PBS at room temperature . Cells were then stained with anti-HA antibody 1:50 ( Roche ) and either anti-Cnx99a ( 1:5 ) , anti-Hrs 8 . 2 ( 1:5 ) , anti-Golgin 84 ( 1:5 ) , anti-Rab7 ( 1:5 ) , anti- GMAP ( 1:50 ) or anti- Golgin 245 ( 1:50 ) ( all antibodies from DSHB ) ( Riedel et al . , 2016 ) . Cells were washed in 20% FBS ( Sigma ) , 0 . 25% Triton X-100 in PBS 5 times and then incubated with anti-rat Alexa Fluor 633 1:50 and either anti-mouse Alexa Fluor 488 or anti-goat Alexa Fluor 488 1:100 ( Thermo Fisher ) for 2 hr at room temperature . Cells were washed again 5 times and then mounted in Vectashield Mounting Medium +DAPI ( Vector Laboratories ) and imaged with Zeiss LSM 700 or 800 confocal microscopes . Quantitation of colocalization was performed as indicated below . The cell line was routinely tested for Mycoplasm infection and found to be negative . Single male flies were frozen for at least 3 hr before grinding them in 100 mM Tris-HCl , 100 mM EDTA , 100 mM NaCL and 0 . 5% SDS . Lysates were incubated at 65°C for 30 min . Then 5M KAc and 6M LiCl were added at a ratio of 1:2 . 5 and lysates were incubated on ice for 10 min . Lysates were centrifuged for 15 min at 20 , 000xg , supernatant was isolated and mixed with Isopropanol . Lysates were centrifuged again for 15 min at 20 . 000xg , supernatant was discarded and the DNA pellet was washed in 70% EtOH and subsequently dissolved in ddH20 . Embryos were collected for 1 hr and aged for an additional 5 hr , all at 29°C . Embryos collected from w- flies were processed in parallel and served as a negative control . Embryos were dissociated as described previously ( Gyoergy et al . , 2018 ) . The cells were sorted using a FACS Aria III ( BD ) flow cytometer . Emission filters were 600LP , 610/20 and 502 LP , 510/50 . Data were analyzed with FlowJo software ( Tree Star ) . The cells from the dissociated negative control w- embryos were sorted to set a baseline plot . RNA was isolated from approximately 50 , 000 mCherry positive or mCherry negative FACS sorted macrophages using RNeasy Plus Micro Kit ( Qiagen , Hilden , Germany ) following manufacturer’s protocol . RNA was also isolated from 50 to 100 mg of ovaries ( about 15–20 pairs of ovaries extracted as indicated above ) . Ovaries were homogenized with a pellet homogenizer ( VWR , Radnor , USA ) and plastic pestles ( VWR , Radnor , USA ) in 1 ml of Trizol ( Thermo Fisher Scientific , Waltham , MA , USA ) and centrifuged at 12 , 000xg for 5 min at 4°C . Further steps were according to the manufacturers protocol . The resulting RNA was used for cDNA synthesis using Sensiscript RT Kit ( macrophages ) or Omniscript ( ovaries ) ( Qiagen , Hilden , Germany ) and oligo dT primers . A Takyon qPCR Kit ( Eurogentec , Liege , Belgium ) was used to mix qPCR reactions based on the provided protocol . qPCR was run on a LightCycler 480 ( Roche , Basel , Switzerland ) and data were analyzed in the LightCycler 480 Software and Prism ( GraphPad Software ) . Data are represented as relative expression to a housekeeping gene ( 2-Δct ) or fold change in expression ( 2-ΔΔct ) . Primer sequences utilized for flies were obtained from the FlyPrimerBank ( http://www . flyrnai . org/FlyPrimerBank ) . Minerva/CG8602: Fw pr TGTGCTTCGTGGGAGGTTTC , Rv pr GCAGGCAAAGATCAACTGACC . C1GalTA: Fw pr TGCCAACAGTCTGCTAGGAAG , Rv pr CTGTGATGTGCATCGTTCACG . Ugalt: Fw pr GCAAGGATGCCCAGAAGTTTG , Rv pr GATATAGACCAGCGAGGGGAC . RpL32: Fw pr AGCATACAGGCCCAAGATCG , Rv pr TGTTGTCGATACCCTTGGGC Embryos were collected for 7 hr at 29°C , bleached and hand-picked for the correct stage . 50–200 embryos were smashed in RIPA buffer ( 150 mM NaCl , 0 , 5% Sodiumdeoxychalat , 0 , 1% SDS , 50 mM Tris , pH 8 ) with Protease inhibitor ( Complete Mini , EDTA free , Roche , Basel , Switzerland ) using a pellet homogenizer ( VWR , Radnor , USA ) and plastic pestles ( VWR , Radnor , USA ) and incubated on ice for 30 min . Afterwards , samples were centrifuged at 4°C , 16 , 000 g for 30 min and the supernatant was collected and used for experiments . The protein concentration was quantified using the Pierce BCA Protein Assay Kit ( ThermoFisher Scientific ) . 30 μg of protein samples were loaded on a 4–15% Mini-PROTEAN TGX Precast Protein Gel ( Bio-Rad , Hercules , USA ) and run at 100V for 80 min in 1x running buffer ( 25 mM Tris Base , 190 mM glycine and 0 . 1%SDS ) followed by transfer onto Amersham Protran Premium 0 . 45 μm NC ( GE Healthcare Lifescience , Little Chalfont , UK ) or Amersham Hybond Low Fluorescence 0 . 2 μm PVDF ( GE Healthcare Lifescience , Little Chalfont , UK ) membrane using a wet transfer protocol with 25 mM Tris Base , 190 mM Glycine +20% MeOH at either 100 Volts for 60 min or 200mA for 90 min at Mini Trans-Blot Cell Module ( Bio-Rad , Hercules , USA ) . Membranes were blocked in PBS-T ( 0 . 1% Triton X-100 in PBS ) containing 2% BSA or Pierce Clear Milk Blocking Buffer ( ThermoFisher Scientific ) for 1 hr at RT . Primary antibodies were incubated overnight at 4°C at the following concentrations: α-T antigen ( Copenhagen ) 1:10 , α-profilin ( Verheyen and Cooley , 1994 ) , DSHB ) 1:50 , anti-GFP ( clone 2B6 , Ogris lab , MFPL ) , anti-GAPDH ( ab181603 , Abcam , Cambridge , UK ) . Afterwards , blots were washed 3x for 5 min in blocking solution and incubated with Goat anti Mouse IgG ( H/L ) :HRP ( Bio-Rad , Hercules , USA ) or goat-anti-rabbit IgG ( H + L ) -HRP ( Bio-Rad , Hercules , USA ) at 1:5 000–10 , 000 for 1–2 hr at room temperature . Blots were washed 2 × 5 min in blocking solution and 1 × 5 min with PBS-T . Blots were developed using SuperSignal West Femto Maximum Sensitivity Substrate ( ThermoFisher Scientific , Waltham , Massachusetts , USA ) according to manufacturer’s instructions . Chemiluminescent signal was detected using the Amersham Imager 600 ( GE Healthcare Lifescience ) or VersaDoc ( Bio-Rad ) . Images were processed with ImageJ . S2R+ cells were transfected as described previously with srpGal4 UAS-Qsox1::FLAG::HA . 2 days post-transfection , medium was removed and cells were washed with PBS . Afterwards , serum-free S2 medium was added and incubated for approximately 40 hr . Afterwards , supernatant was collected and concentrated using Amicon Ultra-4 10K Centrifugal Filter Device ( Merck , Kenilworth , New Jersey , United States ) to gain 80 μl of concentrated supernatant . 20 μl of supernatant was loaded on gel and analyzed by anti-HA ( 1:200 , Roche ) . Images were processed with ImageJ . Embryos were dechorionated in 50% bleach for 5 min , washed with water , and mounted in halocarbon oil 27 ( Sigma-Aldrich , Saint Louis , Missouri , USA ) between a coverslip and an oxygen permeable membrane ( YSI ) . The anterior dorsolateral region of the embryo was imaged on an inverted multiphoton microscope ( TrimScope II , LaVision ) equipped with a W Plan-Apochromat 40X/1 . 4 oil immersion objective ( Olympus ) . mCherry was imaged at 1100 nm excitation wavelengths , using a Ti-Sapphire femtosecond laser system ( Coherent Chameleon Ultra ) combined with optical parametric oscillator technology ( Coherent Chameleon Compact OPO ) . Excitation intensity profiles were adjusted to tissue penetration depth and Z-sectioning for imaging was set at 1 µm for tracking and segmentation respectively . For long-term imaging , movies were acquired for 132–277 min with a frame rate of 40 s . All embryos were imaged with a temperature control unit set to 28 . 5°C . Images acquired from multiphoton microscopy were initially processed with InSpector software ( LaVision Bio Tec ) to compile channels from the imaging data , and the exported files were further processed using Imaris software ( Bitplane ) to visualize the recorded channels in 3D . Macrophage speed and persistence were calculated by using embryos in which the macrophage nuclei were labeled with srpHemo-H2A::3XmCherry ( Gyoergy et al . , 2018 ) . The movie from each imaged embryo was rotated and aligned along the AP axis for tracking analysis . Increasing the gain allowed determination of germband position from the autofluorescence of the yolk . Movies for vnc analysis were analyzed for 2 hr from the time point that cells started to dive into the channels to reach the outer vnc . Macrophage nuclei were extracted using the spot detection function and nuclei positions in xyz-dimensions were determined for each time point and used for further quantitative analysis . Cell speeds and directionalities were calculated in Matlab ( The MathWorks Inc . , Natick , Massachusetts , USA ) from single cell positions in 3D for each time frame measured in Imaris ( Bitplane ) . Instantaneous velocities from single cell trajectories were averaged to obtain a mean instantaneous velocity value over the course of measurement . To calculate directionality values , single cell trajectories were split into segments of equal length ( 10 frames ) and calculated via a sliding window as the ratio of the distance between the macrophage start-to-end location over the entire summed distance covered by the macrophage between successive frames in a segment . Calculated directionality values were averaged over all segments in a single trajectory and all trajectories were averaged to obtain a mean directionality value for the duration of measurement , with 0 being the lowest and one the maximum directionality . To estimate the time for entry into the germband , we increased the gain to visualize the germband position from the autofluorescence of the yolk . We assessed the time point when the first macrophage nucleus reached the edge of the germband ( taken as T0 ) and the time point when the first cell nucleus was just within the germband ( taken as T1 ) . T1-T0 was defined as the time for macrophage entry . Embryos were imaged with Plan-Apochromat 63X/1 . 4 Oil Objective on a Zeiss LSM700 inverted . 10 µm stacks ( 0 . 5 µm intervals ) were taken for properly staged and oriented embryos , starting 10 µm deep in the tissue . These images were converted into Z-stacks in Fiji . ROIs were drawn around macrophages ( signal ) , copied to tissue close by without macrophages ( background ) and the average intensity in the green channel of each ROI was measured . For each pair of ROIs background was subtracted from signal individually . The average signal from control ROIs from one imaging day and staining was calculated and all data points from control , mutant and rescue from the same set was divided by this value . This way we introduced an artificial value called Arbitrary Unit ( AU ) that makes it possible to compare all the data with each other , even if they come from different imaging days when the imaging laser may have a different strength or from different sets of staining . Analysis was done on anonymized samples . Transmitted light images of the embryos were used to measure the position of the germband to determine the stages for analysis . The extent of germband retraction away from the anterior along with the presence of segmentation was used to classify embryos . Embryos with germband retraction of between 29–31% were assigned to late Stage 11 . Those with 29–41% retraction ( early Stage 12 ) were analyzed for the number of macrophages that had entered the germband and those with 50–75% retraction ( late Stage 12 ) for the number along the ventral nerve cord ( vnc ) , and in the whole embryo . Macrophages were visualized using confocal microscopy with a Z-resolution of 3 μm and the number of macrophages within the germband or the segments of vnc was calculated in individual slices ( and then aggregated ) using the Cell Counter plugin in FIJI . To check that this staging allows embryos from the control and mrva3102 mutant to be from the same time during development , embryos were collected for 30 min and then imaged for a further 10 hr using a Nikon-Eclipse Wide field microscope with a Plan-Apochromat 20X/0 . 5 DIC water Immersion Objective . Bright field images were taken every 5 min , and the timing of the start of the movies was aligned based on when cellularization occurred . We found no significant difference in when germband retraction begins ( 269 . 6 ± 9 min in control and 267 . 1 ± 3 min in mrva3102 , p=0 . 75 ) or in when the germband retracts to 41% ( 300 ± 9 min for control , 311 ± 5 min in mrva3102 , p=0 . 23 ) , or in when the germband retraction is complete ( 386 . 5 ± 10 min for control , 401 . 6 ± 8 min for mrva3102 , p=0 . 75 ) . n = 10 embryos for control and 25 embryos for mrva3102 . Standard molecular biology methods were used and all constructs were sequenced by Eurofins before injection into flies . Restriction enzymes BSiWI , and AscI were obtained from New England Biolabs , Ipswich , Massasuchetts , USA ( Frankfurt , Germany ) . PCR amplifications were performed with GoTaq G2 DNA polymerase ( Promega , Madison , USA ) using a peqSTAR 2X PCR machine from PEQLAB , ( Erlangen , Germany ) . All Infusion cloning was conducted using an Infusion HD Cloning kit obtained from Clontech’s European distributor ( see above ) ; relevant oligos were chosen using the Infusion primer Tool at the Clontech website . mrva3102 flies which contain the 3102 P element insert in the 5’ region of CG8602 were crossed to a line expressing transposase ( BL-1429: pn1; ry503Dr1P[Δ 2–3] ) . To allow excision of the P Element , males from the F1 generation containing both the P element and the transposase , were crossed to virgins with the genotype Sp/Cyo; PrDr/TM3Ser ( gift from Lehmann lab ) . In the F2 generation white eyed males were picked and singly crossed to Sp/Cyo; PrDr/TM3Ser virgins . Images were taken with a Z-resolution of 0 . 5 μm from the head of late stage 12 embryos using a Zeiss LSM800 confocal microscope and a 40x/1 . 4 Oil DIC objective . A 4 μm long line was drawn over a macrophage with the middle of the line located approximately at the edge of the cell . mCherry and LanA ( 488 ) intensities were measured using the Multichannel Plot Profile Plugin in Fiji . Three lines were drawn on each cell to catch the variability of secretion . Only cells standing alone or in small groups that had at least some small visible amount of extracellular LanA were analyzed . From each embryo , 20 cells were analyzed . Images were anonymized before quantification . MC-38 colon carcinoma cells , 4T1 breast carcinoma ( ATCC , CRL-2539 ) , Lewis Lung carcinoma LLC1 ( ATCC , CRL-1642 ) and B16-BL6 melanoma ( NCI-DTP; B16BL-6 ) ( all gifts from the Borsig lab ) were kept in DMEM supplemented with 10% FCS ( Sigma-Aldrich , Saint Louis , Missouri , USA ) , Non-essential Amino Acids , and Na-Pyruvate ( Thermo Fisher Scientific , Waltham , Massachusetts , USA ) . All cells were kept in a humidified incubator at 37°C with 5% CO2 . Cells were infected with lentiviral particles containing pInducer20-MFSD1-eGFP . Expression of MFSD1-eGFP was induced with 20 ng/ml ( for MC-38 ) and 100 ng/ml ( for 4T1 , LLC , B16-BL6 ) of Doxycycline for 24 hr prior subsequent analysis . Cell lines were routinely tested for Mycoplasm infection and found to be negative . The identity of the cell lines was confirmed by STR analysis by the cell bank from which they were obtained . Cells were lysed in alysis buffer ( 25 mM Tris , 150 mM NaCl , 1 mM EDTA , 1% Triton X-100 ) supplemented with a protease inhibitor cocktail ( Complete , Roche , Basel , Switzerland ) for 20 min on ice , followed by centrifugation at 14 , 000x g , 4°C for 5 min . The protein lysates were stored at −80°C . Protein concentration was determined with the Pierce BCA Protein Assay Kit ( Thermo Fisher Scientific ) . Cells were fixed with 4% formaldehyde ( Thermo Fisher Scientific ) in PBS for 15 min at room-temperature . Cells were washed three times with PBS followed by blocking and permeabilization with 1% BSA ( Sigma-Aldrich , Saint Louis , Missouri , USA ) /0 . 3% Triton X-100 in PBS for 1 hr . Antibodies were diluted in blocking/permeabilization buffer and incubated for 2 hr at room temperature . Primary antibodies used were: anti-GFP ( clone 5G4 , Ogris lab , MFPL ) , anti-GRASP65 ( Thermo Fisher , PA3-910 ) , anti-Rab5 ( Cell Signaling Technology , #C8B1 ) , anti-Rab7 ( Cell Signaling Technology , #D95F2 ) and anti-LAMP1 ( Abcam , Cambridge , UK , #ab24170 ) . Cells were washed three times with PBS-Tween20 ( 0 . 05% ) for 5 min each , followed by secondary antibody incubation in blocking/permeabilization buffer for 1 hr at room-temperature . Secondary antibodies used were: goat anti-mouse IgG ( H + L ) Alexa Fluor 488 ( Thermo Fisher A11001 ) , goat anti-rabbit IgG ( H + L ) Alexa Fluor 555 ( Thermo Fisher , A21428 ) , Cells were counterstained with DAPI ( Thermo Fisher ) for 10 min in PBS . Cells were mounted with ProLong Gold Antifade Mountant ( Thermo Fisher #P36930 ) . Images were acquired using a Plan-Apochromat 40x/1 . 4 Oil DIC objective M27 on a Zeiss LSM880 confocal microscope . Pictures were processed with ImageJ . Colocalization analysis was performed by ImageJ’s ( NIH ) Coloc two plugin and determined with the pixel intensity spatial correlation analysis ( Pearson’s correlation coefficient ) . 150 mg fly embryos were homogenized in 2 ml 0 . 1% RapiGest , 50 mM ammonium bicarbonate using a dounce homogenizer . The lysed material was left on ice for 40 min with occasional vortexing followed by probe sonication ( 5 s sonication , 5 s pause , 6 cycles at 60% amplitude ) . The lysate was cleared by centrifugation ( 1 , 000 × g for 10 min ) . The cleared lysate was heated at 80°C , 10 min followed by reduction with 5 mM dithiothreitol ( DTT ) at 60°C , 30 min and alkylation with 10 mM iodoacetamide at room temperature ( RT ) for 30 min before overnight ( ON ) digestion at 37°C with 25 µg trypsin ( Roche ) . The tryptic digests were labeled with dimethyl stable isotopes as described ( Boersema et al . , 2009 ) . The digests were acidified with 12 µL trifluoroacetic acid ( TFA ) , 37°C , 20 min and cleared by centrifugation at 10 , 000 g , 10 min . The cleared acidified digests were loaded onto equilibrated SepPak C18 cartridges ( Waters ) followed by 3 × CV 0 . 1% TFA wash . Digests were labeled on the column by adding 5 mL 30 mM NaBH3CN and 0 . 2% formaldehyde ( COH2 ) in 50 mM sodium phosphate buffer pH 7 . 5 ( Light , mrva3102 ) , or 30 mM NaBH3CN and 0 . 2% deuterated formaldehyde ( COD2 ) in 50 mM sodium phosphate buffer pH 7 . 5 ( Medium , control ) . Columns were washed using 3 CV 0 . 1% FA and eluted with 0 . 5 mL 50% MeOH in 0 . 1% FA . The eluates were mixed in a 1:1 ratio , concentrated by evaporation , and resuspended in Jacalin loading buffer ( 175 mM Tris-HCl , pH 7 . 4 ) Glycopeptides were separated from non-glycosylated peptides by Lectin Weak Affinity Chromatography ( LWAC ) using a 2 . 8 m column packed in-house with Jacalin-conjugated agarose beads . The column was washed with 10 CVs Jacalin loading buffer ( 100 µL/min ) before elution with Jacalin elution buffer ( 175 mM Tris- HCl , pH 7 . 4 , 0 . 8M galactose ) 4 CVs , 1 mL fractions . The glycopeptide-containing fractions were purified by in-house packed Stage tips ( Empore disk-C18 , 3M ) . The glycopeptide quantification based on M/L isotope labeled doublet ratios was evaluated to estimate a meaningful cut-off ratio for substantial changes ( Schjoldager et al . , 2015 ) . The labeled glycopeptides produced doublets with varying ratios of the isotopic ions as well as a significant number of single precursor ions without evidence of ion pairs . Labeled samples from control srpHemo-3xmCherry embryos and mrva3102 srpHemo-3xmCherry mutant embryos were mixed 1:1 and subjected to LWAC glycopeptide enrichment . The distribution of labeled peptides from the LWAC flow-through showed that the quantitated peptide M/L ratios were normally distributed with 99 . 7% falling within ±0 . 55 ( Log10 ) . We selected doublets with less/more than ±0 . 55 ( Log10 ) value as candidates for isoform-specific O-glycosylation events . EASY-nLC 1000 UHPLC ( Thermo Scientific ) interfaced via nanoSpray Flex ion source to an -Orbitrap Fusion mass spectrometer ( Thermo Scientific ) was used for the glycoproteomic study . A precursor MS1 scan ( m/z 350–1 , 700 ) of intact peptides was acquired in the Orbitrap at a nominal resolution setting of 120 , 000 . The five most abundant multiply charged precursor ions in the MS1 spectrum at a minimum MS1 signal threshold of 50 , 000 were triggered for sequential Orbitrap HCD-MS2 and ETD-MS2 ( m/z of 100–2 , 000 ) . MS2 spectra were acquired at a resolution of 50 , 000 . Activation times were 30 and 200 ms for HCD and ETD fragmentation , respectively; isolation width was four mass units , and one microscan was collected for each spectrum . Automatic gain control targets were 1 , 000 , 000 ions for Orbitrap MS1 and 100 , 000 for MS2 scans . Supplemental activation ( 20% ) of the charge-reduced species was used in the ETD analysis to improve fragmentation . Dynamic exclusion for 60 s was used to prevent repeated analysis of the same components . Polysiloxane ions at m/z 445 . 12003 were used as a lock mass in all runs . The mass spectrometry glycoproteomics data have been deposited to the ProteomeXchange Consortium ( Vizcaíno et al . , 2014 ) via the PRIDE partner repository with the dataset identifier PXD011045 . Data processing was performed using Proteome Discoverer 1 . 4 software ( Thermo Scientific ) using Sequest HT Node as previously described ( Schjoldager et al . , 2015 ) . Briefly , all spectra were initially searched with full cleavage specificity , filtered according to the confidence level ( medium , low and unassigned ) and further searched with the semi-specific enzymatic cleavage . In all cases the precursor mass tolerance was set to six ppm and fragment ion mass tolerance to 20 mmu . Carbamidomethylation on cysteine residues was used as a fixed modification . Methionine oxidation as well as HexNAc and HexHexNAc attachment to serine , threonine and tyrosine were used as variable modifications for MS2 data . All spectra were searched against a concatenated forward/reverse Drosophila melanogaster-specific database ( UniProt , March 2018 , containing 39034 entries with 3494 canonical reviewed entries ) using a target false discovery rate ( FDR ) of 1% . FDR was calculated using target decoy PSM validator node . The resulting list was filtered to include only peptides with glycosylation as a modification . Glycopeptide M/L ratios were determined using dimethyl 2plex method as previously described ( Schjoldager et al . , 2015 ) Statistical tests as well as the number of embryos/cells assessed are listed in the Figure legends . All statistical analyses were performed using GraphPad Prism and significance was determined using a 95% confidence interval . Data points from individual experiments/embryos were pooled to estimate mean and standard error of the mean . Sample size refers to biological replicates . No statistical method was used to predetermine sample size and the experiments were not randomized . For major questions , data were collected and analyzed masked . Normality was evaluated by D’Agostino and Pearson or Shapiro-Wilk normality test . Unpaired t-test or Mann-Whitney test was used to calculate the significance in differences between two groups and One-Way Anova followed by Tukey post-test or Kruskal-Wallis test followed by Conover or Dunn’s post-test for multiple comparisons . All measurements were performed in 3–38 embryos and at least 37 oocytes . Representative images shown in Figure 1E–G , I , Figure 2F , I , Figure 3A–C , Figure 5A , Figure 6B , D and G and Figure 2—figure supplement 1B-J , Figure 3—figure supplement 1B , H , K , Figure 5—figure supplement 1G , K were from separate experiments repeated 3 to 6 times . The stainings underlying Figure 1—figure supplement 1A-M , Figures 2Hand Figure 6-figure supplement 1C-F are from separate experiments that were repeated at least twice . Stills shown in Figure 3I , L and Figure 5D are representative images from two-photon movies , which were repeated at least 3 times .	Proteins , the workhorses of the body , participate in virtually every single process in a cell . Different types of molecules , such as sugars , can be added onto a protein to change its role or location , but this process may also play a role in cancer . Indeed , tumor cells that contain certain sugar modifications are more likely to be able to spread through the body . For example , a specific combination of sugars called T antigen is rarely present in healthy adult cells; yet , it is commonly found in cancer cells that leave the tumor where they were born and invade another tissue to form a new tumor . However , it is not clear whether T antigen actively helps this process inside the body , or is simply present during it . To answer this question , Valosková , Biebl et al . used genetic and biochemistry tools to study developing fruit fly embryos , where certain immune cells carry T antigen on their proteins . Like invading cancer cells , these immune cells can get inside tissues during development . The experiments revealed that a protein called Minerva helps attach T antigen onto proteins . When embryos were engineered to contain less Minerva , the amount of T antigen in the immune cells dropped , and the cells could not easily make their way into tissues anymore . When the mouse version of Minerva was then added to the embryos , the immune cells of the fruit flies had higher T antigen levels on their proteins and could invade tissues again . Some of the proteins targeted by Minerva were known to be involved in cancer , but not all of them . Future experiments will investigate which role the human version of Minerva plays in cancer cells that get inside new tissues , and if it could help us predict whether a cancer is likely to spread .	[ "Abstract", "Introduction", "Results", "Materials", "and", "methods" ]	[ "developmental", "biology", "cancer", "biology" ]	2019	A conserved major facilitator superfamily member orchestrates a subset of O-glycosylation to aid macrophage tissue invasion
Forskolin is a unique structurally complex labdane-type diterpenoid used in the treatment of glaucoma and heart failure based on its activity as a cyclic AMP booster . Commercial production of forskolin relies exclusively on extraction from its only known natural source , the plant Coleus forskohlii , in which forskolin accumulates in the root cork . Here , we report the discovery of five cytochrome P450s and two acetyltransferases which catalyze a cascade of reactions converting the forskolin precursor 13R-manoyl oxide into forskolin and a diverse array of additional labdane-type diterpenoids . A minimal set of three P450s in combination with a single acetyl transferase was identified that catalyzes the conversion of 13R-manoyl oxide into forskolin as demonstrated by transient expression in Nicotiana benthamiana . The entire pathway for forskolin production from glucose encompassing expression of nine genes was stably integrated into Saccharomyces cerevisiae and afforded forskolin titers of 40 mg/L . Plants synthesize an impressive diversity of specialized metabolites enabling them to communicate and adapt to environmental challenges ( Mithöfer and Boland , 2012; Woldemariam et al . , 2011 ) . Throughout history , humans have benefited from the medicinal properties of many of these phytochemicals ( Hardy et al . , 2012 ) . Specialized plant metabolites and direct derivatives thereof still constitute more than a third of approved pharmaceuticals ( Cragg and Newman , 2013; David et al . , 2015 ) . With over 50 , 000 known structures according to the ‘Dictionary of natural products’ ( http://dnp . chemnetbase . com/ ) , terpenoids are the largest class of plant specialized metabolites and constitute a vast repository of bio-active natural products including many structurally complex compounds ( Pateraki et al . , 2015 ) . Examples of widely used plant-derived terpenoid pharmaceuticals are the anticancer drug paclitaxel ( taxol ) ( Liu and Khosla , 2010 ) , the therapeutic ingenol mebutate ( picato ) that is used for treatment of actinic keratosis ( King et al . , 2016; Luo et al . , 2016 ) and artemisinin which is the most efficient treatment against malaria caused by Plasmodium parasites ( Graham et al . , 2010; Paddon and Keasling , 2014 ) . Traditional chemical synthesis of plant-derived diterpenoid pharmaceuticals remains economically challenging , despite recent examples of elegant strategies mimicking natural routes ( Appendino , 2014; Kawamura et al . , 2016; Yuan et al . , 2016 ) . Extraction from plant biomass and semisynthesis from biotechnologically produced intermediates have been approached as alternative strategies ( Graham et al . , 2010; Paddon et al . , 2013; Roberts , 2007 ) . In contrast to recent examples demonstrating complete pathway reconstruction and production of opiate alkaloids in yeast ( Galanie et al . , 2015; Nakagawa et al . , 2016 ) , engineered total biosynthesis of terpenoid therapeutics—including paclitaxel and ingenol esters—has not yet been achieved . Challenges on the way to achieving this goal include the identification of pathway enzymes in native systems , particularly for those belonging to multi-enzyme families catalyzing the biosynthesis of specialized metabolites in plants , engineering of poorly understood multi-step enzymatic pathways and difficulties encountered in heterologous expression of key enzymes catalyzing monooxygenations critical for diterpenoid biosynthesis ( Pateraki et al . , 2015; Renault et al . , 2014 ) . The diterpenoid forskolin is the active hypotensive principle accumulating in the root cork of Coleus forskohlii ( Pateraki et al . , 2014 ) , a perennial shrub of the Lamiaceae family , indigenous to India and Southeast Asia with numerous reported applications in traditional medicine ( Alasbahi and Melzig , 2010b; Kavitha et al . , 2010 ) . The pharmaceutical properties of forskolin are based on its ability to directly activate the adenylate cyclase enzyme resulting in elevated levels of the second messenger cyclic adenosine monophosphate ( cAMP ) ( Doseyici et al . , 2014; Seamon et al . , 1981 ) . Approved applications of forskolin range from alleviation of glaucoma ( OcuforsEye drop solutions , Sabinsa , India ) , treatment of hypertension and heart failure ( Colforsin daropate hydrochloride , a water-soluble derivative of forskolin , Nippon Kayaku , Japan ) to lipolysis and body weight control ( Godard et al . , 2005; Kikura et al . , 2004; Toya et al . , 1998; Wagh et al . , 2012; Yoneyama et al . , 2002 ) . Therapeutic opportunities were also suggested in animal tests , where forskolin-induced pigmentation of the skin , increasing protection against UV-associated carcinogenesis ( D'Orazio et al . , 2006 ) . The complex chemical structure of forskolin with a decalin core , characteristic of labdane-type diterpenoids , a tetrahydropyran ring , five oxidized positions and eight chiral centers ( Figure 1A ) represents a challenge for classical organic chemical synthesis , although a key intermediate for stereoselective total synthesis has been reported ( Ye et al . , 2009 ) . Hence , commercially available forskolin is extracted from C . forskohlii roots and purified from a mixture of over 60 structurally related abietane and epoxylabdane diterpenoids with a forskolin content varying from 0 . 013% to 0 . 728% of root dry weight ( Alasbahi and Melzig , 2010a; Asada et al . , 2012; Srivastava et al . , 2017 ) . As the demand for forskolin grows , reliable and sustainable commercial production from C . forskholii will become unachievable due to low yields , susceptibility of this species to diseases , changing climatic conditions and the resource intensive extraction and purification procedure required to obtain pharmaceutical grade forskolin ( Mora-Pale et al . , 2014 ) . Elucidation of the biosynthetic pathway to forskolin and subsequent engineering of the pathway into microbial hosts offers a more clear and stable alternative production system that will be better able to address future needs . 10 . 7554/eLife . 23001 . 003Figure 1 . Biosynthesis of forskolin in the root cork cells of C . forskohlii . ( A ) Scheme showing the structures of the diterpene precursor 13R-manoyl oxide , deacetylforskolin and forskolin on a background of root cork cells with forskolin containing oil bodies . ( B ) Transcript profiles of biosynthetic candidate genes in selected tissues of C . forskohlii as shown on the illustrations . DOI: http://dx . doi . org/10 . 7554/eLife . 23001 . 00310 . 7554/eLife . 23001 . 004Figure 1—source data 1 . cDNAs identified in the C . forskohlii root cork transcriptome and cloned during this work , with the GeneBank accession numbers . DOI: http://dx . doi . org/10 . 7554/eLife . 23001 . 00410 . 7554/eLife . 23001 . 005Figure 1—source data 2 . Table of FPKM ( Fragments Per Kilobase of transcript per Million mapped reads ) values of the first 20 most abundant cDNAs identified in the root cork transcriptome library . cDNAs involved in terpenoid metabolism are marked in bold . DOI: http://dx . doi . org/10 . 7554/eLife . 23001 . 00510 . 7554/eLife . 23001 . 006Figure 1—source data 3 . Table of primers used in this study . Construction of plasmids for expression of CfTPS2 , CfTPS3 , CfTPS1 is described in Andersen-Ranberg et al . ( 2016 ) . U ( uracil , marked in bold ) , represents the cleavage site , used in the USER cloning ( Nour-Eldin et al . , 2006 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 23001 . 006 Recently , we reported specific accumulation of forskolin and its diterpene scaffold 13R-manoyl oxide in the root cork cells of C . forskohlii . A pair of diterpene synthases ( CfTPS2 and CfTPS3 ) , exclusively present in the root cork , was found to catalyze cyclization of the C20 diterpenoid precursor geranylgeranyl diphosphate ( GGPP ) into 13R-manoyl oxide , the diterpene scaffold of forskolin ( Pateraki et al . , 2014 ) . As a proof of concept , biosynthesis of 13R-manoyl oxide in enantiomerically pure form but in low yields was achieved by expressing CfTPS2 and CfTPS3 in Saccharomyces cerevisiae , E . coli and Synechocystis sp . ( Andersen-Ranberg et al . , 2016; Englund et al . , 2015; Nielsen et al . , 2014 ) . Taking into account the functionalization steps needed for conversion of 13R-manoyl oxide to forskolin ( Figure 1A ) , involvement of enzymes from the families of cytochrome P450s ( CYPs ) and acetyltransferases was predicted . Here , we report an integrated biochemical and functional genomics approach , including metabolomics , single-cell-type transcriptome studies and a synthetic biology modular approach involving transient combinatorial expression of candidate genes in Nicotiana benthamiana to identify the panel of enzymes catalyzing functionalization of the C . forskohlii diterpene backbones and more specifically the biosynthesis of forskolin . Pathway intermediates were identified using GC- or HPLC-HRMS-SPE-NMR . To demonstrate the downstream application of the present work regarding biotechnological production of forskolin , the entire forskolin biosynthetic pathway was reconstituted in engineered S . cerevisiae for fermentation-based production of forskolin from glucose . Forskolin is the first example of a pharmaceutical diterpenoid produced entirely in yeast at titers relevant for industrial consideration . The outlined combinatorial biochemistry approach paves the way for development of yeast-engineered platforms for biosynthesis of other known or new-to-nature diterpenoids The conversion of 13R-manoyl oxide to forskolin requires six regio- and stereospecific monooxygenations and a single regiospecific acetylation ( Figure 1A ) . Considering the strict localization of forskolin in the root cork cells of C . forskohlii and the almost exclusive expression of the pair of diterpene synthases forming 13R-manoyl oxide within the same tissue ( Pateraki et al . , 2014 ) , the root cork was selected for deep RNA-Seq transcriptome analysis . The generated transcriptome contained 263 , 652 assembled putative cDNAs . The transcriptome was queried for transcripts encoding CYPs belonging to the CYP71 clan , based on their established role in monooxygenation reactions in the biosynthesis of specialized metabolites ( Nelson , 2013; Werck-Reichhart and Feyereisen , 2000 ) . Their relative levels in the root cork transcriptome were also taken into consideration . Within the CYP71 clan , focus was also placed on P450 subfamilies that showed extensive , recent expansions in the cork transcriptome ( Nelson and Werck-Reichhart , 2011; Werck-Reichhart and Feyereisen , 2000 ) . Members of the CYP76AH subfamily , part of the CYP71 clan , have recently been shown to catalyze monooxygenation of abietane-type diterpenoids like miltiradiene and dehydroabietadiene in Lamiaceae species , closely related to C . forskohlii ( Božić et al . , 2015; Ignea et al . , 2016a; Zi and Peters , 2013 ) . Members of this P450 subfamily were therefore of high interest as enzymes putatively involved in diterpenoid biosynthesis in C . forskohlii . Based on these considerations , a total of 29 CYP candidates ( Figure 1—source data 1 ) were selected and cloned in full length using as template cDNA synthesized from root cork total RNA . Among these CYPs , seven members were assigned to the CYP76AH subfamily by the ‘P450 Nomenclature committee’ ( Nelson , 2009 ) , rendering this CYP subfamily the highest represented in the transcriptome ( Figure 1—source data 1 ) . Five were full length sequences ( CfCYP76AH8 , CfCYP76AH9 , CfCYP76AH10 , CfCYP76AH11 , and CfCYP76AH11 ) , whereas two ( CfCYP76AH15 and CfCYP76AH16 ) were represented by partial cDNAs . For the latter , 5’RACE experiments afforded the full-length cDNAs . Similarly to the previously identified diTPSs , CfTPS2 and CfTPS3 ( Pateraki et al . , 2014 ) , gene expression studies showed that the identified members of the CYP76AH family were highly or exclusively expressed in the root cork cells ( Figure 1B ) . We have recently reported an Agrobacterium-mediated modular transient expression system in N . benthamiana enabling biosynthesis of labdane-type diterpenes in quantities permitting purification and structural elucidation ( Andersen-Ranberg et al . , 2016 ) . Utilizing this system , all candidate P450 genes were heterologously expressed in combination with genes necessary for the production of high amounts of 13R-manoyl oxide ( Andersen-Ranberg et al . , 2016; Pateraki et al . , 2014 ) . Of the CYPs tested , six efficiently converted 13R-manoyl oxide into oxygenated derivatives ( Figures 2A and 3 and Figure 3—figure supplements 1 and 2 ) . CfCYP76AH15 , CfCYP76AH8 and CfCYP76AH17 catalyzed formation of 11-oxo-13R-manoyl oxide ( 2 ) as the main product ( Figures 2 and 3 ) . Forskolin harbors a keto-group at the C-11 position , like the majority of 13R-manoyl oxide-derived diterpenoids found in C . forskohlii ( Asada et al . , 2012; Zhang et al . , 2009 ) . Of the three CfCYP76AHs tested in this experiment , CfCYP76AH15 showed the highest efficiency and specificity for the conversion of 13R-manoyl oxide to 2 with no concomitant formation of multi-oxygenated products ( Figure 2 ) . Compound 5d produced by CfCYP76AH8 as well as by CfCYP76AH17 was identified as 1 , 11-dihydroxy-13R-manoyl oxide ( Figures 2B and 4 and Tables 1 and 2 ) . Noticable , this specific oxygenation pattern is found in forskolin . Minor amounts of several di- and trihydroxylated 13R-manoyl oxide-derived compounds were also produced by CfCYP76AH8 and CfCYP76AH17 ( Figure 2A and Figure 3—figure supplement 1 ) . Although we managed to identify the chemical structures of a number of 13R-manoyl oxide derivatives ( Figure 2B ) , it was not possible to do so for all the compounds shown in Figure 3—figure supplement 1 . The main obstacle was the high complexity as well as the small amounts of the diterpenoids produced in N . benthamiana leaves expressing the CfCYP76AHs . Additional limiting factors were the instability of several of these compounds and the limiting plant material available . Production of higher amounts of these compounds in microbial hosts was not pursued , because the terpenoid profiles observed following expression of the enzymes in plants and yeast cells were not identical . 10 . 7554/eLife . 23001 . 007Figure 2 . 13R-manoyl oxide oxide-derived hydroxylated products formed following transient expression of CfCYP76AHs in N . benthamiana leaves . ( A ) Molecular formulas of the hydroxylated products obtained using different CfCYP76AHs . The number of hydroxylations of each compound was deduced from its accurate molecular mass ( <5 ppm , Supplementary file 1 ) as determined by LC-qTOF-MS or NMR . Each different compound is marked by a number . ( B ) Chemical structures of the compounds marked with numbers in bold in A as determined by NMR ( Tables 1 and 2 ) . MO: manoyl oxideDOI: http://dx . doi . org/10 . 7554/eLife . 23001 . 00710 . 7554/eLife . 23001 . 008Figure 3 . GC-MS analysis of 13R-manoyl oxide ( 1 ) derived diterpenoids obtained by transient expression of CYP76AHs from C . forskohlii in N . benthamiana leaves . ( A ) GC-MS total ion chromatograms ( TIC ) of extracts from N . benthamiana transiently expressing CfCXS , CfGGPPS , CfTPS2 and CfTPS3 ( 13R-manoyl oxide biosynthesis ) genes in combination with water ( - ) , CfCYP76AH15 , CfCYP76AH17 , CfCYP76AH8 , CfCYP76AH11 or CfCYP76AH16 . 1-eicosene was used as internal standard ( IS ) . 13R-manoyl oxide ( 1 ) was identified only in ( - ) , indicating that it is further metabolized by the CfCYP76AH15 , CfCYP76AH17 , CfCYP76AH8 , CfCYP76AH11 and CfCYP76AH16 enzymes . ( B ) m/z spectrum of 11-oxo-13R-manoyl oxide ( 2 ) . ( C ) m/z spectrum of 9-hydroxy-13R-manoyl oxide ( 3a ) . The structure of both compounds was verified by NMR analysis ( Tables 1 and 2 ) . The compounds have been identified previously in C . forskohlii as putative intermediates in the in planta biosynthesis of forskolin ( Asada et al . , 2012 ) . For each combination , extracts from leaves of three different N . benthamiana plants have been analyzed and representative chromatograms are shown . DOI: http://dx . doi . org/10 . 7554/eLife . 23001 . 00810 . 7554/eLife . 23001 . 009Figure 3—figure supplement 1 . LC-qTOF-MS analysis of 13R-manoyl oxide-derived diterpenoids obtained by transient expression of C . forskohlii CYP76AH encoding genes in N . benthamiana leaves . Total ion chromatograms ( TIC ) of extracts expressing the 13R-manoyl oxide biosynthesis genes ( CfCXS , CfGGPPS , CfTPS2 , CfTPS3 ) in combination with water ( - ) , CfCYP76AH8 , CfCYP76AH17 , CfCYP76AH15 , CfCYP76AH11 or CfCYP76AH16 are shown . 13R-manoyl oxide-derived oxygenated compounds formed ( marked with grey bars ) and their identity including their molecular formulas was confirmed by their accurate mass ( 5 ppm tolerance , Supplementary file 1 ) . The identity of 1 , 11-dihydroxy-13R-manoyl oxide ( 5d ) and 9-deoxydeactylforskolin ( 10b ) was confirmed by NMR analysis ( Figure 4 and Tables 1 and 2 ) . No 13R-manoyl oxide-derived diterpenoids were detected in the water control ( - ) . For each combination , extracts from leaves of three different N . benthamiana plants have been analyzed and representative chromatograms are shown . DOI: http://dx . doi . org/10 . 7554/eLife . 23001 . 00910 . 7554/eLife . 23001 . 010Figure 3—figure supplement 2 . GC-MS analysis of 13R-manoyl oxide-derived diterpenoids following transient expression in N . benthamiana leaves of the C . forskohlii gene encoding CfCYP71D281 together with genes encoding the required enzymes for biosynthesis of 13R-manoyl oxide ( CfCXS , CfGGPPS , CfTPS2 , CfTPS3 ) . ( A ) GC-MS total ion chromatograms ( TIC ) of extracts from N . benthamiana transiently expressing 13R-manoyl oxide biosynthesis genes in combination with water ( - ) or CfCYP71D381 , respectively . 1-Eicosene was used as internal standard ( IS ) and 13R-manoyl oxide ( 1 ) was identified in both ( - ) and the CfCYP71D381 samples . Compounds 3b and 3c were identified in extracts from N . benthamiana leaves expressing CfCYP71D381 together with the genes in 13R-manoyl oxide biosynthesis . CfCYP71D381 efficiently converted compound 1 to a mixture of two mono-hydroxylated 13R-manoyl oxide derivatives ( 3b and 3c ) . Structural elucidation by NMR ( Figure 4 and Tables 1 and 2 ) showed hydroxylation of 1 at positions C-2 ( 3b ) and C-19 ( 3c ) . These hydroxylation positions do not coincide with those found in forskolin and to our knowledge have not been observed in other diterpenoids known from C . forskohlii . ( B ) m/z spectrum of 2-hydroxy-13R-manoyl oxide ( 3b ) . ( C ) m/z spectrum of 19-hydroxy-13R-manoyl oxide ( 3c ) . For each combination , extracts from leaves of three different N . benthamiana plants have been analyzed and representative chromatograms are shown . DOI: http://dx . doi . org/10 . 7554/eLife . 23001 . 01010 . 7554/eLife . 23001 . 011Figure 4 . Structures of key compounds presented in this work . ( A ) Compounds confirmed using authentic standards . ( B ) Compounds which structure was confirmed/identified by comparison of 13C NMR data with existing literature . ( C ) Compounds which structure was confirmed/identified by HMBC and NOE correlations for assigning position of OH-groups ( marked in red ) , whereas couplings identified in the previously uncharacterized compounds 3b , 3c and 5d are marked in black . All other molecular structures were confirmed by 13C chemical shifts in comparisons to reference values ( Table 1 , Figure 4—source data 1 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 23001 . 01110 . 7554/eLife . 23001 . 012Figure 4—source data 1 . NMR spectra’s of selected 13R-manoyl oxide derived molecules . DOI: http://dx . doi . org/10 . 7554/eLife . 23001 . 01210 . 7554/eLife . 23001 . 013Table 1 . 1H-NMR and 13C-NMR chemical shifts ( Figure 4—source data 1 ) of novel oxygenated 13R- ( + ) -manoyl oxide-derived diterpenoids formed following transient expression of CYP encoding genes from C . forskohlii . DOI: http://dx . doi . org/10 . 7554/eLife . 23001 . 01319-hydroxy- 13R-manoyl oxide ( 3c ) 2-hydroxy- 13R-manoyl oxide ( 3b ) 1 , 11-dihydroxy- 13R-manoyl oxide ( 5d ) Pos . 1H ( nH; m; J ( Hz ) ) 13C1H ( nH; m; J ( Hz ) ) 13C1H ( nH; m; J ( Hz ) ) 13C10 . 89 ( 1H;m ) 1 . 63 ( 1H; m ) 39 . 11 . 10 ( 1H; t ( br ) ; 11 . 9 , 11 . 9 ) 1 . 77 ( 1H; m ) 51 . 33 . 49 ( 1H; dd;11 . 1 , 4 . 5 ) 79 . 021 . 44 ( 1H; m ) 1 . 56 ( 1H; m ) 18 . 13 . 92 ( 1H; m ) 65 . 31 . 75 ( 1H; td; 13 . 5 , 11 . 1 , 3 . 9 ) 1 . 60 ( 1H; m ) 29 . 030 . 95 ( 1H; m ) 1 . 78 ( 1H; m ) 35 . 80 . 76 ( 1H; t ( br ) ; 11 . 9 , 11 . 9 ) 1 . 99 ( 1H; d ( br ) ; 11 . 9 ) 48 . 21 . 47 ( 1H; dd; 13 . 6 , 3 . 9 ) 1 . 39 ( 1H; td; 13 . 5 , 3 . 6 ) 39 . 6438 . 534 . 933 . 451 . 10 ( 1H; dd; 2 . 3 , 12 . 6 ) 56 . 90 . 95 ( 1H; dd; 2 . 2 , 12 . 4 ) 55 . 90 . 84 ( 1H; dd; 11 . 3 , 2 . 0 ) 55 . 661 . 36 ( 1H; dd; 3 . 6 , 12 . 6 ) 20 . 11 . 68 ( 1H; m ) 19 . 71 . 47 ( 1H; m ) 20 . 21 . 75 ( 1H; m ) 1 . 27 ( 1H; m ) 1 . 64 ( 1H; m ) 71 . 42 ( 1H; m ) 1 . 83 ( 1H; dt; 3 . 3 , 12 . 2 ) 43 . 61 . 45 ( 1H; dd ( br ) ; 3 . 6 , 12 . 5 ) 1 . 85 ( 1H; dt ( br ) ; 2 . 9 , 12 . 5 ) 43 . 21 . 48 ( 1H; m ) 1 . 85 ( 1H; m ) 44 . 0875 . 175 . 175 . 391 . 35 ( 1H; dd; 4 . 3 , 12 . 0 ) 55 . 71 . 40 ( 1H; dd; 4 . 2 , 11 . 9 ) 55 . 41 . 54 ( 1H; d; 5 . 8 ) 55 . 81037 . 338 . 743 . 8111 . 48 ( 1H; m ) 1 . 58 ( 1H; m ) 15 . 41 . 53 ( 1H; m ) 1 . 61 ( 1H; m ) 15 . 64 . 38 ( 1H; br q; ≈8 . 6 ) 65 . 6121 . 78 ( 1H; m ) 1 . 64 ( 1H; m ) 35 . 71 . 78 ( 1H; m ) 1 . 66 ( 1H; m ) 35 . 52 . 02 ( 1H; dd; 14 . 3 , 8 . 7 ) 2 . 27 ( 1H; dd; 14 . 3 , 8 . 7 ) 35 . 81373 . 473 . 472 . 8145 . 87 ( 1H; dd; 10 . 8 , 17 . 4 ) 147 . 75 . 87 ( 1H; dd; 10 . 8 , 17 . 4 ) 147 . 75 . 90 ( 1H; dd; 17 . 4 , 10 . 8 ) 147 . 1154 . 92 ( 1H; dd; 1 . 5 , 10 . 8 ) 5 . 14 ( 1H; dd; 1 . 5 , 17 . 4 ) 110 . 24 . 92 ( 1H; d; 10 . 8 ) 5 . 14 ( 1H; d; 17 . 4 ) 110 . 34 . 94 ( 1H; dd; 10 . 7 , 1 . 5 ) 5 . 17 ( 1H; dd; 17 . 4 , 1 . 5 ) 111 . 2161 . 27 ( 3H; s ) 28 . 51 . 27 ( 3H; s ) 28 . 71 . 27 ( 3H; s ) 32 . 1171 . 28 ( 3H; s ) 25 . 31 . 29 ( 3H; s ) 25 . 71 . 49 ( 3H; s ) 27 . 8180 . 97 ( 3H; s ) 26 . 80 . 93 ( 3H; s ) 33 . 50 . 78 ( 3H; s ) 13 . 5193 . 70 ( 1H; d; 10 . 9 ) 3 . 46 ( 1H; d; 10 . 9 ) 65 . 40 . 85 ( 3H; s ) 22 . 20 . 85 ( 3H; s ) 32 . 8200 . 78 ( 3H; s ) 15 . 70 . 84 ( 3H; s ) 16 . 50 . 79 ( 3H; s ) 21 . 1 1H and 13C NMR data acquired at 600 and 150 MHz , respectively , in methanol-d4 , at 300 K . s = singlet , d = doublet , t = triplet , m = multiplet , br = broad10 . 7554/eLife . 23001 . 014Table 2 . Structural identification of four oxygenated 13R-manoyl oxide-derived diterpenoids formed following transient expression of CYP encoding genes from C . forskohlii based on comparison of their 1H-NMR and 13C-NMR ( Figure 4—source data 1 ) chemical shifts to literature data . Chemical shifts for reference compounds marked with * have not been assigned to a specific carbon . The 13C chemical shifts of 9-deoxyforskolin ( Gabetta et al . , 1989 ) were used as reference for 6 , 7-dihydroxy-11-oxo-13R-manoyl oxide ( 7h ) . DOI: http://dx . doi . org/10 . 7554/eLife . 23001 . 0149-Deoxydeacetylforskolin ( 10b ) †1 , 9-Dideoxydeacetylforskolin ( 7h ) †11-oxo-13R-manoyl oxide ( 2 ) †Coleorol ( 3a ) †Pos . 1H ( nH; m; J ( Hz ) ) 13C ( Gabetta et al . , 1989 ) 1H ( nH; m; J ( Hz ) ) 13C ( Gabetta et al . , 1989 ) 13C ( Gabetta et al . , 1989 ) 13C ( Asada et al . , 2012 ) 14 . 38 ( 1H; t; 2 . 8 ) 71 . 671 . 22 . 45 ( 1H , d ( br ) ; 13 . 1 ) 0 . 78 ( H; m ) 41 . 543 . 142 . 141 . 931 . 731 . 621 . 47 ( 1H; m ) 2 . 14 ( 1H; m ) 25 . 825 . 61 . 78 ( H; m ) 1 . 40 ( H; m ) 18 . 718 . 418 . 518 . 418 . 618 . 431 . 12 ( 1H; dt; 3 . 4 , 13 . 2 ) 1 . 62 ( 1H; dt; 3 . 5 , 13 . 5 ) 36 . 436 . 31 . 36 ( H; m ) 1 . 15 ( H; m ) 43 . 843 . 743 . 443 . 341 . 941 . 8434 . 234 . 134 . 434 . 133 . 433 . 233 . 333 . 251 . 34 ( 1H; d; 2 . 1 ) 47 . 547 . 4n . d . 55 . 755 . 256 . 055 . 845 . 745 . 564 . 44 ( 1H; t; 2 . 6 ) 70 . 870 . 24 . 39 ( 1H; m ) 70 . 470 . 219 . 819 . 719 . 519 . 473 . 68 ( 1H; d; 3 . 6 ) 80 . 781 . 13 . 71 ( 1H; d; 3 . 8 ) 81 . 080 . 739 . 639 . 436 . 636 . 4880 . 078 . 580 . 179 . 977 . 577 . 278 . 077 . 893 . 32 ( 1H; s ) 58 . 058 . 22 . 59 ( 1H; s ) 65 . 565 . 466 . 966 . 775 . 375 . 21042 . 241 . 738 . 037 . 837 . 337 . 141 . 140 . 911207 . 7207 . 6206 . 3205 . 7207 . 7207 . 121 . 121 . 0122 . 63 ( 1H; d; 18 . 0 ) 2 . 69 ( 1H; d; 18 . 0 ) 49 . 849 . 92 . 60 ( 1H; d; 18 . 1 ) 2 . 66 ( 1H; d; 18 . 1 ) 50 . 049 . 850 . 450 . 231 . 631 . 51375 . 174 . 875 . 175 . 175 . 174 . 472 . 972 . 8145 . 94 ( 1H; dd; 10 . 8 , 17 . 4 ) 146 . 2145 . 85 . 95 ( 1H; dd; 10 . 7 , 17 . 4 ) 146 . 9146 . 4146 . 9146 . 0147 . 4147 . 3155 . 04 ( 1H; d; 10 . 8 ) 5 . 14 ( 1H; d; 17 . 4 ) 112 . 4112 . 75 . 04 ( 1H; d; 10 . 7 ) 5 . 17 ( 1H; d; 17 . 4 ) 112 . 3112 . 1112 . 3111 . 9110 . 1110 . 0161 . 30 ( 3H; s ) 31 . 531 . 51 . 28 ( 3H; s ) 31 . 633 . 231 . 431 . 228 . 928 . 8171 . 54 ( 3H; s ) 24 . 124 . 51 . 50 ( 3H; s ) 23 . 531 . 428 . 127 . 927 . 029 . 9181 . 38 ( 3H; s ) 33 . 118 . 20 . 97 ( 3H; s ) 33 . 423 . 915 . 615 . 533 . 733 . 6191 . 21 ( 3H; s ) 23 . 723 . 61 . 21 ( 3H; s ) 24 . 023 . 721 . 821 . 621 . 521 . 4201 . 01 ( 3H; s ) 18 . 532 . 81 . 30 ( 3H; s ) 17 . 216 . 733 . 633 . 5*17 . 016 . 8†1H and 13C NMR data acquired at 600 and 150 MHz , respectively , in methanol-d4 , at 300 K . s = singlet , d = doublet , t = triplet , m = multiplet , br = broad Two additional CYPs of the CYP76AH subfamily catalyzed oxygenation of 13R-manoyl oxide at different positions without substantial formation of C-11 keto derivatives . CfCYP76AH16 yielded predominantly 3a which was identified by NMR as 9-hydroxy-13R-manoyl oxide ( or coleorol ) and CfCYP76AH11 produced a range of monooxygenated derivatives including traces of 10b which was identified by NMR as 9-deoxydeacetylforskolin ( Figure 2 ) . The positions of the carbonyl and hydroxyl-groups in 3a and 10b were consistent with those in forskolin . Thus , the individual activities of the CYPs catalyzing formation of these compounds can be considered complementary in forskolin biosynthesis: 10b carries the carbonyl-function at C-11 and the three hydroxyl groups observed in forskolin at positions C-1 , C-6 and C-7 , but lacks hydroxylation at the C-9 position , which is observed in 3a . The only enzyme outside the CYP76AH subfamily that displayed activity toward 13R-manoyl oxide was CYP71D381 , which resulted in oxidized derivatives at positions not compatible with forskolin ( Figure 3—figure supplement 2 ) . To probe the role of the different CfCYP76AH enzymes in forskolin biosynthesis , they were co-expressed by combining one of the three P450s catalyzing formation of 2 with the functionally distinct CfCYP76AH11 and CfCYP76AH16 , first in pairs , then in all possible permutations as triplets ( Figure 5 and Figure 5—figure supplement 1 ) . The combination of CfCYP76AH15 and CfCYP76AH11 afforded production of 6 , 7-dihydroxy-11-keto-manoyl oxide ( 7h ) as part of a complex mixture . Formation of 7h demonstrated combined introduction of the carbonyl group at C-11 together with two hydroxyl groups at positions C-6 and C-7 , again consistent with the oxygenation pattern of forskolin ( Figure 2 and Figure 5—figure supplement 1 ) . 10 . 7554/eLife . 23001 . 015Figure 5 . LC-qTOF-MS analysis of 13R-manoyl oxide-derived diterpenoids obtained by transient expression of combinations of C . forskohlii CYP encoding genes , together with genes encoding the required enzymes for biosynthesis of 13R-manoyl oxide in N . benthamiana leaves . Total ion chromatograms ( TIC ) of extracts expressing the 13R-manoyl oxide biosynthesis genes ( CfCXS , CfGGPPS , CfTPS2 , CfTPS3 ) , in combination with ( from the top ) water ( - ) , CfCYP76AH8 + CfCYP76AH11 + CfCYP76AH16 , CfCYP76AH17 + CfCYP76AH11 + CfCYP76AH16 , and CfCYP76AH15 + CfCYP76AH11 + CfCYP76AH16 are shown . Hydroxylated 13R-manoyl oxide-derived diterpenoids ( marked with grey bars ) and their identity including their molecular formulas were confirmed by accurate mass ( 5 ppm tolerance , Supplementary file 1 ) . Compounds present in trace amounts are not marked . The identity of 1 , 11-dihydroxy-13R-manoyl oxide ( 5d ) , 9-deoxydeacetylforskolin ( 10b ) and 1 , 9-dideoxydeacetylforskolin ( 7h ) was confirmed by NMR analysis ( Figure 4 and Tables 1 and 2 ) , whereas the identity of deacetylforskolin ( 13b ) was confirmed by comparison to an authentic chemically synthesized standard . No 13R-manoyl oxide-derived diterpenoids were identified in the water control ( - ) . For each combinaton , extracts from leaves of three different N . benthamiana plants have been analyzed and representative chromatograms are shown . DOI: http://dx . doi . org/10 . 7554/eLife . 23001 . 01510 . 7554/eLife . 23001 . 016Figure 5—figure supplement 1 . LC-qTOF-MS analysis of 13R-manoyl oxide-derived diterpenoids obtained by transient expression of combinations of C . forskohlii CYP76AH encoding genes in N . benthamiana leaves . Total ion chromatograms ( TIC ) of extracts expressing the 13R-manoyl oxide biosynthesis genes ( CfCXS , CfGGPPS , CfTPS2 , CfTPS3 ) in combination with ( from the top ) water ( - ) , CfCYP76AH15 + CfCYP76AH11 , CfCYP76AH8 + CfCYP76AH11 , CfCYP76AH17 + CfCYP76AH11 , CfCYP76AH15 + CfCYP76AH16 , CfCYP76AH8 + CfCYP76AH16 and CfCYP76AH17 + CfCYP76AH16 are shown . Oxygenated 13R-manoyl oxide-derived diterpenoids ( marked with grey bars ) and their identity including their molecular formulas were confirmed by their accurate mass ( 5 ppm tolerance , Supplementary file 1 ) . Compounds present in trace amounts are not marked . The identity of 1 , 11-dihydroxy-13R-manoyl oxide ( 5d ) , 9-deoxydeacetylforskolin ( 10b ) , 1 , 9-dideoxydeacetylforskolin ( 7h ) was confirmed by NMR analysis ( Figure 4 and Tables 1 and 2 ) . No 13R-manoyl oxide-derived diterpenoids were detected in the water control ( - ) . For each combination , extracts from leaves of three different N . benthamiana plants have been analyzed and representative chromatograms are shown . DOI: http://dx . doi . org/10 . 7554/eLife . 23001 . 016 When the CfCYP76AH enzymes were assayed in triplet combinations , the product profiles were further shifted towards multi-oxygenated 13R-manoyl oxide derivatives . The formation of minor amounts of deacetylforskolin ( 13b ) and several compounds with identical mass to charge ratio ( m/z ) but different retention times were detected using different enzyme combinations ( Figure 5 , Supplementary file 1 ) . The triplet combination CfCYP76AH15 , CfCYP76AH11 and CfCYP76AH16 led to the highest amounts of 13b . Thus , this combination of multifunctional P450s appeared to constitute the optimal biosynthetic pathway for specific formation of 13b from 1 ( Figure 5 ) . In addition to 13R-manoyl oxide-derived diterpenoids , the root cork of C . forskohlii contains numerous abietane diterpenoids derived from miltiradiene ( Alasbahi and Melzig , 2010a ) . Recently a pair of diterpene synthases ( CfTPS1 and CfTPS3 ) mainly expressed in the root cork of C . forskohlii was demonstrated to produce miltiradiene ( Pateraki et al . , 2014 ) . It has been shown previously that members of the CYP76AH subfamily in Lamiaceae are able to oxygenate miltiradiene or derivatives thereof . CYP76AH1 from Salvia miltiorrhiza ( Guo et al . , 2013 ) , CYP76AH4 as well as RoFS1 and RoFS2 from Rosmarinus officinalis ( Božić et al . , 2015; Zi and Peters , 2013 ) and CYP76AH24 from S . pomifera catalyze synthesis of ferruginol , the precursor of carnosic acid and tanshinones ( Guo et al . , 2013; Ignea et al . , 2016a ) , from miltiradiene or dehydroabietadiene . Additionally , CYP76AH3 from S . miltiorrhiza has been shown to accept ferruginol as a substrate to produce sugiol , 11-hydroxy-ferruginol and 11-hydroxy-sugiol ( Guo et al . , 2016 ) . These miltiradiene-accepting CYP76AHs show high-sequence homology , ranging from 60% to 85% at the amino acid level , with those identified in C . forskohlii . Therefore , it was tempting to study the ability of the CfCYP76AHs to metabolize miltiradiene . The CfCYP76AHs were co-expressed individually in N . benthamiana leaves producing miltiradiene ( Andersen-Ranberg et al . , 2016 ) , and the product profile monitored by unbiased LC-MS analysis . CfCYP76AH15 was shown to convert miltiradiene to ferruginol ( Figure 6 ) . Ferruginol was identified in extracts of C . forskohlii root cork ( Figure 6 ) , so it is possible that CfCYP76AH15 is also involved in the biosynthesis of ferruginol in planta . In a parallel series of experiments , CYP76AHs from rosemary and salvia known to accept miltiradiene as substrate were tested for their ability to use 13R-manoyl oxide as a substrate . Transient expression of RoCYP76AH4 ( Zi and Peters , 2013 ) , RoFS1 and SfFS ( Božić et al . , 2015 ) in N . benthamiana leaves able to synthesize 13R-manoyl oxide demonstrated that RoCYP76AH4 was able to efficiently convert 13R-manoyl oxide to 11-oxo-13R-manoyl , while RoFS1 and SfFS were able to produce 11-hydroxy manoyl oxide , in addition to 11-oxo-13R-manoyl oxide ( Figure 7 ) . 10 . 7554/eLife . 23001 . 017Figure 6 . GC-MS analysis of miltiradiene-derived diterpenoids obtained by transient expression of CfCYP76AH15 in N . benthamiana leaves . ( A ) Total ion chromatograms ( TIC ) of extracts transiently expressing CfCXS , CfGGPPS , CfTPS1 and CfTPS3 ( miltiradiene biosynthesis genes ) in combination with water ( - ) or CfCYP76AH15 . Dehydroabietadiene ( DE ) and miltiradiene ( MT ) were observed in the ( - ) extract , whereas ferruginol was observed in extracts from tissue expressing the miltiradiene biosynthesis genes together with CfCYP76AH15 . In root cork extracts , ferruginol was detected together with dehydroabietadiene . Presence of ferruginol was confirmed by comparison to an authentic standard ( Ignea et al . , 2016a ) , while presence of miltiradiene ( B ) and dehydroabietadiene ( C ) were confirmed by comparison of m/z spectra with previously characterized compounds ( Andersen-Ranberg et al . , 2016 ) . For every combination , extracts from leaves of three different N . benthamiana plants have been analyzed and representative chromatograms are shown . DOI: http://dx . doi . org/10 . 7554/eLife . 23001 . 01710 . 7554/eLife . 23001 . 018Figure 7 . GC-MS analysis of 13R-manoyl oxide-derived diterpenoids obtained by transient expression of CYP76AHs in N . benthamiana leaves . ( A ) Total ion chromatograms ( TIC ) of extracts transiently expressing CfCXS , CfGGPPS , CfTPS2 and CfTPS3 ( 13R-manoyl oxide biosynthesis genes ) in combination with water ( - ) , CfCYP76AH15 , RoCYP76AH4 , RoFS1 and SpFS are shown . 13R-manoyl oxide was observed in the ( - ) extracts , while 11-oxo-13R-manoyl oxide ( 2 ) was observed in the CfCYP76AH15 , RoCYP76AH4 , RoFS1 and SfFS extracts . 11-Hydroxy-13R-manoyl oxide ( 3d ) is observed only in extracts expressing the RoFS1 and SfFS1 genes . Presence of 11-hydroxy-13R-manoyl oxide was verified by comparison to an authentic standard ( Ignea et al . , 2016b ) while identification of 11-oxo-13R-manoyl oxide was confirmed by comparison of m/z spectra with a previously characterized compound ( 2 ) . The results show RoCYP76AH4 has an activity similar to CfCYP76AH15 , able to convert efficiently and specifically 13R-manoyl oxide to 2 . RoFS1 , as well as SfFS , can also convert 13R-manoyl oxide to 2 but they catalyze the synthesis of an additional product , 11-hydroxy-13R-manoyl oxide ( 3d ) . For every combination , extracts from leaves of three different N . benthamiana plants have been analyzed and representative chromatograms are shown . DOI: http://dx . doi . org/10 . 7554/eLife . 23001 . 018 To complete the biosynthetic route to forskolin , specific acetylation of the C-7 hydroxyl group of deacetylforskolin ( 13b ) is required . The root cork transcriptome was mined for acyltransferases ( ACTs ) from clade III of the BAHD family earlier reported to predominantly use acetyl CoA as acetyl donor for acetylation of hydroxyl groups ( D'Auria , 2006 ) . Ten ACTs ( Figure 1—source data 1 ) were identified , cloned and tested functionally by Agrobacterium-mediated transient expression in N . benthamiana leaves , engineered to produce deacetylforskolin by co-expression of the enzymes CfDXS , CfGGPPS , CfTPS2 , CfTPS3 , CfCYP76AH15 , CfCYP76AH11 and CfCYP76AH16 . Two ACT candidates , CfACT1-6 and CfACT1-8 , were found to catalyze acetylation of 13b ( Figure 8 ) . Expression of CfACT1-6 resulted in formation of a broad range of acetylated products of which forskolin constituted a minor fraction . In contrast , CfACT1-8 exhibited high activity and specificity , with efficient conversion of 13b to forskolin and absence of detectable acetylated side products . Identification of this enzyme establishes the entire and highly specific biosynthetic route to forskolin from its precursor , GGPP ( Figure 9 ) . 10 . 7554/eLife . 23001 . 019Figure 8 . De novo biosynthesis of forskolin by transient expression of C . forskohlii genes in N . benthamiana as monitored by LC-MS-based extracted ion chromatograms ( EIC ) . To monitor both deacetylforskolin ( 13b ) and forskolin ( 16c ) , the EIC were selected as the sum of m/z 433 . 2 ± 0 . 1 and m/z 391 . 2 ± 0 . 1 . Chromatograms represent LC-MS analysis of extracts from leaves expressing the 13R-manoyl oxide biosynthesis genes ( CfDXS , CfGGPPS , CfTP2 and CfTPS3 ) in combination with ( from the top ) : water ( - ) ; CfCYP76AH15 , CfCYP76AH11 and CfCYPAH16; CfCYP76AH15 , CfCYP76AH11 , CfCYPAH16 and CfACT1-6; CfCYP76AH15 , CfCYP76AH11 , CfCYPAH16 and CfACT1-8 , shown together with authentic standards ( deacetylforskolin and forskolin ) . Forskolin ( 16c ) was identified together with two other acetylated compounds ( e . g . 16a , 16b ) with the same molecular mass in leaves expressing CfACT1-6 together with forskolin-specific CYPs ( Supplementary file 1 ) . When CfACT1-8 was expressed instead of CfACT1-6 , a predominant accumulation of forskolin was observed , with a drastic reduction of non-specific acetylated products . For all combinations , extracts from leaves of three different N . benthamiana plants have been analyzed and representative chromatograms are shown . DOI: http://dx . doi . org/10 . 7554/eLife . 23001 . 01910 . 7554/eLife . 23001 . 020Figure 9 . LC-qTOF-MS analysis of 13R-manoyl oxide-derived diterpenoids obtained by transient expression of combinations of C . forskohlii CYP and ACT encoding genes in N . benthamiana leaves . Total ion chromatograms ( TIC ) from extracts expressing the 13R-manoyl oxide biosynthesis genes ( CfCXS , CfGGPPS , CfTPS2 , CfTPS3 ) in combination with ( from the top ) water ( - ) , CfCYP76AH15 + CfCYP76AH11+ CfCYP76AH16 + CfACT1-6 , and CfCYP76AH15 + CfCYP76AH11 + CfCYP76AH16 + CfACT1-8 are shown . Hydroxylated and acetylated 13R-manoyl oxide-derived diterpenoids ( marked with grey bars ) and their identity , including their molecular formulas , was confirmed by their accurate mass ( 5 ppm tolerance , Supplementary file 1 ) . Compounds present in trace amounts are not marked . The identity of deacetylforskolin ( 13b ) and forskolin ( 16c ) was confirmed by comparison to authentic standards . No 13R-manoyl oxide-derived diterpenoids were detected in the water control ( − ) . For all combinations , extracts from leaves of three different N . benthamiana plants have been analyzed and representative chromatograms are shown . DOI: http://dx . doi . org/10 . 7554/eLife . 23001 . 020 Expression and engineering of plant biosynthetic pathways in microbial organisms provides a method for sustainable production of high value compounds like the structurally complex bioactive diterpenoids ( Guo et al . , 2013; Ignea et al . , 2016a; Jia et al . , 2016 ) . With the genes encoding the entire biosynthetic pathway of forskolin identified , we proceeded to establish stable forskolin production in S . cerevisiae , an excellent host organism for biosustainable and scalable production of numerous bio-active natural products ( Brochado et al . , 2010; Brown et al . , 2015; Galanie et al . , 2015; Hansen et al . , 2009; Ignea et al . , 2016a; Jeandet et al . , 2012 ) . For expression in yeast , all C . forskohlii genes were codon-optimized and stably integrated in neutral loci in the yeast genome . Genomic integration was chosen versus expression via episomal plasmids as the former strategy favors the simultaneous expression of a large number of genes as well as effective selection marker recycling ( Jensen et al . , 2014 ) . Additionally , a sequence encoding a NADPH-dependent cytochrome P450 oxidoreductase ( CfPOR ) , required to support the P450 activity , was identified from the C . forskohlii root cork transcriptome and cloned for co-integration in the yeast genome . The isolated CfPOR was the only one present in the root cork transcriptome . Genomic integration enabled stable , simultaneous expression of a total of eight heterologous genes in the microbial host . CfPOR , CfCYP76AH15 , CfCYP76AH11 , CfCYP76AH16 and CfACT1-8 were co-expressed in the yeast strain EFSC4498 , previously engineered to produce 350 mg/L 13R-manoyl oxide ( Andersen-Ranberg et al . , 2016 ) . Transformed yeast strains verified for the integration of all forskolin biosynthetic cDNAs into their genome , were further analyzed and production titers of forskolin and pathway intermediates were monitored . The highest forskolin producing strain , EVST21543 , which demonstrated genetic stability through several rounds of cultivation , was grown in a 5 L fermenter using minimal medium under glucose-limited conditions . During fermentation , accumulation of forskolin ( 16c ) , 13R-manoyl oxide ( 1 ) , 9-hydroxy-13R-manoyl oxide ( 3a ) , and biomass formation ( Figure 10 ) were monitored over time . Forskolin levels reached more than 40 mg/L of yeast culture . Simultaneous accumulation of high titers of 13R-manoyl oxide ( 1 ) and 9-hydroxy-13R-manoyl oxide ( 3a ) at levels of 200 and 500 mg/L , respectively , indicated that the conversion of 13R-manoyl oxide ( 1 ) to forskolin was far from complete . Interestingly , these intermediates did not accumulate in the C . forskohlii root cork . A comparison of the total intermediate profiles of C . forskohlii root cork versus the fermenter grown EVST21543 yeast strain is shown in ( Figure 10—figure supplement 1 ) . 10 . 7554/eLife . 23001 . 021Figure 10 . Forskolin production in S . cerevisiae following stable genomic integration of codon-optimized C . forskohlii genes . ( A ) Forskolin ( 16c ) accumulation in a fermenter batch using the EVST21543 strain ( expressing CfCYP76AH15 , CfCYP76HA11 , CfCYP76AH16 and CfACT1-8 encoding genes in the EFSC4498 S . cerevisiae strain , optimized for the production of 13R-manoyl oxide [Andersen-Ranberg et al . , 2016] ) . ( B ) 13R-manoyl oxide ( 1 ) accumulation in EVST21543 strain . ( C ) 9-Hydroxy-13R-manoyl oxide ( 3a ) accumulation in EVST21543 strain . ( D ) EVST21543 strain biomass monitored during the fermentation process . ( E ) The biosynthetic pathway used for the production of forskolin in yeast . The fermentation event occurred once , and a triplicate of samples were analysed from each time course . DOI: http://dx . doi . org/10 . 7554/eLife . 23001 . 02110 . 7554/eLife . 23001 . 022Figure 10—figure supplement 1 . Comparison of metabolite profiles between fermenter grown yeast culture of the EVST21543 strain and C . forskohlii root extract analyzed by LC-MS . Forskolin ( 16c ) and 13R-manoyl oxide ( 1 ) were identified based on co-elution with standards and 9-hydroxy-13R-manoyl oxide ( 3a ) was identified based on the presence of the [M+Na]+ ion , 329 . 2457 ( C20H34O2Na+ , Δ 0 . 7 ppm ) . DOI: http://dx . doi . org/10 . 7554/eLife . 23001 . 022 The terpenoid biosynthetic pathways active in the root of C . forskohlii produce an array of 13R-manoyl oxide and miltiradiene-derived diterpenoids , including forskolin . Forskolin is one of the most complex and highly oxygenated diterpenoids reported in C . forskohlii . In the current study , the genes encoding the entire biosynthetic pathway for forskolin were identified . Availability of the transcriptome from the root cork cells of C . forskohlii , where forskolin biosynthesis takes place , the option to achieve rapid functional characterization of the gene candidates in planta by transient expression in N . benthamiana and high-sensitivity techniques for structural characterization made the pathway elucidation possible . With the genes encoding the entire forskolin biosynthetic pathway in hand , de novo production of forskolin in engineered yeast was achieved . Initially , a number of genes encoding CYP76AH subfamily members , expressed mainly in the root cork of C . forskohlii , was cloned and transiently expressed in N . benthamiana leaves being able to produce 13R-manoyl oxide . The products profiles obtained with these enzymes revealed that the identified CfCYP76AHs have discrete roles in forskolin biosynthesis . Efficient monooxygenation at C-11 is catalyzed mainly by CfCYP76AH15 ( but also by CfCYP76AH8 , CfCYP76AH17 and CfCYP76AH11 ) . Monooxygenation at position C-9 is catalyzed exclusively by CfCYP76AH16 and monooxygenation at C-1 , C-6 and C-7 mainly by CfCYP76AH11 . Monooxygenation at C-1 was also observed using CfCYP76AH8 and CfCYP76AH17 ( 4c , Figure 3—figure supplement 1 ) . Collectively , this set of data displays the multifunctional roles of these enzymes . Together , they could account for all the oxygenated positions in forskolin . Co-expression of CfCYP76AH15 , CfCYP76AH11 and CfCYP76AH16 resulted in specific and efficient formation of the final intermediate , deacetylforskolin . Despite the complementary multifunctionality of the CYP76AH enzymes , partial functional redundancy is possible , as demonstrated by the varied oxygenation patterns observed in experiments with single enzymes . Overlapping functionalities may contribute to a coordinated action for efficient conversion of 1 to 13b . Moreover , certain C . forskohlii CYP76AH enzymes seem able to accept oxygenated forms of 1 as substrates which results in the observed shifts in profile toward higher decorated products when they are co-expressed in N . benthamiana ( Supplementary file 1 ) . Although co-expression of CfCYP76AH11 and CfCYP76AH16 with either one of the three CfCYP76AH8 , CfCYP76AH17 or CfCYP76AH15 in the engineered system resulted in biosynthesis of deacetylforskolin ( 13b ) ( Figure 5 ) , the precise sequence of in planta 13R-manoyl oxide oxygenations cannot be deduced from the experimental results partly because all identified CfCYP76AHs accept 1 as substrate . The co-expression of the CfCYP76AH encoding genes in the root cork of C . forskohlii and their partial functional redundancy or complementarity may in vivo constitute the basis for the chemical diversity of labdane terpenoids present in the root cork of C . forskohlii . In planta , the forskolin biosynthetic pathway would thus appear to be entangled within a metabolic grid offering simultaneous production of a multitude of other diterpenoids . Recently , CYP76AH enzymes accepting miltiradiene ( a non-epoxylabdane , abietane diterpenoid , which is also present in C . forskohlii roots ) as substrate were reported from other Lamiaceae species ( Božić et al . , 2015; Guo et al . , 2016; Ignea et al . , 2016a; Zi and Peters , 2013 ) . This prompted us to examine whether the promiscuous and multifunctional CfCYP76AH could accept miltiradiene as substrate , and vice versa , e . g . if different Lamiaceae CYP76AHs can catalyze oxygenations of 13R-manoyl oxide ( Figures 6 and 7 ) . According to our results CfCYP76AH15 was found to have very similar catalytic activities compared to the rosemary CYP76AH4 , an enzyme with a suggested role in the oxygenation of miltiradiene toward the synthesis of ferruginol ( Zi and Peters , 2013 ) . Efficient formation of ferruginol as well as 11-oxo-13R-manoyl oxide , by both enzymes indicates that they may represent orthologues . Two additional ferruginol synthases of the CYP76AH subfamily , one from Salvia fruticosa ( SfFS ) and one from Rosemary officinalis ( RoFS ) , were found to catalyze the conversion of 13R-manoyl oxide to 11-oxo-13R-manoyl oxide and 11-hydroxy-manoyl oxide when expressed in N . benthamiana leaves ( Figures 6 and 7 ) . These findings are not reflected in the phylogenetic analysis of the known CYP76AHs . All C . forskohlii CYP76AHs able to produce 11-oxo-13R-manoyl oxide are clustered together , while CYP76AHs from Salvia spp . and R . officinalis that can catalyze the formation of the same compound , as well as those CYPs able to accept miltiradiene as substrate , form a different cluster when analyzed with currently known CYP76AHs . Thus , it is likely that the ability of CYP76AHs to catalyze 11-oxo-13R-manoyl oxide has evolved convergently in these plants ( Figure 11 ) . 10 . 7554/eLife . 23001 . 023Figure 11 . Phylogeny of known full-length CYP76AHs . The enzymes used are listed below with their accession numbers or source of publication: CfCYP76AH15 , KT382358; CfCYP76AH17 , KT382360; CfCYP76AH8 , KT382348; CfCYP76AH11 , KT382349; CfCYP76AH16 , KT382359; CfCYP76AH9 , KT382347; CfCYP76AH10 , KT382346; CfCYP71D381 , KT382342; RoFS1 , AJQ30187 ( Božić et al . , 2015 ) ; SmCYP76AH3 , KR140168 ( Guo et al . , 2016 ) ; RoFS2 , AJQ30188 ( Božić et al . , 2015 ) ; SfFS , AJQ30186 ( Božić et al . , 2015 ) ; RoCYP76AH4 , ( Zi and Peters , 2013 ) ; RoCYP76AH5v1 , ( Zi and Peters , 2013 ) ; RoCYP76AH5v2 , ( Zi and Peters , 2013 ) ; RoCYP76AH6 , ( Zi and Peters , 2013 ) ; RoCYP76AH7 , ( Zi and Peters , 2013 ) ; SmCYP76AH1 , AGN04215 ( Guo et al . , 2013 ) ; SpCYP76AH24 , ALM25796 ( Ignea et al . , 2016a ) . Coleus forskohlii enzymes are indicated by a solid black triangle . CfCYP71D381 was chosen as a root because it can accept 13R-manoyl oxide as a substrate , but does not catalyze the synthesis of forskolin-related products . The number subscripts indicated at each enzyme refer to their respective enzymatic products , the structures of which are given on the right . Only the main products of each enzymes are mentioned . MO stands for manoyl oxide . DOI: http://dx . doi . org/10 . 7554/eLife . 23001 . 023 These data highlight the functional versatility of the CYP76 family . The enzymes can exhibit broad substrate specificity which may advance metabolic evolution as they provide metabolic plasticity and flexibility affording synthesis of new diterpenoids . This facilitates the expansion of the number of possible diterpenoids produced in nature and potentially serves to diversify and augment the phytochemical defense of plants . The promiscuity of the CYP76AHs also provides potentials for their use in combinatorial approaches for synthesizing a range of diterpenoids with pharmaceutical relevance . The exclusive presence of the CYP76AH subfamily in Lamiaceae species may reflect gene duplications in a Lamiaceae ancestral species followed by expansion and neofunctionalization after speciation ( Figure 11 ) . The identification of CfACT1-8 as an ACT catalyzing regiospecific acetylation of deacetylforskolin ( 13b ) to afford forskolin completed the entire biosynthetic pathway for forskolin from its precursor , GGPP . Interestingly , the only currently identified acyltransferases in diterpenoid biosynthesis are those involved in the biosynthesis of paclitaxel . Those acetyltransferases show substantial regioselective promiscuity ( Ondari and Walker , 2008; Walker and Croteau , 2000 ) and belong to Clade V ( D'Auria , 2006; Tuominen et al . , 2011 ) , whereas the majority of the ACTs identified in the root cork transcriptome of C . forskohlii , including ACT1-6 and ACT1-8 , belong to Clade III ( Figure 12 ) . 10 . 7554/eLife . 23001 . 024Figure 12 . Phylogenetic tree of CfACT encoding candidate genes together with BAHD family acyltransferase representatives from all clades according to D'Auria ( 2006 ) . Accession numbers of the non-Coleus forskohlii selected protein sequences are shown next to the tree taxon names , while C . forskohlii peptide accession numbers are provided in Figure 1—source data 1 . The analysis only includes functionally characterized members . Coleus forskohlii enzymes are indicated by a solid black triangle . The majority of the selected CfACTs belong to Clade III , which includes mainly members which accept a diverse range of hydroxylated substrates and use acetyl-CoA as the main acyl donor ( D'Auria , 2006 ) . Interestingly , the ACTs known to be involved in Taxol biosynthesis belong to Clade V . DOI: http://dx . doi . org/10 . 7554/eLife . 23001 . 024 With all forskolin biosynthetic pathway genes identified , we moved to the generation of a stable forskolin producing S . cerevisiae strain . To engineer a stable microbial production platform , we proceeded to integrate the minimum required set of functional parts into the S . cerevisiae genome . Specific de novo production of the highly functionalized diterpenoid at titers above 40 mg/L was achieved through a pathway consisting of a total of 10 enzymatic steps catalyzed by eight heterologously co-expressed enzymes . This high forskolin titer , achieved with no optimization steps , highlights the potential to develop a microbial manufacturing platform for efficient and stereospecific production of forskolin with further fine-tuning of the biosynthetic pathway . Currently , it is not possible to accurately estimate the forskolin titers necessary for industrially profitable production , as the exact commercial applications , market size and price as well as the production cost including downstream processing cannot be determined . Given the knowledge gained in our present study and experiences with other compounds ( Paddon and Keasling , 2014 ) we find it realistic to aim for yields ranging from a single to double digits of gram per liter of yeast culture . To achieve higher forskolin yields , it is important upfront to ascertain a proper flux toward GGPP through the mevalonate pathway ( Kampranis and Makris , 2012 ) . Specifically for forskolin pathway , it seems clear that there is a limitation in flux through one or several of the P450s involved as we encounter accumulation of the intermediates , 300 mg/L and 500 mg/L of compounds 1 and 3a respectively , compared to forskolin ( 40 mg/L ) . Accumulation of only minute amounts of deacetylforskolin shows that CfACT1-8 is not a limiting step in the pathway . Accumulation of 1 and 3a intermediates was not observed in planta ( Figure 10—figure supplement 1 ) . This likely signifies that the expression levels of the heterologous CYPs expressed in yeast are not properly balanced or their efficiency and activity can be affected negatively after incoorporation into the yeast membrane , while in the native plant host , the pathway exhibits optimized carbon flux and enzymatic efficiency . P450s are notorious difficult to express in high amounts in yeast and recognized as exhibiting rather low Kcat values ( Jung et al . , 2011; Renault et al . , 2014 ) . Hence , to increase forskolin production in the yeast system , efforts should obviously be focused on optimizing CfCYP76AHs expression and enzyme kinetics , specificity and catalytic efficiency as well as pathway scaffolding to facilitate formation of a metabolon which will result in improved pathway flux and efficiency and reduced accumulation of pathway intermediates ( Laursen et al . , 2016 ) . The high forskolin titers already obtained though in the engineered yeast strain highlights the potential to develop a microbial manufacturing platform for efficient and stereospecific production of forskolin and other labdane terpenoids by fine tuning the biosynthetic pathways . A yeast-based production platform constitutes a sustainable alternative to traditional crop-based production but the gains always need to be compared to yield improvements obtained by classical or molecular breeding of the traditional host plant ( Graham et al . , 2010 ) . The model-example from the literature and industry toward production of a pharmaceutically relevant terpenoid in S . cerevisiae is the sesquiterpenoid artemisinic acid , a pathway intermediate to the antimalarial compound artemisinin ( Paddon et al . , 2013 ) . However , this prominent model example is dependent on a costly organic chemical synthesis component to chemically convert artemisinic acid to artemisinin ( Peplow , 2016 ) . Recent approaches of engineered de novo production of structurally complex diterpenoids , triterpenoids or alkaloids in microbial systems have also been limited to proof-of-concept studies , expression of partial pathways and sub-milligram yields , highlighting the challenges in synthetic biology to offer an economically realistic and sustainable alternative to isolation of the desired medicinal compounds from medicinal plants bred to produce elevated levels . The constraints to achieve high yields are connected to expression of multiple CYPs and reconstruction of pathways with multiple functionally divergent steps ( Brown et al . , 2015; Li and Smolke , 2016; Zhou et al . , 2015 ) . Strategies addressing these issues are for example the development of synthetic microbial consortia of S . cerevisiae and E . coli , optimization of CYPS for functional expression in E . coli , optimization of interactions between the CYPs and their reductase partner and N-terminal modifications ( Biggs et al . , 2016; Laursen et al . , 2016; Vazquez-Albacete et al . , 2017; Zhou et al . , 2015 ) . Our current study demonstrates that mining for additional members of the CYP76AH family has the potential to facilitate the assembly of further optimized panels of mixed-species P450s for the biosynthesis of bioactive diterpenoids . This shows the great promise that combinatorial assembly including CYPs outside the CYP76AH subfamily may offer and the opportunity to design production systems for diterpenoids that are currently inaccessible due to their exclusive presence in rare or red-listed plants and to further expand the chemical diversity of diterpenoids to production of compounds currently not known in nature . All chemicals including forskolin were acquired from Sigma-Aldrich . An authentic standard of 13R-manoyl oxide was prepared as previously described ( Nielsen et al . , 2014 ) . CYP76AH4 ( Zi and Peters , 2013 ) was cloned from rosemary plants acquired at a local market in Copenhagen , Denmark . RoFS1 and SfFS ( Božić et al . , 2015 ) were kindly provided by Dr . Angelos Kanellis ( University of Thessaloniki , Greece ) . The deacetylation of forskolin has been achieved previously with the use of methanolic potassium carbonate which can provide 7-desacetylforskolin in 65% yield ( Kosley and Cherill , 1989 ) . Here , we carried out the deacetylation of forskolin using a solution of methanolic ammonia solution ( 2M ) to afford 7-desacetylforskolin quantitatively . The 1 hr NMR data of the deacetylated forskolin were in agreement with the reported data in literature . Coleus forskohlii root cork total RNA was extracted as previously described ( Pateraki et al . , 2014 ) . RNA was prepared for sequencing using the Illumina TruSeq sample preparation kit v2 using poly-A selection ( Illumina San Diego , USA ) . The fragments were clustered on cBot and sequenced with paired ends ( 2 × 100 bp ) on a HiSeq 2500 ( Illumina San Diego , USA ) according to the manufacturer's instructions . A total 106 . 2 million read-pairs were generated . Adaptor sequences were removed from raw reads and reads were trimmed at the ends to phred score 20 , using the fastq-mcf tool from ea-utils ( https://code . google . com/p/ea-utils/ ) . Processed reads were assembled using Trinity ( r2013-02-16 ) resulting in a total of 263 , 652 assembled putative transcripts . Transcript abundance estimation was performed using RSEM and the scripts provided with Trinity . Likewise , the putative coding sequences were predicted using the TransDecoder scripts from Trinity . The high-throughput RNA sequences reported here have been submitted to the short read archive ( SRA ) at the NCBI [accession no SAMN06013363] . Mining of the C . forskohlii transcriptome database was performed as previously described ( Zerbe et al . , 2013 ) using tBLASTx software and known CYP or acetyl transferase ( ACT ) sequences as query . The identified contigs were amplified from single stranded cDNA generated from root cork total RNA using the ‘SuperScript III First-Strand Synthesis System for RT-PCR’ ( Invitrogen ) and oligo-dT primer . Cloning of the putative CYP and ACT cDNAs was achieved after PCR amplification using gene-specific primers ( Figure 1—source data 3 ) that were designed based on the in silico sequences of the identified CYP and ACT contigs ( Figure 1—source data 1 ) . PCR products were cloned into the pJET1 . 2 vector and verified by sequencing . For the identified non-full-length cDNAs , ( CfCYP76AH15 , CfCYP76AH16 and CfACT1-8 ) , full-length transcripts were obtained using 5’RACE techniques . Amino acid sequences of the CYP76AHs of C . forskohlii and of currently available CYP76AHs from other plant species were used to construct a phylogenetic tree to infer the evolutionary history of these enzymes . Sequences were retrieved from the current work , GenBank database ( http://www . ncbi . nlm . nih . gov/ ) and original articles . Peptide alignments were performed using the MUSCLE program included in the MEGA6 software . Phylogenetic analyses were performed by the Maximum Likelihood method based on the Dayhoff matrix-based model with ‘uniform rates’ ( Schwarz and Dayhoff , 1979 ) and using all sites with ‘Nearest-Neighbor-Interchange’ heuristic method conducted using MEGA6 software ( Tamura et al . , 2013 ) . The tree is drawn to scale , with branch lengths measured in the number of substitutions per site . Bootstrap values shown in % were inferred from 1000 replicates . Branches supported by bootstrap values higher than 75% are shown . For the phylogenetic analyses of CfACT candidates , the identified sequences from C . forskohlii were analyzed together with BAHD family acyltransferase representatives from all clades ( D'Auria , 2006 ) . The analysis includes only functionally characterized members . The analysis was inferred by using the Maximum Likelihood method based on the Dayhoff matrix-based model with ‘uniform rates’ and using all sites ( Schwarz and Dayhoff , 1979 ) . Initial trees for the heuristic search were obtained automatically by applying Neighbor-Join and BioNJ algorithms to a matrix of pairwise distances estimated using a JTT model , and then selecting the topology with superior log likelihood value . The tree is drawn to scale , with branch lengths measured in the number of substitutions per site . Evolutionary analyses were conducted using MEGA6 software ( Tamura et al . , 2013 ) . Total RNAs from C . forskohlii tissues were extracted as previously described ( Pateraki et al . , 2014 ) and digested by DNase I on-column . The integrity of the RNA samples was evaluated using the RNA-nano assay using an Agilent 2100 Bioanalyser ( Agilent Technologies ) . First-strand cDNAs were synthesized from 0 . 5 µg of total RNA from an oligo-dT primer , using the ‘SuperScript III First-Strand Synthesis System for RT-PCR’ ( Invitrogen ) . The resulting cDNAs were diluted 10-fold . For the qRT-PCR reactions , gene specific primers were used ( Figure 1—source data 3 ) with Maxima SYBR Green/Fluorescein qPCR Master Mix ( Fermentas ) on a Rotor-Gene Q cycler ( Qiagen ) using the following cycling parameters: 95°C for 7 min , 35 cycles of 95°C for 15 s , 60°C for 30 s and 72°C for 30 s followed by a melting curve cycle from 60°C to 90°C . Eukaryotic initiation factor 4A ( TIF4a ) and Elongation Factor 1A ( EF1a ) were used as reference genes because they showed the lowest variation across different tissues ( Pateraki et al . , 2014 ) . No statistically significant differences were observed between the results obtained using the two different reference genes and the results presented were normalized based on TIF4a . Relative transcript abundance was calculated as the mean of three biological replications obtained using three different C . forskohlii plants , while the reactions were performed in three technical replicates . Amplification efficiency was calculated with the ‘Real Time PCR Miner’ ( http://www . miner . ewindup . info/Version2 ) . Efficiency-corrected ΔCT values were used to quantify relative differences in target gene transcript accumulation . Primer specificity was assessed by agarose gel analysis and sequencing of amplicons from representative reactions , as well as from melting curve analysis of every reaction . Functional characterization of the selected candidate genes was obtained using transient expression in N . benthamiana which offers optimal native plant protein translation and processing , convenient rapid and optional combinatorial co-expression of multiple genes from independent vectors , native subcellular location of diTPS and CYPs as well as the availability of an endogenous native pathway providing GGPP . For the functional characterization of C . forskohlii selected CYPs and testing of their ability to hydroxylate 13R-manoyl oxide ( 1 ) , genes encoding candidate CfCYPs were transiently co-expressed in N . benthamiana leaves together with C . forskohlii enzymes boosting the formation of 1 , namely 1-deoxy-D-xylulose 5-phosphate synthase ( CfDXS ) , geranylgeranyl diphosphate synthase ( CfGGPPS ) , CfTPS2 and CfTPS3 ( Pateraki et al . , 2014 ) . Transient expression in N . benthamiana was performed as previously described ( Bach et al . , 2014 ) . CfCYP cDNAs selected for functional testing were subcloned into pCAMBIA130035Su by USER cloning ( Nour-Eldin et al . , 2010 ) . Vectors carrying the selected cDNAs were then transformed into the agrobacterium strain AGL-1-GC3850 ( Bach et al . , 2014 ) . For the agro-infiltration , the OD of the agrobacterium cultures of all transformed agrobacteria strains was normalized to OD600 = 1 . Different combinations prepared from equal volumes of each culture of transformed agrobacterium strains expressing individual genes encoding CfCYPs or 13R-manoyl oxide biosynthetic enzymes were infiltrated into the leaves of 4–6 weeks old N . benthamiana plants . Controls encompassing expression of the CfCYP encoding genes without the 13R-manoyl oxide biosynthetic genes were used to assess the possibility of CfCYPs cross-reactivity with tobacco endogenous diterpenoids . Metabolites from transgenic N . benthamiana leaves were extracted by hexane and 85% MetOH and were analyzed by GC-MS and LC-MS-qTOF , respectively . LC-MS-qTOF analysis was performed on the system that was comprised of an Agilent G1312B SL binary pump , Agilent G1367C SL WP autosampler , Agilent G1316B column oven , Agilent G1315C Starlight DAD detector and Bruker microTOF II Mass Spectrometer using Electron Spray Ionization ( ESI ) . Samples were separated on a Synergi 2 . 5 mm Fusion-RP C18 column ( 50 × 3 . 2 mm i . d . , Phenomenex Inc . , Torrance , CA , USA ) at a flow rate of 0 . 2 mL/min with column temperature held at 25°C . The mobile phase consisted of water with 0 . 1% formic acid ( v/v; solvent A ) and 80% acetonitrile with 0 . 1% formic acid ( v/v; solvent B ) . The gradient program was 0 min , 60% B; 25 min , 98% B; 31 min , 98% B; 32 min , 60% B; 42 min , 60% B . Mass spectra were acquired in positive ion mode using a drying temperature of 200°C , a nebulizer pressure of 3 . 0 bar , and a drying gas flow of 7 L/min ( Luo et al . , 2016 ) . GC-MS analysis was performed on a Shimadzu GCMS-QP2010 Ultra using a 3 Agilent HP-5MS column ( 20 m × 0 . 180 mm i . d . , 0 . 18 µm film thickness ) . Injection volume and temperature was set to 1 µL and 250°C . GC program: 60°C for 1 min , ramp at rate 30°C min-1 to 180°C , ramp at rate 10°C min-1 to 290°C , ramp at rate 30°C min-1 to 320°C and hold for 2 min . H2 was used as carrier gas . Transfer line temperature was set to 280°C and electron impact ( EI ) was used as ionization method in the mass spectrometer ( MS ) with the ion source temperature and voltage set to 300°C and 70 eV . MS spectra were recorded from 50 m/z to 400 m/z . 11-Oxo-13R-manoyl oxide ( 2 ) , 2-hydroxy-13R-manoyl oxide ( 3b ) , 19-hydroxy-13R-manoyl oxide ( 3c ) , 1 , 11-dihydroxy-13R-manoyl oxide ( 5d ) , 1 , 9-deoxydeacetylforskolin ( 7 hr ) and 9-deoxydeacetylforskolin ( 10b ) were produced using the biosynthetic scheme described above by large-scale expression of the relevant gene combinations in N . benthamiana ( Andersen-Ranberg et al . , 2016 ) . CYP76AH8 was expressed to obtain biosynthesis of compound ( 2 ) , CYP71D381 to obtain ( 3b ) and ( 3c ) , whereas combined expression of CfCYP76AH8 and CfCYP76AH11 afforded ( 5d ) , ( 7 hr ) and ( 10b ) . For the large-scale experiments , agroinfiltration was performed by vacuum infiltration . For biosynthesis of each compound in amounts sufficient for NMR analysis , 100–200 g of fresh weight N . benthamiana leaf material were harvested 7 days after infiltration . Leaf material chopped into small pieces was extracted using 0 . 5 L of n-hexane . The solvent was thrice evaporated and recovered by rotor evaporation for repeated extraction of the same leaf material . Concentrated extracts from N . benthamiana leaves biosynthesizing target compounds were subjected to solid phase extraction ( SPE ) , using silica gel with n-hexane and ethyl acetate mixtures ( 100:0 , 99:1 , 98:2 , 94:6 , 92:8 , 88:12 ( v/v ) ) in steps of 100 mL . Fractions containing diterpenoids were identified by GC-MS . Diterpenoid-containing fractions were combined and the solvent removed by rotor evaporation and the samples resuspended in 1 mL n-hexane affording a crude fraction for further purification . Final isolation of individual diterpenoids was achieved using an Agilent 7890B GC installed with an Agilent 5977A MSD , GERSTEL Preparative Fraction Collector ( PFC ) AT 6890/7890 and a GERSTEL CIS 4C Bundle injection port . Separation was carried out using a RESTEK Rtx-5 column ( 30 m × 0 . 53 mm i . d . ×1 µm df ) with H2 as carrier gas . At the column outlet , a splitter was mounted with a split vent ratio of 1:100 to the MS and the PFC , respectively . Sufficient amounts of oxidized 13R-manoyl oxide-derived diterpenoids for NMR analysis ( 0 . 5–1 mg ) were obtained by 100 repeated injections of 5 µL extract aliquots . Injection port was set in solvent vent mode with a H2 gas flow of 100 mL/min until 0 . 17 min , combined with a sample injection speed of 1 . 5 mL/min . Purge flow was set to 3 mL/min from 0 . 17 min to 2 . 17 min . Injection temperature was held at 40°C for 0 . 1 min followed by ramping at 12°C/s until 320°C , which was held for 2 min . Column flow was set to 7 . 5 mL , which was held constant throughout the GC program . The GC program was set to hold at 60°C for 1 min , ramp 30°C/min to 220°C , ramp 2°C/min to 250°C and a final ramp of 30°C/min to 220°C , which was held for 2 min . Temperature of the transfer line from GC to PFC and the PFC itself was set to 250°C . The PFC was customized to collect the peaks of 2 , 3b , 3c , 5d and 7 hr , and 10b by their retention time identified by the MS . The MS for monitoring the PFC purification was set in scan mode from m/z 35 to m/z 500 , with a threshold of 150 ion counts . Solvent cut-off was set to 4 min and the temperature of the MS source and the MS quadropole - to 300°C and 150°C , respectively . For biosynthesis of 3a in amounts sufficient for NMR analysis , the S . cerevisiae strain EFSC4494 carrying chromosomally integrated CYP76AH16 was inoculated into a pre-culture of 5 mL selective media ( SC-Ura ) and grown for 16 hr at 30°C and 400 rpm . A 1 mL aliquot of the pre-culture was used for inoculation of 100 mL non-selective media ( SC ) and grown in a 500 mL Erlenmeyer shake-flask for 120 hr at 30°C with horizontal shaking at 180 rpm . Following addition of 100 mL UV-grade 99 . 9% ethanol and maintenance of the sample at 60°C for 20 min , 200 mL of n-hexane was added and the sample shaken for 2 hr at room temperature . The hexane phase was concentrated by rotor evaporation and subjected to column chromatography ( dual layer Florisil/Na2SO46 mL PP SPE TUBE , Supelco Analytical ) with a gradient composed of n-hexane and 1–15% ethyl acetate . All genes selected for functional expression in S . cerevisiae were codon optimized for efficient expression in S . cerevisiae , and purchased as DNA STRINGS ( Geneart , LifeTechnologies ) . Genomic integration was chosen over expression via episomal plasmids to favor simultaneous expression of a number of genes as well as to enable the use of selection marker recycling ( Jensen et al . , 2014 ) . The 13R-manoyl oxide producing S . cerevisiae strain EFSC4498 ( Andersen-Ranberg et al . , 2016 ) was used to test all the selected gene combinations for their ability of affording synthesis of forskolin and intermediate products . All genes were cloned into yeast genome integration plasmids by the USER technique ( Nour-Eldin et al . , 2010 ) targeting incorporation into site XI-2 ( Mikkelsen et al . , 2012 ) . Transformants were verified by PCR on genomic DNA for correct insertion of heterologous genes and grown and tested in 96-deep-well plates ( Andersen-Ranberg et al . , 2016 ) . The yeast strain found to produce the highest amount of forskolin and which exhibited stability through several cultivation rounds ( EVST21543 ) was selected for cultivation for 140 hr in a 5 L fermenter using minimal medium and glucose-limited conditions . Forskolin production was monitored using withdrawn culture aliquots . Forskolin was extracted from the mixture of yeast cells and culture broth using 85% methanol and incubation for 20 min at 75°C and the extract centrifuged ( 10 , 000 g for 5 min ) to precipitate yeast debris . The supernatant obtained was used after filtration for LC-MS analysis and forskolin quantification . For forskolin quantification , aliquots of the yeast samples ( with broth ) collected at specific time points ( Figure 3 ) were combined with methanol to give a concentration of 85% methanol , incubated at 75°C for 20 min , filtered and then analyzed by LC-MS . Quantification was based on a standard calibration curve of forskolin purchased from Sigma-Aldrich . An Ultimate 3000 UHPLC+ Focused system ( Dionex Corporation , Sunnyvale , CA ) coupled to a Bruker Compact ESI-QTOF-MS ( Bruker Daltonik , Bremen , Germany ) was used to quantify forskolin . Samples were separated on a Kinetex XB-C18 column ( 100 × 2 . 1 mm i . d . , 1 . 7 μm particle size , 100 Å pore size; Phenomenex Inc . , Torrance , CA ) maintained at 40°C with a flow rate of 0 . 3 mL/min and mobile phase consisting of 0 . 05% ( v/v ) formic acid in water ( solvent A ) and 0 . 05% ( v/v ) formic acid in acetonitrile ( solvent B ) . The gradient LC method used for quantification was as follows: solvent B was held at 20% for 30 s , then ramped to 100% over 8 . 5 min , held at 100% for 2 min , decreased to 20% over 30 s and held for 3 . 5 min to give an overall run time of 15 min . The ESI source parameters were as follows: capillary voltage , 4500 V; nebulizer pressure 1 . 2 bar; dry gas flow , 8 l/min; dry gas temperature , 250°C . The QTOF-MS was operated in MS only mode with collision cell energy of 7 eV and collision cell RF of 500 Vpp . Ions were monitored in the positive mode over a range of 50–1300 m/z and spectra collected at a rate of 2 Hz . For comparison of yeast profiles to C . forskohlii root extract , roots were grinded and then extracted with 85% methanol , incubated at 75°C for 30 min , filtered and analyzed by LC-MS . Analysis was performed as described for forskolin quantification but with the following gradient method: 20% B for 1 min , increased to 100% B over 22 min and then returned to 20% B in 0 . 5 min and held for 4 min . Yeast samples were collected at specific time points ( Figure 10 ) and samples kept at −20°C in glass vials . For diterpenoid extraction , 500 μL of n-hexane was added to 500 μL yeast broth , shaken for 1 hr at room temperature and separated into two phases by centrifugation at 2500 rpm and stored overnight at 4°C . The hexane phase was then diluted 10 times and run on a SCION 436 GC-FID ( Bruker ) . Sample ( 1 μL ) was injected in splitless mode at 280°C . The GC-program was as follows: 60°C for 1 min , ramp at 20°C/min to 160°C , ramp at 10°C/min from 160°C to 240°C , ramp at 20°C/min from 240°C to 320°C , hold at 320°C for 2 min . H2 was used as carrier gas with a linear flow of 50 mL/min . The FID was set at 300°C , with N2 flow of 25 mL/min , H2 at 30 mL/min and air at 300 mL/min . Data sampling rate was 10 Hz . Compounds 1 and 3a were identified by comparing the retention time with an authentic standard and quantification was based on FID peak area and a standard curve of 1 . All NMR experiments were recorded at 300 K in CDCl3 using a Bruker Avance III 600 MHz NMR spectrometer ( 1H operating frequency 600 . 13 MHz ) equipped with a Bruker SampleJet sample changer and a cryogenically cooled gradient inverse triple-resonance 1 . 7 mm TCI probe-head ( Bruker Biospin , Rheinstetten , Germany ) . The experiments were acquired in automation ( temperature equilibration to 300 K , optimization of lock parameters , gradient shimming , and setting of receiver gain ) . Both one-dimensional 1H and 13C spectra were acquired with 30°-pulses and 64k data points . The 1H spectra were recorded with 3 . 66 s inter-pulse intervals and the FID was multiplied with an exponential function corresponding to line-broadening of 0 . 3 Hz prior to Fourier transform . An acquisition time of 0 . 9 s were used for the 13C experiments with an additional relaxation delay of 2 . 0 s . Protons were decoupled during acquisition using waltz16 composite pulse sequence . Backward linear prediction was used to correct the first complex data points of 13C FIDs before zero-filling to 128k data points and application of exponential window function with a line-broadening factor of 1 . 0 Hz . Two-dimensional homo- and heteronuclear experiments were acquired with 2048 data points in the direct dimension and 128 ( DQF-COSY and HMBC ) or 256 ( multiplicity edited HSQC and phase sensitive NOESY ) data points in the indirect dimension; with spectral widths optimized from the corresponding 1H spectra . The HMBC and HSQC experiments were optimized for nJH , C = 10 Hz and 1JH , C = 145 Hz , respectively . Acquisition and processing of NMR data were performed using Topspin ver . 3 . 0 ( Bruker Biospin GmbH ) , and IconNMR ver . 4 . 2 ( Bruker Biospin GmbH ) was used for controlling automated sample change and acquisition .	Unlike animals , plants cannot move away from a herbivore or other threats . Instead , they have evolved to produce a vast array of chemical compounds to protect themselves . Some of these compounds are also important to humans , for example , as medicines or fragrances . Plants usually only produce small amounts of these compounds in mixtures with many other compounds , which makes it difficult to purify them . As a result , the methods of purifying the compounds may require huge amounts of plant material , or be expensive and not environmentally friendly . One solution to this would be to genetically engineer microbes like bacteria or yeast to produce the compounds instead . In order to do that , we need to understand exactly which enzymes the plant uses to make each compound and introduce them into suitable microbes . A compound called forskolin has been used since ancient times in traditional Indian medicine to treat conditions like high blood pressure , asthma and heart complications . Forskolin is found exclusively in the root of a plant called Coleus forskohlii , which is native to India and south-east Asia . It is stored inside cells within the bark of the root in structures called oil bodies , which are similar to oil drops . However , it is not known where forskolin is made , or which enzymes are involved . Pateraki , Andersen-Ranberg et al . set out to uncover how C . forskohlii produces this compound . The experiments show that forskolin is produced within the cells that contain the oil bodies . A technique called RNA sequencing was used to identify several genes that are highly active in these cells and encode enzymes that could potentially be involved in producing forskolin . Further experiments demonstrated that these enzymes drive a cascade of chemical reactions that convert a molecule called 13R-manoyl oxide into forskolin . Next , Pateraki , Andersen-Ranberg et al . inserted the genes into yeast cells that could already produce 13R-manoyl oxide , which allowed the yeast to produce relatively high amounts of forskolin . These findings show that it is possible to identify the genes involved in the production of medicinal compounds in a relatively short amount of time . This knowledge will aid the development of a method that can be used to produce forskolin and other similar compounds on a large scale without needing to harvest C . forskohlii plants .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "biochemistry", "and", "chemical", "biology" ]	2017	Total biosynthesis of the cyclic AMP booster forskolin from Coleus forskohlii
Macrophages derive from multiple sources of hematopoietic progenitors . Most macrophages require colony-stimulating factor 1 receptor ( CSF1R ) , but some macrophages persist in the absence of CSF1R . Here , we analyzed mpeg1:GFP–expressing macrophages in csf1r-deficient zebrafish and report that embryonic macrophages emerge followed by their developmental arrest . In larvae , mpeg1+ cell numbers then increased showing two distinct types in the skin: branched , putative Langerhans cells , and amoeboid cells . In contrast , although numbers also increased in csf1r-mutants , exclusively amoeboid mpeg1+ cells were present , which we showed by genetic lineage tracing to have a non-hematopoietic origin . They expressed macrophage-associated genes , but also showed decreased phagocytic gene expression and increased epithelial-associated gene expression , characteristic of metaphocytes , recently discovered ectoderm-derived cells . We further demonstrated that juvenile csf1r-deficient zebrafish exhibit systemic macrophage depletion . Thus , csf1r deficiency disrupts embryonic to adult macrophage development . Zebrafish deficient for csf1r are viable and permit analyzing the consequences of macrophage loss throughout life . Tissue resident macrophages ( TRMs ) are phagocytic immune cells that also contribute to organogenesis and tissue homeostasis . Therefore , perturbations in TRM production or activity can have detrimental consequences ranging from abnormal organ development to neurodegeneration and cancer ( Cassetta and Pollard , 2018; Mass et al . , 2017; Yang et al . , 2018; Zarif et al . , 2014 ) . In vertebrates , including mammals , birds , and fishes , TRMs derive from successive waves of hematopoiesis that initiate early during development reviewed in: McGrath et al . ( 2015 ) . The initial two embryonic waves give rise to primitive macrophages , born in the embryonic yolk sac in mammals and birds or the rostral blood island ( RBI ) in fishes , and erythro-myeloid precursors ( EMPs ) , which also originate in the yolk sac and expand in the fetal liver of mammals or emerge from the posterior blood island ( PBI ) of fishes . A third embryonic wave that generates definitive hematopoietic stem cells ( HSCs ) begins in the aorta-gonad-mesonephros ( AGM ) region , where HSCs bud from the hemogenic endothelium ( Bertrand et al . , 2010; Boisset et al . , 2010; Kissa and Herbomel , 2010 ) . In zebrafish , newly born hematopoietic stem cells ( HSCs ) migrate to the caudal hematopoietic tissue ( CHT ) , and later seed hematopoietic organs such as the kidney marrow , which is equivalent to the bone marrow in mammals ( Henninger et al . , 2017; Murayama et al . , 2006 ) . Most TRM populations are established by the end of fetal life and are subsequently maintained through the proliferation of local progenitors or through the partial contribution of bone marrow-derived cells ( Liu et al . , 2019 ) . During their colonization of the embryo , macrophages acquire distinct properties adapted to their microenvironment and allowing them to execute tissue niche-specific functions ( Bennett and Bennett , 2020; Gosselin et al . , 2014; Gosselin et al . , 2017; Lavin et al . , 2014; Matcovitch-Natan et al . , 2016 ) . The ontogeny of TRMs within a specific organ is heterogeneous and thought to be determined by the availability of the niche and accessibility of the host tissue reviewed in:Guilliams et al . ( 2020 ) . The microenvironment has a major role in determining TRM phenotype and function , largely regardless of ontogeny , but giving rise to heterogeneous populations of cells ( Lavin et al . , 2014; Shemer et al . , 2018; van de Laar et al . , 2016 ) . Colony stimulating factor 1 receptor ( CSF1R ) is an evolutionarily conserved regulator of macrophage development , directly inducing DNA and protein synthesis as well as proliferation upon ligand binding ( Hume et al . , 2016; Tushinski and Stanley , 1985 ) . Recessive and dominant mutations in CSF1R can cause severe brain disease ( Konno et al . , 2018a; Konno et al . , 2018b; Oosterhof et al . , 2019; Rademakers et al . , 2012 ) , associated with lower microglia density ( Oosterhof et al . , 2018 ) , but whether such mutations affect other myeloid cells , and how , remains unknown . Recently , patients carrying homozygous mutations in CSF1R and presenting with both leukodystrophy and osteopetrosis , phenotypes attributed to an absence of TRMs in the brain and bone , have been described ( Oosterhof et al . , 2019 ) . In mice and rats , the absence of CSF1R results in a complete lack of microglia , Langerhans cells ( LCs ) , and osteoclasts , while other subsets of TRMs are affected to varying degrees ( Cecchini et al . , 1994; Dai et al . , 2002; Erblich et al . , 2011; Ginhoux et al . , 2010; Oosterhof et al . , 2018; Pridans et al . , 2018 ) . It is unknown whether CSF1R is required for the development of early , embryonic TRM precursors and it remains elusive as to why only specific TRM populations are lacking in the absence of Csf1r . Furthermore , it is unclear whether macrophages that persist in Csf1r-deficient mice and rats have a normal macrophage phenotype . Detailed analysis of the Csf1r mutant phenotypes could therefore contribute to the identification of specific and universal features of organism-wide macrophage development . In addition , it is important to understand the systemic effects of CSF1R inhibition on macrophages , as inhibition of CSF1R is a clinical strategy for the intentional depletion of macrophages in various disease contexts , including Alzheimer’s disease , brain injury and cancer ( Edwards et al . , 2019; Lloyd et al . , 2019; Tap et al . , 2015; Webb et al . , 2018 ) . Zebrafish are particularly suitable to study immune cell development in vivo as they develop ex utero , are genetically tractable , and are transparent during early development ( Ellett and Lieschke , 2010; Gore et al . , 2018 ) . We used our previously generated zebrafish line that is deficient for both csf1ra and csf1rb paralogs ( csf1rDM ) , since the phenotypes of these fish , such as osteopetrosis and a lack of microglia , resemble those observed in mice , rats and humans ( Caetano-Lopes et al . , 2020; Chatani et al . , 2011; Dai et al . , 2002; Guo et al . , 2019; Meireles et al . , 2014; Oosterhof et al . , 2019; Oosterhof et al . , 2018; Pridans et al . , 2018 ) . The strong homology of basic developmental cellular processes has proven this model as indispensable for the identification of novel basic features of immune cell development and function ( Barros-Becker et al . , 2017; Bertrand et al . , 2010; Espín-Palazón et al . , 2014; Kissa and Herbomel , 2010; Madigan et al . , 2017; Tamplin et al . , 2015; Tyrkalska et al . , 2019 ) . Here , we aimed to determine how and when loss of Csf1r affects macrophage development . We found that primitive myelopoiesis is initially csf1r-independent , although csf1rDM embryonic macrophages subsequently ceased to divide and failed to colonize embryonic tissues . Surprisingly , a detailed examination of csf1rDM larval zebrafish revealed another wave of mpeg1+ cells in the skin from 15 days of development onwards , but these cells lacked the branched morphology typical of Langerhans cells ( He et al . , 2018 ) . Using fate mapping and gene expression profiling , we identified csf1rDM mpeg1+ cells as metaphocytes , a population of ectoderm-derived macrophage-like cells recently reported in zebrafish ( Alemany et al . , 2018; Lin et al . , 2019 ) . Extending our analyses , we further demonstrated that adult csf1rDM fish exhibit a global defect in macrophage generation . In conclusion , our study highlights distinct requirements for Csf1r during macrophage generation and metaphocyte ontogeny , resolving part of the presumed macrophage heterogeneity and their sensitivity to loss of Csf1r . To determine whether the earliest embryonic macrophages , called primitive macrophages , are still formed in the absence of Csf1r signaling , we analyzed csf1rDM zebrafish embryos carrying the macrophage transgenic reporter mpeg1:GFP ( Ellett et al . , 2011; Oosterhof et al . , 2018 ) . Zebrafish primitive macrophages are born in the rostral blood island on the yolk and can be detected by mpeg1:GFP expression from 22 hr post fertilization ( hpf ) as they migrate on the yolk ball—equivalent to the mammalian yolk sac—and progressively invade peripheral tissues ( Herbomel et al . , 1999; Herbomel et al . , 2001 ) . These constitute the main macrophage population during the first 5 days of development ( Wu et al . , 2018 ) . Indeed , in vivo imaging of GFP-expressing macrophages in control embryos showed that , at 24 hpf , ~13 mpeg1+ primitive macrophages were present on the yolk , increasing to ~49 cells at 42 hpf ( Figure 1A , Video 1; Ellett et al . , 2011 ) . In csf1rDM embryos , even though primitive macrophage numbers were slightly lower at 24 hpf ( ~5 mpeg1+ cells ) , macrophage numbers did not significantly differ from controls at 42 hpf ( ~46 mpeg1+ cells ) ( Figure 1A ) . This indicates that Csf1r is dispensable for the emergence of primitive macrophages . We next investigated whether embryonic macrophages in csf1rDM animals retained the ability to invade peripheral tissues . At 52 hpf , 50% of mpeg1+ cells had exited the yolk epithelium in controls and were observed in the periphery ( Figure 1B ) . In contrast , only 15% of all macrophages were found outside of the yolk in csf1rDM embryos . At this stage , macrophage numbers were significantly lower in csf1rDM larvae than controls ( Figure 1B ) . Migration trajectories of embryonic macrophages into the embryonic tissues , as shown by maximum intensity projections of images acquired over 16 hr , were more widespread in controls than csf1rDM and covered the entire embryo ( Figure 1C , Video 2 ) . Thus , although the generation of embryonic macrophages appeared independent of csf1r , after two days of development macrophage failed to expand in the csf1r mutants and their migration was reduced , suggesting functional deficits caused by the loss of Csf1r . We hypothesized that the reduced macrophage numbers in csf1r mutants could be explained by a reduction in their proliferative activity . To test this , we performed live imaging on mpeg1+ cells and quantified cell divisions . Between 32 and 48 hpf , the proliferative rates were not significantly different between control ( ~12 events ) and csf1rDM embryos ( ~10 events ) ( Figure 1D , Video 1 ) . However , whereas control macrophages actively proliferated between 56 and 72 hpf ( ~11% of macrophages divided ) , csf1rDM macrophages did not ( none of the macrophages divided ) ( Figure 1E ) . This indicates that the expansion of primitive macrophages is halted between 48 and 56 hpf . Thus , while the initial proliferation of emerging primitive macrophages occurs independent of csf1r , by 48 hpf Csf1r signaling becomes necessary for embryonic macrophage proliferation . To explore specific developmental and molecular processes affected by the loss of Csf1r signaling , and to discern a potential effect on proliferation , we performed RNA sequencing on macrophages isolated from 28 and 50 hpf mpeg1:GFP embryos using fluorescence-activated cell sorting ( FACS ) . These time points were chosen to study the primitive macrophages soon after their emergence from the RBI ( 28 hpf ) and as they subsequently transition to a tissue colonizing , migratory phenotype ( 50 hpf ) ( Figure 2A ) . Principal component analysis ( PCA ) of the macrophage gene expression data sets showed clustering of triplicate samples based on genotype ( component 1 ) and developmental stage ( component 2 ) ( Figure 2B ) . This suggests that , even though gene expression differed between control and csf1rDM macrophages at both time points , most of the changes that occurred over time in control embryos also occurred in csf1rDM embryos ( Figure 2B , C ) . To determine macrophage identity we analyzed the expression of genes highly expressed in macrophages , including genes used in zebrafish as macrophage markers ( e . g . csf1ra , mfap4 ) , chemokine and pathogen recognition receptors ( e . g . marco , mrc1 , tlr1 ) , and myeloid transcription factors ( e . g . irf8 , spi1a , cebpb ) , but we did not observe major differences between genotypes ( Figure 2D–E ) . Also , when we compared our gene expression profiles with a zebrafish macrophage expression profile determined by single cell RNA-seq ( Tang et al . , 2017 ) , only ~5% of the reported 2031 macrophage-specific genes were differentially expressed in csf1rDM macrophages , suggesting Csf1r-independent expression of the majority of these macrophage-expressed genes ( Figure 2F ) . Together , this shows that csf1r-deficient embryonic macrophages display a core gene expression profile similar to that seen in controls . The nature of the differences in gene expression profiles between control and csf1rDM macrophages was studied by gene set enrichment analysis ( GSEA ) . GSEA revealed that , at both time points , csf1rDM macrophages had lower expression of genes associated with RNA metabolism and DNA replication ( Figure 3A ) , with transcripts encoding all components of the DNA replication complex being ~2 fold reduced ( Figure 2—figure supplement 1 3B ) . In addition , csf1rDM macrophages showed lower expression of genes in GO classes related to cell cycle at 50 hpf ( Figure 3A , Figure 2—figure supplement 1 ) . Thus , at 28 hpf , DNA replication genes were downregulated , followed by a decrease in expression of genes involved in general cell cycle related processes at 50 hpf . Together , and in line with our in vivo findings , these analyses suggest that proliferation is reduced or halted in csf1rDM macrophages from 2 dpf onwards . Of the three Csf1r ligand genes , both csf1a and csf1b are expressed at 20 hpf , whereas il34 is not detectable at that time , barely detectable at 24 hpf , and moderately expressed at 36 hpf ( Figure 2—figure supplement 1C ) . Therefore , it is possible that the reduced expression of cell cycle related genes in csf1rDM macrophages could be attributed largely to a lack of interaction between the two Csf1 ligands and Csf1r . Additionally , this suggests that these two ligands likely do not influence the specification of embryonic macrophages at this stage . Previous analyses of macrophage development in il34-/- deficient zebrafish around 30 hpf showed primarily a deficiency in the migration of macrophages across the embryo and into the brain ( Kuil et al . , 2019; Wu et al . , 2018 ) . Microglia are the first TRM population present during embryonic development and they are highly proliferative during this time ( Ginhoux et al . , 2010; Herbomel et al . , 2001; Xu et al . , 2016 ) . Therefore , we determined whether loss of Csf1r signaling also affects microglial proliferation . Pcna/L-plastin double immunostaining in control embryos showed that total microglia numbers increase between 2 and 4 dpf . At 2 dpf almost no macrophages in the brain are proliferating , whereas ~20% of the population is at 4 dpf ( Figure 2—figure supplement 1B ) . In csf1rDM larvae a few microglia were occasionally present in the brain between 2 and 4 dpf , however none were Pcna+ ( Figure 2—figure supplement 1B ) . EdU pulse labeling experiments , marking cells that proliferated between 4 and 5 dpf , showed no EdU+ microglia in csf1r mutants , suggesting that csf1r-deficient microglia fail to proliferate ( Figure 3C ) . Thus , proliferation is impaired in both csf1rDM primitive macrophages and early microglia . Next , we assessed the presence of macrophages in developing csf1rDM animals by in vivo fluorescence imaging of one lateral side of entire , individual larvae on 4 consecutive days , starting at 5 dpf . We visualized ~450 macrophages in control animals , whereas csf1rDM animals contained >4 fold fewer ( ~100 ) ( Figure 3D ) . Over the next 4 days , macrophage numbers in both groups remained stable ( Figure 3D ) . This suggests that , at this stage , there is neither proliferative expansion of embryonic macrophages nor supply of macrophages from an alternative source , causing macrophage numbers in csf1rDM larvae to remain much lower than those in controls up to 9 dpf . Together these data indicate that , onwards from the initiation of embryonic tissue colonization , proliferative expansion of macrophages remains halted in csf1rDM animals . Given that macrophages are produced by consecutive waves of primitive and definitive myelopoiesis , and that embryonic csf1rDM macrophages ceased to proliferate , we wondered whether macrophages would be present at later developmental stages in csf1rDM zebrafish . By live imaging at 20 dpf we detected mpeg1+ cells in the skin of control animals , as expected , but also in the skin of csf1rDM animals ( Figure 4A ) . To pinpoint the emergence of these mpeg1+ cells we live imaged entire zebrafish unilaterally from 8 until 24 dpf ( Figure 4B ) . Between 10 and 13 dpf , control mpeg1+ cell numbers increased ~1 . 6 fold and csf1rDM mpeg1+ cell numbers increased 2 . 4 fold ( Figure 4B ) . From 15 to 17 dpf onwards , mpeg1+ cell numbers continued to increase exponentially both in controls and in csf1rDM fish . As we noticed differences in the size of the zebrafish , as they grew older , both among controls and mutants , we also plotted mpeg1+ cell numbers against fish size ( Figure 4B ) . Larval zebrafish rapidly grow in size , and their size often correlates better with developmental hallmarks than their age in days ( Parichy et al . , 2009 ) . In larval fish smaller than 5 mm , mpeg1+ cell numbers did not increase , whereas in fish that were larger than 5 mm mpeg1+ cell numbers correlated almost linearly with size . Taken together , we show that particularly in larvae older than 15 dpf , or over 5 mm in size , mpeg1+ cell numbers increase significantly , independent of csf1r mutation status . Despite the overall similar kinetics of mpeg1+ cell emergence , we observed major morphological differences in these cells between control and csf1rDM animals . In the skin of 22 dpf control zebrafish , we found two distinct cell morphologies: those presenting with a branched and mesenchymal cell shape reminiscent of mammalian Langerhans cells , the macrophage population in the epidermis , and those that display a compact , amoeboid morphology with short , thick , primary protrusions ( Figure 4C ) . In 22 dpf csf1rDM fish , only the more amoeboid cell type was present . These persisting amoeboid mpeg1+ cells in csf1r mutant animals could represent a subtype of macrophages , or skin metaphocytes , a newly identified macrophage-like cell type ( Alemany et al . , 2018; Lin et al . , 2019 ) . Metaphocytes are ectoderm-derived cells that display gene expression overlapping partly with macrophages , including mpeg1 , but with much lower expression of phagocytosis genes; these cells also lack a phagocytic response upon infection or injury ( Alemany et al . , 2018; Lin et al . , 2019 ) . As metaphocytes have also been reported to migrate faster than skin macrophages and morphologically resemble the mpeg1+ cells that remain in csf1rDM fish , we used time-lapse imaging and showed that , both in controls and in csf1rDM fish , the smaller , amoeboid mpeg1+ cells were highly motile ( Video 3; Lin et al . , 2019 ) . In contrast , the branched mpeg1+ cells that were found only in controls presented long , continuously extending and retracting protrusions and an evenly spaced distribution , but were largely confined to their location during 3 hr imaging periods . These highly branched macrophages , which were absent in csf1rDM fish , were located in the skin epidermis and , based on their location , morphology , migration speed , and behavior , may represent the zebrafish counterpart to mammalian Langerhans cells ( Video 3 Lugo-Villarino et al . , 2010 ) . In support of this notion , branched mpeg1+ cells were hardly detected in the skin of zebrafish deficient for interleukin-34 ( Figure 4—figure supplement 1A; 4C ) , the Csf1r ligand that selectively controls the development of Langerhans cells in mice ( Greter et al . , 2012; Wang et al . , 2012 ) . In larval zebrafish , csf1a and csf1b expression were detected in skin ( Figure 2—figure supplement 1 ) , more specifically in interstripe iridophores and hypodermal and fin cells ( Patterson and Parichy , 2013 ) . Although we found that il34 was also expressed in adult skin , this expression was about 10-fold lower than that of csf1a or csf1b ( Figure 2—figure supplement 1D ) . However , our in vivo imaging data suggests that the loss of Il34 , but not of both Csf1a and Csf1b , affects branched skin macrophages in particular ( Figure 4—figure supplement 1B ) . We reasoned that macrophages , and/or possibly Langerhans cells , could be absent in csf1rDM and il34 mutant skin , and that remaining mpeg1+ cells may be metaphocytes . Unlike macrophages , metaphocytes are of non-hematopoietic , likely ectodermal origin ( Lin et al . , 2019 ) . We recently proposed that skin macrophages and metaphocytes , based on these different ontogenies , could be discriminated in the adult zebrafish using the Tg ( kdrl:Cre; ßactin2:loxP-STOP-loxP-DsRed ) fate-mapping model that labels EMPs , HSCs and their progenies ( Bertrand et al . , 2010; Ferrero et al . , 2018 ) . Genetic , permanent labeling with DsRed of adult leukocytes , including branched skin macrophages is induced by constitutive expression of Cre recombinase in endothelial cells and hemogenic endothelium ( Bertrand et al . , 2010 ) . As suggested by restricted expression of the metaphocyte marker cldnh in mpeg1-GFP+DsRed- cells , non-hematopoietic metaphocytes lack DsRed labeling ( Ferrero et al . , 2020 ) . The presence or absence of DsRed expression could thus be used to discriminate between metaphocytes ( GFP+DsRed- ) and macrophages ( GFP+DsRed+ ) . Of note , a possible caveat is that mpeg1+ primitive macrophages , which derive directly from kdrl-negative mesoderm , are also not marked by DsRed in this setting , which could complicate the interpretation of results . However , as we previously documented , there seems to be no contribution from primitive hematopoiesis to mpeg1-expressing cells in the adult skin ( Ferrero et al . , 2020 ) . In addition , primitive macrophages appear virtually absent in Csf1r-deficient zebrafish , thus making this approach suitable to address the identity of mpeg1+ cells in csf1rDM skin . We generated csf1r-deficient animals carrying these three transgenes and examined their skin by confocal imaging . In control adult zebrafish skin , populations both of GFP+DsRed+ and of GFP+DsRed- cells were present , while only GFP+DsRed- cells could be detected in csf1rDM animals ( Figure 5A ) . This phenotype was further validated by flow cytometry analysis , showing a ~ 90% decrease in the GFP+DsRed+ population in csf1rDM zebrafish skin but no change in the frequency of GFP+DsRed- cells ( Figure 5B ) . Collectively , these results suggest that the generation of skin definitive macrophages is largely Csf1r-dependent and point to metaphocytes as the remaining mpeg1+ cells in csf1rDM skin . To further characterize cell identity , we FAC-sorted GFP+DsRed+ and GFP+DsRed- cells from control fish skin and GFP+DsRed- cells from csf1rDM skin and performed bulk RNA sequencing . PCA shows clustering of duplicates and segregation of GFP+DsRed- and GFP+DsRed+ ( PC1 ) and genotype ( PC2 ) ( Figure 5C ) . Consistent with their expected hematopoietic identity , GFP+DsRed+ cells expressed the pan-leukocyte marker ptprc ( Figure 5—figure supplement 2A ) . In contrast , GFP+DsRed- cells were negative for this marker . To address whether GFP+DsRed- cells overlap with metaphocytes , we selected genes expressed at higher levels in zebrafish metaphocytes than in macrophages , LCs and neutrophils ( Lin et al . , 2019 ) ( TPM logFC >2 ) , and analyzed their expression in our data . This revealed that GFP+DsRed- cells display a robust ‘metaphocyte’ gene signature ( e . g . cdh1 , epcam , cldnh , cd4-1 ) , regardless of their genotype ( Figure 5D–E ) . Additionally , many genes involved in phagocytosis and engulfment were downregulated in GFP+DsRed- cells ( e . g . mertka , havcr1 , stab1 , Figure 5F ) , as were genes that were previously shown to be expressed at lower levels in metaphocytes than in LCs and neutrophils ( e . g . itgb7 , cdk1 , cmklr1 , cebpb , Figure 5—figure supplement 2B ) . In line with the transcriptome similarities previously reported for metaphocytes and LCs , all cell populations in our analyses express mpeg1 as well as genes related to antigen presentation ( mhc2dab , cd74a , cd83 ) ( Figure 5—figure supplement 2A ) . Together , these findings validate the qualification of skin GFP+DsRed- cells as metaphocytes . Moreover , further analysis showed no major changes in the transcriptome of metaphocytes in the absence of csf1r , as only relatively few genes ( 359 out of 20 . 382 ) were found to differ significantly in expression between control and csf1rDM GFP+DsRed- cells ( Figure 5G ) . Unexpectedly , many of these genes are involved in pigment cell differentiation . Taken together with our imaging analyses ( Figures 4 and 5A ) , these data show that the skin of csf1rDM zebrafish lack mpeg1+ macrophages , but exclusively contain mpeg1+ metaphocytes , which are not reliant on Csf1r-signaling . We wondered whether the macrophage deficiency observed in the skin represents a general feature of csf1rDM fish . To address this question , we quantified total mpeg1+ cell numbers in 33 dpf and 1 . 5 months post fertilization ( mpf ) ( juvenile zebrafish: between 30–90 dpf ) control , csf1rDM and il34-/- fish by FACS ( Figure 5—figure supplement 1 ) . Fish deficient for il34 were included as an extra control , since they exhibit a selective loss of branched skin macrophages and contain lower embryonic microglia numbers , but retain other macrophage populations ( Figure 4C; Kuil et al . , 2019; Wu et al . , 2018 ) . Indeed , mpeg1+ cell numbers , with macrophage scatter properties , obtained from whole csf1rDM animals , were much lower than those in controls and il34 mutants ( Figure 4—figure supplement 1A-B ) . These findings are analogous to results reported for various organs of Csf1r-deficient mice and rats ( Dai et al . , 2002; Pridans et al . , 2018 ) . We next performed bulk RNA-sequencing on the total population of mpeg1+ cells isolated from controls , csf1rDM , and il34-/- ( Figure 6A ) . PCA showed clustering of triplicates and segregation based on genotype ( component 1: csf1rDM versus controls/il34-/- , component 2: il34 mutants versus controls ) ( Figure 6B ) . In addition , gene expression profiling identified transcriptional programs consistent with phagocytic macrophages in control and il34-/-mpeg1+ cells , but profiles consistent with metaphocytes only in csf1rDM cells ( Figure 6C–E ) . As overall il34-/- animals have a relatively small and selective macrophage depletion , we argue that this could have prevented the detection of a metaphocyte signature . Collectively , this suggests that csf1rDM fish specifically exhibit a profound deficiency in mononuclear phagocytes , whereas numerous remaining mpeg1+ cells appear to be metaphocytes rather than macrophages . We further tested this possibility by lineage-tracing and surveyed , through flow cytometry , the presence of GFP+DsRed+ macrophages and GFP+DsRed- metaphocytes among adult organs isolated from control and csf1rDM fish . As previously reported , in the zebrafish brain , primitive hematopoiesis-derived mpeg1+ microglia are completely replaced by HSC-derived mpeg1+ cells , and therefore all adult microglia , as well as CNS-associated macrophages are GFP+DsRed+ ( Ferrero et al . , 2018 ) . In addition , the lack of GFP+DsRed- cells in the adult brain indicates that metaphocytes are not present in the central nervous system ( Figure 6F ) . Brains of csf1rDM zebrafish were largely devoid of GFP+DsRed+ cells ( Figure 6F ) , in line with our previous studies ( Oosterhof et al . , 2019; Oosterhof et al . , 2018 ) . Similarly , livers from control and csf1rDM animals contained solely GFP+DsRed+ cells , which were virtually absent in csf1rDM animals ( Figure 6G ) . The intestine on the other hand contained both GFP+DsRed+ and GFP+DsRed- cells ( Figure 6H ) . However , these GFP+DsRed+ cells were lost and GFP+DsRed- cell numbers were increased in csf1rDM . As the presence of metaphocytes was reported in skin but also in the intestine ( Ferrero et al . , 2020; Lin et al . , 2019 ) , intestinal GFP+DsRed- cells are likely also csf1r-independent metaphocytes . In all , mpeg1+ macrophages are largely Csf1r-dependent , whereas mpeg1+ cells present in the skin and intestine are Csf1r-independent non-hematopoietic metaphocytes ( Figure 7 ) . Here , we showed that embryonic macrophages , develop , proliferate , and also initially acquire macrophage behavior and gene expression profile independently of Csf1r . However , without functional Csf1r , these cells subsequently fail to distribute across the embryo and cease to expand in numbers . This phenotype explains particularly the strong effect on microglial precursors , as these invade the brain and expand in numbers early in embryonic development and microglia are absent throughout life in zebrafish , mice , rats and humans deficient for CSF1R . Around 15 days of age , however , a strong increase in mpeg1+ macrophages in skin was detected by in vivo imaging in control but also in csf1rDM animals . Nevertheless , skin of both csf1rDM and mutants for the Csf1r ligand Il34 lacked the branched macrophages , which were present in controls , and only contained amoeboid mpeg1+ cells . Based on their non-hematopoietic origin and shared transcriptome profile , we identified these cells as metaphocytes . As metaphocytes share markers , morphology , and gross behavior with macrophages , they are easily mistaken for macrophages . We further showed that csf1rDM adults lacked virtually all blood-derived mpeg1+ mononuclear phagocytes , revealing the presence of mpeg1+ metaphocytes in the gut , as well as in the skin . Our data shows that in zebrafish Csf1r is critical for generation of both embryonic and adult macrophages , but is dispensable for the development of metaphocytes . Therefore , csf1r-deficient zebrafish are macrophage-less in most organs , and as they are viable , enable us to study the in vivo consequences of the absence of macrophages for developmental and homeostatic cellular processes . Two recent studies identified metaphocytes in zebrafish using distinct lineage tracing techniques , namely laser-mediated localized Cre-activation and CRISPR/Cas9 mediated genetic scarring followed by single cell DNA sequencing ( Alemany et al . , 2018; Levraud and Herbomel , 2019; Lin et al . , 2019 ) . Metaphocytes show reduced expression of engulfment genes , do not show a phagocytic response to injury or bacterial infection , have a rounded morphology and are highly motile ( Alemany et al . , 2018; Lin et al . , 2019 ) . Our transcriptome analysis showed high resemblance between metaphocytes and the remaining mpeg1+ cells in csf1rDM zebrafish ( total juvenile population and isolated from adult skin ) . Control and csf1rDM metaphocytes showed overall high similarity , but csf1rDM metaphocytes showed lower expression of genes involved in pigment cell differentiation . It is possible that this is an indirect consequence of the altered pigmentation status of csf1rDM deficient zebrafish , since they lack most of their xantophores , and lack stripes due to abnormal melanocyte patterning . As markers labeling macrophages will likely also label metaphocytes , this could perhaps explain the presumed incomplete depletion of macrophages in Csf1r mutant animals , or after CSF1R pharmacological inhibition ( Dai et al . , 2002; Erblich et al . , 2011; Pridans et al . , 2018 ) . Even though , particularly in vitro , CSF1R is considered essential for macrophage development , macrophages are nevertheless detected , in numbers ranging between 10–70% of the numbers found in controls , in tissues , other than brain , epidermis and bone , of Csf1r-deficient mice and rats ( Dai et al . , 2002; Pridans et al . , 2018 ) . Therefore , at least in zebrafish , macrophage numbers in Csf1r-deficient mutants were initially overestimated ( Oosterhof et al . , 2018 ) . As CSF1R mutations cause pleiotropic effects on various tissues in vertebrates and in human disease , that are likely caused by the absence of macrophages , our results further stress the importance of macrophages for development and homeostatic regulation of tissues . In addition , this raises the question whether metaphocytes exist in mammals ( Oosterhof et al . , 2019; Oosterhof et al . , 2018 ) . In mouse Csf1r knockouts embryonic macrophages were reported to be largely absent from the yolk sac at E12 . 5 ( Ginhoux et al . , 2010 ) . However , at E10 . 5 embryonic macrophages normally have already migrated away to the fetal liver and embryonic organs ( Stremmel et al . , 2018 ) . Therefore , it is unknown whether primitive macrophages would be present in Csf1r-deficient mice at a stage earlier than E10 . 5 and can be generated independently of Csf1r . In csf1rDMfish we found initially normal embryonic macrophage numbers , but at 2–2 . 5 dpf , concordant with E12-13 in mice , we also found reduced macrophage numbers compared to controls . It remains to be determined whether CSF1R signaling is essential for embryonic development in mice and other mammals at earlier stages as well . Homozygous mutations in CSF1R cause severe congenital brain disease with osteopetrosis , and absence of microglia ( Monies et al . , 2017; Oosterhof et al . , 2019 ) . Our data in zebrafish show multiple Csf1r-dependent steps of early microglia development that together illustrate how CSF1R-deficiency could underlie the absence of microglia already early in development . In zebrafish , only few Csf1r-deficient microglial progenitors reach the developing brain , since they stop to expand , and they are unable to respond to neuronal expressed Interleukin-34 , which normally facilitates brain colonization ( Greter et al . , 2012; Kuil et al . , 2018; Wang et al . , 2012; Wu et al . , 2018 ) . Thereafter , these few microglia do not expand , which eventually leads to their extinction . We propose that such a mechanism may underlie the absence of microglia , and osteoclasts , in patients with homozygous mutations in CSF1R ( Figure 7 ) . We find in il34-/- zebrafish that branched skin macrophages are lacking , but we did not find substantially lower numbers of macrophages or obvious gene expression changes overall , as in csf1rDM zebrafish . This phenotype is reminiscent of that of Il34 mutant mice that selectively lack microglia and Langerhans cells ( Greter et al . , 2012; Wang et al . , 2012 ) . Previous studies claimed skin mpeg1+ hematopoietic branched cells in zebrafish to be Langerhans cells ( He et al . , 2018; Lin et al . , 2019 ) . It remains unclear whether these are true Langerhans cells , as there is no known zebrafish ortholog of langerin ( CD207 ) , the main marker of LCs in humans and mice . LCs are likely to exist in zebrafish , as Birbeck granules , the morphological markers of LCs , have been identified in zebrafish skin macrophages ( Lugo-Villarino et al . , 2010 ) , and we recently demonstrated that zebrafish branched skin macrophages , develop independently of the transcription factor Irf8 ( Ferrero et al . , 2020 ) , similar to mammalian LCs ( Chopin et al . , 2013; Hambleton et al . , 2011 ) . Their dependence on Il34 provides additional evidence for the conservation of LCs in zebrafish . The effect of Il34 loss on macrophage development is relatively subtle , and overall gene expression of mpeg1+ cells in il34 mutants is likely to be dominated by gene expression from all Il34-independent macrophage populations and the effect of the loss of branched skin macrophages is therefore masked in the bulk RNA expression . TRMs retain the ability to proliferate , partly due to the relief of transcriptional suppression of proliferative enhancers by MAFB ( Soucie et al . , 2016 ) . Our findings suggest that Csf1r plays a central role in the maintenance of macrophage proliferative capacity . Our embryonic macrophage transcriptome analysis revealed two-fold lower expression of the majority of DNA replication genes in csf1rDM embryos , pointing towards a Csf1r-dependent induction of DNA replication , underlying the lack of macrophage proliferation . CSF1 can indeed rapidly stimulate S-phase entry and DNA replication of macrophages in vitro ( Tushinski and Stanley , 1985 ) . The Csf1r-independent proliferation of the earliest primitive macrophages on the yolk , could be explained by signaling through other members of the type III receptor tyrosine kinase family , including Csf3r , Flt3 , or C-kit , of which two in zebrafish have been shown to be involved in the expansion of primitive macrophages ( Flt3 ) or HSPCs ( Kitb ) ( Bartelmez and Stanley , 1985; He et al . , 2014; Mahony et al . , 2018; Sarrazin et al . , 2009; Williams et al . , 1992 ) . This could explain how the initial proliferation of progenitors is independent of Csf1r while later differentiation then becomes dependent . In sum , our work provides new insight into the dynamics of embryonic and adult macrophage development , but also metaphocyte ontogeny in zebrafish , as well as the developmental requirements for Csf1r therein . The csf1rDM zebrafish are highly suitable for studying the effects of macrophage absence systemically and metaphocyte function in isolation . In addition , we provide an approach to discern Csf1r-independent metaphocytes from Csf1r-dependent macrophages . Our findings here provide insight into the mechanism that could also underlie the absence of microglia in CSF1R-related leukodystrophy and could help predict the effects on other TRM populations in response to CSF1R mutations or pharmacological inhibition . Zebrafish deficient for both Csf1ra ( csf1raj4e1/j4e1 ) and Csf1rb ( csf1rbre01/ re01 ) , csf1rDM , were used as we described previously ( Oosterhof et al . , 2018 ) . The csf1raj4e1/j4e1 mutant was combined with a second csf1rb allele , csf1rbsa1503/sa1503 , affecting an essential splice site , leading to a premature STOP codon , for flow cytometry and lineage tracing experiments . Zebrafish deficient in Csf1a/Csf1b ( csf1are05/re05; csf1bre07/re07 ) or Il34 ( il34re03/re03 ) are described previously ( Kuil et al . , 2019 ) . Tg ( mpeg1:egfp ) ; Et ( shhb:KalTA4 , UAS-E1b:mCherry ) zf279 ) were used as control animals ( Ellett and Lieschke , 2010; van Ham et al . , 2014 ) . For the genetic lineage tracing the following transgenic lines were crossed: Tg ( kdrl:Cre ) s898 and Tg ( actb2:loxP-STOP-loxP-DsRedexpress ) sd5 ( Bertrand et al . , 2010 ) . All control animals used throughout the manuscript are wild-type controls carrying the trangene reporter constructs only . Adult and larval fish were kept on a 14h/10h light–dark cycle at 28°C . Larvae were kept in HEPES-buffered E3 medium . Media was refreshed daily and at 24 hpf 0 . 003% 1-phenyl 2-thiourea ( PTU ) was added to prevent pigmentation . Animal experiments were approved by the Animal Experimentation Committees of the Erasmus MC and ULB . Intravital imaging in zebrafish brains was largely performed as previously described ( van Ham et al . , 2014 ) . Briefly , zebrafish larvae were mounted in 1 . 8% low melting point agarose containing 0 . 016% MS-222 as sedative and anesthetic in HEPES-buffered E3 . The imaging dish containing the embedded larva was filled with HEPES-buffered E3 containing 0 . 016% MS-222 . For the experiment where larvae were followed over time between 5 and 9 dpf , larvae were removed from the low melting point agarose after imaging and put individually in wells of a 6 wells-plate containing HEPES-buffered E3 with PTU in which they were fed paramecia . For the experiment with larval fish between 8 and 24 dpf fish were kept in E3 medium until 5 dpf . From 5 dpf onwards , wild-type controls , il34 , and csf1r mutants were raised under standard conditions ( 14h/10h light–dark cycle , 28°C ) in the aquaria ( Tecniplast , Italy ) in the Erasmus MC fish facility and fed paramecia and dry food . From 13 dpf onwards they were also fed brine shrimp . Animals from all experimental groups were raised with the same number of fish per tank , in tanks of the same size throughout the experiment . Confocal imaging was performed using a Leica SP5 intravital imaging setup with a 20x/1 . 0 NA water-dipping lens . Imaging of mpeg1-GFP was performed using the 488 nm laser . Analysis of imaging data was performed using imageJ ( FIJI ) and LAS AF software ( Leica ) . Immunohistochemistry was performed as described ( van Ham et al . , 2014; van Ham et al . , 2012 ) . Briefly , larvae were fixed in 4 % PFA at 4°C overnight . Subsequently , they were dehydrated with an increasing methanol concentration methanol series , stored in 100% methanol at -20°C for at least 12 hours , and rehydrated , followed by incubation in 150 mM Tris-HCl ( pH=9 . 0 ) for 15 minutes at 70°C . Samples were then washed in PBS containing 0 . 04% Triton ( PBST ) and incubated in acetone for 20 minutes at -20°C . After washing in PBST and ddH2O , larvae were incubated for three hours in blocking buffer ( 10 % goat serum , 1 % Triton X-100 ( Tx100 ) , 1% BSA , 0 . 1 % Tween-20 in PBS ) at 4°C , followed by incubation in primary antibody buffer at 4°C for three days . Larvae were washed in 10 % goat serum 1 % Tx100 in PBS and PBS containing 1 % TX100 for a few hours , followed by incubation in secondary antibody buffer at 4°C for 2 . 5 days . Hereafter the secondary antibody was washed away using PBS . Primary antibody buffer: 1 % goat serum , 0 . 8 % Tx100 , 1 % BSA , 0 . 1 % Tween-20 in PBS . Secondary antibody buffer: 0 . 8 % goat serum , 1 % BSA and PBS containing Hoechst . Primary antibodies: PCNA ( 1:250 , Dako ) , L-plastin ( 1:500 , gift from Yi Feng , University of Edinburgh ) . Secondary antibodies used were DyLight Alexa 488 ( 1:250 ) and DyLight Alexa 647 ( 1:250 ) . Samples were imaged as described above . Scales were manually detached from anesthetized fish and pre-treated with 100mM DTT ( Invitrogen ) before O/N fixation in 4 % PFA . Immunostaining on floating scales was performed as described , using the following primary and secondary antibodies: chicken anti-GFP polyclonal antibody ( 1:500; Abcam ) , rabbit anti-DsRed polyclonal antibody ( 1:500; Clontech ) , Alexa Fluor 488-conjugated anti-chicken IgG antibody ( 1:500; Invitrogen ) , Alexa Fluor 594-conjugated anti-rabbit IgG ( 1:500; Abcam ) . Images were taken with a Zeiss LSM 780 inverted microscope , using a Plan Apochromat 20× objective . Image post-processing ( contrast and gamma adjust ) were performed with the Zeiss Zen Software . Larvae of 4 dpf were placed in a 24 wells plate in HEPES buffered ( pH = 7 . 3 ) E3 containing 0 . 003% PTU and 0 . 5 mM EdU for 24 hours . Next , larvae were fixed in 4% PFA for 3 hours at room temperature , dehydrated with a 25% , 50% , 75% , 100% MeOH series and stored at -20°C for at least 12 hours . Rehydrated in series followed by a proteinase K ( 10 µg/ml in PBS ) incubation for an hour . Followed by 15 minute post fixation in 4% PFA . Larvae were further permeabilized in 1% DMSO in PBS-T . Thereafter 50µl Click-iT ( Invitrogen ) reaction cocktail was added for 3 hours at room temperature protected from light . After washing steps larvae were subjected to immunolabelling using L-plastin ( see section immunofluorescent labelling ) . Samples were imaged as described above . The number of cells was manually quantified using ImageJ ( FIJI ) or Leica LASX software . To generate an overview of the gross migratory patterns maximum intensity projections of timelapse recordings were generated in FIJI . At 28 hpf , 35 larvae were collected in 0 . 16 % MS-222 solution to euthanize them before adding 5x Trypsin-EDTA ( 0 . 25% Trypsin , 0 . 1 % EDTA in PBS ) . For csf1rDM cells , at 50 hpf , 70 larvae were used as these mutants had fewer mpeg1-GFP positive cells . Micro centrifuge tubes containing zebrafish embryos were incubated on ice on a shaking platform to dissociate the cells . At 33 dpf and 1 . 5 mpf , single fish were euthanized in ice water , imaged to measure their length , and they were cut in small pieces with a razor blade and incubated in 5x Trypsin-EDTA on ice for 1 hour to dissociate . Next , the cell suspension was transferred to FACS tubes by running it over a 35 μm cell strainer cap . PBS containing 10 % fetal calf serum ( FCS ) was added over the strainer caps and the samples were centrifuged for 10 minutes 1000 rpm at 4°C . The pellet was taken up in 300 µl PBS-10% FCS containing DAPI ( 1:1000 ) . After analysis based on myeloid scatter , singlets , dapi and mpeg1-GFP signal cells were FAC-sorted by FACSAria IIIu and collected in Trizol , followed by RNA isolation according to the manufacturer’s instructions ( SMART-Seq v4 Ultra Low Input RNA Kit for Sequencing , Takara Bio USA ) ( Figure 5—figure supplement 1 ) . Single-cell suspensions of dissected adult zebrafish organs were prepared as previously described ( Wittamer et al . , 2011 ) . Flow cytometry and cell sorting were performed with a FACS ARIA II ( Becton Dickinson ) . For RNA-sequencing , mpeg1-GFP-positive cells from the skin were collected in Qiazol and RNA was extracted using the miRNeasy Micro Kit ( Qiagen ) . Analyses were performed using the FlowJo software ( Treestar ) . RNA sequencing cDNA was synthesized and amplified using SMART-seq V4 Ultra Low Input RNA kit for Sequencing ( Takara Bio USA , Inc ) following the manufacturer’s protocol . Amplified cDNA was further processed according to TruSeq Sample Preparation v . 2 Guide ( Illumina ) and paired end-sequenced ( 2×75 bp ) on the HiSeq 2500 ( Illumina ) . Experiment 1 , embryonic macrophages were sequenced at between 12 and 21 million reads per sample . Experiment 2 , juvenile macrophages , were sequenced at between 5 and 106 million reads per sample . Reads were mapped using Star v2 . 5 against the GRCz10 zebrafish genome ( Dobin et al . , 2013 ) . For differential gene expression analysis and GSEA we used the Bioconductor packages edgeR and GAGE , respectively ( Durinck et al . , 2009; Luo et al . , 2009; Robinson et al . , 2010 ) . For analyses on adult skin mpeg1+ cells , RNA quality was checked using a Bioanalyzer 2100 ( Agilent technologies ) . Indexed cDNA libraries were obtained using the Ovation Solo RNA-Seq System ( NuGen-TECAN ) with the SoLo Custom AnyDeplete Probe Mix ( Zebrafish probe set ) following manufacturer recommendation . The multiplexed libraries were loaded on a NovaSeq 6000 ( Illumina ) using a S2 flow cell and sequences were produced using a 200 Cycle Kit . On average 65 million paired-end reads were mapped against the Danio rerio reference genome GRCz11 . 94 using STAR software to generate read alignments for each sample . Annotations Danio_rerio . GRCz11 . 94 . gtf were obtained from ftp . Ensembl . org . After transcripts assembling , gene level counts were obtained using HTSeq . Genes differentially expressed were identified used the Bioconductor packages edgeR ( Durinck et al . , 2009; Robinson et al . , 2010 ) . Relative amount of each transcript was quantified via the ΔCt method , using MOB family member 4 ( mob4 ) or elongation-Factor-1-alpha ( ef1α ) expression for normalization , or via the ΔΔCt method , using mob4 or ef1α and WKM for normalization . Primers are listed in Table 1 . The number of biological replicates are listed in Table 2 . For statistical analysis GraphPad was used to perform Student’s t-tests , one-way ANOVA with Dunnett’s multiple comparison test , linear regression and non-linear regression analysis . Results were regarded significant at p < 0 . 05 .	Immune cells called macrophages are found in all organs in the body . These cells are highly effective at eating and digesting large particles including dead cells and debris , and microorganisms such as bacteria . Macrophages are also instrumental in shaping developing organs and repairing tissues during life . Macrophages were , until recently , thought to be constantly replenished from cells circulating in the bloodstream . However , it turns out that separate populations of macrophages become established in most tissues during embryonic development and are maintained throughout life without further input . Previous studies of zebrafish , rodents and humans have shown that , when a gene called CSF1R is non-functional , macrophages are absent from many organs including the brain . However , some tissue-specific macrophages still persist , and it was not clear why these cells do not rely on the CSF1R gene while others do . Kuil et al . set out to decipher the precise requirement for the CSF1R gene in macrophage development in living zebrafish . The experiments used zebrafish that make a green fluorescent protein in their macrophages . As these fish are transparent , this meant that Kuil et al . could observe the cells within the living fish and isolate them to determine which genes are switched on and off . This approach revealed that zebrafish with a mutated version of the CSF1R gene make macrophages as embryos but that these cells then fail to multiply and migrate into the developing organs . This results in fewer macrophages in the zebrafish’s tissues , and an absence of these cells in the brain . Kuil et al . went on to show that new macrophages did emerge in zebrafish that were about two to three weeks old . However , unexpectedly , these new cells were not regular macrophages . Instead , they were a new recently identified cell-type called metaphocytes , which share similarities with macrophages but have a completely different origin , move faster and do not eat particles . Zebrafish lacking the CSF1R gene thus lose nearly all their macrophages but retain metaphocytes . These macrophage-free mutant zebrafish constitute an unprecedented tool for further studies looking to discriminate the different roles of macrophages and metaphocytes .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "developmental", "biology" ]	2020	Zebrafish macrophage developmental arrest underlies depletion of microglia and reveals Csf1r-independent metaphocytes
Methods for analysing correlated mutations in proteins are becoming an increasingly powerful tool for predicting contacts within and between proteins . Nevertheless , limitations remain due to the requirement for large multiple sequence alignments ( MSA ) and the fact that , in general , only the relatively small number of top-ranking predictions are reliable . To date , methods for analysing correlated mutations have relied exclusively on amino acid MSAs as inputs . Here , we describe a new approach for analysing correlated mutations that is based on combined analysis of amino acid and codon MSAs . We show that a direct contact is more likely to be present when the correlation between the positions is strong at the amino acid level but weak at the codon level . The performance of different methods for analysing correlated mutations in predicting contacts is shown to be enhanced significantly when amino acid and codon data are combined . The effects of mutations that disrupt protein structure and/or function at one site are often suppressed by mutations that occur at other sites either in the same protein or in other proteins . Such compensatory mutations can occur at positions that are distant from each other in space , thus , reflecting long-range interactions in proteins ( Horovitz et al . , 1994; Lee et al . , 2008 ) . It has often been assumed , however , that most compensatory mutations occur at positions that are close in space , thus motivating the development of computational methods for identifying co-evolving positions as distance constraints in protein structure prediction ( Göbel et al . , 1994 ) . These methods , which rely on multiple sequence alignments ( MSA ) of homologous proteins as inputs , will become increasingly more useful in the coming years owing to the explosive growth in sequence data . The output of methods for correlated mutation analysis ( CMA ) is a rank order of the pairs of columns in the alignment according to the statistical and/or physical signficance attached to the correlation observed for each pair . The various methods for CMA that have been developed in the past 15 years differ in the measures they employ for attaching significance to the correlations ( Livesay et al . , 2012; de Juan et al . , 2013; Mao et al . , 2015 ) . Early measures include , for example , mutual information ( MI ) from information theory ( Gloor et al . , 2005 ) and observed-minus-expected-squared ( OMES ) in the chi-square test ( Kass and Horovitz , 2002 ) . Statistically significant correlations in MSAs that do not reflect interactions between residues in contact , that is , false positives , can stem from ( i ) various indirect physical interactions and ( ii ) common ancestry . The extent of false positives due to the latter source is manifested in the large number of correlations between positions in non-interacting proteins that can be observed when the sequences of non-interacting proteins from the same organism are concatanated and subjected to CMA ( Noivirt et al . , 2005 ) . Several methods for removing false positives owing to common ancestry were developed ( Pollock et al . , 1999; Wollenberg and Atchley , 2000; Noivirt et al . , 2005; Dunn et al . , 2008 ) but their success in contact prediction using CMA remained limited . False positives due to the former source , that is , indirect physical interactions , can occur when , for example , correlations corrersponding to positions i and j that are in contact and positions j and k that are in contact lead to a correlation for positions i and k that are not in contact . Methods that remove such transitive correlations have been developed in recent years and include , for example , Direct Coupling Analysis ( DCA or DI for Direct Information ) ( Weigt et al . , 2009; Morcos et al . , 2011 ) , Protein Sparse Inverse COVariance ( PSICOV ) ( Jones et al . , 2012 ) and Gremlin's pseudolikelihood method ( Kamisetty et al . , 2013 ) . These methods have been found to be very successful in identifying contacting residues ( Marks et al . , 2012 ) and they outperform earlier methods ( Mao et al . , 2015 ) . Nevertheless , their accuracy , which is ∼80% for the correlations in the top 0 . 1% ( ranked by their scores ) , drops to ∼50% for the top 1% ( Mao et al . , 2015 ) . Given that the number of contacts in a protein with N residues is ∼N ( Faure et al . , 2008 ) , it follows that for proteins with , for example , 100 residues ( i . e . with 4560 potential contacts between residues separated by at least 5 residues in the sequence ) only about 25% of the contacts ( i . e . 23 of the top 1% 46 predictions ) will be identified by these CMA methods . In addition , these methods require large MSAs comprising thousands of sequences in order to perform well and such sequence data are not always available . Consequently , it is clear that much can be gained from further improvements in methods of CMA . Here , we describe the development and application of a new method for CMA that uses both amino acid and codon MSAs as inputs instead of relying exclusively on amino acid MSAs as done before . We show that contact prediction is improved in a meaningful manner when amino acid and codon information are combined . The input for methods for analysing correlated mutations has exclusively been multiple amino acid sequence alignments . Here , we have shown that improved contact prediction can be achieved by analysing both amino acid and codon MSAs together . The premise of our approach is that direct contacts are more likely if the correlation at the amino acid level is high but at the codon level is low . The score we propose , which reflects this expectation , can be used in conjunction with different methods of CMA but other possible scores should be examined in future work . Importantly , we find cases where contacts between residues that are distant in sequence and , thus , of greatest value for structure prediction are predicted only by using the combined method . Future work should test other potential applications of combined analysis of amino acid and codon MSAs such as predicting protein–protein interactions and , more generally , in feature selection in machine learning . Protein sequence datasets were collected from Pfam version 27 . 0 ( Finn et al . , 2014 ) based on representative proteomes ( Chen et al . , 2011 ) at 75% co-membership threshold ( RP75 ) . Protein coding sequences ( CDS ) of the collected proteins from Pfam were retrieved based on Uniprot cross reference annotations ( for Refseq , Ensembl , EMBL and Ensembelgenomes databases in that order of priority ) using the EMBL-EBI's WSDbfetch services ( McWilliam et al . , 2009 ) and Ensembl REST API ( Beta version ) ( Yates et al . , 2015 ) . All collected CDSs were aligned in accordance to the Pfam HMM based MSAs using tranalign tool from the EMBOSS package ( Rice et al . , 2000 ) . Pfam domain families with more than 2000 successfully retrieved coding sequences were used for further analysis ( total of 551 MSAs ) . Only families with a known crystal structure at a resolution of 3 Å or better ( more than 95% of the families have at least three such structures ) and with an overlap of at least 80% of the domain sequence to the ATOM sequence in the solved structure were included in the analysis ( total of 460 MSAs ) . Our analysis was also restricted for proteins with more than 200 residues that have a large number of potential contacts for prediction ( 114 MSAs ) . PDB structures were assigned to Pfam families in accordance to the mapping in the files downloaded from http://www . rcsb . org/pdb/rest/hmmer ? file=hmmer_pdb_all . txt and ftp://ftp . ebi . ac . uk/pub/databases/msd/sifts/text/pdb_chain_uniprot . lst . PDB structures were retrieved and their coordinates were extracted using the bio3D R package ( Grant et al . , 2006 ) . Pairwise sequence alignments for mapping were performed using Biostrings ( Pages H . , Aboyoun P . , Gentleman R . and DebRoy S . Biostrings: String objects representing biological sequences , and matching algorithms . R package version 2 . 34 . 1 ) . The evaluation was based on the all structures with the highest resolution ( at least 3 Å ) but , in cases where families have more than 30 known structures with unique sequences , only 30 with the best resolution were used ( in cases of structures with the same resolution we arbitrarily chose one ) . The average accuracy of contact predictions for all the crystal structures of each domain family was then calculated so that domain families with many crystal structures would not be over-represented . Two definitions for a contact between two amino acids were employed: a distance of less than 8 Å between Cβ atoms and a distance of less than 8 Å between any two heavy atoms . Only pairs of residues that are separated by at least five amino acids in the protein sequence were considered . Accuracy was calculated as the proportion of true contacts from the N pairs with the highest score in that set . We evaluated the improvement of our method using the difference in the area under the curve ( AUC ) of the accuracy vs number of predicted pairs of our method relative to the results of the original OMES , MI , MIp , PSICOV and DCA methods . AUC was calculated using the auc function in MESS package in R with the default parameters . A non-redundant set of 2481 PDB entries with a percentage identity cutoff of 20% , resolution better than 1 . 6 Å and an R-factor cutoff of 0 . 25 was downloaded from the pre-compiled CullPDB lists ( Wang and Dunbrack , 2003 ) at http://dunbrack . fccc . edu/Guoli/pisces_download . php on February 25 , 2015 . Two residues were defined to be in a physical contact if they have at least one pair of atoms with a distance ≤3 . 5 Å . The number of true physical contacts , that is , those with a distance ≤3 . 5 Å between two of their respective heavy atoms , was determined for each protein in the set and divided by the number of residue pairs defined to be in contact if at least one inter-atomic distance between them is ≤8 Å ( designated ‘All’ ) or if the distance between their Cβ atoms is ≤8 Å . Only pairs of residues that are separated by at least five amino acids along the protein sequence were considered . The score for a pair of positions i and j , S ( i , j ) , for the OMES ( Observed Minus Expected Squared ) method was calculated , as follows ( Kass and Horovitz , 2002; Fodor and Aldrich , 2004 ) :SOMES ( i , j ) = ∑a∑b ( OBSaibj−EXPaibj ) 2where OBSaibj and EXPaibj are the respective observed and expected number of sequences in the MSA with residue type a at position i and residue type b at position j . The score for the mutual information , MI , method was calculated as follows ( Gloor et al . , 2005 ) :SMI ( i , j ) = ∑a=121∑b=121f ( i , a;j , b ) logf ( i , a;j , b ) f ( i , a ) f ( j , b ) where f ( i , a ) and f ( j , b ) denote the respective frequencies of occurrence of residue type a at position i and residue type b at position j and f ( i , a; j , b ) denotes the joint probability of occurrence of residue type a at position i and type b at position j . In the case of the MIp method ( Dunn et al . , 2008 ) , an average product correction ( APC ) term is subtracted from the MI score for each pair of positions . The APC term , which is a measure of the background MI shared by positions i and j , is given by:APC ( i , j ) = MI ( i , x¯ ) MI ( j , x¯ ) MI¯where terms in the nominator are the respective average MI values of positions i and j with all other positions in the alignment and the term in the denominator is the average background MI of all the positions in the alignment . The MIp score is given by: SMIp ( i , j ) = SMI ( i , j ) −APC ( i , j ) The Direct Coupling Analysis ( DCA ) method ( Morcos et al . , 2011 ) was implemented in R for amino acid and codon MSAs based on a Matlab source code provided by Weigt et al . ( http://dca . rice . edu/portal/dca/download ) . The PSICOV code was downloaded from http://bioinfadmin . cs . ucl . ac . uk/downloads/PSICOV/ and used for the predictions based on amino acid MSAs with the default parameters for faster options as recommended by the authors ( -p -r 0 . 001 and with the -l option in order to avoid using the APC term ) . The PSICOV code was modified in order to carry out the same analysis for codon MSAs and a python script was implemented to perform the whole analysis as done for the other methods using Pfam MSA files in Stockholm format and fasta MSA files as inputs . PSICOV was used here either with the APC for amino acid MSAs or without the APC for the predictions based on both amino acid and codon MSAs . The R and Python source codes for the contact prediction by all methods , C source code modifications to PSICOV V2 . 1b3 , R source code for structure-domain sequence mapping and python scripts for generating codon MSAs are available at https://etaijacob . github . io/ . Details on the relevant R packages that will be available on CRAN will also be provided at: https://etaijacob . github . io/ .	Genes contain instructions to make proteins from building blocks called amino acids . The instructions are encoded in units called codons that each specify a single amino acid in the chain . A small mutation in a particular codon can change the amino acid found at the corresponding position in the protein . Some amino acids interact with other amino acids in the chain , thereby enabling the protein to adopt the three-dimensional shape it needs to work properly . Therefore , a mutation that affects one of these amino acids may have a large impact on the ability of the protein to work . A mutation at one position in the protein may , however , have little effect if it is accompanied by a ‘compensatory’ mutation at another position . Such compensatory mutations are more likely to occur when the two positions in the protein are close to each other . To identify such mutations , the amino acid sequences of similar proteins from different organisms are aligned and compared . A computational method called ‘correlated mutation analysis’ searches for pairs of positions in the alignment that display co-variation , i . e . where particular mutations at one position tend to be accompanied by certain mutations at the second position . These pairs are then ranked according to the strength of their correlation and those with the highest ranking are predicted to be in close contact . Such predictions are , however , far from perfect and can give false results . Jacob et al . developed and tested a new technique of correlated mutation analysis by examining codon sequences as well as amino acid sequences . The rationale behind the technique relies on the fact that several different codons can encode the same amino acid , so that a mutation in a codon does not always change the amino acid it encodes . Therefore , a strong correlation at the amino acid level can be accompanied by a weak correlation at the codon level . In such cases the positions are more likely to be in contact than in cases where there is a strong correlation also at the codon level since the correlation can then be due to constraints at the DNA or RNA level . Jacob et al . tested their approach using different methods for analyzing correlated mutations that were proposed in previous studies . This showed that the predictions obtained using both amino acid and codon data are significantly more accurate than those obtained by comparing amino acid sequences only . Future work will test whether combining amino acid and codon data can also be used to predict interactions between different proteins .	[ "Abstract", "Introduction", "Results", "and", "discussion", "Materials", "and", "methods" ]	[ "structural", "biology", "and", "molecular", "biophysics", "computational", "and", "systems", "biology" ]	2015	Codon-level information improves predictions of inter-residue contacts in proteins by correlated mutation analysis
Human-based modelling and simulations are becoming ubiquitous in biomedical science due to their ability to augment experimental and clinical investigations . Cardiac electrophysiology is one of the most advanced areas , with cardiac modelling and simulation being considered for virtual testing of pharmacological therapies and medical devices . Current models present inconsistencies with experimental data , which limit further progress . In this study , we present the design , development , calibration and independent validation of a human-based ventricular model ( ToR-ORd ) for simulations of electrophysiology and excitation-contraction coupling , from ionic to whole-organ dynamics , including the electrocardiogram . Validation based on substantial multiscale simulations supports the credibility of the ToR-ORd model under healthy and key disease conditions , as well as drug blockade . In addition , the process uncovers new theoretical insights into the biophysical properties of the L-type calcium current , which are critical for sodium and calcium dynamics . These insights enable the reformulation of L-type calcium current , as well as replacement of the hERG current model . Human-based computer modelling and simulation are a fundamental asset of biomedical research . They augment experimental and clinical research through enabling detailed mechanistic and systematic investigations . Owing to a large body of research across biomedicine , their credibility has expanded beyond academia , with vigorous activity also in regulatory and industrial settings . Thus , human in silico clinical trials are now becoming a central paradigm , for example , in the development of medical therapies ( Pappalardo et al . , 2018 ) . They exploit mature human-based modelling and simulation technology to perform virtual testing of pharmacological therapies or devices . Human cardiac electrophysiology is one of the most advanced areas in physiological modelling and simulation . Current human models of cardiac electrophysiology include detailed information on the ionic processes underlying the action potential such as the sodium , potassium and calcium ionic currents , exchangers such as the Na/Ca exchanger and pumps such as the Na/K pump . They also include representation of the excitation-contraction coupling system in the sarcoplasmic reticulum , an important modulator of the calcium transient , through the calcium-induced calcium-release mechanisms and the SERCA pump . Several human models have been proposed for ventricular electrophysiology , and amongst them the ORd model ( O'Hara et al . , 2011 ) . Its key strengths are the representation of CaMKII signalling , capability to manifest arrhythmia precursors such as alternans and early afterdepolarisation , and good response to simulated drug block and disease remodelling ( Dutta et al . , 2016; Dutta et al . , 2017a; Passini et al . , 2016; Tomek et al . , 2017 ) . Consequently , ORd was selected by a panel of experts as the model best suited for regulatory purposes ( Dutta et al . , 2017a ) . Most of the ORd model development has focused on repolarisation properties such as its response to drug block , repolarisation abnormalities and its rate dependence . However , a more holistic comparison of ORd-based simulations with human ventricular experimental data reveals important inconsistencies . Firstly , the plateau of the action potential ( AP ) is significantly higher in the ORd model than in experimental data used for ORd model construction ( O'Hara et al . , 2011; Britton et al . , 2017 ) and in data from additional studies using human cardiomyocytes ( Coppini et al . , 2013; Jost et al . , 2013 ) . Secondly , the dynamics of accommodation of the AP duration ( APD ) to heart rate acceleration , which are known to be modulated by sodium dynamics , show only limited agreement with a comparable experimental dataset ( Franz et al . , 1988; O'Hara et al . , 2011 ) . Thirdly , we identify that simulations of the sodium current block has an inotropic effect in the ORd model , increasing the amplitude of the calcium transient , in disagreement with its established negatively inotropic effect in experimental/clinical data ( encainide , flecainide , and TTX ) ( Gottlieb et al . , 1990; Tucker et al . , 1982; Legrand et al . , 1983; Bhattacharyya and Vassalle , 1982 ) . All those properties , namely AP plateau potential , APD adaptation and response to sodium current block , have strong dependencies on sodium and calcium dynamics . We therefore hypothesise that ionic balances during repolarisation require further research . We specifically focus on an in-depth re-evaluation of the L-type calcium current ( ICaL ) formulation , given its fundamental role in determining the AP , the calcium transient and sodium homeostasis through the Na/Ca exchanger . The second main focus is the re-assessment of the rapid delayed rectifier current ( IKr ) , the dominant repolarisation current in human ventricle , under conditions that reflect experimental data-driven plateau potentials . Using a development strategy based on strictly separated model calibration and validation , we sought to design , develop , calibrate and validate a novel model of human ventricular electrophysiology and excitation contraction coupling , the ToR-ORd model ( for Tomek , Rodriguez – following ORd ) . Our aim for simulations using the ToR-ORd model is to be able to reproduce all key depolarisation , repolarisation and calcium dynamics properties in healthy ventricular cardiomyocytes , under drug block , and in key diseased conditions such as hyperkalemia ( central to acute myocardial ischemia ) , and hypertrophic cardiomyopathy . Table 1 lists the properties ( left column ) and key references ( right column ) of experimental and clinical datasets considered for the calibration ( top ) and independent validation ( bottom ) of the ToR-ORd model . This represents a comprehensive list of properties , known to characterize human ventricular electrophysiology under multiple stimulation rates , and also drug action and disease . The recordings in were obtained in human ventricular preparations primarily using measurements with microelectrode recordings , unipolar electrograms , and monophasic APs , therefore avoiding photon scattering effects or potential dye artefacts present in optical mapping experiments . In addition , the ToR-ORd model was calibrated to manifest depolarisation of resting membrane potential in response to an IK1 block , based on evidence in a range of studies summarised in Dhamoon and Jalife ( 2005 ) . The calibration criteria are chosen to be fundamental properties of ionic currents , action potential and single-cell pro-arrhythmic phenomena ( described in more detail in Appendix 1-1 ) . The validation criteria include response to rate changes , drug action and disease , to explore the predictive power of the model under clinically-relevant conditions . We initially performed the evaluation of the ORd model ( O'Hara et al . , 2011 ) by conducting simulations for each of the calibration criteria in Table 1 . Further details are described throughout the Materials and methods section and Appendix 1-15 . 1 . Simulations with the existing versions of the ORd model failed to fulfil key criteria such as AP morphology , calcium transient duration , several properties of the L-type calcium current , negative inotropic effect of sodium blockers , or the depolarising effect of IK1 block . The results are later demonstrated in Figures 2 and 3 , and Methods: Calibration of IK1 block and resting membrane potential . Secondly , we attempted parameter optimisation using a multiobjective genetic algorithm ( Torres et al . , 2012 ) . However , simulations with the ORd-based models were unable to fulfil key criteria such as AP and Ca morphology , and the effect of sodium and calcium block on calcium transient amplitude and APD , respectively . We then proceeded to reevaluate the ionic current formulations based on experimental data and biophysical knowledge . Key currents included ICaL and specifically its driving force and activation , as well as the INa , IKr , IK1 and chloride currents . The multiobjective genetic algorithm optimisation was repeated several times , throughout the introduction of structural changes to the model . Once simulations with an optimised model fulfilled all calibration criteria , validation was conducted through evaluation against additional experimental recordings for drug block , disease , tissue and whole-ventricular simulations . Details concerning the simulations are given in Appendix 1-15 . 1 , namely the description of simulation protocols and ionic concentrations used ( Appendix 1-15 . 1 . 1 ) , representation of heart disease ( Appendix 1-15 . 1 . 2 ) , 1D fibre simulations ( Appendix 1-15 . 1 . 3 ) , population-of-models and drug safety assessment ( Appendix 1-15 . 1 . 4 ) , transmurality and whole-heart simulations with ECG extraction ( Appendix 1-15 . 1 . 5 ) , and a technical note on the update to the Matlab ODE solver which facilitates efficient simulation of the multiobjective GA ( Appendix 1-15 . 1 . 6 ) . Unless specified otherwise , the baseline ORd model ( O'Hara et al . , 2011 ) was used for comparison with the ToR-ORd model . The ToR-ORd model follows the general ORd structure ( Figure 1A ) . The cardiomyocyte is subdivided into several compartments: main cytosolic space , junctional subspace , and the sarcoplasmic reticulum ( SR , further subdivided into junctional and network SR ) . Within these compartments are placed ionic currents and fluxes described by Hodgkin-Huxley equations or Markov models . The main ionic current formulations altered compared to ORd are highlighted in orange in Figure 1A . The ICaL current was deeply revisited , particularly with respect to its driving force , based on biophysical principles . This reformulation is of relevance to almost all models of cardiac electrophysiology . The ICaL formulation in the ORd model is based on Hodgkin-Huxley equations , with the total current being a product of three components: 1 ) Open channel permeability , 2 ) A set of gating variables determining the fraction of channels being open , 3 ) The electrochemical driving force which acts on ions to move through the open channel based on the membrane potential and ionic concentrations on both sides of the membrane ( more details in Appendix 1-5 ) . In most Hodgkin-Huxley models of cardia currents , the driving force is computed as ( V-Eion ) , that is , the membrane potential minus equilibrium potential , either computed from the Nernst equation , or measured experimentally . However , starting with the Luo-Rudy model ( LRd ) of 1994 ( Luo and Rudy , 1994 ) , the driving force of ions via ICaL in cardiac models is modelled based on the Goldman-Hodgkin-Katz ( GHK ) flux equation . The driving force based on the GHK equation is:φCaL=z2⋅ V⋅F2R⋅T⋅ [S]i⋅ez⋅V⋅FR⋅T−[S]oez⋅V⋅FR⋅T−1 , where z is the charge of the given ion , V is the membrane potential , F , R , T are conventional thermodynamic constants , and [S]i , [S]o are intracellular and extracellular activities of the given ionic specie . S=γ ∙m , where γ is the ionic activity coefficient and m the concentration ( in either the intracellular or extracellular space , yielding Si or So ) . The calibration of the ToR-ORd model’s AP morphology to experimental data resulted in problematic response to calcium blockade during an early phase of the model development when the original IKr formulation was used ( further details in Appendix 1-12 ) . ICaL block is known to shorten APD experimentally ( O'Hara et al . , 2011 ) but resulted in a major APD prolongation in simulations instead . This discrepancy could not be resolved through parameter optimisation . A mechanistic analysis revealed that this follows from the lack of ORd IKr activation , which is however not consistent with relevant experimental data ( Lu et al . , 2001 ) . We therefore considered alternative IKr formulations and specifically the Lu-Vandenberg ( Lu et al . , 2001 ) Markov model ( Figure 1B ) . The Lu-Vandenberg IKr model is based on extensive experimental data allowing the dissection of activation and recovery from inactivation and provided the best agreement with experimental data , specifically when considering the AP plateau potentials reported experimentally . In Appendix 1-12 , we: ( 1 ) provide a detailed explanation of origins of AP prolongation following ICaL block in a model which manifests experimental data-like plateau potentials and which contains the ORd IKr formulation; ( 2 ) explain why this phenomenon occurs only in a model with experimental data-like plateau potentials , but not in the original high-plateau ORd model; ( 3 ) compare the ORd and Lu-Vandenberg IKr formulations with experimental data , demonstrating the good agreement with experimental data of the Lu-Vandenberg formulation but not the ORd . Following the inclusion of the Lu-Vandenberg IKr formulation , all models generated during model calibration exhibited APD shortening in response to ICaL block . The INa current formulation was replaced by an alternative human-based formulation ( Grandi et al . , 2010 ) , given established limitations of the original model with regards to conduction velocity and excitability ( O'Hara et al . , 2011 ) , comment on article from 05 Oct 2012 ) . The Grandi INa model was updated to account for CaMKII phosphorylation ( Appendix 1-15 . 3 . 1 ) . Also from the Grandi model , we added the calcium-sensitive chloride current I ( Ca ) Cl and background chloride current IClb formulation ( Grandi et al . , 2010 ) . Neither model was changed compared to the original formulations , but the intracellular concentration of Cl- was slightly increased ( Appendix 1-15 . 1 . 1 ) . In accordance with recent observations , I ( Ca ) Cl was placed in the junctional subspace ( Magyar et al . , 2017 ) . The motivation to add these currents was to facilitate the shaping of post-peak AP morphology ( via I ( Ca ) Cl ) , with IClb playing a dual role stemming from its reversal potential of ca . −50 mV . It slightly reduces plateau potentials during the action potential , but during the diastole , it depolarises the cell slightly , improving the reaction to IK1 block as explained in the next subsection . The IK1 model was replaced with the human-based formulation by Carro et al . ( 2011 ) , as it was shown to be key for simulations of hyperkalemic conditions . The IK1 replacement was done before hyperkalemia simulation , not violating the classification of hyperkalemia criterion as a validation step . Extracellular potassium concentration in a healthy cell was reduced from 5 . 4 to 5 mM to fall within the physiological range ( Zacchia et al . , 2016 ) . When evaluating the baseline ORd model against the selected criteria , we observed that a reduction in IK1 results in hyperpolarisation of the cell ( from −88 to −88 . 16 mV at 1 Hz pacing ) . However , it is established that IK1 reduction depolarises cells experimentally ( Dhamoon and Jalife , 2005 ) . Changes made during ToR-ORd calibration ( predominantly the altered balance of currents during diastole and the inclusion of background chloride current ) result in ToR-ORd manifesting depolarization in response to IK1 block , consistent with experimental data . We applied a multiobjective genetic algorithm ( MGA , @gamultiobj function in Matlab , Deb , 2001 ) to automatically re-fit various model parameters . Based on preliminary experimentation , we used a two-dimensional fitness . We used MGA rather than an ordinary genetic algorithm or particle swarm optimisation , given that MGA optimises towards a Pareto front rather than a single optimum , implicitly maintaining population diversity . The Pareto front is the set of all creatures which are not dominated by any other creature in the population , that is creatures for which there is no other creature better in all fitness dimensions . Therefore , a subpopulation of diverse solutions is maintained , and the optimiser consequently has less of a tendency to converge to a single local optimum compared to single-number fitness approaches . In addition , the crossover operator of GA is well suited for a task where multiple criteria are optimised , given that creatures in the population may efficiently share partial solutions to various subcriteria . The fitness used in this study is described in greater detail in Appendix 1-1 . To facilitate the model validation and future work , we also provide an automated ‘single-click’ evaluation pipeline . It runs automatic simulations to extract and visualise single-cell biomarkers including those related to AP morphology , effect of key channel blockers , early afterdepolarisations ( EAD ) , and alternans measurement . The pipeline generates a single HTML report containing all the results; see Appendix 1-15 . 2 for a visualisation . The code for our model ( Matlab and CellML ) , the validation pipeline , and the experimental data on human AP morphology are available at https://github . com/jtmff/torord ( Tomek , 2019; copy archived at https://github . com/elifesciences-publications/torord ) . An informal blog giving further insight into the choices we made , as well as general thoughts on the development of ToR-ORd and computer models in general , is available at https://underlid . blogspot . com/ . We designed the Matlab code used to simulate our model so that the simulation core is structured into functions computing currents , making the high-level organisation of code clear , and facilitating inclusion of alternative current formulations . In addition , a CellML file encoding our model is also provided . This makes the model readily runnable in several simulators in addition to Matlab ( e . g . Chaste [Pitt-Francis et al . , 2009] and OpenCOR [Garny and Hunter , 2015] ) . Furthermore , the Myokit library ( Clerx et al . , 2016 ) enables conversion of the CellML file to other languages ( such as C or Python ) . The AP morphology of the ToR-ORd is within or at the border of the interquartile range of the Szeged-ORd experimental data ( Figure 2A ) . This is a major improvement compared to the original ORd morphology , which overestimates plateau potentials , particularly during early plateau ( Figure 2A ) . The fact that the early plateau potential is around 20–23 mV is clearly apparent from experimental recordings and is further corroborated by additional studies in human tissue samples ( Jost et al . , 2013 , Figure 6 ) and isolated human cardiomyocytes ( Coppini et al . , 2013 ) . We note that compared to the Szeged-ORd dataset ( Britton et al . , 2017 ) , our model manifests a slightly increased peak membrane potential in the single-cell form , similar to single-cell experimental data ( Coppini et al . , 2013 ) . This is a design choice related to the fact that the Szeged-ORd dataset contains recordings of small tissue samples , which are expected to manifest a reduced peak potential compared to single-cell . When coupled in a fibre , ToR-ORd manifests conduction velocity of 65 cm/s , which is consistent with clinical data ( Taggart et al . , 2000 ) . Both time to peak calcium and duration of calcium transient at 90% recovery obtained with the ToR-ORd model are within the standard deviation of experimental data in isolated human myocytes ( Coppini et al . , 2013 ) , whereas ORd slightly overestimated the calcium transient duration ( Figure 2B ) . The calcium transient amplitude of ToR-ORd also matches the Coppini et al . data after accounting for the different APD ( Appendix 1-8 ) . As described in Materials and methods , the ToR-ORd ICaL activation curve was extracted from experimental data , using the Goldman-Hodgkin-Katz formulation of ionic driving force , ensuring theoretical consistency , unlike the ORd ICaL formulation ( Figure 2C ) . This considerably improves the results of simulated protocols to obtain IV relationship ( Figure 2D ) , validating the theory-driven changes ( see Appendix 1-4 for the demonstration of how the updated activation curve underlies the improvement ) . The simulation of the protocol measuring steady-state inactivation also reveals improved agreement of ToR-ORd with experimental data compared to ORd ( Figure 2E ) . The difference between measured ORd steady state inactivation and the experimental data ( ca . two times stronger inactivation at around −15 mV , which is relevant for EAD formation ) is initially surprising , given that the equation of ORd ICaL steady-state inactivation curve provides a good fit to the same experimental data . This difference follows from the formulation of calcium-dependent inactivation of ICaL ( see Appendix 1-5 for details ) . We observed that in cases of elevated ICaL ( e . g . in midmyocardial cells ) , ORd reverses current direction towards positive values , which is an unexpected behaviour given its reversal potential of 60 mV . Conversely , the ToR-ORd model yields negative ICaL values in such conditions , consistent with it being an inward current ( Figure 2F ) . This is a direct consequence of the updates to the extracellular/intracellular calcium activity coefficients ( as explained in Appendix 1-6 ) , which supports their credibility and it is important for cases of elevated ICaL , such as under ß-adrenergic stimulation . We have also simulated a P2/P1 protocol as measured experimentally by Fülöp et al . ( 2004 ) , where two rectangular pulses are applied with varying interval between them . Both ORd and ToR-ORd qualitatively agree with the experimental data ( Appendix 1-7 ) . Figure 3A-D illustrates AP and calcium transient changes caused by block of sodium currents in ToR-ORd ( left ) and ORd ( right ) . As sodium blockers act on channel Nav1 . 5 mediating both the fast ( INa ) and late ( INaL ) sodium current ( Makielski , 2016 ) , we simulate the effect of combined partial INa and INaL block . The ToR-ORd model manifests a small reduction in calcium transient amplitude ( Figure 3C ) , unlike ORd , which gives a sizeable increase ( Figure 3D ) ; ToR-ORd is thus consistent with the observed negative inotropy of sodium blockers ( Gottlieb et al . , 1990; Tucker et al . , 1982; Legrand et al . , 1983; Bhattacharyya and Vassalle , 1982 ) . This is a major improvement in the ToR-ORd model , as sodium current reduction is involved in a range of disease conditions in addition to pharmacological block . Experimental evidence shows that the ratio of INa and INaL block is drug and dose-dependent , with INaL usually being blocked more than INa ( Appendix 1-9 ) . Figure 3E , F illustrates the change in calcium transient amplitude obtained with the ToR-ORd and ORd models , respectively , for several combinations of INa and INaL availability . Both models show a similar general trend where reduced INa availability increases calcium transient amplitude and reduced INaL availability diminishes it; however , the models differ strongly in relative contributions of these components . The ToR-ORd model shows negative inotropy for almost all combinations of blocks . A mild increase in inotropy may be achieved only under near-exclusive INa block . Conversely , ORd shows a general tendency for increased calcium transient amplitude; a reduction occurs only when the sodium current block targets near-exclusively INaL . ToR-ORd presents a greater calcium transient amplitude reduction than ORd in response to INaL block , as the current has a greater role in indirect modulation the cell’s calcium loading via APD change . At the same time , ToR-ORd shows a much smaller calcium transient amplitude increase in response to INa block than ORd because of the updated ICaL activation curve ( Figure 2C ) , as well as closer-to experimental data AP morphology ( Figure 2A ) and its effect on ICaL . A detailed explanation is given in Appendix 1-10 . Fibre simulations carried out to assess the effect of cell coupling on the effect of sodium block are consistent with the single-cell simulations ( Appendix 1-11 ) . The difference in response to half-block of INa and INaL between ToR-ORd and ORd is even larger , as ToR-ORd in fibre predicts a greater reduction in calcium transient amplitude than in single cell ( −14% vs −6% respectively ) , while ORd in fibre predicts a slightly greater increase in calcium transient amplitude than in single cell ( +25% vs +24% ) . With the ToR-ORd model , the 14% reduction in CaT amplitude in the electrically coupled fibre with 50% block of both INa and INaL is generally consistent with clinical data on sodium blockers: Encainide reduced stroke work index by 15% and cardiac index by 8% ( Tucker et al . , 1982 ) . In another study using encainide , the cardiac index was reduced by 18% and the stroke volume index by 28% ( Gottlieb et al . , 1990 ) . Flecainide reduced left ventricular stroke index by 12% and the left ventricular ejection fraction by 9% ( Legrand et al . , 1983 ) . Simulations with the ToR-ORd model show overall agreement with the clinical data . However , a direct quantitative comparison is challenging given the different indices of contractility measured ( CaT amplitude versus clinical indices ) and that it is not possible to estimate the exact ratios of INa and INaL block in clinical data ( Appendix 1-9 ) . EADs are an important precursor to arrhythmia , manifesting as a membrane potential depolarisation during late plateau and/or early repolarisation . They are thought to arise mainly from ICaL current reactivation ( Weiss et al . , 2010 ) . The ToR-ORd model manifests EADs at conditions used experimentally in nondiseased human endocardium ( Guo et al . , 2011; Figure 4A ) . The amplitude of simulated EADs is 14 mV ( Figure 4B ) , which matches the maximum EAD amplitude shown by Guo et al . ( 2011 ) . We also note that the experimental data by Guo et al . manifest early plateau potential of ca . 23 mV ( which is matched by ToR-ORd ) , in line with other studies we referred to previously regarding this matter . Repolarisation alternans is another established precursor to arrhythmia , facilitating the formation of conduction block ( Weiss et al . , 2006 ) . It is induced by rapid pacing and it is mostly thought to arise from calcium transient amplitude oscillations being translated to APD oscillations ( Pruvot et al . , 2004 ) , although purely voltage-driven mechanism was also proposed ( Nolasco and Dahlen , 1968 ) . Alternans in the ToR-ORd model is calcium-driven and appears via the same mechanism as in the ORd: sarcoplasmic reticulum calcium cycling refractoriness ( Tomek et al . , 2018 ) . It occurs at rapid pacing , in both calcium and APD ( Figure 4C–F ) . The peak APD alternans amplitude ( difference in APD between consecutive beats ) is 12 ms , which is matches the value 11 ± 2 ms reported in human hearts without a structural disease ( Koller et al . , 2005 ) . Direct quantitative comparison is however slightly limited by the fact that the data were recorded in RV septum , which may or may not differ from endocardial cells in alternans amplitude . Figure 5 illustrates simulations of drug action using the ToR-ORd model ( red traces ) , compared to experimental data ( black traces ) and to simulations with the ORd model reparametrised by Dutta et al . ( 2017a ) ( blue dashed lines ) . APD is shown in the presence of IKr block ( E-4031 , Figure 5A ) , IKs block ( HMR-1556 Figure 5B ) , multichannel block of INaL , ICaL , IKr ( mexiletine , Figure 5C ) , and a ICaL block ( nisoldipine , Figure 5D ) , at base cycle lengths of 500 , 1000 , and 2000 ms . We note that while the Dutta et al . model was specifically optimised for response of APD to these drug blocks , no such treatment was applied to the ToR-ORd model , making the results presented here an independent validation . Appendix 1-13 contains further details on the choice and use of the drug data . The predictions produced by the ToR-ORd model are in good agreement with experimental data , particularly given the lack of optimisation towards this result . Simulating E-4031 , ToR-ORd provides a prediction similar to the experimental data mean and the Dutta model ( Figure 5A ) . This is crucial , given the key role of IKr in the repolarisation reserve of human cardiomyocytes . The response to IKs blockade via HMR-1556 is even better in ToR-ORd than in the Dutta model , which is also within standard deviation of the data , but carries a clear trend towards AP prolongation ( Figure 5B ) . When simulating the multichannel blocker mexiletine , ToR-ORd prediction is within standard deviation of the experimental data , with the Dutta model giving similar or closer-to-mean predictions at 0 . 5 and 1 Hz ( Figure 5C ) . The predicted effect of the calcium blocker nisoldipine in the ToR-ORd model matches well the experimental data mean ( Figure 5D ) , even better than the Dutta model ( also within standard deviation ) . We note that the good performance of the simulated nisoldipine effect critically relies on the IKr replacement ( Materials and methods and Appendix 1-12 ) . Experimental measurements in human cardiomyocytes ( Franz et al . , 1988; Bueno-Orovio et al . , 2012 ) show how the APD shortens upon increase in pacing frequency , and then prolongs again , as the pacing frequency returns to control ( Figure 6A , top ) . APD adaptation dynamics with changes in heart rate are regulated by changes in sodium homeostasis ( Pueyo et al . , 2011 ) , and their manifestation in QT adaptation have been shown to be useful for arrhythmia risk prediction ( Pueyo et al . , 2004 ) . While simulations with the ORd model capture the general trend of APD accommodation , there are differences compared to the experimental data ( Figure 6A ) . First , changes in pacing rate are followed by slow-dynamics ( ~30 s ) APD prolongation not present in the experimental recordings . Second , the time constant of accommodation is generally slow . Conversely , the ToR-ORd model reproduces the pattern of accommodation well , where the change in APD soon after change in frequency is relatively fast , and then gradually slows down ( Figure 6B ) . This suggests that the ionic balance in ToR-ORd is likely to have been improved compared to ORd . A second indicator of how a model responds to a change in pacing frequency is the S1-S2 restitution protocol . The S1-S2 restitution curve obtained with the ToR-ORd model is given in Figure ( Figure 6C ) , showing a good agreement with the experimental data ( O'Hara et al . , 2011 ) . Drug safety testing is one of the key applications of computer modelling which has yielded highly promising results ( Passini et al . , 2017 ) . To assess the suitability of ToR-ORd for drug safety testing , we replicated the study by Passini et al . ( 2017 ) , which was carried out using populations of models based on the ORd model . Two populations were created based on ToR-ORd similarly to the original study , altering conductances of important currents within the ranges of 50–150% and 0–200% . Models in both populations are stable under significant perturbation of ionic conductances , which supports the robustness of the model ( Figure 7A ) . Prediction of the risk of drug-induced Torsades de Pointes based on simulated drug-induced repolarisation abnormalities using ToR-ORd population yielded similar results to the original study , with predicted risk being correct for 54 out of 62 compounds ( 87% accuracy ) . Compared to Passini et al . ( 2017 ) , the assessment of Mexiletine ( a predominantly sodium blocker that is safe ) was improved from false positive to true negative . High-dose Mexiletine led to formation of many EADs in ORd , but not in ToR-ORd ( Figure 7B ) , highlighting the importance of the advances on sodium blockers presented in this work . At the same time , Procainamide and Metrodinazole were misclassified as false negatives compared to Passini et al . ( 2017 ) . However , these drugs are controversial , as Metrodinazole is considered non-torsadogenic by Lancaster and Sobie ( 2016 ) , and this study predicted both the drugs to be non-risky . Torsadogenic risk for all evaluated compounds and the confusion matrix of the classification are given in Figure 7C . We conducted 3D electrophysiological simulations using the ToR-ORd model , representing the membrane kinetics of endocardial , epicardial and mid-myocardial cells to investigate their ability to simulate the ECG ( see Appendix 1-15 . 1 . 5 ) . Transmural and apex-to-base spatial heterogeneities as well as fibre orientations based on the Streeter rule were incorporated into a human ventricular anatomical model derived from cardiac magnetic resonance ( Lyon et al . , 2018 ) . Figure 9A shows the resulting electrocardiogram computed based on virtual electrodes positioned on a torso model shown in Figure 9B . The ECG manifests a QRS duration of 80 ms ( normal range 78 ± 8 ms ) , and a QT interval of 350 ms ( healthy:<430 ms ) ; all of these quantitative measurements are in the range of ECGs of healthy persons ( Engblom et al . , 2005; van Oosterom et al . , 2000 ) . ECG morphology also showed normal features , such as R wave progression in the precordial leads from V1 to V6 , isoelectric ST segment , and upright T waves in leads V2 to V6 , with inverted T wave in aVR . Figure 9C shows the activation sequence is in agreement with Durrer et al . ( 1970 ) . The APD map shows longer APD in the endocardium and the base , and shorter APDs in the epicardium and the apex , respectively ( Figure 9D ) . In this study , we present a new model of human ventricular electrophysiology and excitation contraction coupling , which is able to replicate key features of human ventricular depolarisation , repolarisation and calcium transient dynamics . The ToR-ORd model was developed using a defined set of calibration criteria and subsequently validated on features not considered during calibration to demonstrate its predictive power . This article also unravels several important theoretical findings with implications for computational electrophysiology reaching beyond the ToR-ORd model and cardiac electrophysiology: firstly , the reformulation of the L-type calcium current , which is broadly relevant and generally applicable to human and other species , and secondly , the mechanistically guided replacement of IKr . Discovering the necessity to carry out these theoretical reformulations was enabled by the comprehensive set of calibration criteria and the use of a genetic algorithm to fulfil them . Finally , to enable reproducibility , we openly provide an automated model evaluation pipeline , which provides a rapid assessment of a comprehensive set of calibration or validation criteria . The AP morphology of ToR-ORd is in agreement with the Szeged endocardial myocyte dataset used to construct the state-of-the art ORd model ( O'Hara et al . , 2011 ) . The agreement is considerably better than that of ORd itself , which has important implications for multiple aspects studied in this work . The calcium transient also recapitulates key features of human myocyte measurements ( Coppini et al . , 2013 ) . The validation of ToR-ORd shows that the model responds well to drug block with regards to APD ( Dutta et al . , 2017a ) . Good APD accommodation ( reaction to abrupt , but persisting changes in pacing frequency ) indicates a good balance between ionic currents ( Franz et al . , 1988; Pueyo et al . , 2010 ) . Replication of arrhythmia precursors such as early afterdepolarisations ( Guo et al . , 2011 ) and alternans ( Koller et al . , 2005 ) makes the model useful for simulations and understanding of arrhythmogenesis . This is particularly important in the context of heart disease , where ToR-ORd is shown to replicate key features of hyperkalemia ( Coronel et al . , 2012 ) and hypertrophic cardiomyopathy ( Coppini et al . , 2013 ) . The model is also shown to be promising in drug safety testing , and whole-heart simulations demonstrate physiological conduction velocity ( Taggart et al . , 2000 ) and produce a plausible ECG signal . Among the improved behaviours compared to the state-of-the-art ORd model ( O'Hara et al . , 2011 ) , the good response of the ToR-ORd model to sodium blockade is particularly noteworthy . ToR-ORd predicts the negative inotropic effect of sodium blockade , consistent with data ( Gottlieb et al . , 1990; Tucker et al . , 1982; Legrand et al . , 1983; Bhattacharyya and Vassalle , 1982 ) , unlike ORd , which suggests a strong pro-inotropic effect . The improvement in ToR-ORd follows from the relatively complex interplay of the theoretically driven reformulation of the L-type calcium current and data-driven changes to the AP morphology . This result is of great importance in the context of pharmacological sodium blockers , but it also plays a crucial role in disease modelling , where both fast ( Pu and Boyden , 1997 ) and late ( Coppini et al . , 2013 ) sodium current are altered . An important feature of a model is its predictive power , and validation of a model using data not employed in model calibration is a central aspect of model credibility ( Pathmanathan and Gray , 2018; Carusi et al . , 2012 ) . With this in mind , we designed our study to first calibrate the developed model using a set of given criteria , with subsequent validation of the model using separate data that were not optimised for during development . The fact that ToR-ORd manifests a wide range of behaviours consistent with experimental studies , even though it was not optimised for these purposes , suggests its generality and a large degree of credibility . To facilitate future model development , we also created an automated ‘single-click’ pipeline , which evaluates a wide range of calibration and validation criteria and creates a comprehensive HTML report . New follow-up models can thus be immediately tested against criteria presented here , making it clear which features of the model are improved and/or deteriorated by any changes made . The greatest theoretical contribution of this work is the theory-driven reformulation of the L-type calcium current , namely the ionic activity coefficients and activation curve extraction . Activation curve of the current in previous cardiac models was based on the use of Nernst driving force in experimental studies , but the models then used Goldman-Hodgkin-Katz driving force to compute the current . This yields a theoretical inconsistency present in existing influential models of guinea pig , rabbit , dog , or human , for example ( Luo and Rudy , 1994; Hund et al . , 2008; O'Hara et al . , 2011; Shannon et al . , 2004; Grandi et al . , 2010; Carro et al . , 2011 ) . We propose and demonstrate that in order to obtain consistent behaviour , the experimental I-V relationship measurements are to be normalised using the Goldman-Hodgkin-Katz driving force instead . Updated ionic activity coefficients and activation of the L-type calcium current improve key features of the current observed in the study underlying the ORd L-type calcium current model ( Magyar et al . , 2000 ) , and strongly contribute to the improved reaction of the model to sodium blockade . The changes made are relevant in development of future models which use the Goldman-Hodgkin-Katz equation for L-type calcium current or other currents . A second major contribution of this work reaching beyond the model itself is the set of observations on modelling of IKr , the dominant repolarising current in human ventricle . We noticed limitations of the ORd IKr model , which may be a result of the single-pulse voltage clamp protocol to characterise the current behaviour . Approaches enabling the dissection of activation and recovery from inactivation based on more comprehensive experimental data , such as Lu et al . ( 2001 ) used in our work , may yield a more general and plausible model . In this study , this change was important predominantly for the response of the ventricular cell to calcium block , but our observations are highly relevant also for models of cells with naturally low plateau , such as Purkinje fibres or atrial myocytes . We anticipate that the main future development of the presented model will focus on the ryanodine receptor and the respective release from sarcoplasmic reticulum . Similarly to most existing cardiac models , the equations governing the release depend directly on the L-type calcium current , rather than on the calcium concentration adjacent to the ryanodine receptors , which is the case in cardiomyocytes . Future development of the ryanodine receptor model and calcium handling will extend the applicability of the model to other calcium-driven modes of arrhythmogenesis , such as delayed afterdepolarisations . Also , while the model represents to a certain degree the locality of ICaL calcium influx and calcium release via the utilization of the junctional calcium subspace , a more direct representation of local control ( Stern , 1992; Hinch et al . , 2004 ) , realistic spatially distributed calcium handling ( Colman et al . , 2017 ) , or representation of stochasticity , may improve the insights the model can give into calcium-driven arrhythmogenesis . However , we note that such changes ( particularly the detailed distributed calcium handling ) will increase computational cost of the model's simulation . In addition , further research on the mechanisms regulating AP dependence on extracellular calcium concentration is needed to update this feature , not currently reproduced by most current human models ( Passini and Severi , 2014 ) .	Decades of intensive experimental and clinical research have revealed much about how the human heart works . Though incomplete , this knowledge has been used to construct computer models that represent the activity of this organ as a whole , and of its individual chambers ( the atria and ventricles ) , tissues and cells . Such models have been used to better understand life-threatening irregular heartbeats; they are also beginning to be used to guide decisions about the treatment of patients and the development of new drugs by the pharmaceutical industry . Yet existing computer models of the electrical activity of the human heart are sometimes inconsistent with experimental data . This problem led Tomek et al . to try to create a new model that was consistent with established biophysical knowledge and experimental data for a wide range of conditions including disease and drug action . Tomek et al . designed a strategy that explicitly separated the construction and validation of a model that could recreate the electrical activity of the ventricles in a human heart . This model was able to integrate and explain a wide range of properties of both healthy and diseased hearts , including their response to different drugs . The development of the model also uncovered and resolved theoretical inconsistencies that have been present in almost all models of the heart from the last 25 years . Tomek et al . hope that their new human heart model will enable more basic , translational and clinical research into a range of heart diseases and accelerate the development of new therapies .	[ "Abstract", "Introduction", "Materials", "and", "methods", "Results", "Discussion" ]	[ "cell", "biology", "computational", "and", "systems", "biology" ]	2019	Development, calibration, and validation of a novel human ventricular myocyte model in health, disease, and drug block
Cytosolic and nuclear iron-sulfur ( Fe-S ) proteins are involved in many essential pathways including translation and DNA maintenance . Their maturation requires the cytosolic Fe-S protein assembly ( CIA ) machinery . To identify new CIA proteins we employed systematic protein interaction approaches and discovered the essential proteins Yae1 and Lto1 as binding partners of the CIA targeting complex . Depletion of Yae1 or Lto1 results in defective Fe-S maturation of the ribosome-associated ABC protein Rli1 , but surprisingly no other tested targets . Yae1 and Lto1 facilitate Fe-S cluster assembly on Rli1 in a chain of binding events . Lto1 uses its conserved C-terminal tryptophan for binding the CIA targeting complex , the deca-GX3 motifs in both Yae1 and Lto1 facilitate their complex formation , and Yae1 recruits Rli1 . Human YAE1D1 and the cancer-related ORAOV1 can replace their yeast counterparts demonstrating evolutionary conservation . Collectively , the Yae1-Lto1 complex functions as a target-specific adaptor that recruits apo-Rli1 to the generic CIA machinery . Maturation of iron-sulfur ( Fe-S ) proteins in ( non-plant ) eukaryotes depends on the coordinated action of complex proteinaceous machineries in mitochondria and cytosol ( Lill , 2009; Stehling and Lill , 2013; Lukes and Basu , 2015; Maio and Rouault , 2015 ) . The mitochondrial iron-sulfur cluster ( ISC ) assembly machinery not only generates the organellar Fe-S proteins but is also essential for the biogenesis of cytosolic and nuclear Fe-S proteins ( Kispal et al . , 1999; Gerber et al . , 2004; Biederbick et al . , 2006 ) . The core part of the mitochondrial ISC machinery generates a sulfur-containing compound that is exported to the cytosol via the ABC transporter Atm1 ( Kispal et al . , 1999; Lill et al . , 2014 ) . Maturation of cytosolic and nuclear Fe-S proteins is further assisted by the cytosolic Fe-S protein assembly ( CIA ) machinery that is conserved in virtually all eukaryotes ( Sharma et al . , 2010; Netz et al . , 2014; Paul and Lill , 2015 ) . Cell biological and biochemical investigations have revealed nine CIA proteins , and an initial mechanistic model of their function has been generated mainly from work in the model organism Saccharomyces cerevisiae and in human cell culture . The first known step of this biosynthetic process involves the de novo assembly of a [4Fe-4S] cluster on the scaffold protein complex consisting of the two related P-loop NTPases Cfd1 and Nbp35 ( Roy et al . , 2003; Hausmann et al . , 2005; Netz et al . , 2007 , 2012a; Basu et al . , 2014 ) . Cluster assembly additionally depends on the electron transfer chain composed of NADPH , the diflavin reductase Tah18 and the Fe-S protein Dre2 ( Zhang et al . , 2008; Netz et al . , 2010 ) . The second step includes the release of the [4Fe-4S] cluster from the scaffold complex , cluster transfer and subsequent insertion into apoproteins . These reactions are assisted by the iron-only hydrogenase-like protein Nar1 ( human IOP1 ) which fulfils an intermediary , so far not well-defined function , by binding to both Nbp35 and the CIA targeting complex ( Balk et al . , 2004; Song and Lee , 2008 ) . The latter is composed of the WD40-repeat protein Cia1 , the DUF59 domain protein Cia2 and the HEAT-repeat protein Mms19 . All three proteins are conserved from yeast to man , and physically associate with multiple Fe-S target proteins suggesting a direct function in cluster transfer and/or insertion ( Balk et al . , 2005; Weerapana et al . , 2010; Gari et al . , 2012; Stehling et al . , 2012 , 2013 ) . Impairment of either the ISC or CIA components leads to a maturation defect of numerous Fe-S proteins with a function in genome maintenance including DNA polymerases and DNA helicases . In turn , ISC or CIA depletion causes severe defects in DNA metabolism including DNA synthesis and repair , chromosome segregation and telomere length regulation ( Veatch et al . , 2009; Thierbach et al . , 2010; Gari et al . , 2012; Stehling et al . , 2012 ) . As a result , the DNA damage response pathway is activated . Another interesting target of the CIA machinery is the essential ABC protein Rli1 ( RNase L inhibitor; human ABCE1 ) . It belongs to the most conserved proteins in Eukarya and Archaea ( Barthelme et al . , 2007; Becker et al . , 2012 ) , and harbors an N-terminal ferredoxin-like domain with eight conserved cysteine residues that coordinate two [4Fe-4S] clusters ( Karcher et al . , 2005; Kispal et al . , 2005; Barthelme et al . , 2007 ) . Depletion of Rli1 or disruption of Fe-S cofactor binding leads to nuclear export defects of both ribosomal subunits and consequently to translational arrest ( Kispal et al . , 2005; Yarunin et al . , 2005 ) . Moreover , Rli1 directly influences protein synthesis via different roles in translation initiation ( Dong et al . , 2004; Chen et al . , 2006 ) , termination ( Khoshnevis et al . , 2010; Shoemaker and Green , 2011 ) , and splitting of ribosomal subunits ( Barthelme et al . , 2011; Shoemaker and Green , 2011; Becker et al . , 2012 ) . Maintenance of Rli1 function was recently suggested to be the ‘Achilles' heel’ of aerobic organisms , because Fe-S cofactor supply to Rli1 is sensitive to diverse pro-oxidants ( Alhebshi et al . , 2012 ) . We hypothesized that additional CIA factors may exist . These may include target-specific biogenesis factors that , similar to the mitochondrial ISC proteins Ind1 , Nfu1 , and BOLA3 ( Stehling and Lill , 2013 ) , assist the maturation of selected apoproteins . Such dedicated CIA proteins are expected to bind to both the late-acting CIA components and specific target Fe-S proteins . Here , we performed systematic protein interaction screens in S . cerevisiae to identify new CIA interaction partners . We describe the essential proteins Yae1 and Lto1 as binding partners of the CIA targeting complex . Previously , both proteins were shown to be associated with each other and to interact with the ribosome-associated Fe-S protein Rli1 ( Krogan et al . , 2006; Zhai et al . , 2014 ) . While the role of Yae1 is unknown , Lto1 ( ‘required for biogenesis of the large ribosomal subunit and initiation of translation in oxygen’ ) was linked to the maturation of the 60S ribosomal subunit and translation initiation under aerobic conditions , yet its precise task remained unclear ( Zhai et al . , 2014 ) . In this work , we functionally characterized the yeast and human Yae1-Lto1 complexes as novel target-specific CIA maturation factors for Fe-S cluster assembly on Rli1 . Newly developed biochemical approaches and mutagenesis of important structural domains of these proteins elucidated a unique molecular mechanism of how these CIA proteins assist the Fe-S maturation process . To systematically investigate the interaction network of the CIA proteins and to identify new interaction partners of the CIA machinery , we performed tandem affinity purifications ( TAPs ) in yeast ( Rigaut et al . , 1999; Puig et al . , 2001 ) using the known CIA factors with a C-terminal TAP-tag as bait . Co-purified proteins were analyzed by mass spectrometry ( Figure 1—figure supplement 1A ) . The procedure yielded numerous interactions between the various CIA components ( Figure 1—figure supplement 1B , C ) . For instance , complex formation between Tah18 and Dre2 as well as between Cfd1 and Nbp35 were detected . Moreover , tight interactions were observed within the CIA targeting complex ( Cia1-Cia2-Mms19 ) . In keeping with functional observations ( Balk et al . , 2005; Hausmann et al . , 2005 ) , Nar1 was found to undergo interactions with both the early- and late-acting CIA components . Overall , this interaction pattern fully confirms previous findings on the CIA interactome , thus verifying our experimental strategy ( Netz et al . , 2014 ) . Three additional protein interactions caught our attention . First , all three components of the CIA targeting complex showed an interaction with the Fe-S protein Rli1 , a known target of the CIA machinery ( Figure 1—figure supplement 1B , C ) ( Balk et al . , 2005; Gari et al . , 2012; Stehling et al . , 2012 ) . Second , by using several interaction analysis criteria these CIA factors were also found to associate with the essential proteins Yae1 and Lto1 . Previous work had provided evidence that Yae1 and Lto1 interact with each other and form a complex with the essential ribosome-associated Fe-S protein Rli1 ( Krogan et al . , 2006; Zhai et al . , 2014 ) . To verify the CIA protein interactions with Yae1 and Lto1 , we expressed Myc- and HA-tagged versions of Yae1 and Lto1 in wild-type ( WT ) yeast for affinity purifications . Isolated proteins and their associated partners were analyzed by immunostaining . In addition to the tight interaction between Yae1 and Lto1 ( Figure 1A , lanes 1 and 2 ) , complex formation of these two proteins with any of the CIA targeting complex partners was observed when Lto1 and Yae1 were co-expressed . Affinity purification of the Yae1-Lto1 complex with the late-acting CIA proteins was significantly enhanced , when the analysis was performed in cells depleted of the early-acting CIA factor Nbp35 ( Figure 1A , lanes 6–9 ) . This indicated that the interactions of Yae1-Lto1 with CIA components are strengthened , when cytosolic Fe-S protein maturation is impaired . Nevertheless , the efficiency of Yae1-Lto1-Rli1 complex formation ( Zhai et al . , 2014 ) was not enhanced by Nbp35 depletion ( Figure 1B ) , demonstrating that inactivation of the CIA activity affects the association of Yae1-Lto1 with the CIA targeting complex more efficiently than with Rli1 . These findings verify our systematic interaction analyses and raise the question whether Yae1 and Lto1 are Fe-S protein targets or perform a function in cytosolic-nuclear Fe-S protein biogenesis . 10 . 7554/eLife . 08231 . 003Figure 1 . Yae1 and Lto1 interact with the cytosolic Fe-S protein assembly ( CIA ) targeting complex and the Fe-S protein Rli1 . ( A ) Wild-type ( WT ) and Gal-NBP35 yeast cells were co-transformed with 2µ vectors containing either no insert or genes encoding HA-Yae1 , HA-Lto1 , Yae1-Myc and Lto1-Myc as indicated . Cells were cultivated in minimal medium containing glucose leading to Nbp35 depletion ( ↓ ) in Gal-NBP35 cells . Cell lysates were prepared , and a HA-tag immunoprecipitation ( IP ) was performed . The immunoprecipitate was analyzed for the indicated proteins or tags by immunoblotting . ( B ) WT and Gal-NBP35 yeast cells were co-transformed with 2µ vectors containing either no insert or genes encoding HA-Yae1 , HA-Lto1 and Rli1-Myc and treated as in part A . The lower band in HA-Lto1-containing lanes is a HA-Lto1 degradation product . DOI: http://dx . doi . org/10 . 7554/eLife . 08231 . 00310 . 7554/eLife . 08231 . 004Figure 1—figure supplement 1 . Yae1 and Lto1 interact with the CIA targeting complex and with the Fe-S protein Rli1 . ( A ) Interactions between core components of the CIA machinery ( i . e . , Tah18 , Dre2 , Cfd1 , Nbp35 , Nar1 , Cia1 , Cia2 and Mms19 ) identified by TAP-MS ( tandem affinity purification ( TAP ) coupled to protein identification by mass spectrometry ) with additional STRING evidence . Baits used for affinity purifications are represented as either red ( CIA2 ) or olive diamonds , and preys are shown as grey diamonds . The interaction network around the core CIA components is extended by nodes imported from the STRING database with reported connections to baits and preys identified in our screen with a STRING confidence score >0 . 9 . Edges between all nodes and their attributes correspond to bait-prey interactions of different confidence levels , either by probabilistic scoring ( SAINT ) of TAP-MS data and/or from experimental data ( STRING ) . Thick green lines depict high confidence interactions observed by TAP-MS , which , in addition , are supported by experimental evidence in STRING . Thick black lines indicate robust connections , resistant to changes in parameters specified in the SAINT algorithm ( SAINT score >0 . 8 , and robustness >0 . 5 ) . Thin grey lines represent connections of lower confidence level , which are sensitive to changes in parameters for the SAINT algorithm ( >0 . 8 SAINT , robustness of <0 . 5 ) . Dashed green lines show interactions observed with a SAINT score threshold between 0 . 7 and 0 . 8 , and with additional experimental evidence . Dashed grey lines are interactions only reported in STRING with a STRING confidence score >0 . 9 . We summarized nodes in a semi-automatic manner , and performed a GO enrichment analysis on the resulting clusters . Clusters were defined by STRING nodes that only had one neighbor which had to be either a prey or bait from our screen . Three additional clusters were defined manually based on common neighborhood . ( B ) Detection of selected CIA protein interaction partners by identification of unique peptide numbers . TAP was performed with CIA-TAP fusion proteins as baits and associated partners identified via mass spectrometry . The amount of unique peptides referring to selected interaction partners of each CIA protein was determined ( n = 4 ) . As a control , we used the strain SC0000 , which does not express any TAP-fusion protein . ( C ) Interactions between the CIA proteins and some prominent potential partners were verified by dedicated co-IP experiments ( Co-IP ) . Verification procedures are represented by different colors . Green: Detection via TAP-MS ( based on identification of unique peptides , see also part B ) . Blue: Co-IP using C-terminally HA-tagged proteins as baits in WT background . Yellow: Co-IP using HA-tagged proteins as baits in a strain depleted for the early-acting CIA component Nbp35 . Orange: Interactions reported by a previous systematic analysis ( Krogan et al . , 2006 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 08231 . 004 Yae1 and Lto1 are conserved from yeast to man and show weak sequence similarity to each other in a central region of the protein ( Figure 2—figure supplement 1 ) . The proteins do not contain any conserved cysteine residues refuting the idea that they might function as recipient Fe-S proteins . To study the potential role of Yae1 and Lto1 in the maturation of cytosolic and nuclear Fe-S proteins , we employed an established in vivo 55Fe radiolabeling assay ( Kispal et al . , 1999 ) . To this end , we constructed regulatable GALL-promoter-containing yeast strains ( Gal-YAE1 , Gal-LTO1 ) which allowed expression of the proteins by cell growth in the presence of galactose and their depletion by growth in the presence of glucose ( Janke et al . , 2004 ) . In this context , we noted that the start codon of the LTO1 gene was erroneously annotated ( for further details of this aspect see Figure 2—figure supplements 2 , 3 , and ‘Materials and methods’ ) . After 55Fe radiolabeling of WT , Gal-YAE1 , Gal-LTO1 and control Gal-NBP35 cells , selected cytosolic and nuclear Fe-S proteins were immunoprecipitated . The protein-associated radioactivity was determined by scintillation counting as a measure of Fe-S cluster assembly . Depletion of Yae1 and Lto1 led to a strong diminution of 55Fe association with Rli1 , and this decrease was comparable to that observed upon depletion of Nbp35 ( Figure 2A ) . Despite structural similarities between Yae1 and Lto1 , these proteins could not mutually replace each other in Rli1 maturation , even after overexpression ( Figure 2B ) . Surprisingly , no significant decline in 55Fe-S cluster formation was detectable for three canonical nuclear Fe-S target proteins ( Rad3 , Ntg2 , and Pol3 ) upon Yae1 or Lto1 depletion ( Figure 2C–E ) . As expected , these proteins were not assembled with a 55Fe-S cluster in cells depleted for the CIA component Nbp35 . Immunostaining showed that the levels of Rli1 and the other Fe-S proteins remained unchanged upon depletion of Yae1 , Lto1 and Nbp35 , with the notable exception of Pol3 in Nbp35-depleted cells due to the known instability of its apoform ( Figure 2F ) ( Netz et al . , 2012b ) . In conclusion , our in vivo 55Fe labeling analysis suggests that Yae1 and Lto1 are specifically required for Fe-S cluster association with Rli1 , yet these factors are dispensable for assembly of other Fe-S proteins . 10 . 7554/eLife . 08231 . 005Figure 2 . Yae1 and Lto1 specifically mediate Fe-S cluster association with Rli1 . WT , Gal-YAE1 , Gal-LTO1 and Gal-NBP35 yeast cells were transformed with plasmids encoding Rli1-HA ( A , B ) , Rad3-HA ( C ) , Ntg2-HA ( D ) and the C-terminal domain of Pol3 N-terminally fused to HA ( E ) . To increase the amount of Pol3-CTD-bound Fe-S clusters the accessory subunit Pol31 was co-expressed ( Netz et al . , 2012b ) . In part B Gal-YAE1 and Gal-LTO1 cells contained 2µ plasmid-borne YAE1 or LTO1 or the empty vector . Cells were cultivated for 24 hr in glucose-containing minimal medium for depletion of the respective proteins ( ↓ ) . After an additional 16 hr in minimal medium lacking iron cells were radiolabeled for 2 hr with 55FeCl3 , cell extracts were prepared , and the Fe-S target proteins were immunoprecipitated with anti-HA antibodies . The amount of 55Fe associated with target proteins was quantified by scintillation counting . Error bars indicate the SEM ( n > 4 ) . ( F ) A representative immunostain of the indicated proteins from parts A , C–E . DOI: http://dx . doi . org/10 . 7554/eLife . 08231 . 00510 . 7554/eLife . 08231 . 006Figure 2—figure supplement 1 . Yae1 and Lto1 contain a deca-GX3 sequence motif that is conserved in eukaryotes . ( A ) The multi-sequence alignment of Yae1 and Lto1 from various eukaryotic species ( Sc , Saccharomyces cerevisiae; Ca , Candida glabrata; Kl , Kluyveromyces lactis; Hs , Homo sapiens; Mm , Mus musculus ) was calculated by Multalin ( Corpet , 1988 ) . The consensus sequence is shown below the alignment . The ScLto1 sequence highlighted in cyan corresponds to the erroneously annotated N-terminus of this protein ( see also Figure 2—figure supplements 2 , 3 ) . ( B ) A multi-sequence alignment of the protein family pfam09811 was conducted and a weblogo generated to highlight the conserved deca-GX3 domain ( weblogo . berkeley . edu ) . The bottom part shows a cartoon of Yae1 and Lto1 from S . cerevisiae and man ( YAE1D1 and ORAOV1 ) to highlight the relative positions of the deca-GX3 motif . The number of amino acids ( aa ) of each protein is depicted at the C-terminus . DOI: http://dx . doi . org/10 . 7554/eLife . 08231 . 00610 . 7554/eLife . 08231 . 007Figure 2—figure supplement 2 . Generation of a regulatable strain for depletion of Lto1 questions the previously predicted physiological translation start site of LTO1 . During the generation of the galactose-regulatable strain of LTO1 we noted that the previously predicted translation start site may be incorrect . ( A ) Cartoon of the 5′ region of the LTO1 gene . For functional analyses of Lto1 and Yae1 the GALL cassette was inserted directly in front of the start codons of YAE1 and LTO1 . In the latter case the GALL promoter was inserted at different sites . First , at the original translation start codon ( ATG1 ) of Saccharomyces Genome Database ( strain Gal-LTO1long ) , or second , at the corrected translation start site ( ATG3 ) located 108 bp downstream of the annotated start as determined in this work ( strain Gal-LTO1 ) . LTO1P: promoter of LTO1 . ( B ) Cells from part A were grown over night in glucose-containing rich medium . A serial dilution ( 1:5 ) was spotted onto rich medium agar plates supplemented with glucose ( Glc ) or galactose ( Gal ) . DOI: http://dx . doi . org/10 . 7554/eLife . 08231 . 00710 . 7554/eLife . 08231 . 008Figure 2—figure supplement 3 . Determination of the correct physiological translation start site of LTO1 . ( A ) Ribosome footprinting analysis of LTO1 . The physiological start site of LTO1 was determined by bioinformatic analysis using the published data from ribosome footprinting ( Ingolia et al . , 2009 ) . Ribosome footprinting and mRNA data were obtained from GEO ( Gene Expression Omnibus , #GSE13750 ) and imported into Microsoft Access . Reads corresponding to chromosome XIV and coordinates corresponding to the LTO1-coding region were selected and transferred to Excel . The sequence reads were binned in a 30 bp window . The vertical dashed line represents the translation start site as predicted by SGD ( ATG1 ) , and the vertical dotted line is the start site defined in this work ( ATG3 ) . ATX1 and ORC5 are neighboring reference genes . ( B ) The region in front of codon ATG3 of LTO1 is not essential for cell growth . Gal-LTO1 cells were transformed with an empty centromeric plasmid or plasmids containing various LTO1 genes under control of the natural promoter . As indicated on the left part the ATG codons 1 and 2 of LTO1 were mutated to stop codons and analyzed for their relevance to support cell growth . Cells were cultivated 16 hr in glucose-containing minimal medium to deplete endogenous Lto1 and spotted onto plates supplemented with glucose ( Glc ) or galactose ( Gal ) ( serial 1:5 dilution , start OD600 = 0 . 5 ) . ( C ) The region in front of codon ATG3 of LTO1 is critical for gene expression . LTO1 promoter ( LTO1P ) fragments were fused to the luciferase ( LUX ) reporter gene . WT cells harboring the luciferase-based reporter constructs were cultivated in glucose-containing minimal medium . Promoter activities were measured in exponentially growing cells and normalized to protein amounts . Error bars indicate the SEM ( n > 4 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 08231 . 008 To further investigate the intriguing target specificity of Yae1 and Lto1 in cytosolic-nuclear Fe-S protein assembly , we analyzed the effects of Yae1 or Lto1 deficiency on the activities of the cytosolic Fe-S enzymes isopropylmalate isomerase Leu1 and sulfite reductase . While depletion of Nbp35 strongly diminished the Leu1 activity , depletion of Yae1 or Lto1 had hardly any effect ( Figure 3A ) . Likewise , Yae1 and Lto1 were not required for sulfite reductase activity as estimated by an in vivo color assay ( Figure 3B ) . Further , we analyzed Fe-S cluster assembly on the early-acting CIA proteins Nbp35 and Nar1 in order to explore whether Yae1 or Lto1 deficiency affects the function of the CIA system ( Figure 3C–E ) . As estimated by 55Fe radiolabeling , neither Yae1 nor Lto1 were involved in the maturation of these two Fe-S cluster-containing CIA components . In contrast , depletion of the mitochondrial ISC component Yah1 strongly diminished Fe-S cluster assembly on Nbp35 and Nar1 . These findings indicate that Yae1-Lto1 operate downstream of these CIA factors in Rli1 Fe-S cluster association . Finally , activities of the non-Fe-S protein alcohol dehydrogenase and the mitochondrial Fe-S protein aconitase were unchanged upon depletion of both Yae1 and Lto1 ( Figure 3—figure supplement 1 ) . Taken together , these results , combined with the tight association of Yae1 and Lto1 to the CIA targeting complex , suggest that both proteins are specific for Fe-S cluster formation on Rli1 , and thus act late in the CIA pathway . A similar Fe-S protein target specificity is unprecedented in yeast . 10 . 7554/eLife . 08231 . 009Figure 3 . Yae1 and Lto1 do not perform a general role in Fe-S protein maturation . ( A ) WT , Gal-YAE1 , Gal-LTO1 and Gal-NBP35 cells were cultivated on glucose-containing minimal medium for 40 hr . Cell extracts were analyzed for the activities of isopropylmalate isomerase ( Leu1 ) . ( B ) Cells were analyzed for sulfite reductase activity in vivo after growth for 3 days on minimal medium agar plates supplemented with ammonium bismuth citrate and sodium sulfite . Sulfide produced by sulfite reductase yields the brown precipitate Bi2S3 . ( C , D ) WT , Gal-YAE1 , Gal-LTO1 and Gal-YAH1 cells were transformed with plasmids containing genes encoding HA-Nbp35 ( C ) and Nar1-HA ( D ) . Cells were radiolabeled with 55FeCl3 , and 55Fe-S protein formation was estimated as in Figure 2 . ( E ) A representative immunostain of the indicated proteins from parts C , D . ( F ) WT , Gal-YAE1 and Gal-LTO1 cells were cultivated in rich medium containing galactose ( Gal ) or glucose ( Glc ) for 40 hr , and radiolabelled with 35S-methionine for 10 min at 30°C . Extracts were prepared , and the protein synthesis efficiency was analyzed by SDS-PAGE and autoradiography . Error bars indicate the SEM ( n > 4 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 08231 . 00910 . 7554/eLife . 08231 . 010Figure 3—figure supplement 1 . Depletion of Yae1 , Lto1 and Nbp35 does not affect enzyme activities of the cytosolic alcohol dehydrogenase and the mitochondrial Fe-S enzyme aconitase . WT , Gal-YAE1 , Gal-LTO1 and Gal-NBP35 cells were cultivated in glucose-containing minimal medium to deplete the respective proteins ( ↓ ) . Cell extracts were prepared using glass beads , and the activities of alcohol dehydrogenase ( A ) and the mitochondrial aconitase ( B ) were measured . Activities were normalized to protein amounts . Error bars indicate the SEM ( n > 4 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 08231 . 010 Rli1 and its essential Fe-S cofactors are involved in the export of ribosomal subunits from the nucleus and are required for various aspects of protein translation ( Hopfner , 2012 ) . To address the influence of Yae1 and Lto1 on cytosolic protein synthesis , yeast cells containing or lacking Yae1 and Lto1 were briefly radiolabelled with 35S-methionine . Cell extracts were prepared and analyzed for incorporation of the radiolabeled amino acid into nascent proteins . Depletion of Yae1 or Lto1 impaired protein translation in a similar fashion as the deficiency in Rli1 ( Figure 3F; cf . [Kispal et al . , 2005] ) . We conclude that the influence of Yae1-Lto1 on protein biosynthesis is explained by their role in Fe-S cluster formation on Rli1 . The specific requirement of Yae1 and Lto1 for the presence of Fe-S clusters on Rli1 suggested that these either act as specific CIA maturation factors or stabilize the assembled Fe-S clusters on Rli1 . Interestingly , assembly of Rli1 was previously shown to be sensitive to oxidative stress-inducing agents such as paraquat , which may destroy the ROS-labile Fe-S cofactors of Rli1 ( Alhebshi et al . , 2012 ) . Further , Yae1 and Lto1 were found to be essential under aerobic but not under anaerobic growth conditions ( Snoek and Steensma , 2006 ) . These findings may indicate a function of these proteins in oxidative stress protection of the Rli1 Fe-S clusters . To discriminate between a maturation or stabilization function of Yae1 or Lto1 for Rli1's Fe-S clusters , we established a method which allowed us to rapidly degrade Yae1 or Lto1 and then follow the fate of Rli1's 55Fe-S clusters . Targeted degradation of Yae1 and Lto1 was achieved by fusing a photosensitive degron ( psd ) to their C-termini . This degron is composed of the light-reactive LOV2 domain of Arabidopsis thaliana phot1 and the murine ornithine decarboxylase-like degradation sequence cODC1 ( Renicke et al . , 2013 ) . Exposure to blue light triggers proteasomal degradation of the psd-fusion proteins ( Figure 4A ) . Induced degradation of Yae1-psd and Lto1-psd was maximal after 2 hr of blue light exposure ( Figure 4—figure supplement 1A ) . The extent of Yae1 and Lto1 depletion affected the growth rate of the mutant cells only weakly ( Figure 4—figure supplement 1B ) . To further increase the depletion efficiency , cells were treated with the translation inhibitor cycloheximide during light exposure to fully inhibit synthesis of Yae1 protein ( Renicke et al . , 2013 ) . This trick further decreased the amount of Yae1 ( compare immunoblots of Figure 4B and Figure 4—figure supplement 1A ) . We then investigated the effect of light-induced Yae1 or Lto1 depletion on the de novo assembly of Rli1's Fe-S clusters by employing our 55Fe radiolabeling procedure . Light-induced depletion of Yae1 caused a strong reduction of 55Fe-S cluster insertion into Rli1 in comparison to the non-exposed sample ( Figure 4B ) . In contrast , the amount of 55Fe associated with Leu1 was unchanged upon exposure to blue light . The latter result showed that cycloheximide had no general negative effect on 55Fe-S cluster insertion into apoproteins . The Lto1-psd mutant strain could not be used in this approach because Rli1 incorporated only little 55Fe suggesting that the C-terminal psd module interfered with normal function of Lto1 . Collectively , these results demonstrate that the light-induced degradation of Yae1 leads to a similar diminution of Fe-S clusters on Rli1 as that observed for Gal-YAE1 cells ( cf . Figure 2A ) . 10 . 7554/eLife . 08231 . 011Figure 4 . Yae1 is a Fe-S cluster maturation rather than stabilization factor for Rli1 . ( A ) Schematic representation of the light-induced rapid degradation of the Yae1-3Myc-AtLOV2-cODC1 fusion protein ( psd; photosensitive degron ) . In the dark the cODC1 degron is inactive and thus the fusion protein is stable . Irradiation with blue light induces a structural rearrangement within the LOV2 domain leading to the activation of the degron and direct degradation of the fusion protein by the 26S proteasome ( adapted from [Renicke et al . , 2013] ) . ( B ) WT and Yae1-psd ( photosensitive degron ) cells were transformed with a 2µ plasmid encoding RLI1-HA . Cells were grown overnight in iron-free minimal medium in the dark . Half of the cells ( 0 . 5 g ) were exposed to blue light in low fluorescence medium ( LFM ) , and the other half was kept in the dark for 2 hr . Cycloheximide ( 200 µg/ml ) was added , and cells were radiolabeled with 55Fe with or without blue light irradiation . The amount of 55Fe associated with Rli1 or Leu1 was quantified by IP and scintillation counting ( cf . Figure 2 ) . The radioactivity associated with Fe-S proteins in Yae1-psd mutant cells is presented relative to that of WT cells . Error bars indicate the SEM ( n > 3 ) . The bottom part shows a representative immunostain of the indicated proteins in cell extracts . ( C ) Three independent cultures of WT and Yae1-psd cells were grown overnight in iron-free minimal medium in the dark and radiolabeled with 55Fe for 2 hr . Culture 1 was assessed immediately for 55Fe-S cluster association to Rli1-HA or Leu1 by IP and scintillation counting ( t0 ) . Cultures 2 and 3 were washed with H2O and supplemented in LFM with cycloheximide . Cells were kept in the dark ( − ) or exposed to blue light ( + ) for additional 4 hr , and then analyzed for 55Fe-S cluster association . The bottom part shows a representative immunostain of the indicated proteins in extracts of Yae1-psd cells . DOI: http://dx . doi . org/10 . 7554/eLife . 08231 . 01110 . 7554/eLife . 08231 . 012Figure 4—figure supplement 1 . Fusion of Yae1 and Lto1 with a photosensitive degron allows their efficient degradation upon irradiation with blue light . ( A ) Blue light-induced degradation of psd-fusion proteins . Yeast cells expressing Yae1-psd or Lto1-psd were grown in liquid LFM supplemented with glucose in the dark . After removal of an aliquot ( t = 0 hr ) , cells were split and one half was exposed to blue light ( LED lamp 465 nm , 30 µmol × m−2 × s−1 ) , whereas the other half was kept in the dark . At indicated time points cells were lysed . The degradation of the fusion proteins was followed by immunostaining using α-Myc antibodies ( porin served as a loading control ) . ( B ) WT , Yae1-psd and Lto1-psd cells were serially diluted ( 1:5 , start OD600 = 0 . 5 ) and spotted onto YP agar plates supplemented with glucose . Plates were incubated in the dark or in blue light overnight . DOI: http://dx . doi . org/10 . 7554/eLife . 08231 . 012 We then investigated whether Yae1 might have a stabilizing effect on pre-assembled 55Fe-S clusters of Rli1 . WT and Yae1-psd cells were radiolabeled with 55Fe in the dark , and then exposed to blue light to induce the rapid degradation of Yae1 . A portion of the cells was harvested immediately after the radiolabeling reaction , and 55Fe associated with Rli1 and Leu1 was assessed ( Figure 4C , t0 ) . The remaining cells were incubated for 4 hr in the absence or presence of blue light , and 55Fe association to Rli1 and Leu1 was determined ( Figure 4C ) . Even though the amount of Rli1-bound 55Fe-S cluster dropped by 20% during the 4 hr incubation ( see below ) , exposure to blue light had no significant detrimental effect on cluster stability . Similar results were observed for the control Fe-S protein Leu1 , even though its Fe-S clusters appeared to be more stable over the 4 hr incubation period ( Figure 4C ) . These results strongly suggest that Yae1 deficiency does not significantly affect the Fe-S cluster stability of Rli1 . Hence , Yae1 functions as a specific CIA maturation factor for Rli1 rather than a Fe-S cluster-stabilizing protein . Most CIA components are essential for cell viability ( Netz et al . , 2014 ) . Strikingly , Yae1 and Lto1 are indispensable under aerobic , but not under anaerobic growth conditions ( Snoek and Steensma , 2006 ) ( Figure 5A ) . This raises the question whether Yae1 and Lto1 are dispensable for Fe-S cluster assembly on Rli1 under anaerobic conditions . When we analyzed the de novo 55Fe-S cluster assembly of Rli1 under anaerobic conditions , depletion of Yae1 or Lto1 led to similar Fe-S cluster assembly defects as depletion of Nbp35 ( Figure 5B ) . These effects were not caused by diminished Rli1 protein levels in the different cell types ( inset ) , demonstrating that Yae1 and Lto1 are involved in Rli1 maturation also under anaerobic conditions . 10 . 7554/eLife . 08231 . 013Figure 5 . Yae1 and Lto1 are required for Fe-S cluster assembly on Rli1 also under anaerobic conditions . ( A ) The indicated cells were cultivated overnight in rich medium containing glucose . Serial dilutions ( 1:5 ) were spotted onto glucose-containing rich medium agar plates . Growth was in the presence or absence of O2 . ( B ) WT , Gal-YAE1 , Gal-LTO1 and Gal-NBP35 cells were transformed with plasmids encoding RLI1-HA ( + ) or no gene ( − ) . Cells were cultivated for 30 hr in glucose-containing minimal medium under anaerobic conditions and transferred to minimal medium lacking iron for additional 16 hr . After radiolabeling for 2 hr with 55Fe under anaerobic conditions , cell extracts were prepared anaerobically and analyzed for 55Fe incorporation into Rli1-HA ( cf . Figure 2 ) . Error bars indicate the SEM ( n > 3 ) . The inset shows a representative immunostain of Rli1-HA and porin . ( C ) Tet-RLI1 cells were transformed with an empty vector , a plasmid containing RLI1 under its native promoter ( 300 bp upstream of RLI1 ) or plasmids containing RLI1 coding for cysteine to serine mutants of Rli1 as indicated . Cells were cultivated in glucose-containing minimal medium for 16 hr . Serial dilutions ( 1:5 ) were spotted under anaerobic conditions onto glucose-containing minimal medium agar plates supplemented with 5 µg/ml doxycycline to deplete endogenous Rli1 . Further growth was in the presence or absence of O2 . DOI: http://dx . doi . org/10 . 7554/eLife . 08231 . 013 These observations may suggest that Rli1 is not essential under anaerobic conditions ( Zhai et al . , 2014 ) . To test this idea , Rli1 was depleted under anaerobic conditions in the tetracycline-repressible Tet-RLI1 strain ( Kispal et al . , 2005 ) . No growth was observed , unless cells were transformed with a plasmid containing WT RLI1 ( Figure 5C , top rows ) demonstrating that Rli1 is essential even under anaerobiosis . A clue how to explain the non-essentiality of Yae1-Lto1 in the absence of oxygen came from the inspection of Rli1 mutant proteins in which cysteine residues 25 and 61 were exchanged either alone or simultaneously to serine ( Kispal et al . , 2005; Zhai et al . , 2014 ) . These residues coordinate two different [4Fe-4S] clusters of Rli1 ( Barthelme et al . , 2007 ) . Depleted Tet-RLI1 cells expressing single cysteine mutants of RLI1 were viable under anaerobic but not aerobic conditions ( Figure 5C , middle rows ) . In contrast , the RLI1 double mutation rendered cells inviable under both conditions ( bottom row ) . These results show that single mutations of RLI1 surprisingly support WT growth under anaerobic conditions ( see also [Zhai et al . , 2014] for the C25S mutation ) , apparently because Fe-S cluster binding to these Rli1 proteins is maintained despite the loss of one cluster coordination site . Such a behavior is not unusual for Fe-S proteins ( Urzica et al . , 2009; Netz et al . , 2012a , 2012b ) . Under aerobic conditions , and even more so in the presence of oxidants ( Alhebshi et al . , 2012 ) , the Rli1 cluster binding stability is too low to sustain growth . The observed higher stability of the Rli1 Fe-S clusters in the absence of oxygen may readily explain the non-essentiality of Yae1 and Lto1 , also indicating that their requirement in Fe-S cluster insertion into Rli1 can be bypassed to some extent under anaerobic conditions . We next investigated the functional importance of characteristic structural features of Yae1 and Lto1 . Both proteins contain an evolutionary conserved deca-GX3 motif of 40 residues ( Figure 6A ) that is not found in any other eukaryotic protein . Lto1 additionally carries a conserved C-terminal tryptophan ( phenylalanine in some organisms ) . We generated Lto1 mutant proteins at positions G17;G21 , G33;G37;G41 and G49;G53 of the deca-GX3 motif , and exchanged the conserved aspartate D4 and the C-terminal tryptophan to alanine ( Figure 6A ) . The HA-tagged Lto1 mutant versions were co-expressed together with Yae1-Myc to investigate their interaction . Since CIA protein depletion increased the complex formation between the CIA targeting complex and Yae1-Lto1 ( cf . Figure 1A ) , we used Nbp35-depleted Gal-NBP35 cells for this analysis , yet similar results were observed with WT cells . Protein amounts of HA-Lto1 were not affected by these substitutions . Immunoprecipitation ( IP ) showed that Lto1 mutations at positions D4 or G17;G21 slightly affected the interaction with the CIA targeting complex in comparison to WT Lto1 , whereas the mutations at positions G33;G37;G41 and G49;G53 almost fully abrogated this association . Likewise , these Lto1 mutations decreased complex formation with Yae1 ( Figure 6B ) . Interestingly , mutation of the C-terminal tryptophan strongly impaired complex formation with Cia1 , Cia2 and Mms19 , but not with Yae1 . 10 . 7554/eLife . 08231 . 014Figure 6 . The conserved deca-GX3 motif and the C-terminal tryptophan are functionally crucial elements of Lto1 . ( A ) Cartoon of S . cerevisiae Lto1 to highlight the mutated residues ( red ) within the deca-GX3 motif and the N- and C-termini ( cf . Figure 2—figure supplement 1B ) . ( B ) Gal-NBP35 yeast cells were co-transformed with 2µ vectors containing either no insert or genes encoding HA-Lto1 , Yae1-Myc or HA-tagged Lto1 mutants as indicated . Cells were cultivated in minimal medium containing glucose . Cell lysates were prepared , and a trichloroacetic acid ( TCA ) precipitation was performed to assess the expression levels of the fusion proteins ( Input ) . After HA- or Myc-tag IP the precipitate was analyzed for the indicated tags or proteins by immunoblotting . ( C ) Cells were co-transformed with plasmids containing RLI1-HA and single copy plasmids containing LTO1 or mutated versions as indicated . Cells were cultivated for 24 hr in glucose-containing minimal medium and the 55Fe radiolabeling-IP procedure was performed to estimate the amount of 55Fe associated with Rli1-HA . Error bars indicate the SEM ( n > 4 ) . The inset shows a representative immunostain of Rli1-HA and porin . DOI: http://dx . doi . org/10 . 7554/eLife . 08231 . 014 To study the consequences of these mutations for Lto1 function , we analyzed their influence on Rli1 maturation by 55Fe radiolabeling of Lto1-depleted cells producing the modified Lto1 versions ( Figure 6C ) . Mutations of residues D4 or G17;G21 had only mild effects on 55Fe-S cluster assembly of Rli1 . In contrast , mutations within the middle or C-terminal region of the deca-GX3 motif or exchange of the C-terminal tryptophan severely impaired Rli1 maturation . Collectively , these data show that the deca-GX3 domain is crucial for Lto1 complex formation with both Yae1 and the CIA targeting complex , while the C-terminal tryptophan of Lto1 is specifically required for the interaction with the CIA targeting complex . These impaired protein interactions account for the functional deficit of mutated Lto1 in Rli1 maturation , suggesting that the deca-GX3 motif-tethered Lto1-Yae1 complex binds to the CIA targeting complex via the Lto1 tryptophan and to Rli1 via Yae1 ( Zhai et al . , 2014 ) . Searches of the human genome for homologs of yeast Yae1 and Lto1 identified YAE1D1 ( Yae1 domain-containing protein 1 ) and ORAOV1 ( oral cancer-overexpressed protein 1 , [Huang et al . , 2002; Jiang et al . , 2008] ) , respectively ( Figure 2—figure supplement 1 ) . We previously have found these proteins amongst the components interacting with the human CIA targeting complex ( Stehling et al . , 2012 , 2013 ) . We investigated whether the two human proteins can replace the essential function of their yeast counterparts upon expression in Gal-YAE1 or Gal-LTO1 cells . Neither YAE1D1 nor ORAOV1 restored the growth defects of yeast cells depleted for Yae1 or Lto1 ( Figure 7A ) . In contrast , co-expression of both YAE1D1 and ORAOV1 fully restored growth , indicating the function of these proteins as a homologous complex . As expected from the lack of growth rescue , 55Fe-S cluster assembly on Rli1 was not improved upon expression of the single human proteins in Yae1- or Lto1-depleted yeast cells ( Figure 7B ) . However , co-expression of both human complex partners restored Rli1 maturation to WT efficiency ( Figure 7C ) . These data demonstrate that the human YAE1D1-ORAOV1 complex functionally replaces the yeast counterpart suggesting that the human proteins perform a conserved function in the maturation of the Rli1 homolog ABCE1 . 10 . 7554/eLife . 08231 . 015Figure 7 . The human YAE1D1-ORAOV1 complex can functionally replace Yae1-Lto1 . ( A ) WT , Gal-YAE1 and Gal-LTO1 cells were transformed with 2µ vectors containing either no insert ( empty ) or genes encoding human YAE1D1 and/or ORAOV1 as indicated . Cells were cultivated overnight in minimal medium containing glucose to deplete Yae1 and Lto1 ( ↓ ) . Serial dilutions ( 1:5 ) were spotted onto minimal medium agar plates supplemented with glucose ( Glc ) or galactose ( Gal ) . ( B ) Cells expressing RLI1-HA plus YAE1D1 or ORAOV1 were radiolabeled with 55Fe , and the amount of 55Fe associated with Rli1-HA was quantified by scintillation counting . Error bars indicate the SEM ( n > 4 ) . The inset shows a representative immunostain of Rli1-HA and porin . ( C ) In a similar analysis as in part B the effects of the expression of YAE1 , LTO1 , YAE1D1 or ORAOV1 for Rli1-HA maturation in Yae1- and Lto1-depleted cells were estimated . DOI: http://dx . doi . org/10 . 7554/eLife . 08231 . 015 In this work we have identified and functionally characterized the essential proteins Yae1 and Lto1 as two novel members of the CIA machinery . Their function is unique , because , unlike the previously identified CIA proteins , these components do not act as general Fe-S maturation factors . Rather , they function downstream of the known CIA proteins by specifically assembling the essential Fe-S protein Rli1 . Our findings allowed us to expand the current model of CIA ( Figure 8 ) ( Paul and Lill , 2014 ) . Yae1 and Lto1 form a complex that acts as a dedicated adaptor to recruit Rli1 to the late-acting part of the CIA machinery . This model was derived from the interaction of Yae1-Lto1 with both the CIA targeting complex ( Cia1-Cia2-Mms19 ) and the Rli1 client . The high target specificity of Yae1-Lto1 is further documented by the fact that the complex is also dispensable for the maturation of Fe-S cluster-containing CIA proteins such as Nbp35 and Nar1 . Yae1 and Lto1 perform non-overlapping , individual functions , since even upon overexpression they cannot mutually complement each other , despite some relationship in amino acid sequence ( see below ) . Nevertheless , their function is conserved in higher eukaryotes as the human proteins YAE1D1 and ORAOV1 can substitute their yeast counterparts , but only when co-expressed , again documenting the function as a complex . Collectively , these data demonstrate that the Yae1-Lto1 complex acts as a conserved specificity factor for Fe-S cluster maturation of the highly conserved eukaryotic protein Rli1 ( human ABCE1 ) . 10 . 7554/eLife . 08231 . 016Figure 8 . Working model for the specific function of the CIA proteins Yae1 and Lto1 in the maturation of the cytosolic Fe-S protein Rli1 . Maturation of cytosolic and nuclear Fe-S proteins is a multi-step process conducted by different CIA protein subcomplexes . First , a [4Fe-4S] cluster is assembled on the scaffold protein complex composed of Cfd1-Nbp35 . This reaction depends on a yet unknown sulfur-containing compound ( X-S ) which is produced by the mitochondrial iron-sulfur cluster ( ISC ) assembly machinery and is exported by the mitochondrial ABC transporter Atm1 to the cytosol . Further , the electron transfer chain NADPH-Tah18-Dre2 is required . The Grx3-Grx4 complex mediates a yet undefined function , possibly in delivering iron , for cytosolic-nuclear Fe-S protein biogenesis . Second , the newly assembled [4Fe-4S] cluster is released from Cfd1-Nbp35 and inserted into target apoproteins via Nar1 and the CIA targeting complex ( Cia1-Cia2-Mms19 ) . The precise mechanism of the latter steps in still unclear . As reported in this study , maturation of the essential Fe-S protein Rli1 additionally depends on the function of the two specific adaptor proteins Yae1 and Lto1 . The Yae1-Lto1 complex uses a unique binding cascade to recruit Rli1 to the CIA targeting complex for Fe-S cluster insertion . The CIA targeting complex interacts with the conserved C-terminal tryptophan residue ( W ) of Lto1; the conserved deca-GX3 motifs ( yellow boxes ) of Yae1-Lto1 are crucial for their complex formation; and Yae1 associates with Rli1 . Such an adaptor function is dispensable in the maturation pathway of the virus-induced Fe-S protein viperin because it directly binds to the CIA targeting complex via its conserved C-terminal tryptophan residue . DOI: http://dx . doi . org/10 . 7554/eLife . 08231 . 016 The amino acid sequences of Yae1 and Lto1 do not contain any characteristic similarities to other eukaryotic proteins , yet both proteins share the short deca-GX3 sequence motif that is unique for these proteins in eukaryotes . A similar , but shorter ‘GX3G’ motif has been recognized in a number of membrane proteins , where the GX3G motif forms trans-membrane helix dimerization by inter-helical hydrogen bonds ( Senes et al . , 2004 ) . The function of the deca-GX3 motif has not been explored until now , but we speculate that it serves a role in Yae1-Lto1 heterodimer formation as in membrane proteins . The deca-GX3 motif is also found in bacterial FliH , a protein involved in the assembly and substrate export of flagella ( Bai et al . , 2014 ) . FliH forms a dimer and transiently binds to the ATPase FliI . Bacterial FliI is similar in structure to the αβ-subunits of F1F0 ATPase and hence not related to the ABC-type ATPase Rli1 , leaving open any functional relevance of the deca-GX3 protein—ATPase interaction . Our Lto1 mutational studies demonstrated that the deca-GX3 motif is functionally important for Yae1-Lto1 complex formation , and consequently for complex interaction with the CIA machinery ( Figure 8 ) . Exchanges of conserved glycine residues of Lto1 , for instance in the middle and C-terminal parts of the deca-GX3 motif , strongly affected its association with both Yae1 and the CIA targeting complex , thus explaining the strongly impaired efficiency of these Lto1 variants to support Fe-S cluster assembly on Rli1 . In contrast , mutation of two N-terminal glycine residues caused only moderate effects on complex formation and Rli1 maturation suggesting some residual function of this disrupted deca-GX3 motif . A strikingly different result was obtained for mutation of the conserved C-terminal tryptophan residue of Lto1 . Exchange to alanine had no detectable effect on complex formation with Yae1 , yet the association of the Yae1-Lto1 complex with the CIA targeting complex was severely disrupted . The decreased Fe-S cluster assembly of Rli1 observed for this Lto1 W → A mutation is therefore best explained by the weakened contact of Lto1 to the CIA machinery thus limiting the recruitment of the Rli1 client ( Figure 8 ) . In summary , our mutational studies strongly suggest a model for Yae1-Lto1 acting as a dedicated adaptor complex mediating the contacts between the CIA targeting complex and Rli1 . The CIA targeting complex binds Lto1 via its C-terminal W , the two deca-GX3 motifs are crucial for Yae1-Lto1 complex formation , and , as shown earlier ( Zhai et al . , 2014 ) , Yae1 mediates the contact to Rli1 ( Figure 8 ) . This chain of binding events facilitates efficient Rli1 maturation . Interestingly , a characteristically different but related maturation strategy is followed by the radical SAM Fe-S protein viperin , an interferon-induced antiviral defense component ( Upadhyay et al . , 2014 ) . This protein uses its conserved C-terminal tryptophan to directly associate with the CIA targeting complex . Removal of this residue abolishes complex formation with the CIA targeting complex , assembly of viperin's Fe-S cluster , and consequently antiviral function . The presence of a tryptophan at the viperin C-terminus thus renders an adaptor function like that of Yae1-Lto1 dispensable . Lto1 has recently been characterized as a protein involved in the biogenesis of the large ribosomal subunit and in the initiation of translation ( Zhai et al . , 2014 ) . Based on our current results this suggested role of Lto1 in protein synthesis seems to be indirect , and rather is mediated via its ( primary ) function in Rli1 maturation . As noted above , this Fe-S protein performs key functions in ribosomal subunit biogenesis as well as translation initiation , termination , and recycling . For all these functions , its two Fe-S clusters are essential ( Kispal et al . , 2005; Khoshnevis et al . , 2010; Barthelme et al . , 2011; Becker et al . , 2012 ) . Consistent with the ( indirect ) roles of Lto1 and also Yae1 in ribosomal protein translation , we found protein synthesis defects in cells deficient in these proteins , similar to what we previously described for a deficiency in Rli1 ( [Kispal et al . , 2005] and this study ) . The Fe-S clusters of Rli1 were reported to be particularly sensitive to added oxidants suggesting that the clusters may need stabilization , in particular under oxidative conditions ( Alhebshi et al . , 2012 ) . We designed an experimental approach that allowed us to test the potential stabilizing function of Yae1-Lto1 for the Rli1 Fe-S clusters . This involved the combination of blue light-induced rapid degradation of Yae1 and the estimation of the Fe-S cluster stability of radiolabeled Rli1 . Depletion of Yae1 by blue light exposure diminished de novo Fe-S cluster assembly of Rli1 , yet did not affect the stability of the clusters , clearly documenting that Yae1 ( and presumably Lto1 ) function as maturation rather than stabilization factors for Rli1 . This maturation function is particularly needed under oxidative conditions to generate enough functional Rli1 to support normal growth . In contrast , under anaerobic conditions both YAE1 and LTO1 are dispensable for normal growth ( [Snoek and Steensma , 2006]; this work ) , indicating that their function can be bypassed to some extent in the absence of oxidative stress . Nevertheless , as documented by our radiolabeling assay these proteins are still beneficial for Rli1 maturation under anaerobic conditions . The decreased need to ( re- ) generate damaged Fe-S clusters of Rli1 may render them dispensable . These findings are fully consistent with a recent study concluding that the oxygen sensitivity of Rli1 is best explained by affecting Fe-S cluster synthesis and/or transfer steps rather than the Rli1 Fe-S cluster stability ( Alhebshi et al . , 2012 ) . This view nicely fits to the observation that RLI1 functions as a high copy suppressor of LTO1 mutations ( Zhai et al . , 2014 ) . Collectively , the sum of these studies suggest that Yae1-Lto1 largely increase the efficiency of Rli1 maturation , in particular under aerobic conditions , yet in the absence of oxygen these factors can be bypassed to some extent without losing cell viability . The sensitivity of Fe-S protein biogenesis pathways towards oxygen and/or oxidative stress has also been noted for members of the bacterial ISC assembly system . For instance , Azotobacter vinelandii IscA was required for the maturation of [4Fe-4S] proteins only under high oxygen concentrations ( Johnson et al . , 2006 ) , and Escherichia coli IscA-SufA paralogs are needed for Fe-S protein assembly especially under aerobic growth conditions ( Tan et al . , 2009 ) . Similarly , E . coli ErpA is essential for Fe-S cluster maturation of isoprenoid biosynthesis enzymes under aerobic conditions , but can be replaced by IscA under anaerobiosis ( Vinella et al . , 2009 ) . As a second example , the bacterial NfuA protein functions as a Fe-S cluster transfer protein that is essential for maturation of dedicated Fe-S proteins such as aconitase particularly in the presence of oxygen or oxidative stress conditions , yet can be bypassed under low oxygen conditions ( Angelini et al . , 2008; Bandyopadhyay et al . , 2008; Py et al . , 2012 ) . In comparison and on the contrary , the mitochondrial ISC assembly pathway is crucial under both aerobic and anaerobic conditions . Deletion of S . cerevisiae GRX5 encoding the mitochondrial monothiol glutaredoxin largely impairs growth rates on minimal medium even under anaerobiosis , and functional inactivation of the Isa1-Isa2 proteins involved in [4Fe-4S] protein biogenesis does not allow cell growth and mitochondrial [4Fe-4S] protein maturation in the presence or absence of oxygen ( Rodriguez-Manzaneque et al . , 2002; Muhlenhoff et al . , 2011 ) . The characterization of Yae1-Lto1 as a dedicated adaptor complex of the CIA machinery for recruitment of specific Fe-S proteins such as Rli1 may be a paradigm for other Fe-S clients and may suggest the existence of additional devoted CIA maturation factors . In mitochondria , several specific ISC assembly factors have been described including the P-loop NTPase IND1 involved in complex I maturation , NFU1 and BOLA3 required for assembly of complex Fe-S proteins such as respiratory complexes I and II and lipoic acid synthase ( Stehling and Lill , 2013 ) . Their molecular mode of action in the assembly process is still unclear , but an adaptor role similar to Yae1-Lto1 can now be tested . Future elucidation of the molecular mechanisms of these specificity factors will clarify whether they follow common or distinct strategies for Fe-S cluster insertion . Solving their 3D structure will certainly be helpful for understanding how these proteins perform their adaptor role between the late parts of the ISC or CIA machineries and their dedicated Fe-S clients . S . cerevisiae strain W303-1A was used as WT strain . Detailed information on the yeast strains , oligonucleotides , and expression vectors used in this study is provided in Supplementary file 1 . Human YAE1D1 and ORAOV1 genes ( IMAGE ID 4544931/AU101 H10 M13F ) were obtained from Eurofins and SourceBioscience , respectively . Yeast strains were cultivated in rich ( YP ) or minimal ( SC ) medium containing the required carbon sources at a concentration of 2% ( wt/vol ) ( Sherman , 2002 ) . For growth under anaerobic conditions media were supplemented with 0 . 2% ( vol/vol ) ethanol , 0 . 2% ( vol/vol ) Tween-80 and 30 µg/ml ergosterol . Low fluorescence medium was used for cultivation of cells expressing psd-fusion proteins ( Taxis et al . , 2006 ) . TAP was performed as described ( Rigaut et al . , 1999; Puig et al . , 2001; Gavin et al . , 2002 ) . Eluted proteins were reduced by 5 mM DTT for 30 min at 56°C , alkylated in 7 mM iodoacetamide for 30 min at room temperature in the dark , and then precipitated in 20% ice-cold trichloroacetic acid . Protein pellets were resuspended in trypsin solution ( 30 µl of 10% acetonitrile ( ACN ) , 50 mM ammonium bicarbonate ( Ambic ) , and 0 . 025 µg trypsin ) and digested at 37°C overnight . Digestion was stopped by addition of 30 µl of 50% ACN in 5% formic acid ( FA ) . Peptide solutions were dried in a speed vac and resuspended in 0 . 1% trifluoroacetic acid for StageTIP purification ( Rappsilber et al . , 2007 ) . StageTIPs were prepared with 2 layers of 3 M Empore C18 membranes . Peptides were eluted with 40 µl of 50% ACN in 1% acetic acid and dried in a speed vac prior to MS analysis . After resuspension in 20 µl of 5% ACN in 5% FA , 10% of each sample were analyzed by LC-MS/MS during a 45 min gradient by reversed-phase chromatography coupled to an LTQ-Orbitrap Velos mass spectrometer ( Thermo Scientific , Waltham , MA ) in Higher Energy Collision Dissociation mode . Peak lists were extracted using MaxQuant ( version 1 . 1 . 1 . 36 ) ( Cox and Mann , 2008 ) and submitted to Mascot 2 . 2 . 03 ( www . matrixscience . com ) searches against the yeast SGD protein database including common contaminants , with carbamidomethylation ( C ) as a fixed modification , oxidation ( M ) , as well as deamidation ( N , Q ) and protein N-terminal acetylation as variable modifications . Mass error tolerances were set to 10 ppm ( MS ) and 0 . 05 Da ( MS/MS ) . Search results were loaded into Scaffold ( version 3 . 3 . 1 , www . proteomesoftware . com ) and into ProHITS ( Liu et al . , 2010 ) for subsequent SAINT analysis ( Choi et al . , 2011 ) . To extract bait-prey interactions from our TAP-MS data , we loaded our Mascot search results without any Mascot score threshold into the ProHITS database ( Liu et al . , 2010 ) and made use of the integrated SAINT algorithm ( Choi et al . , 2011 ) to perform a probabilistic scoring . This algorithm assigns a confidence score to the probability of observing a true interaction using Poisson distributions . In order to assess the dependency of the observed interactions to changes in the SAINT parameters , we applied the SAINT algorithm by using a range of different parameters , for example , the average confidence score threshold ( 0 . 8–0 . 95 in steps of 0 . 01 , range taken from Choi et al . ( 2011 ) ) , Mascot scores ( 10–50 in steps of 10 ) , with or without taking extremely high counts into account , and with or without normalization for spectral counts . In total , we had 320 parameter combinations . By varying these parameters we then counted the number of times that we observed an interaction against the number of variations , thus giving a fraction between 0 and 1 for each bait-prey pair . We only considered interactions that have been assigned an average SAINT score above 0 . 8 in at least one of the parameter sets . We then built a network with interactions of high ( ≥0 . 5 ) and low ( <0 . 5 ) robustness , that is , resistance or sensitivity to variations in SAINT parameters . To support our findings and to expand our network of observed bait-prey interactions , we integrated data from the STRING database ( Szklarczyk et al . , 2011 ) . For interactions observed in our screen , we kept all experimental STRING data without a confidence score threshold . For additional STRING interactions , we only considered experimentally observed interactions with the highest predefined STRING confidence score ( 0 . 9 ) . For ( co- ) IP cells were cultivated for 40 hr in glucose-containing minimal medium and 0 . 5 g cells were resuspended in 0 . 55 ml lysis buffer ( 50 mM Tris-HCl pH 7 . 5 , 5% [vol/vol] glycerol , 100 mM NaCl , 1 . 5 mM MgCl2 , 0 . 2% [vol/vol] NP-40 , 1 mM DTT , 1 mM PMSF ) with Complete Protease Inhibitor Cocktail ( Roche ) . Cells were disrupted by vortexing with glass beads ( three 1 min bursts ) and debris removed ( 1500×g , 5 min , 4°C ) . The supernatant was further clarified by centrifugation ( 13 , 000×g , 10 min , 4°C ) . Anti-HA or anti-Myc beads were added to the clarified cell extracts , and the samples rotated for 1 . 5 hr at 4°C . Beads were washed three times with 0 . 5 ml lysis buffer and analyzed by immunoblotting using monoclonal antibodies against the HA or Myc-tag ( Santa Cruz Biotechnology ) or rabbit antibodies raised against the indicated proteins . During the construction of the GALL promoter-containing Yae1 and Lto1 depletion strains we noted that a promoter insertion according to the translation start site of LTO1 ( as listed by the SGD data base ) resulted in a strain ( termed Gal-LTO1long ) that showed no growth defect under depletion conditions , that is , growth on glucose-containing media ( Figure 2—figure supplement 2B ) . This is in contrast to the findings for the Gal-YAE1 strain , and the fact that both genes are essential for cell viability . Upon examination of the reason for the lack of a growth defect , we suspected that the Gal-LTO1long strain did not allow a critical reduction of the Lto1 protein . A combination of bioinformatic , cell biological and biochemical analyses provided clear information that the reason for this observation was an erroneous translation start site for LTO1 . As summarized in detail below the correct physiological start site of the LTO1 gene is located 108 bp downstream of the annotated start codon ( ATG3 in Figure 2—figure supplement 2A ) . The first hint for an annotation error of the LTO1 gene was provided by a multi-sequence alignment of Lto1 homologs . S . cerevisiae Lto1 contained an unusual N-terminal extension of 36 amino acid residues ( highlighted in cyan in Figure 2—figure supplement 1A ) that was not present in any other Lto1 homolog . Inspection of the S . cerevisiae Lto1 protein sequence suggested another putative translation start site for Lto1 ( residue Met37 corresponding to codon ATG3; Figure 2—figure supplement 2A ) which roughly matches the N-termini of the other Lto1 homologs . Ribosome foot-printing data ( Ingolia et al . , 2009 ) supported that the actively translated region of the LTO1 mRNA does not encompass the N-terminal extension of 36 residues ( Figure 2—figure supplement 3A ) . To directly test the importance of the N-terminal 36 residues , we created a LTO1 mutant in which the GALL promoter was inserted right in front of ATG3 . This mutant strain ( termed Gal-LTO1 ) showed normal growth on galactose-containing medium demonstrating that the N-terminal 36 residue extension of Lto1 is not critical for cell viability ( Figure 2—figure supplement 2B ) . Conspicuously , the Gal-LTO1 cells were unable to grow on glucose-containing medium suggesting efficient depletion of Lto1 , unlike that seen for Gal-LTO1long cells . To investigate the role of the N-terminal Lto1 segment , we exchanged the coding information for methionine 1 ( ATG1 ) and an additional methionine ( residue 10 , ATG2 ) within the 36 residue segment for a stop codon ( Figure 2—figure supplement 3B ) . These proteins and the non-mutated Lto1 were expressed from a centromeric plasmid under control of the endogenous promoter in Gal-LTO1 cells grown with galactose or glucose . Both LTO1 mutations did not affect cell growth . This result clearly documented that the N-terminal segment of Lto1 including its previously proposed start methionine is not essential suggesting that ATG3 rather than ATG1 is the physiologically correct translation start site ( Figure 2—figure supplement 3B ) . To further support this conclusion , we analyzed the importance of the DNA region between codon ATG1 and ATG3 for LTO1 expression efficiency . To this end , we fused different DNA segments of the LTO1 gene in front of the luciferase reporter gene ( Figure 2—figure supplement 3C ) . The fusion constructs were transformed in WT yeast cells and the luciferase-based luminescence was recorded as a measure of LTO1 promoter efficiency . Any truncation within the ATG1-ATG3 segment strongly impaired the transcriptional activity suggesting that this region has gene regulatory rather than coding function . Collectively , these findings demonstrate that the physiological start site of LTO1 is located 108 bp downstream of the previously annotated start codon . The following published methods were used: manipulation of DNA and PCR ( Sambrook and Russell , 2001 ) , transformation of yeast cells ( Gietz and Woods , 2002 ) , immunostaining ( Harlow and Lane , 1998 ) , in vivo labeling of yeast cells with 55FeCl3 and measurement of 55Fe incorporation into Fe-S proteins by IP and scintillation counting ( Pierik et al . , 2009 ) , pulse labeling of yeast cells with 35S-methionine ( Kispal et al . , 2005 ) , TAP tag affinity purification ( Puig et al . , 2001; Gavin et al . , 2002 ) , determination of promoter strength using luciferase ( Molik et al . , 2007 ) , light-induced degradation of psd-fused target proteins ( Renicke et al . , 2013 ) and enzyme activity measurements ( Molik et al . , 2007 ) . Sulfite reductase activity was assessed in vivo ( Stehling et al . , 2012 ) . In this case , cells were cultivated on rich medium agar plates supplemented with galactose before they were spotted onto minimal medium agar plates supplemented with galactose or glucose and 1% ( wt/vol ) β-alanin , 0 . 1% ( wt/vol ) ammonium bismuth citrate , 0 . 3% ( wt/vol ) sodium sulfite .	Many proteins depend on small molecules called cofactors to be able to perform their roles in cells . One class of proteins—the iron-sulfur proteins—contain cofactors that are made of clusters of iron and sulfide ions . In yeast , humans and other eukaryotes , the clusters are assembled and incorporated into their target proteins by a group of assembly factors called the CIA machinery . Several components of the CIA machinery have previously been identified and most of them appear to be core components that are needed to assemble many different proteins in cells . Since these iron-sulfur proteins are involved in important processes such as the production of proteins and the maintenance of DNA , losing of any of these CIA proteins tends to be lethal to the organism . Paul et al . used several ‘proteomic’ techniques to study the assembly of iron-sulfur proteins in yeast and identified two new proteins called Yae1 and Lto1 that are involved in this process . Unlike other CIA proteins , Yae1 and Lto1 are only required for the assembly of just one particular iron-sulfur protein called Rli1 , which is essential for the production of proteins . Most newly made iron-sulfur proteins can bind directly to a group of CIA proteins called the CIA targeting complex , but Rli1 cannot . The experiments show that Lto1 binds to both the CIA targeting complex and to Yae1 , which in turn recruits the Rli1 to the CIA complex . Paul et al . also show that humans have proteins that are very similar to Yae1 and Lto1 . Inserting the human counterparts of Yae1 and Lto1 into yeast lacking these proteins could fully restore the assembly of iron-sulfur clusters into Rli1 . This suggests that Yae1 and Lto1 proteins evolved in the common ancestors of fungi and humans and have changed little since . Taken together , Paul et al . 's findings reveal that Yae1 and Lto1 act as adaptors that link the rest of the CIA machinery to their specific target protein Rli1 in yeast and humans . A future challenge is to find out the three-dimensional structures of Yae1 and Lto1 to better understand how these proteins work and interact .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "biochemistry", "and", "chemical", "biology", "cell", "biology" ]	2015	The deca-GX3 proteins Yae1-Lto1 function as adaptors recruiting the ABC protein Rli1 for iron-sulfur cluster insertion
Endothelial cells respond to molecular and physical forces in development and vascular homeostasis . Deregulation of endothelial responses to flow-induced shear is believed to contribute to many aspects of cardiovascular diseases including atherosclerosis . However , how molecular signals and shear-mediated physical forces integrate to regulate vascular patterning is poorly understood . Here we show that endothelial non-canonical Wnt signalling regulates endothelial sensitivity to shear forces . Loss of Wnt5a/Wnt11 renders endothelial cells more sensitive to shear , resulting in axial polarization and migration against flow at lower shear levels . Integration of flow modelling and polarity analysis in entire vascular networks demonstrates that polarization against flow is achieved differentially in artery , vein , capillaries and the primitive sprouting front . Collectively our data suggest that non-canonical Wnt signalling stabilizes forming vascular networks by reducing endothelial shear sensitivity , thus keeping vessels open under low flow conditions that prevail in the primitive plexus . Functional blood vessel networks are essential for vertebrate development , tissue growth and organ physiology ( Potente et al . , 2011 ) . Vessel assembly and sprouting establish the major axial vessels and a primary network , which undergoes extensive remodelling to become functional . Also in the adult , vascular networks can be reactivated , expanded to meet changing metabolic demands , or remodelled , as a consequence of injury or local occlusion ( Carmeliet , 2005; Potente et al . , 2011 ) . A large number of mouse mutants have been described as having defects in vascular remodelling . Yet , in contrast to vascular sprouting , very little is known about the intrinsic cellular and molecular mechanisms controlling vascular remodelling . One aspect of remodelling is vessel segment regression , in which existing connections are lost . Endothelial cell ( EC ) death drives programmed regression of the ocular hyaloid vessels ( Lobov et al . , 2005 ) and pupillary membrane ( Meeson et al . , 1999 ) . Whilst a similar mechanism was thought to be driving developmental vascular remodelling , recent reports proposed that vessel segment regression in the remodelling retinal blood vessels involves dynamic rearrangement of ECs , which actively migrate from regressing vessel segments to integrate into neighbouring vessels ( Franco et al . , 2015; Udan et al . , 2013 ) . Chen et al . postulated that ECs in zebrafish brain vessels sense a threshold of low blood flow below which vessel regression is triggered irreversibly ( Chen et al . , 2012 ) . Our recent rheology modelling of the retinal plexus also predict regression of poorly perfused vessel segments ( Bernabeu et al . , 2014; Franco et al . , 2015 ) , and demonstrated that EC axial polarization against the blood flow direction is a conserved feature in remodelling vessels ( Franco et al . , 2015 ) . Our observations lead us to propose that flow-induced EC polarization directs migration of ECs that reside in low flow or oscillatory flow segments towards juxtapose high flow segments . As a consequence , this movement of ECs between vessel segments with differential flow regimes leads to regression of low-flow branches and stabilization of the higher flow segments ( Franco et al . , 2015 ) . Blood flow is critical for vascular remodelling ( Hahn and Schwartz , 2009 ) , but the relevance and the mechanistic understanding of how physical forces and signalling pathways collectively stabilize or disrupt vessel connections remains unknown . Here we show that ECs use non-canonical Wnt ligands in a short-range , paracrine manner to stabilize connections during vascular remodelling . We show that loss of endothelial-derived Wnt5a and Wnt11 sensitizes ECs to polarize against the blood flow direction at lower levels of wall shear stress , in vitro and in vivo , thereby leading to premature and excessive vessel regression in mouse . We postulate that the enhanced sensitivity to flow in non-canonical Wnt-deficient endothelium promotes earlier discrimination of flow asymmetries between neighbouring vessel segments in the capillary plexus , thus driving premature vessel regression and accelerated remodelling . Wnt/β-catenin signalling has been shown to both promote and inhibit vessel regression ( Lobov et al . , 2005; Phng et al . , 2009 ) . To gain further insight into the role of Wnt ligands in vessel regression , we conditionally inactivated Wnt-ligand secretion by recombination of the floxed Wls/Evi/Gpr177 allele ( Carpenter et al . , 2010 ) in ECs ( Pdgfb-iCreERT2 or Tie2-Cre ) . Wls encodes for a transporter chaperone protein required for secretion of all Wnt ligands ( Banziger et al . , 2006; Bartscherer et al . , 2006 ) . Embryonic endothelial-specific Wls deletion ( Wlsfl/fl::Tie2-Cre ) leads to mid-gestation lethality , demonstrating an important vascular function for endothelial-derived Wnt ligands ( Table 1 ) . Tamoxifen-inducible Wls deletion in ECs ( Pdgfb-iCreERT2::Wlsfl/fl , hereafter Wls iEC-KO ) led to significantly decreased vascular density compared to littermate controls ( Figure 1a ) . Quantification revealed increased regression profiles ( quantified by the Col . IV-sleeves and Icam2-breakage profiles ) , while sprout frequency , proliferation , EC density and apoptosis rates were unaffected ( Figure 1b and Figure 1—figure supplement 1a , b ) . Surprisingly , these results did not recapitulate recent findings by Korn et al . who reported that Wls iEC-KO causes increased vessel regression through increased apoptosis ( Korn et al . , 2014 ) . Our recent work established that regression in the mouse vasculature follows a sequence of events that begin with vessel stenosis , followed by cell retraction that finally leads to resolution , leaving only empty matrix behind ( Franco et al . , 2015 ) . The frequency distribution of regression profiles at these distinct stages of segment regression , i . e . stenosis , retraction or resolution , was similar in Wls iEC-KO and Wls WT mice ( Figure 1c ) indicating that the lack of secretion of Wnt ligands from ECs affects the frequency but not the mechanism of vessel regression . Experimental hyperoxia-induced vessel obliteration ( which is driven by endothelial apoptosis ( Alon et al . , 1995 ) caused similar central capillary network dropout in Wls WT and Wls iEC-KO , suggesting that EC Wnt-ligands are not able to significantly protect from endothelial cell apoptosis-mediated vessel regression events ( Figure 2a , b ) . Defects in pericyte recruitment have been linked to increased vessel instability and vessel regression ( Benjamin et al . , 1998 ) . We analysed pericyte coverage using NG2 marker and observed no significant changes between Wls WT and Wls iEC-KO retinas ( Figure 3a , b ) . 10 . 7554/eLife . 07727 . 003Table 1 . Tie2-Cre Wlsfl/fl embryos die at mid-gestation . Table showing number of embryos collected at the specified embryonic time point post-coitum ( E ) and at birth . Relative frequency each genotype of embryos/pups is shown as percentage . DOI: http://dx . doi . org/10 . 7554/eLife . 07727 . 003♂ Tie2-Cre::Wls fl/wt X ♀ Wls fl/flGenotypeE12 . 5%E13 . 5%E16 . 5%Adults%Wlsfl/fl60 . 2540 . 2970 . 41200 . 38Wlsfl/wt50 . 2140 . 2940 . 24130 . 25Wlsfl/fl::Cre+60 . 2540 . 294 ( 3 dead ) 0 . 2400 . 00Wlsfl/wt::Cre+70 . 2920 . 1420 . 12200 . 3810 . 7554/eLife . 07727 . 004Figure 1 . EC-derived Wnt ligands protect against vessel regression . ( a ) Overview of retinal vascular plexus of P6 control and Wls iEC-KO mice , labeled for lumen ( ICAM2 ) , ECs ( IB4 ) and ECM ( Col . IV ) . ( b ) Quantification of vascular parameters demonstrating that increased vessel regression is a main feature of Wls iEC-KO in P6 retinas . IB4/Col . IV regression profiles correspond to the number Col . IV-positive segments negative for IB4 staining . ICAM2/Col . IV regression profiles correspond to number of Col . IV-positive vessel segments partially or totally negative for ICAM2 staining . p values from unpaired , two-tailed t-test . Mean +/-SEM; N = 6 mice; 3 litters . ( c ) High magnification images of the vascular plexus of control and Wls iEC-KO mice marked for each stage of vessel regression ( stenosis , blue arrows; retraction , green arrow; resolution , red arrows ) . ( d ) Quantification of the each specific vessel regression stage in Wls iEC-KO and Wls WT P6 retinas ( stenosis , blue bars; retraction , green bars; resolution , red bars ) . Two-way ANOVA with Sidak multiple comparisons test . Mean +/-SEM; N = 5 mice; 3 litters; N = 288 and N = 398 regression events for Wls WT and Wls iEC-KO mice , respectively . DOI: http://dx . doi . org/10 . 7554/eLife . 07727 . 00410 . 7554/eLife . 07727 . 005Figure 1—figure supplement 1 . Wls iEC-KO mice show normal proliferation and apoptosis rates . ( a ) Representative confocal images of proliferation profile in Wls iEC-KO and Wls WT P6 mouse retinas labeled with ICAM2 ( red ) , endothelial cell nuclei ( Erg , green ) and EdU ( blue ) . ( b ) Representative segmentation of confocal images stained for active caspase-3 and blood vessels ( IB4 ) showing distribution of endothelial cells positive for active caspase-3 ( green dots ) in Wls iEC-KO and Wls WT P6 mouse retinas . DOI: http://dx . doi . org/10 . 7554/eLife . 07727 . 00510 . 7554/eLife . 07727 . 006Figure 2 . Oxygen-induced vessel regression is not enhanced in Wls iEC-KO mice . ( a ) Representative confocal images of Wls iEC-KO and Wls WT P12 mouse retinas under 70% oxygen concentration from P7 until P12 , and labeled for vessel lumen ( Icam2 , red ) and extracellular matrix ( Col . IV , green ) . Quantification shows no significant difference in area of vessel obliteration between Wls iEC-KO and Wls WT mouse . p values from unpaired , two-tailed t-test . Mean +/-SEM; N = 6 mice; 2 litters . ( b ) Representative confocal images of Wls iEC-KO and Wls WT P16 mouse retinas under 70% oxygen concentration from P4 until P6 , and labeled with vessel lumen ( Icam2 , red ) , endothelial cells ( IB4 , blue ) and extracellular matrix ( Col . IV , green ) . Quantification shows no significant difference in area of vessel obliteration between Wls iEC-KO and Wls WT mouse . p values from unpaired , two-tailed t-test . Mean +/-SEM; N = 6 mice; 3 litters . DOI: http://dx . doi . org/10 . 7554/eLife . 07727 . 00610 . 7554/eLife . 07727 . 007Figure 3 . Normal pericytic coverage in Wls iEC-KO . ( a ) Overview of retinal vascular plexus of P6 control and Wls iEC-KO mice , labeled for endothelial cells ( CD31 ) , and pericytes ( Ng2 ) , in capillary plexus ( top panels ) and main vessels ( bottom panels ) . ( b ) Quantification of pericyte coverage of retinal blood vessels , showing no significant change between control and Wls iEC-KO in P6 retinas . p values from unpaired , two-tailed t-test . Mean +/-SEM; N = 4 mice; 2 litters . DOI: http://dx . doi . org/10 . 7554/eLife . 07727 . 007 RT-PCR profiling on RNA extracts from isolated P7 retinal ECs ( Figure 4a ) identified expression of Wnt ligands associated with canonical ( Wnt3 , Wnt3a , Wnt6 , Wnt7b , Wnt9a and Wnt10a ) and non-canonical ( Wnt5a and Wnt11 ) Wnt signalling . Expression of the canonical Wnt/β-catenin-dependent targets Axin2 , CyclinD1 and Lef1 ( Clevers and Nusse , 2012 ) were unaffected in Wls iEC-KO ( Figure 4b , c ) , and nuclear Lef1 levels were even slightly increased ( Figure 4d ) . Intercrossing the canonical Wnt signalling reporter mouse BAT-gal ( Maretto et al . , 2003 ) also revealed no differences in X-gal positive ECs ( Figure 4e ) . Also expression of endothelial Dll4/Notch signalling components , potentially influenced by canonical Wnt signalling ( Corada et al . , 2010 ) , was unaffected ( Figure 4b , c ) . Together , these findings identify that canonical Wnt signalling is intact in Wls iEC-KO , and suggest that the observed increase in regression was possibly due to loss of endothelial non-canonical Wnt signalling . 10 . 7554/eLife . 07727 . 008Figure 4 . Wls iEC-KO show no significant defects in canonical Wnt signalling . ( a ) RT-PCR for all known mouse Wnt ligands using mRNA extracts from isolated retinal endothelial cells of P7 wild-type retinas ( red text represents positive bands ) . ( b ) Semi-quantitative real-time analysis of mRNA expression levels of different genes in P6 Wls iEC-KO retinas normalized to Wls WT retinas from whole retina extracts . p values from unpaired , two-tailed t-test . Mean +/- SD; N = 4 mice; 2 litters . ( c ) Semi-quantitative real-time analysis of mRNA expression levels of different genes in P6 Wls iEC-KO normalized to Wls WT from isolated lung endothelial cells . p values from unpaired , two-tailed t-test . Mean +/-SEM; N = 3 mice; 2 litters . ( d ) Lef1 immunostaining and quantification of fluorescence intensity ( graph right ) in endothelial cells ( IB4 ) from Wls iEC-KO and Wls WT retinas . p values from unpaired , two-tailed t-test . Mean +/- SEM; N = 234 Wls WT cells and N = 128 Wls iEC-KO cells; 4 mice . ( e ) X-gal staining , indicative of canonical Wnt signalling activation , in Wls iEC-KO; BAT-gal and Wls WT; BAT-gal retinas . No correlation was found with X-gal positive cells and regression profiles ( visualized by ECM ( Col . IV ) and lumen ( Icam2 ) stainings ) . DOI: http://dx . doi . org/10 . 7554/eLife . 07727 . 008 Indeed , endothelial-specific Wnt5a inactivation ( Wnt5afl/fl::Pdgfb-iCreERT2 , hereafter Wnt5a iEC-KO ) led to increased vessel regression , decreased vascular density and a mild decrease in radial vascular expansion ( Figure 5a , b ) . Constitutive Wnt11 KO mice showed a milder phenotype with a slight decrease in radial expansion , but no significant differences in vascular density ( Figure 5a , b ) . However , compound Wnt5a endothelial-specific KO and Wnt11 KO mice , named Wnt5a iEC-KO; Wnt11 KO hereafter , largely phenocopied the vascular defects of Wls iEC-KO mice ( Figure 6a , b ) . As in Wls iEC-KO mice , ECs numbers and apoptosis rate were unaffected in Wnt5a iEC-KO; Wnt11 KO ( Figure 6a , b ) . Also the tracheal vasculature , undergoing post-natal remodelling ( Baffert et al . , 2004 ) , showed a significant decrease in vascular density in Wnt5a iEC-KO; Wnt11 KO , and an associated increase in vessel regression events ( Figure 6c , d ) . We conclude that endothelial-derived non-canonical Wnt ligands prevent excessive and premature vessel disconnection . 10 . 7554/eLife . 07727 . 009Figure 5 . Characterization of vascular parameters in Wnt5a iEC-KO and Wnt11 KO retinas . ( a ) Overview of retinal vascular plexus of P6 Wnt5a iEC-KO , Wnt11 KO , and corresponding control mice , labeled with ECM ( Col . IV , grey ) . ( b ) Quantification of different vascular parameters in P6 retinas on the different mouse strains . p values from unpaired , two-tailed t-test . Mean +/-SEM; N = 5 mice; 3 litters . DOI: http://dx . doi . org/10 . 7554/eLife . 07727 . 00910 . 7554/eLife . 07727 . 010Figure 6 . Non-canonical Wnt signalling regulates vessel regression . ( a ) Overview of retinal vascular plexus of control and compound Wnt5a iEC-KO; Wnt11 KO mice , labelled with lumen ( Icam2 ) and ECM ( Col . IV ) markers . ( b ) Quantification of different vascular parameters showing increased vessel regression in Wnt5a iEC-KO; Wnt11 KO retinas . p values from unpaired , two-tailed t-test . Mean +/-SEM; N = 7 mice; 3 litters . ( c ) Overview of trachea vascular plexus of P6 Wnt5a iEC-KO; Wnt11 KO and control mice labelled for CD31 ( green ) . ( d ) Quantification of vessel density and regression profiles in the trachea of Wnt5a iEC-KO; Wnt 11 KO and WT P6 mice . p values from unpaired , two-tailed t-test . Mean +/-SEM; N = 4 mice; 2 litters . DOI: http://dx . doi . org/10 . 7554/eLife . 07727 . 010 We recently showed that EC nucleus-to-Golgi axial polarity predicts migration patterns at sites of vessel regression in vivo , and that differential flow/shear patterns in juxtaposed vessels drive asymmetries in cellular movements , thus promoting stabilization of high-flow and regression of low-flow vessel segments ( Franco et al . , 2015 ) . Further , we have developed a computational approach to simulate blood flow in retinal networks and calculate wall shear stress and flow patterns ( Bernabeu et al . , 2014 ) . Given the known involvement of non-canonical Wnt signalling in Planar Cell Polarity ( PCP ) and cell polarization ( Devenport , 2014; Segalen and Bellaiche , 2009 ) we hypothesized that Wls iEC-KO could have defects in cell polarization . We therefore analysed if non-canonical Wnt signalling could influence coordinated polarization of ECs in vivo and in response to flow . We stained Wls iEC-KO and control mice for Golgi , lumen , and EC nuclei and extracted maps of axial polarity for entire retinal vascular networks ( Franco et al . , 2015 ) , in a novel analysis methodology that we call hereafter Polarity Network ( PolNet ) analysis ( Figure 7a and Figure 7—figure supplement 1 ) . To measure efficiency of endothelial polarization in response to flow , we calculated the angle between the axial polarity vectors and the predicted flow vectors ( Figure 7b ) . Similar to controls , Wls iEC-KO EC cells significantly polarize against blood flow direction across all assessed regions of the network ( Figure 7c ) . Surprisingly , Wls iEC-KO ECs in capillaries , the regions of active remodelling , showed significantly better polarization against the blood flow compared to control cells ( Figure 7d ) . When plotting the percentage of cells polarized within 45 degrees of anti-parallel orientation to flow against the computationally predicted wall shear stress , WT and Wls iEC-KO cells segregated such that Wnt-deficient cells reached 60 percent of cells polarizing already at less than 4 Pa whereas WT cells needed more than 7 Pa of shear to reach 60 percent ( Figure 7e ) . These data suggest that loss of endothelial-derived non-canonical Wnt ligands substantially increases the sensitivity of cells to shear . 10 . 7554/eLife . 07727 . 011Figure 7 . Endothelial-derived Wnt ligands modulate endothelial polarization in response to wall shear stress . ( a ) Example of axial polarity of an Wls iEC-KO P6 retina , labeled for EC nuclei ( Erg ) , lumen ( Icam2 ) and Golgi ( Golph4 ) ( i ) , corresponding image segmentation of the vascular plexus with axial polarity vectors in red ( ii ) , flow pattern simulation of the selected area ( iii ) , and correlation between axial polarity and blood flow direction at the endothelial nuclear position ( iv ) . ( b ) Representation of the principle of angle calculation between axial and flow polarities of endothelial nuclei in ( a ) , highlighting the lumen of blood vessels ( grey ) , and the axial polarity of all ECs ( red arrows ) . ( c ) Analysis of the endothelial axial polarity angle in the main vessels , relative to predicted blood flow direction by the rheology in silico model in Wls WT and Wls iEC-KO mice ( n = 3 retinas ) . ( d ) Quantitative analysis of the percentage of ECs polarized at 180° ( ± 45° ) degrees compared to the flow direction in the different vascular beds of Wls WT and Wls iEC-KO mice ( n = 3 retinas ) . ( e ) Correlative analysis of wall shear stress and EC polarization in the capillary vascular bed of Wls WT and Wls iEC-KO mice . ( f ) Representative graph showing the distribution of scalar products in function to wall shear stress levels for ECs from Wls iEC-KO capillaries . Scalar product corresponds to the product between length of the axial polarity vector and the cosine of the angle between the axial polarity vector and the flow direction vector . ( g ) Linear regression analysis of positive ( polarized with flow ) and negative ( polarized against the flow ) scalar product points for each endothelial cell nucleus . Gradient , R-value and number of cells analyzed for each vascular bed and genotype are shown . N = 3 retinas . DOI: http://dx . doi . org/10 . 7554/eLife . 07727 . 01110 . 7554/eLife . 07727 . 012Figure 7—figure supplement 1 . Endothelial polarization patterns in Wls iEC-KO . ( a ) Image segmentation of the vascular plexus of an Wls iEC-KO mouse retina , highlighting the lumen of blood vessels ( grey ) and the nucleus-to-golgi ( axial ) polarity of all endothelial cells ( red arrows ) . DOI: http://dx . doi . org/10 . 7554/eLife . 07727 . 012 To directly investigate whether non-canonical Wnt ligands regulate the sensitivity of ECs to polarize against the flow , we used a microfluidic device to test endothelial polarization in cultured monolayers exposed to laminar flow induced shear stress ( Ziegler and Nerem , 1994 ) . Intriguingly , siRNA-mediate knockdown of Wnt5a and Wnt11 in HUVECs led to a significant increase in polarization against the blood flow in this reductionist in vitro system suggesting that loss of endothelial Wnts directly affects aspects of endothelial cell biology involved in shear sensing and/or transduction ( Figure 8a , b ) . 10 . 7554/eLife . 07727 . 013Figure 8 . Non-canonical Wnt signalling modulates flow-induced polarity but not flow-induced transcriptional gene expression response . ( a ) Rose-plot representation of axial polarity of ECs treated with Control or Wnt5a and Wnt11 specific siRNAs in static conditions or in response to 2 Pa flow in a microfluidic device ( n = 3 independent experiments ) . Line running from the center represents the Mean of the dataset . Arcs extending to either side represent the 95% confidence limits of the mean ( red means not significantly polarised; black means significantly polarised ) . ( b ) Representative images of endothelial cell polarity in flow chamber stained for nuclei ( Dapi , Blue ) and Golgi apparatus ( GM130 , green ) , and corresponding axial polarity vectors ( black arrows ) . ( c ) Quantitative analysis of EC polarization related to the flow direction in the microfluidic device . p values from non-parametric two-tailed Kuiper’s test . ( d ) Semi-quantitative real-time analysis of mRNA expression levels of different genes in Control or Wnt5a and Wnt11 specific siRNAs in static or stimulated with 2 Pa conditions in a microfluidic device . p values from one-way ANOVA with multiple comparisons . Mean +/-SD; N = 3 independent experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 07727 . 013 In addition to polarizing cells , flow triggers a well-documented transcriptional response of a number of genes including Klf2 , Klf4 , and Ptgs2 ( Hahn and Schwartz , 2009 ) . Moreover , VE-cadherin , VEGFR2 and PECAM1 have been described to act in a complex sensing and relaying flow-mediated shear forces ( Tzima et al . , 2005 ) . To understand whether Wnt affected a more general sensitivity to flow , we therefore studied transcriptional levels following Wnt5a and Wnt11 knockdown . Interestingly , HUVECs depleted of Wnt5a and Wnt11 reacted transcriptionally to flow in the same order of magnitude as control cells for all genes analysed ( Figure 8c ) . These data indicate that the mechanism of sensing and transducing flow signals into axial polarization differs from the mechanism triggering the transcriptional responses , such that endothelial Wnt-ligands only affect the former but not the latter . To understand how increasing shear influences the remodelling process in dependence on Wnt signalling , we injected systemically angiotensin II to increase blood flow and therefore augment wall shear stress levels . Quantification of regression profiles showed an increase in the number of regression profiles in retinas from both Wls WT and Wls iEC-KO mice ( Figure 9a , b ) , with angiotensin II-treated control mice having similar numbers of regression profiles as non-treated Wls iEC-KO mice , with a corresponding decrease in vessel density ( Figure 9b ) . We then used our previously described PolNet analysis to evaluate polarisation patterns of ECs after systemic angiotensin II treatment ( Figure 9c ) . Notably , we observed a significant increase in the polarisation of endothelial cells against the blood flow direction in the capillary network ( Figure 9d ) . Thus , increasing shear forces , or the sensitivity of the endothelium to shear-induced polarization appears to have the same effect on remodelling , and can act synergistically . 10 . 7554/eLife . 07727 . 014Figure 9 . Systemic Angiotensin-II treatment accelerates vessel regression independent of Wnt signalling . ( a ) Overview of retinal vascular plexus of P7 control and Wls iEC-KO mice treated with angiotensin II , labelled for lumen ( Icam2 ) and ECM ( Col . IV ) . ( b ) Quantification of radial expansion , vessel density and regression profiles in angiotensin II-treated and PBS-treated control retinas . p values from one-way ANOVA with Holm-Sidak’s multiple comparison test . Mean +/-SEM; N = 4 mice; 3 litters . ( c ) Analysis of the endothelial axial polarity angle in the main vessels , correlated to predicted blood flow direction by the rheology in silico model in Wls WT and Wls iEC-KO mice treated with angiotensin II . ( d ) Quantitative analysis of the percentage of ECs polarized at 180° ( ± 45° ) degrees compared to the flow direction in the different vascular beds of Wls WT and Wls iEC-KO mice with and without angiotensin II treatment ( n = 3 retinas ) . DOI: http://dx . doi . org/10 . 7554/eLife . 07727 . 014 Finally , we asked whether increasing non-canonical Wnt signalling can lower the sensitivity to shear , and thus prevent vessel regression . Surprisingly , inducible endothelial overexpression of Wnt5a ( hereafter Wnt5a OE ) had no effect on vessel regression and vascular density under normal conditions ( Figure 10a–c ) , illustrating that increased and sustained Wnt5a levels are not sufficient to prevent physiological vessel remodelling . Moreover , Wnt5a OE failed to prevent the angiotensin-II-mediated increase in vessel regression and decrease in vessel density ( Figure 10a–c ) . Thus , our results show that non-canonical Wnt signalling is not used as a mechanism to drive flow-dependent vessel regression , but its absence sensitizes ECs to flow-induced regression . 10 . 7554/eLife . 07727 . 015Figure 10 . Overexpression of Wnt5a in endothelial cells does not inhibit vessel regression . ( a ) Overview of retinal vascular plexus of P6 endothelial-specific Wnt5a GOF and corresponding control mice , labeled for ECM ( Col . IV , green ) , blood vessels ( IB4 , blue ) and vascular lumen ( Icam2 , red ) . ( b ) Higher magnification of retinal vascular plexus of P6 endothelial-specific Wnt5a GOF and corresponding control mice , labeled for ECM ( Col . IV , green ) , blood vessels ( IB4 , blue ) and vascular lumen ( Icam2 , red ) . ( c ) Quantification of vascular parameters showing no significant differences between Wnt5a GOF and corresponding control retinas , with indicated treatments . p values from one-way ANOVA with Holm-Sidak’s multiple comparison test . Mean +/-SEM; N = 4 mice; 2 litters . DOI: http://dx . doi . org/10 . 7554/eLife . 07727 . 015 Our present results show that ECs have intrinsic molecular mechanisms that regulate the flow-dependent shear stress response in vivo , which are important to control cell polarity and vessel regression in vascular remodelling . We show that ECs secrete non-canonical Wnt ligands ( Wnt5a and Wnt11 ) that decrease the capacity of ECs to orientate against the direction of blood flow specifically in vascular capillaries . Interestingly , we found that non-canonical Wnt signalling acts as a permissive rather than an instructive cue . Deleting Wnt5a and Wnt11 did not change the overall pattern of the vascular network , and overexpressing Wnt5a in ECs of mouse retinas did not affect normal levels of vessel regression or vascular morphology . Thus , our results suggest that flow-induced vessel regression governs the overall mechanism that selects which vessel segments are redundant and non-functional to undergo regression , and that a basal level of non-canonical Wnt signalling is present to decrease the sensitivity of ECs to the flow-dependent remodelling program ( Figure 11 ) . 10 . 7554/eLife . 07727 . 016Figure 11 . Working model for non-canonical Wnt signalling regulation of flow-induced remodeling . Initiation of the vascular remodeling program is dependent on the level of wall shear stress ( threshold ) to which ECs robustly polarize against the flow direction . Below the threshold , ECs movements in the immature plexus are balanced by countermovements of adjacent ECs maintaining vessel connections , in a state of dynamic stability . When exposed to shear stress levels above the response threshold , ECs react by polarizing and migrating against the flow direction , triggering vessel regression and vessel remodeling . Signaling pathways influencing the threshold for flow-dependent EC polarization can delay remodeling ( raising the threshold ) or induce premature remodeling ( lowering the threshold ) , as in the case of deficient non-canonical Wnt signaling . DOI: http://dx . doi . org/10 . 7554/eLife . 07727 . 016 Our results go against a recent proposal by Korn et al . , who concluded that non-canonical Wnt signalling was involved in the regulation of ECs survival and apoptosis ( Korn et al . , 2014 ) . We find no evidence for decreased EC proliferation , or increased EC apoptosis upon loss of Wls or Wnt5a/Wnt11 , in both in vivo and in vitro experiments . Korn et al . used TNP-470 to inhibit non-canonical Wnt signalling , a compound which inhibits the broad-spectrum enzyme methionine aminopeptidase-2 , and that has been suggested to also interfere with VEGF signalling , a major regulator of cell proliferation and survival ( Emoto et al . , 2000; Sin et al . , 1997 ) . Our study instead used selective genetic loss-of-function , potentially explaining some of the differences . Our results point to a distinct effect of non-canonical Wnt signalling on EC polarity in response to flow . In the light of recent reports showing that developmental vessel regression is driven by cell migration/rearrangements under the influence of flow ( Chen et al . , 2012; Franco et al . , 2015; Kochhan et al . , 2013; le Noble et al . , 2004; Lenard et al . , 2015; Sato et al . , 2010; Udan et al . , 2013 ) , our observed effects on EC polarity are likely the main driving force for enhanced vessel regression in our mouse mutants . The levels of vessel regression in both wild-type and Wls iEC-KO mice could be manipulated by administration of angiotensin II , a potent vasoconstrictor . Interestingly , angiotensin II-treated wild-type animals showed similar levels of vessel regression as untreated Wls iEC-KO mice , and the same drug could further raise regression levels in Wls iEC-KO mice . This data further suggests that non-canonical Wnt signalling effects in vessel regression are indeed blood flow-dependent . But most importantly , it suggests the existence of a shear stress threshold for the onset of vessel regression . Our quantitative analysis combining EC axial polarity patterns and in silico flow simulations shows that with increasing shear more cells are polarized better , further advocating for the presence of a shear threshold . We propose that non-canonical Wnt signalling regulates this threshold . How mechanistically Wnt5a and Wnt11 controls this threshold remains unresolved . Interestingly , Martin Schwartz group showed that HUVECs have a defined threshold to polarize parallel to flow , and that VEGFR3 acts as a modulator of blood flow shear response by regulating the flow sensor VEGFR2/VEcadherin/Pecam1 ( Baeyens et al . , 2015 ) . Non-canonical Wnt signalling could interact with this pathway and thus impact on the flow shear stress sensor . However , our analysis demonstrated that Wnt5a and Wnt11-depleted HUVECs activate expression of key components in flow sensing ( Klf2 , Klf4 , Ptgs2 ) in the same order of magnitude as control cells , suggesting that non-canonical Wnt signalling modulates the physical reorganization of cell polarity rather than flow sensing itself . The profound motility and rearrangement of ECs in the immature vascular plexus ( Chen et al . , 2012; Franco et al . , 2015; Jakobsson et al . , 2010; Sato et al . , 2010 ) implies that ECs need to coordinate their cellular movements in order to maintain vessel integrity and vessel connections . We propose that the primitive network before flow onset , or at a subthreshold level of flow , is in a state of dynamic stability where the movement of cells is less directional , but balanced by countermovements of adjacent cells such that the network remains open and lumenised . In this context , non-canonical Wnt signalling is likely to facilitate coordinated EC behaviour balancing cell movements in low-flow segments . Flow-induced polarity will supersede this mechanism driving ECs to polarize against the flow direction . It is tempting to speculate that flow-independent cell rearrangements and flow-induced cell movements stand in some form of competition to each other . Whereas the balancing rearrangements act to maintain vessels open even under low-flow conditions , flow-induced polarization introduces a bias in the system that leads to stenosis and regression of low-flow vessel segments . Thus , one could assume that forces or signals that drive cells to maintain the vessel open need to be overcome by the flow induced polarization event . What drives the rearrangements of cells in the primitive plexus and how flow in one segment initiates regression in another is poorly understood . Recent results show that VE-cadherin organizes the junctional and cortical actin cytoskeleton , ( Sauteur et al . , 2014 ) , and that differential VE-cadherin dynamics drive cell rearrangements ( Bentley et al . , 2014 ) . Cells with higher VEGF signalling and lower Notch activity show increased mobility by displaying a larger mobile fraction of VE-cadherin at their junctions ( Bentley et al . , 2014 ) . Whether this also holds true for events during regression is unclear . However , given that Notch is also active in remodelling ( Ehling et al . , 2013; Lobov et al . , 2011 ) , VE-cadherin is a component of EC-to-EC and fluid shear stress force sensing ( Conway et al . , 2013 ) , and that VE-cadherin is implicated in coordinating endothelial polarity in collective migration ( Vitorino and Meyer , 2008 ) , it is tempting to speculate that rearrangements in the primitive plexus involve Notch signalling as a driver of differences in cell motility , and that non-canonical Wnt works as a signalling pathway to balance net movements through coordination of cell cohesion , enabling symmetry of movements . Flow will break this symmetry in the primitive network as it provides an extrinsic directional signal that will polarize EC movements preferentially out of the low-flow segments and into the high-flow segments . The following mouse strains were used: Wlsfl ( Carpenter et al . , 2010 ) ; Pdgfb-iCreERT2 ( Claxton et al . , 2008 ) ; Wnt5afl ( Miyoshi et al . , 2012 ) ; Wnt11- ( Majumdar et al . , 2003 ) ; Bat-Gal ( Maretto et al . , 2003 ) ; R26mTmG ( Muzumdar et al . , 2007 ) ; and Wnt5a GOF ( unpublished , provided by T . Yamaguchi , details will be published in a different report ) . Mice were maintained at the London Research Institute under standard husbandry conditions . Tamoxifen ( Sigma , Germany ) was injected intraperitoneally ( IP ) ( 20 μl/g of 1 mg/mL solution ) at postnatal day 2 ( P2 ) before eyes were collected at P5 onwards . In mosaic recombination experiments tamoxifen was injected ( 20 μl/g of 0 . 04 mg/mL solution ) at P3 before eyes were collected at P6 , as described previously ( Franco et al . , 2013 ) . For EC proliferation assessment in the retina , mouse pups were injected IP 4 hr before collection of eyes with 20 ul/g of EdU solution ( 0 . 5 mg/mL; Thermo Fischer Scientific , Waltham , Massachusetts , USA , C10340 ) . Oxygen-dependent vessel obliteration was achieved using two different regimes of hyperoxia . At P4 ( regime 1 ) or P7 ( regime 2 ) pups were place in 70% oxygen chamber until P6 ( regime 1 ) or P12 ( regime 2 ) . Animals were sacrificed immediately after hyperoxia treatment and processed for retinal vasculature analysis . Angiotensin II ( 10 mg/mL in PBS; Sigma ) was injected IP 10 μL/g daily at P4 , P5 and P6 and pups were culled at P7 . Control mice were injected using PBS alone . Animal procedures were performed in accordance with the Home Office Animal Act 1986 under the authority of project license PPL 80/2391 . The investigators were not blinded to allocation during experiments and outcome assessment and the experiments were not randomized . HUVECs ( PromoCell , Germany ) were routinely cultured in EGM2-Bulletkit ( Lonza , Switzerland ) and mycoplasma tested . For siRNA experiments , HUVECs were transfected with ON-TARGET smart pool control untargeting ( D-001210-02-20 ) and siRNAs against human Wnt5a ( L-003939-00-0005 ) , Wnt11 ( L-009474-00-0005 ) , were purchased from Dharmacon ( Lafayette , Colorado , USA ) . HUVECs were transfected with 25 nM siRNA using the Dharmafect 1 transfection reagent following Dharmacon protocols . Twenty-four hours post transfection HUVECs were plated at confluence in IBIDI slides ( height 0 . 6 mm , IBIDI , Germany ) . Sixteen hours later , unidirectional laminar shear stress ( SS ) was applied using peristaltic pumps ( Gilson , France ) connected to a glass reservoir ( ELLIPSE , France ) and to the IBIDI slide . Local shear stress was calculated using Poiseuille’s law and averaged 2 Pa . Cells were exposed to shear stress for 4 hr using EGM2 media ( Lonza ) and then fixed using 100% cold-methanol for 10 min and washed three times in PBS . Cells were then stained for Golgi ( GM130 , 1/500 , BDBiosciences , Franklin Lakes , NJ ) and nucleus ( DAPI , 1/10000 , Sigma ) . Polarity of cell was evaluated looking at the angle formed by the nucleus-Golgi main axe compared to flow direction . Orientation of the cell was evaluated by looking at the angle of major axe of the nucleus compared to flow direction . For each experiment , five fields containing more than 90 cells have been analysed . Eyes were collected from P5 onwards and fixed with 2% PFA in PBS for 5 hr at 4°C , thereafter retinas were dissected in PBS . Blocking/permeabilisation was performed using Claudio’s Blocking Buffer ( CBB ) ( Franco et al . , 2013 ) , consisting of 1% FBS ( Thermo Fisher Scientific ) , 3% BSA ( Sigma ) , 0 . 5% triton X100 ( Sigma ) , 0 . 01% Na deoxycholate ( Sigma ) , 0 , 02% Na Azide ( Sigma ) in PBS pH = 7 . 4 for 2–4 hr at 4°C on a rocking platform . Primary and secondary antibodies were incubated at the desired concentration in 1:1 CBB:PBS at 4°C overnight in a rocking platform . A list of primary and corresponding secondary antibodies can be found in Supplementary file 1 . Dapi ( Sigma ) was used for nuclei labeling . Retinas were mounted on slides using Vectashield mounting medium ( Vector Labs , H-1000 , Burlingame , CA ) . For imaging we used a Carl Zeiss LSM780 scanning confocal microscope ( Zeiss , Germany ) . Retinas from neonatal P7 mice were dissected in cold PBS and rinsed in PBS . Retinal cells were dissociated with 1 mg/ml Collagenase A ( Roche , Germany , 10103578001 ) , 3 U/ml DnaseI ( Roche ) in DMEM ( Thermo Fisher Scientific ) at 37°C for 30 min and cell suspension was passed through cell strainer . After several washes in PBS , cell suspension was incubated with PE rat anti-mouse CD31 ( BD Biosciences , 553373 ) and APC rat Anti-Mouse CD45 ( BD Biosciences , 561018 ) antibodies on ice for 15 min . Cells were then washed and applied to FACS . RNA from CD31+CD45- cells was extracted using QIAGEN microRNA kit ( Netherlands ) . PCR for the different Wnt ligands were performed with standard PCR protocols using primers listed in Supplementary file 2 . To isolate murine endothelial cells we adapted the protocol described in Sun et al . ( Sun et al . , 2012 ) . Briefly , lungs of P6 mouse pups were removed , minced using forceps , and digested with 1% Collagenase-A supplemented with DNAseI ( 3 U/mL ) in HBSS with calcium/magnesium for 60 min at 37°C . Digested tissues were passed through a 14-gauge needle and then filtrated through a 40-µm cell strainer . The cell suspension was incubated for 45 min at 4°C with 50 µl of magnetic Dynabeads ( Thermo Fisher Scientific ) that had been conjugated overnight with anti–mouse CD31 antibody in PBS/EDTA 2 mM and 2%BSA in a rocking platform . Cells with beads attached were collected using an MPC magnet ( Thermo Fisher Scientific ) and washed 6 to 8 times with PBS/EDTA 2 mM and 2%BSA . Endothelial cell fraction and non-endothelial fraction was then centrifuged and ressuspended in lysis buffer from RNeasy MiniKit ( Qiagen ) and stored at -80°C prior RNA isolation using RNeasy MicroKit ( Qiagen ) . For mouse retina gene expression profiling , eyes were collected and retinas dissected in RNAlater ( Qiagen ) . Retinal RNA extraction was done using RNeasy MicroKit ( Qiagen ) . For HUVEC gene expression profiling , treated or control cells were collected directly in RLT lysis buffer from the RNeasy MicroKit ( Qiagen ) and further processed for RNA isolation . Reverse transcription of mRNA was performed using First-Strand cDNA Synthesis Kit ( Roche ) using the manufacturer recommended protocol . Semi-quantitative real time-PCR was performed using a 7900HT Fast Real-Time PCR System and Taqman gene expression probes ( Applied Biosystems , Thermo Fisher Scientific ) . A list of primers used for gene expression profiles can be found in Supplementary file 3 . Mouse pups eyes at the desired stage were collected in 1% PFA and kept at 4°C for 4 hr . Retinas were dissected and washed twice in PBS . X-gal staining was developed in 2 mM MgCl2 ( Sigma ) , 0 . 01% Na deoxycholate ( Sigma ) and 0 . 02% Nonidet P-40 ( Sigma ) , 5 mM K3Fe ( CN ) 6 ( Sigma ) , 5 mM K4Fe ( CN ) 6 ( Sigma ) , and 0 . 5 mg/mL X-gal ( Promega ) in PBS pH = 7 . 4 at 35°C in a rocking platform . After X-gal staining , retinas were processed as for further antibody stainings . X-gal signal was obtained by exciting X-gal precipitate with the helium–neon laser 633 nm wavelength . Given the complexity and technical aspects of the PolNet Analysis the full details related to the methodology will be published in a separate manuscript . We present here a brief description of our methodology . First , the plexus was manually segmented using Adobe Photoshop ( San Jose , CA ) , producing a binary mask which was subsequently skeletonized using a Voronoi diagram based method ( http://uk . mathworks . com/matlabcentral/fileexchange/27543-skeletonization-using-voronoi ) . The local vessel diameters were calculated using maximum inscribed circles at multiple positions along each vessel segment and this information was used to construct a 3D model of the plexus ( Bernabeu et al . , 2014 ) , code available at https://github . com/UCL/BernabeuInterface2014 ) . The surface defined by this model was used as the input to a Lattice Boltzmann Computational Fluid Dynamics solver , HemeLB ( https://github . com/UCL/hemelb ) , run on a High Performance Computing cluster . The raw fluorescence images were processed in MATLAB using the built-in 'Ginput' function to add points corresponding to the nucleus and Golgi of each cell and recording their locations . These positions defined a vector with magnitude and angle describing the spatial relationship between the points . Pairs of points were recorded for each cell , one at the center of the nucleus and one at the center of the Golgi , defining a vector with a magnitude and angle describing the spatial relationship between the points . The WSS values from the flow simulation were recorded at the positions of the cell nuclei , with the WSS at each point described by a vector giving the magnitude and angle of the applied shear stress . Each plexus was subdivided into artery , vein , capillary and sprouting front regions and each cell assigned to one of these vascular beds . The angular distributions were compared using the Kuiper test ( the circular statistics equivalent of the Kolmogorov Smirnov test ) with each comparison yielding a p-value indicating the likelihood that the two samples are drawn from the same underlying angular distribution . The calculation was performed using the Circular Statistics Toolbox from MATLAB’s FileExchange ( Berens , 2009 ) . In addition we binned the angular data according to WSS magnitude to plot the proportion of cells within 45° of being anti-aligned with the flow as a function of WSS . We calculated the scalar product of the two vectors , given by magnitude ( cell polarity ) magnitude ( WSS ) cos ( theta ) which combines information about the length and relative angles of the vectors . By plotting the scalar product versus WSS , we were able to extract a gradient corresponding to magnitude ( cell ) *cos ( theta ) , i . e . the projection of the cell polarity vector onto the axis defined by the WSS vector . A larger negative gradient corresponds to a larger polarization effect for a given WSS . Complete high-resolution three-dimensional ( 3D ) rendering of whole mount retinas were acquired using a LSM780 laser-scanning microscope ( Zeiss ) . Tiled scans of whole retinas were analyzed with Imaris ( Bitplane , Andor Technology , United Kingdom ) or ImageJ . Radial expansion corresponds to the mean distance from the optic nerve to the edge of the sprouting blood vessels ( 4 measurements per retina were done and averaged ) . Vessel density corresponds to the vascular area ( measured by thresholding isolectin B4 signal in ImageJ ) divided by the total area of vascularized tissue ( 3–5 20x objective images of regions between artery and vein were used per retina ) . Number of branching points was measured by manually quantifying all branching points in 3–5 20x objective images , of regions between artery and vein per retina , and dividing by the total area of vascularized tissue . Regression profiles were manually measured in 3–5 20x objective images and divided by the total area of vascularized tissue . IB4/Col . IV regression profiles correspond to the number of empty basement membrane collagen sleeves , i . e . Col . IV-positive segments negative for IB4 staining . ICAM2/Col . IV regression profiles correspond to number of Col . IV-positive vessel segments and segments negative for ICAM2 staining or presenting a breakage in the continuity of the luminal staining . Sprouting activity corresponds to the number of filopodia bursts , clusters of filopodia emanating from the leading edge , per field of view in 4–6 20x objective images of the sprouting front for each retina . Proliferation of endothelial cells was measure by quantifying the total number of endothelial cell nuclei ( labeled by Erg immunostaining ) positive for EdU staining in 3–5 20x objective images , in regions containing the sprouting front , and dividing by the total area of vascularized tissue . Quantification of apoptosis in regression profiles was measured as the number of regression profiles positive for cleaved caspase-3 and divided by the total number of regression profiles in regions used for quantification , and given as percentage . All statistical analysis was performed using Prism 5 . 0 ( GraphPad ) , Oriana 4 ( Kovach Computing Services ) and Matlab ( Mathworks ) software .	Blood vessels play an essential role in growth and development as they transport many important molecules that help cells to survive . Throughout life , the forces that act on the blood vessels help to remodel the vessel network to ensure that blood gets to the parts of the body that need it . For example , the movement of blood across the surface of the endothelial cells that line the inside of the blood vessels applies a force called “shear stress” to the cells . The endothelial cells respond and adapt to the stress by altering their shape , patterns of gene activity and internal organization ( known as their polarity ) . It was not fully understood exactly how the forces acting on endothelial cells help to remodel the blood vessel network . Franco et al . have now investigated how a signalling pathway known as non-canonical Wnt signalling affects the remodelling of blood vessels in mice , and found that this pathway stabilizes existing connections between vessels . Disrupting non-canonical Wnt signalling , by genetically engineering mice to lack proteins called Wnt5a and Wnt11 , increased the sensitivity of endothelial cells to shear stress . Franco et al . then built a computer model that simulates blood flow and endothelial cell polarity in a network of blood vessels; this enabled them to measure the endothelial cells’ response to blood flow in complex vascular networks . The model was then used to show that endothelial cells lacking non-canonical Wnt signalling are able to reorient and become polarized against the direction of blood flow at lower levels of shear stress . Thus , non-canonical Wnt signalling helps to raise the threshold of shear stress above which endothelial cells change their properties . Further work is now needed to identify how non-canonical Wnt signalling interferes with the ability of the endothelial cells to sense shear stress levels .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "developmental", "biology", "cell", "biology" ]	2016	Non-canonical Wnt signalling modulates the endothelial shear stress flow sensor in vascular remodelling
Phosphorylation of histone H3 threonine 118 ( H3 T118ph ) weakens histone DNA-contacts , disrupting the nucleosome structure . We show that Aurora-A mediated H3 T118ph occurs at pericentromeres and chromosome arms during prophase and is lost upon chromosome alignment . Expression of H3 T118E or H3 T118I ( a SIN mutation that bypasses the need for the ATP-dependent nucleosome remodeler SWI/SNF ) leads to mitotic problems including defects in spindle attachment , delayed cytokinesis , reduced chromatin packaging , cohesion loss , cohesin and condensin I loss in human cells . In agreement , overexpression of Aurora-A leads to increased H3 T118ph levels , causing cohesion loss , and reduced levels of cohesin and condensin I on chromatin . Normal levels of H3 T118ph are important because it is required for development in fruit flies . We propose that H3 T118ph alters the chromatin structure during specific phases of mitosis to promote timely condensin I and cohesin disassociation , which is essential for effective chromosome segregation . The packaging of the eukaryotic genome into chromatin facilitates the temporal and spatial regulation of all genomic activities , including DNA repair , replication , transcription and mitosis . Chromatin comprises arrays of nucleosomes , where each nucleosome has ~146 base pairs of DNA wrapped 1 . 75 times around a histone octamer composed of two molecules each of core histone H3 , H4 , H2A , and H2B ( Kornberg , 1974 ) . Repetitive arrays of nucleosomes are then further compacted by higher-order folding , requiring additional proteins including linker histones . During mitosis , chromosome condensation plays a critical role in preventing DNA breaks during mitosis and enabling equal chromosome segregation to the two daughter cells ( Ganem and Pellman , 2012 ) . One important means by which the cell achieves accurate regulation of genomic processes , including mitosis , is via post-translational modifications ( PTMs ) of the core histones ( Strahl and Allis , 2000 ) . The PTMs , usually occurring on the N- and C-terminal tails of the histones , generally serve to recruit reader proteins to the chromatin . PTMs also occur on the histone globular domains , but are much less well studied than the histone tail modifications . PTMs at the histone-DNA interface have been proposed to directly modulate nucleosome structure , without the need for reader proteins ( Cosgrove et al . , 2004 ) . Of all the histone PTMs that occur at the histone-DNA interface , one of the best positioned to disrupt the nucleosome structure is phosphorylation of threonine 118 ( T118ph ) of H3 ( Mersfelder and Parthun , 2006 ) . In agreement with its important location within the nucleosome structure ( Figure 1A ) , biochemical studies have confirmed that H3 T118ph causes reduced nucleosome stability , increased nucleosome mobility , and increased DNA accessibility ( North et al . , 2011 ) . Strikingly , H3 T118ph caused the formation of novel populations of alternate DNA-histone complexes that have DNA wrapped around two complete histone octamers arranged edge-to-edge , termed nucleosome duplexes and altosomes ( North et al . , 2014 ) . In agreement with the biochemical data , a substitution of H3 T118 for isoleucine ( T118I ) was identified in S . cerevisiae as a dominant Swi-INdependent ( SIN ) ( Kruger et al . , 1995 ) . The SIN H3 T118I substitution allows nucleosomes to slide along the DNA without the need for SWI/SNF ( Muthurajan et al . , 2004 ) . 10 . 7554/eLife . 11402 . 003Figure 1 . Dynamic mitotic phosphorylation of H3 T118 . ( A ) The side chain of H3 T118 ( red ) is close enough to form a hydrogen bond with the DNA ( grey ) . Histone H3 is depicted in dark blue , Histone H4 is cyan , Histone H2A is green and H2B is yellow . Angstrom distances were drawn using nearest neighbor wizard in pymol . Protein Data Bank ( PDB ) code 1KX5 . ( B ) The indicated amounts of the respective peptides were dotted and the membrane probed with an antibody to histone H3 T118ph . The UnM T118 peptide corresponds to human histone H3 aa 115 to 125 . ( C ) Western blot of crude extract from HeLa cells , using infra-red labeled secondary antibodies . H3 T118ph ( greyscale/red ) and N-term histone H3 ( green ) . ( D ) HeLa cell extracts untreated or treated with phosphatase inhibitor were probed with the indicated antibodies . Full western blot image can be found in Figure 1—figure supplement 1A . ( E ) HeLa cells were synchronized by a double thymidine arrest and released at the indicated times , followed by western blot analysis of whole cell extracts . ( F ) Immunoprecipitation ( IP ) using the H3 T118ph antibody from HeLa cells asynchronous ( Asynch ) or released from a G2 arrest ( with 9 µM Ro-3306 for 16 hr ) for 30 min resulting in pro-metaphase cells ( Pro-M ) . Full western blot image can be found in Figure 1—figure supplement 1B . ( G ) Immunofluorescence analysis of H3 T118ph ( green ) and α-tubulin ( red ) in HeLa cells . Scale bar = 5 μm . ( H ) H3 T118ph antibody was pre-incubated with no peptide ( top ) , H3 phosphorylated at T118 ( middle ) or unmodified ( UnM T118 , bottom ) . The supernatants were used to detect H3 T118ph in pro-metaphase HeLa cells . Scale bar = 5 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 11402 . 00310 . 7554/eLife . 11402 . 004Figure 1—figure supplement 1 . Full size western blots of data shown in Figure 1 . ( A ) Full image of western blot probed with H3 T118ph from Figure 1D . ( B ) Full image of immunoprecipitation analysis western blot probed with antibody to histone H3 for Figure 1F . labels are abbreviated the same as in Figure 1F . The “” marks the non-specific IgG band . DOI: http://dx . doi . org/10 . 7554/eLife . 11402 . 004 Despite the striking biochemical effects of H3 T118ph on nucleosome structure and the phenotype of the yeast T118I mutant , H3 T118ph has not been studied in cells beyond its identification ( Olsen et al . , 2010 ) . Accordingly , we characterized H3 T118ph function in metazoan cells . H3 T118ph , mediated by Aurora-A , is localized to centromeres and chromosome arms during specific phases of mitosis , Mutation of H3 T118 caused a wealth of defects including lagging chromosomes , delayed cytokinesis , reduced cohesion and altered chromosome compaction in mammalian cells and inviability in Drosophila . Given that the H3 T118I mutant or overexpression of Aurora-A led to premature release of cohesin and condensin I from chromosomes , we propose that H3 T118ph alters chromosome structure during mitosis to help dissociate cohesion and condensin I . To characterize the spatiotemporal localization of H3 T118ph ( Figure 1A ) , we first established the specificity of the H3 T118ph antibodies . Here we show only the results obtained with the Abcam antibody , although similar results were obtained with our independently generated H3 T118ph polyclonal antisera ( data not shown ) . The antisera were highly specific in dot-blot assays ( Figure 1B ) and recognize a single protein identical in size to histone H3 in western blot analysis of total protein extracts from HeLa cells ( Figure 1C ) . This signal in western blots was increased by treating the cells with the protein phosphatase 1 and 2A inhibitor calyculin A for 3 hr , indicating that the H3 T118ph antibody recognized phosphorylated H3 ( Figure 1D , Figure 1—figure supplement 1A ) . In concordance with previously published mass spectrometry results ( Olsen et al . , 2010 ) , we observed a dramatic increase in H3 T118ph levels as cells entered mitosis ( Figure 1E ) . The antibody also recognized H3 T118ph in its native conformation , because it immunoprecipitated H3 from cells released into mitosis ( Figure 1F , Figure 1—figure supplement 1B ) . Using immunofluorescence analysis , we found that H3 T118ph was restricted to mitotic cells during prophase through anaphase and was greatly diminished in interphase ( Figure 1G ) . Specifically , H3 T118ph signal was detected as discrete foci on chromatin only in prophase and pro-metaphase . Additionally , H3 T118ph co-localized with centrosomes through all phases of mitosis ( Figure 1G ) . This is a consequence of non-chromatin bound histones localizing to the centrosomes for proteasome-mediated degradation during mitosis ( C . Wike and J . K . Tyler , manuscript submitted ) . During anaphase , the H3 T118ph antibodies also detected the spindle mid-body ( Figure 1G ) . The localization pattern of H3 T118ph was not unique to HeLa cells , nor cancer cell lines , because it was similar in HMEC , WI-38 and MCF10A cells ( data not shown ) . Finally , the H3 T118ph signal was specifically competed away by an H3 T118ph peptide ( Figure 1H ) . Together , these results show that the H3 T118ph antibody is specific . Threonine 118 and the surrounding residues are highly conserved among metazoan H3 proteins . Therefore , we tested whether H3 T118 is phosphorylated in other metazoans and whether this occurs specifically during mitosis . In D . melanogaster S2 cells , H3 T118ph localized to chromatin and centrosomes during mitosis ( data not shown ) . H3 T118ph localization was also conserved in C . elegans . During pro-metaphase , H3 T118ph was localized along the outside edges of chromosomes , indicative of centromeric localization on holocentric chromosomes in C . elegans ( Figure 2A ) . To determine if the localization of H3 T118ph along the arms of chromosomes was dependent on the centromeric chromatin structure , we used siRNA to the centromeric histone variant CENP-A to abolish the centromeres . Upon CENP-A knockdown , H3 T118ph is diminished from the chromatin ( Figure 2A ) . These data demonstrate that mitotic enrichment of H3 T118ph is conserved amongst metazoans . Furthermore , H3 T118ph localizes to centromeres and its localization is dependent on intact centromeres . 10 . 7554/eLife . 11402 . 005Figure 2 . H3 T118ph localizes to pericentromeres and chromosome arms during prophase and pro-metaphase . ( A ) Immunofluorescence of two-cell C . elegans embryos Control ( RNAi ) ( top ) and centromeric protein A CENP-A ( RNAi ) -depleted ( bottom ) embryos were fixed and stained with α-tubulin ( green ) and H3 T118ph ( red ) antibodies . DNA was stained with DAPI ( blue ) . Scale bar = 5 μm . ( B-E ) Immunofluorescence of HeLa cells stained with CENP-A ( red ) and H3 T118ph ( green ) antibodies . ( B ) Images of progressive mitotic stages . ( C ) Mitotic spreads synchronized with colcemid ( no tension across the kinetochores ) . The white box indicates magnified area . Intensity of the signal across centromeres is plotted . Scale bar = 5 μm . ( D ) Unsynchronized mitotic spread , as in C . ( E ) Extended metaphase chromatid fibers showing H3 T118ph localization to discrete regions of chromosome arms . DOI: http://dx . doi . org/10 . 7554/eLife . 11402 . 005 Given our results in C . elegans , we asked if the punctate chromosomal staining of H3 T118ph in mammalian cells ( Figure 1G ) reflects centromeric staining . Indeed , we found that H3 T118ph co-localized with CENP-A in human cells in prophase and pro-metaphase ( Figure 2B ) . Noteworthy , the prophase to metaphase timing of the appearance and disappearance of H3 T118ph on centromeres is distinct from other mitotic H3 phosphorylation events . For example , H3 S10ph ( Crosio et al . , 2002 ) and H3 T3ph ( Polioudaki et al . , 2004 ) remain on chromosome arms and centromeres , respectively , through anaphase . CENP-A S7ph ( Zeitlin et al . , 2001 ) remains through metaphase , while H3 T118ph is lost from centromeres in metaphase coincident with chromosome alignment . H3 T118ph foci did not always perfectly colocalize with CENP-A , but sometimes appeared to be adjacent to CENP-A foci . Indeed , detailed inspection of mitotic spreads revealed that H3 T118ph localized to the inner centromere when cells were treated with the microtubule destabilizing drug colcemid while CENP-A remained on the outer-centromere ( Figure 2C ) . The distinct localization of CENP-A and H3 T118ph emphasizes that the H3 T118ph signal is not due to phosphorylation of CENP-A per se . Importantly , upon colcemid treatment , where microtubule attachment is lost , the interkinetochore distance is decreased ( Uchida et al . , 2009 ) . The single foci that is detected by H3 T118ph antibody could be because of a single population of H3 T118ph localized to the inner-centromere or because the two adjacent centromeres are separated by less than the resolution of light microscopy , which is theoretically 200nm . Therefore we asked when there is dynamic microtubule attachment , promoting tension and kinetochore stretching , what is the true H3 T118ph signal at the centromere . H3 T118ph existed in two foci per chromosome that correlate well with , but are larger than , the two CENP-A foci ( Figure 2D ) consistent with pericentromeric localization . We noted that H3 T118ph also appears to occur along the chromosome arms in mitotic spreads ( Figure 2D ) . Indeed , H3 T118ph was detectable at the centromere partially overlapping with CENP-A and at weaker foci at discrete intervals along the chromosome arms on extended chromatin fibers ( Figure 2E ) . To gain insight into the function of H3 T118ph , we sought to identify the kinase responsible for its phosphorylation . We utilized a ProQinase kinase screen to test 190 recombinant kinases for their ability to phosphorylate H3 T118 ( Figure 3—figure supplement 1 ) . Aurora-A was the only cell cycle regulated kinase able to efficiently phosphorylate H3 T118 from among the three positive kinases , arbitrarily defined as having an activity above 3000 cpm . Absence of phosphorylation of H3 T118 by Aurora-B and Aurora-C further validated Aurora-A as the kinase of H3 T118 ( Figure 3A ) . We independently confirmed that Aurora-B INCENP could not phosphorylate the T118 peptide , including an H3 S10 peptide as a positive control ( data not shown ) . Two different inhibitors to Aurora-A eradicated the H3 T118ph signal ( Figure 3—figure supplement 2A ) . Upon Aurora-A knockdown , which eradicated most of the Aurora-A protein and activity ( Figure 3—figure supplement 2B , Figure 3—figure supplement 3A ) , H3 T118ph was undetectable on chromatin ( Figure 3B ) . In agreement with Aurora-A being the bona fide H3 T118 kinase , knockdown of TPX2 , a known activator of Aurora-A ( Kufer et al . , 2002 ) , greatly reduced H3 T118ph ( Figure 3—figure supplement 3B , C ) . Taken together , these results demonstrate that Aurora-A mediates H3 T118 phosphorylation . 10 . 7554/eLife . 11402 . 006Figure 3 . Aurora-A phosphorylates H3 T118 and mutations that mimic T118 phosphorylation cause mitotic defects . ( A ) In vitro kinase activity of Aurora-A , -B , -C for H3 T118 peptide . ( B ) Immunofluorescence of pro-metaphase HeLa cells cotransfected with H2B:RFP and siRNA to Aurora-A ( bottom ) or control scrambled siRNA ( top ) . Scale bar = 5 μm . ( C ) Cytokinesis in 293TR cells transiently transfected with H3-YFP plasmids . YFP ( yellow ) and DNA stained with DAPI ( blue ) . Scale bar = 5 μm . ( D ) Quantitation of C ( n=30 cells in anaphase , p=0 . 01 , by Fishers exact test ) . Error bars represent SD of the mean ( SDM ) . ( E ) Quantitative data of live cell imaging showing differences in average length in cytokinesis during live cell imaging ( n = 50 cells , p<0 . 01 and *p<0 . 001 by unpaired student t-test ) . ( F ) Error correction assay for 293TR stable cell lines expressing H3 . Inhibition of Aurora-B with ZM447439 represents an extreme case of inability to correct error . Scale bar =10 μm ( G ) Quantitation of cells with misaligned chromosomes on the metaphase plate as in F . ( p<0 . 05 and *p<0 . 001 by Fishers exact test ) . Error bars represent SDM . DOI: http://dx . doi . org/10 . 7554/eLife . 11402 . 00610 . 7554/eLife . 11402 . 007Figure 3—figure supplement 1 . Results of in vitro kinase screen on peptide spanning H3 T118 . Kinase screen was performed by ProQinase . Kinase activity values ( in cpm , corrected for peptide background ) of 190 Ser/Thr kinases performed with 1 µM biotinylated peptide containing amino acids 112 to 123 of H3 . The Y-axis set to zero . DOI: http://dx . doi . org/10 . 7554/eLife . 11402 . 00710 . 7554/eLife . 11402 . 008Figure 3—figure supplement 2 . Aurora-A inhibitors lead to decreased H3 T118ph . ( A ) Aurora-A inhibitors lead to decreased H3 T118ph . Asynchronous HeLa cells were treated with or without Mln8237 or Vx680 and immunostained with antibodies to H3 T118ph ( green ) , and CENP-A ( red ) and DNA was stained with DAPI ( blue ) . Representative pro-metaphase cells are shown . Scale bar = 5 μm . ( B ) Test of efficiency of the Aurora-A knockdown for experiments shown in Figure 3 onwards . Whole cell extracts were resolved by SDS-PAGE and analyzed by western blot with antibodies to Aurora-A and GAPDH . Quantitation of Aurora-A protein levels normalized to GAPDH . Shown is the average and standard deviation from three independent experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 11402 . 00810 . 7554/eLife . 11402 . 009Figure 3—figure supplement 3 . Knockdown of TPX2 leads to reduced H3 T118ph . ( A ) Control for showing that Aurora-A knockdown worked to decrease phosphorylation of a known substrate . HeLa cells were cotransfected with H2B:RFP and siRNA to Aurora-A or control scrambled siRNA . Coverslips were collected 72 hr post transfection and immunostained with primary antibody Aurora-A T288ph ( green ) . Scale bar = 5 μm . ( B ) Knockdown of TPX2 leads to reduced H3 T118ph . HeLa cells were cotransfected with H2B:RFP and siRNA to Aurora-A or control scrambled siRNA . Coverslips were collected 72 hr post transfection and immunostained with primary antibody Aurora-A T288ph ( green ) . Scale bar = 5 μm . ( C ) Knockdown of TPX2 reduces Aurora-A activity in pro-metaphase . Knockdown was performed as in B . Representative images of immunostained with primary antibody to Aurora-A T288ph ( red ) are shown . Scale bars = 5 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 11402 . 00910 . 7554/eLife . 11402 . 010Figure 3—figure supplement 4 . Characterization of transient transfections and stable cell lines of wild type and mutant H3 . ( A ) Consistent level of expression of YFP tagged wild type and mutant H3 . Western blot analysis of whole cell extract from transient transfections following live cell imaging of H3 WT:YFP and H3:YFP mutants . The blot was probed with a GFP specific antibody . Histone H3 and GAPDH are shown as loading controls . ( B ) Time spent in mitosis is not significantly affected by expression of H3 or T118 mutants . Quantitation of the duration pro-metaphase to an anaphase of transiently transfected H3 . 2:YFP , H3 . 2 T118E:YFP , and H3 . 2 T118I:YFP taken from analysis of live cell imaging of YFP . The average and standard deviation of three independent experiments is shown . ( C ) Time spent in mitosis is not significantly affected by addition of H3 or T118 mutants . Quantitative data of live cell imaging of cells progressing through cytokinesis with a lagging chromosome . The average length of cytokinesis for H3:YFP , H3 T118E:YFP , and H3T118I:YFP is marked by the horizontal line . Differences were not statistically significant . ( D ) Equal expression of the FLAG tagged H3 wild type and mutant constructs . Western blot analysis of whole cell extract from 293TR cells stably expressing wild type and mutant H3:FLAG . The blot was probed for histone H3 to detect endogenous H3 and H3:FLAG . GAPDH is shown as a loading control . ( E ) Quantitation of FLAG tagged H3 compared to endogenous H3 . The average and standard deviation of three independent experiments is shown . DOI: http://dx . doi . org/10 . 7554/eLife . 11402 . 01010 . 7554/eLife . 11402 . 011Figure 3—figure supplement 5 . FLAG-tagged wild type and mutant H3 are equally incorporated into chromatin . ( A ) FLAG-tagged wild type and mutant H3 are equally incorporated into chromatin . Chromatin fractionation following mitotic shake off of 293TR stable cell lines expressing wild type and mutant H3:FLAG . The blot was probed with α-FLAG to detect tagged H3; histone H3 and GAPDH were used as fractionation controls . “Supernatant” contains the soluble proteins while “pellet” contains the insoluble proteins including those on chromatin . ( B ) Analysis of distribution of cells in different phases of mitosis upon expression of FLAG-tagged wild type and mutant H3 . Stable cell lines of histone H3 WT:FLAG ( WT ) , H3 T118A:FLAG ( TA ) , H3 T118E:FLAG ( TE ) , or H3 T118I:FLAG ( TI ) were grown on coverslips and arrested in Ro-3306 inhibitor for 24 hr . Coverslips were collected at the times listed following release into fresh DMEM . At each time point the cells were scored for the phases of the cell cycle phase based on DAPI DNA stain ( n=300 mitotic cells for each mutant , per time point , collected over 3 experiments ) . ( C ) Expression of wild type and mutant H3 does not make prophase longer . Quantitation of stable cell lines in pro-metaphase or metaphase upon release from monastrol with arrest in MG132 for 2 hr ( n=75 cells , collected over 3 experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 11402 . 01110 . 7554/eLife . 11402 . 012Figure 3—figure supplement 6 . Overview of the Drosophila system expressing wild type and mutant H3 proteins . ( A ) Overview of the Drosophila system used to replace all H3 with exogenous wild type or mutant H3 expressed from 12 transgenes . Schematic representation of histone gene organization in Drosophila melanogaster . Each histone gene repeat unit contains a single His1 ( red ) , His2B ( blue ) , His2A ( yellow ) , His4 ( aqua ) and His3 ( purple ) gene , which is repeated approximately 100 times on chromosome 2 . Transgenes carrying three histone gene units were added one at a time into phiC31 recombination sites on the left and right arms of chromosome 3 . This supplies 12 copies of each histone gene to rescue the ∆HisC deletion . ( B ) Stage of development at which lethality occurred due to replace all endogenous H3 with exogenous wild type or mutant H3 expressed from 12 transgenes . DOI: http://dx . doi . org/10 . 7554/eLife . 11402 . 012 Given that H3 T118ph is detectable on chromatin during early mitosis ( Figure 1G , 2B ) , we investigated whether H3 T118ph plays a role in mitotic progression . To do this , we mutated T118 to alanine to prevent its phosphorylation . This serves as a negative control that is not expected to yield a phenotype , because there is still phosphorylation of the endogenous H3 . We also mutated T118 to glutamic acid ( E ) , although this mutation does not cause the nucleosome destabilization or altered nucleosome structures that result from T118 phosphorylation in vitro ( North et al . , 2011 , North et al . , 2014 ) . As such , T118E is not an effective mimic of T118 phosphorylation , at least on mononucleosomes in vitro . We also mutated H3 T118 to isoleucine ( I ) to recapitulate the yeast sin mutant . Transient transfection of HEK 293TR cells with plasmids expressing histone H3:YFP where T118 was mutated to E or I led to a significantly increased incidence of lagging chromosomes ( Figure 3C , D ) . Equal expression of the wild type and mutant H3 proteins was verified by western blot analysis ( Figure 3—figure supplement 4A ) . Using time-lapse microscopy , we found that cells expressing H3 T118E:YFP or H3 T118I:YFP had significant delays in cytokinesis ( Figure 3E , Figure 3—figure supplement 4B , see Materials and methods ) . Furthermore , whenever a lagging chromosome was evident , there also was an accompanying delay in the subsequent cytokinesis , regardless of the transfected construct ( Figure 3—figure supplement 4C ) . From these experiments , we conclude that expression of H3 T118I and T118E results in lagging chromosomes that delay cytokinesis . An increase in lagging chromosomes is symptomatic of defects in chromosome congression ( Thompson and Compton , 2011 ) . This prompted us to investigate if phosphorylation of H3 T118 plays a role in correction of chromosome alignment errors , using an error correction assay ( Lampson et al . , 2004 , Santaguida et al . , 2010 ) . For this , we created a panel of stable 293TR cell lines expressing FLAG-tagged wild type H3 , H3 T118E , T118I , or T118A from the same locus . All the H3:FLAG proteins were expressed to equivalent levels , at approximately 10% of the endogenous H3 level ( Figure 3—figure supplement 4D , E ) and all could be incorporated into chromatin ( Figure 3—figure supplement 5A ) . We further verified that the H3 T118 mutations in H3:FLAG did not cause a delay in prophase to anaphase ( Figure 3—figure supplement 5B ) . The error correction assay was as follows: Monastrol was used to induce a monopolar spindle and kinetochore-microtubule attachment errors ( Figure 3F ) . The cell lines were able to recover by washing out Monastrol if proper checkpoints and machinery are in place and the chromosomes will attach to bipolar spindles . Additionally , cells were released in the presence of MG132 to allow time to align the chromosomes to the metaphase plate by preventing cells from entering anaphase . Importantly , the H3 T118 mutations did not delay release from the pro-metaphase arrest ( Figure 3—figure supplement 5C ) . Expression of either H3 T118E or T118I significantly decreased the ability to align chromosomes compared to wild type H3 or T118A ( Figure 3F , G ) . This result suggests that an over abundance of H3 T118E and T118I mutants may hinder chromosome congression . Our results suggest an important role for phosphorylation of H3 T118 in regulating chromosomal dynamics in metazoans . However , these studies were performed in a situation where only 10% of the histone H3 was mutant . In order to examine the consequences of having all or none H3 phosphorylated on T118 , we introduced the T118 mutations into 12 copies of the H3 gene on transgenes and introduced them into Drosophila where the endogenous H3 gene copies were deleted ( Figure 3—figure supplement 6A ) , such that the flies only expressed H3 T118A , T118E , or T118I ( Gunesdogan et al . , 2010 ) . While control animals bearing wild type H3 survived to adulthood , animals expressing the mutant H3 T118A , E and I died as embryos after depletion of the maternal contribution of histones ( Figure 3—figure supplement 6B ) . These results indicate that normal levels of H3 T118ph are essential for development . Having found that phosphorylation of H3 T118 was essential for development in fruit flies , we sought to gain a better understanding of its function . Since cells expressing H3 T118I and T118E showed reduced chromosome congression , we asked if H3 T118ph remains at centromeres of misaligned chromosomes as the cells enter metaphase . Caffeine was used to induce misaligned chromosomes ( Katsuki et al . , 2008 ) . H3 T118ph remained at centromeres of misaligned chromosome along with the spindle assembly checkpoint ( SAC ) kinase BubRI , even in metaphase ( Figure 4A ) . This suggests that removal of H3 T118ph is triggered by chromosome alignment and led us to speculate that H3 T118ph plays a role in achieving efficient chromosome attachment . Accordingly , we investigated the potential molecular reasons for the defect in chromosome congression caused by H3 T118E and T118I . Outer-kinetochore proteins , spindle assembly checkpoint proteins and the heterochromatin landscape were indistinguishable between cells expressing H3 T118A , T118E , T118I or wild type H3 ( data not shown ) . Taken together , these data show that misaligned chromosomes in H3 T118I and T118E mutants are capable of forming proper kinetochores and recruiting SAC proteins10 . 7554/eLife . 11402 . 013Figure 4 . H3 T118I , T118E and Aurora-A overexpression lead to premature loss of cohesion . ( A ) Immunofluorescence of HeLa cells representing pro-metaphase ( top panel ) , metaphase ( middle panel ) and caffeine-treated ( bottom ) . The primary antibodies used were histone H3 T118ph ( green ) , BubR1 ( red ) and DNA was stained with DAPI ( blue ) . Scale bar = 5 μm . ( B ) Chromosome spreads of H3 T118 mutant cell lines following the error correction assay either untreated ( left ) or monastrol ( middle ) then released into MG132 ( right ) . Scale bar = 5 μm . ( C ) The degree of cohesion loss for Monastrol ( - ) and Monastrol washout MG132 ( + ) treatments were scored from B . ( n=100 cells per treatment collected over 3 experiments , p<0 . 01 and *p<0 . 001 by unpaired student t-test ) . Error bars represent the SDM . ( D ) Chromosome spreads of 293TR cell lines with over expression of Aurora-A or Aurora-A KD . The primary antibodies used were against CENP-A ( magenta ) , H3 T118ph ( green ) , and DNA was stained with DAPI ( blue ) . Scale bar = 5 μm . ( E ) Quantitation of Fig . S4C colcemid pro-metaphase arrest ( Pro-M ) . ( n=100 cells per treatment , collected over 3 experiments p<0 . 001 by unpaired student t-test ) . Error bars represent the SDM . DOI: http://dx . doi . org/10 . 7554/eLife . 11402 . 01310 . 7554/eLife . 11402 . 014Figure 4—figure supplement 1 . Metaphase spreads of 293TR stable cell lines expressing wild type H3 or mutant H3 proteins , to demonstrate cohesion defect upon prolonged pro-metaphase arrest . The uncropped images of chromosome spreads in Figure 4B . The white boxes indicate the magnified image . Scale bar = 5 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 11402 . 01410 . 7554/eLife . 11402 . 015Figure 4—figure supplement 2 . Characterization of 293TR stable cell lines expressing wild type Aurora-A:FLAG and Aurora-A Kinase Dead:FLAG . ( A ) Demonstration of equivalent expression of exogenous Aurora-A and kinase dead Aurora-A . Total protein extracts from asynchronous and pro-metaphase ( Pro-M ) mitotically arrested 293TR , Aurora-A:FLAG , and Aurora-A KD:FLAG cells were resolved by SDS-PAGE and analyzed by western blot with antibodies to FLAG , Aurora-A , and GAPDH is used as a loading control . Samples were collected for flow cytometry analysis at the same time . The G2/M population is listed as a percentage of the mitotic population . ( B ) Aurora-A overexpression leads to cohesion defects . Representative chromosome spreads displaying cohesion defects for the control cell lines and overexpressed Aurora-A cell lines untreated ( left ) , arrested with colcemid ( right ) . DNA stained with DAPI ( blue ) . Scale bar = 5 μm . The white boxes indicate the magnified image area . After mitotic shake-off to remove any cells in mitosis , cells were treated with colcemid and chromosome spreads were prepared . DOI: http://dx . doi . org/10 . 7554/eLife . 11402 . 015 Because sister chromatid cohesion is important for chromosome congression , we examined whether the H3 T118I or T118E mutations caused faulty cohesion . Mitotic spreads from cells expressing wild type H3 or H3 T118A upon pro-metaphase arrest ( induced by Monastrol ) and metaphase arrest ( Monastrol arrest released into MG132 ) mostly had closely associated sister chromatids with 'closed' or 'open' arms ( Figure 4B , C , Figure 4—figure supplement 1 ) . By contrast , H3 T118E or H3 T118I caused a higher incidence of chromosomes with “partially separated” arms , indicating loss of arm cohesion and partial loss of centromeric cohesion ( Figure 4B , C ) . Similar defects in cohesion were observed for cells expressing T118E and T118I upon pro-metaphase arrest with the microtubule destabilizing drugs nocodazole and colcemid ( data not shown ) . The loss of cohesion was most pronounced for H3 T118I , where partially separated sister chromatids were predominant in 50% of the cells versus 10% of the cells for wild type H3 ( Figure 4C ) . Furthermore , the proportion of cells where most of the sister chromatids were totally separated , indicating complete loss of cohesion , was 16% for H3 T118I versus 4% for wild type H3 ( Figure 4C ) . These data indicate that H3 T118E or H3 T118I promotes loss of cohesion at the centromere and chromosome arms . Given the correlation between faulty cohesion and chromosome alignment defects , we propose that the faulty cohesion caused by expression of H3 T118I or T118E is responsible for the defects in chromosome alignment . Aurora-A overexpression has been linked to aneuploidy and cancer , presumably through its role in centrosome duplication . Aurora-A overexpression has not been linked to cohesion loss previously , but this could provide an alternate explanation for aneuploidy . Therefore , we made isogenic cell lines overexpressing Aurora-A and kinase dead Aurora-A . The cell lines had equal expression of Aurora-A ( Figure 4—figure supplement 2A ) and proceeded relatively normally through the cell cycle ( data not shown ) . Overexpression of Aurora-A increased levels of H3 T118ph along the chromosome arms ( Figure 4D ) . We asked if overexpression of Aurora-A recapitulates the loss of cohesion caused by expression of H3 T118I and T118E . Upon colcemid-induced pro-metaphase arrest , overexpression of Aurora-A caused 44% of the sister chromatids to be “partially separated” as compared to 20% for the control cell line ( Figure 4E , Figure 4—figure supplement 2B ) . Because overexpression of Aurora-A leads to cohesion loss , it is likely that cohesion loss in the H3 T118E and T118I mutants is due to their structurally mimicking elevated levels of H3 T118ph . Because cohesion defects can be caused by altered chromatin integrity , we measured the length and width of chromosome one from each H3 mutant . We identified chromosome one by using a special DAPI-treatment protocol to highlight the large pericentromeric heterochromatin cluster ( Figure 5—figure supplement 1A ) . Expression of H3 T118E and T118I made chromosome one significantly wider and shorter ( Figure 5A , Figure 5—figure supplement 1B , C ) . To investigate centromere integrity in the T118 mutants , we measured the sister chromatid interkinetochore distance in chromosome spreads collected after arrest in metaphase . We found that H3 T118E and T118I significantly increased sister chromatid interkinetochore distances ( Figure 5B ) , as measured by immunostaining for CENP-A ( Figure 5—figure supplement 1D ) . 10 . 7554/eLife . 11402 . 016Figure 5 . Altered chromosomal compaction due to H3 T118E , H3 T118I or overexpressing Aurora-A . ( A ) Measurement of the width and length of chromosome one for over 50 chromatids for each H3 WT:FLAG and H3 T118I:FLAG stable cell lines ( *p<0 . 001 by Wilcoxon rank sum test ) . ( B ) Interkinetochore distances for pairs of sister chromatids . N=100 centromeres from 5 mitotic chromosome spreads ( p<0 . 01 by student t-test ) . Error bars represent SD of the mean ( SDM ) . ( C ) SEM images taken at 50 K and 100 K magnification upon prolonged mitotic arrest . Scale bar = 1 μm . ( D ) Western analysis of soluble ( free histones ) and pellet ( chromatin ) fractions following successive increasing concentration NaCl extractions . ( E ) Dnase-I digestion analysis on nocodazole arrested cells . Densitometric profiles are shown on the right . ( F ) As in A , comparing 293TR versus Aurora–A overexpressing cell lines for over 30 chromatids ( *p<0 . 001 and p<0 . 01 by Wilcoxon rank sum test ) . ( G ) As in B , comparing 293TR cell lines with over expression of Aurora-A or Aurora-A KD with and without colcemid arrest ( Pro-M arrest , pro-metaphase arrest ) . N=50 centromeres from 5 mitotic spreads ( *p<0 . 001 by student t-test ) . Error bars represent SDM . DOI: http://dx . doi . org/10 . 7554/eLife . 11402 . 01610 . 7554/eLife . 11402 . 017Figure 5—figure supplement 1 . The interkinetochore distance becomes longer upon expression of T118I . ( A ) Demonstration of how we measured chromosome 1 in metaphase spreads . Representative image of spread chromosomes treated with netropsin followed by DAPI staining . Arrows indicate characteristic heterochromatin of chromosome 1 . Dashed lines exemplify the measurements taken of the telomere-to-telomere-length versus the width . Scale bar = 5 μm . ( B ) Packaging of chromosome 1 is unchanged by T118A . Chromosome one arm length was measured and plotted against the width of each chromatid for over 50 chromatids ( p<0 . 001 by Wilcoxon rank sum test ) . ( C ) Packaging of chromosome 1 is shortened and becomes wider by expression of H3 T118E . Chromosome one arm length was measured and plotted against the width of each chromatid for over 50 chromatids ( **p<0 . 001 by Wilcoxon rank sum test ) . ( D ) The interkinetochore distance becomes longer upon expression of T118I . Representative interkinetochore distances are shown for individual centromeres of each stable cell line marked by CENP-A following a mitotic chromosome spread for asynchronous ( left panels ) and metaphase arrest ( right panel ) . Scale bar = 1 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 11402 . 01710 . 7554/eLife . 11402 . 018Figure 5—figure supplement 2 . The Aurora-A kinase dead does not change the packaging of chromosome 1 , as compared to expression of Aurora-A . ( A ) H3 T118I makes chromatin more accessible to nuclease digestion . Analysis of the Dnase-I digestion products ( time 0 , 1 min , 2 min , 5 min , 10 min , 20 min ) carried out on nuclei isolated from either asynchronous H3 WT:FLAG or H3 T118I:FLAG stable cell lines . Densitometric profiles of each time point of digestion products are shown H3 WT:FLAG ( blue ) or H3 T118I:FLAG ( red ) . ( B ) The Aurora-A kinase dead does not change the packaging of chromosome 1 , as compared to expression of Aurora-A ( Figure 5 ) . As in Figure 5 – figure supplement B , C . Analysis was of over 30 chromatids for the 293TR versus Aurora-A KD:FLAG stable cell line . DOI: http://dx . doi . org/10 . 7554/eLife . 11402 . 018 To obtain a higher resolution view of the effects of the H3 T118 mutations on chromosome structure , we performed scanning electron microscopy ( SEM ) . Upon pro-metaphase arrest , chromosomes from the H3 wild type and T118A mutant cell lines were organized into loops and coils to form very tight compact structures ( Figure 5C ) . By contrast , mitotic chromosomes from the H3 T118E and T118I cell lines were less tightly packed with longer radiating DNA loops . These results indicate that H3 T118E and T118I disrupt the higher order chromatin packaging . This grossly altered mitotic chromosome structure led us to test whether the H3 T118I mutation causes the histones to be more readily removed from chromatin . In agreement , H3 T118I was more readily extracted from chromatin than wild type H3 at 600 mM salt ( Figure 5D ) . Expression of H3 T118I also increased DNA accessibility to the nuclease DNase I in both asynchronous and mitotically arrested cells ( Figure 5E , Figure 5—figure supplement 2A ) . Together , these results are consistent with biochemical studies that showed that mononucleosomes with H3 T118ph favor the removal of histone H3 from DNA compared to unphosphorylated mononucleosomes ( North et al . , 2011 ) . Given that overexpression of Aurora-A results in excess H3 T118ph ( Figure 4D ) , we asked if it also disrupts chromosome integrity . Overexpression of Aurora-A caused significant widening and shortening of the chromosome arms of metaphase chromosomes ( Figure 5F ) as was observed for H3 T118E and T118I ( Figure 5A ) , while overexpression of Aurora-A KD did not ( Figure 5—figure supplement 2B ) . Overexpression of Aurora-A also caused increased sister chromatid interkinetochore distances ( Figure 5G ) . These results further indicate that the H3 T118I and T118E mutations are functional mimics of H3 T118 phosphorylation in vivo , and show that H3 T118ph disrupts higher order chromatin packaging . The altered chromatin integrity and cohesion defect caused by excess H3 T118ph or mutations that mimic excess H3 T118ph led us to ask whether there was a dissociation of cohesin proteins from DNA due to excess H3 T118ph . During mitotic delay , the intensity of the Rad21/Scc1 component of the cohesion complex along chromosome arms and at centromeres was drastically reduced in cells expressing H3 T118E and T118I ( Figure 6A , Figure 6—figure supplement 1 ) . Mechanistically , the loss of cohesin and the resulting faulty cohesion phenotype that is caused by excess H3 T118ph ( Figure 4B , C , E ) could result from multiple causes: premature activation of separase , premature removal of cohesion via cohesin phosphorylation , or improper establishment of cohesion . We set out to distinguish amongst these possibilities . To ask if cells expressing H3 T118I and T118E had premature activation of separase during mitotic delay , we analyzed mitotic spreads after incubation with MG132 , which prevents degradation of Cyclin B and Securin and therefore inhibits separase activation ( Rock et al . , 1994 ) . The fact that the T118I and T118E mutants still displayed cohesion loss , despite inhibition of separase ( Figure 6B ) , indicates that cohesin loss in the T118E/I mutants is not due to premature separase activity . The bulk of cohesion is removed in pro-metaphase by phosphorylation of the cohesin subunit SA2 by PLK-1 kinase or Aurora-B kinase ( Hauf et al . , 2005 ) . We found that the PLK-1 inhibitor , BI2536 , and the Aurora-B inhibitor , hesperidin , prevented cohesion loss in all the H3 expressing cell lines ( Figure 6B ) . Sister chromatid cohesion is also facilitated by DNA catenation during DNA replication ( Nitiss , 2009 ) . To prevent DNA decatenation , we used a specific inhibitor of Topo II , ICRF-193 and found that chromosomes became extremely tangled , indicative that DNA catenation is undisturbed by the H3 T118 mutations ( Figure 6B ) . Taken , together , these data indicate the H3 T118I and T118E mutations do not disrupt the proper establishment of sister chromatid cohesion by both DNA and sister chromatid catenation , but are likely to lead to premature cohesion loss via the PLK-1 or Aurora-B mediated pathway . 10 . 7554/eLife . 11402 . 019Figure 6 . Premature cohesion loss in the phosphomimetic and SIN mutants is independent of separase activity , but dependent on proper centromere tension . ( A ) Mitotic spreads following the error correction assay . The primary antibodies used were against Rad21 , cohesion subunit ( magenta ) , CENP-A ( green ) , and DNA was stained with DAPI ( blue ) . Scale bar = 5 μm . ( B ) Quantitation of the degree of cohesion loss for H3:FLAG stable cell lines , upon proteasome inhibition with MG132 , treatment with colcemid , Aurora-B ( hesperidin ) , Plk-1 ( BI-2536 ) , and Topo-II ( ICRF-193 ) inhibitors for 3 hr was scored ( n=75 cells , per treatment collected over 3 experiments ) . Insets show representative chromosomes for each type of defect: closed , open , partially separated , separated or tangled . DOI: http://dx . doi . org/10 . 7554/eLife . 11402 . 01910 . 7554/eLife . 11402 . 020Figure 6—figure supplement 1 . Stable cell lines expressing H3 T118 mutants do not alter Rad21 staining in an asynchronous cell population . Mitotic spreads of an asynchronous cell population from each H3:FLAG stable cell line were subjected to indirect immunofluorescence . The primary antibodies used were Rad21 ( magenta ) , CENP-A ( green ) . DNA was stained with DAPI ( blue ) . The white boxes indicate the magnified image area . Scale bars = 5 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 11402 . 020 During our PLK-1 inhibition studies , we observed that chromosomes from cells expressing H3 T118I were extremely short ( Figure 7A , Figure 7—figure supplement 1A ) , a phenotype observed previously ( van Vugt et al . , 2004 ) . These short chromosomes occurred in 90% of the mitotic spreads from PLK-1 inhibited cells expressing H3 T118I compared to 22% for wild type H3 . This hypercondensation phenotype suggests that H3 T118I may disrupt chromosome scaffolding proteins involved in shaping mitotic chromosomes , including condensin I and II and Topo II . However , H3 T118ph does not co-localize with Topo II ( Figure 7—figure supplement 1B ) or condensin I ( Figure 7B ) . Next , we determined whether the amounts of the scaffold proteins condensin I , condensin II and Topo II were altered on mitotic chromosomes in the H3 T118 mutants . The staining of Topo II ( Figure 7—figure supplement 2A , B ) and condensin II ( Figure 7—figure supplement 3A , B ) was similar among cells expressing wild type or mutant H3 . However upon mitotic delay , by the error correction assay , there was a significant loss of turbo-GFP ( tGFP ) tagged condensin I CAP-H protein in both H3 T118E ( 25% of mitotic cells were tGFP negative ) and T118I ( 50% of mitotic cells were tGFP negative ) cell lines as compared to wild type H3 ( 0% of mitotic cells were GFP negative ) ( Figure 7C , D ) . These data demonstrate that H3 T118I and T118E results in reduced levels of condensin I , but not condensin II or Topo II , on chromatin , suggesting that H3 T118ph plays a role in reducing condensin I occupancy on the chromatin . 10 . 7554/eLife . 11402 . 021Figure 7 . Reduced condensin I association with chromatin due to H3 T118E and T118I . ( A ) Chromosome spreads upon PLK-1 inhibition and quantitation of the degree of cohesion loss for H3: WT:FLAG and H3 T118I:FLAG stable cell line . Insets show representative chromosomes for each type of defect: closed and short . ( n=50 cells ) . Scale bar = 5 μm . ( B ) Extended chromatin fibers from 293TR CAP-H:tGFP cells . Scale bar = 2 μm . The primary antibodies used were against tGFP ( green ) , H3 T118ph ( red ) , and DNA was stained with DAPI ( blue ) . ( C ) Representative mitotic spreads for condensin I ( CAP-H:tGFP ) positive and tGFP negative cell . The primary antibodies used were against tGFP ( green ) , CENP-A ( red ) , and DNA was stained with DAPI ( blue ) . Scale bar = 5 μm . ( D ) Quantitation of number of cells with positive condensin I ( CAP-H:tGFP ) for mutant H3 stable cell lines treatment without Monastrol ( - ) and Monastrol washout followed by MG132 ( + ) treatments . SDM is for three independent experiments ( n=100 per treatment ) . ( E ) As in D , quantitation using 293TR and Aurora-A overexpressing cell line from over 50 mitotic spreads in each condition . Error bars are SDM . DOI: http://dx . doi . org/10 . 7554/eLife . 11402 . 02110 . 7554/eLife . 11402 . 022Figure 7—figure supplement 1 . Topoisomerase II and H3 T118ph display different localization patterns along chromatin fibers . ( A ) PLK-1 inhibition leads to very short chromosomes . The un-cropped images of chromosome spreads in Figure 7B are shown . The white boxes indicate the magnified image . ( B ) Topoisomerase II and H3 T118ph display different localization patterns along chromatin fibers . Extended chromatin fibers were isolated from HeLa cells released for 30 min from a G2 arrest . Cells were stained with primary antibodies to H3 T118ph ( green ) and costained with Topo II ( magenta ) . DNA is stained with DAPI ( blue ) . Scale bar = 5 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 11402 . 02210 . 7554/eLife . 11402 . 023Figure 7—figure supplement 2 . Topoisomerase II and its levels are unaltered in chromatin from cell lines expressiong H3 WT:FLAG , H3 T118A:FLAG , H3 T118E:FLAG and H3 T118:FLAG . ( A ) Topoisomerase II and its levels are unaltered on chromatin from cell lines expressing H3 WT:FLAG , H3 T118A:FLAG , H3 T118E:FLAG and H3 T118I:FLAG . Representative mitotic spreads are shown from asynchronous cultures . The primary antibodies used were against Topo II ( magenta ) , CENP-A ( green ) , and DNA was stained with DAPI ( blue ) . White box indicates magnified image area . Scale bar = 5 μm . ( B ) Topoisomerase II and its levels are unaltered on chromatin from cell lines expressing H3 WT:FLAG , H3 T118A:FLAG , H3 T118E:FLAG and H3 T118I:FLAG upon pro-metaphase arrest . Representative mitotic spreads are shown . The arrest was established via the error correction assay method from each H3:FLAG stable cell line . DOI: http://dx . doi . org/10 . 7554/eLife . 11402 . 02310 . 7554/eLife . 11402 . 024Figure 7—figure supplement 3 . Condensin II and its levels are unaltered on chromatin from cell lines expressing H3 WT:FLAG , H3 T118A:FLAG , H3 T118E:FLAG and H3 T118I:FLAG ( A ) Condensin II and its levels are unaltered on chromatin from cell lines expressing H3 WT:FLAG , H3 T118A:FLAG , H3 T118E:FLAG and H3 T118I:FLAG in asynchronous cultures . Representative mitotic spreads are shown . The primary antibodies used were CapD3 ( condensin II subunit ) ( magenta ) and CENP-A ( green ) . DNA is marked by DAPI ( blue ) . The white boxes indicate the magnified image area . Scale bar = 5 μm . ( B ) Condensin II and its levels are unaltered on chromatin from cell lines expressing H3 WT:FLAG , H3 T118A:FLAG , H3 T118E:FLAG and H3 T118I:FLAG upon pro-metaphase arrest . Representative mitotic spreads are shown . The arrest was established via the error correction assay method from each H3:FLAG stable cell line . DOI: http://dx . doi . org/10 . 7554/eLife . 11402 . 02410 . 7554/eLife . 11402 . 025Figure 7—figure supplement 4 . The binding of Condensin I to nucleosomes is not affected by H3 T118 mutations . ( A ) Purified Condensin I . Silver stain analysis of Streptavidin-Binding Peptide ( SBP ) tagged SMC2 and CAP-H GFP isolated from chicken DT40 mitotically arrested cells . ( B ) The binding of Condensin I to nucleosomes is not affected by H3 T118 mutations . Binding of condensin I to unmodified or H3 T118ph histone octamers reconstituted onto cy5 labeled 247 bp DNA . DOI: http://dx . doi . org/10 . 7554/eLife . 11402 . 025 Given that mutations that mimic H3 T118ph had reduced condensin I occupancy , we asked whether H3 T118ph directly prevents the binding of condensin I to chromatin . We purified the condensin I complex ( Figure 7—figure supplement 4A ) and used expressed protein ligation to generate mononucleosomes that were 100% phosphorylated on H3 T118 ( North et al . , 2011 ) . The histones carrying H3 T118ph generated not only canonical nucleosomes , but also altosomes and disomes ( Figure 7—figure supplement 4B ) as seen previously ( North et al . , 2014 ) . In electrophoretic mobility shift assay ( EMSA ) at higher levels of condensin 1 , we found that condensin I could bind to nucleosomes and the altered histone-DNA forms , irrespective of the phosphorylation status of H3 T118 . This result indicates that H3 T118ph does not directly affect condensin I binding to a mononucleosome . As such , we favor the idea that H3 T118ph promotes changes in global chromatin packaging that may indirectly reduce condensin I occupancy . Therefore , we asked if overexpression of Aurora-A recapitulates the loss of condensin I caused by expression of H3 T118I and T118E . Indeed , upon mitotic delay , there was a significant loss of condensin I from chromatin upon Aurora-A overexpression ( 25% of mitotic cells were GFP negative ) as compared to the control ( 0% of mitotic cells were GFP negative ) ( Figure 7E ) . This result shows that excess H3 T118ph leads to condensin I loss from chromatin . Taken together , these data suggest that the function of mitotic H3 T118ph is to indirectly reduce condensin I and cohesin occupancy on chromatin via its influence on chromosome packaging . H3 T118ph appears at pericentromeric regions during prophase and disappears from each chromosome as it aligns at the metaphase plate . Furthermore , the H3 T118I and T118E mutants resulted in displacement of condensin I and cohesin from chromatin and generated chromosomes with looser chromatin packaging . Accordingly , we propose that H3 T118ph plays an important role in organizing the chromatin structure around centromeres to achieve optimal levels of cohesin and condensin I association to permit enough conformational flexibility for microtubule attachment . Condensin I is highly enriched at centromeres in mitosis in metazoans ( Kim et al . , 2013 ) and promotes the rigidity of the centromere ( Gerlich et al . , 2006 ) . Additionally , Aurora–A has been demonstrated to play a role in error correction by destabilizing microtubule connections of misaligned chromosomes . Upon knockdown or inhibition of Aurora-A the kinetochore , as well as their attachment to microtubules , become more rigid and stable ( Chmatal , Yang et al . 2015 , Ye , Deretic et al . 2015 ) . As such , H3 T118ph at the centromere appears to act to limit condensin I occupancy in order to increase flexibility at the centromeres of misaligned chromosomes ( Figure 8 ) . This idea is supported by chromosomes from cells expressing H3 T118E , T118I or overexpressing Aurora-A having increased interkinetochore distances ( Figure 5B , G ) , which could be indirectly or directly related to the role of H3 T118ph in removal of cohesin and condensin I . However , upon attachment to mitotic spindles from opposite centrosomes , the centromeric regions have to be rigid enough to resist the forces that the microtubules exert on the centromere in order to prevent separation of sister chromatids until anaphase ( Musacchio and Salmon , 2007 ) . Removal of H3 T118ph as soon as tension is sensed across the kinetochores would allow for better centromere rigidity . Consistent with an important role for H3 T118ph in achieving appropriate microtubule attachment , H3 T118ph remained at centromeres of misaligned chromosomes ( Figure 4A ) . H3 T118ph occurs in a punctate periodic pattern along chromosome arms in prophase and pro-metaphase . Excess H3 T118ph ( due to overexpression of Aurora-A or mutations that mimic the effect of phosphorylation ) leads to gross alterations in chromosome compaction , with wider and shorter chromosome arms and longer , less organized chromatin loops ( Figure 5 ) , suggesting that H3 T118ph plays a role in shaping mitotic chromosomes . Mitotic chromosomes have been suggested to be packaged in a two phase process ( Naumova et al . , 2013 ) . In the first phase , a linear array of chromatin loops form at random , but consistent , positions along the chromosome . In the second phase , the loops longitudinally condense around the axes . These two different phases are mediated by the condensins , where condensin II is required for linear compaction along the chromosome axes while condensin I helps organize chromatin loops around the axes ( Shintomi and Hirano , 2010 , Green et al . , 2012 ) . Although the timing of appearance of H3 T118ph and condensin I on chromosome arms is similar , their spatial localization along the arms are distinct ( Figure 7B ) . As such , there is no evidence that H3 T118ph physically recruits or displaces condensin I from chromatin . Indeed , other proteins promote condensin recruitment in yeast including kinetochore proteins ( Tada et al . , 2011 ) and the Ku heterodimer complex , which functions in non-homologous end joining ( Tanaka et al . , 2012 ) . Perhaps related to its recruitment mechanism , yeast condensin interacts with histone H2A and H2AZ in vitro ( Tada et al . , 2011 ) and cross-linking mass spectrometry studies have found interactions between condensin I and H2A and H4 ( Barysz et al . , 2015 ) . In addition , our evidence indicates that H3 T118 phosphorylation is likely to regulate condensin I occupancy on the chromatin , given that expression of H3 T118I , T118E and overexpression of Aurora-A cause loss of condensin I from chromosome arms ( Figure 7C–E ) . This disruption of condensin I function is in agreement with the longer loops of chromatin that were observed by SEM in the T118E and T118I mutants ( Figure 5C ) . Consistent with the delayed cytokinesis that occurs upon condensin I knockdown ( Gerlich et al . , 2006 ) the H3 T118I and T118E mutants caused a delay in cytokinesis ( Figure 3E ) . Given that condensin I interacts with chromosomes in a more dynamic manner than condensin II ( Gerlich et al . , 2006 ) , we propose that the dynamic nature of the association of condensin I with chromatin enables H3 T118ph to regulate the levels of condensin I to shape the mitotic chromosomes as they condense . The ratio of condensin I to condensin II is very tightly controlled within cells , given that changes in the ratio profoundly alters the shape of mitotic chromosomes ( Shintomi and Hirano , 2010 , Bakhrebah et al . , 2015 ) . As such , the removal of H3 T118ph from the chromosome arms by metaphase , either by dephosphorylation or by our preferred model of physical removal of T118 phosphorylated H3 from the DNA , is likely to regulate the ratio of condensin I:condensin II for appropriate chromosome compaction . This function is likely to occur in an indirect manner via H3 T118ph affecting chromatin structure , given that condensin I binding to mononucleosomes is not affected by H3 T118ph in vitro ( Figure 7—figure supplement 4B ) . Similarly , we propose that the loss of cohesin is an indirect consequence of the altered packaging of the chromatin structure caused by excess H3 T118ph , which may expose the cohesin ring to PLK-1 mediated phosphorylation and subsequent removal of cohesin ( Hauf et al . , 2005 ) . However , we were unable to rule out the possibility that the cohesion phenotype may be due to loss of Sgo-1-mediated protection against PLK-1 and Aurora-B kinases . Taken together , our work suggests a model where phosphorylation of H3 T118 at the nucleosome dyad by Aurora-A is a critical step to ensure chromosome congression , via its influence on chromosome compaction and cohesion through physically regulating nucleosome structure . These functions are likely to be conserved in metazoans , as we find similar localization and timing of H3 T118ph in nematodes , flies , and human cells . The importance of the ability to utilize H3 T118ph to alter the nucleosome structure to regulate mitosis is underscored by the embryonic lethality of flies where all of their histones are mutated to prevent T118 phosphorylation or to mimic persistent H3 T118 phosphorylation . Given that Aurora-A is overexpressed in many cancers , it is tempting to speculate that the carcinogenic effect of overexpressed Aurora-A may be mediated at least in part via altering the mitotic chromatin structure by phosphorylation at the nucleosome dyad . Plasmid expressing human H2B:RFP was a kind gift from Walter Hittelman ( MDACC , Houston , TX ) . Plasmids expressing human Aurora-A:FLAG and Aurora-A KD:FLAG were a kind gift from Subrata Sen ( MDACC , Houston , TX ) ( Katayama et al . , 2012 ) . The CMV-histone Drosophila H3-YFP ( dH3 ) plasmid was purchased from Addgene ( plasmid 8694 ) . The CapH:GFP plasmid was purchased from Origene ( Rockville , MD USA , RG201421 ) . The shRNA histone H3 resistant plasmid pOZ-FH-C H3 . 1c:FLAG:HA ( HuH3 . 1:FLAG ) was kindly provided by Zhenkun Lou , Ph . D ( Mayo Clinic , Rochester , Mn ) . Site directed mutagenesis was performed on the CMV-histone dH3-YFP and pOZ-FH-C H3 . 1c:FLAG:HA plasmids listed below using the QuickChange Site-directed Mutagenesis Kit ( Agilent Technologies , Santa Clara , CA , USA 200515 ) . This plasmid has the histone sequence of Drosophila histone H3 and corresponds to the human histone H3 . 2 amino acid sequence . The CMV-histone dH3 YFP T118A plasmid was generated using the following primers: Forward: 5’- TTCATGCCAAGCGTGTCGCCATAATGCCCAAAGAC -3’ Reverse: 5’- GTCTTTGGGCATTATGGCGACACGCTTGGCATGAA -3’ The CMV-histone dH3 YFP T118E plasmid was generated using the following primers: Forward: 5’- GCCATTCATGCCAAGCGTGTCGAGATAATGCCCAAAGACATCCAG -3’ Reverse: 5’- CTGGATGTCTTTGGGCATTATCTCGACACGCTTGGCATGAATGGC -3’ The CMV-histone dH3 YFP T118I plasmid was generated using the following primers: Forward: 5’- TCATGCCAAGCGTGTCATCATAATGCCCAAAGACA -3’ Reverse: 5’-TGTCTTTGGGCATTATGATGACACGCTTGGCATGA -3’ The pOZ-FH-C HuH3 . 1T118A:FLAG primer was generated using the following primers: Forward: 5’- CACGCTAAACGCGTCGCCATCATGCCCAAAG -3’ Reverse: 5’- CTTTGGGCATGATGGCGACGCGTTTAGCGTG -3’ The pOZ-FH-C HuH3 . 1T118E:FLAG plasmid was generated using the following primers: Forward: 5’- GCTATTCACGCTAAACGCGTCGAGATCATGCCCAAAGATATCCAG -3’ Reverse: 5’- CTGGATATCTTTGGGCATGATCTCGACGCGTTTAGCGTGAATAGC -3’ The pOZ-FH-C HuH3 . 1T118:FLAG plasmid was generated using the following primers: Forward: 5’- TCACGCTAAACGCGTCATCATCATGCCCAAAGATA -3’ Reverse: 5’- TATCTTTGGGCATGATGATGACGCGTTTAGCGTGA -3’ The following primers were used for a PCR ligation reaction to amplify HuH3 . 1:FLAG Forward: 5’-ATGGCTCGTACGAAGCAAAC-3’ Reverse: 5’-CTAGGCGTAGTCGGGCACGTCGT -3’ The resulting PCR fragment was cloned into pcDNA5 FRT/TO TOPO TA plasmid ( Life technology Grand Island , NY USA K6510-20 ) Mad2 antibody was a kind gift from Ted Salmon ( UNC , Chapel Hill , NC ) . The following primary antibodies were purchased: polyclonal H3 T118ph ( Abcam Cambridge , MA USA ab33310 , lot 7 for western blots and lot 9 for immunofluorescence ) , H3S10ph ( Abcam , ab14955 ) , C-terminal H3 ( Abcam , ab1791 ) , N-terminal H3 ( Active Motif Carlsbad , CA USA , 39763 ) , γ-tubulin ( Abcam , ab27074 ) , CENP-A ( Abcam , ab8245 ) , CENP-A ( Cell Signalling Technology Danvers , MA USA , 2186 ) , GAPDH ( Abcam , ab8245 ) , M2-FLAG ( Sigma St . Louis , MO USA , F3165 ) , BubR1 ( Abcam , ab4637 ) , Hec1 ( Abcam , ab3613 ) , CENP-E ( Abcam , ab4163 ) , Hp1α ( Active Motif , 39295 ) , HP1β ( Active Motif , 39979 ) HP1γ ( Active Motif , 39981 ) , Aurora-B/AIM-1 ( BD Biosciences , 611082 ) , SA2 ( Bethyl laboratories , Montgomery , TX USA , A310-043A ) , Rad21 ( Millipore Billerica MA , USA 05–908 ) , H3 K9 me3 ( Abcam , ab6001 ) , CapD3 ( Bethyl laboratories , A300-604A ) , Topo II ( Milllipore , MAB4197 ) , ( phospho ) Aurora-A T288 ( Cell Signaling , 3079 ) , Aurora-A Clone 35C1 ( Invitrogen , 45–8900 ) , α-tubulin ( Sigma-Aldrich , T9026 ) , α-tubulin ( AbD Sterotec Raleigh , NC , USA MCA78G ) , anti-GFP ( Roche Indianapolis , IN USA 11814460001 ) , and anti-turboGFP ( Origene TA150041 ) . The secondary antibodies used were as follows: Alexa Fluor 488 goat anti-rabbit ( Life Technologies Carlsbad , CA , A11034 ) , Alexa Fluor 594 goat anti-rabbit ( Life Technologies , A11037 ) , Alexa Fluor 488 goat anti-mouse ( Life Technologies , A11029 ) , Alexa Fluor 555 goat anti-rat ( Cell Signaling ) , HRP-conjugated anti-mouse ( Promega , Madison , WI USA , PR-W4011 ) , and HRP-conjugated anti-rabbit ( Promega , PRW4021 ) . Non-biotinylated peptides used were H3 unmodified ( Abcam , ab12149 ) , H3 S10ph ( Abcam , ab1147 ) , H3 K122ac ( Abcam , ab34466 ) , and H3 T118ph ( Abcam , ab33505 ) . Biotinylated peptides were either purchased from Anaspec ( Freemont , CA , USA ) or were a kind gift from Min Gyu Lee ( MDACC ) . HeLa cells were maintained in Dulbecco's Modified Eagle Medium ( DMEM ) supplemented with 10% fetal bovine serum and 1% penicillin/streptomycin . WI-38 cells were maintained in Eagle's Minimum Essential Medium ( MEM ) supplemented with 10% fetal bovine serum and 1% penicillin/streptomycin . MCF10A cells were maintained in DMEM/Nutrient Mixture F-12 supplemented with 5% horse serum , 1% penicillin/streptomycin , 10 mg/ml insulin , 1 mg/ml hydrocortisone , 25 µg/ml EGF , and 1 mg/ml cholera toxin . The Flp-in T-Rex 293 ( 293TR ) cell line was purchased from Life Technologies ( R780-07 ) and were maintained in DMEM 10% fetal bovine serum and 1% penicillin/streptomycin . Stable cell lines of HuH3 . 1 FLAG:HA were made by transfecting 293TR cells with 1 µg of pcdna5 FRT huH3 . 1FLAG:HA , and 9 ug of POG44 ( Life Technologies , V6005-20 ) , using the Nucleofector kit according to the manufacturers instructions ( Lonza Basel Switzerland , V4XC-2012 ) . One day post transfection , cells were washed with fresh medium . Two days post transfection polyclonal stable cell lines were selected by maintaining cells in 400 μg/ml hygromycin . Stable cell lines expressing Aurora-A:FLAG , Aurora-A KD:FLAG and CapH:tGFP were made by transfecting 293TR cells ( and desired H3 mutant cell lines ) with 1 µg of plasmid , using lipofectamine 2000 according to the manufacturer’s instructions . Two days post-transfection , stable cell lines were selected by maintaining cells in media containing 800 μg/ml G418 . Cells were plated in a six-well dish and were grown to 50% to 60% confluence . For siRNA inhibition studies , the cells were co-transfected with 0 . 5 µg plasmid pBos H2B:RFP and siGENOME Human AURKA siRNA ( Thermo Scientific Lafayette , CO USA , D-003545-05-0005 ) or ON-TARGET plus non-targeting siRNA #1 ( Thermo Scientific , D-001810-01-05 ) ( at a final concentration of 100 nM ) in the presence of Lipofectamine 2000 reagent ( Life Technologies , 11668019 ) , as per the manufacturer's instructions . The cells were harvested at 72 hr post transfection for protein extraction and immunofluorescence analysis . For shRNA knockdown studies , three different shRNA constructs ( pGipz ) were purchased from MD Anderson’s shRNA core . The target sequences of TPX2 shRNA are ( 1 ) TTAGCAGTGGAATCGAGTG; ( 2 ) AACAGGTTAATATCATCCT; ( 3 ) ATCTTGATGAGCACTGCCT . Cells were plated in six-well plates with CELL-TAK ( BD Biosciences San Jose , CA USA , 354240 ) were grown to 50% to 60% confluence , and were cotransfected with all three TPX2 target sequences in the presence of Lipofectamine 2000 reagent ( Life Technologies ) , as per the manufacturer's instructions . After transfection , the cells were split at 72 hr and 1 μg/ml puromycin was added . After 5 days the cells were collected for protein extraction and immunofluorescence analysis . Wild type N2 Bristol C . elegans were grown and maintained at 20°C as described ( Brenner , 1974 ) . The feeding method of RNAi delivery was used to deplete CENP-A/HCP-3 , as previously described by Timmons and Fire ( Timmons and Fire , 1998 ) . RNAi plasmids for CENP-A/hcp-3 were obtained from the Geneservice Ltd . C . elegans feeding library ( Kamath and Ahringer , 2003 ) . E . coli HT115 ( DE3 ) bacteria was transformed with the control or CENP-A/HCP-3 RNAi plasmids . 1 ml LB + 100 µg/µl ampicillin liquid culture was inoculated with a single colony of HT115 bacterial transformation and grown overnight at 37°C . The following day these cultures were expanded into 50 ml LB/amp using a 1:100 dilution and grown for six hours at 37°C . After six hours , 200 µl were spread onto single nematode growth ( NG ) plates supplemented with 20% β-lactose and placed at 25°C for 72 hr . Subsequently , the plates were seeded with L4-stage hermaphrodites and incubated at 25°C for 24 hr ( Arur et al . , 2009 ) . The L4440 RNAi vector was used as an RNAi control . We used an Eg5 inhibitor , Monastrol , to induce a monopolar spindle and kinetochore-microtubule attachment errors ( Sanhaji et al . , 2010 ) . For the chromosome attachment error correction assays ( monastrol-release experiments ) , cells were split into a 6 well dish at least 24 hr prior to treatment . Cells at 75% confluency were treated for 4 hr with monastrol ( 100 µM , Enzo Life Sciences , Farmingdale NY USA , BML-GR322-0005 ) and washed and released into fresh medium containing MG132 ( 20 µM , Calbiochem , Billerica , MA USA , 474790-1MG , in ETOH ) for 2 hr and cells collected for immunofluorescence . All inhibitors were used at the listed concentrations MG132 ( 20 µM in ETOH ) , RO-3306 ( 9 µM , Enzo Life Sciences , ALX-270-463-M001 , in DMSO ) , Nocodazole ( 100 mg/ml , Sigma , M1404 , in ETOH ) , Colcemid ( 0 . 01 µg/mL , Roche 10295892001 ) , PLK-1 inhibitor BI-2536 ( 100 nM , Selleck chemicals , Houston , TX USA , S1109 , in DMSO ) , Caffeine ( 80 nM , Sigma C0750 , in DMEM ) , Aurora-B inhibitor ZM447439 ( 2 µM , Tocris Biosciences , S1103 , in DMSO ) , Calyculin A ( 50 nM , Tocris Biosciences , in EtOH ) , Aurora-B inhibitor Hesperidin ( 1 µM Selleck chemicals S2309 , in DMSO ) , Aurora-A inhibitor VX-680 ( 1 µM , Selleck chemicals , S1048 , in DMSO ) , Aurora-A inhibitor MLN 8237 ( 1 µM , Selleck Chem , S1133 , in DMSO ) , Topoisomerase II inhibitor ICRF 193 ( 10 µM , Sigma , U4659 , in DMSO ) . Immunofluorescence of metaphase chromosome spreads was prepared by cytospin following the pre-extraction method as described previously ( Ono et al . , 2003 ) . Immunofluorescence of adherent cells were grown on poly-D-lysine coated coverslips ( BD Biosciences , 354086 ) and harvested prior to reaching 80% confluency . Coverslips were washed in 1× PBS and fixed in 4% paraformaldehyde/1 x PBS for 10 min at room temperature ( Electron Microscopy Sciences Hatfield , PA USA , 15710 ) . Coverslips were washed in 1 x PBS and then permeabilized with 1 x PBS + 0 . 1% Triton X-100 at RT for 10 min . Coverslips were then washed in 1 x PBS and blocked in 3% BSA/1× PBS for 1 hr . Primary antibodies were diluted into 3% BSA/1 x PBS and incubated overnight at 4ºC . Coverslips were washed 3 times 1× PBS for 15 min prior to adding secondary antibodies . Coverslips were washed 3 times in 1 x PBS for 15 min and mounted onto glass slides with ProLong Gold Antifade mounting reagent containing DAPI ( Life Technologies , Grand Island , NY , USA , Cat# P36931 ) . Immunofluorescence images were acquired as described below . Embryos from adult hermaphrodites were picked into 10 μl egg buffer on a Poly-L-Lysine coated glass slide ( Sigma , St Louis , MO P0425 ) . To release the embryos , a coverslip was placed over the animals and gentle pressure was applied . The slides were subsequently placed on an aluminum plate over dry ice for 1 hr . To crack the embryo’s cuticle and aid its permeabilization , coverslips were quickly snapped off . Slides were fixed in -20°C methanol for 20 min , followed by sequential rehydrations: 80:20 , 50:50 , and 20:80 methanol to 1x PBS with 0 . 1% Tween ( PBST ) . After hydration , samples were blocked in 1X PBST with 1% BSA for 1 hr at room temperature and then incubated overnight in primary antibody diluted in PBST at 4°C . Primary antibodies used were anti-tubulin ( 1:2000 , Sigma ) , and H3 T118ph ( 1:1000 ) . Samples were then washed with PBST and secondary antibodies were applied for 2 hr at room temperature . Secondary antibodies used were: Alexa Fluor 488 goat anti-mouse IgG and Alexa Fluor 594 goat anti-rabbit ( both at 1:1000 ) ( Invitrogen Molecular Probes , Eugene , OR ) . After incubation with the secondary antibodies the samples were washed with PBST and mounted using ProLong Gold Antifade ProLong with DAPI . Immunofluorescence images were acquired as described below . Cells were collected by mitotic shake off . Media was removed and the cells were pelleted at 1000 rpm for 5 min . All but 1 ml of media was removed and gently used to resuspend cells . Cells were swollen in 10 ml of hypotonic solution ( 46 . 5 mM KCl/8 . 5 mM NaCitrate ) and incubated for 20 min at 37°C . Fresh Carnoy’s fixative ( 3:1 methanol:acetic acid ) was added to hypotonic buffer at 10% ( v/v ) . Subsequent to centrifugation cells , were fixed 3 times with 10 mls Carnoy’s fixative for 10 min at RT followed by pelleting the cells at 1000 rpm for 5 min . Pellets were than resuspended in a small volume of Carnoy’s fixative , dropped onto positively charged slides ( Fisher scientific , Ashville , NC USA , 12-550-15 ) air-dried , and stained with 1 mg/ml DAPI solution diluted 1:15 , 000 . Slides were mounted with ProLong Gold Antifade mounting reagent containing DAPI . Immunofluorescence images were acquired as described below . To stain heterochromain , chromosome spreads were treated as in ( Hirota et al . , 2004 ) except 0 . 08 mg/ml netropsin was used instead of distamycin . We followed published methods ( Lai et al . , 2011 ) . Chromosome spreads were prepared as described above except the chromosomes were dropped onto poly-D-lysine coated coverslips ( BD Biocoat , 354086 ) in a 37°C room with minimal drying . The coverslips were flipped onto a larger coverslip with 1 drop of 45% acetic acid and the large coverslip was placed on dry ice for 15 min . The chromosome spreads were then fixed in 2 . 5% glutaraldehyde / 1 x PBS overnight at 4°C . The fixed samples were than washed with distilled water for 5 min , 10 min , and 15 min , then dehydrated with a graded series of increasing concentrations of ethanol ( 5 min in 70% , 10 min in 90% and 15 min in 100% ) . The samples were then chemically dried in a graded series of increasing concentrations of hexamethyldisilazane ( HMDS , Electron Microscopy Services ) 2:1 ( 100% EtOH:HMDS ) , 1:1 ( 100% EtOH: 100% HMDS ) , then 1:2 ( 100% EtOH: HMDS ) , then 3 changes in pure HMDS where all steps were for 5 min each . Then the samples were air dried overnight . Samples were mounted onto an aluminum specimen mount ( Ted pella , INC . ) by carbon conductive double-stick tape ( Ted Pella . Inc . , Redding , CA ) . The samples were then coated under vacuum using a sputter system ( 208HR , Cressington Scientific Instruments , England ) with platinum alloy for a thickness of 30 nm . Samples were examined in a Nova NanoSEM 230 scanning electron microscope ( FEI , Hillsboro , Oregon ) at an accelerating voltage of 10 kV . In general , to produce chromosome spreads , HeLa mitotic cells obtained by mitotic shake off were incubated in pre-warmed hypotonic buffer ( 46 . 5 mM KCl/8 . 5 mM NaCitrate ) at 37°C for 8–10 min . 293TR mitotic cells obtained by selective detachment were incubated in pre-warmed hypotonic buffer #5 ( 10 mM Tris-HCl pH7 . 4 , 40 mM glycerol , 20 mM NaCl , 1 . 0 mM CaCl2 , 0 . 5 mM MgCl2 ) . After attachment to Poly-D-lysine glass coverslips by Cytospin at 1000 rpm for 2 min , chromosome spreads were pre-extracted with 0 . 1% Triton X-100/1 x PBS for 2 min and were than fixed with 2% PFA/1 x PBS at RT for 10 min . Cells were extracted with 0 . 1% Triton X-100/PBS for 10 min . Blocking occurred in 1 x PBS , 3% BSA , and 0 . 1% Triton X-100 , for 30 min at room temperature . Once blocking was complete , the immunofluorescence protocol was followed as described above . Cells were arrested with colcemid and the chromatin fibers were generated as described elsewhere ( Dunleavy et al . , 2011 ) . Briefly , chromatin fibers from human cells were prepared by incubating 6–8 x 104 cell/ml in prewarmed hypotonic buffer at 37°C for 10 min . HeLa cells used hypotonic buffer 46 . 5 mM KCl/8 . 5 mM Na Citrate and for 293TR used the buffer was 10 mM Tris-HCl pH 7 . 4 , 40 mM glycerol , 20 mM NaCl , 1 . 0 mM CaCl2 , 0 . 5 mM MgCl2 . Cells were centrifuged onto charged microscope slides ( Fisher Scientific , 2-550-15 ) and lysed for 14 min in salt detergent buffer supplemented with urea ( 10 mM Tris HCl pH 7 . 5 , 1% Triton X-100 , 500 mM NaCl , and 500 mM urea ) before slowly aspirating the lysis buffer by vacuum and fixing in 2% PFA/1 x PBS . Slides were incubated in 1× PBST ( 1× PBS + 0 . 1% Triton X-100 ) and blocked in 1 x PBS , 1% BSA , 0 . 1% Triton X-100 , for 30 min at room temperature . Once blocking was complete , the immunofluorescence protocol was followed as described above . Two D150 plates , at 80% confluency , were collected by mitotic shake off . Cells were pelleted and washed in TB buffer ( 20 mM Hepes , pH 7 . 3 , 110 nM K-acetate , 5 mM Na-acetate , 2 mM Mg-acetate , 1 mM EGTA , 2 mM DTT , and a protease inhibitor cocktail ( Roche , Complete-mini , cat#1187350001 ) . All steps were done at 4°C . NP40 extraction of detergent soluble proteins was performed by treatment with 0 . 1% NP40 for 5 min , followed by centrifugation at 3000 rpm for 3 min to separate the non-chromatin supernatant and chromatin pellet fractions . The pellet fractions were subsequently digested with 20 µg/ml DNaseI ( Worthington Biochemical Corporation , Lakewood , NJ USA , LS006342 ) for 10 min at 37°C . Total , supernatant ( non-chromatin ) , and pellet ( chromatin ) fractions were resolved by SDS-PAGE and analyzed by western blotting . The method was adapted from ( Henikoff et al . , 2008 ) with the differences detailed below . Five million cells were pelleted during the nuclei extraction on ice samples and were divided into 5 tubes . The Nuclei were washed in NIM buffer ( 0 . 25 M sucrose , 25 mM KCl , 5 mM MgCl2 , 10 mM Tris-HCL pH 7 . 4 ) . Pelleted at 300 rpm for 5 min . The nuclei were resuspened in 5 different extraction buffers at increasing salt concentration ( 0 , 600 , 900 , 1200 , and 1500 mM NaCl ) Incubated on ice for 10 min . The soluble and pellet fractions were collected by centrifugation at 13 , 000 rpm for 10 min . 5xSDS was added to the soluble fractions and boiled at 100°C for 5 min . The pellet fractions were resuspended in 250 μl Laemmli buffer and the Whole Cell Extract protocol was followed ( as detailed below ) . Approximately 2 × 106 cells were lysed with 200 µl Laemmli buffer ( 4% SDS , 20% glycerol , and 120 mM Tris pH 6 . 8 ) . Cells were subsequently vortexed for 30 s and then boiled at 100°C for 5 min . After briefly cooling , samples were vortexed for 30 s and sonicated for 10 s at 20% power . Lastly , 5 x SDS buffer was added to samples to obtain a 1x final concentration and samples were boiled at 100°C for 5 min . Whole cell extracts were prepared using RIPA buffer ( 150 mM NaCl , 1% NP40 , 0 . 5% sodium deoxycholate , 0 . 1% SDS , 50 mM Tris , pH 8 , 10 mM NaF , 0 . 4 mM EDTA , 10% glycerol and protease inhibitors ) supplemented with 10x phostop ( Roche 04906845001 ) , and 25x Protease inhibitor ( Roche 04693132001 ) . The pre-blocked protein-A–Dynabeads ( Thermo-Fischer , 10001D ) was then incubated with whole-cell extracts overnight in 4°C . The antibody was added for 4 hr the next day . Following extensive washes , the bead-bound protein complexes were analyzed by western blotting using H3 C-term antibody . Samples were resolved by 15% SDS-PAGE and transferred to nitrocellulose according to standard procedures . For HRP detection , following transfer the membranes were blocked in 5% non-fat milk ( w/v ) in 1× TBST for 1 hr . The blots were probed with primary antibodies at room temperature for 1 hr or overnight at 4ºC . Blots were washed and incubated in secondary antibodies at room temperature for 1 hr . ECL detection was either by Amersham ECL Western Blotting Detection Reagents ( GE , Pittsburgh , PA USA , RPN2106 ) or Immobilon Western Chemiluminescent HRP Substrate ( Millipore , WBKLS0500 ) . Alpha viewer was used to analyze and quantitate bands ( Proteinsimple , Santa Clara , CA , USA ) . For LICOR Odyssey detection the transfer blots were blocked in Sea Block buffer ( Thermo Scientific , Cat#37527 ) in 1 x PBS for 1 hr . Blots were incubated with primary and secondary antibodies as described above . An Odyssey imager was used to analyze and quantitate bands . Lyophilized peptides were rehydrated in 1 x PBS at a 10 µM concentration . The peptides were serially diluted to the indicated concentrations and dotted out onto activated PVDF membrane . The membrane was air-dried and then stained with amido black to verify the presence of the peptides . The membranes were washed in PBS and then blocked in 3% BSA/1 x PBS . The blots were incubated in primary antibodies overnight at 4ºC . The blots were washed and probed with HRP conjugated secondary antibodies at room temperature for 1 hr . Alpha viewer was used to analyze and quantitate bands ( Proteinsimple ) . Biotinylated peptides ( 4 µg total ) were incubated with 400 µl H3 T118ph antibody for 45 min at room temperature ( RT ) . Samples were centrifuged for 15 min at 4ºC at 13 k rpm . 300 µl of each supernatant was used for indirect immunofluorescence as described above . Approximately 2 × 106 cells were lysed with 2 mL of lysis buffer ( 50 mM Tris-HCl pH 7 . 9 , 100 mM KCl , 5 mM MgCl2 , 0 . 05% v/v saponin , 50% v/v glycerol , 0 . 5M DTT , 10x phostop ( Roche 04906845001 ) , and 25x Protease inhibitor ( Roche 04693132001 ) of asynchronous or synchronized cells ( synchronized for 6 hr in nocodazole ( 100 mg/ml , in ETOH ) ) . Cells were incubated in lysis buffer for 3 min on ice and vortexed every minute . Samples were centrifuged for 10 min at 4ºC at 1000 x g . Nuclei were subsequently digested for increasing times at 37°C with 5U DNase I ( Worthington Biochemical Corporation LS006342 ) in TB buffer ( 20 mM Hepes , pH 7 . 3 , 110 nM Potassium-acetate , 5 mM Sodium-acetate , 2 mM Magnesium-acetate , 1 mM EGTA , 2 mM DTT and a protease inhibitor cocktail ( Roche , Complete-mini , cat#1187350001 ) ) . Fragmented DNA was purified and analyzed by agarose gel electrophoresis followed by Sybr Gold ( Life Technologies , S-11494 ) staining for visualization with a FluorChem E FE05000 ( Protein simple , San Jose , CA ) . Plot profiles were obtained with ImageJ software . A four well chamber was coated with BD Bio TAK according to the manufacturer’s instructions . Approximately 24 hr prior to live cell imaging , HEK293 cells were transfected using the Nucleofector kit according to the manufacturers instructions ( Lonza , V4XC-2012 ) with 0 . 5 µg plasmid CMV:H3 . 2 YFP wild type or T118 mutant . The transfected cells were plated at 50 , 000 cells per well and grown in a humidified chamber for 24 hr . At the time of imaging , the cells were placed in a prewarmed Oko Full Enclosure incubator at 37°C with 5% CO2 . Cells were imaged using a 3i Marianas Spinning Disk Confocal equipped with an Evolve 10 MHz Digital Monochrome Camera ( Photometrics , Tuscon , AZ USA ) and images were taken every 5 min for 16 hr and driven by Slidebook 5 . 5 software ( a 63 x 1 . 49 NA Plan Apo oil immersion objective ) . Three Z-sections were acquired for each cell . The start of cytokinesis was defined when H3:YFP chromatin decondensed after anaphase . The end of cytokinesis was determined by the physical separation of the cytoplasmic membrane . The majority of images were acquired on a 3i Marianas Spinning Disk Confocal equipped with a coolSNAP HQ2 CCD Camera . Slidebook 5 . 5 software was used with a 63 x 1 . 49NA Plan Apo oil immersion objective and Z sections were acquired at 0 . 2 um steps . Intensity measurements were calculated with Slidebook 5 . 5 software . To measure inter-kinetochore distance , the center intensity of foci was determined by Imaris Bitplane software . Some immunofluorescence images were acquired on a Nikon 2000U inverted microscope equipped with a Photometrics Coolsnap HQ camera . Metamorph software was used with a 60x 1 . 49NA Plan Apo oil immersion objective and Z sections were acquired at 0 . 2 µm steps . Histone octamers unmodified and modified at H3T118ph were purified according to the method North et . al . ( North et al . , 2011 ) . The Condensin I complex was purified from 5 x 106 CAP-H-GFP-SBP , SMC2-SBP and GFP-SBP mitotic cells using the method by Kim et . al . ( Kim et al . , 2010b ) . After purification , proteins were eluted in SEB ( 50 mM Tris pH7 . 4 , 250 mM NaCl , 0 . 5% NP-40 , 0 . 1% Deoxycholate and 4 mM Biotin ) . 10 ml samples were subjected to NuPAGE SDS-PAGE and protein and evaluated by silver staining . The EMSA was performed as described previously ( Kimura et al . , 1997 ) . The following genotypes were used in this study: yw; Df ( 2L ) HisC/ CyO , P{ActGFP}JMR1; 6xHisGUVK33 , 27/ TM6B yw; Df ( 2L ) HisC/ CyO , P{ActGFP}JMR1; 6xHisGU VK33 , 27 H3T118A/ TM6B yw; Df ( 2L ) HisC/ CyO , P{ActGFP}JMR1; 6xHisGU VK33 , 27 H3T118E/ TM6B yw; Df ( 2L ) HisC/ CyO , P{ActGFP}JMR1; 6xHisGU VK33 , 27 H3T118I/ TM6B We constructed 6xHisGU VK33 , 27and 6xHisGU VK33 , 27 H3T118A , E and I chromosomes essentially as previously described ( Gunesdogan et al . , 2010 ) with the following changes: ΦC31attB3xHisGU . H3T118A , ΦC31attB3xHisGU . H3T118E , and ΦC31attB3xHisGU . H3T118I plasmids ( further referred to collectively as H3T118A/E/I ) were generated by replacing the EcoR1/Sac1 fragment in pENTR221-HisGU with a synthetic fragment ( Integrated DNA Technologies , Inc . , Iowa , USA ) containing an ACC into GCC codon exchange leading to the H3 T118A mutation , an ACC into GAG codon exchange leading to the H3 T118E mutation , or an ACC into AUC codon exchange leading to the H3 T118I mutation . The pENTRL4R1-HisGU . H3T118A/E/I and pENTRR2L3-HisGU . H3T118A/E/I entry vectors were generated by moving the Acc65I/AgeI fragment from the pENTR221-HisGU . H3T118A/E/I mutant vectors to the pENTRL4R1 and the pENTRR2L3 vectors . Recombination of pENTR221-HisGU . H3T118A/E/I , pENTRL4R1-HisGU . H3T118A/E/I and pENTRR2L3-HisGU . H3T118A/E/I with pDESTR3R4-ΦC31attB resulted in the ΦC31attB3xHisGU . H3T118A/E/I transgenic constructs . We utilized ΦC31-mediated transgenesis to integrate these constructs , as well as ΦC31attB3xHisGU , site specifically into the Drosophila genome using the landing sites VK27 and VK33 ( Venken et al . , 2006 ) . Homozygous viable insertions from each site were recombined to generate 6xHisGU VK33 , 27and 6xHisGU VK33 , 27 H3T118A , E and I chromosomes and crossed into the Df ( 2L ) HisC mutant background ( Gunesdogan et al . , 2010 ) . Df ( 2L ) HisC was kept heterozygous over CyO , P{ActGFP}JMR1 to identify mutant embryos lacking green fluorescent protein expression , and the viability of 12xHisGU transgene containing mutant and wild type animals was assessed . Wild type controls were either non-mutant sibling embryos ( internal control ) or embryos which contain 12xHisGU ( WT control ) , which both survive to adult viability .	In every one of our cells , our DNA is wrapped together with histone proteins to make a structure called chromatin . When a cell divides , each newly formed daughter cell must receive an identical set of chromatin . As part of this process , the chromatin is copied and then compacted , which causes a characteristic “X”-shaped chromosome to form . This “X” shape is actually made up of two identical parts , or chromatids , that are joined together until a specific time during cell division . If chromosomes separate too early or too late , the DNA will not distribute evenly to daughter cells , which could lead to diseases including cancer . Histone modifications are small chemical changes on the histone proteins that the DNA wraps around . Previous research identified a new histone modification that is located at an important contact point between the DNA and a particular histone protein . However , the role of this modification in living cells was not clear . Wike et al . have now determined that in animal cells , this histone modification occurs immediately before the chromatids separate and at specific locations along the chromosomes . The amount of this histone modification is very important: in cells with too much of the modification , the chromosomes compacted incorrectly and the chromatids separated too soon . As a result , the chromosomes were incorrectly distributed among the daughter cells . Wike et al . also show that an enzyme called Aurora-A kinase is responsible for making this histone modification . The next challenge will be to understand how the Aurora-A kinase knows when and where to add the histone modification to the chromosome . This will help us to understand how the overproduction of Aurora-A causes cancer .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "chromosomes", "and", "gene", "expression" ]	2016	Aurora-A mediated histone H3 phosphorylation of threonine 118 controls condensin I and cohesin occupancy in mitosis
Adult hippocampal neurogenesis provides the dentate gyrus with heterogeneous populations of granule cells ( GC ) originated at different times . The contribution of these cells to information encoding is under current investigation . Here , we show that incoming spike trains activate different populations of GC determined by the stimulation frequency and GC age . Immature GC respond to a wider range of stimulus frequencies , whereas mature GC are less responsive at high frequencies . This difference is dictated by feedforward inhibition , which restricts mature GC activation . Yet , the stronger inhibition of mature GC results in a higher temporal fidelity compared to that of immature GC . Thus , hippocampal inputs activate two populations of neurons with variable frequency filters: immature cells , with wide‐range responses , that are reliable transmitters of the incoming frequency , and mature neurons , with narrow frequency response , that are precise at informing the beginning of the stimulus , but with a sparse activity . The dentate gyrus ( DG ) is the main entrance of information to the hippocampus ( Andersen et al . , 1971 ) . Activity arriving to the hippocampus from the entorhinal cortex ( EC ) should be processed by heterogeneous populations of granule cells ( GC ) of different stages of maturation ( Zhao et al . , 2008; Lepousez et al . , 2015 ) . Neural progenitor cells of the adult subgranular zone give rise to dentate GC that develop and mature over several weeks . Developing GC follow a precise sequence to establish their afferent connectivity and functional maturation . New born neurons are initially contacted by dendritic GABAergic terminals , followed by glutamatergic axons and last by GABAergic perisomatic contacts . In parallel , their membrane resistance decreases and excitability becomes mature ( Espósito et al . , 2005; Ge et al . , 2006 ) . When fully developed , adult-born neurons achieve a functional profile that is indistinguishable from that of all other GC as reflected by their inputs , intrinsic membrane properties , and firing behavior ( Laplagne et al . , 2006 , 2007 ) . In addition , adult-born GC form functional glutamatergic synapses onto DG interneurons and CA3 pyramidal cells ( Toni et al . , 2008; Temprana et al . , 2015 ) , indicating that new neurons receive , process , and convey information onto target neurons and participate in hippocampal function . As activity arrives to the DG , it does not only activate GC but also inhibitory circuits ( Buzsaki , 1984 ) . The interaction of the excitatory activity arriving from EC with the recruited inhibition will ultimately determine GC's activity . Inhibitory circuits can have a profound impact in the processing of afferent activity ( Markwardt et al . , 2009; Isaacson and Scanziani , 2011; Ikrar et al . , 2013 ) ; for example , they can ensure a temporal fidelity of responses ( Pouille and Scanziani , 2001 ) and also modulate neuronal firing in response to afferent inputs ( Pouille et al . , 2009; Dieni et al . , 2013 ) . We have previously demonstrated that due to a smaller and slower inhibition , immature GC are preferentially recruited in response to a single stimulation of its afferent entorhinal pathway ( Marín-Burgin et al . , 2012 ) . However , activity arriving to the hippocampus does not only come as a single stimulus but varies during different brain states in frequencies ranging from theta to gamma ( Buzsáki and Draguhn , 2004; Buzsáki and Moser , 2013 ) . How is this information transformed into activity of the granule cells ? Are immature and mature populations of neurons differentially recruited at different frequencies ? If that was the case , information arriving to DG could be differentially channeled into activation of different populations of granule cells depending on the frequency . In this work , we have addressed these questions focusing on the understanding of how excitation and inhibition interact to determine the response of GC . We found that due to a weaker influence of feedforward inhibitory circuits , immature neurons can respond to a wider range of frequencies than mature neurons , however , they are less time locked to the incoming stimuli . Mature neurons on the contrary show a higher temporal fidelity but are less effective in their responses to frequency . We suggest that immature and mature GC reflect complementary aspects of incoming activity . We investigated how immature GC process afferent activity from entorhinal inputs arriving at different frequencies and how they compare to mature GC in the adult mice hippocampus . We selected 4-week-old neurons as immature GC because this is the earliest stage at which adult-born GC can be reliably activated by an excitatory drive ( Mongiat et al . , 2009 ) , exhibit funtional properties that distinguish them from mature neurons ( Wang et al . , 2000; Snyder et al . , 2001; Schmidt-Hieber et al . , 2004; Espósito et al . , 2005; Ge et al . , 2007; Marín-Burgin et al . , 2012 ) , and are already connected with postsynaptic targets ( Gu et al . , 2012; Marín-Burgin et al . , 2012; Temprana et al . , 2015 ) . Adult-born GC were retrovirally labeled to express RFP or GFP and acute hippocampal slices were prepared 4 weeks post retroviral injection ( 4 wpi ) ( Figure 1A ) . 10 . 7554/eLife . 08764 . 003Figure 1 . Frequency-dependent activation of mature and immature granule cells . ( A ) The image shows a hippocampus slice with 4-weeks-old GC ( 4wpiGC ) expressing RFP ( pseudo-colored in blue ) . Scale bar: 50 µm . The upper timeline indicates the time of retroviral injection . The lower scheme shows the recording configuration: a stimulating electrode was placed in the medial perforant path ( mPP ) to deliver 10 stimuli at different frequencies; the stimulation intensity was kept at 50% fEPSP . Loose patch recordings were obtained from mature GC ( matGC ) and 4wpiGC to detect spikes . ( B ) Raster plots from one matGC ( black ) and one 4wpiGC ( blue ) at 1 Hz , 10 Hz , 20 Hz , and 40 Hz . Each color dash denotes a spike . Columns: stimulation pulses at the four frequencies . Rows: stimulation trains . Lower panels are the average action potential probability at each pulse of the train for all the data from matGC ( black ) and 4wpiGC ( blue ) . ( C ) Average of the sum of action potentials evoked by stimulation trains of 1 Hz , 10 Hz , 20 Hz , and 40 Hz , in matGC ( black bars ) and 4wpiGC ( blue bars ) . Activation decreases with frequency in both cells and is higher in 4wpiGC than in matGC at all stimulation frequencies ( two-way ANOVA , variation between GC: p < 0 . 05; variation in frequency: p < 0 . 001; interaction: ns , p > 0 . 05 ) . N ( 4wpiGC ) = 12 , 13 , 7 , and 11 cells and N ( matGC ) = 15 , 16 , 11 , 14 cells for 1 Hz , 10 Hz , 20 Hz , and 40 Hz , respectively . The frequency range lines at the bottom shows the range of frequencies that GC responded with a number of action potentials significantly different from 1 ( Wilcoxon signed-rank test , at 20 and 40 Hz , p < 0 . 05 for 4wpiGC , p > 0 . 05 for matGC ) . ( D ) Average number of action potentials when stimulus was at 100% spiking of each cell after the first stimulation pulse ( above threshold ) in matGC ( black ) and 4wpiGC ( blue ) at 40 Hz stimulation . ( E ) Upper scheme shows the recording configuration . Whole cell current clamp ( I-Clamp ) recordings were obtained from matGC and 4wpiGC . Synaptic activity was blocked by kynurenic acid ( KYN ) and picrotoxin ( PTX ) . Intrinsic spiking activity was evaluated by injecting 10 depolarizing brief square current pulses at an intensity that evoked action potentials in the 10 pulses at 1 Hz . Cells were tested at frequencies every 20 Hz . The lower graph shows the mean number of action potentials evoked at 1 Hz , 20 Hz , and 40 Hz . All recorded cells in whole-cell fired 10 action potentials when 10 current pulses were delivered at these frequencies ( N = 4 cells for both GC ) . Error bars indicate SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 08764 . 00310 . 7554/eLife . 08764 . 004Figure 1—figure supplement 1 . Input normalization . ( A ) Left , recording configuration . A stimulation electrode ( input ) was placed to stimulate the medial perforant path ( mPP ) ; one field recording electrode placed in the GCL to simultaneously monitor the pop spike and fEPSP . Right , field recordings from the GCL at increasing stimulus intensities . The dash indicates the simulation time ( the stimulation artifact was erased for better visualization of the data ) . The black trace corresponds to the intensity which evokes the maximum pop spike ( 100% pop spike ) . The dotted green line indicates the slope of the fEPSP corresponding to the maximum pop spike , considered 100% fEPSPslope . Bottom , input strength calculation formula: the input strength is the slope of the fEPSP elicited at any given stimulus intensity , normalized to the slope of the fEPSP evoked at the stimulus intensity that evokes a pop spike of maximal amplitude ( 100% pop spike ) . See ‘Materials and methods’ for details . ( B ) Normalized slope of the fEPSP ( input strength ) plotted against the current amplitude applied to the stimulating electrode ( stimulation intensity ) for the experiment illustrated in B . A sigmoid curve was fitted to the data . Inset: linear relationship between the percentage fEPSPslope and the percentage fiber volley amplitude for the example slice; R = 0 . 88 . ( C ) Sigmoid fits from input strength normalization curves for three example slices reveal the need for a normalization procedure to compare physiologically equivalent input intensities among experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 08764 . 00410 . 7554/eLife . 08764 . 005Figure 1—figure supplement 2 . Immature GC are more effective at responding to frequency than matGC . Left , recording configuration . Loose patch recordings were obtained from matGC and 4wpiGC in response to mPP stimulation with 10 pulses at different frequencies . Right , efficacy to reproduce the stimulating frequency with the spiking was calculated as the probability of occurrence of action potentials at the time interval of stimulation . The efficacy decreases with frequency in both matGC and 4wpiGC , and is greater in 4wpiGC than in matGC at all stimulation frequencies ( two-way ANOVA , variation between GC: p < 0 . 01; frequency variation: *p < 0 . 001; interaction: ns , p > 0 . 05 ) . N ( 4wpiGC ) = 12 , 13 , 7 , and 11 cells and N ( matGC ) = 15 , 16 , 11 , 14 cells for 1 Hz , 10 Hz , 20 Hz , and 40 Hz , respectively . Data expressed as average ±SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 08764 . 005 To investigate how 4-week-old ( 4wpiGC ) and mature ( matGC ) GC respond to conditions of activity that resemble those occurring during hippocampus-dependent behaviors ( Leutgeb et al . , 2007 ) , we measured spiking in response to trains of different frequencies delivered to the medial perforant pathway ( mPP ) . We used electrophysiological loose patch recordings to characterize the activation profile of GC at the single cell level ( Figure 1A ) . The intensity of the stimulus delivered to mPP was normalized using field recordings to compare physiologically equivalent inputs across slices and a fixed stimulation intensity ( 50% fEPSPslope ) was chosen for all experiments ( Figure 1—figure supplement 1 ) . Five trains of 10 pulses were delivered at 1 , 10 , 20 , and 40 Hz for each cell ( Figure 1B ) . Immature GC activate with a higher probability than mature GC at all frequencies , yet both GC populations showed decreased responsiveness for trains delivered at increasing frequency ( Figure 1B ) . Activation of mature and immature GC decreased within each train , and this reduction was more pronounced for the higher frequencies ( Figure 1B ) . To quantitatively compare activation between mature and immature GC across frequencies , we calculated the average spike number for all 10 pulses for each frequency . Results indicate that the activation of GC decreases drastically with frequency . However , immature GC spiked more reliably than mature GC at all frequencies ( Figure 1C ) . Stimulation at 1 Hz represents a situation similar to applying a single pulse; accordingly , spiking probability in matGC was around 50% ( 5 spikes average ) , consistent with the responses evoked using stimulation intensities of about 50% of the fEPSP and with previous results ( Marín-Burgin et al . , 2012 ) . While 4wpiGC significantly responded with more than 1 spike to all frequencies , matGC only responded with more than 1 spike to 1 and 10 Hz , indicating that matGC could not be synaptically activated to fire at 20 and 40 Hz ( Figure 1C ) . Furthermore , matGC could not spike two consecutive action potentials even if the stimulation intensity was chosen to record neurons above threshold ( Figure 1D ) . The higher activation levels of 4wpiGC at all frequencies indicate that they could reproduce a wider range of frequencies than matGC . These data indicate that dentate GC have frequency filters with variable gain depending on their age . The ability of GC to spike at a given frequency reflects its ability to re-transmit time information to their target neurons . To analyze this aspect , we calculated the efficacy to reproduce each frequency . This was computed as the average of the probability of occurrence of action potentials in the same frequency range as that given by the stimulus . The results show that 4wpiGC have higher efficacy than matGC at all frequencies ( Figure 1—figure supplement 2 ) . This observation indicates that the population of immature GC has a higher capability to retransmit temporal information to their postsynaptic targets than the population of mature neurons . However , 4wpiGC and matGC display lower spiking efficacy at higher frequencies , indicating that both neuronal populations act as low pass filters . The decreased activation with frequency could simply be due to intrinsic physiological restrictions of GC or instead to characteristics of the activated circuit . Therefore , we studied the intrinsic capacity of 4wpiGC and matGC to fire action potentials at 20 and 40 Hz in response to injection of current pulses . Whole-cell current-clamp recordings were obtained from 4wpiGC and matGC , and trains of square current pulses were injected at increasing frequencies . All recorded GC started showing failures at frequencies near 100 Hz . Importantly , both 4wpiGC and matGC consistently followed 20 and 40 Hz ( Figure 1E ) . These results indicate that the decrease in GC’ activation observed when the mPP was stimulated with increasing frequencies was not due to an intrinsic incapacity of GC to spike but , rather , to properties of the activated circuit . To understand the mechanisms underlying the differential filtering found in matGC vs 4wpiGC , we studied the individual circuit components . Activation of mPP axons not only produces monosynaptic glutamatergic excitation of GC but also recruits feedforward GABAergic inhibitory circuits , which can modulate neuronal firing in response to afferent inputs ( Buzsaki and Eidelberg , 1981; Pouille et al . , 2009; Ewell and Jones , 2010; Marín-Burgin et al . , 2012 ) . Thus , we evaluated the effect of blocking GABAergic inhibition with picrotoxin ( PTX ) in the response of 4wpiGC and matGC to stimulation of the mPP with 10 pulses at 1 , 10 , 20 , and 40 Hz ( Figure 2A ) . Notably , in the absence of inhibition activation of 4wpiGC vs matGC was similar at all frequencies ( Figure 2B ) . This observation indicates that the inhibitory circuit is responsible for generating activation differences among 4wpiGC and matGC after train stimulation of the mPP . 10 . 7554/eLife . 08764 . 006Figure 2 . GABAergic inhibition generates the difference between mature and immature GC . ( A ) Left , the scheme shows the recording configuration , a stimulating electrode was placed in the medial perforant path ( mPP ) to deliver 10 stimuli at different frequencies; the stimulation intensity was kept at 50% fEPSP . Loose patch recordings were obtained from mature GC ( matGC ) and 4wpiGC to detect spikes in the presence of picrotoxin ( PTX ) . Right , Raster plots from one matGC ( in black ) and one 4wpiGC ( in blue ) at 1 Hz , 10 Hz , 20 Hz , and 40 Hz . Each color line denotes a spike . Responses of neurons were recorded in PTX . ( B ) Average of the sum of action potentials evoked by stimulation trains of 1 Hz , 10 Hz , 20 Hz , and 40 Hz , in matGC ( gray columns ) and 4wpiGC ( light blue columns ) in the presence of PTX . Activation slightly decreases with frequency in both cells , but there were not significant differences among matGC and 4wpiGC ( two-way ANOVA , variation between GC: ns; variation in frequency: p < 0 . 05; interaction: ns , p > 0 . 05 N = 23 cells for both GC at the four frequencies ) . ( C ) Comparison of matGC activation to stimulation with trains in control conditions and when inhibition was blocked with PTX . Activation of matGC was higher in the presence of PTX at all frequencies ( two-way ANOVA , variation between treatments: *p < 0 . 001; frequency variation . p < 0 . 01; interaction: ns , p > 0 . 05 ) . ( D ) Comparison of 4wpiGC activation to stimulation with trains in control conditions and when inhibition was blocked with PTX . The effect of blocking inhibition in 4wpiGC varies with frequency . Activation increases only at high stimulation frequencies ( two-way ANOVA variation between treatments: p < 0 . 05; variation between frequencies: p < 0 . 05; positive interaction p < 0 . 01; at 40 Hz , p < 0 . 01 , Bonferroni post-test ) . Error bars indicate SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 08764 . 006 Activation of matGC is strongly controlled by inhibition , since their activity significantly increases with PTX at every frequency ( Figure 2C ) . Interestingly , activation levels in 4wpiGC did not change with the blockage of inhibition at 1 and 10 Hz , indicating that inhibitory circuits do not control activity in immature neurons when stimuli arrive every 100 ms or more . At 20 and 40 Hz however , when the separation of stimuli is 50 ms or less , the recruitment of inhibitory circuits does restrict the response of immature neurons ( Figure 2D ) . These results are extremely interesting because they indicate that there is a different temporal window in which inhibition restricts mature and immature GC activation . Immature neurons have a wider time window during which excitation can elicit spikes without being affected by inhibition . The differential effect of inhibition on immature and mature GC explains the wider range of responses of immature GC to stimuli arriving at frequencies ranging from 1 to 40 Hz , as compared to mature neurons which rarely reproduce high frequencies ( Figure 1C ) , demonstrating that each population have different gains to filter frequency . In addition , the results reveal that inhibitory circuits generate a wide variability of responses in the population of GC , since not only are they responsible for the differences in activation between 4wpiGC and matGC , but also for the frequency dependence of the activation . To study how the recruited inhibition affects the activation profiles of 4wpiGC and matGC , we investigated the precise contribution of excitatory and inhibitory components that control the activation of GC at different frequencies . We performed whole-cell voltage-clamp recordings to measure excitatory and inhibitory responses elicited at each stimulation frequency at a fix intensity as wherein before . Activation of mPP produced excitatory postsynaptic currents ( EPSC ) and inhibitory postsynaptic currents ( IPSC ) , indicating that glutamatergic entorhinal axons directly activate immature and mature GC and also recruit inhibitory interneurons that synapse onto both populations ( Figure 3A ) . In order to assess the contribution of the IPSC and EPSC to the activation of GC , we analyzed the excitatory postsynaptic conductance ( EPSG ) and inhibitory postsynaptic conductance ( IPSG ) at a time close to the generation time of an action potential ( Figure 3A ) . Comparison of latencies of evoked action potentials and EPSC peaks show that these events occur close in time ( Figure 3B ) . Therefore , we quantified the peak EPSG and IPSG value at the peak of the EPSC at all frequencies ( Figure 3A–E ) . 10 . 7554/eLife . 08764 . 007Figure 3 . Interaction between excitation and inhibition determines the response of GC to frequency . ( A ) Top , the schemes show the recording configuration . Whole-cell voltage-clamp recordings were obtained from matGC and 4wiGC in response to stimulation of mPP with trains at different frequencies . Bottom , example of recordings . IPSCs were recorded at the reversal potential of excitation ( −60 mV ) , and EPSCs were recorded at the reversal potential of inhibition ( 0 mV ) . The dash indicates the time of the stimulus . In subsequent quantifications , excitation and inhibition were measured at the peak of the EPSC , marked with blue in the trace . Traces in black correspond to matGC and in blue to 4wpiGC . ( B ) Latency to action potentials compared to latency of the peak EPSC measured from the stimulating artifact . The action potentials and the peak of the EPSC occur close in time . ( C ) Average latency to reach the 20% of the peak IPSC evoked by the first stimulation pulse at 1 Hz in matGC ( black ) and 4wpiGC ( blue ) . The IPSC is slower in 4wpiGC than in matGC ( **p < 0 . 001 , t test ) . ( D ) Average excitatory ( EPSG ) and inhibitory ( IPSG ) conductance evoked by each pulse in the train at 1 Hz , 10 Hz , 20 Hz , and 40 Hz . The values were calculated as indicated in A . For comparison , the spiking probability for each pulse in the train obtained from Figure 1A is plotted as the gray shadow in the back . The insets show representative current traces obtained for each frequency . Dashes indicate stimuli . ( E ) Average of the mean of excitatory and inhibitory conductances evoked by the 10 pulses in matGC and 4wpiGC . Excitation decreases with frequency and is higher in matGC than in 4wpiGC at all frequencies of stimulation ( two-way ANOVA paired between frequencies; variation between GC: p < 0 . 05; variation between frequencies . *p < 0 . 001; interaction: ns , p > 0 . 05; N ( 4wpiGC ) = 17 cells N ( matGC ) = 14 cells at the four frequencies ) . Inhibition increases with frequency and is higher in matGC than in 4wpiGC that at all frequencies of stimulation ( two-way ANOVA paired between frequencies; variation between GC: p < 0 . 001; frequency variation . p < 0 . 001; interaction: ns , p > 0 . 05 . N ( 4wpiGC ) = 9 cells and N ( matGC ) = 11 cells at the four frequencies ) . For comparison , the average number of action potentials is plotted as a gray shadow in the back . Bottom: ratios between mean EPSG and IPSG evoked along train stimulation at 1 , 10 , 20 , and 40 Hz for 4wpiGC ( blue ) and matGC ( black ) . Dotted lines show the switch between higher excitation ( above ratio = 1 ) and inhibition ( below ratio = 1 ) balance . Error bars indicate SEM . Stimulation artifacts were erased from traces for better visualization . DOI: http://dx . doi . org/10 . 7554/eLife . 08764 . 00710 . 7554/eLife . 08764 . 008Figure 3—figure supplement 1 . Excitation/inhibition ratios evoked by train stimulation . Top , recording configuration . Bottom , ratios of EPSG and IPSG shown in Figure 3D , evoked after each pulse of train stimulation at 1 , 10 , 20 , and 40 Hz for 4wpiGC ( blue ) and matGC ( black ) . Dotted lines show the switch between higher excitation ( above ratio = 1 ) and inhibition ( below ratio = 1 ) balance . N ( 4wpiGC ) = 9–18 cells and N ( matGC ) = 11–15 cells . Error bars indicate SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 08764 . 00810 . 7554/eLife . 08764 . 009Figure 3—figure supplement 2 . Contribution of residual and recruited inhibition in immature and mature GC . ( A ) Top , recording configuration . Bottom , example trace of an IPSC to show the measurement of residual and recruited inhibition: residual inhibition ( gray arrows ) was measured as the remaining inhibition at 25 ms , 50 ms , 100 ms , and 1000 ms ( corresponding to 40Hz , 20 Hz , 10 Hz , and 1 Hz respectively ) , after the first stimulation pulse at 1 Hz . Recruited inhibition ( black arrows ) was measured as the inhibition evoked after the stimulation pulse at the spike time ( time of the peak EPSC ) indicated with a black dash and dotted lines . ( B ) Average residual inhibition at the times related to the studied frequencies . Average recruited inhibition was plotted on top of each value and was obtained from the mean inhibition measured at 1 Hz . The red line depicts the mean EPSG of matGC and 4wpiGC . Error bars indicate SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 08764 . 009 Stimulation of the mPP evokes slower IPSC in 4wpiGC than in matGC ( Figure 3C ) , and as a result inhibition around action potential time is weak in 4wpiGC at frequencies were stimuli are at least 100 ms apart ( at 1 Hz and 10 Hz ) ( Figure 3D ) . The slower IPSC rising kinetics results in a delayed inhibition that poorly influences spiking of 4wpiGC at low frequencies . IPSG evoked at 1 to 10 Hz is maintained under EPSG values in both matGC and 4wpiGC . However , IPSG values in immature neurons are very small , resulting in a higher excitation to inhibition balance than that of mature neurons , which explains their higher spiking probability ( Figure 3—figure supplement 1 ) . At 20 and 40 Hz , IPSG values increase over EPSG along the stimulation train , restricting the ability of the neurons to spike ( Figure 3D and Figure 3—figure supplement 1 ) . Interestingly , in 4wpiGC the IPSG exceeds the EPSG later in the train compared to matGC , which might explain the higher efficacy of 4wpiGC to transmit spike trains within the first pulses of the train . To make quantitative comparisons between the currents evoked in 4wpiGC and matGC at different frequencies , we calculated the mean peak EPSG , and the mean IPSG value at the peak EPSG , for the 10 pulses at each frequency . Thus , we obtained a single value of the EPSG and a single value of the IPSG , for 4wpiGC and matGC at each frequency . These data show that the evoked inhibition in both populations of GC increases with frequency . Excitation , on the contrary , slightly decreases ( Figure 3E ) . This slight decrease in the total excitation with the frequency explains the slight decrease in the activation of GC when inhibition was blocked ( Figure 2B ) . On the other hand , the large increase in inhibition with frequency , explains the large differences found in the activation of GC among frequencies in the conditions in which the circuit was intact ( Figure 1C ) . Notably , inhibition particularly increases at 20 Hz and 40 Hz , and this coincides with the greatest decrease in the activation of GC . Both excitatory and inhibitory conductances are smaller in 4wpiGC than in matGC ( Figure 3E ) , which reflects that these connections are not fully mature at this developmental stage , as previously shown ( Marín-Burgin et al . , 2012 ) . Nonetheless , differences in inhibition between immature and mature neurons are greater than in excitation , giving higher mean excitation/inhibition ratios in 4wpiGC than matGC at 1 and 10 Hz ( Figure 3E ) . At 20 Hz and 40 Hz higher ratios are only observed within the first pulses of the train , where spiking differences reside ( Figure 3D , Figure 3—figure supplement 1 ) . Therefore , higher excitation/inhibition ratios in 4wpiGC explain their higher levels of activation after train stimulation . On the other hand , in the absence of inhibition both 4wpiGC and mature GC display similar firing properties . This similarity results from the balance between a strong excitatory drive combined with a low input resistance in mature GC , and a weak excitation matched to high resistance in 4wpiGC . The latter combination facilitates reaching spiking threshold with small postsynaptic currents ( Mongiat et al . , 2009 ) . The different rising kinetics of the recruited inhibition between matGC and 4wpiGC determines different time windows in which inhibition restricts excitation . To better understand the extent to which this difference affects activation as a function of frequency , we calculated the residual inhibition at different periods corresponding to the studied frequencies , and added the recruited inhibition observed on average at 1 Hz on top ( Figure 3—figure supplement 2 ) . The results clearly show that due to the faster kinetics of the IPSC in matGC , the contribution of the recruited inhibition in matGC is higher than in 4wpiGC . Therefore , the sum of residual plus recruited inhibition in matGC is higher than excitation even if stimuli are 100 ms apart ( restricting their responses to 10 Hz stimulation ) , whereas 4wpiGC have a wider temporal window in which inhibition does not overcome excitation ( Figure 3—figure supplement 2 ) . Overall , these results show that the higher activation levels of 4wpiGC are due to a higher balance between the excitation and inhibition , which is mainly generated by a smaller inhibition at the time around spike generation . The above findings show that inhibitory circuits are responsible for generating differences in the responsiveness to different frequencies between mature and immature GC . Therefore , we focused on the understanding of which components of the activated inhibitory circuit are responsible for the observed differences . As we have mentioned , activation of mPP not only produces excitation on GC but also activates inhibitory circuits in a feedforward manner . In addition , activation of GC could also recruit feedback inhibition that could influence their activity , especially during trains of stimuli . The contribution of each type of inhibition could vary depending on the train frequency . To study the specific contribution of feedforward and feedback inhibition to the total inhibition recruited with trains of stimulation , whole-cell voltage-clamp experiments were performed in 4wpiGC and matGC , and trains were applied to the mPP at different frequencies . EPSC and IPSC were recorded in control conditions and in the presence of DCG4 , an agonist of group II metabotropic glutamate receptors ( mGluR2/3 ) that reduces release probability in mossy fiber terminals ( Kamiya et al . , 1996 ) . Application of DCG4 abolished feedback inhibition , and the remaining inhibition corresponded then to feedforward inhibition ( IPSC-FF ) . The contribution of feedback inhibition ( IPSC-FB ) was then calculated by subtraction of the IPSC-FF to total IPSC ( Figure 4A , Figure 4—figure supplement 1 ) . 10 . 7554/eLife . 08764 . 010Figure 4 . Feedforward inhibition generates differences between mature and immature GC . ( A ) Recording configuration showing the activated circuit . Stimulation of mPP-activated monosynaptic excitation and feedforward inhibition ( FF I , green ) . Feedback inhibition ( FB I , purple ) is recruited by the activity of GC . Right traces show inhibitory currents recorded in whole cell , voltage clamping the cell at the reversal potential of excitation ( ∼0 mV ) . The recorded total inhibitory post-synaptic current ( total IPSC ) obtained under control conditions is in gray . Application of DCG4 abolished feedback IPSC leaving only the feedforward IPSC ( FF-IPSC , green trace ) . Feedback IPSC ( FB-IPSC , purple trace ) was measured by subtraction of the FF-IPSC from the total IPSC . ( B ) Left , traces from representative recordings of FF-IPSC and FB-IPSC from matGC and 4wpiGC . FF-IPSC in 4wpiGC is slower . The dash indicates the time of the stimulus . Right , average latency to reach 20% of the peak IPSC evoked by the first stimulation pulse at 1 Hz for FF and FB inhibition in matGC and 4wpiGC . The IPSC-FF is slower in 4wpiGC than in matGC ( p < 0 . 001 , t test ) . The latency of the IPSC-FB did not show significant differences between 4wpiGC and matGC ( ns , p > 0 . 05 , t test ) . ( C ) Left , mean FF inhibitory postsynaptic conductance ( IPSG-FF ) evoked by stimulation of 10 pulses to mPP measured at the peak of the EPSC in matGC and 4wpiGC . The IPSG-FF increases with frequency and is higher in matGC than in 4wpiGC at all frequencies of stimulation ( two-way ANOVA paired between frequencies; variation between GC: p < 0 . 01; frequency variation . p < 0 . 001; interaction: ns , p > 0 . 05 ) . Right , mean FB inhibitory postsynaptic conductance ( IPSG-FB ) evoked by stimulation of 10 pulses to mPP measured at the peak of the EPSC in matGC and 4wpiGC . The IPSG-FB shows no differences between matGC and 4wpiGC , but increases in both GC with frequency ( two-way ANOVA paired between frequencies; variation between GC: ns , p > 0 . 05; variation between frequencies . *p < 0 . 001; interaction: ns , p > 0 . 05 ) . N ( 4wpiGC ) = 6–7 cells , N ( matGC ) = 10–11 cells , in the four frequencies . Error bars indicate SEM . Stimulation artifacts were erased from traces for better visualization . DOI: http://dx . doi . org/10 . 7554/eLife . 08764 . 01010 . 7554/eLife . 08764 . 011Figure 4—figure supplement 1 . Feedforward and feedback inhibition recruited by train stimulation . Left: recording configuration showing the activated circuit . Stimulation of mPP activated monosynaptic excitation and feedforward inhibition ( FF I , green ) . Feedback inhibition ( FB I , purple ) is recruited by the activity of GC . Right: representative traces show feedforward inhibitory postsynaptic currents ( FF-IPSC , green ) and feedback IPSC ( FB-IPSC , purple ) recorded in whole-cell voltage clamp at ∼0 mV , after train pulses at 10 and 40 Hz , for a 4wpiGC and a matGC , as shown in Figure 4A . Dashes indicate stimuli . Stimulation artifacts were erased from traces for better visualization . DOI: http://dx . doi . org/10 . 7554/eLife . 08764 . 01110 . 7554/eLife . 08764 . 012Figure 4—figure supplement 2 . Response of PV+ interneurons to mPP stimulation is not affected with DCG4 . ( A ) Left , the scheme shows the crossing to obtain adult PVCre; CAGfloxStopTom ( PV-Tom ) mice . Right , confocal images of dentate gyrus ( DG ) from hippocampal slices obtained from PV-Tom animals . ( B ) Left , recording configuration . A stimulating electrode was placed to stimulate mPP and loose patch recordings were performed in PV-Tom+ interneurons from the DG . DIC image with a fluorescent PV-Tom interneuron with a loose patch pipette and delineation of the GCL . Scale bar: 50 µm . Right , representative traces showing the evoked spikes from a PV interneuron in control conditions and after the application of DCG4 at 50% fEPSP input strength . The graph shows the spiking probability in control conditions and after the application of DCG4 for five PV interneurons . The spiking probability does not change with DCG4 ( ns , paired t test , p > 0 . 05 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 08764 . 012 Quantification of the kinetics of the IPSC-FF and IPSC-FB recruited by the stimulation of the first pulse at 1 Hz reveals that the IPSC-FB is recruited after the IPSC-FF and after the peak of the EPSC ( Figure 4B ) . Notably , the IPSC-FF is slower in 4wpiGC than in matGC . Conversely , the IPSC-FB does not show significant differences between 4wpiGC and matGC . To compare IPSC-FF and IPSC-FB between matGC and 4wpiGC , we calculated the mean of the IPSG-FF and the IPSG–FB of the 10 pulses at the peak of the EPSC ( Figure 4C ) . Interestingly , the results reveal differences between matGC and 4wpiGC only in the IPSG-FF , which is higher in the four studied frequencies of stimulation . However , the FB-IPSG does not differ between matGC and 4wpiGC at any frequency . These differences could be explained by the observed differences in the kinetics of FFI but not FBI . Since DCG4 can also affect neurotransmitter release in the mPP pathway ( Macek et al . , 1996 ) , and therefore , modify recruitment of inhibitory neurons in a feedforward manner , we tested the possibility that the recorded FF-IPSC in GC could be affected by DCG4 . Parvalbumin ( PV ) interneurons are known to respond to the PP pathway and participate in feedforward inhibition ( Sambandan et al . , 2010 ) . We studied the effect of DCG4 on the activation ( spiking ) of PV interneurons using hippocampal slices from PV-Tom mice ( Hippenmeyer et al . , 2005 ) . As expected , PV interneurons already showed responses at very low stimulation intensities ( 10% fEPSPslope ) . Thus , activation in response to stimulation of the mPP at the intensities used in the experiments ( 50% fEPSPslope ) was well above threshold and unaltered by the application of DCG4 ( Figure 4—figure supplement 2 ) . The above results indicate that the differences in the recruitment of mature and immature neurons are dictated specifically by feedforward inhibition and are given by the fact that the IPSC-FF is slower in 4wpiGC compared to matGC . However , both IPSG-FB and IPSG-FF increase with frequency , indicating that both types of inhibition contribute to generate differences in activation among frequencies . Immature neurons are better followers of spike trains arriving to the mPP due to a weak influence of inhibition . However , a weak and slower inhibition can lead to increased temporal variation in the responses , since temporal fidelity is ensured by inhibition ( Pouille and Scanziani , 2001 ) . The slower inhibition in immature neurons would allow a wider temporal window in which neurons can summate excitation to spike . Therefore , we compared the latency and the jitter of action potentials evoked in matGC and 4wpiGC in response to the first pulse of each stimulus ( Figure 5A ) . 10 . 7554/eLife . 08764 . 013Figure 5 . Temporal fidelity in mature and immature GC ( A ) Recording configuration . Left , the mPP was stimulated at an intensity of 50% fEPSP . Recordings were performed from 4wpiGC and matGC in loose patch configuration . Right , representative traces of evoked action potentials in response to stimulation of the mPP . The latency from the stimulation artifact to action potentials ( asterisks ) and spiking jitter and were calculated for matGC ( black ) and 4wpiGC ( blue ) . ( B ) Top , latencies measured in 100 spikes from matGC ( black ) and 4wpiGC ( blue ) . Bottom , distribution of latencies binned along 20 bins . The red arrow denotes the average pop spike time of all included experiments ( 4 . 36 ± 0 . 19 ms ) . ( C ) Top , jitter measured as the time of action potential occurrence relative to the mean spike timing of 3–5 trials . Bottom , jitter distribution in matGC and 4wpiGC; solid lines represent a normal fitting lines . Distribution is wider in 4wpiGC than in matGC ( p < 0 . 001 , two-sample F-test for equal variances ) . N ( matGC ) = 114 spikes from 8 cells; N ( 4wpiGC ) = 96 spikes from 6 cells . Std: standard deviation . ( D ) Jitter distribution in matGC and 4wpiGC in the presence of picrotoxin ( PTX ) ; solid lines represent normal fitting lines . Distribution is wider in 4wpiGC than in matGC ( p < 0 . 001 , two-sample F-test for equal variances ) and is wider in the presence of PTX for both matGC ( p < 0 . 05 , two-sample F-test for equal variances ) and 4wpiGC ( p < 0 . 001 , two-sample F-test for equal variances ) . N ( matGC ) = 142 spikes of 9 cells; N ( 4wpiGC ) = 170 spikes from 10 cells . DOI: http://dx . doi . org/10 . 7554/eLife . 08764 . 013 Indeed , mature neurons show a time locked response to afferent stimulation , since the majority of mature cells is activated after 3 ms of the stimulation of the mPP and all neurons tend to spike around the time of the population spike ( pop spike ) ( Figure 5B ) . Immature neurons , on the contrary , show longer latencies ( most cells spike after the pop spike ) and broader distribution of latencies to spike than mature neurons . In addition , 4wpiGC also show a higher variance of their spike times than mature neurons . The little jitter of matGC indicates that this population of neurons has a higher temporal fidelity ( Figure 5C ) . In the absence of inhibition , both populations increased their spike time variability , indicating that inhibition restricts their temporal responses ( Figure 5D ) . Even in the absence of inhibition mature neurons show more precise activation which could be given by the nature of the excitatory input , which is stronger in mature GC than in immature GC . On the other hand , immature neurons have slower membrane time constant ( Mongiat et al . , 2009 ) that could also contribute to their higher spiking jitter . These results suggest a division of labor in the populations of immature and mature GC , in which immature neurons respond with higher efficacy to the frequency of the afferent inputs but with imprecise responses to the timing of the afferent spike , while mature neurons respond with a narrow tuning to the time of the afferent inputs , though with less efficiency to respond to frequency . Numerous studies have shown that brain activity occurs with certain temporal patterns and proposed that the timing of the signals is itself a code ( Ahmed and Mehta , 2009; Singer , 2009 ) . Extracellular field potentials recorded in vivo present oscillations that change with different brain states ( Womelsdorf et al . , 2014 ) . For example , layer II of the EC and the DG of the hippocampus , tend to oscillate in frequencies theta ( ∼7 Hz ) and gamma ( ∼40 Hz ) ( Roux and Buzsaki , 2015 ) . Neurons from EC , the main afferents to DG , are activated by brief bursts of frequencies in the gamma order spaced by intervals in the order of theta frequency ( Chrobak and Buzsaki , 1998; Fyhn et al . , 2004; Hafting et al . , 2005; Bonnevie et al . , 2013 ) . Our results indicate that differences in the activation frequency of afferent MEC inputs arriving to DG differentially activate immature and mature GC . Although both populations of neurons tend to behave as low pass filters , immature neurons are more efficient in responding to trains ( Figure 1 ) . In fact , immature neurons would be able to spike with more than two spikes in a theta cycle , whereas mature neurons would not . The implications of this result are important also in terms of the effect that GC could have in CA3 . It has been shown that direct stimulation of one GC is capable of activating a pyramidal cell in CA3; however , one action potential is not enough to cause activation of the pyramidal cell , but it requires at least two , and the activation probability increases exponentially with the frequency of GC stimulation ( Henze et al . , 2002 ) . Because the connectivity of 4wpiGC in CA3 is very similar to the connectivity of matGC ( Temprana et al . , 2015 ) , our results suggest that immature GC could be more efficient in transmitting high frequency stimuli which can cause a robust activation of CA3 pyramidal cells . A higher probability of efferent activation by immature GC may have a role to reinforce these connections , since 4wpiGC also have greater levels of LTP in these synapses ( Gu et al . , 2012 ) , suggesting that perhaps immature neurons could select networks in CA3 that correspond to a particular moment in which those GC are still immature . Local inhibitory circuits are primarily responsible for differences in activation generated by trains of stimuli . The large increase in inhibition at high frequencies determines very low levels of activation . Thus , due to inhibitory circuits , the activity arriving to DG at different frequencies can be differentially decoded by mature and immature GC . We observed that at low frequencies , feedforward inhibition plays an important role in controlling activation whereas feedback inhibition is very low . As frequency increases , feedback inhibition also increases and adds to the control of activity at high frequencies . Despite the difference in intrinsic properties , activation differences between immature and mature GC are entirely generated by inhibition . The relation between recruited excitation and inhibition is crucial to determine GC activation ( Marín-Burgin et al . , 2012; Dieni et al . , 2013 ) . Therefore , the slow feedforward inhibition observed in immature neurons is essential to determine differences in activation between them and mature neurons . The time interval between trains of pulses at 1 and 10 Hz stimulation is such that the delayed IPSC in 4wpiGC does not compensate the EPSC , and thus inhibition at these frequencies does not affect the activation of immature cells . This could have important functional consequences in light of studies suggesting that activation of DG is set to frequencies near theta ( Jung and McNaughton , 1993; Leutgeb et al . , 2007; Neunuebel and Knierim , 2012 ) . At frequencies higher than 10 Hz , the time interval between pulses is such that the IPSC recruited pulse to pulse can summate and exceedingly compensate the EPSC in both mature and immature GC . Additionally , the growth of inhibitory currents in GC at high frequencies could be the result of specific recruitment of different types of interneurons affecting equally both GC . A recently published study in mature GC shows that dendritic inhibition is preferentially recruited with high frequency stimulation and grows along the train of pulses . Instead perisomatic inhibition is recruited at low and high frequencies , and decreases within a train ( Liu et al . , 2014 ) . In the same line , we have previously shown that dendritic inhibition is very similar between mature and immature neurons , but perisomatic inhibition is bigger and faster in mature GC ( Marín-Burgin et al . , 2012 ) . In light of these results and since we have found that the main difference between mature and immature GC resides in feedforward inhibition , it is highly possible that feedforward inhibition has an important perisomatic component , like the one PV interneurons exert . The slow inhibition on 4wpiGC could be a consequence of the immature nature of perisomatic inhibitory contacts from PV interneurons , which would require future investigation . Also , there could be a differential contribution of mossy cells , which can affect inhibition and excitation on GC ( Chancey et al . , 2014 ) . Our results show , in addition , that spikes in mature neurons are more time-locked to the incoming input than immature neurons , partly due to inhibition , but also probably to the stronger excitatory synapses they receive , as has been suggested . In vivo experiments , recording in the DG in animals with reduced neurogenesis show that they have reduced number of neurons with spiking broadly tuned to theta cycle , and most cells show a narrow tuning to a specific phase of theta , suggesting that immature cells could presumably contribute to the broadly tuned population . This notion may have functional implications in the influence that mature and immature neurons could have on CA3 ( Rangel et al . , 2013 ) . It has been proposed that there are two coding strategies in the nervous system ( Harris and Mrsic-Flogel , 2013 ) , a sparse code , in which information is encoded at any instant by the spiking of a small subset of neurons within the population , in general with low mean firing rates and high selectivity , that could increase information storage . And a dense code , in which most neurons are active at any moment and information is encoded by variations in firing rate . Our results suggest that mature and immature GC could represent these two different codes coexisting in the same structure and presumably serving at different functions . While mature GC could specifically represent a particular input , immature GC , with a higher ability to drive CA3 , could be generating networks in CA3 that would presumably represent a particular experience occurring while they are immature . A replication-deficient retroviral vector based on the Moloney murine leukemia virus was used to express RFP or GFP under a CAG promoter ( Marín-Burgin et al . , 2012 ) . Retroviral particles were assembled using three separate plasmids containing the capside ( CMV-vsvg ) , viral proteins ( CMV-gag/pol ) , and the transgene ( CAG-RFP or CAG-GFP ) . Plasmids were transfected onto HEK 293T cells using deacylated polyethylenimine . Virus-containing supernatant was harvested 48 hr after transfection and concentrated by two rounds of ultracentrifugation . Virus titer was typically ∼105 particles/μl . Female C57Bl/6J mice 6–7 weeks of age were housed at 4 mice per cage , with two running wheels . Running wheel housing started 2–4 days before surgery and continued until the day of slice preparation . For surgery , mice were anesthetized ( 150 µg ketamine/15 µg xylazine in 10 µl saline/g ) , and virus ( 1 µl at 0 . 125 µl/min ) was infused into the dorsal area of the right DG using sterile microcapillary calibrated pipettes and stereotaxic references ( coordinates from bregma: −2 mmanteroposterior , -1 . 5 mm lateral , −1 . 9 mm ventral ) . Animals were killed for acute slice preparation 4 weeks after the surgery . To generate PvalbCre;CAGFloxStopTom ( PV-Tom ) mice , B6;129P2-Pvalbtm1 ( cre ) Arbr/J ( PVcre ) mice ( Hippenmeyer et al . , 2005 ) were crossed to B6 . Cg-Gt ( ROSA ) 26Sortm14 ( CAG-tdTomato ) Hze/J ( Ai14 ) conditional reporter mice . Animals heterozygous for Cre and tomato were used . Experimental protocols were approved by the Institutional Animal Care and Use Committee Leloir Foundation ( Protocols Number 2009 08 37 and 64/2015 , IACUC , Leloir Institute Foundation ) according to the Principles for Biomedical Research involving animals of the Council for International Organizations for Medical Sciences and provisions stated in the Guide for the Care and Use of Laboratory Animals . Mice were anesthetized and decapitated at 4 weeks post injection ( wpi ) . Brains were removed into a chilled solution containing ( mM ) 110 choline-Cl− , 2 . 5 KCl , 2 . 0 NaH2PO4 , 25 NaHCO3 , 0 . 5 CaCl2 , 7 MgCl2 , 20 dextrose , 1 . 3 Na+-ascorbate , 3 . 1 Na+-pyruvate , and 4 kynurenic acid . The right hippocampus was dissected and slices ( 400 µm thick ) were cut transversally to the longitudinal axis in a vibratome and transferred to a chamber containing artificial cerebrospinal fluid ( ACSF; mM ) : 125 NaCl , 2 . 5 KCl , 2 . 3 NaH2PO4 , 25 NaHCO3 , 2 CaCl2 , 1 . 3 MgCl2 , 1 . 3 Na+-ascorbate , 3 . 1 Na+-pyruvate , and 10 dextrose ( 315 mOsm ) . Slices were bubbled with 95% O2/5% CO2 and maintained at 30°C for >1 hr before experiments started . Salts were acquired from Sigma-Aldrich ( St . Louis , MO ) . Recorded neurons were visually identified by fluorescence and infrared DIC videomicroscopy . The mature neuronal population encompassed RFP− neurons localized in the outer third of the GCL ( Mongiat et al . , 2009 ) . Whole-cell recordings were performed using microelectrodes ( 4–5 MΩ ) filled with ( in mM ) 130 CsOH , 130 D-gluconic acid , 2 MgCl2 , 0 . 2 EGTA , 5 NaCl , 10 HEPES , 4 ATP-tris , 0 . 3 GTP-tris , 10 phosphocreatine . In experiments where the intrinsic responses to current pulses were evaluated , a potassium gluconate internal solution was used ( in mM ) : 120 potassium gluconate , 4 MgCl2 , 10 HEPES buffer , 0 . 1 EGTA , , 5 NaCl , 20 KCl , 4 ATP-tris , 0 . 3 GTP-tris , and 10 phosphocreatine ( pH = 7 . 3; 290 mOsm ) . Loose-patch recordings were performed with ACSF-filled patch pipettes ( 5–6 MΩ ) . Field recordings were performed using patch pipettes ( 5 MΩ ) filled with 3 M NaCl . Recordings were obtained using Axopatch 200B and Multiclamp 700B amplifiers , ( Molecular Devices , Sunnyvale , CA ) , digitized , and acquired at 20 KHz onto a personal computer using the pClamp10 software . Membrane capacitance and input resistance were obtained from current traces evoked by a hyperpolarizing step of 10 mV . Series resistance was typically 10–20 MΩ , and experiments were discarded if higher than 30 MΩ . The input strength is proportional to the number of activated mPP axons , and it was assessed as percentage of the field excitatory postsynaptic potential slope ( fEPSPslope , Figure 1—figure supplement 1 ) , which increases linearly with the fiber volley ( Figure 1—figure supplement 1B , inset ) . Unlike the fiber volley , the fEPSP can be well visualized in all recordings . For this purpose , a field-recording microelectrode was placed on the granule cell layer ( GCL ) to record the fEPSP and the pop spike in response to the mPP stimulation . To compare input strengths across experiments , the fEPSPslope elicited at any given stimulus intensity was normalized to the fEPSPslope evoked at a stimulus intensity that evokes a maximal pop spike ( 100% , Figure 1—figure supplement 1A , B ) . Input strength was kept at 50% for all experiments and frequencies ( Figure 1—figure supplement 1C ) . A similar approach for input strength calibration has been previously used ( Pouille et al . , 2009; Marín-Burgin et al . , 2012 ) . The input strength was always assessed and calibrated in the absence of PTX . Thus , for a given input strength within a slice , the number of mPP axons stimulated in control conditions or in the presence of PTX ( 100 µM , Sigma-Aldrich ) were the same . Evoked monosynaptic EPSCs and disynaptic IPSCs were recorded after mPP stimulation at 50% input strength . To minimize the contribution of IPSCs mediated by direct stimulation of inhibitory axons , we only considered experiments in which kynurenic acid ( KYN ) ( 6 mM , bath applied at the end of the experiment ) blocked >70% of IPSCs . EPSCs were isolated by voltage clamping GC at the reversal potential of the IPSC measured for each individual neuron ( ∼60 mV ) . In turn , disynaptic IPSCs were recorded at the reversal potential of the EPSC ( ∼0 mV ) . When present , direct monosynaptic IPSC recorded in KYN ( always <30% of the peak IPSC amplitude ) was subtracted from the IPSC . Synaptic excitatory and inhibitory conductances were computed as the EPSC or IPSC divided by the driving force at which the synaptic currents were recorded . 2-amino-5-phosphonovaleric acid ( AP5 , 50 mM , Tocris , United Kingdom ) was perfused during recordings to block NMDA currents . We ruled out a possible influence of NMDA receptor-mediated currents on the activation of GC by recording spiking in the presence of AP5 . No differences were found between spiking in ACSF and after application of AP5 in the bath ( data not shown ) . Efficacy was computed as the average of the probability of occurrence of action potentials in the same frequency range as that given by the stimulus . Thus , a granule cell that responds with five consecutive action potentials in response to the train of 10 pulses , has an efficacy = 4/9 ( 4 intervals were represented out of the 9 that were in the train ) , while a GC with five discontinuous action potential firing , ( without presenting two continuous ) has an efficacy of 0/9 . Efficacy was calculated for matGC and 4wpiGC ( Figure 1—figure supplement 2 ) . To dissect feedforward and feedback inhibitory components , after registering the total inhibition evoked by the stimulation of the mPP , mossy fiber synapses were blocked with DCG4 ( 1 μM , Tocris ) ( Kamiya et al . , 1996 ) . In this condition , the remaining feedforward current was measured , and the feedback component was calculated by subtracting the feedforward current to the total current offline . In parallel experiments , we controlled for the effect of DCG4 on the activation of inhibitory interneurons , by measuring the change in their activation in response to mPP stimulation after the application of DCG4 , using a stimulation intensity of 50% fEPSP as in the rest of the work ( Figure 4—figure supplement 2 ) . To record from identified interneurons , we used transgenic mice with fluorescently labeled PV interneurons , PV-Tom mice . Significant differences were assessed by two-tailed t-test , Wilcoxon signed-rank test , two-sample F-test for equal variances , one-way or two-way ANOVA , and Bonferroni's post-hoc test , as indicated in the figure legends .	A number of cell types in the body are capable of dividing to produce two new cells . These are used either to replace damaged or worn out tissue , or to satisfy a need for additional cells . By contrast , the vast majority of the billions of neurons in the brain are produced before birth , and only a handful of brain regions retain the ability to generate new neurons throughout life . One of these brain regions is the hippocampus , which has roles in memory and navigation . The zone of the hippocampus that receives signals from other parts of the brain—known as the dentate gyrus—contains cells called granule cells that are still able to divide in adult animals . Newly formed cells mature over the course of a few weeks , but whether immature and mature cells make different contributions to processing the information that enters the hippocampus is unclear . By stimulating slices of mouse hippocampus using sequences of electrical pulses similar to those that occur naturally in the brain , Pardi et al . have shown that the granule cells' responses vary with age . Both mature and 4-week-old granule cells responded more strongly when the electrical pulses were applied at a slower rate . However , the immature cells could also respond to pulses applied at a faster rate more reliably than their mature counterparts . In contrast , mature cells signaled the arrival of the first pulse in a sequence more precisely than immature cells , with the majority of mature granule cells firing within the first three milliseconds of receiving a pulse . Additional experiments revealed that these differences arise because mature cells are more easily prevented from firing by inhibitory neurons that contact them , particularly in the presence of rapid sequences of pulses . By using immature and mature granule cells to encode different aspects of an incoming stimulus , the hippocampus may thus be able to maximize its ability to process information . These results raise questions regarding how parallel processing by immature and mature granule cells could affect the processing in CA3 , the area that receives the information sent by these cells .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "neuroscience" ]	2015	Differential inhibition onto developing and mature granule cells generates high-frequency filters with variable gain
Conventional approaches for antiparasitic drug discovery center upon discovering selective agents that adversely impact parasites with minimal host side effects . Here , we show that agents with a broad polypharmacology , often considered ‘dirtier’ drugs , can have unique efficacy if they combine deleterious effects on the parasite with beneficial actions in the host . This principle is evidenced through a screen for drugs to treat schistosomiasis , a parasitic flatworm disease that impacts over 230 million people . A target-based screen of a Schistosoma serotoninergic G protein coupled receptor yielded the potent agonist , ergotamine , which disrupted worm movement . In vivo , ergotamine decreased mortality , parasite load and intestinal egg counts but also uniquely reduced organ pathology through engagement of host GPCRs that repressed hepatic stellate cell activation , inflammatory damage and fibrosis . The unique ability of ergotamine to engage both host and parasite GPCRs evidences a future strategy for anthelmintic drug design that coalesces deleterious antiparasitic activity with beneficial host effects . Schistosomiasis ( infection with parasitic blood flukes ) affects over 230 million people , with severe infections causing ~200 , 000 deaths per year . Chronic infections are associated with morbidities such as anemia , impaired cognitive development and stunted growth ( Colley et al . , 2014; King and Dangerfield-Cha , 2008 ) . Infection is acquired upon exposure to freshwater , larval cercariae which penetrate the host skin . Inside host tissue , cercariae transform into schistosomula that access and traverse the circulatory system , pass through the lungs and liver before finally residing as mature adults within either the mesenteric vessels ( Schistosoma mansoni and Schistosoma japonicum ) or vasculature surrounding the urinary bladder ( Schistosoma haematobium ) ( Wilson , 2009; Georgi et al . , 1986 ) . The sexually mature male and female worms pair up and generate a prodigious number of eggs , depositing hundreds to thousands of eggs per female worm per day ( Cheever , 1969; Cheever et al . , 1994; Barlow and Meleney , 1949 ) . Eggs lodge within the peritoneal organs , evoking a massive Th2 response which cues granuloma formation and consequent pathology in affected organs ( reviewed in [Cheever , 1969] ) . In the case of urogenital schistosomiasis ( S . haematobium infection ) eggs are deposited in the bladder mucosa and urinary tract , resulting in hematuria and increased rates of squamous-cell carcinoma . In the case of intestinal schistosomiasis ( S . mansoni and S . japonicum infections ) , eggs are deposited in the liver and hepatic portal system leading to periportal fibrosis , pulmonary hypertension and ascites . Broad spectrum chemotherapy of schistosomiasis relies upon praziquantel ( PZQ ) , which has been in clinical use for nearly 40 years ( Colley et al . , 2014; King and Dangerfield-Cha , 2008; Andrews et al . , 1983 ) . Mass drug administration ( MDA ) programs will require ~250 million tablets of PZQ per year for at risk populations ( Osakunor et al . , 2018 ) . The widespread distribution of schistosomiasis , coupled with observations that PZQ can be subcurative in areas of high intensity of infection and transmission ( Danso-Appiah and De Vlas , 2002; King et al . , 2011 ) raise concerns that treatment resistant parasites may emerge ( Fallon and Doenhoff , 1994; Fallon et al . , 1995; Ismail et al . , 1996; Wang et al . , 2012 ) . Therefore , it is important to identify new flatworm drug targets and lead compounds to expand the arsenal of drugs available to treat schistosomiasis and other parasitic flatworm infections currently resolved through PZQ administration . In this regard , targeting parasite neuromuscular physiology has proven to be a highly successful anthelmintic approach ( Geary et al . , 1992 ) . In schistosomes , serotonin ( 5-HT ) controls motor function and a specific serotonergic G-protein coupled receptor ( GPCR ) mediating this effect has been implicated by both RNAi ( Patocka et al . , 2014 ) and drug screening ( Chan et al . , 2016a , 2016b ) . Here , we cloned full-length 5-HT receptor sequences from each of three major species causing human infection worldwide and expressed these targets in a high-throughput capable assay enabling screening of thousands of compounds from natural product libraries . The dataset identified anti-schistosomal chemotypes that conveyed anti-parasitic efficacy . Most importantly these activities led to the discovery that the ergot alkaloid ergotamine ameliorated both infection and the pathological sequelae of infection . These properties highlight an opportunity to combat schistosome infections and their pathological impact on the host using single agents that coalesce deleterious actions on parasites with beneficial activities on host responses . Full-length sequences for a serotoninergic GPCR ( 5-HTR ) were cloned from the three major schistosome species causing infections worldwide; Schistosoma mansoni ( Sm . 5HTRL[Chan et al . , 2016b] ) , S . haematobium ( Sh . 5HTR ) and S . japonicum ( Sj . 5HTR ) . The three 5HTRs shared high amino acid similarity ( 84–94% ) , and clustered with 5-HT7 receptors from other organisms ( Figure 1—figure supplements 1 and 2 ) . In mammalian HEK293 cells , GFP-tagged variants of each 5-HTR localized at the cell surface ( Figure 1A ) , enabling functional profiling . To measure signaling output , GPCR activity was assessed using a luciferase based cAMP reporter ( GloSensor 22F , Promega ) engineered to undergo a conformational change on cAMP binding that increases luciferase activity ( Figure 1B ) . The choice of this reporter was guided by prior evidence demonstrating coupling of the flatworm SER-7 GPCRs to Gs ( Patocka et al . , 2014; Chan et al . , 2016b , 2016c ) and the ability of this approach to resolve real time kinetics of cAMP generation with high sensitivity in intact cells . The 5-HTRs from each individual schistosome species were transiently transfected into a stable HEK293 cell line expressing the cAMP reporter and luminescence values monitored during addition of 5-HT . Each of the GPCRs responded to the addition of 5-HT with a rapid elevation in luminescence values ( ~70–110 fold increase , Figure 1C ) with sensitivity in the low nanomolar range for Sj . 5HTR and Sh . 5HTR ( EC50 = 3 . 0 ± 0 . 3 , 3 . 6 ± 0 . 3 nM ) and 76 ± 1 nM for Sm . 5HTRL ( Figure 1D ) . To facilitate high-throughput drug screening of these targets , dual stable cell lines expressing both the cAMP reporter and individual 5-HTR clones were derived and individual lines evaluated based on performance in a 384-well format . Conditions were iterated to deliver robust resolution of the signals elicited by 5-HT ( ‘agonist screen’ , Z’ score >0 . 7 , signal window >10 ) and the known inhibitory action of bromocriptine ( ‘antagonist screen’ , Z’ score >0 . 6 , Figure 1—figure supplement 3 ) , a previously identified antagonist ( Chan et al . , 2016b ) . Having optimized these methods , we proceeded to screen the S . mansoni 5HTRL against a compilation of natural product libraries ( 4288 compounds , Supplementary file 1 ) . The screening protocol consisted of an ‘agonist screen’ ( addition of compound alone , 10 µM ) followed by an ‘antagonist screen’ ( inhibition of response to 5-HT , 1 µM ) in the same cells for every compound . The workflow , and criteria for putative ‘hit’ designation , from the primary screen are shown in Figure 2A . An example of the data profile generated for a putative agonist ‘hit’ and an antagonist ‘hit’ is shown in Figure 2B . Compilation of the screening dataset is shown in Figure 2C , where the fold-increase ( ‘agonist screen’ ) or decrease ( relative to 5-HT , ‘antagonist screen’ ) in luminescence values are plotted for individual compounds . From the whole dataset ( 4288 ligands ) , 92 putative hits were identified that exceeded threshold criteria ( Figure 2C ) . These compounds were then counter-screened against the parental HEK293 reporter cell line lacking any schistosome 5-HTR ( Figure 2D ) . This was necessary to exclude false positive hits , for example compounds that stimulated cAMP production through endogenous HEK293 targets ( GPCRs , direct adenylate cyclase activators ) , or compounds that non-specifically inhibited cAMP production . Counter-screening triaged 40 of the original 92 hits , invalidating compounds that either stimulated cAMP production or inhibited forskolin-evoked cAMP generation in cells lacking Sm . 5HTRL . Overall , the screening dataset yielded 52 compounds ( 12 agonists and 40 antagonists at Sm . 5HTRL , ~1 . 2% hit rate ) for subsequent investigation . Chemotype analysis of the ‘hit’ compounds ( Figure 2E ) revealed two main groups of favored structures . The first core group of ligands were benzylisoquinolines with aporphine or protoberberine cores . These compounds were exclusively antagonists . The second group of compounds contained an indole ring system , either bicyclic tryptamines , or ergot alkaloids with a four ring ergoline core . This second group of compounds comprised both agonist and antagonist ligands . This range of efficacies underpinned prioritization of these ligands , and ergot alkaloids as the predominant chemotype , for further analysis of the structure-activity relationship at the parasitic 5-HTRs . Ergot alkaloids are a diverse group of compounds with a long history of therapeutic use . Their interaction with mammalian bioaminergic GPCRs encompasses a range of activities and kinetic profiles ( Knight et al . , 2009; Wacker et al . , 2017 ) . Based on the results from the natural product screen , an expanded series of ergot alkaloid ligands was sourced for structure-activity screening ( ‘SAR by commerce’ ) . Complete dose-response relationships were performed for twenty compounds containing the tetracyclic ergoline skeleton ( 5 compounds from the primary screen plus 15 commercially sourced ligands ) against all three schistosome 5 . HTRs ( Sm . 5HTRL , Sh . 5HTR and Sj . 5HTR ) . Ergot alkaloid interactions with the schistosome 5-HTRs spanned a range of effects . To summarize these actions , agonist efficacy ( relative to peak levels of 5-HT evoked cAMP generation ) and antagonist potency ( pKi ) were represented as heat-maps where increasingly intense coloration represents higher efficacy ( Figure 3A ) and potency ( Figure 3B ) . Several features of this dataset merit comment . First , these representations convey a shared responsiveness across the three 5-HTRs to tested ligands , suggesting a conservation of pharmacological profile between the 5-HTR orthologs from the different schistosome species . This is important if any putative lead compounds are to be broad spectrum , that is effective against each of the major infective strains . Second , the data evidence a broad spectrum of potencies and efficacies within the ergot alkaloid series evidencing full agonists ( for example , compound ‘1’ ergotamine , ‘2’ dihydroergotamine ) , partial agonists ( compound ‘3’ , lisuride ) , antagonists ( compounds ‘16’ terguride , ‘19’ , bromocriptine , ‘20’ 2-bromo-LSD ) as well as closely related ‘inactive’ compounds ( compound ‘14’ , α-ergocryptine ) . Third , marked changes in efficacy and inhibitory potency resulted from minimal structural modifications to the core scaffold . For example , hydrogenation of the D9-10 double bond of lisuride ( compound ‘3’ ) eliminated efficacy to yield the antagonist terguride ( compound ‘16’ , IC50 = 400 ± 50 nM , Figure 3C ) . The partial agonist LSD ( compound ‘6’ ) is converted into a potent antagonist ( IC50 = 100 ± 20 nM ) on bromination at the B2 indol ring position ( 2-Bromo-LSD , compound ‘20’ ) . Distinct structural modifications impacting efficacy could be seen in an ergopeptine series ( Figure 3D ) , through comparison of the chemical structures of ergotamine ( compound ‘1’ ) , a potent full agonist at each schistosome receptor ( Emax95 ± 4 , 96 ± 5 and 114 ± 3% of 5-HT for Sm . 5HTRL , Sh . 5HTR and Sj . 5HTR , respectively ) , the inactive α-ergocryptine ( compound ‘14’ ) and bromocriptine ( compound’ 19 ) which acted as a potent blocker ( IC50 = 1 . 6 ± 0 . 4 µM ) of Sm . 5HTRL ( Figure 3D ) . Such SAR data underscore the potential for modifying the ergoline scaffold to maximize , or engineer away , efficacy at schistosome 5HTRs ( Figure 3C and D ) . Overall , structure-activity analysis across the three schistosome 5-HTRs identified ergotamine ( compound ‘1 ) as the most potent , full agonist and 2-Bromo-LSD ( compound ‘20’ ) as the most potent antagonist for further evaluation . The effects of prioritized compounds from the screen were examined on the movement of adult parasites cultured ex vivo . Male and female worms respond to 5-HT with increased movement ( Figure 4A ) . This responsiveness permits the basic screening assay shown in Figure 4B , where changes in worm motility are resolved in response to compound addition , and in the presence of 5-HT . For example , 5-HT ( 200 µM ) increased worm movement , and 5-HT stimulation was not observed following preincubation with bromocriptine ( 1 µM , Figure 4B ) . Compounds that exhibited agonist activity at each of the three 5-HTRs in vitro also stimulated worm movement ex vivo ( Figure 4C ) . While the rank order of potency varied slightly from the in vitro screen , an increase in potency of ergot alkaloids relative to 5-HT was evident ( EC50 for 5-HT = 51 . 6 ± 3 . 0 µM , EC50 = 97 . 8 ± 31 . 6 nM for ergotamine , EC50 = 1 . 4 ± 0 . 5 µM for LSD ) . As expected , the potent antagonist BOL-148 ( Figure 3C ) blocked 5-HT evoked hyperactivity ( Figure 4D ) . BOL-148 displayed ~10 fold higher potency than bromocriptine consistent with the results from the in vitro screen ( IC50 = 28 . 8 ± 23 . 1 nM for BOL-148 versus IC50 = 331 ± 271 nM for bromocriptine , Figure 4D ) . Do these schistosome 5-HTR ligands display anthelmintic activity in vivo ? As processes such as parasite feeding , mate pairing and movement within the host vasculature depend upon coordinated neuromuscular function , exposure to ligands that stimulate hyperactivity ( 5-HTR agonists ) or cause paralysis ( 5-HTR antagonists ) may promote dysfunction and parasite clearance . One indicator of acute drug efficacy that may prove indicative of anthelmintic activity is a shift in location of the parasites from the mesenteric vessels to the liver ( the ‘hepatic shift’ ) , where worms are subsequently targeted for elimination by the host immune system ( Buttle and Khayyal , 1962; Pellegrino et al . , 1977; Mehlhorn et al . , 1981 ) . To examine the ability of the prioritized 5-HT ligands to cause this hepatic shift , mice harboring a mature schistosome infection ( 42 days p . i . ) were given 5-HTR ligands by intraperitoneal injection and sacrificed three hours later to assess parasite distribution ( Figure 5A ) . Numbers of worms residing within the mesenteries , portal vein or liver were assessed . Following treatment with PZQ ( 100 mg/kg ) , most of the worms were found within the liver ( 68 . 8 ± 14 . 1% of recovered parasites , Figure 5B ) , whereas worms were infrequently found within the liver in mice injected with vehicle control ( 9 . 7 ± 8 . 5% of recovered parasites ) . The performance of a selection of 5-HTR agonists or antagonists was then benchmarked relative to these standards . Treatment with 5-HTR agonists elicited a hepatic shift ( Figure 5B ) . However , treatment with 5-HTR antagonists did not alter worm distribution ( Figure 5B ) . Furthermore , the extent of the hepatic shift observed in mice treated with 5-HTR agonists correlated with the measured efficacy of these agonists at the parasite 5 . HTR ( Figure 5B ) . Full agonists promoted the greatest shift: for example , ergotamine treatment of mice yielded a similar percentage of worms in the liver ( 75 . 1 ± 16 . 9% ) as seen in mice treated with PZQ . Partial agonists promoted a lesser response , and 5-HTR antagonists were ineffective . For example , neither bromocriptine ( 2 . 0 ± 2 . 8% of worms in liver ) or BOL-148 ( 5 . 2 ± 2 . 3% of worms in liver ) caused a hepatic shift of worms ( Figure 5B ) despite the effectiveness of these compounds at arresting worm movement ex vivo ( Figure 4D ) . Based on these data , we prioritized treatment of infected mice with ergotamine to assess amelioration of chronic schistosomiasis . Mice were injected with vehicle control ( DMSO ) , PZQ ( 50 mg/kg ) or ergotamine ( 60 mg/kg ) twice daily for one week starting 42 days post infection ( p . i . ) . Control animals succumbed to infection ( ~8–10 weeks p . i . ) over a timeframe consistent with the onset of egg laying by sexually mature parasites as expected ( Figure 6A , [Cheever , 1969] ) . By 15 weeks p . i . , the majority of control mice were dead . Death rates were significantly reduced in ergotamine-treated mice ( 16 . 5% versus 68 . 9% mortality in control cohort , log-rank test p<0 . 001 , hazard ratio 5 . 02 ) . PZQ-treated mice survived through 15 weeks with no observed lethality ( Figure 6A ) . As ergotamine increased survival of infected mice , we assessed various metrics of infection ( worm burden , egg number , hepatosplenomegaly ) immediately following ergotamine treatment ( 42–49 days p . i ) . First , ergotamine treatment significantly reduced worm burden from 28 ± 8 worm pairs per mouse to 9 ± 6 worm pairs per mouse ( ~65% reduction , p<0 . 002; Figure 6B ) . Next , egg counts were performed . Eggs deposited within the mesenteric system progress through the intestinal mucosa for excretion from the host , thereby propagating the parasite life cycle ( Pellegrino and Faria , 1965 ) . Therefore , it is possible to determine whether drug treatment impacted parasite egg laying by counting eggs in sections of the large and small intestines . Control mice contained large numbers of viable eggs in both the small and large intestine . However , treatment with either ergotamine or PZQ resulted in a considerable reduction ( 96 . 3 ± 1 . 2 and 99 . 9 ± 0 . 1% , respectively ) in intestinal egg burden ( Figure 6C ) . We also assessed whether serotonergic manipulation influenced egg laying in worms cultured ex vivo ( Figure 6—figure supplement 1A ) . Egg laying was decreased following treatment with ergotamine ( 1 µM , 54 . 2 ± 12 . 6% reduction ) , serotonin ( 100 µM , 82 . 4 ± 2 . 2% reduction ) or forskolin ( 100 µM , 61 . 6 ± 12 . 6% reduction ) ( Figure 6—figure supplement 1B&C ) . Eggs laid by ergotamine treated worms were often small and deformed , indicating abnormal development ( Figure 6—figure supplement 1D ) . We conclude ergotamine was highly effective at blocking parasite egg production and development . Next , liver and spleen sizes were measured . A consequence of the enormous production of schistosome eggs is a host Th2 wave ( Grzych et al . , 1991 ) that drives granuloma formation and fibrosis in affected peritoneal organs ( Pearce and MacDonald , 2002 ) . Hepatosplenomegaly is a hallmark of chronic schistosomiasis and splenomegaly and liver fibrosis are observed in the murine model ( Fallon , 2000 ) . This is reflected by comparison of the weights of uninfected mouse spleens ( 0 . 12 ± 0 . 02 g ) and the spleens of infected littermates ( ~6 fold enlargement , weighing 0 . 76 ± 0 . 20 g at 49 days p . i ) , or increases in liver mass ( ~2 fold enlargement from 2 . 0 ± 0 . 3 g in uninfected mice to 4 . 8 ± 0 . 8 g in infected mice , Figure 6D ) . PZQ treatment ( one week , starting at 42 days p . i ) did not prevent spleen and liver enlargement ( Figure 6D ) . There was no difference in the weights of the liver and spleen between PZQ-treated and vehicle treated infected mice ( Figure 6E and F ) . This is likely due to the fact that , while PZQ is highly effective at eliminating worms at six weeks infection and beyond ( Sabah et al . , 1986; Keiser et al . , 2009 ) , sexual maturation and egg laying commence as early as 5 weeks post infection ( Sabah et al . , 1986 ) , meaning that eggs deposited prior to treatment initiation still evoke an immune response which PZQ exposure does not ameliorate . In contrast to PZQ , ergotamine exposure prevented hepatosplenomegaly . Ergotamine reduced the infection-promoted increase in spleen mass by 55 . 8 ± 14 . 0% and liver mass by 52 . 0 ± 14 . 2% relative to vehicle treated controls ( Figure 6E and F ) . Similarly , liver sections from ergotamine-treated infections displayed decreased granuloma formation and fibrosis when investigated by various staining methods ( Figure 6G and Figure 6—figure supplement 2 ) . Movat's stain and Masson’s trichrome showed decreased levels of collagen deposition surrounding granulomas ( Figure 6G , II–III ) . Aldehyde fuchsin stain showed elastin surrounding the granulomas of DMSO and PZQ treated mice , while the livers of uninfected or ergotamine treated mice exhibited little staining ( Figure 6G , IV ) . Oil red O staining revealed lipid stores present in the livers of uninfected mice that were absent in infected control livers . Treatment with ergotamine , but not PZQ , preserved regions of Oil red O positive staining - albeit at lower levels than healthy livers ( Figure 6G , V ) . Pro-apoptotic cleaved caspase-3 was present in cells within granulomas of all three infected livers ( DMSO control , ergotamine and PZQ treated – Figure 6G , VI ) . Finally , we tested whether ergotamine displayed efficacy as a treatment against immature infections , where PZQ is ineffective ( Sabah et al . , 1986 ) . Both 5-HT and ergotamine stimulated movement of juvenile worms ex vivo ( Figure 6—figure supplement 3A and B ) . In vivo , however , early dosing with ergotamine prior to egg laying ( 3–4 weeks post infection ) did not resolve infections , mirroring the lack of effect of PZQ ( Figure 6—figure supplement 3C ) . In conclusion , these data show that ergotamine was highly effective at promoting survival in mice with mature infections ( Figure 6A ) through reductions in worm ( Figure 6B ) and egg number ( Figure 6C ) . However - unlike PZQ - ergotamine was effective at attenuating hepatosplenomegaly associated with the chronic pathology of schistosomiasis . How does ergotamine treatment protect against hepatosplenomegaly resulting from schistosome infection ? Both ergotamine and PZQ clear parasites , but unlike ergotamine , PZQ-treated mice did not show a noticeable difference in liver and spleen size relative to control infections ( Figure 6 ) . As ergotamine is a human therapeutic with affinity for various host GPCRs ( O'Connor and Roth , 2005 ) , in addition to high affinity for schistosome 5-HT receptors ( Figure 3D ) , engagement of host signaling pathways likely underpins these protective effects . To investigate mechanisms contributing to the differential in vivo effects of these drugs , we performed RNA-Seq analyses on the livers and spleens of schistosome-infected mice following drug treatment ( PZQ , ergotamine ) over a window 6 to 7 weeks post-infection ( Figure 7A ) . Livers and spleens of infected mice showed widespread changes in gene expression relative to uninfected littermates; of the 24 , 421 transcripts with mapped sequencing reads , 9684 transcripts were differentially expressed in infected livers and 7325 transcripts were differentially expressed in infected spleens . It is well established that the host pathology of schistosomiasis infection is driven by an initial Th1 response to the worms themselves followed by a sustained Th2 response that drives egg-induced fibrosis ( reviewed in [Pearce and MacDonald , 2002; Wynn et al . , 2004] ) . Such a gene expression signature has been observed in numerous studies on livers or spleens of infected mice ( Sandler et al . , 2003; Burke et al . , 2010; Gobert et al . , 2015; Perry et al . , 2011 ) and is borne out here through our data comparing infected and uninfected controls . Most components of Th1 and Th2 signaling pathways were significantly changed in both infected livers and spleens relative to uninfected littermates ( Figure 7B and C ) . Drug treatment with either PZQ or ergotamine attenuated transcriptional changes associated with infection . The identity of transcripts changing with drug treatment ( either with PZQ or with ergotamine treatment versus control infections ) largely overlapped with those changing with infection ( uninfected animals versus control infections ) but as expected drug-treatment was associated with a decrease of transcripts up-regulated in control infections ( relative to uninfected animals ) or an increase of transcripts down-regulated in control infections ( relative to uninfected animals , Figure 7D ) . Specifically , PZQ or ergotamine treatment decreased expression of pathways associated with host immune response to either parasites ( acute phase response signaling , Th1 and Th2 pathway ) or bacteria entering the circulation with the breakdown of the intestinal epithelium that occurs with heavy parasite egg production ultimately leading to host death ( LPS/IL-1 mediated inhibition of RXR function [Feingold et al . , 2004; Wang et al . , 2005] ) . Expression of pathways associated with drug metabolism ( xenobiotic metabolism signaling ) and liver function ( hepatic cholestasis ) increased with both treatments ( Figure 7D ) . These observations that the PZQ and ergotamine cohorts shared similar transcriptional patterns relative to control infection is consistent with both treatments resolving infection and parasite egg production ( Figure 6B and C ) . Next , we sought to filter the transcriptomic dataset to identify transcriptional changes unique to ergotamine treatment ( Figure 7E ) . This involved selecting transcripts that ( i ) displayed significant changes between uninfected and control infection cohorts , and ( ii ) were differentially expressed between control infection and ergotamine-treated cohorts , and ( iii ) showed little change between control infection and praziquantel treated cohorts ( Figure 7E ) . Ingenuity pathway analysis using this triple-filtered criteria identified a cohort of transcripts uniquely associated with ergotamine , but not PZQ , treatment . These transcripts were tagged for involvement in hepatic fibrosis and hepatic stellate cell ( HSC ) activation ( p-value 0 . 00015 , Fisher's exact test ) . Ergotamine treatment down-regulated gene products associated with HSC activation to control infections , while PZQ treatment had little to no effect on the same suite of transcripts . HSCs represent a quiescent , non-proliferative cell population in normal liver but once activated during injury or disease , drive fibrotic responses by transforming into proliferative , contractile myofibroblasts that deposit extracellular matrix components ( Tsuchida and Friedman , 2017 ) . Differentially expressed transcripts in ergotamine-treated samples ( Supplementary file 4 ) included known markers of HSC activation responsible for extracellular matrix production ( COL1A1 , COL1A2 , COL3A1 ) , transcripts associated with HSC contraction ( CACNA1E , CACNA1H ) , migration ( IGFBP3 , RANTES ) , survival ( IGFBP5 , GAS6 ) and transdifferentiation into myofibroblasts ( MMP13 , GLI1 , PAK1 ) ( Figure 7F and G ) . Ligands that control HSC activation and survival were also differentially expressed ( PDGFB , IL13RA2 , BMP4 , IL-4 , SAA ) . Therefore , ergotamine and PZQ were both effective at promoting survival ( Figure 6A ) through reductions in worm ( Figure 6B ) and egg number ( Figure 6C ) , as well as reversing many transcriptional signatures that drive host pathological outcomes in response to parasite infection ( Figure 7D ) . However , unlike PZQ , ergotamine was also effective at attenuating the hepatosplenomegaly associated with the chronic pathology of schistosomiasis ( Figure 6D–G ) , due to decreased activation of profibrotic HSCs ( Figure 7E–G ) . Ergotamine is a potent Sm . 5HTRL agonist ( Figure 3 ) . Sm . 5HTRL and homologs of this receptor in other flatworms , regulate movement ( Patocka et al . , 2014; Chan et al . , 2015 ) . Sm . 5HTRL is downregulated upon pairing of adult S . mansoni females and males , implicating an additional role in sexual maturation ( Lu et al . , 2016 ) . As a Gs coupled receptor , the role of 5-HT acting through cAMP to regulate flatworm metabolism is also relevant ( Mansour , 1984 ) . These broad physiological roles - motor function , female egg production , metabolism – underscore consideration of Sm . 5HTRL as an attractive antischistosomal target . Screening hits ( Figure 2 ) , and subsequent structure-activity studies ( Figure 3 ) confirmed the ergoline core as a customizable scaffold for ligand optimization at Sm . 5HTRL . These analyses provided a toolbox of compounds with varied potencies and efficacies against Sm . 5HTRLin vitro and Sm . 5HTRL agonists and antagonists stimulated and inhibited adult worm motility ex vivo as predicted ( Figure 4 ) . However in vivo , only agonists of Sm . 5HTRL signaling caused a hepatic shift of worms and mitigated worm burden , Sm . 5HTRL blockers were ineffective ( Figure 5 ) . This observation was surprising given the presumed beneficial effect of causing potent ( BOL-148 ) or long-lasting ( bromocriptine ) worm paralysis ( Chan et al . , 2016a , 2016c ) , but nevertheless informative for future phenotypic screens to prioritize ‘hyperactive’ over ‘hypoactive’ serotoninergic hits . From these analyses , ergotamine emerged as a potent full agonist ( Figure 3D ) that stimulated movement ex vivo ( Figure 4C ) and reduced worm and disease burden in vivo ( Figure 6 ) . While ergotamine stimulates schistosome movement through 5 . HTRs , the drug also acts on host GPCR targets . Ergotamine is known to display a broad polypharmacology , interacting with numerous bioaminergic GPCRs ( O'Connor and Roth , 2005 ) . These interactions likely engage physiological responses that mitigate the damage caused by egg laying and possibly contribute to worm elimination . Acutely , ergotamine acts as a vasoconstrictor , including within the mesenteric vasculature where the adult worms reside ( Mikkelsen et al . , 1981 ) . Disruption of the parasite environment consequent to ergotamine-evoked mechanical changes and alterations of blood flow within the hepatic portal system could drive parasites from the mesenteries to the liver for elimination . This action would mirror recent data demonstrating R-PZQ , which also causes a hepatic shift , acts a human 5-HT2B receptor partial agonist that contracts host mesenteric vessels ( Chan et al . , 2017 ) . Chronically , drug treatment with either PZQ or ergotamine improved various aspects of liver function ( hepatic cholestasis , retinol metabolism , bile acid synthesis , Figure 7D and Supplementary file 4 ) relative to untreated infections . Beneficial changes highlighted by RNA-Seq also encompassed protection against intestinal damage ( Cao et al . , 2010 ) . Various pathways involved in response to bacterial infection were down regulated with PZQ and ergotamine treatment relative to control infections ( Supplementary file 4 ) . This is likely because mice experience a relative infection burden orders of magnitude higher than typical for humans ( Cheever , 1969 ) resulting in compromised intestinal integrity and release of microbiota into the circulation as mature parasites commence egg production . These effects likely contribute to the observed lethality post-infection and are shown to be mitigated by both drug treatments in the RNA-Seq analysis . However , ergotamine additionally provided protection against the spleen and liver enlargement that persisted in PZQ-treated mice ( Figure 6D and E ) , suggesting a further action of ergotamine on the host immune response . Analysis of RNA-Seq datasets identified transcriptional changes unique to ergotamine and not PZQ treatment , most notably a decrease in gene products involved in hepatic fibrosis ( Figure 7E–G ) . Hepatic stellate cells ( HSCs ) are key drivers of fibrosis during liver disease . Upon their transformation into myofibroblasts , HSCs initiate active production of extracellular matrix components ( collagen , elastin – Figure 6G , II-IV ) and decrease lipid storage ( Figure 6G , V , [Tsuchida and Friedman , 2017] ) . HSCs are known to express a portfolio of GPCRs that regulate activation and transformation ( Tsuchida and Friedman , 2017 ) . These include several serotonergic GPCRs ( 5HT1B , 5HT1F , 5HT2A , 5HT2B , 5HT7 , [Ruddell et al . , 2006] ) . It is tantalizing that the known polypharmacological signature of ergotamine ( O'Connor and Roth , 2005 ) closely matches the serotoninergic GPCR expression profile in HSCs ( Ruddell et al . , 2006 ) . This may imply a cell-autonomous action of ergotamine on HSCs , although further insight would be needed to distinguish this possibility from non cell-autonomous mechanisms given that serotonin impacts immune responses and liver regeneration in a multiplicity of ways ( Herr et al . , 2017; Lesurtel et al . , 2006; Ebrahimkhani et al . , 2011 ) . For example , ergotamine could impair HSCs indirectly by inhibiting macrophage polarization ( de las Casas-Engel et al . , 2013 ) or immune mediators such as TNF-α and ICAM-1 ( 54 ) . ICAM-1 is found at elevated levels within egg-evoked granulomas ( Ritter and McKerrow , 1996 ) and TNF-α drives the host immune response to schistosome eggs ( Amiri et al . , 1992; Leptak and McKerrow , 1997 ) . Prior studies on schistosome infection of immune compromised mice have shown that T-cell deprived mice exhibit hepatic microvesicular steatosis , since granulomas also have a beneficial function of facilitating excretion of eggs from the mesenteries and sequestering toxic egg products ( reviewed in [Hams et al . , 2013; Doenhoff et al . , 1986] ) . Ergotamine treatment impaired the granulomatous response to eggs ( Figure 6G , Figure 6—figure supplement 2 ) , but increased microvascular damage was not observed ( Figure 6—figure supplement 4 ) . Additional experiments will be needed to investigate stellate cell responsiveness and whether ergotamine is effective against other parasitic diseases currently treated with PZQ . While the polypharmacology of ergotamine evidences the principle of combining host and parasite responses for therapeutic effect , it is not itself a likely candidate for repurposing . Repeated dosing of ergotamine as an antimigraine therapy is associated with known side effects: it is contraindicated in pregnancy and poorly suited for the large pediatric population susceptible to schistosomiasis . For future studies , analogs with improved pharmacokinetic profile and bioavailability will need to be evaluated for clinical potential as safe and effective agents . Indeed , for treatment of parasitic infections , a single bolus dose may provide more flexibility and avoid cumulative side effects associated with repeated dosing . Our limited structural-activity studies to date ( e . g . Figure 3 ) on schistosome GPCRs do suggest modifications of the core ergot alkaloid scaffold can generate analogs with a range of potencies and efficacy . In conclusion , ergotamine is fortuitously suited to engage pathways in both host and parasite that are simultaneously deleterious to the blood fluke but beneficial to the infected host in terms of worm clearance and mitigation of damage caused by eggs . Rather than prioritizing ‘selective’ antiparasitic therapies with no host effects , these data demonstrate merit in evaluating agents with broader polypharmacology across parasite and host as a route to novel discovering effective treatments for parasitic disease . The sequence for Sm . 5HTRL ( GenBank accession KX150867 [Chan et al . , 2016a] ) was used to BLAST the S . haematobium and S . japonicum genomes for putative homologs . The resulting S . haematobium ( NCBI Reference Sequence XM_012944163 and XM_012944164 ) and S . japonicum ( GenBank FN332592 . 1 ) hits were used as templates to clone out full-length mRNA sequences by 5’ and 3’ RACE ( Marathon cDNA Amplification Kit , Clontech ) using S . haematobium and S . japonicum total RNA provided by the Schistosomiasis Resource Center . Finalized sequences have been deposited with NCBI ( Sh . 5HTR = GenBank accession number MG813903 , Sj . 5HTR = GenBank accession number MG813904 ) . Plasmids for heterologous expression were generated by codon optimization of the coding sequence for mammalian expression and subcloning into either pcDNA3 . 1 ( - ) ( between NotI and EcoRI ) or pEGFP-N3 ( EcoRI ) . HEK293 cells ( ATCC CRL-1573 , an authenticated cell line validated by STR profiling ) were cultured in growth media consisting of DMEM supplemented with GlutaMAX ( Gibco cat # 10566016 ) + 10% heat inactivated fetal bovine serum and penicillin-streptomycin ( 100 units/mL , ThermoFisher ) . Stable cell lines for the pGloSensor-22F cAMP Plasmid ( Promega ) and schistosome 5HTRs in pcDNA3 . 1 ( - ) were selected with hygromycin ( 200 µg/mL ) and G418 ( 400 µg/mL ) . The correct identity of each stably expressed sequence was verified by isolating total RNA from each stable line ( TRIzol Reagent , Ambion ) and amplifying the sequence of interest with gene specific primers flanking the 5’ and 3’ regions of the coding sequence ( SuperScript III One-Step RT-PCR System , Invitrogen ) . PCR products were ligated into a TA cloning vector ( pGEM-T Easy , Promega ) and verified by Sanger sequencing . HEK293 lines stably expressing both the GloSensor 22F cAMP sensor and Sm . 5HTRL were cultured in growth media supplemented with 10% dialyzed FBS . For cAMP assays in 96 well format , cells cultured in T-75 flasks were trypsinized ( TrypLE Express , Gibco ) and plated in solid white 96 well plates ( Costar cat # 3917 ) the day prior to assay at a density of 2 . 5 × 104 cells/well . The following day media was removed and replaced with assay buffer consisting of HBSS buffered with HEPES ( 20 mM , pH 7 . 4 ) + BSA ( 0 . 1% w/v ) and D-luciferin sodium salt ( 1 mg/mL , Gold Biotechnology ) . Plates were equilibrated at room temperature for 1 hr , at which point 3-Isobutyl-1-methylxanthine ( IBMX 200 µM , Sigma Aldrich ) was added . Plates were equilibrated a further 30 min prior to test compound addition . Test compounds were added and luminescence was read for 45 min on a GloMax-Multi Detection System plate reader ( Promega ) to screen for agonists , after which plates were removed , 5-HT was added to each well at a concentration determined to achieve a maximal response for each stable cell line ( Sm . 5HTRL = 500 nM , Sj . 5HTR = 500 nM , Sh . 5HTR = 60 nM ) , and plates were read a second time to screen for antagonist activity . Readings performed in 384 well format were modified so that cells were assayed in suspension . Cells were grown to 70% confluence in T-75 flasks , trypsinized and pelleted ( 300 RCF x 5 min ) . Media was removed and cells were resuspended in 25 mL assay buffer supplemented with IBMX ( 200 µM ) . A 96 channel semi-automated pipet ( Eppendorf epMotion 96 ) delivered 55 µL of cell suspension per well ( 5000 cells/well ) into solid white 384 plates ( Corning 3574 ) . After one hour of equilibration , test compounds were screened at a fixed concentration of 10 µM , and control wells ( 64 wells per plate ) were treated with vehicle alone ( 1% DMSO in HBSS ) , providing an internal negative control reference . Following 60 min incubation at ambient temperature , plates were read for agonist activity , defined as a >10 fold increase in luminescence relative to control , DMSO treated wells on the same plate . Serotonin ( 1 µM ) was then added to each well , and after 60 min the plate was read again to identify compounds with antagonist activity , defined as compounds that inhibiting 5-HT evoked signal by >90% relative to control wells . The initial high-throughput screen in 384 well format was performed as a single replicate , with all ‘hits’ meeting agonist or antagonist thresholds re-screened in 96 well format ( technical quadruplicates performed for three biological replicates ) . Counter screening of primary hits against the parental HEK293 cell line expressing the GloSensor 22F cAMP sensor and lacking Sm . 5HTRL was performed to eliminate compounds stimulating cAMP ( >3 fold change in basal luminescence ) or inhibiting forskolin evoked cAMP production ( >25% inhibition relative to DMSO controls ) . A complete list of chemical vendors , catalog numbers and SMILES is contained in Supplementary file 1 . Natural product libraries were sourced from the National Institutes of Health ( NCI Natural Products Set IV ) and TimTec ( Natural Product Library [NPL-800] and Natural Derivatives Library [NDL-3000] ) . LSD and BOL-148 were sourced from the National Institute on Drug Abuse ( NIDA ) . The following compounds were sourced from Sigma Aldrich: praziquantel ( PZQ , P4668-1G ) , ergotamine tartrate ( 1241506–150 MG ) , dihydroergotamine tartrate ( D1952000 ) , bromocriptine ( B2134-100MG ) . LAMPA was sourced from Cerilliant ( L-004–1 ML ) . Lisuride maleate ( 4052 ) , dihydroergotoxine mesylate ( 0474 ) and methylergometrine maleate ( 0549 ) were sourced from Tocris . All animal experiments followed ethical regulations approved by the Medical College of Wisconsin and Iowa State IACUC committees and additionally reviewed in the context of funding by the National Institutes of Health ( NIAID ) . Female Swiss Webster mice infected with S . mansoni cercariae ( NMRI strain ) were sacrificed 49 days post infection by CO2 euthanasia . Adult schistosomes were recovered by dissection of the mesenteric vasculature . Harvested schistosomes were washed in RPMI 1640 Medium with GlutaMAX + 5% heat inactivated FBS ( Gibco ) and Penicillin-Streptomycin ( 100 units/mL ) . For movement assays , worms were cultured 37°C and 5% CO2 in vented 100 × 25 mm petri dishes ( ThermoFisher cat # 4031 ) containing 50 mL of media and used 1–3 days after harvesting . Prior to assessing worm movement , male worms were transferred to a six well dish ( 4–5 individual worms per well ) containing 3 mL drug solution in RPMI 1640 supplemented with HEPES ( 25 mM ) and FBS ( 5% ) . Videos were recorded using a Zeiss Discovery v20 stereomicroscope and a QiCAM 12-bit cooled color CCD camera controlled by Metamorph imaging software ( version 7 . 8 . 1 . 0 ) . 1 min recordings were acquired at 4 frames per second , saved as a . TIFF stack , and movement was analyzed using ImageJ software as described in ( Chan et al . , 2016a , 2016b ) . Data represents mean ± standard error for ≥3 independent experiments . EC50 and IC50 values are reported ± 95% confidence interval of fitted curves . For egg laying assays , adult schistosomes were transferred to 24 well plates the day after harvesting from mice ( 5 pairs of male and female worms in 2 mL media/well ) . Eggs were counted daily using a stereomicroscope , after which worms were transferred to a new well with fresh drug-containing media . Egg counts were recorded for 5 days , and data processed as the mean number of eggs laid per worm pair per day . Measurements of egg dimensions were quantified using the ‘analyze>measure’ function in ImageJ to record the length and width of individual eggs . Female Swiss Webster mice were exposed to 200 s . mansoni cercariae ( NMRI strain ) at between 4–6 weeks old . For hepatic shift assays , infections matured to 7 weeks , at which point mice were given test compounds by intraperitoneal injection and euthanized 3 hr later . Compounds were solubilized in 50 µL DMSO , and diluted in 200 µL 5% w/v Trappsol ( Cyclodextrin Technologies Development , THPB-p-31g ) in saline ( NaCl 0 . 9% ) solution . Immediately after being euthanized , mice were dissected to separate the liver from the portal vein , so that the number of worms recovered from the portal vein , mesenteric vasculature and liver could be recorded for each mouse . Hepatic shift assays were performed on at least 3 mice per drug treatment . Assays testing the curative efficacy of compounds against schistosome infections were performed on mice at 3 weeks post infection ( immature parasites ) and 6 weeks post infection ( mature parasites ) . Drugs were dosed as follows . Ergotamine ( 60 mg/kg ) was solubilized in 50 µL DMSO and diluted in 200 µL Trappsol-saline and delivered by intraperitoneal injection twice a day for one week . Praziquantel ( 50 mg/kg ) was similarly solubilized and injected intraperitoneally once a day for one week . The negative control cohort was given twice daily injections of DMSO ( 50 µL ) added to 200 µL Trappsol-saline solution . Mice were weighed and euthanized at 49 days post infection . Worms were harvested and counted as described for the hepatic shift assay . Spleen and livers were weighed , and a segment of intestine was excised from the most distal region of the rectum to 10 cm above the cecum . The small and large intestines were separated by cutting immediately above and below the cecum . Each was cut lengthwise to expose the lumen and thoroughly washed in NaCl ( 1 . 2% ) to remove excrement . Intestines were laid flat with the intestinal mucosa facing upwards and clamped between two glass plates to allow visual inspection of egg morphology and number using a stereo microscope . Scored egg counts reflect ‘viable’ eggs containing a developing embryo or a mature miracidium , while empty egg shells and granulomas were not included ( Pellegrino and Faria , 1965; Mati and Melo , 2013 ) . Mouse livers were removed immediately after animals were euthanized , rinsed in ice cold PBS , and fixed in 10% neutral buffered formalin . For H and E stain , Movat’s stain , Masson’s trichrome , and aldehyde fuchsin staining , samples were embedded in paraffin and sectioned at 4 microns ( Microm HM355S ) . For Oil red staining , livers were equilibrated in freezing medium ( 10% w/v and then 30% w/v sucrose solution in PBS ) prior to cryosectioning ( Leica CM1850 UV Cryostat ) . If paraffin embedded , slides were deparaffinized , hydrated , and the following staining procedures were then performed: H and E stain , slides were stained with Harris hematoxylin and eosin working solution ( Poly Scientific ) . Movat’s stain , Alcian blue ( 1% w/v ) , hematoxylin ( 10% ) , crocein scarlet –acid fuchsin solution , and alcoholic saffron ( 6 . 5% w/v ) . Masson’s trichrome , slides were pre-treated with Bouin’s solution and successively stained with hematoxylin , Biebrich-Scarlet-Acid Fuchsin and Aniline Blue solution . Aldehyde Fuchsin . slides were stained with aldehyde fuchsin and counterstained with fast green . Oil red O staining , after cryosectioning onto Superfrost slides ( Fisherbrand ) , samples were stained with Oil Red ( 0 . 3% w/v in 60% isopropanol ) followed by hematoxylin counterstain . Cleaved Caspase-3 staining , slides were treated with Leica BOND Epitope Retrieval Solution 2 ( 20 min ) , peroxidase-blocking solution ( Dako ) , avidin/biotin blocking kit ( Vecgtor ) and background sniper blocking reagent ( Biocare ) per manufacturer’s instructions . This was followed by Cleaved Caspase-3 primary antibody incubation ( Biocare , CP-229C ) at 1:100 dilution for 90 min at room temperature , 3 × 1 min washes ( Lieca BondTM Wash Solution ) , and incubation in biotinylated secondary antibody ( Jackson Immuno , 771-066-152 ) at 1:500 dilution for 30 min , 3 × 1 min washes , and finally incubation in Streptavidin HRP ( Dako , P0397 ) at 1:300 dilution . After washing ( 3 × 1 min ) , staining was visualized using DAB + Substrate Chromogen System ( Dako , K3468 ) . As a negative control , samples were treated as stated above but without primary antibody . Mice were infected and treated with drugs as described above . After drug treatment from weeks 6–7 post infection , mice were euthanized and livers and spleens were immediately removed and placed on ice . For liver samples , a section of the larger , right lobe was excised and homogenized in TRIzol Reagent ( Invitrogen ) . For spleen samples , the entire organ was homogenized in TRIzol . Samples were stored in −80°C until processed . Total RNA was extracted from TRIzol according to manufacturer’s protocols and quantified with RiboGreen RNA Assay Kit . Libraries were generated using the Clontech StrandedRNA Pico Mammalian kit and sequenced using the Illumina HiSeq 2500 system ( high-output mode , 50 bp paired-end reads ) at a depth of approximately 20 million reads/sample . Trimmed reads were mapped to the mouse genome ( mm10/GRCm38 ) using HISAT2 . Expression was quantified using featureCounts ( read counts ) and cuffquant ( FPKM ) . EdgeR was used to identify differentially expressed genes ( tagwise dispersion model , adjusted p-value<0 . 05 ) which were processed using Ingenuity Pathway Analysis ( Qiagen ) and DAVID functional annotation tool ( version 6 . 8 ) . RNA-Seq data has been deposited in the NCBI SRA database under accession number SRP131511 . Metrics for evaluating cell based screening assays include coefficient of variation ( CV% ) , signal window ( SW ) , Z factor and Z’ factor . Equations for each metric are included in the legend of Figure 1—figure supplement 3 . Sample sizes for in vivo mouse experiments measuring worm burden were chosen to exceed minimum sample sizes determined by power calculations assuming a difference in means of 60% , standard deviation of 20% , power of 0 . 9 ( β = 0 . 1 ) and significance criterion ( α ) of 0 . 05 . Sample size for in vivo mouse survival analysis experiments also assumed a significance criterion ( α ) of 0 . 05 and power of 0 . 9 ( β = 0 . 1 ) , with half of the mice being treated with experimental drug and half with vehicle control and a hazard ratio of 5 . RNA-Seq data was analyzed for differential expression in EdgeR which generated FDR adjusted p-values using the Benjamini-Hochberg method - transcripts with p-values under 0 . 05 were considered differentially expressed .	More than 200 million people worldwide are infected with parasitic worms that cause the disease schistosomiasis . Most cases occur in sub-Saharan Africa . Long-term infections can damage organs , and children who are affected may suffer delayed growth and learning difficulties . Despite its significant health and economic impact , schistosomiasis is still considered a ‘neglected’ tropical disease . This means there has not been adequate investment into developing new treatments or cures . A drug called praziquantel is currently the only treatment for schistosomiasis . However , the drug has unpleasant side effects , cannot cure all infected individuals , and there is a concern that worms may develop resistance to its effects . This means there is an urgent need to develop new therapies . One possible approach would be to develop drugs that interfere with the worm’s ability to move . Chan et al . screened thousands of existing chemicals for interactions with a protein that is known to control how the worms move . A drug called ergotamine , which is currently used to treat migraines , strongly interacted with the protein . Treating infected mice with ergotamine eliminated the parasites and reduced the organ damage caused by the infection . Praziquantel also reduced the number of parasites in the mice but it did not prevent organ damage . The results presented by Chan et al . show that a single drug can interact with targets in both the worm and the animals it infects . Searching for drugs that have this dual effect may help to develop more effective treatments for schistosomiasis and other diseases caused by parasites . Ergotamine itself is unlikely to be used to treat people for schistosomiasis because of the side effects produced when using it repeatedly . However , these findings will help researchers identify and develop safer drugs with similar benefits .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "cell", "biology", "microbiology", "and", "infectious", "disease" ]	2018	Coalescing beneficial host and deleterious antiparasitic actions as an antischistosomal strategy
Antibiotic-induced perturbation of the human gut flora is expected to play an important role in mediating the relationship between antibiotic use and the population prevalence of antibiotic resistance in bacteria , but little is known about how antibiotics affect within-host resistance dynamics . Here we develop a data-driven model of the within-host dynamics of extended-spectrum beta-lactamase ( ESBL ) producing Enterobacteriaceae . We use blaCTX-M ( the most widespread ESBL gene family ) and 16S rRNA ( a proxy for bacterial load ) abundance data from 833 rectal swabs from 133 ESBL-positive patients followed up in a prospective cohort study in three European hospitals . We find that cefuroxime and ceftriaxone are associated with increased blaCTX-M abundance during treatment ( 21% and 10% daily increase , respectively ) , while treatment with meropenem , piperacillin-tazobactam , and oral ciprofloxacin is associated with decreased blaCTX-M ( 8% daily decrease for all ) . The model predicts that typical antibiotic exposures can have substantial long-term effects on blaCTX-M carriage duration . Antibiotic use can increase resistance prevalence in a host population through multiple pathways ( Lipsitch and Samore , 2002 ) . It may: ( i ) affect the duration of resistance carriage and hence transmission potential; ( ii ) increase bacterial load of resistant organisms and thus increase transmission; or ( iii ) selectively suppress host microbial flora where resistance is lacking , which may reduce the potential for transmission of sensitive organisms and also render hosts more susceptible to acquiring resistant bacteria . As well as being important for understanding population dynamics , levels of intestinal resistance are also likely to be important from an individual patient perspective . It has been shown , for instance , that the digestive tract is the primary source of enterobacteria causing bloodstream infections in haematological patients , and a high abundance of beta-lactam resistant enterobacteria in the gut flora is predictive of a high risk of a corresponding drug-resistant bloodstream infection ( Woerther et al . , 2015 ) . Moreover , colonization with extended-spectrum beta-lactamase ( ESBL ) -producing Enterobacteriaceae amongst patients receiving cephalosporin-based prophylaxis prior to colorectal surgery is associated with a more than two-fold increase in risk of surgical site infection ( Dubinsky-Pertzov et al . , 2019 ) . Therefore , quantifying within-host selection dynamics should lead to a better understanding of both individual patient-level and population-level risks and benefits of antibiotic use . Here we focus on Enterobacteriaceae , a bacterial family that is commonly found in the healthy mammalian gut microbiome ( Donnenberg , 1979 ) . Some member genus-species—Klebsiella pneumoniae , Escherichia coli , Enterobacter spp . —are important opportunistic human pathogens that can cause urinary tract , bloodstream , and intra-abdominal infections , as well as hospital-acquired respiratory tract infections . A major concern is the global increase in extended-spectrum beta-lactamase-producing organisms in this family ( Coque et al . , 2008; Tacconelli et al . , 2018; Valverde et al . , 2004 ) . ESBL genes – of which the most important and globally widespread is the blaCTX-M gene family – confer resistance to clinically important broad-spectrum antimicrobials , such as third generation cephalosporins ( Paterson , 2000 ) . These genes commonly reside on large conjugative plasmids ( Bonnet , 2004 ) , and are co-carried with other antibiotic resistance determinants , making them a good marker for multi-drug resistance ( MDR ) in strains of Enterobacteriaceae ( Schwaber et al . , 2005 ) . Because Enterobacteriaceae have their main biological niche in the gut microbiome ( Masci , 2005 ) , these bacteria are exposed to substantial collateral selection from antibiotics used to treat or prevent infections with other organisms ( ‘bystander selection’ [Tedijanto et al . , 2018] ) . Quantifying the effects of antibiotic therapy on within-host resistance dynamics will help us to better understand the potential for selection of drug-resistant Enterobacteriaceae associated with different patterns of antibiotic usage . In this work , we analysed sequential rectal swabs ( n = 833 ) from 133 ESBL positive hospitalised patients from three hospitals ( Italy , Romania , Serbia ) to study the dynamics of antibiotic resistance gene abundance . Both blaCTX-M gene and , as a proxy for total bacterial load , 16S rRNA gene abundance were determined using quantitative polymerase chain reaction ( qPCR ) . Previously , using a subset of these data , Meletiadis et al . demonstrated a statistical association between exposure to ceftriaxone and increases in blaCTX-M normalised by total bacterial load . Here , we addressed some broader questions . We studied the effects of a range of different antibiotics on the abundance of blaCTX-M and of 16S rRNA , and we aimed to fully characterise the within and between host variation of blaCTX-M and 16S rRNA and their within-host dynamics . For this purpose we developed a novel dynamic model , a state-space model that we fit to fine-grained patient-level measurements and antibiotic exposure data . By incorporating hidden-state dynamics our model allowed us to dissect and quantify different types of data variability , such as noise from qPCR measurement or from the DNA extraction process , and to separate this from the within-host processes . In this way we directly estimated ecologically important parameters such as strength of resistance amplification during antibiotic treatment or the rate of decline of blaCTX-M . We then used our model to make counterfactual predictions about how alternative choices of treatment would impact blaCTX-M carriage duration . The development of this data-driven within-host model and its use in exploring the impact of antibiotic treatment on amplification and loss of resistance is an important step in furthering our quantitative mechanistic understanding of how antibiotic use drives changes in the prevalence of resistance in a population . The study enrolled a total of 1102 patients who were screened positive for ESBL producing Enterobacteriaceae at admission , and 133 patients ( 12% ) gave consent to be included in the study: 51 ( 38% ) from Romania; 52 ( 39% ) from Serbia; and 30 ( 23% ) from Italy . The median age was 59 years ( range of 23–88 ) , and 46% were female . The median length of hospital stay was 15 days ( maximum of 53 days ) . All patients apart from one had two or more rectal swabs taken , with a median of five swabs per patient ( range of 1–15 ) . 114 out of 133 ( 86% ) enrolled patients received antibiotics during their stay and 85 of these 114 ( 75% ) received two or more different antibiotics , which were given both in mono- and combination therapy ( see Figure 1 ) . A total of 3993 patient days were observed , of which 2686 ( 67% ) were days with antibiotic therapy ( mono- or combination therapy ) . Table 1 summarises important details of the study . Note that the antibiotics that we considered in this study were exclusively antibacterial drugs , and we ignored treatment with anti-tuberculosis drugs ( pyrazinamide and isoniazid ) , which occurred only in two patients . The different antibiotic classes , ranked by proportion of treatment days , were cephalosporins 25% ) , fluoroquinolones ( 18% ) , penicillins ( 9% ) , nitroimidazole derivatives ( metronidazole ) ( 9% ) , glycopeptides ( 8% ) , carbapenems ( 5% ) , and others ( 26% ) . Two thirds of antibiotic treatment days were from intravenously administered antibiotics and one third from oral administration . Details on individual antibiotics are given in Table 2 . The time-varying blaCTX-M abundance exhibits a diverse range of dynamic patterns , including monotonic increases and decreases , as well as highly variable non-monotonic behaviour ( Figure 1 ) . Qualitatively similar fluctuations in blaCTX-M abundance were seen both in the presence and absence of antibiotic treatment . To determine whether this high level of dynamic variation contained a meaningful biological signal , we first studied temporal autocorrelation . If the observed variability is driven by observation uncertainty – for instance through the swab procedure , DNA extraction , or qPCR process – we expect autocorrelation close to zero in the time series . Conversely , if the observed fluctuations reflect true within-host dynamics in carriage levels , we would generally expect to see positive autocorrelation . We found a clear signal of first-order autocorrelation for both the blaCTX-M and the 16S rRNA gene time series , though autocorrelation was substantially stronger for the blaCTX-M data ( Figure 1—figure supplement 1a and b ) . Using a Bayesian state-space model that decomposes the time series data into an observation component ( representing noise due to variability in qPCR runs , and in the procedure of swab taking and sample processing ) and a process component ( due to the within-host dynamics ) , we estimated that much of the variability in blaCTX-M and 16S rRNA outcomes was due to measurement error associated with the swab procedure ( median estimate of the proportion of total abundance variability attributable to swab error [90% credible interval [CrI]] of 54% [44% , 57%] and 73% [68% , 77%] , respectively ) ( see Figure 1—figure supplement 1c ) . However , the blaCTX-M data in particular were found to also contain a strong process component signal , indicating that a median estimate of 36% ( 90% CrI 30% , 43% ) of the variability in the qPCR outcomes was due to underlying within-host dynamics ( Figure 1—figure supplement 1c ) . To further investigate the determinants of blaCTX-M gene variation , we explored how much the blaCTX-M gene load varied between different patients or , over time , within the same patient . Using a Bayesian state-space model ( see Methods and Materials ) we found 16S rRNA gene abundance to be two orders of magnitude higher than blaCTX-M ( median ratio 16S / blaCTX-M [90% CrI] 158 [88 , 181] ) , with an estimated coefficient of variation ( ratio of standard deviation to the mean ) of 5 . 5 for 16S rRNA and 32 . 1 for blaCTX-M . Between-patient abundance of blaCTX-M showed substantially more variability than within-patient abundance ( median ratio [90% CrI] 134 [18 , 1422] ) . In contrast , 16S rRNA gene abundance had similar between-patient and within-patient variability ( median ratio [90% CrI] 0 . 8 [0 . 4 , 1 . 7] ) ( Figure 2 ) . We noted that the rank plots ( Figure 1—figure supplement 1a ) indicate some convergence problems of σb⁢i⁢o , 16⁢S , but several independent runs of the MCMC algorithm with different initial values consistently arrived at the same mean and standard deviation of the posterior estimate . The change in relative resistance between samples , measured as blaCTX-M abundance divided by 16S rRNA gene abundance , was only slightly elevated in time intervals where antibiotics were given compared to those where they were not ( Figure 3a ) . However , use of antibiotics with activity against Enterobacteriaceae to which carriage of blaCTX-M does not confer resistance ( colistin , meropenem , ertapenem , imipenem , amoxicillin-clavulanic acid , ampicillin-sulbactam , piperacillin-tazobactam , gentamicin , amikacin , ciprofloxacin , ofloxacin , levofloxacin , tigecycline , doxicycline ) was associated with a modest decrease in blaCTX-M abundance ( Figure 3b ) . In contrast , the use of antibiotics with broad spectrum killing activity and to which carriage of blaCTX-M does confer resistance ( cefepime , ceftazidime , ceftriaxone , cefotaxime , cefuroxime , amoxicillin , ampicillin ) was associated with substantially higher increases in relative blaCTX-M abundance ( Figure 3c ) . Fitting a dynamic model of blaCTX-M abundance and 16S rRNA abundance to the data ( 133 patients , 833 swabs , 3361 qPCR measurements ) , we found that cefuroxime and ceftriaxone were associated with increases in both absolute blaCTX-M abundance ( mean daily increase [90% CrI] 21% [1% , 42%] and 10% [4% , 17%] , respectively ) and relative blaCTX-M abundance ( 14% [-1% , 30%] and 11% [5% , 17%] , respectively ) ( Figure 4 ) . Piperacillin-tazobactam , meropenem and ciprofloxacin ( when given orally ) were negatively associated with both blaCTX-M ( −8% [−18% , 2%] , −8% [−17% , 1%] , and −8% [−17% , 2%] , respectively ) and 16S rRNA gene abundance ( −3% [−8% , 1%] , −3% [−7% , 1%] , and −1% [−6% , 3%] , respectively ) , although uncertainty was large ( Figure 4 ) . Their effect on relative resistance ( blaCTX-M/16S rRNA ) also appeared to be negative ( −5% [−14% , 5%] for piperacillin-tazobactam , −5% [−14% , 4%] for meropenem , −7% [−15% , 3%] for oral ciprofloxacin ) . Intravenously administered ciprofloxacin did not show these effects . Imipenem and meropenem had similar effects on blaCTX-M abundance , while no clear effects were evident for amikacin , metronidazole , and amoxicillin-clavulanic acid ( Figure 4 ) . The out of sample prediction accuracy using approximated leave-one-out cross-validation ( loo-cv ) ( Watanabe , 2010 ) ( see Materials and methods ) for the dynamic model with antibiotic effects is higher than the accuracy of the model without antibiotic effects ( Δloo-cv = 4 . 2 ) . With the dynamic model we are able to make predictions about the time required for the blaCTX-M gene to fall below detection levels . To achieve this , we added to our stochastic model a threshold below which the blaCTX-M gene cannot be detected ( see Materials and mthods ) . Below this threshold the gene may either be lost from the bacterial community , or it may exist in very small reservoirs for example in persister cells ( Balaban et al . , 2019 ) . The predictions of detectable carriage duration show a high degree of uncertainty , visible as long-tailed predictive distributions ( Figure 5 ) . Because of the skew , we report here the median instead of the mean together with 80% credible intervals . We chose the duration of different antimicrobial therapies according to clinical guidelines . Assuming that the estimated antibiotic associations represent causal effects , we find that a single eight day course of cefuroxime or a 14 day course of ceftriaxone substantially prolongs carriage of blaCTX-M , by a median estimate of 147% ( 80% CrI 13 . 4% , 577% ) for cefuroxime and 120% ( 80% CrI −8 . 6% , 492% ) for ceftriaxone versus no exposure ( Figure 5 ) . Addition of oral ciprofloxacin to a course of amoxicillin-clavulanic acid or ceftriaxone reduces blaCTX-M carriage duration ( by approximately 51% [80% CrI −115% , 89%] and 48% [80% CrI −71 . 1% , 86%] ) ( Figure 5 ) . A typical 14 day course of meropenem or a 8 day course of piperacillin-tazobactam reduce blaCTX-M carriage duration relative to no treatment ( by approximately 42% [80% CrI −25% , 75%] and 41% [80% CrI −45% , 71%] , respectively ) , and each reduces blaCTX-M carriage even more relative to a 7 day course of combined ceftriaxone plus amikacin ( by approximately 69% [80% CrI 20% , 89%] and 66% [80% CrI −7% , 88%] , respectively ) ( Figure 5 ) . Finally , a 14 day course of meropenem reduces blaCTX-M resistance carriage relative to a shorter 5 day course ( by approximately 69% [80% CrI 20% , 89%] ) ( Figure 5 ) . By fitting a dynamic model accounting for both observation noise and within-host dynamics to time series data from 133 patients , we quantified the association between antibiotic exposure and changes in rectal swab abundance of gut bacteria and blaCTX-M resistance genes . The largest effects were found for exposures to the second and third generation cephalosporins , cefuroxime and ceftriaxone , both of which were associated with increases in blaCTX-M abundance . Forward simulations indicated that if these associations are causal , exposure to typical courses of these antibiotics would be expected to more than double the carriage duration of blaCTX-M . Both cefuroxime and ceftriaxone have broad-spectrum killing activity ( Nahata and Barson , 1985; Neu and Fu , 1978 ) , but have limited activity against ESBL-producing organisms ( Livermore and Brown , 2001; Sorlózano et al . , 2007 ) . Therefore , a direct selective effect of these two antibiotics is biologically plausible to account for the above finding . Though credible intervals were wide , meropenem , piperacillin-tazobactam , and oral ciprofloxacin – all common agents for treating hospital-acquired infections ( Lautenbach et al . , 2001; Masterton et al . , 2003; Paterson , 2006 ) – were associated with reductions in blaCTX-M abundance . All three are broad-spectrum antibiotics with activity against ESBL producers in the absence of specific co-resistance , suggesting that this association may at least in part be a causal effect . These antibiotics were also associated with a negative effect on relative resistance ( blaCTX-M divided by 16S rRNA gene abundance ) . This observation can be explained by a general reduction of bacterial biomass that leads the blaCTX-M abundance to drop below detection levels . In line with this , our simulations suggested that a typical course of meropenem or of piperacillin-tazobactam would reduce blaCTX-M carriage duration relative to no treatment by about 40% , and each course reduces blaCTX-M carriage duration by about 70% relative to a combined course of ceftriaxone plus amikacin . Pharmacokinetic models suggest that bacteriocidal serum concentrations of ceftriaxone persist for relatively short time periods after treatment ( typically between 1 hr and 4d , Garot et al . , 2011 ) . In contrast , our model predicts that treatment effects on the gut flora can be much longer lasting ( on the order of weeks ) . Also , a 14 day course of meropenem is predicted to reduce ESBL resistance carriage duration relative to a shortened course ( 5 days ) by approximately 70% . While these findings suggest suppression of ESBL producing bacteria by use of carbapenem as a measure to reduce the risk of infection in high-risk patients , such a measure clearly needs to be balanced against the considerable risk of selecting for carbapenem resistance . Finally , we also found that adding oral ciprofloxacin to amoxicillin-clavulanic acid or ceftriaxone was associated with reductions in predicted median carriage duration of ESBL-producing bacteria by approximately 50% . This is in line with recent work indicating that hospital monotherapy with cephalosporins is more strongly association with later ESBL carriage relative to combination therapy ( Tacconelli et al . , 2020 ) . Administration of oral fluoroquinolone to reduce faecal load of ESBL-producers in asymptomatic carriers has been used in outbreak settings with ESBL E . coli and K . pneumoniae ( Paterson et al . , 2001 ) , where risk for fluoroquinolones resistance was low . However , the wide variation in rates of ciprofloxacin resistance amongst ESBL-producing Enterobacteriaceae across settings ( Winokur et al . , 2001 ) is likely to limit the generalisability of this finding . Although oral ciprofloxacin showed an association with reduced blaCTX-M abundance , intravenous ciprofloxacin showed near zero effect . Antibiotic selection for resistance due to antibiotics with different routes of administration has been previously explored in a mouse model , suggesting that , indeed , oral drug administration has stronger selective effect on resistance than intravenous administration ( Zhang et al . , 2013 ) , but similar studies for humans are lacking . Delineating the relationship between the various routes of antibiotic administration and resistance selection will be important for a better understanding of advantages and disadvantages of different routes of administration . There are are number of advantages to our modelling approach over more classical , associational methods used in related work ( Meletiadis et al . , 2017 ) . First , because we use a mechanistic model it allows us to directly estimate ecologically important parameters such as strength of resistance amplification under antibiotic selection , which are of inherent interest and can inform further modelling work . Indeed , our predictions of resistance carriage duration ( Figure 5 ) are good examples of the latter . Second , our model uses the variability present in the data to quantify different types of data variability due to the data collection ( noise from the qPCR machine , noise from taking the swab and DNA extraction ) . This allows our model to fully propagate uncertainty to the final estimates . Finally , rather than using only aggregate data ( which loses information ) , our analysis is designed to fully exploit the information available in the time series data . Our work also has a number of important limitations , aside from the obvious risks of confounding present in this observational dataset . Our analysis did not explicitly model changing antibiotic concentrations in the gut , nor did it attempt to explicitly model how antibiotics affect the ecology of the gut bacterial community . While it would have been straightforward to include a pharmacokinetic model of antibiotic concentrations ( similar studies have been performed in mice [Jumbe et al . , 2003] and pigs [Nguyen et al . , 2014] ) , disentangling direct effects of antibiotic concentrations from indirect effects mediated by other components of the gut flora is far more challenging and beyond the scope of what we considered appropriate with the available data . Instead , our model assumed multiplicative antibiotic effects , which we considered a reasonable simplification of the underlying mechanisms . Multiplicative effects imply that antibiotics alter the daily total bacterial growth rate ( 16S abundance ) and the growth of resistant bacteria relative to the average bacterium ( blaCTX-M/16S ) , but it does not allow for more complex fitness effects due to , for example , synergies between antibiotics ( MacLean et al . , 2010 ) , or density-dependent effects , whereby antibiotic-mediated killing may depend on bacterial density ( Udekwu et al . , 2009 ) . Further , many non-antibiotic drugs have been shown to have an impact on human gut bacteria ( Maier et al . , 2018 ) , but only antibacterials were considered in our analysis . Lastly , all patients in this study were identified ( and consenting ) ESBL-carriers . Therefore , apart from the potential for selection bias , we assumed that all changes in blaCTX-M abundance were due to within-host dynamics , neglecting the possibility of new acquisitions , which should be the scope for other modelling frameworks that integrate both within- and between-host dynamics . Antibiotic impact on the human gut microbiome is likely to be an important mediator for the increase of bacterial resistance globally ( Donskey , 2004; Relman and Lipsitch , 2018 ) . A large body of theory has been developed that demonstrates the role of within-host processes for understanding population-wide selection of resistance through antibiotic use ( Blanquart , 2019; Davies et al . , 2019; Webb et al . , 2005; Knight et al . , 2018; Lipsitch et al . , 2000; Lipsitch and Samore , 2002; Webb et al . , 2005 ) . However , surprisingly few studies have used data to quantify the effect of antibiotic treatment on resistance abundance within individual gut microbiomes . Two studies involving patients admitted to intensive care units looked at the effect of a preventative antibiotic cocktail ( selective digestive decontamination ) on gut microbiome resistance in patients , with one study ( including n = 13 patients ) finding no clear effect ( Buelow et al . , 2014 ) , and the other ( n = 10 ) showing increases of four different resistance genes associated with treatment ( Buelow et al . , 2017 ) . Gibson et al . ( 2016 ) studied the faecal metagenomics of preterm infants ( n = 84 ) over time and found that treatment with antibiotics was correlated with enrichment of both cognate and noncognate resistance markers . But none of these studies attempted to model resistance dynamics . The modelling framework we have developed enables testable predictions about the impact of different antibiotics and lengths of treatment on the duration of carriage of resistant determinants above detection threshold . Such understanding is an important step toward understanding the spread of antibiotic resistance . The development of a mechanistic understanding of the relationship between antibiotic use in a population and the proportion of this population in whom resistance can be detected relies on quantifying the antibiotic effects in individual exposed patients , as we do here , but also on quantifying the knock-on effects on transmission to contacts . These indirect effects are likely to be considerable . A recent study in Dutch travellers returning to the Netherlands who had acquired ESBL-producing Enterobacteriaceae carriage overseas found that their new carriage status was associated with a 150% increase in the daily risk of non-carrying household members also becoming ESBL-positive ( Arcilla et al . , 2017 ) . Developing mechanistic models for the spread of ESBLs and other resistance determinants within host populations accounting for direct and indirect antibiotic effects is an important priority for future research . Such models would help us to understand and predict how changes in antibiotic usage patterns affect the prevalence of antimicrobial resistance in a community and ultimately help to prioritise interventions to reduce the burden of antimicrobial resistance . Participants were recruited as part of an observational , prospective , cohort study that included three hospitals ( Italy , Serbia and Romania ) , with known high endemic prevalence of antibiotic resistance in bacterial infections . The hospitals were serving a general urban population . The study was conducted over two years from January 2011 to December 2012 as part of the multi-centre SATURN ( ‘Impact of Specific Antibiotic Therapies on the prevalence of hUman host ResistaNt bacteria’ ) project ( NO241796; clinicaltrials . gov NTC01208519 ) . The study enrolled adult ( >18 y ) inpatients of medical and surgical wards , excluding pregnant patients . Enrolled patients were screened at admission for carriage of ESBL-producing Enterobacteriaceae with rectal swabs ( E swab , Copan , Italy ) . Patients who tested positive for ESBL producing Enterobacteriaceae carriage ( details below ) were included in the follow-up cohort . The target cohort size was calculated to be 400 patients , based on the number of patients that would be required to detect one log difference in resistance abundance with 90% certainty . However , patient recruitment we slower than anticipated . For all follow-up patients ( n = 133 ) rectal swabs were taken every two to three days ( as per study protocol ) during hospitalisation , which includes one swab at admission and one at discharge . The swabs were stored at −80 degrees Celsius and sent to a central laboratory for processing . Using patient charts , the study also collected information on antibiotics treatment , including antibiotic type , duration , and route of administration . See Table 1 for an overview of the study details . Written informed consent and consent to publish was obtained from all patients before study enrolment . All collected data was entered de-identified into the central study database which sat in Tel Aviv in accordance with the local rules of personal data privacy protection . The study protocol was reviewed and approved by the Catholic University Ethics Commission in Rome , Italy ( protocol P/291/CE/2010 approved on 6 . 4 . 2010 ) and the Clinical Center of Serbia Ethic Committee ( protocol 451/34 approved on 18 . 03 . 2010 ) . At the site in Romania patient screening for multidrug-resistant bacteria was considered , due to the local epidemiology , a quality improvement intervention and did not require institutional ethical approval . Samples taken at admission were cultured on chromogenic agar ( Brilliance ESBL , Oxoid , Basingstoke , UK ) to test for ESBL producing Enterobacteriaceae . Characteristically coloured colonies for Enterobacteriaceae were isolated ( single colony per colour ) , replated on blood agar and incubated overnight in air . ESBL status was then confirmed with the double disk diffusion method according to CLSI guidelines . These methods were performed in the laboratories on the respective hospitals ( all laboratories were ISO accredited ) . The above methods are not specific to any single bacterial species , but instead identify ESBL producing specimens of the Enterobacteriaceae family , including Escherichia coli , Klebsiella pneumoniae , or Enterobacter cloacae and others . According to the standard definition by the Centers of Disease Control ( CDC/NHSN , 2018 ) , samples taken at admission identify community acquired organisms . DNA was extracted from the swab samples using QIAampDNA Stool Mini Kit ( Qiagen ) and a fixed volume ( 4 µl ) of DNA solution was used as a template for quantitative PCR ( qPCR ) assays . Two singleplex qPCR assays were conducted , one to assess quantity of blaCTX-M gene family with primers CTX-M-A6 ( TGGTRAYRTGGMTBAARGGCA ) and CTX-M-A8 ( TGGGTRAARTARGTSACCAGAA ) ( product length , 175 bp ) and one targeting a conserved bacterial 16S rRNA gene region bacteria using the following primer set , 16S_E939F ( GAATTGACGGGGGCCCGCACAAG ) and 16S_1492R ( TACGGYTACCTTGTTACGACTT ) ( product length , 597 bp ) to assess total bacterial quantity . Each singleplex qPCR run targeted either 16S gene or blaCTX-M , and included reaction tubes with negative controls and tubes containing a standard ( 16S or blaCTX-M depending on the run ) of different concentrations ( eight different dilutions ) , which also served as a positive control . For details on reaction mix and cycling conditions see Lerner et al . , 2013 . All reactions were carried out in duplicates , sometimes triplicates , representing technical replicates . The qPCR was performed at the Laboratory of Medical Microbiology , University of Antwerp ( ISO accredited ) . While we did not validate the 16S qPCR measurements through , for example , spiked standard DNA preparations , the ability of qPCR to quantify bacterial amounts in faecal samples has been shown previously ( Rinttilä et al . , 2004 ) . We first transformed all qPCR measurements onto a log-scale . For all patients and each time point we then computed the mean of the qPCR duplicates ( or triplicates ) for blaCTX-M and 16S rRNA . To get reliable estimates of autocorrelation , we selected only patients with more than five time points . Separately for the blaCTX-M and 16S rRNA gene data , we computed the first-order autocorrelation ( disregarding varying spacing between time points ) for each patient , and we averaged these values across the patients . We then simulated serially uncorrelated ‘white noise’ time series , again separately for blaCTX-M and 16S rRNA , with the same length as the patient data and with identical time series mean and variance . Similar to the real data , we computed mean autocorrelations for the simulated data and show their distribution for a large number of such simulations ( n = 10 , 000 ) together with the observed autocorrelation ( Figure 1—figure supplement 1a and b ) . We also computed the proportion of simulated datasets that showed an average autocorrelation equal to , or larger than , the observed data , and we show those numbers on the arrows in Figure 1—figure supplement 1a and b . To estimate the amount of observation noise and process noise in the time series we constructed a Bayesian state-space model that included qPCR noise , swab noise , and biological noise . This model is given through: ( 1 ) qi , j , k , g∼N ( si , j , g , σqpcr ) , si , j , g∼N ( xi , j , g , σswabg ) , xi , j+1 , g∼N ( si , j , g , σbiog ) , where i denotes a given patient , j denotes a swab ( one per time point ) , k denotes a qPCR measurement ( multiple repeats per swab ) , and g denotes the genetic target , either blaCTX-M or 16S rRNA . The term qi , j , k , g , represents the measured quantity of genetic target g , of the kth qPCR replicate ( on a log-scale ) from patient i , at time point j . In addition , there are two hidden-state parameter vectors: si , j , g is the underlying , true sequence abundance of genetic target g that a qPCR assay with 100% efficiency could ( in theory ) measure at time point j for patient i , and xi , j , g is the actual gene abundance of genetic target g , in the patient at time point j for patient i , before the added noise through the swab process and gene extraction . The unobserved variables of interest are σq⁢p⁢c⁢r , the qPCR machine error ( assumed to be the same for blaCTX-M and 16S rRNA ) , σs⁢w⁢a⁢bg , the swab variation of the genetic target g , and σb⁢i⁢og , the variation of genetic target g from biological processes . We assigned improper uniform ( ‘flat’ ) priors for the hidden-state parameters and generic weakly informative priors ( half-normal , N+ ( 0 , 1 ) ) for the the noise parameters σq⁢p⁢c⁢r , σs⁢w⁢a⁢bg , and σb⁢i⁢og . We then fitted this model to the blaCTX-M and 16S rRNA measurements . The posteriors of the three noise parameters are shown in Figure 1—figure supplement 1c , where we expressed each type of noise as a fraction of the total noise . The model was fitted using Stan software ( v2 . 19 . 1 ) ( Carpenter et al . , 2017 ) and with additional analysis in R ( R Development Core Team , 2016 ) , and we sampled 80 , 000 samples from the posterior with four independent chains and a burn-in period of 20 , 000 samples . We assessed convergence by checking that R̂ ( Gelman and Rubin , 1992 ) was low ( <1 . 01 ) for all parameters , and visually by looking at the rank plots for all parameters ( shown in Figure 2—figure supplement 1a ) . For rank plots , posterior draws are ranked across all chains and then ranks are plotted as histograms separately for each chain . A uniform shape of all histograms then indicates that all chains target the same posterior ( Vehtari et al . , 2019 ) . For the estimation of between and within time series variation we used a Bayesian hierarchical model , which accounted for unbalanced sampling between patients . This model used the mean posterior estimates of xi , j ( actual gene abundance in time point j for patient i ) from the previous model , and it took the form ( 2 ) xi , j , g∼N ( μi , g , σwithin , g ) , μi , g∼N ( μg , σbetween , g ) , where m⁢ui , g is the mean abundance of genetic target g ( blaCTX-M or 16S gene ) and patient i , around which the log-scaled measurements were assumed to be normally distributed with standard deviation σw⁢i⁢t⁢h⁢i⁢n , g , the within time series variation . The mean abundances were assumed to follow a normal distribution with a population mean µ and between-patient standard deviation σb⁢e⁢t⁢w⁢e⁢e⁢n . We assigned improper uniform priors for the population and the patient means , and generic weakly informative priors for the standard deviations ( N+ ( 0 , 1 ) ) . We fitted the model using Stan ( v2 . 19 . 1 ) , with 80 , 000 posterior draws after a burn-in period of 500 iterations . The R̂ statistic ( <1 . 01 ) and MCMC rank histogram plots ( Figure 2—figure supplement 1b ) were used to assess convergence . Model estimates are shown in Figure 2 . To calculate the coefficient of variation for the non-log-scaled blaCTX-M and 16S rRNA measurements , we use the transform described by Koopmans et al . , 1964: ( 3 ) cv=esl⁢n2-1 , where sl⁢n is the estimated standard deviation of the log-scaled data . To study the association between antibiotic treatment and resistance we looked at relative abundance of resistance ( blaCTX-M abundance/16S rRNA gene abundance ) as a marker of natural selection . First , we computed the changes in relative resistance for every pair of adjacent time points and for each antibiotic we used a binary variable indicating whether or not a given antibiotic was administered between these time points . When an antibiotic treatment was on the same day as a swab , this treatment was allocated to the time interval between this day and the next swab . We first looked at how changes in relative resistance are associated with courses of any antibiotics , then with courses of anti-enterbacteriaceae antibiotics to which carriage of blaCTX-M does not confer direct resistance ( colistin , meropenem , ertapenem , imipenem , amoxicillin-clavulanic acid , ampicillin-sulbactam , piperacillin-tazobactam , gentamicin , amikacin , ciprofloxacin , ofloxacin , levofloxacin , tigecycline , doxicycline ) , and finally with antibiotics that have a broad-spectrum activity and to which blaCTX-M does confer resistance ( cefepime , ceftazidime , ceftriaxone , cefotaxime , cefuroxime , amoxicillin , ampicillin ) . Results are shown in Figure 3 , upper panel . We evaluated how likely the observed differences between treatments are under the assumption of no association between treatment and resistance . For this we did a permutation or ‘reshuffling’ experiment: we randomly reassigned ( without replacement ) the antibiotic treatment labels to the data intervals . We compute the distribution of mean differences from 50 , 000 permutations and compare this to the observed difference ( Figure 3 , lower panel ) . We extended previous modelling approaches to extracting ecological parameters from microbial ecosystem dynamics ( Faust and Raes , 2012; Stein et al . , 2013 ) by applying a Bayesian hidden-state model , which featured two layers of hidden-state variables: the unobserved mean of the qPCR measurements , and the unobserved true abundances of blaCTX-M or 16S rRNA in the gut . This model structure allowed us to separate process noise ( stochastic effects impacting the gene abundance change from one day to the next ) from observation noise ( stochastic effects impacting the swab efficiency , DNA extraction or the qPCR measurements ) and also to account for different spacing between measured time points . We analysed antibiotic treatment separately by type and route of administration , only including treatments that occurred in five or more patients ( amoxicillin-clavulanic acid ( iv ) , piperacillin-tazobactam ( iv ) , cefuroxime ( iv ) , ceftriaxone ( iv ) , meropenem ( iv ) , imipenem ( iv ) , ciprofloxacin ( iv ) , ciprofloxacin ( or ) , amikacin ( iv ) , metronidazole ( iv ) ) . Since the blaCTX-M and 16S rRNA measurements are expected to be proportional to the absolute abundance of the resistance gene and bacterial load respectively , the ratio blaCTX-M/16S is a measure of the relative abundance of the blaCTX-M gene in the gut microbiota . Positive or negative selective effects by antibiotics on blaCTX-M mediated resistance are expected to cause shifts in bacteria carrying blaCTX-M versus non-carriers . As a result they affect blaCTX-M/16S , but quantifying their effects on absolute blaCTX-M abundance is important for predicting extinction and persistence of the blaCTX-M gene . Under the assumption that 16S rRNA gene abundance is independent of antibiotic treatment , variation in 16S rRNA would be caused mainly by the swab procedure and DNA extraction ( and other steps in the protocol ) , and it could be used to normalise blaCTX-M abundance . However , as we found in Figure 4 , certain antibiotic treatments were associated with changes in 16S rRNA abundance . Thus , we used a dynamic model that explicitly modelled both antibiotic effects on 16S rRNA and on blaCTX-M/16S , from which the effects on blaCTX-M could then be computed . Studying the standard deviation between qPCR measurement repeats as a function of the mean , we observed that qPCR variation remained relatively stable over five orders of magnitude of the mean measurement ( from 1 . 5 to 6 . 5 on the log-scale ) , but it increased quickly for lower magnitudes ( Figure 4—figure supplement 1 ) . In the Bayesian model for different sources of variation described above , the parameter σq⁢p⁢c⁢r assumed that the qPCR uncertainty is the same across measurements . Here , we aimed to account for the fact that low measurements of gene copy numbers have higher uncertainty . We fitted a smooth spline ( choosing five degrees of freedom ) to the qPCR measurements ( red line in Figure 4—figure supplement 1 ) . This let us assign an estimated qPCR standard deviation to every set of qPCR repeats . We provided those estimates as data to the Bayesian model . This allowed us to use all qPCR measurements , including extremely low values , without removing any data points from the analysis . Our model then took the form: ( 4 ) qi , j , g , k∼N ( si , j , g , σqpcri , j , g ) , si , j , g=16S∼N ( xi , j , b=16S , σswab ) , xi , j , b=ratio=si , j , g=CTX−M−si , j , g=16S , xi , j+1 , b∼N ( xi , j , b+f ( abx ) i , j , b , tj+1−tjσbio , b ) , f ( abx ) i , j , b=∑tjtj+1−1 ( ab+∑z=1nzcz , byz , t ) , where g denotes either blaCTX-M or 16S rRNA , and b denotes either relative resistance ( blaCTX-M/16S rRNA ) or 16S rRNA . Then , qi , j , g , k is the k-th qPCR result ( log-scaled ) of patient i , measured in the sample with index j , and genetic target g ( blaCTX-M or 16S rRNA ) . The qPCR standard deviation for sample j and patient i is given through σq⁢p⁢c⁢ri , j , g , and it is estimated as described above . si , j , g is the mean of the qPCR measurements in sample j of patient i , and genetic target g , and xi , j , b=16⁢S is the 16S sequence abundance that is actually present at time point j in the gut of patient i . The error introduced by the swab procedure , DNA extraction etc . is given through σs⁢w⁢a⁢b , and we assume that this error causes the same perturbations to blaCTX-M and to 16S , such that this error cancels out when computing the ratio of blaCTX-M and 16S rRNA xi , j , b=r⁢a⁢t⁢i⁢o , which on a log-scale is computed as the difference ( si , j , g=C⁢T⁢X-M-si , j , g=16⁢S ) . Further , tj denotes the calendar day of sample j in patient i , and tj+1 denotes the calendar day of the following sample in the same patient i . The daily biological variability of the blaCTX-M/16S ratio and of 16S are given through σb⁢i⁢o , b=r⁢a⁢t⁢i⁢o and σb⁢i⁢o , b=16⁢S , respectively , with tj+1-tj adjusting the expected random walk variation by the number of days between observations ( Lemons and Langevin , 2002 ) . The ecological dynamics are modelled with the function f⁢ ( a⁢b⁢x ) i , j , b , which is the change in the expected value of xi , j , b between sample j and j+1 . In the definition of f⁢ ( a⁢b⁢x ) i , j , b in line 5 of Equation 4 , ab denotes the neutral growth or loss of g , cz , b denotes the effect of antibiotic z on b , and yz , t is a boolean variable indicating whether or not antibiotic z was given on day t . The term inside the bracket takes , for a single calendar day t , the neutral growth/loss term ( ab ) and adds to this the summed effect of all antibiotics given on day t . This is computed for each calendar day from tj until a day before the subsequent sample ( tj+1-1 ) . The effects of all of these days are then summed up . Note , that xi , j , b denotes the abundance of 16S rRNA , or the relative abundance of blaCTX-M/16S rRNA , on the log-scale . Exponentiating this variable gives the copy numbers ( or copy number ratio ) on the real scale . Therefore , summing all effects on the scale of xi , j , b is equivalent to multiplying the exponentiated effects on the scale of copy numbers . For example , consider that genetic target b=16⁢S in patient i=1 has at the time of sample j=1 an abundance of 10xi=1 , j=1 , b=16⁢S=100 copy numbers and a neutral trend of ab=16⁢S=-0 . 5 . Suppose on the day of this sample ( ti=1 , j=1 ) two antibiotics z=1 and z=2 are given with effects cz=1 , b=16⁢S=+0 . 5 and cz=2 , b=16⁢S=+0 . 1 , then one day after this sample the genetic target abundance has an expected abundance of 10010-0 . 510+0 . 5*10+0 . 1=126 . In this model , qi , j , g , k and yz , t correspond to measured data , σq⁢p⁢c⁢ri , j , g is computed from this data ( see above ) . All other parameters are estimated: the hidden-state variables are si , j , g , xi , j , g=16⁢S , and xi , j , g=r⁢a⁢t⁢i⁢o , the noise variables are σs⁢w⁢a⁢b , σb⁢i⁢o , g=16⁢S , and σb⁢i⁢o , g=r⁢a⁢t⁢i⁢o , and finally the variables describing the ecology are ag , cz , g . The model has three likelihood functions . The first ( line 1 ) applies to each single qPCR result and it relates repeat qPCR measurements to their variability and underlying mean . The second ( line 2 ) relates qPCR means of 16S rRNA to their variability and the underlying 16S rRNA gene abundance . The third likelihood ( line 4 ) applies to all sample pairs where a previous and following sample exists from the same patient . This likelihood relates changes in underlying 16S rRNA abundance or underlying blaCTX-M/16S rRNA to the parameters ab and cz , b , and σb⁢i⁢o . All parameters ( including all hidden-states ) are estimated simultaneously . We can express the posterior distribution over all estimated parameters ( Θ ) , which is conditional on the set of all data ( D ) , and the prior over the parameter space p⁢ ( Θ ) . For readability , here we only keep necessary subscripts: ( 5 ) p ( Θ\|D ) =p ( s , x , a , c , σswab , σbio\|q , σqpcr , y ) , p ( Θ\|D ) ∝p ( Θ ) ∏i , j , g , kp ( qi , j , g , k\|si , j , g , σqpcri , j , g ) ∏i , jp ( si , j , g=16S\|xi , j , b=16S , σswab ) ∏i , j , bp ( xi , j+1 , b\|cb , ab , σbio , b , xi , j , b , y ) . On the hidden-state variables we assigned improper uniform priors , on the standard deviations describing swab and biological variability we assigned standard weakly informative priors ( N+ ( 0 , 1 ) ) , and on the antibiotic effects ( c ) we assigned conservative priors of the form N ( 0 , 0 . 1 ) . We fitted the model using Stan software ( v2 . 19 . 1 ) , and we sampled 80 , 000 samples from the posterior with four independent chains after a burn-in phase of 10 , 000 samples . The marginal posterior draws of the cz , g parameters are exponentiated to be on the scale of gene counts . Subtracting one allows us to express daily effects as percent of increase or decrease relative to the previous day ( Figure 4 ) . We also show marginal posterior distributions together with prior distributions for the cz , g parameters ( Figure 4—figure supplement 2 ) . Diagnostic plots for the MCMC sampling of the cz , g parameters are shown in Figure 4—figure supplement 3 . We compared our model as given through model definition four to a model without antibiotic effects ( all c parameters set to zero ) . The number of patients treated with the same antibiotic is too small to perform cross-validation ( iteratively fitting the model to all time series while leaving out data from one patient , which is then used to assess model predictions ) . Therefore , we used an efficient approximation of Bayesian leave-one-out cross-validation using Pareto-smoothed importance sampling ( Vehtari et al . , 2017 ) . We forward simulated blaCTX-M data using the dynamic model above and the posterior distributions from the model fit . We added to the model a threshold below which the blaCTX-M gene becomes extinct or at least undetectable . According to a study of returning European travelers to Southeast Asia , ESBL carriers lose detectable resistant bacteria after a median of 30 days ( Arcilla et al . , 2017 ) . Accordingly , we simulated blaCTX-M time series without antibiotic treatment and chose an extinction threshold ( 0 . 25 blaCTX-M copy numbers ) that achieved the same median extinction time . We then used this model to repeatedly ( 2000 times ) simulate blaCTX-M carriage durations , with each simulation using a new draw from the parameter posterior . The resulting distribution of carriage times contains both the uncertainty in the parameter estimates and uncertainty from the Markov process ( Figure 5 , right-hand side ) . We also draw a set of parameter values from the posterior to simulate blaCTX-M carriage durations repeatedly ( 200 times ) with the same parameters , and taking the median carriage duration to remove Markov process uncertainty . We repeated this for 300 draws of parameters ( Figure 5 , left-hand side ) . We used both of the above methods to simulate carriage time under different alternative antibiotic treatments . The resulting distributions are shown in Figure 5 .	Bacteria that are resistant to antibiotics are a growing global health crisis . One type of antibiotic resistance arises when certain bacteria that can produce enzymes called extended-spectrum beta-lactamases ( or ESBLs for short ) become more common in the gut . These enzymes stop important antibiotics , like penicillin , from working . However , exactly which antibiotics and treatment durations contribute to the emergence of this antibiotic resistance remain unknown . Now , Niehus et al . find certain antibiotics that are associated with an increase in the number of gut bacteria carrying antibiotic resistance genes for ESBL enzymes . First , rectal swabs collected from 133 patients from three European hospitals were analysed to measure the total gut bacteria and the number of genes for ESBL enzymes . These samples had been collected at several time points including when the patient was first admitted to hospital , then every two to three days during their stay , and finally when they were discharged . Combining the analysis of the samples with details of the patients’ charts showed that treatment with two antibiotics: cefuroxime and ceftriaxone , was linked to an increase in ESBL genes in the gut bacteria . Other antibiotics – namely , meropenem , piperacillin-tazobactam and oral ciprofloxacin – were associated with a decrease in the number of bacteria with ESBL genes . Niehus et al . then performed further analysis to see if different treatment regimens affected how long patients were carrying gut bacteria with ESBL genes . This predicted that a longer course of meropenem , 14 days rather than 5 days , would shorten the length of time patients carried ESBL-resistant bacteria in their guts by 70% , although this effect will likely depend on the location of the hospital and the local prevalence of other types of antibiotic resistance . This analysis reveals new details about how antibiotic treatment can affect ESBL resistance genes . More studies are needed to understand how antibiotics affect other antibiotic resistance genes and how resistant bacteria spread . This will help scientists understand how much specific antibiotic regimens contribute to antibiotic resistance . It may also help scientists develop new antibiotic treatment strategies that reduce antibiotic resistance .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "epidemiology", "and", "global", "health", "microbiology", "and", "infectious", "disease" ]	2020	Quantifying antibiotic impact on within-patient dynamics of extended-spectrum beta-lactamase resistance
Cyclic nucleotide-modulated channels have important roles in visual signal transduction and pacemaking . Binding of cyclic nucleotides ( cAMP/cGMP ) elicits diverse functional responses in different channels within the family despite their high sequence and structure homology . The molecular mechanisms responsible for ligand discrimination and gating are unknown due to lack of correspondence between structural information and functional states . Using single particle cryo-electron microscopy and single-channel recording , we assigned functional states to high-resolution structures of SthK , a prokaryotic cyclic nucleotide-gated channel . The structures for apo , cAMP-bound , and cGMP-bound SthK in lipid nanodiscs , correspond to no , moderate , and low single-channel activity , respectively , consistent with the observation that all structures are in resting , closed states . The similarity between apo and ligand-bound structures indicates that ligand-binding domains are strongly coupled to pore and SthK gates in an allosteric , concerted fashion . The different orientations of cAMP and cGMP in the ‘resting’ and ‘activated’ structures suggest a mechanism for ligand discrimination . Cyclic nucleotide modulated ion channels are physiologically important for visual and olfactory signal transduction and pacemaking activity in the heart and brain ( Kaupp and Seifert , 2002; Craven and Zagotta , 2006; Robinson and Siegelbaum , 2003; Biel et al . , 2009 ) . Their activity is dependent on the binding or dissociation of cyclic nucleotides ( cAMP or cGMP ) to a cytoplasmic cyclic nucleotide-binding domain ( CNBD ) within the channel . This in turn leads to changes in the ionic flow across the membrane and alterations in the membrane potential , allowing an intracellular signal such as an increase in cAMP concentration to be transmitted across the membrane of the cell . The magnitude of the ion channel response to ligand binding modulates the final cellular response and can lead to different physiological outcomes ( James and Zagotta , 2018 ) . Thus , it is important to understand the molecular mechanism by which binding of ligands leads to modification of the open-closed equilibrium in these ion channels . Cyclic nucleotide-modulated channels comprise a large family of ion channels in both eukaryotes and prokaryotes ( Kaupp and Seifert , 2002; Craven and Zagotta , 2006 ) . In eukaryotes , they are categorized into two subfamilies: cyclic nucleotide-gated ( CNG ) and hyperpolarization-activated and cyclic nucleotide-modulated ( HCN ) channels . HCN channels are activated by hyperpolarizing voltages and this activity is modulated by cyclic nucleotide binding ( Santoro et al . , 1998; Lyashchenko et al . , 2014; Wang et al . , 2002; Zhou and Siegelbaum , 2007; Kusch et al . , 2010; Ludwig et al . , 1998 ) , while CNG channels are activated by cyclic nucleotides and their activity is only slightly modulated by depolarizing voltages ( Kaupp et al . , 1989; Benndorf et al . , 1999; Clayton et al . , 2004; Nache et al . , 2006 ) . Prokaryotic cyclic nucleotide-modulated channels , homologous in both sequence and structure to their eukaryotic counterparts , have been identified ( Nimigean et al . , 2004; Brams et al . , 2014; James et al . , 2017 ) , but not investigated enough to warrant a categorization into one of these two groups . Ligand binding and activation have been extensively studied in eukaryotic CNG and HCN channels using electrophysiology on channels expressed in heterologous expression systems ( Kaupp and Seifert , 2001 , 2002 ) . However , only recently , high resolution structural information for human HCN1 ( Lee and MacKinnon , 2017 ) and TAX-4 ( Li et al . , 2017 ) ( a C . elegans CNG channel homolog ) , became available with the advent of single-particle cryo-electron microscopy ( cryo-EM ) ( Kuhlbrandt , 2014 ) . For HCN1 , two structures were reported , in an apo and a cAMP-bound , closed conformation , while TAX-4 is in a cGMP-bound open conformation only . HCN1 channels are activated by hyperpolarization , while the addition of saturating cAMP only marginally shifts the activation ( Santoro et al . , 1998 ) , questioning whether cAMP is an agonist for this channel . Different models for ligand gating have been proposed for HCN and CNG channels based on functional studies ( Biskup et al . , 2007; Li et al . , 1997; Ulens and Siegelbaum , 2003 ) . These models are all versions or combinations of two fundamental models proposed to work in allosteric proteins: the concerted and the sequential model ( Koshland et al . , 1966; Monod et al . , 1965 ) . In the concerted model , all subunits of the tetrameric channel change their conformation together , all at once , upon ligand binding and accordingly all subunits in one channel adopt the same conformation whether open or closed and irrespective of how many ligands are bound . The sequential model , in contrast , allows individual domains and/or subunits to change conformation independently upon ligand binding , one after the other , on the pathway towards the open state . The existing structures ( James et al . , 2017; Lee and MacKinnon , 2017; Li et al . , 2017 ) are not sufficient to argue in favor of either model because there are not enough conformations available and because of the lack of direct correspondence between conformations and functional states . In order to address this void , we selected SthK , a prokaryotic cyclic nucleotide-modulated channel from Spirochaeta thermophila ( Brams et al . , 2014 ) . We showed previously that SthK is activated by cyclic nucleotides and its activity is modulated by depolarizing voltages , making it more similar in this respect to CNG rather than HCN channels ( Schmidpeter et al . , 2018 ) . The channel shares about 35% sequence similarity with eukaryotic CNG channels , and , unlike its eukaryotic counterparts , lends itself easily to high-resolution single-channel functional studies as well as structural investigations , providing a unique model to investigate ligand gating ( Schmidpeter et al . , 2018 ) . We present here the high-resolution cryo-EM structures of SthK in the apo , cAMP-bound , and cGMP-bound states in lipid nanodiscs . This is , to our knowledge , the first apo , resting structure of a member of the CNG channel family . The single channel electrophysiology data allow us to specifically assign functional states to the SthK structures in different ligand-bound conformations , and we propose a concerted mechanism for CNG channel gating , as well as a molecular basis for cyclic nucleotide discrimination . SthK has been previously reported to function as a CNG-like channel when expressed in oocytes ( Brams et al . , 2014 ) , and structures of the soluble ligand-binding domains in complex with both cAMP and cGMP have been reported ( Kesters et al . , 2015 ) . We recently described an SthK ion channel construct that robustly expresses in E . coli and we established that the purified protein is active in binding , flux , and single-channel electrophysiology assays ( Schmidpeter et al . , 2018 ) . Purified and reconstituted SthK showed no activity in cyclic nucleotide-free conditions , became active upon cAMP application , channel activity was increased upon membrane depolarization and the channel showed no evidence of inactivation ( Schmidpeter et al . , 2018 ) ( Figure 1 , Figure 4B ) . Nevertheless , the maximal open probability of SthK in saturating concentrations of cAMP was less than 1 , even in the presence of depolarizing voltages , indicating that cAMP is a partial agonist for SthK . The cAMP-induced activity is inhibited by cGMP application , although infrequent single-channel openings are still observed ( Schmidpeter et al . , 2018 ) . Unlike in cyclic nucleotide-free conditions , the infrequent openings are still present in high concentrations of cGMP alone , indicating that cGMP is not an antagonist for SthK , as has been previously proposed based on patch-clamp recordings from oocytes expressing SthK ( Brams et al . , 2014 ) , but rather an extremely poor partial agonist ( Figure 1E ) . The difference in attribution is likely due to the extremely low open probability of SthK in cGMP making it easy to miss when recording from cells , or from the difference in the experimental setup ( purified channels in controlled lipid bilayer composition versus channels expressed in oocytes ) . In order to get insights into partial agonism in SthK , we solved the structures of SthK with single-particle cryo-EM in three different states: apo ( cyclic nucleotide-free conformation ) , cAMP-bound ( partial agonist-bound conformation ) , and cGMP-bound ( poor partial agonist-bound conformation ) ( Figure 1A , C and D ) . The three structures are with the SthK channel reconstituted in lipid nanodiscs , in order to recapitulate a native-like membrane environment ( Figure 1—figure supplement 1 ) . The structures were solved to similar resolutions , 3 . 42 , 3 . 35 , and 3 . 46 Å for the apo , the cAMP-bound , and the cGMP-bound states , respectively ( Figure 1—figure supplements 2–5 , Supplementary file 1 ) . The apo structure adopts an architecture different from any of the existing CNG/HCN structures ( Lee and MacKinnon , 2017; Li et al . , 2017 ) , which is expected , as it is the first apo , resting state structure of a member of the CNG channel family ( Figure 2 ) . Unexpectedly however , the cAMP- and cGMP-bound SthK structures also display a resting , apo-like ligand binding pocket , despite the resolved , bound ligands ( Figures 1 and 2 ) . Thus , the CNBDs of the ligand-bound full-length SthK adopt a different conformation than that observed in the crystal structures of the isolated ligand-bound CNBDs ( Figures 1 and 2 and Figure 6—figure supplement 1 ) . Since the three SthK structures have largely similar features , and alignments between them yield RMSDs that are very small ( apo/cAMP: 0 . 50 Å , apo/cGMP: 0 . 40 Å , cAMP/cGMP: 0 . 46 Å ) , we will focus first on the apo structure , and then discuss the unanticipated effects of ligand binding and their implication for mechanism . Similar to the eukaryotic cyclic nucleotide-modulated ion channels whose structures were solved recently by cryo-EM ( HCN1 , TAX-4 ) ( Lee and MacKinnon , 2017; Li et al . , 2017 ) , the SthK channel is a functional tetramer and the structure can be divided in three layers as viewed from the membrane plane ( Figure 1A ) . The top layer contains the transmembrane domain ( TMD ) with the ion-conducting pore , and a voltage sensor-like four transmembrane helix bundle , which presumably is responsible for voltage modulation of the channel ( Figures 1 and 4 ) . The voltage sensor-like domains are in a non-swapped configuration , similar to HCN1 , TAX-4 , and LliK ( James et al . , 2017 ) , but unlike MloK1 ( Kowal et al . , 2018 ) . The bottom layer , at the intracellular side , consists of four cyclic nucleotide binding domains ( CNBDs ) , which contain the ligand-binding pockets ( Figure 1A ) . Sandwiched in between the top and bottom layers is the C-linker domain , strategically located to relay the ligand binding event from the CNBDs to the channel pore ( Figure 2B ) . The C-linker/CNBD complex is domain-swapped with respect to the transmembrane region ( Figures 1A and 2A ) . Some prokaryotic homologs of cyclic nucleotide-modulated channels with structures solved by cryo-EM completely lack a C-linker ( such as MloK1 ) ( Kowal et al . , 2018; Chiu et al . , 2007 ) , while others have an intact C-linker ( such as SthK and LliK ) ( James et al . , 2017; Kesters et al . , 2015 ) . Out of these prokaryotic channels , only SthK is conducive to being analyzed in detail at the functional level with single-channel recordings . Sequence alignments as well as structure overlays of SthK with HCN1 , TAX4 , and LliK , show that all channels are similar in coarse overall architecture ( Figure 1—figure supplement 6 and Figure 2C ) . SthK is a potassium selective channel ( Brams et al . , 2014 ) , and displays a traditional potassium channel pore architecture ( Zhou et al . , 2001; McCoy and Nimigean , 2012 ) where the selectivity filter , encoded by the TIGYGD signature sequence for potassium selection ( Heginbotham et al . , 1992 , 1994 ) ( Figure 1—figure supplement 6 ) , is the same as that of KcsA and other potassium channels ( Figure 3 ) . The pore widens below the selectivity filter into an aqueous cavity . Roughly half way down the membrane , in the middle of this cavity , the side chains of I215 pointing radially towards the pore axis , constrict the pore to a diameter of the size of a dehydrated potassium ion similar to what is seen in the HCN1 and LliK channels pore ( James et al . , 2017; Lee and MacKinnon , 2017 ) ( Figure 3A–B ) . However , instead of further narrowing into a bundle-crossing , like in HCN1 or KcsA ( Lee and MacKinnon , 2017; Liu et al . , 2001 ) , the SthK pore widens towards the intracellular side . The expectation from the functional data ( Schmidpeter et al . , 2018 ) is that the SthK pore in the cryo-EM structures is in a closed conformation as will be discussed in further detail ( Figure 1E , Figure 4—figure supplement 2 ) . Thus , it is possible that the gate in SthK is located at a previously unreported location , in the middle of the cavity , which is the second narrowest point in the permeation pathway , after the selectivity filter . However , we cannot exclude the possibility that an additional gate exists at the level of the selectivity filter , similar to other ligand-gated ion channels ( Heer et al . , 2017; Posson et al . , 2015; Wilkens and Aldrich , 2006 ) , including eukaryotic CNG channels ( Flynn and Zagotta , 2001 ) . SthK is activated by cAMP binding and this activity is further increased with depolarization ( Schmidpeter et al . , 2018 ) ( Figure 4B and C ) similar to other voltage-gated K+ and Na+ channels ( Bezanilla , 2000 ) , but unlike HCN channels that activate with hyperpolarization ( Santoro et al . , 1998; Wang et al . , 2002 ) . The channel has a voltage sensor-like domain S1-S4 , and the S4 transmembrane helix contains 4 basic residues ( Figure 4A , Figure 4—figure supplement 1 ) . Although we do not have direct evidence that this is a bona fide voltage sensor , the similarities to other known voltage sensor domains and the increase in channel activity with depolarization lead us to hypothesize that the S1-S4 helices in SthK also form a voltage sensor . Two out of the four basic residues in S4 , R111 and K114 , are located within the electric field drop across membrane and are likely to contribute to the gating charge , while R120 and R124 are solvent exposed due to a cavity on the intracellular side of the S4 helix ( Figure 4A ) . The small number of arginine and lysine residues in the S4 helix of SthK is in agreement with the lower voltage dependence of this channel compared with other voltage-gated channels , which have up to 9 positively-charged residues in the membrane ( Bezanilla , 2000; Tao et al . , 2010 ) . L54 on the S2 helix , at the equivalent position of the phenylalanine , the so-called charge transfer center in voltage-gated K+ channels ( Tao et al . , 2010 ) ( Figure 1—figure supplement 6 ) , has its side chain protruding right in between R111 and K114 ( Figure 4A ) , suggesting that L54 is the charge transfer center in SthK . Although most charge transfer centers are either phenylalanines or tyrosines , the voltage-gated KvAP potassium channel also has a leucine at this position , similar with SthK ( Jiang et al . , 2003 ) ( Figure 4—figure supplement 1 ) . Since the channel experiences a membrane potential of 0 mV on the EM grid , and the median value for voltage activation is about 85 mV ( Schmidpeter et al . , 2018 ) , the voltage sensor is likely in the down , resting conformation in the SthK structure ( Schmidpeter et al . , 2018 ) ( Figure 4B–C ) . Furthermore , both R111 and K114 appear strongly anchored in place by multiple attractive interactions within short distances with residues from both S2 and S3 helices ( Figure 4A ) , in agreement with the requirement for strong depolarizations needed to activate the channel . In comparison , the positively-charged residues on the S4 helix of HCN1 ( easier to open with voltage than SthK ) appear less strongly anchored in place with counter charges than those in SthK , while those on TAX-4 ( quasi voltage-insensitive ) appear more strongly anchored ( Figure 1—figure supplement 6 , Figure 4—figure supplement 1 ) . Thus , there is a correlation between the strength of the interactions the voltage-sensitive charges in S4 make with their counter charges in S2 and S3 and the energy necessary to stimulate the channels with voltage . All three SthK structures revealed multiple extraneous elongated densities running along and in direct contact with the transmembrane helices , at interfaces between helices , and in crevices along subunit interfaces ( Figure 5 ) . Since these channels are in lipid nanodiscs , and we see these densities in each of the three structures at the same positions , we assigned them to lipid molecules binding specifically to SthK ( POPG , since the nanodiscs were assembled with this lipid only ) . In all cases , only partial lipid molecules were modeled , since the densities did not support modelling of most headgroups or the entire length of the alkyl chains . The lipid molecules modeled are all annular lipids and most of them are observed to partition in the top leaflet of the bilayer ( Figure 5 ) , suggesting that the top transmembrane part of the channel is strongly anchored in the membrane . The lipids also bind at the interfaces between adjacent subunits and in the crevices between the voltage sensors and pore domains , perhaps acting as a greasy slide during voltage gating , consistent with our previous finding that SthK gating is dependent on the lipid composition of the bilayer ( Schmidpeter et al . , 2018 ) . The dearth of observed lipids on the inner leaflet is consistent with the idea that the inner helices are moving during gating in CNG channels ( Flynn and Zagotta , 2001 ) . We identified only one lipid molecule at the inner leaflet of the bilayer , bound along the length of the S6 helix all the way up to the end of the pore helix/beginning of the selectivity filter ( Figure 5C–D ) . Its unusual position near the moving channel parts that form the gates is also consistent with the role of lipids in SthK gating ( Schmidpeter et al . , 2018 ) . The lipid-SthK interactions are mostly hydrophobic , although multiple serines and threonines were found near the putative locations of the headgroups at the extracellular side suggesting that some specific interactions exist . Similarly located lipids have been identified in the structure of Kv1 . 2 channels crystallized in the presence of lipids ( Long et al . , 2007 ) , and in the cryo-EM structure of other channels in nanodiscs ( Gao et al . , 2016; Autzen et al . , 2018; Jin et al . , 2017; Dang et al . , 2017; Chen et al . , 2017 ) . The C-linker/CNBD domain of SthK is overall highly homologous in both sequence and architecture/structure with the C-linker/CNBD of all other cyclic nucleotide-modulated channels ( Figure 1—figure supplement 6 , Figure 6—figure supplement 1 ) , with an important difference . Previous crystal structures of isolated C-linker/CNBD domains of other CNG/HCN channels , and , importantly , also of SthK itself , allowing direct comparisons , have the most C-terminal helix of the CNBD ( the C-helix ) packed tightly like a lid against the binding pocket and interacting closely with the ligand ( Figure 6D–E , grey and wheat , Figure 6—figure supplement 1 ) ( Lee and MacKinnon , 2017; Li et al . , 2017; Kesters et al . , 2015; Saponaro et al . , 2014 ) . We will call this the ‘activated’ CNBD conformation . In all our full-length SthK structures , whether apo , or ligand-bound , the C-helix is swung away from the binding pocket ( Figure 6D–E green ) , most similar to the only reported apo CNBD conformation of HCN2 ( Goldschen-Ohm et al . , 2016 ) ( Figure 6F ) . We will call this the ‘resting’ CNBD conformation . Furthermore , due to the lower positioning of the C-helix in our full-length SthK structures , the loop that connects the C-linker with the CNBD ( which we previously called the siphon loop [Kowal et al . , 2018] ) also has more room to expand downward further than in the ‘activated’ CNBDs ( Figure 6—figure supplement 1 ) . Interestingly , not only our apo , but also the cAMP- and cGMP-bound full-length SthK structures display the CNBD in a ‘resting’ conformation . The presence of the ligand in the ligand-binding pocket of our full-length SthK structures is shown as experimental maps ( grey mesh in Figure 6C ) as well as difference maps between ligand-bound and apo density maps ( blue mesh Figure 6C ) . The entire ligand density is recovered in the difference maps indicating that there is no density in the binding pocket in the experimental apo map ( compare blue and grey mesh in Figure 6C ) . In order to investigate whether each individual CNBD within a tetramer is actually occupied by ligand , we also performed 3D refinement without imposing C4 symmetry on the ligand-bound particles ( C1 , unsymmetrized maps for apo , cAMP-bound and cGMP-bound SthK are shown in Figure 7A–C ) . Although at a lower resolution ( 3 . 6 Å , see Supplementary file 1 ) , the final refined structures were nearly identical to the fourfold symmetrized structures we present here , with minor differences between the subunits within the tetramer , suggesting that all subunits within the tetramer are the same . The ligand is bound with an occupancy close to 1 in each of the 4 subunits of the tetramer for both cAMP ( Figure 7D ) and cGMP ( Figure 7E ) , suggesting that all four subunits are ligand-bound . We modeled cAMP to bind to the CNBD in an anti-configuration ( Figure 6A ) , although the density can also accommodate the syn , albeit not as well ( Figure 6—figure supplement 2 ) . On the other hand , cGMP can be modeled much better in the syn configuration than in anti in the experimental map , likely due to the favorable interaction between the hydroxyl group of T378 and the 2-amino group of the guanosine ( Figure 6B–C , Figure 6—figure supplement 1 ) , which cannot occur if cGMP is in anti ( Kesters et al . , 2015 ) . The equivalent threonine in other CNG channels has been previously shown to play a role in the cGMP binding ( Altenhofen et al . , 1991 ) . The other interactions between the ligands and the binding pocket are conserved ( Figure 6A–B ) , and have been described before in detail ( Kesters et al . , 2015 ) . Given its accommodating density , the lack of interaction between the base of the cAMP molecule and the protein binding pocket , as well as the 70–30% solution distribution of the anti-syn conformers of cAMP , respectively ( Yathindra and Sundaralingam , 1974 ) , it makes sense that both conformers would bind to the protein , likely in a similar distribution as seen in solution . For cGMP on the other hand , the fact that the experimental map accommodates the syn conformer much better , only the syn conformer can make a favorable interaction with the protein , and the solution distribution of the syn-anti cGMP is 95–5% , respectively ( Yathindra and Sundaralingam , 1974 ) , strongly indicates that cGMP binds in syn to SthK . The somewhat featureless aspect of the experimental cAMP map despite its higher resolution ( cAMP-bound SthK is 3 . 3 Å , while cGMP-bound SthK is 3 . 5 Å resolution , Supplementary file 1 ) also argues towards it representing a mixture of the two cAMP configurations , although we cannot rule out that the features distinguishing syn from anti cAMP are absent due to a low overall definition of the cAMP in the map . As described above , the full-length channel , including the ligand binding pocket , adopts the same resting/closed conformation not only in the apo but also in the cAMP-bound state . We used single-channel analysis to define the functional state of SthK ( open/closed ) in the conditions employed to determine the structure with cryo-EM . In saturating cAMP concentrations , the open probability ( Po ) of SthK is low ( Schmidpeter et al . , 2018 ) ( ~0 . 2–0 . 5 , at +100 mV , Figures 1 and 4 ) , indicating that cAMP is only a partial agonist and not a full agonist for SthK . Additionally , as we showed before ( Figure 4B , C ) , SthK activity is increased by depolarization , and thus in the conditions on the microscope grid ( at 0 mV , where the voltage sensor is not activated yet ) the open probability is even lower ( ~0 . 05–0 . 1 ) and the open intervals are very short ( Schmidpeter et al . , 2018 ) ( Figure 4 , Figure 4—figure supplement 2 ) . Thus , we expect that at most 5–10% of the channels imaged from a grid frozen in saturating cAMP conditions are in the open state . Moreover , despite our efforts towards detecting the open state , 3D classification never produced a class that could correspond to an open state , thus the channels that contribute to the final density map are mostly in a closed state . The potential mixed-configuration binding of the cAMP in the binding pocket may provide an explanation for its partial agonist nature . As discussed above , cAMP may bind to apo SthK in both the syn and the anti configurations ( Figure 6A , C , Figure 6—figure supplement 2 ) . However , the ‘activated’ SthK CNBD ( Kesters et al . , 2015 ) displays cAMP bound in the anti configuration ( Figure 6D ) . Clashes are observed if we artificially dock syn cyclic nucleotides in the binding pocket of the ‘activated’ SthK CNBD suggesting that only anti cAMP binding can lead to activation ( Figure 6D–E , insert , shown for cGMP ) . Thus , only a subset of all cAMP binding events ( those where cAMP is in anti ) can lead to channel opening , likely contributing to the subunitary open probability of SthK in saturating cAMP . On the other hand , in HCN2 channels , where cAMP is an efficacious agonist ( Wainger et al . , 2001; Zagotta et al . , 2003 ) , syn cAMP may not clash with the C-helix of HCN2 channels and may even make favorable interactions due to the different amino acid composition of the HCN2 C-helix ( Figure 1—figure supplement 6 ) . The cGMP-bound SthK structure is closed , in agreement with the role of cGMP as very poor partial agonist ( Po ~0 . 003 , Figure 1E ) , where less than 0 . 3% of the channels on the cryo-EM grid are expected to be in the open state . SthK discriminates strongly between cAMP and cGMP in the degree with which each of these ligands promote channel opening . The weak activation of SthK by cGMP may be explained by the different binding orientations of cGMP between the resting and ‘activated’ CNBDs . As we pointed out above , cGMP can be better accommodated in the syn configuration in the density map ( Figure 6B , C , E , Figure 6—figure supplement 2 ) . This was surprising , given that cGMP was found to bind in an anti configuration in the ‘activated’ CNBD of SthK ( Kesters et al . , 2015 ) ( Figure 6—figure supplement 1 ) . The binding pocket ( Figure 6B ) likely prefers the syn cGMP due to the additional interaction with T378 also mentioned above , which cannot occur when the cGMP is in anti , and also because of the syn-biased distribution of cGMP configurations in solution . On the other hand , syn cGMP cannot be easily accommodated in the ‘activated’ CNBD binding pocket , likely due to a clash with the side chain of L422 on the C-terminal C-helix ( Figure 6E , insert ) . We propose that the reason for the weak partial agonism of cGMP for SthK is that the apo conformation strongly prefers to bind the syn cGMP , which does not allow the CNBD to ‘activate’ . This is in contrast to TAX-4 and the isolated CNBD from HCN2 , where syn cGMP stabilizes the ‘activated’ CNBD via favorable interactions between the base and the C-helix , which are absent in SthK ( Li et al . , 2017; Zagotta et al . , 2003 ) . In these channels , cGMP is able to increase activity to a large degree , so it makes sense that the form of cGMP that binds with higher affinity to the apo conformation is able to support the conformational change . Surprisingly , the binding of either cAMP or cGMP to full-length SthK does not induce the local conformational changes in the binding pockets that have been observed in the crystal structures of ligand-bound isolated SthK CNBDs ( Kesters et al . , 2015 ) ( Figure 6D–E ) . This can occur if strong allosteric coupling exists between the CNBD and the pore in the full-length channel . We propose that when ligand binds to the isolated CNBD , the binding energy is sufficient to activate it and change its conformation , but it is not enough to activate the CNBD when it is part of the full-length channel , suggesting that the CNBD and the pore domain must change conformation together , in a concerted way . The absence of a local conformational change in the binding pocket despite the bound ligand can thus be understood in the context of a concerted Monod-Wyman-Changeux ( Monod et al . , 1965 ) ( MWC ) model , previously proposed to describe the gating of CNG/HCN channels and other multi-subunit ligand-gated ion channels ( Li et al . , 1997; Colquhoun and Sivilotti , 2004; Lape et al . , 2008 ) . This model stipulates that the channel exists in two conformations: closed ( or resting ) , and open ( or activated ) ( Figure 8E , top ) . All closed conformations , whether apo or ligand bound , are identical , differing only in the number of ligands bound ( all squares in Figure 8E , top ) . Each closed channel can undergo a concerted conformational change to an open channel , meaning that all subunits undergo the closed to open transition at the same time . All open conformations are also identical to each other ( all circles in Figure 8E , top ) . This model requires that the difference between partial agonists and full agonists lies only in the efficacy with which the agonist can induce the concerted conformational change ( the vertical transition from all squares to all circles in Figure 8E , top ) . The conformational change is favored when efficacious agonists are bound , so that most bound channels are open . Conversely , the conformational change is less favored when less efficacious agonists are bound , so that most bound channels are closed . Our data shows identical , closed structures for the apo and inefficacious , partial agonist-bound SthK channels , which is consistent with the concerted gating model ( the functional states - closed - that we solved with cryo-EM are indicated by dashed green circles in Figure 8E , top ) . In contrast , identical structures are not expected from ion channels gating according to the sequential ( or Koshland-Nemethy-Filmer , KNF [Koshland et al . , 1966] ) model , which predicts that binding of partial agonists and full agonists will induce specific closed conformational states ( the red and green ligands leading to circles and octagons , respectively in Figure 8E , bottom ) , different than the apo state ( squares ) and perhaps different from each other ( circles versus octagons in Figure 8E ) , leading to channel opening . Further ruling out the sequential model , we found no major differences between the subunits within the tetramer when the structural data was processed without fourfold symmetrization ( Figure 7A–C ) and all ligands were bound to an occupancy close to 1 in each of the subunits within the tetramer ( Figure 7D–E ) . Previous functional investigations employed concerted , sequential , or a combination of both models to describe gating in CNG and HCN channels ( Biskup et al . , 2007; Li et al . , 1997; Ulens and Siegelbaum , 2003 ) . To our knowledge , this study provides the first structural evidence that a concerted step exists in the gating of CNG channels . Conversely , other ligand-gated channels , such as the AMPA receptors , have been proposed to gate according to the sequential model , based on findings that the structures of the ligand binding domains display major differences depending on the type of agonist/antagonist bound ( Armstrong and Gouaux , 2000; Armstrong et al . , 2003; Jin et al . , 2003; Sobolevsky et al . , 2009; Twomey et al . , 2017 ) . Additional evidence frequently invoked to support concerted channel gating and that we also found for SthK are the lack of subconductance states observed at the single-channel level and the equal single-channel amplitudes when the channel opens in response to cAMP or cGMP-binding ( Figure 1E , Figure 4B and Figure 4—figure supplement 2 ) . This suggests that there is only one structurally distinct open state , with the caveats that we are not able to observe states shorter than the dead time of the recording ( ~180 µs ) and that structurally distinct open channel conformations may display the same single-channel current amplitude . Furthermore , in addition to the single ‘resting’ CNBD conformation reported here for SthK , there is also only one ‘activated’ CNBD conformation reported by the crystal structures of the isolated CNBDs of SthK ( where the cAMP-bound is identical to the cGMP-bound conformation ( Kesters et al . , 2015 ) , despite their quite different phenotypes at the functional level ) . In contrast to what we observe with SthK , the CNBDs in the full-length apo HCN1 channel appear to be almost identical to those in the cAMP-bound HCN1 , which are in the same ‘activated’ conformation seen in other CNBDs ( Lee and MacKinnon , 2017 ) ( Figure 4—figure supplement 2 ) . Thus , in HCN1 , cAMP binding does not lead to a drastic conformational change likely because the CNBDs are already ‘pre-activated’ in the absence of ligand . This is in agreement with functional data showing not only that cAMP binding has minimal effect on HCN1 channel activity but also that they are the fastest activating HCN channels , requiring the least amount of hyperpolarization for opening ( Santoro et al . , 1998 ) . We do not have the structure of the full-length open SthK channel . However , based on the near identity structure-wise between the ‘activated’ CNBD structures of SthK ( Kesters et al . , 2015 ) and TAX-4 ( Li et al . , 2017 ) ( Figure 6—figure supplement 1 ) , we will use the full-length structure of TAX-4 , the only open cGMP-bound CNG structure , as a proxy for open SthK . Thus , an overlay of SthK and TAX-4 structures reveal the regions in the channel that likely have to rearrange and how , in order to go from a closed to an open conformation . In TAX-4 , the CNBD/C-linker domains are closer to the membrane , and are rotated outwards , ~9 degrees counterclockwise viewed from the extracellular side compared to SthK apo ( Figure 8C–D ) . The B- and C-helices are shifted together as a rigid body to bring the C-helix in close contact with the ligand , and consequently the siphon is also displaced vertically towards the membrane to make room ( Labeled as 1 and 2 in Figure 8A , Video 1 ) . The C-linker from the adjacent subunit ( labeled as 3 in Figure 8A ) appears to have undergone a translation and rotation to accommodate for the displaced siphon , and since it is directly connected to the end of the S6 helices , they are consequently widened at the intracellular side . The resulting enlargement of the intracellular entryway likely has to be propagated to the selectivity filter to open the channel ( Figure 8A , C–D ) . Movements of the C-linker can also be transmitted to the S4-S5 linker of the adjacent subunit due to direct interactions between these regions ( dashed circle Figure 8A , B ) . Displacements of the S5 transmembrane helix can be directly sensed by the pore helix and selectivity filter , a putative gate for these channels ( Flynn and Zagotta , 2001 ) , and bias its opening ( Mazzolini et al . , 2018 ) . Additionally , this direct interaction between the S4-S5 linker and the CNBD also provides a direct link between cAMP binding and voltage sensor moving and may explain channel modulation by voltage . We report here the cryo-EM structures of apo , cAMP-bound and cGMP-bound SthK potassium channels in lipid nanodiscs . The conformations are identical , and functionally correspond to resting/closed states as indicated by single-channel recordings of SthK . Identical apo , and partial agonist-bound conformations indicate that SthK gating can be described with a concerted model , with strong coupling between the ligand-binding and pore domains . We hypothesize that the strong discrimination against cGMP in SthK is based on its preferred binding configuration to the apo conformation which differs from the configuration that leads to channel activation . The C-terminally truncated gene for SthK 1–420 ( UniProtKB G0GA88 ) was cloned into pCGFP-BC using restriction sites HindIII and Xhol ( Kawate and Gouaux , 2006 ) , which adds 19 additional amino acids from the multiple cloning site at the N-terminus as described previously ( Schmidpeter et al . , 2018 ) . These 19 amino acids were not removed because they helped protein expression . The GFP and four out of the eight histidines were removed by Quikchange mutagenesis keeping the thrombin cleavage site . The construct with four histidines instead of eight displayed increased expression and solubility . SthK was expressed in E . coli C41 ( DE3 ) cells ( Lucigen ) . Cells were grown in 4 L of LB media supplemented with 100 mg/L ampicillin at 37°C until reaching OD600nm=0 . 4 and then transferred to 20°C . Protein expression was induced at OD600nm=0 . 8 with 0 . 5 mM IPTG for 12 hr . The cells were harvested by centrifugation ( 5660 x g , 15 min , 4°C ) and resuspended in 50 mL ice-cold lysis buffer ( 20 mM HEPES , pH 8 . 0 , 100 mM KCl ) supplemented with PMSF ( 85 μg/mL ) , Leupeptine/Pepstatin ( 0 . 95/1 . 4 μg/mL ) , 1 mg of DNaseI ( Sigma ) , 1 mg of Lysozyme ( Sigma ) , and one cOmplete ULTRA mini Protease Inhibitor ( Roche ) tablet . All following steps were performed at 4°C unless otherwise stated . The cells were lysed by sonication ( Sonic Dismembrator 500 , Fisher scientific ) and solubilized with 30 mM n-Dodecyl-β-D-Maltopyranoside ( DDM , Anatrace ) for 1 . 5 hr . Non-solubilized material was removed by centrifugation ( 36 , 500 x g , 45 min , 4°C ) . The supernatant was supplemented with 40 mM imidazole , filtered through a 0 . 22 µm filter and applied to a 5 mL HiTrap chelating HP Co2+ column ( GE Lifesciences ) equilibrated in wash buffer ( 20 mM HEPES , pH 8 . 0 , 100 mM KCl , 40 mM imidazole , 200 µM cAMP , 1 mM DDM ) . The column was washed with 10–15 column volumes of wash buffer until a stable baseline was reached before eluting the protein with elution buffer ( wash buffer supplemented with 250 mM imidazole ) . The protein was concentrated to 10 mg/mL using a 100 kDa cut-off concentrator ( Amicon Ultra , Millipore ) and applied to a Superdex 200 10/300 GL column ( GE Lifesciences ) equilibrated in running buffer ( 20 mM HEPES , pH 8 . 0 , 100 mM KCl , 200 µM cAMP , 0 . 3 mM DDM ) at room temperature . The peak fraction at ~11 mL containing SthK tetramer was collected and concentrated to 10–12 mg/mL ( 200–230 µM monomer ) using a 100 kDa cut-off concentrator ( Amicon Ultra , Millipore ) . The final protein concentration was determined using the molar extinction coefficient at 280 nm: 55 , 900 M−1cm−1 ( ProtParam [Wilkins et al . , 1999] ) . The plasmid of the membrane scaffold protein 1E3 ( MSP1E3 ) was obtained from Addgene ( #20064 ) . Expression and purification of MSP1E3 was carried out as previously described , with some modifications ( Ritchie et al . , 2009 ) : BL21 ( DE3 ) cells were used for protein expression at 37°C in 4 L LB media supplemented with 50 mg/L kanamycin . Expression was induced at OD600nm=0 . 8 by 1 mM IPTG for 3 hr . Cells were harvested by centrifugation ( 5660 x g , 15 min , 4°C ) and resuspended in 50 mL ice cold lysis buffer ( 40 mM Tris/HCl ( pHRT 8 . 0 ) , 300 mM NaCl , 1% Triton X-100 ) supplemented with PMSF ( 85 μg/mL ) , Leupeptine/Pepstatin ( 0 . 95/1 . 4 μg/mL ) , 1 mg of DNaseI ( Sigma ) , 1 mg of Lysozyme ( Sigma ) , and one cOmplete ULTRA mini Protease Inhibitor ( Roche ) tablet . All following steps were carried out at 4°C . The cells were lysed by sonication ( Sonic Dismembrator 500 , Fisher scientific ) and cell debris was removed by centrifugation ( 36 , 500 x g , 45 min , 4°C ) . The clarified lysate was incubated with 3 mL Ni-NTA resin ( Millipore ) ( equilibrated in 40 mM Tris/HCl ( pHRT 8 . 0 ) , 300 mM NaCl , 1% Triton X-100 ) for 45 min . The resin was transferred into an EasyPack column ( Bio-rad ) and washed with 10 column volumes for each following buffers: wash buffer I ( 40 mM Tris/HCl ( pHRT 8 . 0 ) , 300 mM NaCl , 1% Triton X-100 ) , wash buffer II ( 40 mM Tris/HCl ( pHRT 8 . 0 ) , 300 mM NaCl , 20 mM imidazole , 50 mM sodium cholate ) , wash buffer III ( 40 mM Tris/HCl ( pHRT 8 . 0 ) , 300 mM NaCl , 50 mM imidazole ) . The protein was eluted with wash buffer III containing 400 mM imidazole , concentrated using a 30 kDa cut-off concentrator ( Amicon Ultra , Millipore ) and applied to a PD10 desalting column ( GE Lifesciences ) equilibrated with 50 mM Tris/HCl ( pHRT 8 . 0 ) , 150 mM KCl , 0 . 5 mM EDTA for buffer exchange . The final MSP1E3 sample was concentrated to ~250 µM and stored at −80°C . For the lipid stock , 1-palmitoyl-2-oleoyl-sn-glycero-3-phospho- ( 1'-rac-glycerol ) ( sodium salt ) ( POPG , Avanti ) in chloroform was dried under constant nitrogen stream forming a thin lipid film . Residual chloroform was further removed under vacuum for 3 hr . The lipid film was dissolved in 20 mM HEPES , pH 8 . 0 , 100 mM KCl , 33 mM CHAPS by sonication at a final concentration of 25 mg/mL ( 32 . 4 mM ) POPG and stored at −80°C . Purified SthK was mixed with MSP1E3 and POPG prepared as above , in a molar ratio 1:1:70 ( SthK ( monomer ) :MSP1E3:POPG ) , supplemented with 1 mM cAMP and incubated for 1 hr at 4°C . Reconstitution was started by detergent removal using bio-beads SM-2 ( Bio Rad , 20 mg per 100 µl mixture ) . After 3 hr of gentle shaking at 4°C , bio-beads were removed and the same amount of bio-beads was added for an additional 12 hr of continuous shaking at 4°C . The supernatant was collected , filtered through a 0 . 22 µm Spin-X centrifugation tube filter ( Costar ) and applied to a Superose 6 Increase ( GE Lifesciences ) equilibrated in running buffer ( 20 mM HEPES , pH 8 . 0 , 100 mM KCl , 200 µM cAMP ) at room temperature . The peak fraction corresponding to SthK in lipid nanodiscs was collected and concentrated to 5–5 . 5 mg/mL ( approximated as 1 Absorption unit = 1 mg/mL ) using a 10 kDa cut-off concentrator ( Amicon Ultra , Millipore ) . For the preparation of SthK-apo and SthK-cGMP the size exclusion chromatography was equilibrated in cAMP-free running buffer . After concentrating , the sample was dialyzed against the cAMP-free running buffer for 5 hr at 4°C . A negative stain protocol was published before ( Ohi et al . , 2004 ) and was used with minor modifications: 3 μL of purified SthK in lipid nanodics ( 20 μg/mL ) was incubated for 60 s on a glow discharged homemade carbon coated copper grid ( Electron Microscopy Sciences , 400 mesh ) . The grid was washed on four drops ddH2O ( 50 μL ) and stained twice using a 1 . 5% uranyl acetate solution ( 50 μL ) . Excess liquid was removed by blotting after each step . For the second staining step the grid was incubated for 20 s in staining solution before blotting . Finally , the grid was air-dried for 5 min . The prepared grids were loaded in a Tecnai T12 ( FEI ) electron microscope operated at 120 kV acceleration voltage equipped with a F416 ( TVIPS ) camera ( 2048 × 2048 pixel ) . Micrographs were taken at 49000x magnification with a pixel size of 3 . 44 Å/pixel using Leginon ( Suloway et al . , 2005 ) . Reference-free particle picking was done using DoG Picker ( Voss et al . , 2009 ) and extracted particles were 2D classified with Xmipp ( Scheres et al . , 2005a; Scheres et al . , 2005b ) both included in the Appion ( Lander et al . , 2009; Voss et al . , 2010 ) processing package . For cryo-EM grid preparation , 2 mM cAMP and 7 mM cGMP were added to purified nanodiscs of SthK-cAMP and SthK-apo , respectively , to obtain SthK-cAMP and SthK-cGMP , and incubated for 30 min before freezing . Immediately prior to freezing , all samples were spiked with 3 mM fluorinated Fos-choline 8 ( Anatrace ) . 3 . 5 µL of SthK in lipids nanodiscs were applied to glow-discharged UltrAuFoil R1 . 2/1 . 3 300-mesh gold grids ( Quantifoil ) . The sample was incubated for 10 s and then plunge frozen in liquid ethane using a Vitrobot Mark IV ( FEI ) with a blot force of 0 and a blot time of 2 s , at 22°C and 100% humidity . Automated data collection was done in Leginon . The SthK-cAMP datasets were collected on a 300 keV Titan Krios ( FEI ) equipped with a K2 summit direct electron detector ( Gatan ) at a calibrated pixel size of 1 . 07325 A/pixel and a nominal defocus of −1 . 0 to −2 . 2 µm . The total dose of 70 e-/Å2 was distributed over 50 frames ( 1 . 4 e-/Å2/frame ) at a total exposure time of 10 s ( 200 ms/frame ) . The SthK-apo and SthK-cGMP datasets were collected on a different Titan Krios equipped with both a Quantum GIF and a K2 detector ( Gatan , slit width 20 eV ) and a Cs corrector at a calibrated pixel size of 1 . 0961 A/pixel and a nominal defocus of −1 . 0 to −2 . 2 µm . A total dose of 52 e-/Å2 was accumulated over 40 frames ( 1 . 3 e-/Å2/frame ) and a total exposure time of 8 s ( 200 ms/frame ) . The frame stacks were motion corrected using MotionCor2 ( Zheng et al . , 2017 ) . CTFFIND4 ( Rohou and Grigorieff , 2015 ) was used to determine the contrast transfer functions ( CTF ) on non-dose weighted summed images . All following image processing steps were carried out using the dose weighed summed images . Dogpicker ( Voss et al . , 2009 ) as part of the Appion processing package ( Lander et al . , 2009 ) was used for reference-free particle picking . All processing steps were done with RELION 2 . 1 ( Scheres , 2012a , 2012b; Kimanius et al . , 2016 ) unless indicated otherwise . Three datasets for SthK-cAMP containing 5473 micrographs ( 823 , 2210 , 2440 for each dataset , respectively ) were first separately subjected to two rounds of 2D classification using two-times binned particles . 572844 particles from all three datasets were combined after the 2D classification and further classified into 20 classes in 3D without applying symmetry using a 40 Å low-pass filtered map of HCN1 generated in Xmipp ( Sorzano et al . , 2004 ) as an initial reference ( PDB: 5U6O ) . 233053 particles from four 3D classes showing a tetrameric channel arrangement were re-extracted , unbinned , and subjected to 3D Refinement applying C4 symmetry resulting in a 3 . 43 Å resolution map . From this point on , all processing steps were done with C4 symmetry applied . Further improvement of the resolution was achieved by focused classification using a mask excluding the nanodisc . The highest resolution class out of ten with most detailed features containing 81501 particles was subjected to a final 3D refinement . After conversion , the refinement was restarted applying a nanodisc-excluding mask . The resolution of refined maps was assessed by RELION postprocessing and the gold standard FSC value 0 . 143 using a mask that excluded the nanodisc . This resulted in a final resolution of 3 . 35 Å ( Figure 1—figure supplement 3 , EMDB-7484 ) . In addition , the final particle set was subjected to the same refinement approach but without applying symmetry . The final resolution of the C1 refinement was 3 . 66 Å ( Figure 7B ) . A similar processing approach was used for SthK-cGMP combining two sets of data containing 2744 curated micrographs ( 1404 , 1340 , respectively ) resulting in combined 199965 particles after individual 2D classification . 145635 selected particles after 3D classification ( five out of ten classes ) were refined to 3 . 60 Å . After focused classification 91800 particles ( two out of five classes ) were selected for 3D refinement with application of a mask excluding the nanodisc yielding a final map at 3 . 46 Å resolution with C4 symmetry ( Figure 1—figure supplement 4 , EMDB-7483 ) and 3 . 84 Å resolution without symmetry ( Figure 7C ) . SthK-apo structure was obtained from a single data collection session using the procedure described above . From 2130 micrographs 179504 particles were selected after 2D classification and subjected to 3D classification . The selected 121101 particles from the 3D classification ( two out of five classes ) resulted in a 3 . 79 Å resolution refined map . Focused classification resulted in 51115 particles ( one out of five classes ) refining to final resolution of 3 . 42 Å in C4 ( Figure 1—figure supplement 2 , EMDB-7482 ) and 3 . 90 Å in C1 ( Figure 7A ) . Local resolution distributions were calculated with blocres ( Cardone et al . , 2013 ) using the two half maps and the mask applied in postprocessing . For atomic model building a semi-de novo approach was used . One subunit of SthK-CNBD-cAMP crystal structure ( PDB 4D7T ) was docked into the density map of SthK-cAMP using UCSF Chimera ( Pettersen et al . , 2004 ) . After a first round of refinement using phenix . real_space_refine ( Afonine et al . , 2013 ) miss-aligned regions were manually adjusted in COOT ( Emsley et al . , 2010 ) . The transmembrane domain of one SthK subunit was then built de novo in COOT using bulky side chains ( Phe , Trp , Arg ) to establish the register of the alpha helices . The finished model of one SthK-cAMP subunit was subjected to real space refinement in PHENIX ( Afonine et al . , 2013 ) . The tetramer was obtained by placing monomers into all four subunits ( UCSF Chimera ) . The new model was refined in real space applying non-crystallographic symmetry and secondary structure restraints ( PHENIX [Adams et al . , 2010] ) . The atomic model of an individual SthK-cAMP subunit was used as initial model for both the SthK-cGMP and the SthK-apo structures and refined against the respective density maps , following the same process . Outliers were fixed manually , and specific regions in the protein that display very low resolution and lack of definition ( the S2-S3 loop , and the last ~2 helical turns of the C-terminal helix ) were removed using COOT in all final models . The side chains of several residues were not 100% defined , but since they were within a recognizable stretch with a clear secondary structure and register , we decided to model them as the amino acid indicated in the sequence . The models were analyzed with Molprobity ( Chen et al . , 2010 ) and EMRinger ( Barad et al . , 2015 ) for validation . For further cross validation and to check for overfitting , all atoms of each model were randomly displaced by 0 . 5 Å and each resulting model was refined against the first half-map obtained from processing . We then calculated the FSC between the refined models and the half-maps used during refinement and compared them with the FSC between the refined models and the other half-maps , not used during refinement . In addition , the FSC between the refined model and sum of both half maps was calculated . The nanodisc area in all maps was excluded by applying the same mask used in RELION postprocessing to prevent non-meaningful correlation at low resolution . Correlations at higher resolution were not influenced by the mask . The resulting FSC curves were similar showing no evidence of overfitting . The final models contained amino acids 10–62 and 76–412 ( SthK-cAMP , PDB 6CJU ) , 10–68 and 76–412 ( SthK-cGMP , PDB 6CJT ) , and 10–65 and 80–415 ( SthK-apo , PDB 6CJQ ) . HOLE ( Smart et al . , 1996 ) was used to measure the pore size and water accessible cavities in the voltage sensor domain . ‘Omit’ maps between experimental data and atomic models were calculated using phenix . real_space_diff_map ( Afonine , 2017 ) ( PHENIX ) . Difference maps between two experimental maps were obtained using the program diffmap from the Grigorieff laboratory ( http://grigoriefflab . janelia . org/diffmap ) . Figures were prepared in Chimera ( Pettersen et al . , 2004 ) , VMD ( Humphrey et al . , 1996 ) and PyMOL Molecular Graphics System , Version1 . 8 Schrodinger , LLC . Chloroform-solubilized lipids: 1 , 2-dioleoyl-sn-glycero-3-phosphocholine ( DOPC ) , POPG and 18:1 Cardiolipin were mixed in a glass tube at a ratio 5:3:2 ( w:w:w ) , respectively . Lipids were dried under nitrogen , washed with pentane , and redried . The thin lipid film was solubilized immediately by sonication in reconstitution buffer ( 10 mM HEPES , pH 7 . 6 , 400 mM KCl , 5 mM NMDG ) containing 33 mM CHAPS , at a concentration of 10 mg lipids/mL . Solubilized lipids were mixed with freshly purified cAMP-bound SthK ( 30 µg protein/mg lipid ) . Detergent was removed by applying 0 . 5 mL of the mixture to an 18 mL self-packed , degassed , and pre-swollen in reconstitution buffer Sephadex G-50 ( GE Lifesciences ) column . Liposomal aliquots ( eluting at about 7 mL ) are collected in 0 . 5 mL fractions , visually inspected for turbidity and frozen and stored in small aliquots at −80°C . Electrophysiological experiments were performed using a horizontal lipid bilayer setup as previously described ( Heginbotham et al . , 1999; Posson et al . , 2013 ) . Briefly , an artificial lipid bilayer was formed over a ~100 µm diameter hole in the partition separating the cis ( top ) and trans ( bottom ) chamber of the recording system by painting a 6 . 25 mg/mL solution of 1 , 2-dioleoyl-sn-glycero-3-phosphocholine ( DPhPC ) in n-decane ( Sigma ) . The solutions in both chambers contain 97 mM KCl , 3 mM KOH , 10 mM Hepes , pH 7 . 0 . Agar bridges in the two chambers connect via Ag/AgCl pellet electrodes to an Axopatch 200A/B amplifier ( Molecular Devices ) . All electrophysiological signals are filtered at 1000 Hz with an 8-pole low-pass Bessel filter , digitized at 20 KHz with Digidata 1440A ( Molecular Devices ) and collected using Clampex software ( V10 , Molecular Devices ) . Bilayer formation was monitored electronically using an in-house designed protocol in Clampex . Freshly-thawed and briefly sonicated SthK-containing liposomes were applied to the cis-side of the bilayer , and fusion events were monitored in the Gap-Free recording mode in Clampex under +100 mV applied voltage . The channel orientation was determined by having cAMP only on one side of the bilayer , since SthK displays no activity in the absence of cAMP . The system was set up so that only channels with the CNBDs facing the trans chamber were recorded . The trans chamber was serially perfused to different conditions to determine the channel response to cyclic nucleotides . The single-channel data was analyzed in Clampfit ( Molecular Devices ) with no additional filter applied unless stated otherwise . The dwell-time analysis was performed using the Single-Channel Search module in Clampfit . The dwell-time histograms include only events longer than 360 µs ( two times the dead time ) . The closed dwell-time histograms additionally do not contain the very rare events longer than 500 ms . All plots and fits were made in Clampfit and then imported into Adobe Illustrator for final figure assembly . The single-channel measurements were repeated for at least 7 separate bilayers for each condition reported in Figure 1 and Figure 4 and yielded similar results . The all-amplitude histograms and dwell-time distributions for all the single-channel data obtained are qualitatively similar with the representative ones shown in Figure 4—figure supplement 2 .	Ion channels are essential for transmitting signals in the nervous system and brain . One large group of ion channels includes members that are activated by cyclic nucleotides , small molecules used to transmit signals within cells . These cyclic nucleotide-gated channels play an important role in regulating our ability to see and smell . The activity of these ion channels has been studied for years , but scientists have only recently been able to look into their structure . Since structural biology methods require purified , well-behaved proteins , the members of this ion channel family selected for structural studies do not necessarily match those whose activity has been well established . There is a need for a good model that would allow both the structure and activity of a cyclic nucleotide-gated ion channel to be characterized . The cyclic nucleotide-gated ion channel , SthK , from bacteria called Spirochaeta thermophila , was identified as such model because both its activity and its structure are accessible . Rheinberger et al . have used cryo electron microscopy to solve several high-resolution structures of SthK channels . In two of the structures , SthK was bound to either one of two types of activating cyclic nucleotides – cAMP or cGMP – and in another structure , no cyclic nucleotides were bound . Separately recording the activity of individual channels allowed the activity states likely to be represented by these structures to be identified . Combining the results of the experiments revealed no activity from channels in an unbound state , low levels of activity for channels bound to cGMP , and moderate activity for channels bound to cAMP . Rheinberger et al . show that the channel , under the conditions experienced in cryo electron microscopy , is closed in all of the states studied . Unexpectedly , the binding of cyclic nucleotides produced no structural change even in the cyclic nucleotide-binding pocket of the channel , a region that was previously observed to undergo such changes when this region alone was crystallized . Rheinberger et al . deduce from this that the four subunits that make up the channel likely undergo the conformational change towards an open state all at once , rather than one by one . The structures and the basic functional characterization of SthK channels provide a strong starting point for future research into determining the entire opening and closing cycle for a cyclic nucleotide-gated channel . Human equivalents of the channel are likely to work in similar ways . The results presented by Rheinberger et al . could therefore be built upon to help address diseases that result from deficiencies in cyclic nucleotide-gated channels , such as loss of vision due to retinal degradation ( retinitis pigmentosa or progressive cone dystrophy ) and achromatopsia .	[ "Abstract", "Introduction", "Results", "and", "discussion", "Materials", "and", "methods" ]	[ "structural", "biology", "and", "molecular", "biophysics" ]	2018	Ligand discrimination and gating in cyclic nucleotide-gated ion channels from apo and partial agonist-bound cryo-EM structures
Embryonic stem cells co-express Oct4 and Oct1 , a related protein with similar DNA-binding specificity . To study the role of Oct1 in ESC pluripotency and transcriptional control , we constructed germline and inducible-conditional Oct1-deficient ESC lines . ESCs lacking Oct1 show normal appearance , self-renewal and growth but manifest defects upon differentiation . They fail to form beating cardiomyocytes , generate neurons poorly , form small , poorly differentiated teratomas , and cannot generate chimeric mice . Upon RA-mediated differentiation , Oct1-deficient cells induce lineage-appropriate developmentally poised genes poorly while lineage-inappropriate genes , including extra-embryonic genes , are aberrantly expressed . In ESCs , Oct1 co-occupies a specific set of targets with Oct4 , but does not occupy differentially expressed developmental targets . Instead , Oct1 occupies these targets as cells differentiate and Oct4 declines . These results identify a dynamic interplay between Oct1 and Oct4 , in particular during the critical window immediately after loss of pluripotency when cells make the earliest developmental fate decisions . The mammalian blastocyst inner cell mass ( ICM ) contains undifferentiated , pluripotent cells capable of generating all tissue lineages of the embryo proper . Cultured embryonic stem cells ( ESCs ) are derived from these cells and have similar capabilities ( Abranches et al . , 2009 ) . The POU transcription factor Oct4/Pou5f1 is an indispensable component of the regulatory circuitry underlying these properties ( Morey et al . , 2015 ) . It is expressed in the ICM and in ESCs where its loss accompanies differentiation ( Nichols et al . , 1998 ) . Oct4 is also widely used to generate induced pluripotent stem cells ( iPSCs ) from somatic cells ( Takahashi and Yamanaka , 2006 ) . Together with other factors , Oct4 sustains pluripotency by activating ‘core’ targets such as Pou5f1 ( encoding Oct4 itself ) and Nanog ( Boyer et al . , 2005 ) . It also maintains ‘poised’ targets , including developmentally critical transcription regulators , in a silent but readily inducible state ( Bernstein et al . , 2006; Meissner et al . , 2008 ) . These genes frequently encode developmentally important transcription factors and are marked with a bivalent chromatin signature defined by the simultaneous presence of H3K4me3 and H3K27me3 ( Azuara et al . , 2006; Bernstein et al . , 2006; Ku et al . , 2008; Pan et al . , 2007 ) . Oct1/Pou2f1 is a widely expressed protein related to Oct4 . The two proteins have similar DNA-binding specificity ( Tantin , 2013 ) . In somatic cells , it regulates stem cell and immune memory phenotypes ( Maddox et al . , 2012; Shakya et al . , 2015b ) and is associated with cytotoxic stress resistance , glycolytic metabolism and malignant transformation ( Bellance et al . , 2012; Shakya et al . , 2009; Tantin et al . , 2005 ) . Oct1 amplification and/or overexpression correlates with tumor aggressiveness in esophageal , gastric , prostate , lung , cervical , and colorectal cancer ( Vázquez-Arreguín and Tantin , 2016 ) . It is also co-expressed with Oct4 in ESCs ( Okamoto et al . , 1990; Rosner et al . , 1990 ) . Oct1-deficient mice undergo implantation but show defects following gastrulation , most prominently in extra-embryonic tissues , where trophoblast stem cell development is arrested and expression of the direct Oct1 target Cdx2 is defective ( Sebastiano et al . , 2010 ) . Tetraploid complementation bypasses this developmental restriction , allowing embryos to survive to E8 . 5–9 . 5 where they die from an embryo-intrinsic block . These embryos are runted , developmentally arrested , and lack beating hearts . ( Sebastiano et al . , 2010 ) . A slightly less severe germline allele dies in mid-gestation and manifests runting , anemia , hemorrhaging , and other defects with variable penetrance ( Wang et al . , 2004 ) . Here , we show that ESCs lacking Oct1 have no discernable defects when maintained in an undifferentiated state , but that silent , normally poised developmental-specific genes fail to induce properly upon differentiation . Additionally , genes specific for alternative developmental lineages are inappropriately expressed . Most prominently , placenta-specific genes not normally expressed in any ESC-derived lineage are induced , indicating that Oct1 restricts extra-embryonic gene expression in differentiating ESCs . Additionally , these cells show phenotypic defects when differentiated into multiple lineages , form smaller and less differentiated teratomas , and fail to generate chimerism when injected into blastocysts . ChIPseq identifies a group of targets co-bound by Oct1 and Oct4 in ESCs associated with non-classical binding sites termed MOREs ( More Palindromic Octamer Related Elements , ATGCATATGCAT ) . These sites are inducibly bound by Oct1 in somatic cells lacking Oct4 . The function of Oct1 at these genes is to insulate their expression against repression by oxidative stress , and consistently Oct1-deficient ESCs are hypersensitive to oxidative stress . Oct1 associates with developmentally poised targets upon differentiation and Oct4 loss , explaining the altered gene expression observed with RNAseq . These results establish Oct1 as a key mediator of both developmental-specific gene induction and repression , and identify a dynamic interplay in which Oct1 replaces Oct4 at target genes as ESCs differentiate and early decisions about induction or repression of lineage-specific genes are made . We derived Oct1-deficient ESC lines by intercrossing Pou2f1 germline heterozygotes ( Wang et al . , 2004 ) . Oct1-deficient animals die in utero ( Sebastiano et al . , 2010; Wang et al . , 2004 ) , but survive long enough to derive ESCs . Two Oct1-deficient lines and two littermate WT controls were generated . All had normal karyotypes ( not shown ) . Oct1-deficient ESCs proliferate at normal rates ( not shown ) , are morphologically normal ( Figure 1A ) and can be propagated for a month in culture with no loss of ESC morphology ( not shown ) . They express normal levels of Oct4 , Sox2 , and Nanog protein but no Oct1 ( Figure 1B ) . In addition , cells express the pluripotency-associated Pou5f1 ( Oct4 ) , Sox2 , Nanog , and Dppa4 mRNAs at normal levels ( Figure 1C ) . Ahcy , a stress-inducible Oct1 target in which the function of Oct1 is to prevent stress-associated repression ( Kang et al . , 2009; Shakya et al . , 2011 ) , was also unaltered . 10 . 7554/eLife . 20937 . 003Figure 1 . Abnormal developmental gene induction in ESCs lacking Oct1 . ( A ) Phase microscopy images of four ESC lines ( two Oct1 deficient , two WT littermate controls ) derived from Pou2f1-/+ intercrosses . Passage 5 ESCs on feeder fibroblasts are shown . ( B ) Immunoblot comparing lysates of a WT control line and littermate Oct1-deficient line . GAPDH is shown as a loading control . ( C ) mRNA expression of six genes in WT control and littermate Oct1-deficient ESC lines . Data were obtained by RT-qPCR using three biological replicates of a single line of each genotype . Error bars denote standard deviations . p-values: NS=non-significant , * < 0 . 05 , < 0 . 01 , * < 0 . 001 . ( D ) Phase microscopy images of 4-day EBs derived from ESCs ±Oct1 . Three representative images of each genotype from wells of a 96-well plate are shown . ( E ) EBs were collected at 4 , 9 , and 14 days , and cDNA was prepared and subjected to RT-qPCR . Expression levels were normalized to GAPDH . Pluripotency genes ( Pou5f1 , Sox2 ) and Pou2f1 were tested . Three biological replicates were performed . Error bars denote ±standard deviation . ( F ) Additional genes representative of all three germ layers , Sox17 , T , and Fgf5 , were tested as in E . ( G ) Three known poised Oct4 target genes , Hoxa5 , Hoxc6 , and Gata2 , were tested as in E . DOI: http://dx . doi . org/10 . 7554/eLife . 20937 . 00310 . 7554/eLife . 20937 . 004Figure 1—figure supplement 1 . Abnormal morphology in differentiating Oct1-deficient cells manifests by day 5 of EB formation . ( A ) Phase microscopy images of WT and Oct1-deficient ESCs aggregating into EBs . Top: 1 day of culture . Bottom: day 2 . ( B ) Day 5 . EBs are in low attachment dishes . DOI: http://dx . doi . org/10 . 7554/eLife . 20937 . 004 To study differentiation , we used early-passage Oct1-deficient and WT control ESCs to form embryoid bodies ( EBs ) . Oct1-deficient ESCs were able to aggregate into EBs at d four with morphology similar to WT ( Figure 1D ) . Similar results were obtained at days 1 and 2 ( Figure 1—figure supplement 1A ) . During EB formation , Pou5f1 and Sox2 were down-modulated with similar kinetics in Oct1-deficient and WT cells , while Pou2f1 ( Oct1 ) remained undetectable ( Figure 1E ) . Sox17 ( endoderm ) , Brachyury ( T , definitive mesoderm ) , and Fgf5 ( definitive ectoderm ) expression in Oct1-deficient EBs was grossly similar to WT at some ( days 4 , 9 , or 14 ) timepoints ( Figure 1F ) , consistent with findings that Oct1 is dispensable for gastrulation ( Sebastiano et al . , 2010; Wang et al . , 2004 ) . However , there were consistent defects in expression in the Oct1-deficient condition at day 14 for Sox17 and day 4 for T and Fgf5 . Sox17 , T and Fgf5 are known Oct4 targets ( Chen et al . , 2008 ) . By day 5 , Oct1-deficient EBs were somewhat smaller in appearance ( Figure 1—figure supplement 1B ) . We therefore looked for further evidence of defects in induction kinetics in three other known silent but developmentally inducible Oct4 target genes: Hoxa5 , Hoxc6 , and Gata2 ( Chen et al . , 2008 ) . Each of these genes showed a similar pattern of defective induction in Oct1-deficient EBs relative to WT controls ( Figure 1G ) . To study gene induction using a more developmentally restricted system , we analyzed expression of known developmentally inducible Oct4 target genes during RA-mediated differentiation of WT and Oct1-deficient ESCs . RA treatment of ESCs ultimately results in a largely neuronal phenotype , but waves of gene expression , differentiation , proliferation , and cell death take place during the course of RA treatment ( Walker et al . , 2007 ) . Upon differentiation , ESCs ±Oct1 lose their clustered , spherical , refractile morphology with similar kinetics ( not shown ) . Pou5f1 and Sox2 were also lost with similar kinetics ±Oct1 , while Pou2f1 was not detectable in KO cells ( Figure 2A ) . To study developmental gene expression , we tested Hoxa5 , Hoxc6 , Cdx2 , and Sox17 . These genes encode developmentally important transcription factors and are known Oct4 targets ( Chen et al . , 2008 ) , but are silent in ESCs . Upon RA-mediated differentiation , lineage-appropriate ( ectoderm ) genes such as Hoxa5 and Hoxc6 ( Jiang et al . , 2011 ) normally ‘resolve’ their bivalent state by losing H3K27me3 and becoming induced , while lineage-inappropriate ( e . g . endoderm ) genes such as Cdx2 and Sox17 normally resolve by losing H3K4me3 , gaining DNA methylation , and becoming stably silenced . Induction of Hoxa5 and Hoxc6 was robust following RA-mediated differentiation of WT cells , but defective in the Oct1 KO condition ( Figure 2B ) . In contrast , Cdx2 was ectopically activated upon RA-mediated differentiation of Oct1-deficient ESCs ( Figure 2B ) . Similarly , the definitive endoderm-specific gene Sox17 is not normally induced upon RA-mediated differentiation , but showed ectopic expression in the absence of Oct1 ( Figure 2B ) . The ectopic Sox17 expression observed with RA differentiation differed from the expression defects observed in Oct1-deficient EBs , which include endodermal lineages and which showed 100-fold stronger Sox17 expression ( Figure 1F ) . These results indicate that Oct1-deficient ESCs induce lineage-appropriate developmental genes poorly , while ectopically expressing lineage-inappropriate genes . 10 . 7554/eLife . 20937 . 005Figure 2 . Effect of Oct1 loss on RA-mediated differentiation and neurogenesis . ( A ) Quantitative RT-PCR results are shown for Pou5f1 ( Oct4 ) , Sox2 , and Pou2f1 ( Oct1 ) mRNA relative to a GAPDH standard . Average of three biological replicates ±standard deviation is shown . ( B ) Similar analysis performed for the Oct4 targets Hoxa5 , Hoxc6 , Cdx2 , and Sox17 . ( C ) 293 T cells were transiently transfected with a lentiviral vector ( pHAGE ) expressing mouse Oct1 . Lysates were prepared 48 hr later and immunoblotted for Oct1 . Un-transfected cells are shown as a negative control ( lane 1 ) . Oct1 is not visible in these cells because of the lightness of the exposure . The same cells transduced with a retroviral vector encoding Oct1 ( lane 3 ) are shown as a positive control . β-actin is shown as a loading control . ( D ) Oct1-deficident ESCs were differentiated using RA for 14 days . 4 days into the timecourse , cells were infected with lentiviruses expressing Oct1 and a puromycin resistance cassette , or an empty vector ( EV ) control . Cells were selected with puromycin for the remainder of the timecourse . cDNAs from the endpoint cultured were used to study expression of Hoxa5 relative to an RPL13 ribosomal protein internal standard . Cells were prepared in triplicate for each condition . Error bars denote ±standard deviation . denotes p<0 . 05 . ( E ) Immunofluorescence images of WT and Oct1-deficient ESCs differentiated into neurons . Cells were cultured as EBs for 8 days , followed by culture for a further 8 days in neuralizing media ( see Materials and methods ) . β-tubulin III and DAPI staining are shown . ( F ) Quantification of 300–400 cells from three individual differentiation experiments . Error bars denote ±standard deviation . ( G ) Similar to ( E ) except cells were cultured for eight additional d in neuralizing media . ( H ) Similar to ( F ) except using cells cultured for eight additional d . ( I ) 16 d-differentiated neuron cultures of similar genotypes were pooled and subjected to RT-qPCR using primers specific for Tubb3 , Nestin , and Map2 . Expression was assessed relative to GAPDH . Averages of three biological replicates are shown . Error bars denote ±standard deviation . DOI: http://dx . doi . org/10 . 7554/eLife . 20937 . 00510 . 7554/eLife . 20937 . 006Figure 2—figure supplement 1 . β-tubulin III staining of neuralizing WT and Oct1-deficient EBs . ( A ) WT cells . ( B ) Oct1-deficient cells . Two day-old EBs were used . DOI: http://dx . doi . org/10 . 7554/eLife . 20937 . 006 In order to determine whether forced Oct1 expression during differentiation was sufficient to correct defects in gene expression , we differentiated Oct1-deficient ESCs using RA and infected the cell during the differentiation timecourse with lentiviral vectors expressing Oct1 and a puromycin reistance cassette , or empty vectors containing the puromycin resistance cassette alone . We confirmed that cells transduced with this vector overexpressed Oct1 by immunoblotting ( Figure 2C ) . Cells were infected over 2 consecutive days and selected with puromycin throughout the remained of the 14-day differentiation timecourse . At timepoints after day 6 , infection and selection with empty vector skewed the expression of genes such as Hoxa5 , suggesting that infection and selection were skewing the populations of cells in the culture . Cells infected at 4 and 5 days , however , did not show major differences ( not shown ) , suggesting that the composition of cells in culture was not being significantly altered . We therefore infected differentiating Oct1-deficient cells consecutively on days 4 and 5 , prepared cDNAs at day 14 and examined gene expression . By RT-qPCR , Oct1 was undetectable in cells transduced and selected with empty vector but robustly expressed by cells transduced with Oct1 ( not shown ) . Expression of the developmentally-inducible Hoxa5 gene was significantly augmented ( p=0 . 026 ) by ectopic Oct1 expression ( Figure 2D ) . These results indicate that restoration of Oct1 expression at these times and conditions can correct at least some of the gene expression defects associated with Oct1 deficiency . RA-mediated differentiation yields neuronal precursor cells but not neurons . We used a differentiation system involving EB generation and culture in insulin , transferrin and selenium ( see Materials and methods ) to generate arborized neurons that express the marker β-tubulin III ( Tubb3 ) and the neuroectoderm genes Nestin ( Nes ) and Map2 . Staining of 2-day-old EBs for β-tubulin III prior to laminin/poly-L-lysine dish attachment – early in the differentiation protocol - revealed fewer β-tubulin III-positive cells in the Oct1 deficient condition ( Figure 2—figure supplement 1 ) . Upon complete differentiation ( 8 d EBs , 8 days in neuralizing monolayer culture ) , WT ESCs formed neurons robustly ( Figure 2E , asterisk ) while few β-tubulin III-expressing neurons were formed from Oct1-deficient ESCs . Oct1-deficient cells that did induce β-tubulin III tended to do so at lower levels , and the few cells that did express β-tubulin III robustly were nevertheless abnormal ( Figure 2E–F , arrow ) . To test if Oct1 loss induced a kinetic delay that could be overcome by longer culture , cells were incubated for 8 or 16 additional days ( 16 or 24 days in neuralizing medium , 24 or 32 days total differentiation ) . In neither case were neurons formed ( Figure 2G–H and data not shown ) . To study gene expression , individual wells of common genotypes differentiated for 16 days were pooled and subjected to RT-qPCR for Tubb3 , Nes and Map2 . Each of these genes showed defective expression in the absence of Oct1 ( Figure 2I ) . To test an unrelated developmental system , we performed cardiomyocyte differentiation by culturing EBs in hanging drops followed by culture with gelatin ( see Materials and methods ) . Oct1-deficient ESCs failed to form beating cardiomyocytes , unlike WT ( Figure 3A and Videos 1–12 ) . RNA was collected from pooled beating and non-beating WT colonies , and Oct1-deficient colonies , and used to analyze Mef2c and Hand1 , regulators of cardiomyocyte differentiation , and the terminal differentiation markers Mlc2v and Mlc2a . We observed Mlc2v expression defects in the Oct1-deficient condition equivalent to non-beating cardiomyocyte colonies from WT ESCs . Mlc2a and Mef2c expression was even weaker than non-beating WT cardiomyocyte colonies ( Figure 3B ) . In contrast , Hand1 showed no expression defects ( Figure 3B ) , indicating that Oct1 deficiency does not globally down-regulate genes associated with cardiomyocyte differentiation . Cumulatively , the results indicate that although Oct1-deficient ESCs appear normal in the absence of differentiation cues , they do not induce poised developmentally inducible genes and fail to repress lineage-inappropriate genes such as Cdx2 and Sox17 , resulting in multiple cellular defects following differentiation . 10 . 7554/eLife . 20937 . 007Figure 3 . Defective cardiomyocyte differentiation in ESCs lacking Oct1 . ( A ) Cardiomyocytes were generated from individual EBs using 24-well dishes with gelatin . Functionality ( ±beating ) was assessed for each well ( 16 per genotype ) and plotted . ( B ) The wells assessed in ( A ) were pooled according to genotype and function ( beating WT , non-beating WT and non-beating Oct1 deficient ) , cDNA was prepared and used for RT-qPCR using primers for Mlc2v , Mlc2a , Mef2c and Hand1 . Averaged results from three replicates are shown . Error bars denote standard deviation . p-Values: NS=non-significant , < 0 . 05 , < 0 . 01 , * < 0 . 001 . DOI: http://dx . doi . org/10 . 7554/eLife . 20937 . 00710 . 7554/eLife . 20937 . 008Video 1 . Example WT ESC line cardiomyocyte differentiation 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 20937 . 00810 . 7554/eLife . 20937 . 009Video 2 . Example WT ESC line cardiomyocyte differentiation 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 20937 . 00910 . 7554/eLife . 20937 . 010Video 3 . Example WT ESC line cardiomyocyte differentiation 3 . DOI: http://dx . doi . org/10 . 7554/eLife . 20937 . 01010 . 7554/eLife . 20937 . 011Video 4 . Example WT ESC line cardiomyocyte differentiation 4 . DOI: http://dx . doi . org/10 . 7554/eLife . 20937 . 01110 . 7554/eLife . 20937 . 012Video 5 . Example WT ESC line cardiomyocyte differentiation 5 . DOI: http://dx . doi . org/10 . 7554/eLife . 20937 . 01210 . 7554/eLife . 20937 . 013Video 6 . Example WT ESC line cardiomyocyte differentiation 6 . DOI: http://dx . doi . org/10 . 7554/eLife . 20937 . 01310 . 7554/eLife . 20937 . 014Video 7 . Example WT ESC line cardiomyocyte differentiation 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 20937 . 01410 . 7554/eLife . 20937 . 015Video 8 . Example WT ESC line cardiomyocyte differentiation 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 20937 . 01510 . 7554/eLife . 20937 . 016Video 9 . Example WT ESC line cardiomyocyte differentiation 3 . DOI: http://dx . doi . org/10 . 7554/eLife . 20937 . 01610 . 7554/eLife . 20937 . 017Video 10 . Example WT ESC line cardiomyocyte differentiation 4 . DOI: http://dx . doi . org/10 . 7554/eLife . 20937 . 01710 . 7554/eLife . 20937 . 018Video 11 . Example WT ESC line cardiomyocyte differentiation 5 . DOI: http://dx . doi . org/10 . 7554/eLife . 20937 . 01810 . 7554/eLife . 20937 . 019Video 12 . Example WT ESC line cardiomyocyte differentiation 6 . DOI: http://dx . doi . org/10 . 7554/eLife . 20937 . 019 Although the ESC lines described above were derived from littermate animals and had normal karyotypes , it was possible that the developmental phenotypes and altered gene expression patterns resulted from differences unrelated to Oct1 status . Furthermore , the observed gene expression defects could result from compensatory changes due to development in an Oct1-deficient environment . Finally , the allele used to generate these lines is a severe hypomorph rather than a complete null ( Wang et al . , 2004 ) . To circumvent these issues , we generated tamoxifen-inducible , Oct1 conditional-deficient ESCs . We previously described Pou2f1 conditional ( floxed ) mice ( Shakya et al . , 2015b ) . We generated inducible-conditional Oct1 ESCs by crossing the floxed allele onto Rosa26-Cre-ERT2 and Rosa26-lox-stop-lox-YFP ( see Materials and methods ) . Pregnant animals were used to isolate Pou2f1f1/fl;Rosa26-Cre-ERT2;Rosa26-lox-stop-lox-YFP ESC lines in which Oct1 could be acutely deleted and YFP induced by 4-hydroxytamoxifen ( 4-OHT ) administration . Treatment of parent ESCs with 4-OHT resulted in variegated YFP+ colonies ( Figure 4A , step one at top ) . Colonies with good morphology were picked ( red arrow ) , trypsinized and expanded into derived ESC lines ( Figure 4A , step two at bottom ) that genotyped as Pou2f1Δ/Δ or Pou2f1fl/Δ ( Figure 4B ) . The designation Δ will be used to differentiate this allele from the germline deficient allele used in Figures 1–3 . As with Oct1 germline-deficient ESCs , derived Pou2f1Δ/Δ ESC lines displayed normal colony morphology ( Figure 4C ) , proliferated normally ( not shown ) yet expressed no Oct1 ( Figure 4D ) . The derived cells showed normal karyotype profiles and could be propagated for >1 month without loss of an undifferentiated phenotype ( not shown ) . Similar to germline Oct1 deficient ESCs , derived Pou2f1Δ/Δ cells also expressed Oct4 , Sox2 and Nanog at normal levels ( Figure 4D ) . 10 . 7554/eLife . 20937 . 020Figure 4 . Gene expression defects upon differentiation of Oct1 inducible-conditional ESCs . ( A ) YFP-epifluorescence and phase microscopy images of inducible-conditional ESCs . Top: parent Pou2f1fl/fl cells were treated with 500 nM 4-OHT for 24 hr . A colony with good morphology and variegated YFP expression was picked , trypsinzed , replated and expanded . Bottom: derived Pou2f1Δ/Δ ESCs . ( B ) PCR genotyping of parent ( C2 ) and derived ( 13-1 , 13-2 ) lines . Feeder fibroblasts were depleted by two serial 1 hr platings on plastic . The residual WT band in lane two is due to feeder contamination . The rightmost lane ( lane 4 ) shows control tail DNA from a Pou2f1fl/+ ( top ) or Pou2f1fl/Δ animal . ( C ) Epifluorescence and phase microscopy images of single colonies . Images were taken at the same magnification . ( D ) Immunoblots comparing lysates of vehicle-treated parent ESCs and derived KO cells . GAPDH is shown as a loading control . ( E ) Feeder-depleted ESCs were treated continuously with RA on gelatin-coated plates in the absence of LIF for 14 days . Media was changed every other day . cDNA was prepared every other day and used in RT-qPCR with primers against Pou5f1 and Sox2 . Averages of three biological replicates ±standard deviation are shown . Methodologically , the experiments were performed identically to Figure 2A–B . ( F ) Additional RT-qPCR using primers against the Oct4 target genes Hoxa5 , Hoxc6 and Cdx2 . ( G ) Pou2f1Δ/Δ ESCs were differentiated into neurons as in Figure 2C–G by forming EBs for 8 days followed by 8 day in medium containing insulin , transferrin and selenium . Cells were fixed and used for immunofluorescence using DAPI and antibodies against YFP and β-tubulin III . ( H ) Immunofluorescence images were quantified based on arborized morphology . Approximately 700 total cells were analyzed . ( I ) Similar to ( G ) except parent Pou2f1fl/fl , ESCs were used and after 4 days in culture cells were treated with 500 nM 4-OHT for 24 hr to delete Oct1 and induce YFP . Two representative images are shown . ( J ) Immunofluorescence images were quantified based on YFP and β-tubulin III positivity . Percent total cells showing single or double staining , or percent β-tubulin III+ cells with and without YFP are shown . Approximately 700 total cells were analyzed . DOI: http://dx . doi . org/10 . 7554/eLife . 20937 . 02010 . 7554/eLife . 20937 . 021Figure 4—figure supplement 1 . Steady-state metabolite levels in parental WT and 4-OHT-treated Pou2f1Δ/Δ ESCs as determined by GC-MS . ( A ) Identified metabolites with total ion current ( TIC ) levels between 0 . 25 and 2 . 5 . Normalization were achieved by setting the total ion current area under curve for all metabolites equal to each other . Average of four biological replicates is shown . Error bars depict ±standard deviation . ( B ) TICs between 1 and 15 . * denotes student T-test p<0 . 05 . ( C ) TICs between 10 and 200 . ( D ) TICs between 100 and 1500 . DOI: http://dx . doi . org/10 . 7554/eLife . 20937 . 02110 . 7554/eLife . 20937 . 022Figure 4—figure supplement 2 . Images of parental WT and 4-OHT-treated KO ESCs differentiating in the presence of RA . Left: phase microscopy images . Right: epifluorescence images taken using a YFP filter . DOI: http://dx . doi . org/10 . 7554/eLife . 20937 . 022 In differentiated cells , Oct1 promotes glycolysis and dampens mitochondrial function . Oct1 deficiency dramatically increases mitochondrial amino acid oxidation and oxygen consumption while decreasing glycolysis and to a lesser extent glucose oxidation ( Shakya et al . , 2009 ) . These changes contribute to failure of fibroblasts to undergo oncogenic transformation , despite the fact that they grow at normal rates and can be immortalized by serial passage . To test if similar changes occur in ESCs lacking Oct1 , we analyzed the metabolic profile of these cells . Few differences were noted , with only phosphoethanolamine ( also known as phosphorylethanolamine , p=0 . 047 ) and inositol ( p=0 . 037 ) showing significant changes ( Figure 4—figure supplement 1 ) . The lack of difference may be due to redundant functions of co-expressed Oct4 and Oct6 , or co-selection for metabolic stability when selecting and propagating ESCs . Derived Pou2f1Δ/Δ ESC lines , and parent cell line controls , were subjected to RA-mediated differentiation . Similar to results using Oct1 germline-deficient ESCs , the derived Pou2f1Δ/Δ ESCs lost Oct4 and Sox2 expression with kinetics identical to the parent line ( Figure 4E ) . Microscopic imaging of the differentiating cells revealed that they were morphologically similar until approximately d 12 , at which point Pou2f1Δ/Δ cells showed an increase in columnar/epithelial appearance ( Figure 4—figure supplement 2 ) . Also as before , the induction of silent , developmentally poised genes was defective: Hoxa5 and Hoxc6 both showed reduced expression in timecourse assays ( Figure 4F ) . The cells also showed ectopic Cdx2 expression upon RA treatment ( Figure 4F ) . As with germline-deficient ESCs , Pou2f1Δ/Δ ESCs did not generate true neurons efficiently ( Figure 4G ) . To determine the effect of conditional Oct1 loss during differentiation , parent cells were treated with 4-OHT following 8 d EB formation and 4 d in insulin , transferrin and selenium . After an additional 4 days , cells were fixed and stained with antibodies against β-tubulin III to score neurogenesis and YFP to score deletion . 40–50% of the treated cells induced YFP . Nearly all cells that induced β-tubulin III and/or generated neuron morphology lacked YFP expression ( Figure 4I and Figure 4J ) . A few cells ( 2/ ~ 700 ) were both YFP- and β-tubulin III-positive ( not shown ) , though it is possible that these cells are Pou2f1 heterozygous as 4-OHT treatment can result in recombination of only one allele ( Figure 4B ) . Parent and Pou2f1Δ/Δ ESCs were injected subcutaneously into contralateral flanks of NCr Nude immunocompromised animals to generate teratomas . ESCs lacking Oct1 consistently generated smaller tumors ( Figure 5A–C ) . Immunoblotting confirmed that recovered tumors maintained their original Oct1 status ( Figure 5D ) . Histological analysis confirmed that parent cells generated mature teratomas that included , e . g . , glial tissue , and glandular epithelial and squamous elements ( Figure 5E ) . In contrast , Oct1 deficient ESCs generated areas of focally immature cells , consistent with reduced differentiation . Occasionally tumors were comprised virtually entirely of primitive malignant cells resembling a germ cell tumor ( Figure 5E , lower right ) . 10 . 7554/eLife . 20937 . 023Figure 5 . Smaller , less differentiated teratomas and lack of contribution to adult mouse tissues in Pou2f1Δ/Δ ESCs . ( A ) 1 × 106 ESCs were injected into flanks ( left flank: control Pou2f1fl/fl parent cells , right flank: derived Pou2f1Δ/Δ ESCs ) of NCr Nude mice . Images are shown at 4 weeks . ( B ) Images of dissected teratomas . Left side: white light . Scale in cm shown on the right . Right side: YFP fluorescence . ( C ) The mass from ten tumors was averaged and plotted . Error bars denote standard deviation . ( D ) Immunoblots are shown for Oct1 in lysates prepared from part of the individual teratomas . Lane 7: extracts from 293 T cells transiently over-expressing recombinant Oct1 . α-Tubulin is shown as a loading control . The lack of α-Tubulin in lane seven arises from the fact that less protein was loaded due to high levels of recombinant Oct1 . ( E ) H and E images of parent and KO teratomas . Top left: normal teratoma morphology comprised of mature elements , e . g glial tissue ( arrowhead ) , mature squamous ( left arrow ) and ciliated glandular epithelial ( right arrow ) . Top right: teratoma comprised predominantly of mature elements , but with immature elements ( approximately 5% of the tumor , arrows ) . Bottom left: teratoma with both mature and immature elements represented . Mature squamous epithelium ( right arrow ) is abundant , while immature neuroepithelium ( left arrow ) is noted focally . Bottom right: tumor comprised almost entirely of a primitive malignant neoplasm that does not recapitulate any recognizable line of differentiation . Arranged in sheets and irregular nests , these cells exhibit marked cytologic atypia , with nuclear pleomorphism and coarsely-clumped chromatin . Nucleoli are variably prominent . Mitotic activity is brisk ( arrows ) . ( F ) Parent ESCs ( left side ) or derived KO cells ( right side ) were injected into albino C57BL/6 blastocysts and implanted in pseudo-pregnant animals . Representative images are shown . ( G ) Average contribution is shown for the two cell types . ESCs contribution was assessed subjectively based on dark coat and eye color . 33 animals were tested in the case of the parent line and 36 animals were assessed in the case of the derived Pou2f1Δ/Δ line . DOI: http://dx . doi . org/10 . 7554/eLife . 20937 . 02310 . 7554/eLife . 20937 . 024Figure 5—figure supplement 1 . Pluripotent phenotype of ESCs immediately prior to blastocyst injection . Microscopy images of parental WT and two derived 4-OHT-treated , YFP-expressing , Pou2f1Δ/Δ ESC lines cultured in the absence of feeder fibroblasts . Images were taken by the University of Utah Transgenic Core Facility immediately prior to injection . An established WT eYFP-expressing ESC line ( which fluoresces more brightly ) is shown as a positive control . DOI: http://dx . doi . org/10 . 7554/eLife . 20937 . 024 A standard measure of pluripotency is the ability to contribute efficiently to adult cells and tissues ( De Los Angeles et al . , 2015 ) . We injected parent and Pou2f1Δ/Δ ESCs into albino C57BL/6 blastocysts , resulting in high contribution in the case of the parent line ( Figure 5F , left ) , but no contribution in the case of the derived lines ( right ) . The average percent chimerism from two separate sets of injections ( 33 animals from parent cell line injections , 36 combined from two different Pou2f1Δ/Δ lines ) confirmed the lack of contribution ( Figure 5G ) . 18/33 animals injected with parent ESCs showed some detectable chimerism ( 55% ) , while 1/36 animals injected with conditional knockout ESCs showed transient trace chimerism in the eye ( 0 . 03% ) . The cells were imaged immediately prior to blastocyst injection to confirm an undifferentiated phenotype ( Figure 5—figure supplement 1 ) . To identify gene expression changes stemming from loss of Oct1 , we performed RNAseq with undifferentiated and 14 d RA-differentiated parent and Pou2f1Δ/Δ ESCs . Three independent replicates were performed for each of the four conditions . Between 18 . 1 and 24 . 9 million sequence reads were generated for each sample , 73% to 82% of which aligned uniquely to the mouse Mm10 reference genome . 99 . 6% of the reads within coding regions aligned to the correct strand . Variance between replicates was similar regardless of genotype or differentiation state ( not shown ) . Unsupervised hierarchical clustering indicated that 0 and 14 days samples separated clearly from each other regardless of genotype , while within each timepoint the KO and parent WT samples clustered together ( Figure 6—figure supplement 1A ) . These results indicated that the effect of RA treatment and differentiation on gene expression was far stronger than the effect of Oct1 deletion . Plotting gene expression levels in the parent vs Pou2f1Δ/Δ cells ( Figure 6A ) showed relatively few gene expression changes in the undifferentiated condition ( >2 . 5 fold , p<0 . 01 , 253 total genes ) . These genes never changed by >7 fold ( Supplementary file 1 ) . In contrast , 1123 genes change expression in differentiated Pou2f1Δ/Δ cells , some of which varied by >200 fold . Plotting gene expression fold change vs . p-value ( Figure 6B ) recapitulated these findings . Comparing genes differentially expressed at the two timepoints revealed little overlap ( 23 genes , Figure 6C ) . Analysis of the genomic alignments revealed that expression of control genes such as Tbp was unaltered , while pluripotency genes such as Nanog were silenced equivalently ( Figure 6D ) . Other pluripotency genes such as Pou5f1 , Klf4 , Dnmt3l , and Dppa4 behaved similarly to Nanog ( not shown ) . One gene showing increased expression in the undifferentiated state was Pou2f3 ( Oct11 , Figure 6—figure supplement 1B ) . Pou2f3 shows low but detectable mRNA expression in WT ESCs , and is repressed upon differentiation to nearly undetectable levels . It is slightly elevated in Oct1-deficient ESCs but decreases to an even greater extent upon differentiation . RT-qPCR confirmed these changes in the context of overall low expression ( Figure 6—figure supplement 1C ) . It is therefore unlikely that this protein provides a compensatory function upon differentiation . 10 . 7554/eLife . 20937 . 025Figure 6 . Genome-wide changes in developmental gene expression following differentiation of Oct1 conditional-inducible deficient ESCs . ( A ) For each gene , averaged RNAseq FKPM ( aligned Fragments Per Kilobase per Million aligned reads ) values from three undifferentiated or 14-day RA-differentiated parent and KO ESCs were plotted on a log10 scale . Red genes signify significantly changed gene expression ( Adj . p<0 . 01 , fold change >2 . 5 ) at d 14 . For each timepoint , genes showing <50 total reads in both genotypes conditions were called unexpressed and are not displayed . ( B ) Volcano plots showing log2 averaged difference in gene expression vs . –10×log10 significance . Significantly altered genes ( Adj . p<0 . 01 , 2 . 5-fold change ) are shown in blue ( down-regulated ) or red ( up-regulated ) . ( C ) Venn diagram showing total numbers of significantly ( p<0 . 01 ) differentially expressed ( >2 . 5 fold ) genes in undifferentiated and 14-day RA-differentiated Oct1-deficient ESCs . Overlap shows genes differentially expressed at both timepoints . ( D ) Genome tracks of averaged RNAseq read densities ( genome build mm10 ) for two control genes: Tbp ( a constitutively expressed gene ) , and Nanog ( expressed in pluripotent but not differentiated conditions ) . Arrows show directionality of gene transcription and size of the transcription unit . ( E ) Additional genome tracks are shown of three genes with poor induction in the KO condition: Hoxa5 , Hoxb9 , and Nppa . Ahcy is also shown , which becomes more strongly down-regulated in the differentiated condition . Hoxa5 physically overlaps with Hoxa3 , Hoxaas3 , and Hoxa6 , which are not highlighted . ( F ) Additional genome tracks are shown of genes showing ectopic expression in the differentiated condition: Sox17 , Cdx2 , Pparg , and Plr8a6 . ( G ) RT-qPCR validations of additional genes identified by RNAseq , Ahcy and Prl8a6 . Tbp is shown as a control . Average of three biological replicates ±standard deviation is shown . DOI: http://dx . doi . org/10 . 7554/eLife . 20937 . 02510 . 7554/eLife . 20937 . 026Figure 6—figure supplement 1 . Differences in gene expression in differentiated Oct1 deficient cells revealed by RNAseq . ( A ) Heat map of gene expression differences across all biological replicates generated using unsupervised hierarchical clustering . 10 , 500 genes that vary at least fourfold across all samples were used . Color shows Log2 deviation of each gene from its average value across all the samples . Yellow= above average . Blue=below average . Each unit in the color key represents a doubling in intensity . Input values were log-scale FPKMs for each gene in each sample . ( B ) Genome tracks of Pou2f3 . Asterisk: spurious background signal sometimes seen with poorly expressed genes . ( C ) RT-pPCR of Pou2f3 expression . Average of three biological replicates ±standard deviation is shown . DOI: http://dx . doi . org/10 . 7554/eLife . 20937 . 02610 . 7554/eLife . 20937 . 027Figure 6—figure supplement 2 . Differences in gene expression in differentiated Oct1 deficient cells revealed by RNAseq . ( A ) Genome tracks of additional genes showing decreased expression in the differentiated state identified by RNAseq . ( B ) Genome tracks of elevated genes . DOI: http://dx . doi . org/10 . 7554/eLife . 20937 . 027 Approximately 300 genes were induced poorly in RA treated Pou2f1Δ/Δ cells relative to normal controls . Many of these encode regulators of neuronal specification and differentiation . Examples include Hoxa5 , Hoxb9 and Nppa ( Figure 6E ) . Other examples include Foxg1 , Pcdh17 , Ptgfr , Akr1c18 , Cts7 , Duox2 , Hoxc5 , Hoxc6 , Hoxc8 , Hoxc10 , and Dcx ( Figure 6—figure supplement 2A ) . In addition , Ahcy , a stress-responsive Oct1 target ( Kang et al . , 2009; Shakya et al . , 2011 ) showed weakened expression in the absence of Oct1 specifically in the differentiated condition ( Figure 6E ) . An even larger cohort of ~800 genes was aberrantly expressed upon differentiation of Pou2f1Δ/Δ cells . These genes are strongly associated with alternative developmental fates ( Figure 6F and Figure 6—figure supplement 2B ) . Examples include Sox17 , Cdx2 and Gata4 ( endoderm ) , Fgb ( endoderm/liver ) , Gata2 ( mesoderm/endothelial ) , Pparg and Irx3 ( mesoderm/mesenchymal ) , Muc13 ( epithelial/hematopoietic ) and Tnfrsf9 ( which encodes CD137/hematopoietic ) . The difference in Sox17 and Gata2 expression between EBs ( defective ) and RA-differentiated cells ( elevated ) likely arises from the additional developmental fates specified in EBs . Unexpectedly , differentiating Pou2f1Δ/Δ cells also resulted in inappropriate expression of genes associated with trophoblast and placental development , the specification of which is normally restricted to trophectoderm cells rather than the inner cell mass ( from which ESCs are derived ) . Examples include Cdx2 ( which is also expressed in endoderm ) , Prl8a6 , Hand1 ( which is also expressed in cardiomyocytes ) , Pappa2 , Prl3b1 , and Psg27 ( Figure 6F and Figure 6—figure supplement 2B ) . Some of these genes are also expressed in other lineages while others are highly specific . In aggregate , they indicate improper activation of an extra-embryonic program . Using RT-qPCR we confirmed unperturbed expression of Tbp , defective expression of Ahcy , and elevated expression of Prl8a6 in differentiating Oct1-deficient ESCs ( Figure 6G ) . These results indicate that Oct1 deficiency results in defective lineage specification upon differentiation . We used ChIPseq to identify common and unique Oct1 and Oct4 target genes in ESCs . We also performed H3K4me3 ChIPseq as a control . The ChIPseq data were of high quality based on measures of signal/noise ratio ( see Materials and methods ) . After filtering , 27 . 3 ( Oct1 ) , and 23 . 7 ( Oct4 ) million alignable reads were generated , corresponding to 692 ( Oct1 ) , and 8673 ( Oct4 ) peaks . Allocating the peaks to nearest genes revealed 209 unique Oct1 target genes , 356 common targets , and 5563 unique Oct4 targets ( Figure 7A ) . The smaller size of the Oct1 target pool relative to Oct4 may be attributable to >10 fold lower Oct1 levels in ESCs as observed by RT-qPCR ( Figure 1C , Figure 1E , Figure 2A ) and RNAseq ( not shown ) . Oct1 may also require a more open chromatin context , and/or the presence of specific co-bound factors , to access DNA . For example Il2 and Ifng are known Oct1 targets in differentiated T cells ( Shakya et al . , 2015b , 2011 ) but were not identified as targets in this analysis . 10 . 7554/eLife . 20937 . 028Figure 7 . Unique and common Oct1 and Oct4 targets in ESCs . ( A ) Venn diagram illustrating Oct1 and Oct4 target gene profile and intersection with RNAseq gene set . ( B ) Motif analysis for peaks occupied uniquely by Oct1 or Oct4 , and for peaks occupied by both proteins . Top shows best consensus sequences associated with binding . Bottom shows best matches to annotated weight matrices . In the case of known motifs , deviation of physiological binding sites from consensus causes recurring sequences meet threshold criteria for the compound ‘Oct4-Sox2’ site but not for a simple octamer site ( ‘Oct4’ ) . This is why the percentage of target sites computationally associated with ‘Oct4-Sox2’ is higher than for ‘Oct4 . ’ ( C ) Genome tracks showing ChIPseq enrichment for Oct1 ( blue ) , Oct4 ( gray ) or H3K4me3 ( red ) . Target gene and orientation is shown at the bottom of each track . Pou5f1 , Polr2a , Taf12 and Pax6 are shown . ( D ) ChIP-qPCR validation of select ChIPseq targets . Fold enrichments using Oct1 and Oct4 antibodies at Polr2a and Pou5f1 are shown . ( E ) RT-qPCR for three identified Oct1 target genes , Tbx3 , Tcf4 and Tbx6 at 0 , 2 , 4 , 6 , 8 , 10 , 12 and 14 d of differentiation . Expression was normalized to the control ribosomal gene RPL13 . Three biological replicates were performed . Error bars denote ±standard deviation . DOI: http://dx . doi . org/10 . 7554/eLife . 20937 . 02810 . 7554/eLife . 20937 . 029Figure 7—figure supplement 1 . Oct1 and Oct4 ChIPseq read density at example co-bound genes . ( A ) Genome tracks of Oct1 and Oct4 enrichment at Ell in ESCs . ( B ) Ahcy . ( C ) Rras2 . DOI: http://dx . doi . org/10 . 7554/eLife . 20937 . 029 Motif analysis of unique and co-bound peaks revealed significant differences in recognized DNA elements . Regions associated exclusively with Oct4 were significantly enriched for Oct-Sox compound elements that likely also associate with Sox2 in ESCs ( Figure 7B ) . In contrast , target regions preferentially associated with Oct1 were enriched for the simple octamer element ATTTGCAT ( shown by the software as an Oct4 motif in Figure 7B ) . Interestingly , co-occupied peaks strongly associate with a motif termed a MORE that is known to bind two Oct protein molecules ( Reményi et al . , 2001; Tomilin et al . , 2000 ) . In differentiated cells lacking Oct4 , oxidative stress induces homodimeric Oct1 binding to MORE-containing genes such as Polr2a , Ahcy , Ell , and Rras2 . Oxidative stress-induced binding occurs via phosphorylation of a conserved serine residue in the DNA-binding domain ( Kang et al . , 2009 ) . These genes were constitutively co-bound by Oct1 and Oct4 in ESCs ( Figure 7—figure supplement 1 ) . Additional examples of genes associated with Oct4 alone ( Pou5f1 ) , or Oct1 alone ( Taf12 ) are shown in Figure 7C . This panel also shows another example of a MORE containing gene ( Polr2a , two tandem MOREs binding four molecules ) that also associates with both proteins but shows an Oct1 bias , as well as an example ( Pax6 ) that is bound by both proteins but in two different locations . Using ChIP-qPCR we validated two genes , Polr2a ( Oct1-enriched ) and Pou5f1 ( Oct4-enriched , Figure 7D ) . The complete set of identified targets is shown in Supplementary file 2 . Intersecting the ChIPseq and RNAseq data revealed little overlap . Only 34 Oct1-bound or Oct1/Oct4 co-bound targets showed differential expression following RA-mediated differentiation ( Figure 7A ) . Examples include Pank4 , Cdh5 and Med16 . 193 Oct1-bound and 325 Oct1/Oct4-co-bound genes did not show expression differences at d 14 . Examples include Tbx3 , Tcf4 , and Txb6 , which also showed no differences throughout the differentiation timecourse ( Figure 7E ) . Instead , 1066 genes with altered expression in differentiated Oct1-deficient cells showed Oct4 but not Oct1 enrichment . These findings indicate that ( 1 ) identified Oct1 targets were not differentially expressed upon differentiation , and ( 2 ) developmental genes shown to be differentially expressed in RA treated Oct1-deficient cells were not Oct1 targets in ESCs . The above findings could be reconciled by postulating that ( 1 ) Oct functions at co-bound targets in ESCs to buffer them against oxidative stress as described previously in fibroblasts ( Kang et al . , 2009; Shakya et al . , 2011 ) , and ( 2 ) developmental genes that are differentially expressed in Oct1-deficient cells but exclusively bound by Oct4 in ESCs become Oct1 targets during the differentiation process as Oct4 is lost . To test the first hypothesis , we studied the effect of H2O2 exposure on the expression of two co-bound genes , Ahcy and Polr2a , in ESCs ±Oct1 . Both genes contain conserved MORE sequences ( Figure 8A ) . Treatment of cells with 2 mM H2O2 resulted in a rapid loss of Ahcy and Polr2a mRNA specifically in Oct1-deficient ESCs ( Figure 8B ) , exactly as observed in fibroblasts ( Kang et al . , 2009; Shakya et al . , 2011 ) . As expected , these cells were hypersensitive to H2O2 ( Figure 8C ) . These results suggest that as in other cell types , Oct1 functions in ESCs to buffer these genes from oxidative stress-associated inhibition . 10 . 7554/eLife . 20937 . 030Figure 8 . Oct1/Oct4 co-binding to MOREs in ESCs , and inducible Oct1 binding to poised targets upon differentiation . ( A ) Conserved MOREs at two genes ( Ahcy and Polr2a ) co-bound by Oct1 and Oct4 in ESCs . The top co-bound de novo motif from Figure 7B is shown at top . The MORE sequence ( Reményi et al . , 2001; Tomilin et al . , 2000 ) is shown at bottom . Mammalian ( mouse , human , dog ) conservation is shown . MORE position relative to TSS is shown in parentheses . Polr2a contains two adjacent MOREs ( Kang et al . , 2009 ) , only one of which is shown here . ( B ) Parent and derived KO ESCs were treated with 2 mM H2O2 for the indicated times . Ahcy and Polr2a log2 mRNA levels were assessed by RT-qPCR . Three biological replicates were performed . Error bars denote ±standard deviation . ( C ) Microscopic images of the same cells during the treatment timecourse . ( D ) Conservation of the octamer sequence in the Hoxc5 3’ UTR and Rest/Nrsf upstream region . Mammalian ( mouse , human , dog ) conservation is shown . Octamer element position relative to TSS is shown in parentheses . ( E ) Genome tracks showing ChIPseq enrichment at Hoxc5 , Rest and upstream of Myf5 , for Oct1 ( blue ) , Oct4 ( gray ) or H3K4me3 ( red ) . ( F ) ChIP-qPCR differentiation timecourse of four targets that exclusively bind Oct4 in ESCs . Fold enrichments using Oct1 and Oct4 antibodies at Hoxc5 , Pou5f1 , Rest and Myf5 are shown . ( G ) Model for Oct1 and Oct4 function in ESCs and their differentiated progeny . The example of neuronal differentiation is shown . In stem cells , Oct1 and Oct4 collaborate at constitutively expressed MORE-containing targets such as Polr2a and Ahcy to insulate them against oxidative stress ( red and black short dashed lines ) . Oct4 poises developmental genes of all embryonic lineages ( long dashed black line ) and repress trophectoderm-specific genes ( solid black block line ) . Oct4 additionally activates pluripotency genes ( solid black arrow ) . In differentiating cells , Oct1 occupies MORE genes in response to oxidative stress and buffers their expression , as described previously ( Kang et al . , 2009; Shakya et al . , 2011 ) . Oct1 also contributes to eventual lineage-specific developmental gene activation ( solid red line ) , and alternate developmental lineage gene repression ( including trophectoderm , solid red block line ) . DOI: http://dx . doi . org/10 . 7554/eLife . 20937 . 030 To test the hypothesis that Oct1 occupies Oct4 targets as cells differentiate and Oct4 is lost , we performed ChIP-qPCR timecourses using differentiating ESCs and antibodies against Oct1 and Oct4 . Material was collected from 0 , 2 , 4 , 6 , 8 , 10 , 12 and 14 d of differentiation with RA . We chose a gene , Hoxc5 , that contains a conserved perfect octamer sequence ( Figure 8D ) , but is not an Oct1 target based on ChIPseq ( Figure 8E ) . Hoxc5 also shows poor induction in upon RA-mediated differentiation of Pou2f1Δ/Δ cells ( Figure 6—figure supplement 2A ) . Oct1 ChIP-qPCR revealed no binding in ESCs , as expected based on the ChIPseq ( Figure 8F ) , however robust binding was transiently observed at 6 d . By 14 d of differentiation Oct1 binding was again undetectable . We also examined a target region between the linked Myf5 and Myf6 ( Mrf4 ) loci on chromosome 10 that contains several near-perfect octamer sites ( not shown ) , and is strongly bound by Oct4 but not Oct1 ( Figure 8E ) . Oct1 inducibly occupied this region even more rapidly ( 2 d ) as Oct4 binding was lost , and in this case Oct1 binding was maintained , at varying levels , during ESC differentiation ( Figure 8F ) . Finally , we studied two additional genes , Pou5f1 and Rest ( Nrsf ) , which also contain conserved perfect octamer sequences in their regulatory regions and also show exclusive Oct4 binding . These genes are both expressed in ESCs and silenced as differentiation proceeds . Here early Oct1 binding was identified , which was maintained at low levels during the differentiation timecourse in the case of Pou5f1 but transient in the case of Rest ( Figure 8F ) . These results indicate a highly dynamic interplay between Oct1 and Oct4 in differentiating cells . Our results indicate that Oct1-deficient ESCs are unperturbed in terms of morphology , growth , metabolism , and gene expression . EBs formed from these cells are microscopically normal at early timepoints and express genes associated with all three germ layers . However , Oct1-deficient ESCs show phenotypic and molecular defects upon differentiation . These cells fail to form neurons and cardiomyocytes , generate smaller and less differentiated teratomas , and fail to contribute to adult mouse tissues . Prior work has shown that partial knockdown of Oct1 also inhibits neuron formation in the context of knockout of the related protein Oct2 ( Theodorou et al . , 2009 ) . Three molecular defects manifest upon differentiation of Oct1-deficient ESCs . First , loss of Oct1 results in a failure to fully induce genes associated with a given developmental lineage . Second , Oct1 is necessary for the repression of alternative embryonic developmental lineages . As a result , upon differentiation gene expression programs are marked not only by poor induction of lineage-appropriate gene expression , but also by ectopic expression of genes specific to alternative lineages ( Figure 8G ) . The third defect is mis-expression of genes associated with extra-embryonic lineages . These genes are normally under tight repression in ESCs and their differentiated progeny . For example , Prl8a6 and Prl3b1 are mis-expressed in RA-differentiated ESCs . Other genes such as Cdx2 and Hand1 are examples of genes expressed both in extra-embryonic and alternative embryonic lineages . For example , the Cdx2 promoter contains a perfect consensus octamer element and is a known Oct1 target in somatic cells ( Jin and Li , 2001; Wang et al . , 2009 ) . In the early embryo , Cdx2 promotes trophectoderm fate and is under tight repression by Oct4 ( Yeap et al . , 2009; Yuan et al . , 2009 ) . Later in development , Cdx2 is induced in the endoderm-derived developing gastrointestinal tract ( Guo et al . , 2004; Lu et al . , 2008 ) and during primitive hematopoiesis ( Wang et al . , 2008 ) , but is not widely expressed in ectoderm ( Suh and Traber , 1996 ) . In RA-differentiated cells , Cdx2 thus represents both a lineage-inappropriate gene and an extra-embryonic lineage . Cdx2 is mis-expressed following RA-mediated differentiation of both germline and inducible-conditional Oct1-deficient ESCs . Interestingly Oct1 may execute the opposite function in extra-embryonic tissue , as germline Oct1-deficient mice show defects in extra-embryonic tissues including poor expression of Cdx2 ( Sebastiano et al . , 2010 ) . ChIPseq experiments reveal that Oct1 and Oct4 regulate common and distinct targets in ESCs . These differences in bound targets lead to functional consequences , as the two proteins recruit different cofactors such as Jmjd1a in the case of Oct1 and Jmjd1c in the case of Oct4 ( Shakya et al . , 2015a , 2011 ) . Oct4 occupies a large group of >5000 genes , including developmentally poised genes such as Hoxc5 and Myf5 , and core pluripotency genes such as Pou5f1 and Nanog . Oct1 does not occupy these genes in ESCs , consistent with the ability of Oct1-deficient ESCs to maintain pluripotency . Instead Oct1 co-occupies a cohort of 325 genes with Oct4 that are highly enriched for a motif known as a MORE ( Reményi et al . , 2001; Tomilin et al . , 2000 ) . Oct proteins are known to homo- and hetero-dimerize ( Kang et al . , 2009; Tantin et al . , 2008; Tomilin et al . , 2000; Verrijzer et al . , 1992 ) . The configuration of Oct proteins can determine cofactor association and hence regulatory output ( Reményi et al . , 2001; Tomilin et al . , 2000 ) . Many of these constitutively co-bound genes were previously shown to become occupied by Oct1 upon oxidative stress exposure in differentiated cells lacking Oct4 ( Kang et al . , 2009 ) . The function of Oct1 at these genes is to insulate them against inhibition by oxidative stress . Fibroblasts lacking Oct1 show inappropriate repression of MORE-containing genes following H2O2 exposure ( Kang et al . , 2009; Shakya et al . , 2011 ) . We demonstrate the identical phenotype using Oct1-deficient ESCs . Oct1 also exclusively associates with a small number ( ~200 ) of other genes including Taf12 , which contains another binding site variant known as a TMFORE ( Kang et al . , 2009 ) . Notably , in undifferentiated cells Oct1 does not associate with developmental targets that become deregulated upon differentiation of Oct1-deficient ESCs . Oct4 is present at higher levels in ESCs compared to Oct1 , suggesting that mass action may contribute to the lack of Oct1 binding . This model predicts that Oct1 would occupy these genes as Oct4 is lost during differentiation . We tested four regions bound by Oct4 but not Oct1 in ESCs , Hoxc5 , Myf5/Myf6 , Rest and Pou5f1 , predicting that Oct1 binding will manifest as cells differentiate and Oct4 is lost . In all cases , Oct1 binding was observed at one or more points during the differentiation timecourse . We propose that Oct1 transiently replaces Oct4 at many such Oct4 target genes upon differentiation , where it promotes lineage-appropriate target gene expression , and represses expression of lineage-inappropriate targets . The binding events occur during a brief but important window during which critical decisions about suppression or potentiation of lineage-specific developmental Oct4 target genes are made . Binding also occurs before many of the affected target genes are induced , suggesting that Oct1 is not the principal driver of expression of these genes , but instead establishes a chromatin context in which these genes remain poised for expression , or become permanently repressed . Of the genes tested in ChIP-qPCR RA differentiation timecourses , the lineage-appropriate Hoxc5 gene shows poor induction in upon differentiation of Oct1-deficient cells ( Figure 6—figure supplement 2A ) , Myf5 and Myf6 are mesoderm-specific and lineage-inappropriate , Rest is both pluripotency-associated and lineage-inappropriate , and Pou5f1 is more restricted to ESCs . These latter genes showed no evidence of ectopic expression . This observation can be reconciled with our model by positing that redundant mechanisms , perhaps mediated by other Oct proteins such as Oct6/Pou3f1 , enforce their repression in differentiating ESCs . The bipotential function of Oct1 is consistent with previous findings in fibroblasts and T cells ( Shakya et al . , 2011 ) . Oct1 functions are mediated in part through association with the inhibitory chromatin remodeling complex NuRD ( Shakya et al . , 2011 ) , or with Jmjd1a/KDM3A , a histone H3K9me2 lysine demethylase ( Shakya et al . , 2015b , 2011 ) . H3K9me2 also controls developmental gene induction ( Wen et al . , 2009; Zylicz et al . , 2015 ) and reprogramming to pluripotency ( Chen et al . , 2013; Sridharan et al . , 2013 ) . The ability of Oct1 to suppress genes for alternative developmental lineages is reminiscent of findings using T cells in which Oct1 suppresses alternative T cell lineage genes via inter-chromosomal communication between gene loci that execute opposing gene expression programs ( Kim et al . , 2014 ) . Oct1 interacts with CTCF ( Kim et al . , 2014 ) , helping it foster exclusive gene expression programs in T cells . More work is required to determine if Oct1 insures mutually exclusive embryonic developmental gene expression programs through similar mechanisms . All mice were C57BL/6J background . Oct1 germline-deficient ESCs were generated by intercrossing heterozygous Pou2f1-/+ mice ( Wang et al . , 2004 ) to generate a 1:2:1 ratio of Pou2f1-/-: Pou2f1-/+: Pou2f1+/+ embryonic offspring . ESCs were derived from preimplantation blastocysts and genotyped . Heterozygous ESCs were not studied further . Littermate WT ESCs lines constituted the controls for these experiments . Oct1 inducible-conditional ESCs were generated by first separately crossing mice with the Pou2f1 conditional ( floxed ) allele ( Shakya et al . , 2015b ) to the YFP reporter B6 . 129 × 1-Gt ( ROSA ) 26Sortm1 ( EYFP ) Cos/J ( Jackson labs #006148 ) and inducible cre transgenic line B6 . 129-Gt ( ROSA ) 26Sortm1 ( cre/ERT2 ) Tyj/J ( Jackson labs #008463 ) . Resulting Pou2f1fl/fl animals were intercrossed to generate embryonic Pou2f1fl/fl offspring in which LSL-YFP was expressed from one Rosa26 allele and Cre-ERT2 was expressed from the other . Parent ESCs were derived from these preimplantation blastocysts . The parent lines constituted the controls for derived 4-OHT-treated , Pou2f1Δ/Δ:YFP+ lines . Cell lines were routinely authenticated by genotyping . Mycoplasma testing was conducted regularly in-house using a previously published method ( Molla Kazemiha et al . , 2009 ) . Cells were negative throughout the study . ESCs were cultured as previously ( Shakya et al . , 2015a ) with 2i conditions: the ERK inhibitor PD0325901 ( 1 μM , LC Laboratories ) and the GSK3 inhibitor CHIR99021 ( 3 μM , LC Laboratories ) . 4-OHT ( Sigma ) was dissolved in ethanol and used at 500 nM for 24 hr . Two methods were used to generate EBs . Low-attachment dishes were used to generate WT and Oct1-deficient EBs for microscopic analysis , RT-qPCR and the generation of neurons . Briefly , ESCs were trypsinized and feeders depleted by binding to gelatin-coated dishes for 30–60 min . ESC suspensions were plated on low-attachment dishes for 5–7 days . For cardiomyocyte differentiation , the hanging drop method ( Wang and Yang , 2008 ) was used in order to generate single EBs in 96-well plates . Individual EBs were then used to generate cardiomyocyte colonies in 24-well plates . Generation of neurons was accomplished as in ( Bain et al . , 1995 ) , with modifications . Briefly , EBs were formed for 4 days using low-attachment dishes , followed by culture for a further 4 days as EBs in 0 . 1 μM RA/DMEM . After 8 days , EBs were trypsinized and cultured for 8 days in 1:1 F12:DMEM , 10 μg/mL insulin ( SAFC Biosciences ) , 5 . 5 μg/mL transferrin and 38 . 7 μM sodium selenite ( ThermoFisher ) on laminin/poly-L-lysine-coated ChamberSlides ( Corning ) . Cells in Figure 2E–F were cultured for eight additional d . For H2O2 treatment , ESCs were seeded 24 hr prior to treatment on 6-well plates with sparse feeders . Cells were treated with 2 mM H2O2 ( Sigma ) for the indicated times . Antibodies for immunoblotting were as follows: Oct4 , Santa Cruz sc-5279; Oct1 , Bethyl A301-716A + A301–171A; Nanog , GeneTex GTX100863; Sox2 , GeneTex GTX101507; GAPDH , EMD Millipore , MAB374; α-Tubulin , Santa Cruz sc-5286 . RNA was isolated using TRIzol ( Thermo Fisher , Waltham MA ) , followed by RNAeasy purification ( Qiagen ) using the RNA cleanup procedure . cDNA was synthesized using SuperScript III and random hexamers ( Thermo Fisher ) . RT-qPCR oligonucleotide primers are listed in Supplementary file 3 . The Oct1 ( Pou2f1 ) cDNA and IRES ( internal ribosomal entry site ) elements were amplified and cloned together by overlap PCR . In the first PCR , primers to the 5’ end of Oct1 containing a NotI restriction site and to the 3’ end of Oct1 that contained a 5’ extension of IRES-complementary DNA were used . The sequences were: Oct1-NotI-For: 5’-AATGAAAAAAGCGGCCGCCATGAATAATCCATCAGAAAC-3’; Oct1-Rev-IRES: 5’-TTAGGGGGGGGGGAGGGATCTTCACTGTGCCTTGGAG-3’ . In the second PCR , an IRES sequence was amplified using primers to the 5’ end of the IRES containing a 5’ extension of DNA complementary of the Oct1 3’ end , and primers to the 3’ end of the IRES containing an NdeI restriction site . The sequences were: IRES-overlap-FOR: 5’ AGATCCCTCCCCCCCCCCTAACGTTACTGGCCGAA-3’; IRES-Rev-NdeI: 5’- GGGAATTCCATATGTGTGGCCATATTATCATCGTGT-3’ . The third PCR used as a template the PCR products from the first two rounds , along with the Oct1-NotI-For and IRES-Rev-NdeI primers . This process generated a DNA fragment containing an Oct1 cDNA fused to an IRES at the 3’ end , along with a NotI site at the 5’ terminus and an NdeI site at the 3’ terminus . The fragment was cloned into the optimized , self-inactivating , nonreplicative pHAGE lentiviral vector using the NotI and NdeI restriction sites . To insert a Puro cassette after the IRES , the cDNA was amplified using primers containing 5’ NdeI and 3’ ClaI restriction sites . The sequences were Puro-NdeI-For: 5’- GGAATTCCATATGATGACCGAGTACAAGCCCACGGT-3’; Puro-ClaI-Rev: 5’ GGTTTATCGATTCAGGCACCGGGCTTGC-3’ . Because the IRES apparently attenuated expression of the Puro resistance cassette in this vector , puromycin selection was performed at 0 . 75 μg/mL . To generate an empty vector control , the vector was cut with NdeI and NotI , filled in with Klenow fragment , and re-ligated . Immunofluorescence was performed as described previously ( Kang et al . , 2013 ) , using mouse monoclonal antibodies against β-tubulin-III ( R and D Systems MAB1195 ) and rabbit polyclonal antibodies against YFP ( Life Technologies A6455 ) . Secondary antibodies used were goat anti-rabbit-Alexa568 ( Life Technologies A-11011 ) and goat anti-mouse-Alexa488 ( Life Technologies A-11001 ) . Teratomas were generated as described ( Nelakanti et al . , 2015 ) by injecting parent or KO ESCs into contralateral flanks of female NCr Nude mice ( NCRNU-F , Taconic ) . Mice were sacrificed at four wk . Tumors were excised , washed with cold PBS and weighed . 1/3 of the excised tumor was used to make lysates for protein analysis using a Dounce homogenizer with RIPA lysis buffer ( 50 mM Tris pH 7 . 4 , 150 mM NaCl , 0 . 1% SDS , 0 . 1% sodium deoxycholate , 1 mM EDTA and protease inhibitors [Roche] ) on ice . Lysates were centrifuged 10 , 000 × g for 10 min . Supernatant protein concentrations were normalized using Bradford assays . 6× Laemmli sample buffer was added . The mixture was boiled for 5 min and resolved using a 10% SDS-PAGE gel . The remainder of the tumor was fixed in formaldehyde , paraffin-embedded , sectioned and H and E stained for histological analysis by a blinded pathologist . RNA was prepared from three independent cultures of undifferentiated or 14 d RA-differentiated parent Pou2f1fl/fl or 4-OHT treated Pou2f1Δ/Δ ESCs . Concentration was determined using a Quant-iT RNA assay kit and a Qubit fluorometer ( Thermo Fisher ) . Intact poly ( A ) RNA was purified from total RNA samples ( 100–500 ng ) with oligo ( dT ) magnetic beads , and stranded mRNA sequencing libraries were prepared as described using the Illumina TruSeq mRNA library preparation kit . Purified libraries were qualified on an Agilent Technologies 2200 TapeStation using a D1000 ScreenTape assay . Molarity of adapter-modified molecules was defined by qPCR using the Kapa Biosystems Library Quant Kit . Individual libraries were normalized to 10 nM and equal volumes were pooled in preparation for Illumina sequencing . Sequencing libraries ( 25 pM ) were chemically denatured and applied to an Illumina HiSeq v4 paired end flow cell using an Illumina cBot . Hybridized molecules were clonally amplified and annealed to sequencing primers with reagents from an Illumina HiSeq PE Cluster Kit v4-cBot . Following transfer of the flowcell to an Illumina HiSeq 2500 instrument ( HCS v2 . 2 . 38 and RTA v1 . 18 . 61 ) , a 125-cycle paired-end sequence run was performed using HiSeq SBS Kit v4 sequencing reagents . Fastq data quality were checked using Fastqc verision 0 . 10 . 1 ( http://www . bioinformatics . babraham . ac . uk/projects/fastqc/ ) . Quality scores dipped towards the 3’ end of the reads , so reads were trimmed at 50 bases to eliminate poor-quality data . The resulting 50-base reads were aligned to the mouse mm10 genome ( GRCm38 , December 2011 ) plus splice junctions using novoalign version 2 . 08 . 01 ( http://www . novocraft . com ) . Alignments to splice junctions were translated back to genome coordinates using the SamTranscriptomeParser application in the USeq package ( Nix et al . , 2010 ) . Aligned reads were quality checked using the Picard tools’ CollectRnaSeqMetrics command ( https://broadinstitute . github . io/picard/ ) . On average 99 . 0% of the reads aligned to the mouse genome , with 78% of reads providing unique alignments , and 86% of reads providing alignments to protein coding and UTR regions of the genome . Tests for differential gene expression were performed with DESeq2 , version 1 . 10 . 0 ( Love et al . , 2014 ) . Genes with a count of at least 50 in one or more samples were tested . Genes showing at least 2 . 5-fold change of expression at an adjusted p-value of <0 . 01 were selected as differentially expressed . Figures were generated in R version 3 . 2 . 3 ( http://www . r-project . org ) using functions from the gdata and gplots libraries . ChIP was performed as described ( Shakya et al . , 2015a ) . ChIP oligonucleotide primers are listed in Supplementary file 3 . Antibodies used were the following: Oct1 ( Bethyl , a mixture of A301-716A + A301–717A ) , Oct4 ( Santa Cruz , sc-8629 ) and H3K4me3 ( Millipore , 07–473 ) . ChIPseq was performed as described previously ( Shakya et al . , 2015a , 2015b ) , using a single IP per condition and clones of parent or derived Oct1-deficient ESCs . For ChIPseq , reads were aligned to the mouse reference genome ( mm10 ) with the Burrows-Wheeler Aligner ( BWA , version: 0 . 5 . 9 ) . Reads were filtered for alignment quality of >Q10 and duplicates were removed using Picard tools ( function MarkDuplicates ) . After filtering there were 21 . 1 ( H3K4me3 ) , 27 . 3 ( Oct1 ) , and 23 . 7 ( Oct4 ) million reads . MACSv2 peak caller ( version: 2 . 1 ) was used to call ChIPseq regions of enrichment with the following parameters ( -p 1e-5 --nomodel --shiftsize <fragment_length/2> for Oct1 , Oct4 and -p 1e-2 --broad for H3K4me3 ) . To estimate the --shiftsize parameter ( predominant fragment length divided by 2 ) we performed strand cross-correlation analysis using SPP R package ( version: 1 . 10 . 1 ) with default parameters . Peaks overlapping with ENCODE blacklisted regions were filtered using BEDtools ( function itersectBed ) . We also discarded peaks localized to mitochondria , chromosome Y , and unmapped contigs . After filtering we had 692 ( Oct1 ) , and 8673 ( Oct4 ) peaks . Signal to noise ratio was assessed by calculating normalized strand coefficient ( NSC ) and relative strand correlation ( RSC ) using the SPP R package with default parameters ( version: 1 . 10 . 1 ) . The obtained values of NSC and RSC ( H3K4me3: 2 . 28 , 1 . 25; Oct1: 1 . 02 , 1 . 45; Oct4: 1 . 05 , 2 . 32 ) indicate highly enriched datasets with large fragment-length peak as compared to read-length peak . The NSC value for Oct1 transcription factor was somewhat smaller but typical for high quality datasets generated for factors with small numbers of genuine binding sites ( 692 MACS2-identified peaks for Oct1 ) . We used MACSv2 function bdgdiff to build fold-enrichment signal tracks for all positions in the genome . Signal tracks were converted to TDF files using igvtools ( https://www . broadinstitute . org/igv/igvtools ) . Peaks were allocated to genes using the annotatePeaks . pl program from HOMER suite ( Hypergeometric Optimization of Motif Enrichment , version: 4 . 7 , http://homer . salk . edu/homer/ ) by determining the closest RefSeq transcription start sites of the genes to the peaks . Functional enrichment analysis was performed using the findGO . pl program from HOMER and Bonferroni as well as Benjamini and Hochberg correction for multiple testing corrections . Robustness of the analysis was confirmed using MEME-ChIP ( Machanick and Bailey , 2011 ) , which generate highly similar motifs . Transcription factor enrichment within ChIPSeq peaks ( de novo motif discovery and known motif matching ) was determined using findMotifsGenome . pl program from HOMER . Motif analysis was run on overlapped and separately on unique Oct1 and Oct4 ChIPseq peaks . Oct1 and Oct4 ChIPseq peak overlaps were defined by requiring the distance between peak summits to be ≤100 bp . Motif lengths of 6–24 bp were identified within 200 bp regions centered on peak summits and an option of random background was selected for motif discovery .	Humans and most other animals are composed of hundreds of different types of cell , including nerve cells , muscle cells and blood cells . Despite performing many different roles , these cells all develop from a single fertilized egg , which divides to make a particular group of cells that when studied in the laboratory are called embryonic stem cells ( or ESCs for short ) . The ability of a cell to become a different cell type is defined as “potency” . ESCs are unique because they can specialize into any type of cell present in the adult organism , and they are therefore called “pluripotent” . However , as the embryo develops , its ESCs gradually lose their potency , and become more and more specialized . The activity of a great number of genes must be regulated during the transition from pluripotent to specialized cells , and some of the mechanisms involved in this transition are still unclear . ESCs are known to need a gene-regulating protein called Oct4 to remain pluripotent and Shen , Kang , Shakya et al . now show that a similar protein named Oct1 is essential for their transition to becoming more specialized . When the gene for Oct1 was deleted from mouse ECSs , they behaved largely like “normal” ESCs , but could not properly mature into certain cell types such as heart and nerve cells . Molecular analyses revealed that Oct4 and Oct1 compete to regulate the activity of many common genes with opposing outcomes: Oct4 keeps ESCs pluripotent while Oct1 leads them to specialize . The Oct4 protein is abundant in ESCs and prevails over Oct1 , but as the cells mature , the levels of Oct4 drop , and Oct1 takes over in the regulation of their common target genes . Going forward , a better understanding of how ESCs become specialized will help basic research in the laboratory and allow scientists to tackle new questions about how the human body develops and how our organs work . In the longer-term , these findings might also have applications in the field of regenerative medicine , which aims to repair or replace a person’s cells , tissues or organs to improve their health .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "chromosomes", "and", "gene", "expression", "developmental", "biology" ]	2017	Enforcement of developmental lineage specificity by transcription factor Oct1
Microbial communities routinely have several possible species compositions or community states observed for the same environmental parameters . Changes in these parameters can trigger abrupt and persistent transitions ( regime shifts ) between such community states . Yet little is known about the main determinants and mechanisms of multistability in microbial communities . Here , we introduce and study a consumer-resource model in which microbes compete for two types of essential nutrients each represented by multiple different metabolites . We adapt game-theoretical methods of the stable matching problem to identify all possible species compositions of such microbial communities . We then classify them by their resilience against three types of perturbations: fluctuations in nutrient supply , invasions by new species , and small changes of abundances of existing ones . We observe multistability and explore an intricate network of regime shifts between stable states in our model . Our results suggest that multistability requires microbial species to have different stoichiometries of essential nutrients . We also find that a balanced nutrient supply promotes multistability and species diversity , yet make individual community states less stable . Recent metagenomics studies revealed that microbial communities living in similar environments are often composed of rather different sets of species ( Zhou et al . , 2007; Lahti et al . , 2014; Lozupone et al . , 2012; Zhou et al . , 2013; Pagaling et al . , 2017; Gonze et al . , 2017 ) . It remains unclear to what extent such alternative species compositions are deterministic as opposed to being an unpredictable outcome of communities’ stochastic assembly . Furthermore , changes in environmental parameters may trigger abrupt and persistent transitions between alternative species compositions ( Shade et al . , 2012; Rocha et al . , 2018; Scheffer and Carpenter , 2003 ) . Such transitions , known as ecosystem regime shifts , significantly alter the function of a microbial community and are difficult to reverse . Understanding the mechanisms and principal determinants of alternative species compositions and regime shifts is practically important . Thus , they have been extensively studied over the past several decades ( Sutherland , 1974; Holling , 1973; May , 1977; Tilman et al . , 1997; Schröder et al . , 2005; Fukami and Nakajima , 2011; Bush et al . , 2017 ) . Growth of microbial species is affected by many factors , with availability of nutrients being among the most important ones . Thus , the supply of nutrients and competition for them plays a crucial role in determining the species composition of a microbial community . The majority of modeling approaches explicitly taking nutrients into account are based on the classic MacArthur consumer-resource model and its variants ( MacArthur and Levins , 1964; MacArthur , 1970; Huisman and Weissing , 1999; Tikhonov and Monasson , 2017; Posfai et al . , 2017; Goldford et al . , 2018; Goyal et al . , 2018; Butler and O'Dwyer , 2018 ) . This model assumes that every species co-utilizes several substitutable nutrients of a single type ( e . g . carbon sources ) . However , nutrients required for growth of a species exist in the form of several essential ( non-substitutable ) types including sources of C , N , P , Fe , etc . Real ecosystems driven by competition for multiple essential nutrients have been extensively experimentally studied ( see recent papers; Fanin et al . , 2015; Browning et al . , 2017; Camenzind et al . , 2018 and references therein ) . The theoretical foundation for all existing consumer-resource models capturing this type of growth has been laid in Tilman ( 1982 ) , where a model with two essential resources has been introduced and studied . Future studies extended Tilman’s approach to three and more essential resources , where it has been shown to sometimes result in oscillations and chaos ( Huisman and Weissing , 1999; Huisman and Weissing , 2001; Shoresh et al . , 2008 ) . However , all the previously studied models accounted for just a single metabolite per each essential nutrient . Here , we introduce and study a new consumer-resource model of a microbial community supplied with multiple metabolites of two essential types ( e . g . C and N or N and P ) . This ecosystem is populated by microbes selected from a fixed pool of species . We show that our model has a very large number of possible steady states classified by their distinct species compositions . Using game-theoretical methods adapted from the well-known stable marriage ( or stable matching ) problem ( Gale and Shapley , 1962; Gusfield and Irving , 1989 ) , we predict all these states based only on the ranked lists of competitive abilities of individual species for each of the nutrients . We further classify these states by their dynamic stability , and whether they could be invaded by other species in our pool . We then focus our attention on a set of steady states that are both dynamically stable and resilient with respect to species invasion . For each state , we identify its feasibility range of all possible environmental parameters ( nutrient supply rates ) for which all of state’s species are able to survive . We further demonstrate that for a given set of nutrient supply rates , more than one state could be simultaneously feasible , thereby allowing for multistability . While the overall number of stable states in our model is exponentially large , only very few of them can be realized for a given set of environmental conditions defined by nutrient supply rates . The principal component analysis of predicted microbial abundances in our model shows a separation between the alternative stable states reminiscent of real-life microbial ecosystems . We further explore an intricate network of regime shifts between the alternative stable states in our model triggered by changes in nutrient supply . Our results suggest that multistability requires microbial species to have different stoichiometries of two essential resources . We also find that well-balanced nutrient supply rates matching the average species’ stoichiometry promote multistability and species diversity yet make individual community states less structurally and dynamically stable . These and other insights from our consumer-resource model may help to understand the existing data and provide guidance for future experimental studies of alternative stable states and regime shifts in microbial communities . Our consumer-resource model describes a microbial ecosystem colonized by microbes selected from a pool of S species . Growth of each of these species could be limited by two types of essential resources , to which we refer to as ‘carbon’ and ‘nitrogen’ . In principle , these could be any pair of resources essential for life: C , N , P , Fe , etc . A generalization of this model to more than two types of essential resources ( e . g . C , N and P ) is straightforward . Carbon and nitrogen resources exist in the environment in the form of K distinct metabolites containing carbon , and M other metabolites containing nitrogen . For simplicity , we ignore the possibility of the same metabolite providing both types . We further assume that each of the S species in the pool is a specialist , capable of utilizing only a single pair of nutrients , that is one metabolite containing carbon and one metabolite containing nitrogen . We assume that for given environmental concentrations of all nutrients , a growth rate of a species α is limited by a single essential resource via Liebig’s law of the minimum ( de Baar , 1994 ) : ( 1 ) gα⁢ ( c , n ) =min⁡ ( λα ( c ) ⁢c , λα ( n ) ⁢n ) . Here , c and n are the environmental concentrations of the unique carbon and nitrogen resources consumed by this species . The coefficients λα ( c ) and λα ( n ) are defined as species-specific growth rates per unit of concentration of each of two resources . They quantify the competitive abilities of the species α for its carbon and nitrogen resources , respectively . Indeed , according to the competitive exclusion principle , if two species are limited by the same resource , the one with the larger value of λ wins the competition . Note that according to Liebig’s law , if the carbon source is in short supply so that λα ( c ) ⁢c<λα ( n ) ⁢n , it sets the value for this species growth rate . We refer to this situation as c-source limiting the growth of the species α . Conversely , when λα ( c ) ⁢c>λα ( n ) ⁢n , the n-source is limiting the growth of this species . Thus , each species always has exactly one growth-limiting resource and one non-limiting resource . In our model , microbes grow in a well-mixed chemostat-like environment subject to a constant dilution rate δ ( see Figure 1A for an illustration ) . The dynamics of the population density , Bα , of a microbial species α is then governed by: ( 2 ) d⁢Bαd⁢t=Bα⁢[gα⁢ ( ci , nj ) -δ] , where ci and nj are the specific pair of nutrients defining the growth rate gα of this species according to the Liebig’s law ( Equation 1 ) . These nutrients are externally supplied at fixed rates ϕi ( c ) and ϕj ( n ) and their concentrations follow the equations: ( 3 ) dcidt=ϕi ( c ) −δ⋅ci−∑allαusingciBαga ( ci , nj ) Ya ( c ) , dnidt=ϕi ( n ) −δ⋅ni−∑allαusingniBαga ( ci , nj ) Ya ( n ) Here , Yα ( c ) and Yα ( n ) are the growth yields of the species α on its c- and n-resources respectively . Yields quantify the concentration of microbial cells generated per unit of concentration of each of these two consumed resources . The yield ratio Yα ( n ) /Yα ( c ) determines the unique C:N stoichiometry of each species . A steady state of the microbial ecosystem can be found by setting the right hand sides of Equations 2-3 to zero and solving them for environmental concentrations of all nutrients ci , and nj , and abundances Bα of all species . We choose to label all possible steady states by the list of species present in the state and by the growth-limiting nutrient ( cor n ) for each of these species . Thus , two identical sets of species , where at least one species is growth limited by a different nutrient are treated as two distinct states of our model . Conversely , our definition of a steady state does not take into account species’ abundances . Examples of such states in a system with two carbon , two nitrogen nutrients and four species ( one species for every pair of carbon and nitrogen nutrients ) with specific values of species’ competitive abilities λα ( c ) and λα ( n ) and yields Yα ( c ) and Yα ( n ) ( see Supplementary files 1 , 2 for their exact values ) are shown in Figure 1B . For the sake of brevity we refer to this model as 2C × 2N × 4S . Because each of the S species in the pool could be absent from a given state , or , if present , could be limited by either its c- or its n-resource , the theoretical maximum of the number of distinct states is 3S ( equal to 81 in our 2C × 2N × 4S example ) . However , the actual number of possible steady states is considerably smaller ( equal to 34 in this case ) . Indeed , possible steady states in our model are constrained by a variant of the competitive exclusion principle ( Gause , 1932 ) ( see Materials and methods for details ) . One of the universal consequences of this principle is that the number of species present in a steady state of any consumer-resource model cannot exceed K + M − the total number of nutrients . We greatly simplified the task of finding all steady states in our model by the discovery of the exact correspondence between our system and a variant of the celebrated stable matching ( or stable marriage ) problem in game theory and economics ( Gale and Shapley , 1962; Gusfield and Irving , 1989 ) . The matching in our model connect pairs of C and N resources via microbial species using both of them . Unlike in the traditional stable marriage model , a given resource can be involved in more than one matching but cannot be limiting for more than one microbe . Thus , the competitive exclusion principle provides a number of constraints on the set of possible matchings and their stability , which are described in detail in Materials and methods and Appendix 3 . Each of the steady states identified in the previous chapter can be realized only for a certain range of nutrient supply rates . These ranges can be calculated using the steady state solutions of Equations 2 , 3 , governing the dynamics of microbial populations and nutrient concentrations , respectively ( see Materials and methods ) . Among all formal mathematical solutions of these equations we select those , where populations of all species and all nutrient concentrations are non-negative . This imposes constraints on nutrient supply rates , thereby determining their feasible range for a given steady state ( shown in green in Figure 1C ) . The volume of such feasible range has been previously used to quantify the so-called structural stability of a steady state ( Rohr et al . , 2014; Grilli et al . , 2017; Butler and O'Dwyer , 2018 ) . States with larger feasible volumes generally tend to be more resilient with respect to fluctuations in nutrient supply . Stability of a steady state could be also disturbed by a successful invasion of a new species ( see Figure 1D ) . We can test the resilience of a given state in our model with respect to such invasions . A state is called uninvadable if none of the other species from our pool can survive in the environment shaped by the existing species . Figure 1B shows all seven states that are uninvadable in our variant of the 2C × 2N × 4S model . Whether or not a given state is uninvadable is determined by the specific choice of parameters λα ( c ) , λα ( n ) . For example , for parameters listed in the Supplementary file 1 the state in which B12 is limited by carbon c1 , and B22 - by carbon c2 could be invaded by the species B11 . Indeed , λ ( c ) of B11 is larger than that of B12 , and the nitrogen concentration n1 is not limited by any species . Hence , this state is not shown in Figure 1B . However , the same state may turn out to be uninvadable for a different combination of parameters . The one-to-one correspondence between our model and a variant of the stable matching problem ( Gale and Shapley , 1962 ) allows us to identify all uninvadable steady states for a given choice of λα ( c ) , λα ( n ) describing species competitiveness for resources ( see Materials and methods and Appendix 3 ) . Note that , in the regime of our model , where the supply of all nutrients is high , that is ϕ ( c , n ) ≫δ2/λα ( c , n ) , invadability of individual states does not depend on supply rates . Indeed , in this regime the outcome of an attempted invasion is fully determined by the competition between species , which in turn depends only on the rank-order of competitive abilities λ of the invading species relative to the species currently present in the ecosystem ( see Materials and methods for details ) . In addition to structural and invasion types of stability described above , there is also a notion of dynamic stability of a steady state actively discussed in the ecosystems literature ( see e . g . May , 1972; Allesina and Tang , 2012; Butler and O'Dwyer , 2018 ) . Dynamic stability can be tested by exposing a steady state to small perturbations in populations of all species present in this state ( see Figure 1E ) . The state is declared dynamically stable if after any such disturbance the system ultimately returns to its initial configuration ( see Materials and methods for details of the testing procedure used in our study ) . We classify all the steady states in our model according to these three types of stability . The example of this classification for our realization of 2C × 2N × 4S model is summarized in Figure 1F . Note , that in general , one type of stability does not imply another . Out of 34 possible steady states realized for different ranges of nutrient supply rates there are only seven uninvadable ones . In the 2C × 2N × 4S model only one of the states ( labelled seven in Figure 1B ) turned out to be dynamically unstable , while for the remaining 33 states small perturbations of microbial abundances present in the state do not trigger a change of the state . Unlike two other types of stability , the structural stability has a continuous range . It could be quantified by the fraction of all possible combinations of nutrient supply rates for which a given state is feasible ( referred to as state’s normalized feasible range ) . We estimated the normalized feasible ranges of all states in the 2C × 2N × 4S model using a Monte Carlo procedure described in Materials and methods . The results are reflected in the second column of Figure 1F , where a structurally stable state is defined as that whose normalized feasible range exceeds 0 . 1 ( an arbitrary threshold ) . In general we find that normalized feasible ranges of uninvadable states in our model have a broad log-normal distribution ( see Figure 1—figure supplement 1 for details ) . It is natural to focus our attention on steady states that are simultaneously uninvadable and dynamically stable . Indeed , such states correspond to natural endpoints of the microbial community assembly process . They would persist for as long as the nutrient supply rates do not change outside of their structural stability range . Therefore , they represent the states of microbial ecosystems that are likely to be experimentally observed . From now on , we concentrate our study almost exclusively on those states and refer to them simply as stable states . The feasible ranges of nutrient supply of different stable states may or may not overlap with each other ( see Figure 2A–B for a schematic illustration of two different scenarios ) . Whenever feasible ranges of two or more states overlap ( see Figure 2B ) - multistability ensues . Note that the states in the overlapping region of their feasibility ranges constitute true alternative stable states defined and studied in the ecosystems literature ( Sutherland , 1974; Holling , 1973; May , 1977; Fukami and Nakajima , 2011; Bush et al . , 2017 ) . The existence of alternative stable states goes hand-in-hand with regime shifts manifesting themselves as large discontinuous and hysteretic changes of species abundances ( Scheffer and Carpenter , 2003 ) . Every pair of states with overlapping feasibility ranges in our model corresponds to a possible regime shift between these states as illustrated in Figure 2D ( note abrupt changes in the population B11 at the boundary of the overlapping region ) . In general , a discontinuous regime shift happens in our model when one of the species ( B12 in this example ) changes its growth-limiting nutrient thereby making the state invadable . It is then promptly invaded by the species present in the new state ( B11 and B22 in our example ) which may lead to immediate changes in populations of multiple species . Conversely , when feasible ranges of a pair of states do not overlap with each other but share a boundary ( Figure 2A ) , the transition between these states is smooth and non-hysteretic ( Figure 2C ) . It manifests itself in continuous changes in abundances of all microbial species at the boundary between states . Such continuous transitions happen in our model when the growth rate of one of the species ( B21 in this example ) falls below the dilution rate δ . This species then slowly disappears from the ecosystem thereby changing its state . When this boundary is crossed in the opposite direction , the same species ( B21 ) gradually appears in the ecosystem . As expected for regime shifts , dynamically unstable states always accompany multistable regions in our model ( Scheffer and Carpenter , 2003 ) ( see below for a detailed discussion of the interplay between multistability and dynamically unstable states ) . We observed that dynamically unstable state #7 in our 2C × 2N × 4S is feasible in the overlapping region between states #1 and #2 in Figure 2B . The population B11 in this state is shown as dashed line in Figure 2D . We identified all possible regime shifts in the 2C × 2N × 4S model by systematically looking for overlaps between the feasible ranges of nutrient supply of all six uninvadable dynamically stable states . These regime shifts can be represented as a network in which nodes correspond to community’s stable states and edges connect states with partially overlapping feasible ranges ( see thick red edges in Figure 2E ) . One can see that regime shifts are possible only for of nine pairs of uninvadable states . We performed additional simulations ( see Materials and mthods ) looking for shared boundaries ( continuous transitions ) between uninvadable states and identified additional four pairs of states bordering each other ( thin blue edges in Figure 2E ) . The pairs of states #5 - #6 and #3 - #4 do not directly transition to each other either continuously or discontinuously . This indicates that their feasible ranges are too far apart from each other , so that they do not have any overlaps or common boundaries . Combining the information in Figure 1B and Figure 2E one can find that all states connected by a discontinuous regime shift in our 2C × 2N × 4S model have two distinct sets of keystone species: B11-B22 in one state and B12-B21 in another . This is because all regime shifts are driven by the same bistable switch in which these pairs of species compete and mutually exclude each other . The dynamically unstable state #7 is formed by the union of all four keystone species and , when perturbed , collapses into a state with either one or another keystone set . Conversely , states connected by a continuous transition share the same pair of keystone species . One of the ‘satellite’ species , that is species distinct from the keystone , gradually goes extinct when the boundary between these states is crossed . When the nutrient supply is changed in the opposite direction this species gradually invades the system . Figure 2F shows a much larger network of 8633 regime shifts between 893 uninvadable dynamically stable states in the 6C × 6N × 36S realization of our model . In this model the microbial community is supplied with six carbon and six nitrogen nutrients and colonized from a pool of 36 microbial species ( one for each pair of C and N nutrients ) ( see Supplementary files 3 , 4 , 5 , 6 for the values of λ’s and yields ) . For simplicity , we did not show the remaining 165 uninvadable stable states that have no possible regimes shifts to any other states . The size of a node is proportional to its degree ( i . e . the total number of other states it overlaps with ) ranging between 1 and 164 with average around 20 ( the degree distribution is shown in Figure 2—figure supplement 1 ) . The network modularity analysis ( see Materials and methods for details ) revealed seven network modules indicating that pairs of states that could possibly undergo a regime shift are clustered together in the multi-dimensional space of nutrient supply rates . This modular structure suggests the existence of distinct sets of keystone species driving regime shifts within each module . However , the complexity of the 6C × 6N × 36S model does not allow a straightforward identification of these drivers ( paired sets of keystone species and dynamically unstable states ) . In a general case , the number of stable states that are simultaneously feasible for a given set of nutrient supply rates can be more than two . Furthermore , as the number of nutrients increases , the multistability with more than two stable states becomes progressively more common . In Figure 3A , we quantify the frequency of multistability with V stable states occur in our 6C × 6N × 36S model across all possible nutrient supply rates ( see Materials and methods for details of how this was estimated ) . V-1 approximately follows a Poisson distribution ( dashed line in Figure 3A ) with λ=0 . 063 . Note that for some supply rates up to five stable states can be simultaneously feasible . However , the probability to encounter such cases is exponentially small . We further explored the factors that determine whether multistability is possible in resource-limited microbial communities . Like in a simple special case of regime shift between two microbial species studied in Tilman ( 1982 ) , multistability in our model is only possible if individual microbial species have different C:N stoichiometry . This stoichiometry is given by the ratio of species’ nitrogen and carbon yields . Our numerical simulations and mathematical arguments show that when all species have exactly the same stoichiometry Yα ( n ) /Yα ( c ) , there is no multistability or dynamical instability in our model ( see Appendix 5 ) . That is to say , in this case for every set of nutrient supply rates the community has a unique uninvadable state , and all these states are dynamically stable . A complementary question is whether multistable states are more common around particular ratios of carbon and nitrogen supply rates . Figure 3B shows this to be the case: the likelihood of multistability has a sharp peak around the well-balanced C:N nutrient supply rates . In this region multiple stable states are present for roughly 15% of nutrient supply rate combinations . Note that the average C:N stoichiometry of species in our model is assumed to be 1:1 . In case of an arbitrary C:N stoichiometry , by redefining the units of nutrient concentrations and supply rates one can transform any ecosystem to have 1:1 nutrient ratio . Hence , in general , we predict that the highest chance to observe multistability will be when the ratio of nutrient supply rates is close to the average C:N stoichiometry of species in the community . To illustrate how multistable states manifest themselves in a commonly performed Principal Component Analysis ( PCA ) of species’ relative abundances , we picked the environment with V=5 simultaneously feasible stable states in our 6C × 6N × 36S model . In natural environments , nutrient supply usually fluctuates both in time and space . To simulate this we sampled a ±10% range of nutrient supply rates around this chosen environment ( see Materials and methods ) and calculated species’ relative abundances in each of the uninvadable states feasible for a given nutrient supply . To better understand the relationship between dynamically stable and unstable states we included the latter in our analysis . Figure 3C shows the first vs the second principal components of relative microbial abundances sampled in this fluctuating environment . ( two more examples calculated for different multistable neighborhoods are shown in Figure 3—figure supplement 1A–B ) . One can see five distinct clusters , each corresponding to a single dynamically stable uninvadable state . Interestingly , in the PCA plot these states are separated by V-1=4 dynamically unstable ones . Furthermore , all states are aligned along a quasi-1D manifold with an alternating order of stable and unstable states . It is tempting to conjecture that some variant of our model may explain similar arrangements of clusters of microbial abundances , commonly seen in PCA plots of real ecosystems . If this is the case , the gaps between neighboring clusters would correspond to dynamically unstable states of the ecosystem , which may be experimentally observable as long transients in community composition . Above we demonstrated that multistable states are much more common for balanced nutrient supply rates , that is to say , when the average ratio of carbon and nitrogen supply rates matches the average C:N stoichiometry of species in the community ( see Figure 3B ) . Interestingly , a balanced supply of nutrients also promotes species diversity . In Figure 4A , we plot the average number of species in a stable state , referred to as species richness , as a function of the average balance between carbon and nitrogen supplies for 6C × 6N × 36S model . The species richness is the largest ( around 10 . 5 ) for balanced nutrient supply rates , while dropping down to the absolute minimal value of six in two extreme cases of very large imbalance of supply rates , where the nutrient supplied in excess becomes irrelevant in competition . In this case , only six species that are teh top competitors for carbon metabolites ( if nitrogen supply is plentiful ) or , respectively nitrogen metabolites ( if carbon is large ) survive , while the rest of less competitive species are never present in uninvadable states . The number of distinct community states also has a sharp peak at balanced nutrient supply ( see 3-orders of magnitude difference in Figure 4—figure supplement 1 ) . For balanced nutrient supply rates the relationship between species’ competitiveness and its prevalence in the community is much less pronounced than for imbalanced ones . It is shown in Figure 4B , where we plot the prevalence of the species as a function of its average competitiveness . Here , the average competitiveness rank of a species is defined as the mean of its ranks of competitive abilities ( λ parameters of the model ) for its carbon and nitrogen resources . The rank 1 being assigned to the most competitive species for a given resource ( the species with the largest value of λ ) , while the rank 6 - to the least competitive species for this resource . Species prevalence is given by the fraction of all environments where it can survive . Note that all 36 species in our pool are present in some of the environments . In general , more competitive species tend to survive in a larger subset of environments ( see the dashed curve in Figure 4B ) . For example , in our pool there is one species which happens to be the most competitive for both its carbon and nitrogen sources . This species is present in all of the states in every environment . However , we also find that some of the least competitive species ( those at the right end of the x-axis in Figure 4B ) survive in a broad range of environments . For example , one species with average competitiveness rank of 5 . 5 corresponding to the last and next to last rank for its two resources still has relatively high prevalence of around 20% . This illustrates complex ways in which relative competitiveness of all species in the pool shapes their prevalence in a broad range of environments . We also explore the relationship between species richness of a state ( i . e . its total number of surviving species ) and its other properties . Figure 4C shows an exponential increase of the number of uninvadable states as a function of species richness . In our 6C × 6N × 36S model all uninvadable states with less than 10 species are dynamically stable ( solid line in Figure 4C ) , while those with 10 or more species can be both stable or unstable ( dashed line in Figure 4C ) . Overall , the fraction of stable states to dynamically unstable ones decreases with species richness . In other words , the probability for a state to be dynamically unstable increases with the number of species . In this aspect , our model behaves similar to the gLV model in Robert May’s study ( May , 1972 ) . In Figure 4D , we show a negative correlation between the species richness of a stable state and its feasible range of nutrient supplies . Thus in our model the number of species in an ecosystem has detrimental effect on the structural stability of the community quantifying its robustness to fluctuating nutrient supply ( Rohr et al . , 2014 ) . The empirically observed exponential decay of state’s feasible range with its number of species is well described by a two-fold decrease per each species added ( see Serván et al . , 2018 and Grilli et al . , 2017 for related results in the gLV model ) . Note that the observed decrease in feasible range with species richness goes hand-in-hand with an increase in the overall number of states . Thus , in well-balanced environments a large number of states are carving all possible combinations of nutrient supply into many small and overlapping ranges . Overall , the results of our model with a large number of nutrients suggest the following picture . In nutrient-balanced environments , we expect to observe a high diversity of species in the existing communities . These species can form a very large number of possible combinations ( uninvadable states ) . Each of these states could be realized only for a narrow range of nutrient supply rates indicating their low structural stability . Moreover , in such environments we predict common appearance of multistability between some of these states . We find that multistability in our model requires a mix of species with different nutrient stoichiometries . In this aspect it is similar to both the Tilman model ( Tilman , 1982 ) , and the MacArthur model ( MacArthur and Levins , 1964; MacArthur , 1970; Chesson , 1990 ) . Common variants of the MacArthur model assume identical biomass yields of different species growing on a given nutrient ( Tikhonov and Monasson , 2017; Posfai et al . , 2017; Goldford et al . , 2018; Goyal et al . , 2018; Butler and O'Dwyer , 2018 ) . In this case , the absence of multistability is guaranteed by a convex Lyapunov function ( MacArthur , 1970 ) guiding any dynamical trajectory of the system to its unique minimum . However , a MacArthur model with different nutrient yields of different species is capable of multistability . For some growth functions g⁢ ( C , N ) multistability is possible even in a community of species with identical nutrient yields/stoichiometries . For example , the growth function g⁢ ( C , N ) =λ⋅C⋅N has been numerically studied in the context of autocatalytic polymer growth and shown to be capable of multistability ( Tkachenko and Maslov , 2018 ) . This model had 1:1 stoichiometry: a ligation always eliminates one left end and one right end of two polymer chains and generates one autocatalytic polymer segment . Another type of growth function with two essential resources has been shown to have bistable solutions even for identical species stoichiometries ( see Figure 5C in Marsland et al . , 2019b ) . The Minimum Environmental Perturbation Principle introduced in this study may provide additional insights on the necessary conditions for multistability in consumer resource models . Given that multistability in our model is impossible in communities of species with identical stoichiometries , it is reasonable to expect that the larger is the variation of C:N ratio of individual microbes , the higher is the likelihood to observe multistability . We investigated this question in the 2C × 2N × 4S model and summarized the results in Figure 5 . It shows that the likelihood of finding nutrient supply rates with multistability systematically decreases with standard deviation of the logarithm of species stoichiometry . We also found that about half of the combinations of species stoichiometries yielded no multistable states at all . Multistability in our model is caused by a complex interplay between species’ competitiveness abilities λ and their C:N stoichiometries Y ( n ) Y ( c ) . Appendix 4 explains why multistability is impossible for half of yield combinations for which the enumerator in Equation S11 exceeds its denominator . Nutrient stoichiometry of phytoplankton species in marine ecosystems has been known to be relatively universal with C:N:P ≃ 106:16:1 known as Redfield ratio ( Redfield , 1958 ) . Thus species-to-species variability of C:N ratio for phytoplankton is rather small with logarithmic standard deviation estimated to be around 0 . 05 based on data from Finkel et al . ( 2016 ) . In this limit , the multistability in our model is rather unlikely ( observed in ∼3% of nutrient supply combinations , see green arrow in Figure 5 ) . The likelihood of multistability is also low ( ∼1% ) for the mammalian gut microbiome , where variability of the logarithm of C:N ratio in different 'keystone’ gut species studied in Reese et al . ( 2018 ) is around 0 . 03 . The chances of multistability increase in terrestrial ecosystems such as soil , where significant deviations from the Redfield ratio have been reported ( Cleveland and Liptzin , 2007 ) . For example , using the data for the microbial species from grassland leaf litter community reported in Mouginot et al . ( 2014 ) , with log ( C:N ) variability of 0 . 12 we predict the likelihood of multistability to be around 10% . Another important factor favoring multistability in our model is the balanced supply of two essential nutrients ( see Figure 3B ) . It occurs when the average ratio of supply rates of two essential nutrients matches the average C:N stoichiometry of community’s species ( see Figure 3B ) . When nutrient supplies are balanced , microbial community multistability is relatively common . Furthermore , for balanced nutrients the community can be in one of many different states , characterized by different combinations of limiting nutrients . These states tend to have high species diversity ( Figure 4A ) – a trend consistent with lake ecosystems in Interlandi and Kilham ( 2001 ) – and relatively small range of feasible supply rates ( Figure 4D ) . Hence , regime shifts can be easily triggered by changes in nutrient supply . The balanced region is characterized by a complex relationship between species competitiveness and survival , so that even relatively poor competitors could occasionally have high prevalence ( species in the upper right corner of Figure 4B ) . In the opposite limit , the supply of nutrients of one type ( say nitrogen ) greatly exceeds that of another type ( say carbon ) . For such imbalanced supply , the community has a unique uninvadable state , where every carbon nutrient supports the growth of the single most competitive species . Nitrogen nutrients are not limiting the growth of any species and thus have no impact on species survival and community diversity . As a consequence , the average diversity of microbial communities in such nutrient-imbalanced environments is low ( about one half of that for balanced supply conditions ) . This is in agreement with many experimental studies showing that addition of high quantities of one essential nutrient ( e . g . as nitrogen fertilizer ) tends to decrease species diversity . This has been reported in numerous experimental studies cited in the chapter 'Resource richness and species diversity’ of Tilman ( 1982 ) as well as in recent experiments in microbial communities ( Mello et al . , 2016 ) . Species in our model are characterized by their competitiveness abilities λ ( c ) , λ ( n ) and nutrient yields Y ( c ) and Y ( n ) . As we showed above , the rank order of the former fully defines the total number of stable states and their invadability . On the other hand , multistability highly depends on combination of species’ nutrient yields . While in the current version of our model we did not assume any specific correlations between these parameters , imposing such correlations due to various biologically motivated tradeoffs may affect multistability and the total number of states of the ecosystem . One possibility is a negative correlation between the competitive abilities of a given species for different nutrients . Such tradeoff may exist due to a limited amount of internal resources ( such as the overall number of transporters ) this species can allocate for consumption of all nutrients . This type of tradeoff was shown to result in an increased species diversity in well-balanced environments , but does not lead to multistability ( Posfai et al . , 2017; Tikhonov , 2016 ) . Similar negative ( positive ) correlations to increase ( decrease ) the number of stable states in a very different consumer-resource model based on the stable marriage problem ( Goyal et al . , 2018 ) . We expect these results to also apply to our model with tradeoff between λα ( c ) and λα ( n ) . Negative correlations between species’ competitive abilities for carbon and nitrogen are expected to increase the total number of stable states in our model , while positive correlations - to decrease it . Another possibility is a negative correlation between species’ competitive ability and its yield for the same nutrient . It is known as a ‘growth-yield tradeoff’ , which states that microbial species with faster growth on a given nutrient tend to use it less efficiently ( have a smaller yield ) ( Pfeiffer et al . , 2001; Beardmore et al . , 2011; Novak et al . , 2006 ) . Growth-yield tradeoff is expected to increase the likelihood of multistability in our model . It could be demonstrated already in the model of Tilman ( 1982 ) with two species competing for two essential resources . If the species with the higher growth rate on , say , carbon source has a smaller yield on this resource than the other species - bistability always ensues . Note that , while growth-yield tradeoff is known to be common among microorganisms , the macroscopic ( e . g . plant ) ecosystems , which are the main focus of Tilman ( 1982 ) , have the opposite correlation in which species’ yield Y is proportional to its competitive ability λ . This type of tradeoff leads to a relative scarcity of multistability in macroscopic ecosystems . Conversely , multistability is expected to be more common in microbial ecosystems due to the growth-yield tradeoff . Ever since Robert May’s provocative question ‘Will a large complex system be stable ? ’ ( May , 1972 ) the focus of many theoretical ecology studies has been on investigating the interplay between dynamic stability and species diversity in real and model ecosystems ( Ives and Carpenter , 2007 ) . May’s prediction that ecosystems with large number of species tend to be dynamically unstable needs to be reconciled with the fact that we are surrounded by complex and diverse ecosystems that are apparently stable . Thus , it is important to understand the factors affecting stability of ecosystems in general and microbial ecosystems in particular . Here , we explored the interplay between diversity and stability in a particular type of microbial ecosystems with multiple essential nutrients . We discussed three criteria for stability of microbial communities shaped by the competition for nutrients: ( i ) how stable is the species composition of a community to fluctuations in nutrient supply rates; ( ii ) the extent of community’s resilience to species invasions; and ( iii ) its dynamical stability to small stochastic changes in abundances of existing species . Naturally-occurring microbial communities may or may not be stable according to either one of these three criteria ( Ives and Carpenter , 2007 ) . The degree of importance of each single criterion is determined by multiple factors such as how constant are nutrient supply rates in time and space and how frequently new microbial species migrate to the ecosystem . Our model provides the following insights into how these three criteria are connected to each other . First , as evident from Figure 1F , the three types of stability are largely independent from each other . Second , communities growing on a well balanced mix of nutrients tend to have high species diversity ( see peak in Figure 4A ) . The similar effect was demonstrated in other consumer resource models ( Posfai et al . , 2017; Tikhonov and Monasson , 2017; Taillefumier et al . , 2017; Marsland et al . , 2019a ) . However , each of the community states in this regime tends to have a low structural stability with respect to nutrient fluctuations . In environments with highly variable nutrient supplies the community will frequently shift between these states . That is to say , some of the species will repeatedly go locally extinct and the vacated niches will be repopulated by others . Furthermore , many of the steady states in this regime are dynamically unstable giving rise to multistability and regime shifts . In this sense our model follows the general trend reported in May ( 1972 ) . Conversely , microbial communities growing on an imbalanced mix of essential nutrients have relatively low diversity ( Figure 4A ) but are characterized by a high degree of structural and dynamic stability ( see Figure 4D and Figure 4C respectively ) . We expect these trends to apply to a broad variety of consumer-resource models . The existence of dynamically unstable states always goes hand in hand with multistability ( Scheffer and Carpenter , 2003 ) ( see the dashed line in Figure 2D for an illustration of this effect in our model ) . Interestingly , in our model we always find V-1 dynamically unstable states coexisting with V dynamically stable ones for the same environmental parameters ( see Figure 3C and Figure 3—figure supplement 1 for some examples ) . All states ( both dynamically stable and unstable ) shown in Figure 3C are positioned along some one-dimensional curve in PCA coordinates . This arrangement hints at the possibility of a non-convex one-dimensional Lyapunov function whose V minima ( corresponding to stable states ) are always separated by V-1 maxima ( unstable stable states ) as dictated by the Morse theory ( Milnor , 1963 ) . This should be contrasted with convex multi-dimensional Lyapunov functions used in MacArthur ( 1970 ) , Case and Casten ( 1979 ) and Chesson ( 1990 ) . Our model can be extended to accommodate several additional properties of real-life microbial ecosystems . First , one could include generalist species capable of using more than one nutrient of each type . The growth rate of such species is given by:gα=min ( ∑iusedbyαλαi ( c ) ci , ∑jusedbyαλαj ( n ) nj ) Here , the sum over i ( respectively j ) is carried out over all carbon ( nitrogen , respectively ) sources that this species is capable of converting to its biomass . One may also consider the possibility of diauxic shifts between substitutable nutrient sources . In this case , each generalist species is following a predetermined preference list of nutrients and uses its carbon and nitrogen resources one-at-a-time , as modelled in Goyal et al . ( 2018 ) . Since at any state each of the species is using a ‘specialist strategy’ , that is to say , it is growing on a single carbon and a single nitrogen source , we expect that many of the results of the current study would be extendable to this model . Interestingly , the stable marriage problem can be used to predict the stable states of microbial communities with diauxic shifts between substitutable resources ( Goyal et al . , 2018 ) and those in communities growing on a mix of two essential nutrients as in this study . It must be pointed out that these models use rather different variants of the stable marriage model . It is straightforward to generalize our model to Monod’s growth equation and to take into account non-zero death rate ( or maintenance cost ) of individual species ( see Appendix 1 ) . One can extend our model to include cross-feeding between the species . In this case some of the nutrients are generated as metabolic byproducts by the species in the community . These byproducts should be counted among nutrient sources and thus would allow the number of species to exceed the number of externally supplied resources . Above we assumed a fixed size of the species pool . This constraint could be modified in favor of an expanding pool composed of a constantly growing number of species . These new species correspond to either migrants from outside of the community or mutants of the species within the community . This variant of the model would allow one to explore the interplay between ecosystem’s maturity ( quantified by the number of species in the pool ) and its properties such as multistability and propensity to regime shifts . In many practical situations we would like to be able to control microbial communities in a predictable and robust manner . That is to say , we would like to be able to reliably steer the community into one of its stable states and to maintain it there for as long as necessary . Alternative stable states and regimes shifts greatly complicate the task of manipulation and control of microbial ecosystems . Indeed , multistability means that the environmental parameters alone do not fully define the state of the community . In order to get it to a desired state , one needs to carefully select the trajectory along which one changes the environmental parameters ( nutrient supply rates ) . Changing these parameters could lead to disappearance ( local extinction ) of some microbial species and open the ecosystem for colonization by others thereby changing its state . However , not all the states could be directly converted to each other in one step due to them being restricted to vastly different environments . Thus , densely interconnected networks of regime shifts shown in Figure 2E–F can be viewed as maps guiding the selection of the optimal environmental trajectory leading to the desired stable species composition . These maps also suggest that microbial ecosystems described by our model might have a relatively small number of key drivers of regime shifts roughly corresponding to network modules ( see Figure 2F ) . Regime shifts in each of the modules are driven by the competition between two mutually exclusive sets of keystone species . In addition to these keystone species , states also include ‘satellite’ species that do not generally affect the bistable switch . The exploration of different manipulation strategies of microbial ecosystems and the role of keystone and peripheral species in regime shifts is the subject of our future research ( Maslov et al . , unpublished ) . The competitive exclusion principle states that , in general , two species competing for the same growth-limiting nutrient cannot coexist with each other . Accounting for non-limiting nutrients present in our model , the competitive exclusion principle can be reformulated as the following two rules: Note that in any state of our model every species has a unique nutrient limiting its growth . By the virtue of the Rule 1 , if a nutrient is limiting the growth of any species at all , such species is also unique . Hence , in a given state the relationship between surviving species and their growth-limiting nutrients ( marked as shaded ovals in Figure 1A ) is an example of a matching on a graph of resource utilization . Rule two imposes additional limitations on this matching . As we show in the Appendix 2 , uninvadable states correspond to stable matchings in a variant of the celebrated stable marriage problem ( Gale and Shapley , 1962; Gusfield and Irving , 1989 ) . Just like in the MacArthur model ( MacArthur and Levins , 1964 ) or any other consumer-resource model for that matter , the number of species present in a steady state of the community cannot exceed the total number of nutrients they consume . Any community constructed using Rules 1 and 2 represents a steady state of the ecosystem feasible for a certain range of nutrient supply rates . This state can be either invadable or uninvadable , and either dynamically stable or not . For simplicity in our simulations we use equal numbers of C and N resources ( L carbons and L nitrogens ) , with one unique species capable of utilization of every pair of resources ( L2 species in total ) . One must reiterate that our theory is not restricted to the specific values of K , M , and S . We first selected the values of λ ( i , j ) ( c ) and λ ( i , j ) ( n ) from a uniform random distribution between 10 and 100 . Note that in the regime of high nutrient supply ( ϕ ( c , n ) ≫δ2λ ( c , n ) ) all steady states of the community can be identified and tested for invadability using only the relative rank order of species’ competitiveness for nutrients . For this we used the following exhaustive search algorithm: Step 1 - Select the subset of species whose growth is limited by C ( C-limited species ) . For every carbon nutrient there are L ways to choose a species using this nutrient . There is also an additional possibility that this nutrient is not limiting the growth of any species . Thus , the total number of possibilities is L+1 for each of L carbon nutrients . There are ( L+1 ) L ways to choose the set of C-limited species and our algorithm will exhaustively investigate each of these potential steady states one-by-one . Step 2 - Given the set of C-limited species selected in Step 1 , we now select the set of N-limited species . We first eliminate from our search any species that doesn’t have enough carbon to grow . That is to say , we go over all carbon nutrients one-by-one and eliminate all species whose λ ( c ) is smaller than that of the C-limited species ( if any ) for this carbon nutrient . Among the remaining species we go over the nitrogen nutrients one-by-one and look for all possible ways to add a species limited by a given nitrogen source nj that satisfy the Rule 2 . More specifically , we identify all species that use nj and can grow on their carbon sources ( those are the only species that remained after the elimination procedure described above ) . We then compare λ ( n ) s of these species to λ ( n ) s of all C-limited species using nj . To satisfy the Rule 2 for each nj we can add at most one N-limited species and its λ ( n ) has to be smaller than λ ( n ) s of all C-limited species using nj . Let Mj be the number of such species ( Mj=0 if there are no such species for a given nj ) . The total number of possible steady states of our model for a given combination of C-limited species selected in Step 1 is given by ∏j=1L ( Mj+1 ) . Here the +1 factor in Mj+1 takes into account an additional possibility to not add any N-limited species for nj . The unique way to construct an uninvadable state by following this algorithm is to go over all nitrogen sources one-by-one and for each of them add the N-limited species with the largest λ ( n ) among all species using this resource , whose growth is allowed by carbon constraints . If for every nj this species is allowed by the Rule 2 , that is to say , if its λ ( n ) is smaller than λ ( n ) of all C-limited species using nj , we successfully constructed a unique uninvadable state for a given set of C-limited species . Indeed , all possible invading species that are allowed to grow by their carbon nutrients will be blocked by their nitrogen nutrients . If , however , for any of nj , the species with the largest λ ( n ) is not allowed by the Rule 2 , that is to say , if its λ ( n ) is larger than λ ( n ) of at least one of the C-limited species , this species would make a successful invader of any state we construct . In this case , there is no uninvadable state for the set of C-limited species selected during the Step 1 . We used the above procedure to identify all possible steady states and to classify them as invadable and uninvadable for different numbers of resources used in our 2C × 2N × 4S and 6C × 6N × 36S examples . Note that , while this method is computationally feasible for a relatively small number of nutrients ( we were able to successfully use it for up to 9 nutrients of each type ) , for larger systems one should rely on computationally more efficient algorithms based on the stable marriage problem ( Gale and Shapley , 1962; Gusfield and Irving , 1989 ) as described in the Appendix 3 . Given the parameters defining all species ( i . e . , the set of their λs and Ys ) and the chemostat dilution constant δ , each state p is feasible within a finite region in the nutrient supply space ( a K+M dimensional space ϕ→={ϕi ( c ) , ϕj ( n ) } ) , where all microbial populations and nutrient concentrations are non-negative and the limiting nutrients of every surviving species do not change . It is easy to show that in a steady state our system satisfies mass conservation laws for each of the nutrients: ( 4 ) ci+∑allαusingciBαYα ( c ) =ϕi ( c ) δ , nj+∑allαusingnjBαYα ( n ) =ϕi ( n ) δ . To simplify the process of calculating the feasible volumes of all states we worked in the limit of high nutrient supply where ϕi ( c ) ≫δ2λα ( c ) and ϕj ( n ) ≫δ2λα ( n ) for all species α . In this case the concentration δ/λα ( c , n ) of any nutrient limiting growth of some species ( α in this case ) is negligible compared to its ‘abiotic concentration’ϕi ( c , n ) /δ , that is to say , its concentration before any microbial species were added to the chemostat . In this case one can ignore the terms ci and nj in Equation 4 for all nutrient limiting growth of some species and leave only the ones that are not limiting the growth of any species . It is convenient to introduce the K + M − dimensional vector X→p of microbial abundances and non-limiting nutrient concentrations in a given state p . For example , for the uninvadable state #5 in the 2C × 2N × 4S model we have: X→5={B ( 1 , 1 ) , B ( 1 , 2 ) , B ( 2 , 2 ) , n2} . The mass conservation laws ( Equation 4 ) can be used to obtain the feasible volumes of all states and can be represented in a compact matrix form for each state p: ( 5 ) ϕ→=Rp^⁢X→p , where ϕ is the vector of K+M nutrient supply rates and Rp^ is a matrix composed of inverse yields Y-1 of surviving species and '1’ for each of the non-limiting nutrients in a given state p . For example , for the state #5 in our 2C × 2N × 4S model the Equation 5 expands to: ( 6 ) [ϕ1 ( c ) ϕ2 ( c ) ϕ3 ( c ) ϕ4 ( c ) ]=[1Y ( 1 , 1 ) ( c ) 1Y ( 1 , 2 ) ( c ) 00001Y ( 2 , 2 ) ( c ) 01Y ( 1 , 1 ) ( n ) 00001Y ( 1 , 2 ) ( n ) 1Y ( 2 , 2 ) ( n ) 1][B ( 1 , 1 ) B ( 1 , 2 ) B ( 2 , 2 ) n2] . Using Equation 5 , it is easy to check if a given state is feasible at a particular nutrient supply rate ϕ→ by multiplying Rp^-1 ( the inverse of the matrix Rp^ ) with ϕ→ . If all of the elements of the resulting vector X→p are positive , then the state p is feasible at ϕ→ . If the matrix Rp^ is not invertible that is det⁡ ( Rp^ ) =0 , the state is feasible only on a low-dimensional subset of nutrient supply rates . This is not possible for a general choice of yields Y and is not considered in our study . The parameters λ and Y for 2C × 2N × 4S and 6C × 6N × 36S realizations of the model were drawn from the uniform random distributions and are listed in Supplementary files 1 , 2 and 3 , 4 , 5 , 6 . For λ’s the distribution ranges between 10 and 100 . For Y’s it is between 0 . 1 and 1 . In our numerical simulations , for each model realization , we sampled 106 random nutrient supply rate combinations ϕ→ . Supply rates of each individual nutrient were independently selected from a uniform distribution on the [10 , 1000] interval . We refer to this procedure as Monte Carlo sampling . The lower bound ensures that the system is always in the limit of high nutrient supply since max ( δ2λα ) =0 . 1≪10 . . Then we checked the feasibility of each of the 33 possible steady states ( both invadable and uninvadable ) in the 2C × 2N × 4S model and each of the 1211 uninvadable steady states in the 6C × 6N × 36S model . That is to say , for every set of nutrient supply rates ϕ→ and for every state p we checked whether all elements of X→p are positive . The feasible range of nutrient supply rates of each state was estimated as the fraction of nutrient supply rate combinations ( out of 1 million vectors ϕ→ sampled by our Monte Carlo algorithm ) where this state was found to be feasible . Two stable states are said to be capable of a regime shift if their feasibility ranges overlap with each other , that is if there exists at least one nutrient supply rate combination at which both these states are feasible . We used the data obtained by the Monte-Carlo sampling to look for such cases and to construct networks shown in Figure 2E , Figure 2F . We performed additional simulations to look for boundaries between uninvadable states in our realization of the 2C × 2N × 4S model . In order to do that , for each state we generated a large ensemble of random vectors of bacterial abundances of surviving species and concentrations of all not-limiting nutrients . The population of one of the species ( also randomly selected ) was set to be a small negative number ( −0 . 01 ) . This represents continuous gradual extinction of this species upon crossing of the boundary , The populations of state’s other surviving species and the concentrations of its non-limiting nutrients were drawn from the uniform distribution between ( 0 , 1] . Using Equation 5 , we calculated the nutrient supply rates ϕ corresponding to this case . For these nutrient supply rates lying just across the feasibility boundary of the originally selected state , we checked the feasibility of other five uninvadable dynamically stable states . If any of these states ended up being feasible , we assumed that this state shares a boundary with the originally selected one . Using these procedure we found four bordering pairs of states shown as red edges in Figure 2E . We used Gephi 0 . 9 . 2 software package to visualize the network in Figure 2F and to perform its modularity analysis . Seven densely interconnected clusters shown with different colors in Figure 2F were identified using Gephi’s built-in module-detection algorithm ( Blondel et al . , 2008 ) with the resolution parameter set to 1 . 5 . We checked the dynamic stability of every 33 possible steady state ( both invadable and uninvadable ) ( for the 2C × 2N × 4S model ) and each of the 1211 uninvadable states ( for the 6C × 6N × 36S model ) using the following two algorithms: To investigate how multistability in our model depends on variation in stoichiometric ratios of different species , we simulated 4000 variants of the 2C × 2N × 4S model . In these variants , we kept the same choice of species competitiveness ( quantified by their λ’s ) but reassigned their yields Y . To cover a broad range of standard deviations of N:C stoichiometry of different species ( their Yα ( c ) /Yα ( n ) ) we randomly sampled yield combinations from gradually expanding intervals . First we simulated 1000 model variants , where yields of four species were independently drawn from uniform distribution U⁢ ( 0 . 45 , 0 . 55 ) . These simulations were followed by 1000 model variants , where yields of four species were drawn from U⁢ ( 0 . 3 , 0 . 7 ) , 1000 model variants with yields from U⁢ ( 0 . 1 , 0 . 9 ) and , finally , 1000 model variants with yields from U⁢ ( 0 . 01 , 1 . 0 ) . In each variant of the model with a particular set of yields of four species , we calculated the fraction of multistable points among 105 nutrient supply rate combinations as described in the section 5 . 2 Monte-Carlo sampling of nutrient supply rates to identify feasible ranges of states of Materials and methods . The results are shown in Figure 5 . The PCA analysis , plots and statistical tests were implemented using R version 3 . 4 . 4 . Other simulations were carried out in C ( using compiler gcc version 5 . 4 . 0 ) and Python 3 . 5 . 2 . Matlab analysis was done using MATLAB and Statistics Toolbox Release 2018a , The MathWorks , Inc , Natick , Massachusetts , United States . The code for both our simulations and statistical analysis can be downloaded from: https://github . com/ssm57/CandN ( Dubinkina and Maslov , 2019; copy archived at https://github . com/elifesciences-publications/CandN ) .	In nature , different species of bacteria and fungi often live together in stable microbial communities . Exactly which species are present in the group and in which proportion may vary between communities . Changes in the environment , and in particular in the availability of nutrients , can trigger abrupt , extensive , and long-lasting changes in the composition of a community: these events are known as regime shifts . For instance , when bodies of water receive large quantities of phosphorus and nitrogen , certain algae can start to multiply uncontrollably and take over other species . A given community can have different stable species compositions , but it was unclear exactly how variations in nutrients can influence regime shifts . To examine this problem , Dubinkina , Fridman , Pandey and Maslov harnessed mathematical techniques used in game theory and economics and modeled all the possible stable compositions of a community . They could then predict which environmental conditions – in this case , the amount of specific nutrients – were necessary for each stable composition to exist . These models also showed which conditions could trigger a regime shift . Finally , how resilient the communities were to different types of perturbations – for instance , an invasion by new species or changes in nutrient supply – was examined . The results show that if competing species require different quantities of the same nutrients , then the community can have several possible stable compositions and it is more likely to go through regime shifts . In addition , a small number of keystone species were identified which can drive regime shifts by preventing other microbes from invading the community . Ultimately , these results suggest ways to control microbial communities in our environment , for example by manipulating nutrient supplies or introducing certain species at the right time . More work is needed however to verify the predictions of the model in real communities of microbes .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "ecology" ]	2019	Multistability and regime shifts in microbial communities explained by competition for essential nutrients
Epithelial cells and their underlying basement membranes ( BMs ) slide along each other to renew epithelia , shape organs , and enlarge BM openings . How BM sliding is controlled , however , is poorly understood . Using genetic and live cell imaging approaches during uterine-vulval attachment in C . elegans , we have discovered that the invasive uterine anchor cell activates Notch signaling in neighboring uterine cells at the boundary of the BM gap through which it invades to promote BM sliding . Through an RNAi screen , we found that Notch activation upregulates expression of ctg-1 , which encodes a Sec14-GOLD protein , a member of the Sec14 phosphatidylinositol-transfer protein superfamily that is implicated in vesicle trafficking . Through photobleaching , targeted knockdown , and cell-specific rescue , our results suggest that CTG-1 restricts BM adhesion receptor DGN-1 ( dystroglycan ) trafficking to the cell-BM interface , which promotes BM sliding . Together , these studies reveal a new morphogenetic signaling pathway that controls BM sliding to remodel tissues . The basement membrane ( BM ) is a cell-associated , dense , sheet-like form of extracellular matrix that underlies all epithelia and endothelial tissue , and surrounds muscle , fat , and Schwann cells ( Halfter et al . , 2015; Yurchenco , 2011 ) . BMs are built on polymeric laminin and type IV collagen networks that arose at the time of animal multicellularity , and may have been required for the evolution of complex tissues ( Hynes , 2012; Ozbek et al . , 2010 ) . Consistent with this idea , BMs provide tissues with mechanical support , barrier functions , and cues for polarization , differentiation and growth ( Breitkreutz et al . , 2013; Hay , 1981; Poschl et al . , 2004; Rasmussen et al . , 2012; Suh and Miner , 2013; Yurchenco , 2011 ) . Although it was generally thought that cell-BM interactions are static , live imaging studies have revealed that cell-BM interfaces are highly dynamic ( Morrissey and Sherwood , 2015 ) . One of the most dramatic examples of this mobility is cell-BM sliding , during which epithelial cell layers and their underlying BM sheets move ( slide ) along one another independently to regulate tissue remodeling or renewal . Examples of cell-BM sliding are varied and include egg chamber rotation in Drosophila , where the follicle cells move along BM and deposit constricting strips of aligned type IV collagen , which direct egg elongation along the anterior-posterior axis ( Cetera and Horne-Badovinac , 2015; Haigo and Bilder , 2011 ) . During salivary gland growth in vertebrates , the BM slides away from the bud tip toward the duct , allowing bud expansion while restricting growth at the duct ( Harunaga et al . , 2014 ) . Further , BM labeling and pulse chase experiments revealed that intestinal epithelial cells derived from the stem cell crypt slide along the BM towards the villus tips during differentiation to rapidly renew the gut epithelium ( Clevers , 2013; Trier et al . , 1990 ) . Cell-BM sliding may be a common morphogenetic process that regulates organ shaping during development , tissue homeostasis , and diseases such as cancer , where remodeling of cell-BM interfaces frequently occur ( Kelley et al . , 2014; Rowe and Weiss , 2008 ) . Because of the challenge of visualizing and experimentally examining dynamic cell-BM interactions in vivo , however , mechanisms controlling cell-BM sliding remain largely unknown . An experimentally tractable model to examine cell-BM sliding during tissue remodeling is uterine-vulval attachment in C . elegans ( Schindler and Sherwood , 2013 ) , a developmental process that is necessary for effective mating and egg laying in the worm . During the mid-L3 larval stage , the uterine-vulval connection is initiated by a specialized uterine cell , the anchor cell ( AC ) , that breaches the BM that separate these tissues and attaches to the underlying vulval cells . Following AC invasion , the gap in the BM widens further , which allows additional connection between uterine and vulval cells ( Ihara et al . , 2011 ) . BM gap widening does not involve BM degradation . Instead , optical highlighting of BM and manipulation of tissue dynamics has shown that growth and morphogenesis of the uterine and vulval tissues generate forces on the BM that drive BM sliding over the vulval and uterine cells to further expand the gap ( Ihara et al . , 2011 ) . The vulval cells have a key role in controlling the extent of BM movement . The centrally located vulval E and F cells , which contact the BM gap boundary , undergo precisely timed divisions to initiate BM sliding--cell rounding during divisions dramatically reduces cell contact with the BM and allows the BM to slide over the vulE and F cells ( Matus et al . , 2014 ) . The BM stops sliding on the non-dividing vulD cell , which concentrates the BM adhesion receptor INA-1/PAT-3 ( integrin ) to stabilize the BM gap boundary ( Ihara et al . , 2011; Matus et al . , 2014 ) . While vulval cell divisions control BM sliding on the vulval side of the uterine-vulval connection , the role of the uterine π cells , which flank the AC , and sit on the opposing side of the BM gap boundary , remain unclear . Many receptors bind BM components and are possible regulators of BM sliding . Two of the most prominent classes of adhesion receptors that link BM to the cytoskeleton are integrin family members and the receptor dystroglycan ( Bello et al . , 2015; Kramer , 2005; Yurchenco , 2011 ) . Most studies have focused on how these receptors are activated or upregulated to strengthen adhesion; however there is a growing appreciation of the importance of integrin and dystroglycan downregulation in morphogenetic , homeostatic , and disease processes ( Agrawal et al . , 2006; Bouvard et al . , 2013; Miller et al . , 2015; Nakaya et al . , 2013 ) . Cells utilize a variety of mechanisms to reduce BM receptor adhesion . These include transcriptional downregulation , production of negative regulators that interfere with receptor activation , alterations in phosphorylation status , and changes in localization and trafficking ( Bouvard et al . , 2013; Nakaya et al . , 2013; Poulton and Deng , 2006 ) . Whether these strategies are used to modulate adhesion to control BM sliding is unknown . To address how the uterine BM gap boundary cells control BM sliding during uterine-vulval attachment in C . elegans , we carried out a mutagenesis screen . Through this screen we identified a putative null mutant in the gene encoding the LIN-29 protein , which is a Kruppel-family EGR ( early growth response ) protein ( Harris and Horvitz , 2011 ) , that is deficient in BM sliding . We show that LIN-29 is expressed and functions in the invading AC to promote BM sliding . We further find that LIN-29 regulates BM sliding through its role in maintaining AC expression of LAG-2 , a transmembrane Notch ligand , which activates Notch signaling in the neighboring BM gap boundary cells , the uterine π cells that sit on BM next to the nascent breach . Through a targeted uterine-specific RNAi screen of putative direct Notch targets , we find that Notch activation in these uterine cells upregulates ctg-1 expression , which encodes a Sec14-GOLD domain phosphatidylinositol transfer protein that belongs to a class of proteins implicated in regulating vesicle trafficking ( Grabon et al . , 2015 ) . Finally , using photobleaching , CRISPR/Cas-9 targeted knockout , and cell-specific rescue experiments , we demonstrate that CTG-1 functions in the uterine π cells to limit the cell surface trafficking of DGN-1 ( dystroglycan ) , and that reduction of DGN-1 is sufficient to promote BM sliding . Together these studies identify a new morphogenetic pathway—from signaling mechanism to effector—that promotes BM sliding , which can be used by invasive cells to enlarge BM openings and allow the direct interaction of cells between tissues . During larval development , the uterine and vulval tissues in C . elegans are initially separated by the juxtaposed gonadal and ventral BMs . A specialized uterine cell , the AC , breaches these BMs during the mid-L3 larval stage to initiate uterine-vulval connection ( Sherwood and Sternberg , 2003 ) . AC invasion and BM remodeling occur in tight coordination with vulval development and can be staged according to the centrally located 1° fated P6 . p vulval precursor cell ( VPC ) divisions . During invasion ( P6 . p 2 and 4-cell stages ) the gonadal and ventral BMs are breached and fuse at the edges of the invading AC ( Figure 1A , Ihara et al . , 2011 ) . Following AC invasion , the vulval cells grow , invaginate , and divide ( Ihara et al . , 2011 ) . The rapid expansion and invagination of the vulval cells generate forces on the BM that promote expansion of the nascent BM opening through BM sliding over the vulval and uterine cells that sit at the boundary of the BM gap . Expansion of the BM gap allows direct contact between the cells that mediate uterine-vulval attachment ( Ihara et al . , 2011 ) . Precisely timed divisions of the BM gap boundary vulval P6 . p descendants ( the vulE and vulF cells ) during the late L3 stage reduce BM adhesion , allowing the BM to slide over these dividing vulval cells ( Matus et al . , 2014 ) . The BM gap halts its expansion over the non-dividing vulD cells during the early L4 stage ( P6 . p 8-cell stage ) ; the vulD cells further stabilize the BM gap boundary by upregulating the integrin heterodimer INA-1/PAT-3 and the adhesion regulator VAB-19 ( the worm ortholog of Kank ) ( see Figure 1A , Ihara et al . , 2011; Matus et al . , 2014 ) . The ventral uterine ( VU ) cells neighboring the AC are also initially in contact with the BM . During the time of AC invasion , the adjacent six VU cells are specified to adopt a π fate through LAG-2/LIN-12 ( Notch ) signaling by the AC . During BM sliding the uterine π cells undergo one round of division ( Newman , 1995 ) . The role of the π cells , the uterine BM gap boundary cells , in regulating BM sliding is unknown . 10 . 7554/eLife . 17218 . 003Figure 1 . BM sliding during uterine-vulval connection and isolation of lin-29 ( qy1 ) . ( A ) Top left: During the mid-L3 larval stage ( vulval P6 . p 2-cell stage ) , the uterine anchor cell ( AC ) breaches the juxtaposed gonadal and ventral BMs ( green ) and contacts the central 1° fated vulval precursor cells ( VPCs; P6 . px cells , red ) . Middle: After AC invasion is completed at the late L3 ( P6 . p 4-cell stage ) , VPC growth , divisions , and invagination drive BM sliding , which widens the BM gap . Ventral uterine ( VU ) π cells ( blue ) also begin to divide at the late L3 . Right: By the early-to-mid L4 ( P6 . p 8-cell stage ) , the BM stops sliding and is stabilized over the non-dividing 2° fated vulD cells ( dotted line ) by vulD expression of cytoplasmic VAB-19 ( Kank ) and cell surface integrin . BM sliding allows direct connection between vulE and F cells with uterine π cells , which form the mature uterine-vulval connection . The regulation of BM sliding by the uterine π cells is unknown ( represented with ' ? ' ) . ( B ) A schematic diagram of the lin-29 gene . Allele qy1 ( red ) creates an early stop codon affecting all three transcripts . Allele ga94 ( blue ) affects the lin-29a/b transcripts but leaves the translation of lin-29c largely unaffected , while n546 ( green ) introduces a nonsense mutation affecting all three encoded isoforms of the LIN-29 protein . ( C ) A lateral , central plane ( with the AC in focus ) DIC image ( top ) with BM ( laminin::GFP ) and AC ( cdh-3 > mCherry::PLCδPH ) fluorescence ( middle ) and ventral view of laminin::GFP ( bottom ) showing the circular opening of the BM at the uterine-vulval connection at the P6 . p 8-cell stage . In lin-29 ( qy1 ) mutants the BM gap ( top , white brackets=diameter; middle , arrowheads=edges of gap; bottom , white dashed lines=outline of circular BM gap ) failed to widen after AC invasion . Scale Bar , 5 μm . ( D ) Quantification of the BM gap diameter at the P6 . p 8-cell stage in wild type ( n = 66 ) and lin-29 ( qy1 ) mutants ( n = 65 ) . Box signifies first and third quartiles and median measurement; whisker ends signify minimum and maximum values that fall within 1 . 5 times the interquartile range . The asterisks denote a statistically significant difference ( * indicates p<0 . 001 Wilcoxon rank sum test ) . DOI: http://dx . doi . org/10 . 7554/eLife . 17218 . 00310 . 7554/eLife . 17218 . 004Figure 1—source data 1 . BM gap diameter in wild type vs . lin-29 ( qy1 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 17218 . 004 To identify genes that regulate the morphogenetic process of BM sliding , we performed an F2 forward mutagenesis genetic screen ( Jorgensen and Mango , 2002; see Experimental Procedures ) . We screened the F2 progeny of 12 , 000 mutagenized F1 animals and selected fertile mutants with a protruding vulva ( Pvl ) phenotype , which often results from defects in uterine-vulval connection . Isolated Pvl mutants were then examined at high magnification ( 1000x ) using differential interference contrast ( DIC ) microscopy at the early L4 stage when BM position can be visualized as a phase dense line separating the uterine and vulval tissue . Through this screen we identified 10 mutants with defects in BM remodeling during uterine-vulval attachment ( see Materials and methods ) . We focused on the mutant qy1 , which had a highly penetrant defect in the BM gap expansion after the AC invasion ( see below ) . To determine the molecular nature of the qy1 mutation , we used a single-nucleotide-polymorphism ( SNP ) based mapping strategy and complementation analysis ( see Experimental Procedures; Davis et al . , 2005 ) . Using this strategy , we mapped the causative mutation of the qy1 allele to lin-29 , which encodes a Kruppel-family EGR zinc-finger protein transcription factor ( Harris and Horvitz , 2011; Rougvie and Ambros , 1995 ) . We sequenced the qy1 allele and found that it introduces a nonsense mutation near the start of the region encoding the DNA binding domain in all three known isoforms of the gene ( Figure 1B ) . As this disrupts translation prior to the previously characterized null allele lin-29 ( n546 ) ( Rougvie and Ambros , 1995 ) , it suggests that qy1 is a null allele of the lin-29 gene . To better visualize and precisely quantify the BM gap defect after the loss of lin-29 , we crossed integrated transgenes expressing a functional fusion of the main structural BM component laminin to GFP ( laminin::GFP ) and an AC membrane marker ( cdh-3 > mCherry::PLCδPH ) into lin-29 ( qy1 ) mutant animals ( Figure 1C , D ) . Laminin::GFP shows the same localization as immunolocalized laminin ( Sherwood and Sternberg , 2003 ) . To quantify the BM gap defect , we measured the diameter of the BM gap at the early L4 ( P6 . p 8-cell stage ) in the plane where the AC is in focus in wild type and qy1 worms ( Figure 1D ) . The gap in the BM was significantly reduced in qy1 mutant animals ( Figure 1C , D ) . Further , we found that that while in wild type animals , the BM moved away from both the anterior and posterior sides of the AC in most animals ( n = 13/20 animals; the remaining 7 moved on one side ) , in qy1 animals the BM never moved away from both sides of the AC ( n = 0/20 animals ) and only occasionally moved slightly away from one side ( n = 5/20 animals ) . We postulated that the defect in BM position in lin-29 mutant animals could result from the absence of BM sliding or alternatively the inappropriate deposition of new BM after sliding . To differentiate between these possibilities we directly examined BM sliding by performing optical highlighting experiments using transgenic animals expressing laminin tagged with the photoconvertible fluorophore Dendra ( laminin::Dendra; Figure 2A , Ihara et al . , 2011 ) . Dendra is a stable , photoconvertible fluorescent protein that irreversibly switches from green to red following exposure to short wavelength light ( Gurskaya et al . , 2006 ) . Segments of the BM adjacent to the invading AC were photoconverted at the P6 . p 4-cell stage and imaged at the P6 . p 8-cell stage . While marked edges of the BM gap moved away from the AC and rested over the vulD cell in wild type animals ( n = 10/10 animals ) , optically highlighted segments maintained contact with the AC in lin-29 ( qy1 ) mutants and failed to move over the vulD cell ( n = 18/18 animals; Figure 2A ) . These observations indicate that the BM in lin-29 mutant animals fails to slide away from the AC and excludes additional BM deposition as a mechanism for the defect . Notably , the morphology of the developing vulva—including the number of VPCs and their relative positions ( n = 15/15 animals , Figure 2B ) and vulval height and width ( Figure 2—figure supplement 1 ) —was the same in lin-29 ( qy1 ) mutants compared with wild type animals . Taken together , these results indicate that lin-29 mutants have a specific defect in BM sliding that is independent of the role of the vulval cell divisions in controlling BM movement ( Matus et al . , 2014 ) . 10 . 7554/eLife . 17218 . 005Figure 2 . lin-29 ( qy1 ) mutants are defective in BM sliding but have a normal vulval precursor morphology . ( A ) Fluorescence overlays of optical highlighting ( photoconversion ) of ~5 µm long laminin::Dendra segments ( magenta in top; white in middle; white arrowheads highlighted regions next to BM gap , yellow arrowheads highlighted regions farther from gap ) in wild type ( left ) versus lin-29 ( qy1 ) mutant ( right ) . The fate of optically highlighted BM at the P6 . p 4-cell stage ( T=0’ ) was assessed at the P6 . p 8-cell stage ( T=240’ ) . In wild type animals , optically highlighted laminin::Dendra at the edges of the BM gap slid from a position over the vulF precursor cell to over the vulD cell ( indicated with dashed outline; n = 10/10 animals ) . In contrast , in lin-29 ( qy1 ) mutants the optically highlighted laminin did not move over the vulD cell ( n = 18/18 animals ) . The arrow in the bottom panels indicates the AC . ( B ) Vulval morphology in wild type ( left ) versus a qy1 mutant ( right ) . The number ( see text ) and positions of vulF , vulE , and vulD ( outlined in CAAX::GFP ) were normal in lin-29 ( qy1 ) animals ( top panels: grayscale , bottom panels: green ) , while the BM ( laminin::mCherry , middle panels: grayscale and bottom panels: magenta ) gap boundaries are narrower in the lin-29 ( qy1 ) mutant . ( bracket ) . The arrow in the middle panel for the wild type denotes a puncta of laminin internalized by the AC during invasion , and does not represent the edge of the BM gap ( which is located above vulD in wild type and is characterized by a build-up of laminin ) . All images lateral , central plane . Scale Bars , 5 um . DOI: http://dx . doi . org/10 . 7554/eLife . 17218 . 00510 . 7554/eLife . 17218 . 006Figure 2—figure supplement 1 . Vulval morphology measurements in lin-29 mutants . Quantification of vulval height ( measured as the distance between the nucleus of vulE and the base of the vulva ) and width ( measured as the distance between the nuclei of the 2 vulD cells ) in wild type ( n = 34 ) and lin-29 ( qy1 ) ( n = 16 ) No significant differences were observed ( p>0 . 05 , Student’s t-test ) . DOI: http://dx . doi . org/10 . 7554/eLife . 17218 . 006 To determine where LIN-29 functions to promote BM sliding , we first examined the expression pattern of the lin-29 gene . We fused 5-kb of the 5’ cis-regulatory element of the lin-29 gene to GFP ( lin-29a/b > GFP; see Experimental Procedures ) . Using this reporter , we detected expression of lin-29 in the AC throughout the course of uterine-vulval connection ( Figure 3A ) , but did not detect expression in the other cells in contact with or bordering the BM boundary . This expression pattern is consistent with published immunofluorescence studies showing that the LIN-29 protein is expressed at high levels in the AC during this time ( Bettinger et al . , 1997 ) . We also fused GFP to the 5-kb region upstream of the lin-29c isoform , but did not see any expression in the uterine or vulval cells at the time of uterine-vulval connection . These results suggest that the lin-29a/b isoforms might act in the AC to promote BM sliding . 10 . 7554/eLife . 17218 . 007Figure 3 . LIN-29 functions within the AC to promote BM gap expansion . ( A ) Expression of GFP from a lin-29a/b > GFP transcriptional reporter in the AC throughout the course of BM gap expansion . ( B ) Quantification of the BM gap in wild type ( n = 16 ) and lin-29 ( ga94 ) ( n = 22 ) at the P6 . p 8-cell stage . ( C ) Quantification of the BM gap in lin-29 ( qy1 ) ( n = 16 ) and AC specific rescue of LIN-29A ( n = 15 ) at the P6 . p 8-cell stage . ( D ) Laminin::GFP overlaid on DIC in wild type , lin-29 ( qy1 ) , and a lin-29 mutant with AC specific rescue of LIN-29A protein showing diameter of the BM gap ( brackets ) . In B and C , indicates p<0 . 01 , Wilcoxon rank sum test . All images lateral , central plane . Scale Bars , 5 um . DOI: http://dx . doi . org/10 . 7554/eLife . 17218 . 00710 . 7554/eLife . 17218 . 008Figure 3—source data 1 . BM gap diameter in wild type vs . lin-29 ( ga94 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 17218 . 00810 . 7554/eLife . 17218 . 009Figure 3—source data 2 . BM gap diameter in lin-29 ( qy1 ) vs . lin-29 ( qy1 ) ; cdh-3 > lin-29a::GFP . DOI: http://dx . doi . org/10 . 7554/eLife . 17218 . 009 To further test the idea that that the lin-29a/b isoforms have a specific function in BM sliding , we scored lin-29 ( ga94 ) mutants . The ga94 mutation creates an early stop codon in exon IV of the lin-29a and lin-29b isoforms but removes only the first two amino acids of the shorter lin-29c isoform ( see Figure 1B ) . Previous studies have reported that lin-29 ( ga94 ) mutants display some , but not all of the phenotypes associated with lin-29 null alleles ( Bettinger et al . , 1997 ) . We measured BM gap expansion and found that ga94 mutants had a similar defect in the BM gap opening to qy1 mutants ( Figure 3B ) . These results suggest that LIN-29A and/or LIN-29B—but not LIN-29C—promote BM gap expansion . We next performed an AC-specific rescue of lin-29a expression in lin-29 ( qy1 ) mutant animals using an AC-specific promoter ( cdh-3 > lin-29a::GFP; [Kirouac and Sternberg , 2003] ) . We found that LIN-29A::GFP protein expressed only in the AC completely rescued BM gap expansion ( Figure 3C , D ) . We conclude that LIN-29 functions within the AC to promote BM sliding . As the BM remained in contact with the AC in most lin-29 mutants , we hypothesized that LIN-29 might be required for AC de-adhesion from the BM . We used laser-directed ablation to specifically kill the AC in qy1 mutants immediately after AC invasion and then assessed BM gap expansion . We found that the BM gap still failed to open in the absence of the AC in lin-29 mutant animals ( n = 8/8 animals , Figure 4A ) . Ablation of the AC at this stage in wild type animals does not prevent BM sliding and gap expansion ( Ihara et al . , 2011 ) . These results suggest that LIN-29 does not facilitate sliding by causing the AC to de-adhere from the BM . 10 . 7554/eLife . 17218 . 010Figure 4 . AC-mediated Notch signaling to the uterine π cells promotes BM sliding . ( A ) Fluorescence overlays of laminin::GFP on DIC images show that the BM gap ( brackets ) failed to expand in lin-29 ( qy1 ) mutants after the AC was ablated ( left , n = 8/8 ) or in mock ablated lin-29 mutants ( middle , n = 8/8 ) . However , the BM gap opened wider on the side of the AC in lin-29 mutants ( BM slid over the vulD cell , outlined with dashed line ) after laser ablation of the uterine π cells on that side ( right , arrow , n = 5/8 animals; p<0 . 001 , Fisher’s Exact Test ) . ( B ) Uterine-specific RNAi targeting the L4440 empty vector control ( left ) and the AC-expressed Notch ligand lag-2 ( right ) revealed that lag-2 is required for BM sliding ( BM visualized with laminin::GFP , green ) away from the AC ( mCherry::PLCδPH , magenta ) . ( C ) Quantification of the BM gap size after treatment L4440 control ( n = 62 ) and uterine-specific lag-2 RNAi ( right , n = 68 ) . ( D ) In sel-12 ( ty11 ) mutants , which perturbs LIN-12 ( Notch ) signaling in the uterine π cells , the BM gap ( laminin::GFP , right; visualized at P6 . p 8-cell stage , DIC left ) failed to expand . ( E ) Quantification of the BM gap in sel-12 ( ty11 ) ( n = 15 ) and wild type ( n = 15 ) . For C and E , * indicates p<0 . 001 Wilcoxon rank sum test . All images lateral , central plane . Scale Bars , 5 µmDOI: http://dx . doi . org/10 . 7554/eLife . 17218 . 01010 . 7554/eLife . 17218 . 011Figure 4—source data 1 . BM gap diameter in L4440 vs . lag-2 RNAi . DOI: http://dx . doi . org/10 . 7554/eLife . 17218 . 01110 . 7554/eLife . 17218 . 012Figure 4—source data 2 . BM gap diameter in wild type vs . sel-12 ( ty11 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 17218 . 01210 . 7554/eLife . 17218 . 013Figure 4—source data 3 . BM gap diameter in lag-2 vs . lin-12 and glp-1 RNAi . DOI: http://dx . doi . org/10 . 7554/eLife . 17218 . 01310 . 7554/eLife . 17218 . 014Figure 4—figure supplement 1 . The AC signals through Notch receptor LIN-12 to promote BM sliding . Quantification of the BM gap size after treatment with uterine-specific lag-2 RNAi , lin-12 RNAi , and glp-1 RNAi ( n = 30 each ) . No significant difference was observed between lag-2 and lin-12 ( p>0 . 05 , Wilcoxon rank sum test ) , but glp-1 knockdown yielded a significantly larger BM gap ( * indicates p<0 . 001 , Wilcoxon rank sum test ) . DOI: http://dx . doi . org/10 . 7554/eLife . 17218 . 014 We next hypothesized that the role of LIN-29 in the AC might be to regulate BM sliding through interactions with the neighboring uterine cells at the BM gap boundary . During the early L3 stage ( just prior to the time of AC invasion ) , LIN-29 is required in the AC to specify the π cell fate in the six ventral uterine cell descendants that directly neighbor the AC ( Newman et al . , 2000; Newman , 1995 ) . LIN-29 maintains the expression of the transmembrane Notch ligand LAG-2 in the AC , which activates LIN-12 in the neighboring ventral uterine cells and specifies the π cell fate ( Newman et al . , 2000; Newman , 1995 ) . We thus investigated whether the π cell fate might be required for BM sliding . We perturbed π cell fate specification by knocking down lag-2 expression specifically in the uterine cells . Uterine-specific RNAi was attained with uterine-specific expression of the Argonaute/PIWI gene rde-1 in an rde-1 mutant background , which restores RNAi only in the uterine tissue ( Hagedorn et al . , 2009 ) . Uterine lag-2 knockdown recapitulated the BM sliding defect seen in lin-29 mutants ( Figure 4B , C ) , suggesting that the LIN-29 protein acts in a LAG-2 dependent manner to promote BM sliding . Notably , the reduction in BM gap opening after lag-2 RNAi was less than in lin-29 mutants ( Figure 4C versus Figure 3C ) , which was likely due to the incomplete loss of lag-2 by RNAi . Consistent with LAG-2 acting through the LIN-12 ( Notch ) receptor , we also used RNAi targeting lin-12 and glp-1 ( the only other encoded Notch receptor in the C . elegans genome ) ( Greenwald , 2005 ) and found that loss of lin-12 , but not glp-1 , phenocopied lag-2 RNAi ( Figure 4—figure supplement 1 ) . Finally , we assessed gap expansion in mutants that lacked the ability to transduce Notch signaling in the π cells ( sel-12 ( ty11 ) ) . SEL-12 is a C . elegans presenilin that cleaves ligand bound LIN-12 ( Notch ) so that the intracellular domain is released to translocate to the nucleus and activate target gene expression . No other C . elegans presenilins act in the prospective π cells ( Cinar et al . , 2001 ) , so loss of sel-12 specifically targets π cell specification and circumvents other phenotypes associated with the absence of Notch signaling . Loss of sel-12 led to a similar defect in BM gap expansion as lin-29 mutants ( Figure 4D , E ) . These observations offer strong evidence that LIN-29 in the AC regulates BM sliding by specifying ventral uterine descendents adjacent to the BM gap boundary to adopt the π cell fate . We next hypothesized that LIN-29 mediated specification of the π cells was required for them to reduce their adhesion from the BM . If this were the case , removal of the unspecified uterine boundary cells should restore BM sliding in lin-29 mutants . To test this notion , we ablated the precursors of the π cells ( ventral uterine cells ) on one side of the AC in lin-29 ( qy1 ) mutants . We found that this treatment rescued the BM sliding defect on that side ( n = 5/8 animals , Figure 4A ) . These results support the idea that π cell fate specification leads to reduced adhesion to the BM , which facilitates BM sliding . When the vulval cells vulE and vulF divide at the BM gap boundary , the cells round and briefly reduce their attachment to the BM . This loss of attachment during division promotes BM movement over vulE and vulF , which then stops at the non-dividing vulD cell ( Matus et al . , 2014 ) . We thus hypothesized that the uterine π cell divisions , which occur near the time of primary vulval cell divisions ( see Figure 1A and Figure 5A ) , might be coordinated with the vulE and vulF divisions to facilitate the sliding of the linked ventral and gonadal BMs . 10 . 7554/eLife . 17218 . 015Figure 5 . π cell division does not regulate BM sliding . ( A ) A schematic diagram presenting the hypothesis that uterine π cell ( blue ) divisions ( along the dorsal-ventral axis ) are coordinated with divisions of the underlying vulE and vulF cells ( red , which divide along the left-right axis ) such that the BM ( green ) is released on both sides when dividing cells lose contact with the BM , allowing it to slide . ( B ) Fluorescence overlay showing a cell membrane marker ( CED-10::GFP , green ) of dividing vulE ( dashed outline ) and adjacent uterine π cells ( asterisk denotes a dividing cell and arrowhead a non-dividing cell ) with the BM ( laminin::mCherry , magenta ) . ( C ) A lateral , central plane DIC image of wild type ( left ) with inset showing divided π cells expressing EGL-13::GFP driven by the egl-13 promoter ( egl-13 > egl-13::GFP ) had a similar BM gap size ( bracket ) to an animal expressing CKI-1::GFP in the π cells ( right , egl-13 > cki-1::GFP ) , thus arresting their divisions . ( D ) Quantification of the BM gap in wild type ( n = 15 ) and egl-13 > cki-1::GFP transgenic animals ( n = 13 ) . Only animals in which the π cells failed to divide were included in the analysis . No significant difference was observed ( p>0 . 05 , Wilcoxon rank sum test ) . DOI: http://dx . doi . org/10 . 7554/eLife . 17218 . 01510 . 7554/eLife . 17218 . 016Figure 5—source data 1 . BM gap diameter in wild type vs . egl-13 > cki-1 . DOI: http://dx . doi . org/10 . 7554/eLife . 17218 . 016 During the time of BM gap expansion ( P6 . p 4-cell to 6-cell transition ) each π cell divides once along the dorsal-ventral axis ( Figure 5A ) . To examine the precise timing of these π cell divisions in relation to the vulval divisions , we used a membrane bound GFP that marks both the VPCs and the uterine cells ( CED-10::GFP; [Ziel et al . , 2009] ) . While the π cells began dividing near the same time as the vuE and vulF cells , we found little correlation between the timing of their divisions ( n = 8/46 π cells in 13 animals dividing coincidently with vulE on the other side of the BM , n = 16/54 π cells in 12 animals dividing coincidently with vulF on the other side of the BM , Figure 5B ) . These observations suggest that coordination of π cell divisions with vulE and vulF divisions are not required for BM sliding . To directly determine if π cell divisions are required to expand the BM gap , we blocked their division by expressing the cyclin dependent kinase inhibitor cki-1 driven with an early π cell specific promoter ( egl-13 > cki-1::GFP ) in wild type animals ( Figure 5C ) . egl-13 initiates expression at the time of π cell birth when there are six π cells in the late L3 stage ( Cinar et al . , 2003 ) . We found no difference in the BM gap size at the P6 . p 8-cell stage between animals ( Hanna-Rose and Han , 1999 ) in which π cell divisions had been arrested compared to wild type controls ( Figure 5D ) . We conclude that π cell divisions are not required for these BM boundary cells to loosen their adhesion to the BM to promote BM sliding . We next examined the relationship of the BM to the π cells during BM sliding and stabilization . Notably , on the vulval side of the developing uterine-vulval connection , the BM moves completely over the dividing vulE and vulF cells and the edge of the BM gap stabilizes on the non-dividing vulD cell ( Matus et al . , 2014 ) . In contrast , we found that as the BM slid along the π cells to expand the gap , the edge of the BM gap stabilized on the π cells; BM maintained contact with a significant portion the ventral π cells ( n = 13/13 animals imaged during the BM gap expansion , Figure 6A–C ) . These observations indicate that , in contrast to the vulval cells that lose attachment to the BM , the uterine π cells remain in contact with and are possibly adherent to the BM as the BM slides . 10 . 7554/eLife . 17218 . 017Figure 6 . The π cells remain in contact with the BM during and after BM sliding . ( A ) A schematic diagram showing image planes of lateral view ( orange ) and ventral view ( purple ) of the π cells in relation to the BM . ( B ) In the early P6 . p 6-cell stage , the π cells ( visualized with GFP driven by the egl-13 promoter , green ) lie almost completely over BM ( magenta ) . ( C ) By the P6 . p 8-cell stage , the BM slides open farther and the π cells loose BM contact in the central region , but maintain contact on the lateral edges of the cell . Scale bars , 5 μm . In B and C , ventral perspective is a max intensity projection of all z-slices with the BM gap edge and the edge of the BM highlighted in merge ( white dashed line ) . DOI: http://dx . doi . org/10 . 7554/eLife . 17218 . 017 Based on our observations , we hypothesized that LIN-29-mediated LIN-12/Notch activation in the π cells may result in transcriptional changes in the π cells that reduce but do not eliminate cell-BM adhesion . During LIN-12/Notch activation , the intracellular domain of LIN-12/Notch is cleaved and enters the nucleus where it forms a complex with the DNA binding protein LAG-1 ( the worm ortholog of CSL ) that promotes expression of LIN-12/Notch target genes ( Greenwald , 2012 ) . Bioinformatic studies have previously identified 163 putative direct targets of LIN-12/Notch that contain clusters of LAG-1 binding sites ( Yoo , 2004 ) . We conducted an RNAi screen of 104 of the 163 presumptive Notch targets to assess BM gap opening . Worms were scored as having a small BM gap diameter if the BM remained in contact with the AC or vulF cells at the P6 . p 8-cell stage—a time when the BM stabilizes over vulD in control animals . Through this screen , we found that RNAi-mediated knock down of ctg-1 , which encodes a Sec14-GOLD protein , a poorly understood group of the Sec14 family of phospholipid transfer proteins ( Mousley et al . , 2007; Saito et al . , 2007 ) , led to a similar BM sliding defect as knock down of the LIN-12/Notch ligand lag-2 ( Supplementary file 1 ) . Furthermore , we found that RNAi-mediated knock down of ctg-1 expression in a uterine-specific RNAi strain also resulted in a BM sliding defect and smaller BM opening ( Figure 7A ) . These results are consistent with the possibility that ctg-1 is a primary target of LIN-12/Notch signaling in the π cells and mediates BM sliding . 10 . 7554/eLife . 17218 . 018Figure 7 . Notch signaling promotes BM sliding through upregulation of the Sec14 family phospholipid transfer protein CTG-1 . ( A ) Quantification of uterine-specific RNAi targeting ctg-1 . Loss of ctg-1 led to a smaller BM gap at the early L4 ( L4440 control , n = 65; ctg-1 RNAi , n = 64 ) . ( B ) DIC and fluorescence images showing lateral views in the π cell nuclei plane ( see Figure 5F ) of animals harboring a genomic insertion of GFP at the ctg-1 locus ( ctg-1 > GFP ) . ctg-1 expression is upregulated in the π cells of wild type worms at the time of Notch mediated induction of π cell fate , which coincides with BM sliding ( n = >5 animals each stage , top ) . Expression of ctg-1 was not induced when the π cell fate was disrupted in a lin-29 mutant ( n = >5 animals each stage; bottom ) . Brackets in DIC images denote the location of the π cells in wild type and non-specified ventral uterine boundary cells in lin-29 ( qy1 ) animals . ( C ) Quantification of fluorescence intensity of the GFP from the ctg-1> GFP genomic locus shows reduced expression in lin-29 mutants relative to wild type ( n = 20 for each ) . Values in the plot are normalized to the wild type mean . The asterisks signify a statistically significant difference . ( D ) A CRISPR-cas9 mediated deletion allele of ctg-1 ( qy11 ) ( lateral central plane; DIC , top; laminin::mCherry , bottom ) has a smaller BM gap ( bracket and arrowheads ) compared to a wild type animal . ( E ) Quantification of BM gap diameter in ctg-1 ( qy11 ) ( n = 18 ) and wild type ( n = 16 ) animals . ( F ) Pan-uterine cell expression of CTG-1 ( fos-1 > ctg-1::GFP ) in a lin-29 ( qy1 ) mutant increased BM gap size ( brackets ) . Lateral central plane; DIC , Top; GFP expression , bottom . ( G ) Quantification of BM gap diameter in lin29 ( qy1 ) ( n = 21 ) and lin-29 ( qy1 ) rescued with fos-1 > ctg-1::GFP ( n = 21 ) . For A , C , E , and G , the asterisks signify a statistically significant difference ( indicates p<0 . 01 and * indicates p<0 . 001 , Wilcoxon rank sum test ) . Scale bars , 5 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 17218 . 01810 . 7554/eLife . 17218 . 019Figure 7—source data 1 . BM gap diameter in L4440 vs . ctg-1 RNAi . DOI: http://dx . doi . org/10 . 7554/eLife . 17218 . 01910 . 7554/eLife . 17218 . 020Figure 7—source data 2 . Fluorescence intensity of GFP from ctg-1 > GFP in wild type vs . lin-29 ( qy1 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 17218 . 02010 . 7554/eLife . 17218 . 021Figure 7—source data 3 . BM gap diameter in wild type vs . ctg-1 ( qy11 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 17218 . 02110 . 7554/eLife . 17218 . 022Figure 7—source data 4 . BM gap diameter in lin-29 ( qy1 ) vs . lin-29 ( qy1 ) ; fos-1 > ctg-1::GFP . DOI: http://dx . doi . org/10 . 7554/eLife . 17218 . 02210 . 7554/eLife . 17218 . 023Figure 7—source data 5 . BM gap diameter in L4440 vs . pifk-1 , C56A3 . 8 , ZC8 . 6 , Y75B8A . 24 , and ppk-1 RNAi . DOI: http://dx . doi . org/10 . 7554/eLife . 17218 . 02310 . 7554/eLife . 17218 . 024Figure 7—figure supplement 1 . LAG-1 binding sites in the ctg-1 promoter and ctg-1 ( qy11 ) schematics . ( A ) A schematic of approximately 3 kb of the 5’ regulatory sequence of ctg-1 . ( B ) The DNA sequence of approximately 3 kb of the 5’ regulatory sequence of ctg-1 . In both ( A ) and ( B ) , gray bars highlight LAG-1 binding site consensus sequences ( YRTGRGAA ) and the yellow bar indicates the start of ctg-1 exon I . ( C ) A schematic of the ctg-1 genomic locus showing the coding region corresponding to the Sec14 domain ( purple brackets ) . ( D ) A schematic of the ctg-1 gene with the homology arms used for CRISPR-cas9 repair vector ( orange bars ) and sgRNA cut site ( red arrows ) . Gray regions represent UTR sequences . ( E ) A schematic of the edited ctg-1 ( qy11 ) GFP knock-in allele . DOI: http://dx . doi . org/10 . 7554/eLife . 17218 . 02410 . 7554/eLife . 17218 . 025Figure 7—figure supplement 2 . CTG-1::GFP expression . Pan-uterine expression of CTG-1::GFP ( fos-1 > ctg-1::GFP ) translational fusion at the P6 . p 8-cell stage ( right ) with DIC ( left ) in wild type worms includes the π cells ( blue dashed outlines ) . DOI: http://dx . doi . org/10 . 7554/eLife . 17218 . 02510 . 7554/eLife . 17218 . 026Figure 7—figure supplement 3 . PI4-kinase PIFK-1 is required for BM sliding . ( A ) Quantification of the BM gap size after treatment with L4440 control and uterine-specific pifk-1 ( yeast Pik1 and vertebrate PI4KIIIβ ) RNAi , C56A3 . 8 ( vertebrate PI4KIIα/β ) RNAi , and ZC8 . 6 ( vertebrate PI4KIIα/β ) RNAi ( n = 34 each ) , ( B ) Quantification of the BM gap size after treatment with L4440 control and uterine-specific Y75B8A . 24 ( vertebrate PI4KIIIa ) ( n = 34 each ) RNAi , and ( C ) Quantification of the BM gap size after treatment with L4440 control and uterine-specific ppk-1 ( vertebrate type I PIP 5-Kinase ) RNAi ( n = 40 each ) revealed that only pifk-1 is required for BM sliding . ( * indicates p<0 . 001 Wilcoxon rank sum test , N . S . indicates no significant difference ( p>0 . 05 ) , Wilcoxon rank sum test ) . DOI: http://dx . doi . org/10 . 7554/eLife . 17218 . 026 The upstream regulatory region of ctg-1 has 19 LAG-1 binding sites ( Figure 7—figure supplement 1 ) , which is why it is predicted to have direct transcriptional regulation by LIN-12/Notch signaling . To determine if ctg-1 is upregulated in the π cells , we created a transcriptional reporter for ctg-1 by replacing the majority of the coding region with GFP using CRISPR/Cas9 genome editing ( Figure 7—figure supplement 1; ctg-1 > GFP; qy11 allele ) ( Dickinson et al . , 2013 ) . Strikingly , we found ctg-1 > GFP was specifically upregulated in the π cells during the time of BM sliding in wild type animals ( Figure 7B ) , but not lin-29 mutants ( Figure 7B , C ) . Furthermore , as this GFP insertion created a ctg-1 loss of function mutation , we examined BM sliding in qy11 worms and confirmed that genetic loss of ctg-1 led to a reduction in BM sliding and gap opening ( Figure 7D , E ) . To assess whether restoration of CTG-1 in the ventral uterine cells in a lin-29 mutant could restore BM sliding , we expressed a full-length GFP translational fusion of ctg-1 under the control of a pan-uterine promoter ( fos-1 > ctg-1::GFP ) in lin-29 mutant animals . CTG-1::GFP predominantly localized to the cytosol of the uterine cells ( Figure 7—figure supplement 2 ) , consistent with reports of Sec14 family protein localization in yeast , plant and mammalian cells ( Schnabl et al . , 2003; Shibata et al . , 2001 ) . Restoration of uterine expression of CTG-1::GFP in lin-29 mutants rescued BM sliding and gap opening ( Figure 7F , G ) . Taken together , these results offer compelling evidence that ctg-1 is a direct and crucial transcriptional target of LIN-29-mediated LIN-12/Notch signaling in the π cells that promotes BM sliding . Sec14 domain-containing genes are a eukaryotic-specific gene family . Six Sec14 family proteins are encoded in the S . cerevisae genome and more than 20 exist in vertebrates ( Bankaitis et al . , 2010 ) . We found that at least 16 Sec14 family members are encoded in the C . elegans genome ( Supplementary file 2 ) , with 12 encoding Sec14-GOLD proteins . Evidence suggests that Sec14 family proteins act as site-specific catalysts of phosphatidylinositol 4-OH kinases that generate phosphatidylinositol-4-phosphate ( PI ( 4 ) P ) ( Bankaitis et al . , 2010; Huang et al . , 2016; Schaaf et al . , 2008 ) . Many Sec14 family proteins are thought to regulate vesicular trafficking , yet very little is known about the precise cellular processes controlled by most of these proteins ( Curwin et al . , 2009; Grabon et al . , 2015; LeBlanc and McMaster , 2010 ) . In S . cerevisiae Sec14 regulates trafficking through the PI4-kinase Pik1 . To determine whether CTG-1 might regulate BM sliding through a PI4-kinase , we examined uterine-cell specific RNAi-mediated knockdown of the four PI4-kinases encoded in the C . elegans genome . We also examined knockdown of the sole C . elegans PI5-kinase , ppk-1 , which uses PI ( 4 ) P to generate PI ( 4 , 5 ) P2 ( Weinkove et al . , 2008 ) . In support of a similar role for CTG-1 as Sec14 , we found that reduction of pifk-1 ( ortholog of Pik1 and vertebrate PI4KIIIβ ) , but not other PI4-kinases or ppk-1 , decreased BM sliding and gap expansion ( Figure 7—figure supplement 3 ) . As Sec14 and Pik1/ PI4KIIIβ are implicated as regulators of vesicle trafficking ( Clayton et al . , 2013; Sciorra et al . , 2005 ) , we hypothesized that ctg-1 might be upregulated in the π cells to alter the trafficking and localization of BM adhesion receptor proteins to promote BM sliding . One of the major BM adhesion receptors in C . elegans is integrin . All integrins are heterodimers composed of a single α and β subunit . C . elegans have two heterodimers , composed of αPAT-2 or αINA-1 subunit bound to the sole β subunit , PAT-3 ( Kramer , 2005 ) . Only the INA-1/PAT-3 heterodimer is expressed in uterine and vulval tissue at the time of uterine-vulval attachment . INA-1/PAT-3 integrin is necessary for the stabilization of the BM over vulD and its loss results in the Pvl phenotype ( Hagedorn et al . , 2009; Ihara et al . , 2011 ) . We did not , however , find any changes in the expression or localization of the functional integrin reporter PAT-3::GFP ( Hagedorn et al . , 2009 ) in the lin-29 ( qy1 ) mutant background , which lacks ctg-1 expression , compared to wild type ( Figure 8—figure supplement 1 ) . Additionally , fluorescence recovery after photobleaching ( FRAP ) experiments indicated that PAT-3::GFP was trafficked to the cell-BM interface normally at the time of BM sliding ( between P6 . p 4-cell and 8-cell stages ) in lin-29 mutants ( Figure 8—figure supplement 1 ) . Finally , RNAi-mediated targeting of ina-1 in the uterine-specific RNAi background did not rescue BM sliding in lin-29 mutants ( BM gap diameter 8 . 13 ± 0 . 16 μm ( L4440 empty vector control , n = 48 ) , 8 . 30 ± 0 . 25 μm [ina-1 RNAi , n = 48] ) . These experiments strongly suggest that CTG-1 does not regulate INA-1/PAT-3 localization in the π cells to promote BM sliding . We next examined the expression and localization of DGN-1 , the sole C . elegans ortholog of the BM receptor dystroglycan . DGN-1 is expressed in the uterine tissue and , like integrin , its loss results in a Pvl phenotype , indicating potential functions in uterine-vulval attachment ( Johnson et al . , 2006 ) . We detected a significant increase in the localization of DGN-1 at the cell-BM interface in the π cells of lin-29 mutants compared to wild type animals ( Figure 8A , B ) . This difference was not a result of increased expression , as the transcriptional reporter for dgn-1 was similarly expressed in lin-29 mutants and wild type ( Figure 8C ) . In addition , the localization of DGN-1 in regions beyond the edges of the BM gap ( over the vulD cell ) were similar in lin-29 and wild type , indicating that the alteration in DGN-1 localization was specific to the region of BM sliding near the AC ( Figure 8—figure supplement 2 ) . FRAP experiments indicated that DGN-1 recovered more quickly after photobleaching in lin-29 mutants versus wild type animals , which may account for the increased DGN-1::GFP localization ( Figure 8F , G ) . Faster recovery of DGN-1 at the cell-BM interface in lin-29 mutants could indicate more rapid cell surface trafficking , slower endocytosis/removal , or faster lateral plasma membrane mobility . Notably , the number of DGN-1 containing Rab-7 marked late endosome vesicles in lin-29 mutants and wild type animals was similar , suggesting that CTG-1 might not direct rapid removal of DGN-1 from the cell surface into the endosomal system ( Figure 8—figure supplement 3 ) . There was , however , a reduction of F-actin at the basal surface in the π cells in lin-29 mutants ( Figure 8—figure supplement 4 ) . As F-actin is thought to act as a barrier to exocytosis ( Eitzen , 2003 ) , these observations support the idea of increased DGN-1 delivery to the cell surface . Given the association of the GOLD domain , Sec14 proteins , and PI4-kinase regulation with vesicle trafficking , ( Anantharaman and Aravind , 2002; Clayton et al . , 2013; Mousley et al . , 2007 ) , CTG-1 may direct a specific trafficking pathway that restricts the delivery of DGN-1 to the cell-BM interface . Importantly , these observations do not rule out the possibility that CTG-1 enhances endocytic removal from the cell surface at levels that we were unable to detect or that it regulates the membrane mobility of DGN-1 to limit DGN-1 at the cell-BM interface . 10 . 7554/eLife . 17218 . 027Figure 8 . LIN-29/CTG-1 restricts dystroglycan ( DGN-1 ) levels and trafficking at the π cell-BM interface . ( A ) A functional translational fusion of the BM-receptor DGN-1 ( DGN-1::GFP , top ) reveals that more DGN-1 localizes on the surface of the ventral uterine cells ( blue dashed outlines ) at the edges of the BM gap in a lin-29 mutant compared to a wild type animal . Enlargement of boxes in top images ( bottom ) shows regions ( red bars ) where DGN-1 levels were measured in ( B ) . ( B ) Quantification of the fluorescence intensity of a functional DGN-1::GFP translational reporter ( dgn-1 > dgn-1::GFP ) at the BM gap boundary in the π cells in wild type and lin-29 ( qy1 ) animals ( n = 22 for each ) . Values in plot are normalized to the wild type mean ( indicates p<0 . 01 , Wilcoxon rank sum test ) . ( C ) Quantification of the fluorescence intensity of a dgn-1 transcriptional reporter in wild type and lin-29 ( qy1 ) animals ( n = 21 for each ) revealed no significant difference in expression levels ( p>0 . 05 Wilcoxon rank sum test ) . Values in plot are normalized to the wild type mean . ( D ) Uterine-specific dgn-1 RNAi rescued the BM gap ( brackets ) defect in a lin-29 ( qy1 ) background expressing BM and AC reporters ( laminin::GFP and cdh-3 > mCherry::PLCδPH ) . ( E ) BM gap diameters in uterine-specific RNAi knockdown of dgn-1 ( n = 50 ) and L4440 control ( n = 63 ) in the lin-29 mutant background ( * indicates p<0 . 001 , Wilcoxon rank sum test ) . ( F ) Fluorescence recovery after photobleaching ( FRAP ) of a 1 . 0-µm region of DGN-1::GFP along the basal surface of the uterine π cells during the late the early P6 . p 6-cell stage ( time of BM sliding ) in wild type and lin-29 ( qy1 ) mutant animals . DGN-1::GFP recovers more quickly in the lin-29 mutant . ( G ) Graph reports DGN-1::GFP recovery in wild type and lin-29 ( qy1 ) mutants ( n = 9 animals for each ) . Error bars represent standard deviation and the asterisks signify a statistically significant difference at comparable time points ( p<0 . 05 Student’s t-test ) . T1/2 denotes time to recovery of 50% of initial fluorescence signal . All images lateral , central plane . Scale bars , 5 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 17218 . 02710 . 7554/eLife . 17218 . 028Figure 8—source data 1 . Fluorescence intensity of DGN-1::GFP in wild type vs . lin-29 ( qy1 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 17218 . 02810 . 7554/eLife . 17218 . 029Figure 8—source data 2 . Fluorescence intensity of GFP from a dgn-1 > GFP transcriptional reporter in wild type vs . lin-29 ( qy1 ) animals . DOI: http://dx . doi . org/10 . 7554/eLife . 17218 . 02910 . 7554/eLife . 17218 . 030Figure 8—source data 3 . BM gap diameter in L4440 vs . dgn-1 RNAi in lin-29 ( qy1 ) background . DOI: http://dx . doi . org/10 . 7554/eLife . 17218 . 03010 . 7554/eLife . 17218 . 031Figure 8—source data 4 . Fluorescence intensity recovered over time in DGN-1::GFP FRAP . DOI: http://dx . doi . org/10 . 7554/eLife . 17218 . 03110 . 7554/eLife . 17218 . 032Figure 8—source data 5 . Fluorescence quantifications for pat-3 transcriptional and translational markers . DOI: http://dx . doi . org/10 . 7554/eLife . 17218 . 03210 . 7554/eLife . 17218 . 033Figure 8—source data 6 . Fluorescence intensity recovered over time in PAT-3::GFP FRAP . DOI: http://dx . doi . org/10 . 7554/eLife . 17218 . 03310 . 7554/eLife . 17218 . 034Figure 8—source data 7 . Fluorescence intensity of DGN-1::GFP in wild type vs . lin-29 ( qy1 ) over vulD ( past region of BM sliding ) . DOI: http://dx . doi . org/10 . 7554/eLife . 17218 . 03410 . 7554/eLife . 17218 . 035Figure 8—source data 8 . RAB-7 and DGN-1 co-localization in wild type vs . lin-29 ( qy1 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 17218 . 03510 . 7554/eLife . 17218 . 036Figure 8—source data 9 . Fluorescence intensity of moesinABD::mCherry in wild type vs . lin-29 ( qy1 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 17218 . 03610 . 7554/eLife . 17218 . 037Figure 8—figure supplement 1 . Integrin expression and localization are unaffected by LIN-29/CTG-1 . ( A ) An integrin translational reporter ( PAT-3::GFP ) localized similarly at the edges of the BM gap in lin-29 mutant ( right ) compared to a wild type ( left ) animal . Enlargement of boxes in top images ( bottom ) shows regions in π cells ( white bars ) where PAT-3::GFP levels were measured in ( C ) . ( B ) Quantification of the fluorescence intensity of a pat-3 transcriptional reporter in wild type and lin-29 ( qy1 ) animals ( n = 21 for each ) revealed no significant difference in expression levels ( p>0 . 05 Wilcoxon rank sum test ) . Values in the plot are normalized to the wild type mean . ( C ) Quantification of the fluorescence intensity of a PAT-3 translational reporter at the BM gap boundary in the π cells in wild type and lin-29 ( qy1 ) animals ( n = 10 for wild type , 11 for lin-29 ) revealed no significant difference in expression levels ( p>0 . 05 Wilcoxon rank sum test ) . Values in the plot are normalized to the wild type mean . ( D ) Quantification of the fluorescence intensity of a PAT-3 translational reporter at the BM gap boundary in the π cells in which GFP signal was bleached from the VPCs and remaining fluorescence intensity was measured ( n = 5 for wild type , 7 for lin-29 ) revealed no significant difference is expression levels ( p>0 . 05 Wilcoxon rank sum test ) . ( E ) Fluorescence recovery after photobleaching ( FRAP ) of a 1 . 0-µm region of PAT-3::GFP along the basal surface of the uterine π cells during the early P6 . p 6-cell stage ( time of BM sliding ) in wild type and lin-29 ( qy1 ) mutant animals revealed similar rates of PAT-3::GFP recovery . ( F ) Graph reports PAT-3::GFP recovery in wild type ( n = 13 ) and lin-29 ( qy1 ) mutants ( n = 12 ) . Error bars represent standard deviation . There was no significant difference in PAT-3 recovery at comparable time points ( p>0 . 05 Students t-test ) . T1/2 denotes time to recovery of 50% of initial fluorescence signal . Scale bars , 5 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 17218 . 03710 . 7554/eLife . 17218 . 038Figure 8—figure supplement 2 . Loss of LIN-29/CTG-1 does not affect dystroglycan ( DGN-1 ) localization in regions beyond the BM gap interface . Quantification of the fluorescence intensity of a DGN-1 translational reporter ( dgn-1 > dgn-1::GFP ) at a 1 µm region in uterine cells opposite vulD in wild type and lin-29 ( qy1 ) animals ( n = 20 each ) . No significant difference was observed ( p>0 . 05 , Wilcoxon rank sum test ) . DOI: http://dx . doi . org/10 . 7554/eLife . 17218 . 03810 . 7554/eLife . 17218 . 039Figure 8—figure supplement 3 . Loss of LIN-29/CTG-1 does not affect DGN-1 localization to late endosomes . ( A ) RAB-7::GFP ( green ) and DGN-1::mCherry ( magenta ) co-localize to a similar degree in wild type ( left ) and lin-29 ( qy1 ) ( right ) mutants . White dashed outline indicates the position of the π cells . Arrows indicate vesicles with co-localization . Gray boxes in merge indicate area of magnification ( far right panels ) . Scale bar , 5 µm . ( B ) Quantification of the total number of RAB-7::GFP marked vesicles indicate a similar number of late endosomes in wild type and lin-29 ( qy1 ) ( n = 8 each , error bars SEM , not significant ( p>0 . 05 , Student’s t-test ) ) . ( C ) Quantification of co-localization of DGN-1::mCherry and RAB-7::GFP in wild type and lin-29 ( qy1 ) ( n = 8 each , error bars SEM , not significant ( p>0 . 05 , Student’s t-test . ) ) . DOI: http://dx . doi . org/10 . 7554/eLife . 17218 . 03910 . 7554/eLife . 17218 . 040Figure 8—figure supplement 4 . F-actin is reduced at the cell-BM interface with loss of LIN-29/CTG-1 . ( A ) moesinABD::mCherry expression in wild type ( left ) and lin-29 ( qy1 ) ( right ) shows that decreased levels of F-actin at the BM gap edge accompany reduced BM sliding . White boxes indicate areas magnified in lower panels . In lower panels , regions of interest are highlighted: yellow indicates a region of the basal surface on the BM gap edge , blue indicates a region on the apical surface of the cell , and magenta indicates a region of the basal surface beyond the boundary of the sliding BM . Scale bar , 5 μm . ( B ) Quantification of moesinABD::mCherry expression in wild type and lin-29 ( qy1 ) ( n=20 each ) . Measurements were made on a 0 . 3-µm x 0 . 3-µm region as indicated by the boxes in the lower panels of ( A ) . Fluorescence intensities are normalized to the expression level of a region along the apical membrane in the same cell . * indicates p<0 . 001 Wilcoxon rank sum test . DOI: http://dx . doi . org/10 . 7554/eLife . 17218 . 040 We next determined if the increase of DGN-1 at the cell-BM interface in the lin-29 mutant could be responsible for stronger adhesion of the uterine cells to the BM , causing BM sliding to fail . Consistent with this idea , reduction of dgn-1 expression via uterine specific RNAi in the lin-29 ( qy1 ) mutant restored BM sliding and gap expansion ( Figure 8D , E ) . Taken together , these results suggest that LIN-29 activity in the AC leads to LAG-2/LIN-12 ( Notch ) -mediated upregulation of ctg-1 expression in the uterine π cells that restricts the cell surface trafficking and accumulation of the BM receptor DGN-1 , which facilitates BM sliding during uterine-vulval attachment . The shifting of cell-BM interfaces is crucial in many diverse morphogenetic events , including intestinal epithelial renewal , branching morphogenesis , and BM deposition ( Clevers , 2013; Glentis et al . , 2014; Morrissey and Sherwood , 2015 ) . How BM sliding is controlled , however , remains poorly understood . We show here that the AC in C . elegans further widens a breach it creates in BM during uterine-vulval attachment by promoting BM sliding in neighboring uterine cells . LIN-29 maintains the expression of the Notch ligand LAG-2 in the AC , which leads to the activation of Notch signaling and induction of uterine π cell fate in the neighboring cells that sit over the nascent BM gap boundary . Notch activation in turn leads to upregulation of the Sec14 family phospholipid transfer protein CTG-1 in these BM gap boundary cells , which restricts trafficking of the receptor DGN-1 ( dystroglycan ) to the cell-BM interace . Our data suggest that the proper regulation of DGN-1 trafficking modulates cell-BM adhesion and allows the BM slide , which further widens the BM breach . This work reveals a new morphogenetic signaling pathway that promotes cell-BM sliding ( See Summary Figure 9 ) . 10 . 7554/eLife . 17218 . 041Figure 9 . Summary of uterine boundary cell regulation of BM sliding . ( A ) A schematic diagram of the mid L3 stage in wild type ( left ) . The Kruppel-family EGR protein LIN-29 acts in the AC to maintain LAG-2 ( Notch ligand ) expression , which induces the π cell fate in the six VU cells that contact the AC . This specification event fails to occur in lin-29 mutants ( right ) . ( B ) During the late L3 stage after AC invasion , invagination , growth , and division of the vulval cells generates forces that promote BM sliding in wild type animals ( left ) . Within the π cells ( inset ) , LAG-2 binding of LIN-12 ( Notch ) activates transcriptional upregulation of ctg-1 , which encodes a Sec14-GOLD phospholipid transfer protein . CTG-1 restricts BM receptor DGN-1 ( dystroglycan ) trafficking to the cell-BM interface of the π cells , thus decreasing their adhesion to the BM and allowing BM sliding . In lin-29 mutants ( right ) ctg-1 expression is not upregulated in the unspecified ventral uterine descendant cells , and more DGN-1 is trafficked to the cell-BM interface , maintaining cell adhesion to the BM and preventing sliding . ( C ) At the early-to-mid L4 , the BM gap edges have slid ( arrowheads ) and stabilize over the vulval D cells and at the edges of uterine π cells in wild type animals ( left ) . In contrast , the BM does not slide in lin-29 mutants and maintains its position adjacent to the AC ( right ) . DOI: http://dx . doi . org/10 . 7554/eLife . 17218 . 041 During cell-BM sliding , the cells , the BM , or both shift position in relationship to each other . During uterine-vulval attachment in C . elegans , the linked ventral and gonadal BMs slide over vulval and uterine cells to expand the BM gap , and the vulval cells also appear to move ( through migration and invagination ) inside the gap to form direct contacts with uterine cells ( Ihara et al . , 2011; Schindler and Sherwood , 2013 ) . It has previously been shown that vulval cell divisions at the BM gap boundary controls BM sliding ( Matus et al . , 2014 ) . During vulval invagination and growth , the centrally located vulE and F cells divide and reduce their contact with the BM during cell rounding , allowing the BM to slide over these cells . The BM ultimately ceases sliding and the BM gap boundary stabilizes on the non-dividing vulD cell . Blocking the divisions of vulE and F halts BM sliding prematurely resulting in a narrower BM gap , while inducing vulD divisions causes the BM to slide further and enlarges the opening ( Matus et al . , 2014 ) . Here we have found that uterine cells at the BM gap boundary regulate BM movement through a distinct mechanism independent of cell division . Uterine π cell divisions are not coordinated with vulval cell divisions during BM movement . Further , blocking uterine π cell divisions did not halt BM sliding . Instead , through a forward genetic screen , site of action , and expression studies , our data indicate that the invading AC directs a morphogenetic signaling program that promotes BM sliding in neighboring uterine cells: ( 1 ) the AC expresses the transcription factor LIN-29 , ( 2 ) LIN-29 activates Notch signaling and induction of the π cell fate in neighboring uterine cells ( Newman et al . , 2000 ) , ( 3 ) as a part of π cell fate induction , Notch activation upregulates ctg-1 , which restricts trafficking of the BM adhesion receptor DGN-1 ( dystroglycan ) to the cell-BM interface and allows BM sliding . Together , these observations suggest that after AC invasion , the uterine π cells loosen BM adhesion , allowing the BM to slide along the π cells to a precise site determined by the dividing vulE and F cells and non-dividing vulD . This model is supported by the observed flexibility of BM boundary position , which can be adjusted by manipulating vulval cell divisions . The distinct , but temporally coordinated , mechanisms controlling BM movement in the uterine and vulval cells precisely position the BM gap boundary to ensure robust uterine-vulval attachment . Very few examples of gene regulatory networks guiding morphogenetic cellular behaviors leading to cell fate specification are known ( Christiaen et al . , 2008 ) . Using a focused RNAi screen we found that ctg-1 , which encodes a Sec14 family phosphatidylinositol-transfer protein ( PITP ) ( Tripathi et al . , 2014 ) , is a crucial target of Notch signaling during π cell fate specification that promotes BM sliding . Consistent with ctg-1 being a direct transcriptional target of Notch , the ctg-1 upstream regulatory region contains 19 predicted binding sites for the Notch transcriptional activator LAG-1 ( Suppressor of Hairless ) ( Yoo , 2004 ) . Furthermore , ctg-1 expression is upregulated in the π cells at the time of activated LIN-12 ( Notch ) signaling , and its expression is dependent on LIN-29-directed Notch signaling . In addition , uterine-cell specific RNAi and genetic loss of ctg-1 reduced BM sliding , which phenocopies loss of Notch signaling in these cells . Sec14 family PITPs are unlikely to be true carriers of phospholipids . Instead , studies suggest that Sec14 family PITPs function as scaffolds or nanoreactors that present phosphatidylinositol to phosphatidylinositol-4 kinase ( PI4K ) to regulate specific protein and membrane trafficking pathways in the trans-Golgi and endosomal recycling system ( Cockcroft and Garner , 2011; Curwin , 2013; Curwin et al . , 2009; Mousley et al . , 2008; Schaaf et al . , 2008; Tripathi et al . , 2014 ) . Consistent with this notion , our work also indicates that the C . elegans PI4K , pifk-1 , is required in the uterine cells to promote BM sliding . The Sec14 family has expanded significantly in multicellular eukaryotes . Whereas yeast have six Sec14 family members , we found C . elegans has at least 16 and vertebrates have at least 20 ( Bankaitis et al . , 2010 ) . Little , however , is known about the cellular process controlled by many these proteins ( Grabon et al . , 2015 ) . Our work suggests that CTG-1 regulates the trafficking of DGN-1 ( dystroglycan ) to the cell surface and promotes de-adhesion from BM to control BM sliding . Notably , localization and photobleaching studies indicate that the integrin heterodimer INA-1/PAT-3 , another key BM receptor , was unaffected by loss of ctg-1 , suggesting that CTG-1 function is highly specific . Loss of LIN-29-mediated Notch signaling ( and thus ctg-1 expression in the uterine cells ) led to an increase of DGN-1 on the surface of uterine cells , but did not alter dgn-1 gene expression . Photobleaching experiments in wild type and lin-29 mutants ( lack upregulation of ctg-1 ) indicated that ctg-1 expression in the π cells restricts the rate of DGN-1 trafficking to the cell-BM interface . Interestingly , in CTG-1 the Sec14 domain is paired with a GOLD ( Golgi Dynamics ) domain , which , though its function has yet to be completely characterized , may regulate the selection of proteins being trafficked from the Golgi , thus affecting their secretion to the plasma membrane or routing to the endosome-lysosomal system ( Anantharaman and Aravind , 2002; Schimmoller et al . , 1995 ) . These results suggest that CTG-1 might reduce cell-BM adhesion by regulating a vesicular pathway with specificity for DGN-1 , possibly slowing DGN-1 delivery to the cell surface or directing it to the endosome-lysosome system . The function of Sec14-GOLD proteins is poorly understood , and CTG-1 does not show strong localization to any specific domain in the uterine π cells . Thus , we cannot rule out possible functions for CTG-1 in mediating increased removal of DGN-1 from the cell surface or slowing lateral plasma membrane trafficking to restrict DGN-1 at the cell-BM interface . While the specific mechanism by which CTG-1 restricts DGN-1 trafficking remains unclear , we found that reduction of DGN-1 restored BM sliding in the absence of ctg-1 , strongly supporting the functional significance of reduced localization and trafficking of DGN-1 . The two most prominent BM receptors that transduce signals and link the cells’ cytoskeleton to the BM are integrin heterodimers and dystroglycan ( Bello et al . , 2015; Yurchenco , 2011 ) . Unlike mammals , which construct 24 known αβ integrin heterodimers , C . elegans make only two , and only one of these , αINA-1/βPAT-3 , is expressed with DGN-1 ( dystroglycan ) in the uterine π cells ( Hagedorn et al . , 2009; Ihara et al . , 2011 ) . Integrin and dystroglycan are co-expressed widely , and likely have many distinct roles , such as having opposing effects on extracellular-regulated kinase ( ERK ) activation ( Ferletta et al . , 2003 ) . However , how these proteins interact and coordinate functions in regulating cell-BM communication and adhesion has been challenging to elucidate as a result of tissue complexity and the large family of integrin heterodimers in vertebrates ( Bello et al . , 2015; Nakaya et al . , 2013 ) . Interestingly , our data suggest that INA-1/PAT-3 is not downregulated to promote BM sliding in the uterine π cells . Further , unlike loss of dgn-1 , uterine-specific loss of ina-1 did not rescue BM sliding in the lin-29 mutant background . Our results indicate that DGN-1 is the primary BM adhesion receptor that controls cell-BM sliding in the uterine π cells , while integrin appears to have a function ( s ) independent of BM sliding in these cells . These results suggest that different BM receptors may allow cells to have diverse and concurrent signaling and adhesive interactions with BM . While dystroglycan has been most extensively studied in its association with the dystrophin complex and muscular dystrophies , it is broadly expressed and implicated in regulating many cellular processes , including BM assembly , cell migration , axon outgrowth , retinal layering , and Schwann cell wrapping ( Bello et al . , 2015; Moore and Winder , 2010 ) . Notably , dystroglycan is also required for branching morphogenesis in glandular epithelium , a process that is thought to require BM sliding ( Durbeej et al . , 2001; Harunaga et al . , 2014 ) . Further , it has been shown that loss of basally localized dystroglycan during chick gastrulation promotes breakdown of BMs during the epithelial-to-mesenchymal transition that forms mesoderm ( Nakaya et al . , 2011 ) . Thus , dystroglycan may have a fundamental role in mediating BM remodeling events during development . Importantly , loss or reduction of dystroglycan is also emerging as a common event in the progression of epithelial derived cancers , including breast , prostate , colon , cervical , and renal adenocarcinomas ( Cross et al . , 2008; Esser et al . , 2013; Sgambato et al . , 2007 , 2003 , 2006 ) . As regulated loss or reduction of dystroglycan during development creates and widens BM gaps through BM loss and sliding , uncontrolled loss of dystroglyan during cancer progression may lead to the breakdown of BM barriers to facilitate tumor spread . Thus , understanding mechanisms that control dystroglycan levels at the cell surface are not only important in understanding tissue morphogenesis , but may provide new strategies to halt cancer progression . Worms were reared under standard conditions at 15°C , 20°C , or 25°C ( Brenner , 1974 ) . N2 Bristol strain was used as wild-type nematodes . Strains were reared and viewed at 20°C or 25°C using standard techniques . In the text and figures , we use a '>' symbol for linkages to a promoter and use a '::' symbol for linkages that fuse open reading frames ( Ziel et al . , 2009 ) . The following alleles and transgenes were used: qyEx439 [egl-13 > cki-1::GFP] , cgEx308 [dgn-1 ( + ) , rol-6 ( su1066 ) , dgn-1 > GFP] , qyEx557 [fos-1 > moesinABD::mCherry , myo-2 > GFP] , qyEx558 [fos-1 > moesinABD::mCherry , myo-2 > GFP] , qyEx561 [fos-1 > rab-7::GFP , dgn-1 > dgn-1::mCherry , unc-119 ( + ) ] , qyIs28[ced-10::GFP] , qyIs43[pat-3 > pat-3::GFP , genomic ina-1] , qyIs102 [fos-1 > rde-1; myo-2 > GFP] , qyIs108 [laminin::Dendra] , qyIs127 [laminin::mCherry] , qyIs251 [cdh-3 > lin-29a::GFP] , qyIs330[laminin::mCherry] , qyIs351 [unc-62 > GFP::CAAX] , qyIs361[lin-29a/b > GFP] , qyIs486 [dgn-1 > dgn-1::GFP] , qyIs508 [fos-1 > ctg-1::GFP]; LGI , ctg-1 ( qy11 ) ; rhIs2 [pat-3::GFP]; LGII , lin-29 ( ga94 ) , lin-29 ( qy1 ) , rrf-3 ( pk1426 ) ; LGIV , qyIs8 [laminin::GFP]; LGV , kuIs29 [egl-13::GFP ( pWH17 ) + unc-119 ( + ) ] , rde-1 ( ne219 ) ; LGX , sel-12 ( ty11 ) , qyIs24 [cdh-3 > mCherry::PLCδPH] , qyIs86 [egl-13 > GFP] . Because the protruding vulva of qy1 homozygotes prevents male mating , the lin-29 ( qy1 ) -containing chromosome was balanced by the balancer mIn1[mIs14 dpy-10 ( e128 ) ] for effective mating ( Edgley and Riddle , 2001 ) . L4 N2 Bristol strain worms were mutagenized in 4 ml M9 buffer with N-ethyl-N-nitrosourea ( ENU ) at a concentration of 0 . 5 mM in a 15 ml conical tube . The tube was placed on the rocker for 4 hr at room temperature then subject to centrifugation to collect worms for recovery . Worms were washed 3 times with 3 ml M9 buffer and resuspended in a few drops of M9 , then transferred to a plate using a glass pipette . One h later , healthy worms were transferred to new plates at a density of 2 worms/plate . Five rounds of mutagenesis were performed . F2 progeny produced from the mutagenized worms were examined for the protruding-vulva ( Pvl ) phenotype . The fertile Pvl mutants were then subject to microscopic examination for defects in uterine-vulval attachment . During this examination , 10 fertile Pvl mutants were identified with defects in BM remodeling during uterine-vulval attachment . Nine of these had defects in AC invasion , and one , qy1 , had a defect in the BM hole opening after invasion , suggesting that mutants that affect BM hole opening are rarer than those that alter AC invasion . Of the nine mutants with defects in AC invasion , four were mapped further . One AC invasion mutant is putative allele of unc-6 and another an allele other unc-40 , genes that are known to promote AC invasion ( Ziel et al . , 2009 ) . A single-nucleotide-polymorphism ( SNP ) based mapping strategy ( Davis et al . , 2005 ) was used to locate the qy1 mutation to between 11777294 and 12029092 on Chromosome II . Briefly , three hundred F2 Pvl progeny of multiple qy1/CB4856 F1 hermaphrodites that were produced by crossing CB4856 males into qy1/mIn1 were individually isolated onto each well of 24-well plates . After four days , self-progeny from these F2 progeny were washed off the plate using water ( >30 worms/plate ) and placed in a single well of a 96-well plate . Worms were allowed to settle to the bottom of the wells at 4°C for 15 min ( min ) . Excess solution was removed to leave a final volume of 90 µl/well . Plates were stored at −80°C . For mapping PCR , the plates were thawed and 1X lysis buffer ( 50 mM KCl , 2 . 5 mM MgCl2 , 10 mM Tris pH 8 . 3 , 0 . 45% Tween 20 , 0 . 04% gelatin , 100 µg/ml proteinase K ( freshly added ) ) was added . Worms were lysed by incubation at 65°C , 1 hr and 95°C , 15 min to extract DNA . PCR templates were frozen at −80°C and thawed prior to each use . For each PCR , 1 µl of worm lysate was added into each well of the 96-well plate that then received 19 µl of a PCR mix containing 14 µl water , 2 µl 10X buffer , 0 . 4 µl 10 mM dNTP , 2 µl each primer ( 10 µM ) , and 0 . 2 µl Taq ( 5 units/µl ) . PCR conditions: 93°C , 2 min , 35 cycles ( 93°C , 20 s ( s ) , 58°C , 30 s , 72°C , 1 min ) , 72°C , 5 min , 10°C for holding . The restriction enzyme DraI 0 . 26 µl ( 20 unit/µl ) with its 10x buffer was added into each PCR well to make final digestion volume of 22 . 26 µl . Within the region identified , 4 genes were previously reported to have a Pvl phenotype: ZK930 . 3 , lin-29 , ash-2 , and mcm-2 , which we confirmed by RNAi . qy1 was identified as lin-29 when it failed to complement lin-29 ( n546 ) . Transgenic animals expressing laminin::Dendra were photoconverted using a Zeiss LSM 510 confocal microscope ( Zeiss Microimaging ) , equipped with a 63x objective , scanning regions of interest with a 405 nm laser at 1 mW power for 30 s . After photoconversion , images were captured using a spinning disc confocal microscope . Animals were recovered from the agar pad , left to develop on the plates with OP50 at 25°C for the specified amount of time then reimaged . Identical settings were used to acquire images at all times . Acquired images were processed using Imaris ( Bitplane ) . Laser-induced cell ablations of the AC and VU were performed as previously described ( Bargmann , 1995 ) . ACs were ablated at the P6 . p 4-cell stage after clearing of the BM by the AC was complete . For ventral uterine ( VU ) cell ablations , the 2 VUs that flanked the AC on either the anterior or posterior side ( Z4 . aaa and Z1 . ppa or Z4 . aap and Z1 . ppp ) were ablated at the early P6 . p 2-cell stage prior to π cell specification . Worms were recovered from slides to OP50 plates at 25°C and imaged at the L4 stage . Images were acquired using a Zeiss AxioImager microscope with a 100x Plan-APOCHROMAT objective and equipped with a Yokogawa CSU-10 spinning disc confocal controlled by iVision ( Biovision Technologies ) or Micromanager software , or using a Zeiss AxioImager microscope with a 100x Plan-APOCHROMAT objective and with a Zeiss AxioCam MRm CCD camera controlled by Axiovision software ( Zeiss Microimaging ) . Acquired images were processed using ImageJ 1 . 40 ImageJ 1 . 47d and Photoshop CS3 Extended or Photoshop CC ( Adobe Systems ) . Fluorescence intensities were measured as Mean Gray Values using ImageJ . Three-dimensional reconstructions were built from confocal z-stacks , analyzed and exported using Imaris 7 . 4 ( Bitplane ) . The diameter of the BM opening along the anterior-posterior axis was measured as the size of the opening using single-channel images of laminin::GFP , laminin::mCherry or the phase dense line of the BM seen by DIC in plane of focus for the AC using ImageJ . For CKI-1 π division arrest , examination of the number of π cell nuclei by DIC as well as GFP expression in the π cells were used to determine whether π cell division had occurred , and only animals in which all 6 π cells failed to divide were scored . Standard molecular biology and transgenic techniques were used to generate PCR fusion products ( Hobert , 2002 ) , plasmids , and transgenic animals ( Mello and Fire , 1995 ) . To generate the transcriptional reporter for the lin-29a/ b gene , the 5’ cis-regulatory element ( 5’ CRE ) 5 . 4 kb upstream of the ATG start codon of the lin-29a/ b gene was amplified . The 5’ CRE sequence was then fused in frame to GFP coding sequence ( PCR amplified from the vector pPD95 . 81 ) using PCR fusion . For the transcriptional reporter of the lin-29c gene , the 5’ CRE 4 . 2 kb upstream of lin-29c coding sequence was PCR amplified and fused in frame to GFP coding sequence ( PCR amplified from the vector pPD95 . 81 ) using PCR fusion . AC-specific lin-29a::GFP was generated by fusing GFP coding sequence with the amplicon cdh-3 > lin-29a using PCR fusion . To generate the amplicon cdh-3 > lin-29a , lin-29b cDNA was first PCR amplified and subcloned into the vector pPD95 . 81 containing the cdh-3 promoter ( Sherwood et al . , 2005 ) at XhoI and HindIII sites . The coding sequence of cdh-3 > lin-29a was PCR amplified from the vector cdh-3 > lin-29b using the New England Laboratory site-directed-mutagenesis protocol with a primer containing a sequence encoding four lin-29a specific amino acids . To generate the egl-13 > cki-1::GFP construct , the 5 kb promoter region of egl-13 was amplified from plasmid pWH17 and fused with to CKI-1::GFP amplified from the recombineered fosmid WRM0626bF02 ( wTRG5 . 1_2491680425634929_G10 ) . To generate the DGN-1 translational reporter , 2 . 6 kb upstream of the start codon to the end of the coding sequence of dgn-1 was amplified from N2 genomic DNA and fused in frame to the GFP coding sequence amplified from vector pPD95 . 75 , based on a previously published construct that rescues the sterility phenotype of dgn-1 ( Johnson et al . , 2006 ) . To generate fos-1 promoter containing constructs , pBlueScript containing the fos-1 promoter ( Sherwood et al . , 2005 ) was digested with ApaI and BamHI , then fused with the ctg-1 coding sequence amplified from genomic DNA and the GFP coding sequence amplified from pPD95 . 75 for fos-1 > ctg-1::GFP , the moesin::ABD coding sequence amplified from pJWZ6 for fos-1 > moesinABD::mCherry , and the rab-7::GFP coding sequence amplified from pHD69 for fos-1 > rab-7::GFP , using NEBuilder HiFi DNA Assembly Cloning . Transgenic worms were created by co-injection of expression constructs with the transformation marker pPD#MM016B ( unc-119 ) , or the co-injection marker ( myo-2 > GFP ) or both into the germline of unc-119 ( ed4 ) mutants or wild type worms . These markers were injected with EcoRI-digested salmon sperm DNA and pBluescript II at 50 ng/µl as carrier DNA along with the expression constructs , which were normally injected at 10–50 ng/µl . Integrated strains were generated as described previously ( Inoue et al . , 2002 ) . To generate Y75B8A . 24 RNAi , the first 800 bp of the coding sequence of the gene were amplified from genomic DNA and inserted into the vector L4440 , which was digested with HindIII and KpnI . A GFP knock-in that removes the majority of the coding region of ctg-1 was generated using the CRISPR-Cas9 system ( Dickinson et al . , 2013 ) . Two sgRNAs with sequences 5’- AATTCGACAACTACTATTCG-3’ and 5’-AATATTGTAGCGCAATTGCT-3’ were used to induce double stranded breaks in the final exon of ctg-1 . A repair template was constructed with 1 . 5 kb upstream of the coding start site plus the first 20 bp of ctg-1 , followed by GFP , cbr-unc-119 , and 1 . 7 kb of sequence downstream of the ctg-1 stop codon . Plasmids containing guide RNAs and Cas9 were co-injected at a concentration of 25 ng/µL each , and the repair template was injected at a concentration of 10 ng/µL . A Zeiss LSM 510 laser scanning confocal microscope with a 100x objective was used for initial bleaching and imaging the DGN-1::GFP reporter . In the plane of focus for the AC , a 1 μm by 1 μm region of interest was defined at the edge of the BM gap or next to the boundary of the AC if no visible gap in the BM was present . Regions of interest were bleached at 488 nm for 100 iterations at 100% power . Further imaging was conducted every 30 s for 9 min . The protocol was repeated for DGN-1::GFP and PAT-3::GFP using a Nikon Ti-U with the ‘Plan Apo VC’ 100x objective , Yokogawa spinning disc unit , Andor iXon EM-CCD , and Andor ALC601 series laser . ROIs were bleached for 20 iterations at 10% intensity . dsRNA was delivered via feeding to synchronyzed L1 larvae . Phenotypes were subsequently scored at the early L4 stage . All RNAi experiments examining BM gap opening were conducted in a blinded manner , where the treatment was masked from person quantifying the opening . Constructs were obtained from the Vidal ( Rual et al . , 2004 ) and Ahringer ( Simmer et al . , 2003 ) libraries . The L4440 empty vector was used as a negative control . Clones targeting lag-2 , ctg-1 , ina-1 , dgn-1 were verified by sequencing . Screening of Notch targets was scored in rrf-3 worms containing laminin::mCherry and the AC marker ( CAAX::GFP ) . Uterine-specific RNAi was scored in strains containing RDE-1 expressed under the control of the fos-1a promoter in an rde-1 mutant background ( Hagedorn et al . , 2009 ) . Data was collected in 3 independent trials for uterine-specific lag-2 RNAi and dgn-1 RNAi in the lin-29 ( qy1 ) background , and 2 trials for ctg-1 RNAi and PI4-kinase RNAi . Fluorescence intensities were measured as the mean gray values of 0 . 3- µm x 0 . 3- µm regions of interest selected either adjacent to the edge of the expanding BM gap or no longer in contact with the BM ( past the zone of BM sliding ) in the π cells at the P6 . p 6-cell stage . In lin-29 ( qy1 ) mutants , only areas adjacent to the edge of the BM gap were measured , as regions beyond the zone of BM sliding were not visible ( due to the loss of BM sliding ) . To account for the varied expression levels seen in extrachromosomal array lines , fluorescence intensities were normalized using the fluorescence intensity of an apical region of the same size . RAB-7::GFP positive vesicles were selected and marked using ImageJ . Regions of interest were then overlaid on images of corresponding DGN-1::mCherry and assessed for overlap . Textpresso homology/orthology data and Gene ontology and BLAST from the WS249 version of Wormbase were searched for hSEC14L2 , SEC14 , and CRAL-TRIO . Statistical analysis was performed using JMP version 12 . 0 ( SAS Institute ) or Microsoft Excel , using a two-tailed unpaired Student’s t test , two-tailed Fisher’s exact test , or nonparametric Wilcoxon rank sum test . Figure legends specify when each test was used . Sample sizes were validated a posteriori for statistical significance and variance ( for parametric tests ) . Normality was assessed using a Shapiro-Wilk’s normality test .	All tissues in the human body are encased with a thin , dense , web of proteins called the basement membrane . These membranes separate and help shape tissues , while protecting them from mechanical damage such as stretching . However , despite serving as important barriers , basement membranes must be rapidly removed at specific times and places to allow tissues to grow and change shape . This process also permits cells to enter and leave tissues , such as immune cells when they search for disease-producing agents . This creates a paradox: how does a tissue create a tough , impassable barrier that can also be rapidly removed at specific times and places ? Recent research has shown that basement membranes can slide over a tissue , and thus rapidly move the barriers to create openings for tissue growth and reshaping . However , it was not known how the cells within tissues that are normally firmly attached to basement membrane control when and where basement membrane can slide . Now , McClatchey , Wang et al . have carefully observed the tissues and basement membranes in a small worm called Caenorhabditis elegans under a microscope . This revealed that a single cell , called the anchor cell , relays a signal that instructs a group of neighboring cells to let go of the basement membrane at a specific time to allow tissue reshaping . Further experiments revealed that this signal causes cells to reduce the amount of a protein named dystroglycan at their surface . Dystroglycan is present in most tissues and helps stick the cells of tissues to basement membranes . The loss of dystroglycan was previously reported to promote the spread of cancer , although its role in cancer progression was not clear . The findings of McClatchey , Wang et al . now suggest that tumors that lose dystroglycan might allow the basement membranes surrounding them to slide , creating openings that allow the cancers to spread . Finally , McClatchey , Wang et al . also found that a protein named CTG-1 , one of a family of proteins thought to regulate the movement of proteins within cells , restricts the levels of dystroglycan at cell surface . As such , the next challenge will be to understand exactly how CTG-1 limits the amount of dystroglycan at the cell surface .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "cell", "biology" ]	2016	Boundary cells restrict dystroglycan trafficking to control basement membrane sliding during tissue remodeling
Studies of neuronal network emergence during sensory processing and motor control are greatly facilitated by technologies that allow us to simultaneously record the membrane potential dynamics of a large population of neurons in single cell resolution . To achieve whole-brain recording with the ability to detect both small synaptic potentials and action potentials , we developed a voltage-sensitive dye ( VSD ) imaging technique based on a double-sided microscope that can image two sides of a nervous system simultaneously . We applied this system to the segmental ganglia of the medicinal leech . Double-sided VSD imaging enabled simultaneous recording of membrane potential events from almost all of the identifiable neurons . Using data obtained from double-sided VSD imaging , we analyzed neuronal dynamics in both sensory processing and generation of behavior and constructed functional maps for identification of neurons contributing to these processes . One of the principal goals in neuroscience is to clarify how neuronal circuits process sensory information and control behavior . Sensory information and behavioral states are represented as dynamic activity patterns of neuronal populations in large neuronal networks . To clarify the neuronal mechanisms underlying sensory processing and behavioral generation , it is necessary to determine which neurons are involved in functionally relevant neuronal dynamics and how those neuronal components interact with each other within the larger network . Technological advances in neuroimaging have enabled brain-wide recording of neuronal activity with sufficiently fine spatial resolution to identify individual neurons within a population ( Ahrens et al . , 2013 ) . Researchers can perform pan-neuronal Ca2+ imaging in selected animals with nervous systems comprising small neuronal populations , including larval zebrafish ( Ahrens et al . , 2013 ) and C . elegans ( Schrödel et al . , 2013; Kato et al . , 2015; Nguyen et al . , 2016 ) . Although Ca2+ imaging is a convenient tool for detecting neuronal activity , it is limited to intracellular events that are associated with a change in Ca2+ concentration . Thus , Ca2+ imaging measures neither subthreshold depolarizing nor hyperpolarizing synaptic events . Accordingly , it is difficult to observe synaptic integration processes using Ca2+ indicators . In contrast , voltage sensitive dyes ( VSDs ) can detect both action potentials and sub-threshold excitatory and inhibitory synaptic potentials . Voltage sensors have enabled neuroscientists to examine ethologically relevant neuronal dynamics and to functionally map parts of the nervous systems of sea slugs ( Bruno et al . , 2015; Hill et al . , 2015; Hill et al . , 2014 ) and the medicinal leech Hirudo verbana ( Briggman et al . , 2005; Briggman and Kristan , 2006; Frady et al . , 2016 ) . The central nervous system of the leech consists of a ventral nerve cord connecting a head brain , 21 nearly identical segmental ganglia and a tail brain . The segmental ganglion of the leech is particularly well suited for comprehensive recording using VSD imaging for two reasons: It consists of only about 400 identifiable neurons ( Pipkin et al . , 2016 ) , mostly as bilateral pairs , arranged in a well-preserved geometry in a single spherical shell surrounding a central neuropil , and it functions as a basic unit of sensory processing and control of several behaviors ( Kristan et al . , 2005 ) . In the leech segmental ganglion , multiple neuronal circuits responsible for reflexive and voluntary locomotor behaviors have already been characterized by electrophysiology and VSD imaging ( Briggman et al . , 2005; Briggman and Kristan , 2006; Frady et al . , 2016; Kristan et al . , 2005 ) . However , existing technology only allowed imaging one side of a ganglion at a time , and hence captured the activity of at most half of the full ensemble of neurons: researchers could record from , at most , approximately 15% of all 6000 neurons in a pedal ganglion of Aplysia , and fewer than 50% of all 400 neurons in a leech segmental ganglion . In addition , VSDs applied in these previous studies were limited by their sensitivity or response speed: electrochromic dyes used in sea slugs did not possess enough sensitivity to detect subthreshold potentials ( Bruno et al . , 2015; Hill et al . , 2015 ) and FRET-based dye previously used in the leech had a non-negligible delay between optical signal and actual voltage change ( Briggman et al . , 2005; Briggman and Kristan , 2006 ) . To overcome these limitations , we developed a double-sided microscope for VSD imaging , consisting of precisely aligned upright and inverted fluorescent microscopes , and imaged voltage changes from the neuronal membrane stained by a highly-sensitive , fast VSD ( Woodford et al . , 2015 ) . This microscope enabled us to record from all cell bodies of a leech ganglion regardless of their location , and allowed us , for the first time , to directly analyze functional relationships between neurons located on opposite surfaces . We combined this double-sided neuronal imaging system with simultaneous electrophysiological recording and stimulation , which allowed us to monitor motor outputs , to verify agreement of VSD signals with actual membrane potentials , and to activate or inhibit selected target cells by injecting current . To demonstrate the utility of the newly developed VSD imaging method , we addressed the following two questions . ( 1 ) How are individual identifiable neurons that exhibit higher discriminability for the different sensory stimuli distributed across different surfaces of the ganglion ? ( 2 ) To what extent are neural circuit components unique or shared between different behaviors ? Double-sided VSD imaging requires simultaneously focusing two fluorescent microscopes . We achieved this by mounting the fluorescence train of an Olympus BX upright microscope with a custom focus rack on top of the body of an Olympus IX inverted microscope . Both microscopes were equipped with 20x objectives . An optically stabilized high-power LED ( Wagenaar , 2012 ) provided excitation light through the top objective , which operated in epifluorescence mode . The top objective also functioned as a condenser lens for imaging with the bottom objective , which thus operated in transfluorescence mode ( Figure 1a ) . Because of the high NA ( 1 . 0 ) of the top objective , inhomogeneities in the imaged tissue did not cause substantive deviations from uniform illumination of the bottom focal plane . The two microscopes were first coarsely aligned ( to within about 200 μm ) by moving the upright microscope’s body and its objective turret , after which micro-alignment was achieved by fine-tuning the position of the upright microscope’s objectives in their turret . We used highly sensitive CCD cameras ( Photometrics QuantEM 512SC ) to image neuronal activity with single cell resolution throughout the ganglion ( Figure 1b ) . We suppressed mechanical vibration noise by replacing the internal fans of the CCD cameras with external blowers . Photon noise was not substantially different between the top and the bottom image ( Top: 72 ± 3 ppm; Bottom: 65 ± 3 ppm ( mean ± SEM over 10 areas size-matched to typical cells ) . We imaged neural activity with a new-generation voltage sensitive dye , VF2 . 1 ( OMe ) . H ( Woodford et al . , 2015 ) , which is sensitive enough to record subthreshold events and fast enough to detect action potentials with accurate timing . The dye was loaded into somatic membranes on both aspects of a ganglion by bath application and a perfusion pump for targeted delivery ( Briggman et al . , 2005 ) . In leech ganglia , the sensitivity reached 2 . 7 ± 0 . 3 % per 100 mV ( mean ± SD across five ganglia in two leeches ) at resting potential ( −50 mV ) ( Figure 1—figure supplement 1 ) . Microscopic motion artifacts can have outsized effects on VSD signals compared to Ca2+ signals because of the limited relative change in fluorescence of VSDs and their location in the cell membrane . Accordingly , we applied a custom motion correction algorithm to all imaging data ( Figure 1—figure supplement 2 and Materials and methods ) . Bleaching artifacts in the optical signals were corrected using locally fitted cubic polynomials ( Wagenaar and Potter , 2002 ) ( Figure 1—figure supplement 3 ) and global fluctuations were subtracted away ( Lippert et al . , 2007 ) ( Materials and methods ) . The voltage sensor faithfully detected various types of membrane potential change , including action potentials , excitatory and inhibitory postsynaptic potentials , and rhythmic oscillation during fictive behaviors ( Figure 1d ) . All motor patterns recorded in this study are fictive patterns . We established a mapping between cells seen in the fluorescent images ( Figure 1b ) and identified neurons on a canonical map ( Figure 1c ) using a semi-automated procedure in a custom user interface ( Materials and methods ) . One of the major advantages of VSDs is that recorded traces can be directly compared to intracellular voltage recordings . This allowed us to identify selected cells in our recordings by comparing our data to previously published intracellular activity of those neurons in the same behaviors . In our setup , the preparation is accessible to intracellular and extracellular electrodes advanced from the upper surface . Optically recorded signals simultaneously recorded from both sides of the ganglion closely matched typical patterns of fictive behaviors that have been previously well characterized by electrophysiology and single-sided VSD imaging ( Briggman and Kristan , 2006; Kristan et al . , 2005 ) . We first focused on fictive swimming , which we induced by electrically stimulating a DP nerve root of a posterior ganglion ( Briggman and Kristan , 2006 ) ( typically , the 13th segmental ganglion ( M13 ) ) . We then imaged ganglion M10 with our double-sided microscope and simultaneously recorded intracellularly from selected cells ( Figure 2a ) . Rhythmic activity associated with swimming was readily observed , and we determined which cells were involved in this rhythm by calculating the phase and magnitude of coherence ( Briggman and Kristan , 2006 ) for each cell at the frequency with the greatest spectral power in the rhythm ( Figure 2b , c ) . The optical signal of dorsal inhibitor motor neuron DI-1 exhibits a well-understood swimming oscillation and was used as the phase reference for other cells . Using the VF2 . 1 ( OMe ) . H dye , we were able to confirm the oscillatory behavior of neurons previously studied using an earlier-generation dye ( Briggman and Kristan , 2006 ) . In addition , we were able to detect weaker oscillations in many other neurons on both sides of the ganglion . Results from coherence analysis obtained from doubly desheathed ganglia imaged using either camera in our double-sided microscope closely matched results from conventional single-sided imaging , as evidenced by the consistency of the coherence maps computed from either method ( Figure 2b and Figure 2—figure supplement 1 ) . The measured amplitudes of swim oscillations in motor neuron DI-1 , the noise levels in those recordings , and the coherence between bilateral homologues of DI-1 were also indistinguishable between single-sided and double-sided imaging experiments ( Figure 2—figure supplement 1 ) , indicating that double-sided imaging does not entail any compromises from an imaging quality perspective . We used double-sided VSD imaging to record the activity of all neurons in one ganglion ( M10 ) within a short chain during a fictive bout of a reflexive behavior known as local bending , a withdrawal response to tactile stimulation in which the leech bends its body away from the stimulated location ( Kristan et al . , 2005 ) . Fictive local bending can be induced readily even in isolated single ganglia or a short chain of ganglia by stimulating one of four pressure-sensitive sensory neurons ( P cells ) . Stimulating P cells causes a combination of excitation and inhibition in identified ‘local bend interneurons’ ( LBIs ) ( Kristan et al . , 2005; Lockery and Kristan , 1990 ) . The LBIs synapse onto several motor neurons to produce an appropriate pattern of contraction and relaxation in the local area of the body wall that depends on which location ( or which P cell ) was stimulated ( Kristan et al . , 2005; Lockery and Kristan , 1990; Baljon and Wagenaar , 2015 ) . We induced fictive local bending by stimulating the left and right ventral P cells ( PVL and PVR ) with trains of depolarizing pulses ( 20 Hz , 50% duty cycle , 1 s ) , which reliably evoked action potentials in those cells ( Figure 3a , b ) . Intracellular recording from an Anterior Pagoda ( AP ) cell , which responds differentially to stimulation of ipsilateral vs . contralateral P cells ( Jin and Zhang , 2002 ) , was simultaneously performed to confirm the different response to PVL and PVR stimulation and the correspondence of VSD and intracellular signals ( Figure 3b ) . Stimuli were presented in order of LRRLLR… , for a total of 10 stimuli per P cell . From each of the resulting VSD traces , we extracted the average fluorescence change ( ΔF/F ) during the first 0 . 5 s of the stimulus as well as during a control phase ( 1–0 . 5 s before stimulus onset ) , both relative to a reference phase ( 0 . 5–0 . 1 s before stimulus onset; Figure 3c ) . Using a leave-one-out procedure , we calculated for each of the cells how reliably their activity could be used to ‘predict’ which of the P cells had been stimulated ( Figure 3d ) . We then established a mapping between cells in the VSD images and identified neurons on the canonical maps to determine for all identified neurons to what degree their activity encoded stimulus identity ( Figure 3e ) . On average across eight experiments , 113 ± 11 ( mean ± SD ) cells on the ventral surface and 129 ± 6 on the dorsal surface could be mapped to identified neurons ( Figure 3f ) . Among those , 28% of ventral cells [35 ± 11 , mean ± SD] and 36% of dorsal cells ( 52 ± 18 ) encoded stimulus identity with prediction success higher than 75% during the first 0 . 5 s of the stimulus . This included one ventral LBI , all dorsal LBIs , and most motor neurons ( MNs; Figure 3g ) . ( All other ventral LBIs had prediction success in the range 65–75% . In contrast , the average prediction success in the control period was at chance level: 50 . 9%±1 . 3% ( mean ± SEM ) for both ventral and dorsal cells . ) The other neurons with high prediction success were AP cells and Leydig cells , as well as cells provisionally identified as cells 56 , 61 , 251 , and 152 on the ventral surface and cells 9 , 10 , 22 , 28 , 107 , and 123 on the dorsal surface . To further establish the utility of double-sided VSD imaging , we set out to determine to what extent neural circuit components are unique or shared between fictive three behaviors: local bending , swimming , and crawling . To do so , we evoked the corresponding fictive behaviors in isolated whole nerve cords using electrical stimulation ( Briggman and Kristan , 2006 ) . Specifically , local bending was activated by intracellular stimulation of a single PVL or PVR ( Kristan , 1982 ) ; swimming was elicited by stimulating a DP nerve from either M11 , M12 , or M13; and crawling was elicited by stimulating tail brain nerve roots . Motor patterns of fictive local bending and swimming were confirmed based on extracellular recordings of DP nerves or intracellular recording of AE cells ( Briggman and Kristan , 2006; Frady et al . , 2016; Gu et al . , 1991 ) . Crawling patterns were confirmed based on simultaneous intracellular recordings from two different motor neurons: the AE and CV cells ( Briggman and Kristan , 2006 ) . All three behaviors could be induced in each of six animals ( Videos 1–4 ) . We calculated the phase and magnitude of the coherence of each imaged neuron to the stimulus train ( 0 . 5 Hz ) during fictive local bending; to the optical signal of motor neuron DI-1 during swimming; and to the intracellular trace of an AE cell during crawling . Results from all behaviors in one animal are shown in Figure 4a–d and Videos 1–4 . Optical signals from representative cells located on both surfaces confirmed stereotyped activity patterns that were highly distinctive for each of the behaviors ( Figure 4e–h ) . We established identities of imaged neurons as before . On average over six preparations , we were able to assign 126 ± 11 cells on the ventral surface and 121 ± 10 on the dorsal surface . This allowed us to construct summary maps showing which neurons were consistently involved in which behaviors ( Figure 4j and Materials and methods ) . Approximately , 10% of cells were involved in all three behaviors , 33% in two out of the three behaviors , 42% in a single behavior , and 9% in none of the three behaviors ( Figure 4j ) . For the remaining 6% of cells , involvement in any of the behaviors could not be established due to lack of samples . Finally , we calculated a correlation matrix between the recorded activity of each of the cells , separately during each of the three fictive behaviors , and performed automated clustering based on these correlations ( Figure 5a ) . For each of the cells in a recording , we then calculated what fraction of the cells in the same cluster was located on the ventral or the dorsal side of the ganglion . We found that during crawling and especially during local bending , most clusters were largely confined to only one side of the ganglion , whereas during swimming they more commonly spanned sides ( Figure 5b ) , which indicates that swimming involves correlated activity among cells located on both surfaces whereas local bending largely does not . We quantified this by calculating an ‘integration coefficient’ ( Materials and methods ) which is equal to zero if all clusters are either wholly on the dorsal or wholly on the ventral side , and equal to one if all clusters are equally spread between the two sides ( Figure 5c ) . We constructed a double-sided microscope that can record fluorescence signals from two sides of a biological preparation . This technique should be broadly applicable to experimental questions that require simultaneous imaging from two widely spaced cell layers in Drosophila ( Kohsaka et al . , 2017 ) , sea slugs ( Bruno et al . , 2015; Hill et al . , 2015 ) and other organisms . The optical system can be assembled from conventional optic parts and devices . In our implementation , we used microscope parts from Olympus , but an equivalent system could be constructed using , for example , Thorlabs CERNA parts . By combining our microscope with next-generation voltage-sensitive dyes ( VF2 . 1 ( OMe ) . H ( Woodford et al . , 2015 ) , we achieved simultaneous large-scale neuronal recording from two widely spaced cell layers at single-cell resolution , capturing not only action potentials but also small excitatory and inhibitory synaptic potentials . A primary feature of the system is its ability to acquire these signals at high speed , and without delay for image capture between the two focal planes . At present , this cannot be achieved by wide-brain volumetric Ca2+ imaging as previously established for C . elegans ( Schrödel et al . , 2013; Kato et al . , 2015 ) . With our newly developed microscope , we simultaneously recorded , for the first time , the activity of the majority of neurons in a leech ganglion . While beyond the scope of this study , the fact that VSD recordings contain both spikes and postsynaptic potentials makes it possible to infer network connectivity among the different individual , identifiable cells . This offers a notable advantage over techniques that only give access to either spike events or intracellular Ca2+ concentration . The leech has 21 nearly identical segmental ganglia containing approximately 400 neurons that are arranged in a highly conserved geometry ( Kristan et al . , 2005 ) . For 148 of these neurons , functional descriptions have been published . ( A gateway to the relevant literature is available online , at http://www . danielwagenaar . net/ganglion . ) The ganglionic neurons are distributed in a single layer on the surface of the ganglion , but this layer wraps around both the dorsal and ventral sides , so that at best half of the neurons can be simultaneously imaged with conventional microscopy . Our double-sided microscope , in contrast , has access to all of them , although surface curvature means that not all neurons can simultaneously be in sharp focus ( Figure 1b ) . A single light source was sufficient for illuminating both top and bottom surfaces , because the leech nervous system is sufficiently translucent to permit even lighting onto both sides . As typical in monopolar cells in invertebrate central nervous systems , the somata of leech ganglionic neurons have passive membrane properties and action potentials are electrotonically attenuated in the cell body from their origins in the neuropil . The somata of typical interneurons and motor neurons show small , attenuated action potentials . However , some other neurons , including sensory neurons and a few neurosecretory cells , exhibit larger , easily detectable spikes . In some neurons , unitary synaptic potentials are relatively easy to detect from the cell body because those potentials are not greatly attenuated . We could capture discrete unitary EPSPs ( approximately 2–4 mV ) as optical signals visually detectable even in single sweep of recordings ( approximately 0 . 02–0 . 04% in ΔF/F ) . In most neurons , synaptic interactions in the leech central nervous system during fictive behaviors usually result in compound synaptic potentials detectable even in the cell body . Hence , a low-noise imaging system using sensitive voltage sensors enables us to record behaviorally-relevant responses even in small neurons in the leech . In addition , our double-sided microscope is compatible with both intra- and extracellular electrode placement at least from one side , enabling detailed electrophysiological interrogation of selected specific neurons along with optical imaging from the whole population . Intriguing features that we observed using our pan-neuronal imaging system are ( 1 ) widespread distribution of neurons that are differentially involved in left and right fictive ventral local bending ( Figure 3c , d ) , and ( 2 ) involvement in multiple fictive behaviors of a large fraction of identifiable neurons ( Figure 4i , j ) . With respect to ( 1 ) , we found that not only the local bend interneurons and the motor neurons previously reported ( Lockery and Kristan , 1990 ) discriminated between the stimuli , but so did many other neurons that had not previously been implicated in fictive local bending . It has long been known that the neural mechanism of local bending involves population coding ( Lewis and Kristan , 1998a; Lewis and Kristan , 1998b; Lewis and Kristan , 1998c; Lewis , 1999; Thomson and Kristan , 2006; Kretzberg et al . , 2016 ) , but its exact algorithm and computation remain unknown . Although the calculation of discriminability here was based on stimulus category ( PVL vs . PVR ) instead of actual local bend patterns in the leech’s body wall , the population dynamics of the highly discriminative cells we identified putatively underlie the neuronal computation . The discriminability maps from our study can thus be utilized for future investigations of mechanosensory information processing: the maps will work as a guide to selectively record or manipulate by intracellular electrodes during local bend in order to assess the role of individual neurons on the behavior . A few neurons can be seen from both sides of the ganglion , and some cells near the lateral edge of the ganglion may be cut off from the imaging . For instance , in Figure 3e , the right Leydig cell showed high discriminability and was seen in both dorsal and ventral aspects , but the left one was lost on the ventral side and had weak discriminability on the dorsal side . The main reason of such asymmetrical appearance on the discriminability maps could be explained as follows: We protected P cells from phototoxic effect by leaving sheath around the cell on the ventral side , thus the remaining sheath often covered other cells locating the lateral edge like Leydig cells . Because those cells are stained weakly or not at all especially on the ventral side , the asymmetrical results appeared . With respect to ( 2 ) , we observed that 43% of identifiable neurons on the ventral and dorsal surfaces were involved in at least two of the three behaviors tested ( local bending , swimming , and crawling ) . This result indicates that the neural circuits for those behaviors share many components while generating unique motor patterns for each behavior . The percentage of circuit components shared between swimming and crawling identified in this study differed from previous work ( Briggman and Kristan , 2006 ) ; in particular , the number of cells we identified as involved in crawling ( 56 ) was lower than in the previous study ( 188 ) . The reason is that double-sided desheathing , which is necessary for double-sided imaging , made long and regular episodes of fictive crawling relatively rare . Thus , crawl episodes in our experiments were somewhat shorter ( typically only 3–4 cycles ) than in the older study , resulting in a weaker coherence signal . In addition , we imaged for 50 s in our crawl trials , 10 s shorter than the previous study , so that we could capture 20 frames per second . This did result in less opportunity of inclusion of extended crawling oscillation cycles . Double-sided imaging revealed a previously unappreciated difference between the swim rhythm and local bending: The cell assemblies that are simultaneously active in the former span both sides of the ganglion , whereas in local bending , they are mostly confined to either the dorsal or the ventral side . Whether this difference has any functional significance remains unknown . In our study , the limited frame rate of the CCD camera ( QuantEM 512SC; Photometrics ) restricted imaging of brief action potentials . In the experiment for multiple behaviors , the frame rate for local bend and swimming was 50 Hz ( for 15 s in duration ) , while the frame rate for crawling was 20 Hz ( for 50 s ) , at the 512 × 128 spatial resolution adequate for making ROIs and for cell identification . In VSD imaging at 50 Hz frame rate , long lasting potentials ( e . g . spikes in Retzius or Leydig cells ) are easy to detect but brief action potentials ( millisecond order ) like an S-cell spike are hard to detect . When imaging at higher frame rate ( e . g . 200 Hz as in the report of dye development [Miller et al . , 2012] ) , the dye realizes sufficient reconstruction of such brief spikes . It will be necessary , then , to apply a higher frame rate than the current setting for connectivity analysis on detected spikes in future studies . Even in the present study , however , voltage imaging with VoltageFluor dye sufficiently reconstructed relatively large , slow action potentials in several cells ( e . g . Leydig cells , Retzius cells and mechanosensory cells ) . This enabled us to confirm cell identity based on the characteristics of spontaneous firing activity and the shape of action potential and afterhyperpolarization . The delay of the optical signal following voltage changes in VoltageFluor is negligible unlike with FRET-based VSDs because the speed of VoltageFluor is nanosecond to microsecond , while that of FRET dyes is millisecond to second ( Miller et al . , 2012 ) . With VoltageFluor dyes ( Frady et al . , 2016; Miller et al . , 2012; Moshtagh-Khorasani et al . , 2013 ) including VF2 . 1 ( OMe ) . H ( Woodford et al . , 2015 ) , we do not have to adjust phase delay which was commonly done to correct for the slow response feature of FRET-based dyes . For instance , in Briggman and Kristan ( 2006 ) , where they made single-sided coherence maps for fictive locomotory motor patterns , they used the slow dyes ( FRET-based dyes ) for imaging , using a frame rate of 10 Hz for swimming and 2 Hz for crawling ( Briggman and Kristan , 2006 ) . This likely explained why several cells like AE cells that show swim-related oscillation in our intracellular and VSD recordings ( Figure 2a ) were not detected in the their coherence analysis . By using a faster dye , we could obtain higher coherence in those neurons , especially during swimming . In this study , we identified imaged cells with known neurons using a semi-automatic mapping algorithm based on cell size and location along with an expert’s assessment based on the physiological properties of cells along with this geometrical information . To gain more insight into the neuronal networks responsible for behavior , it will be necessary to carry out more accurate neurocartography , which we will achieve by combining functional mapping using machine learning methods ( Frady et al . , 2016 ) with a connectomic approach using serial block face scanning electron microscopy ( Pipkin et al . , 2016 ) . The combination of those techniques with double-sided VSD imaging will pave the way for future investigations on how the activity of all neurons in a central nervous system is recruited to process sensory information and to generate distinctive behaviors from overlapping neuronal circuits . We acquired fluorescence images simultaneously from two focal planes using a custom double-sided microscope consisting of the fluorescence train of an upright microscope ( Olympus BX , Tokyo , Japan ) mounted on top of an inverted microscope ( Olympus IX ) . The top microscope was used to image the upper focal plane while the bottom microscope imaged the lower focal plane . We used a 20x , 1 . 0 numerical aperture ( NA ) water-immersion objective , reduced to 0 . 7 NA by way of a custom aperture , for the upright and a 20x , 0 . 7 NA objective with cover-slip adjustment collar for the inverted microscope ( both Olympus ) . The alignment of those two objectives was fine-adjusted manually so that cameras attached to the top and bottom microscopes saw the same field of view to within about 300 nm when the two focal planes were at the same depth . The two objectives served as condenser for each other , so that blue excitation light delivered through the top objective for epifluorescence imaging also served as a transfluorescence light source for the bottom objective . Further , a red LED illuminator attached to the bottom microscope provided wide-field transillumination that enabled us to use the upright objective to visualize intracellular electrodes . Both objectives were mounted on standard turrets so that they could be rotated out of the way to make place for 5x objectives used to visualize extracellular suction electrodes . For VSD imaging , we used excitation light ( bandpass filtered to 470 ± 15 nm ) from a high-power blue LED ( LedEngin LZ1-10B200 ) controlled with optical stabilization ( Wagenaar , 2012 ) . In both the upright and inverted microscopes , we used a 490 nm dichroic mirror and 505 nm LP emission filter . Images were acquired with two cooled CCD cameras ( QuantEM 512SC; Photometrics , Tucson , AZ ) at a resolution of 512 × 128 pixels . The frame rate was set depending on which behavior was recorded: for local bending and swimming , images were acquired at 50 Hz; for crawling , images were acquired at 20 Hz . Imaging data were acquired using custom software VScope ( Wagenaar , 2017 ) . Optical and electrical recordings were synchronized by connecting frame timing signals from each camera to a data acquisition board that also recorded electrophysiology signals ( see below ) . VSD imaging is highly sensitive to even sub-micrometer motions . Because VSDs are located in cell membranes rather than the cytosol , a movement of less than 1% of a cell diameter can cause a signal change of well over 1% due to bright edge pixels moving out of a pre-defined region of interest ( ROI ) . Since typical VSD signals are themselves far less than 1% , this can cause dramatic motion artifacts . To mitigate this problem , we replaced cooling fans inside each CCD camera with external blowers , since we determined that internal fans in cameras caused significant vibrations of the microscope objectives relative to the sample . After removing these fans , the remaining noise in image sequences was dominated by shot noise . Procedures of dissection and dye loading for VSD imaging are described in more detail at Bio-protocol ( Tomina and Wagenaar , 2018 ) . Medicinal leeches ( Hirudo verbana ) were obtained from Niagara Leeches ( Niagara Falls , NY ) and maintained in artificial pond water at 15°C . For each experiment , an adult leech , regardless of its behavioral status at the time , was captured from an aquarium tank to make a preparation . The sample size of experiments of mechanosensory stimulus encoding ( N = 8; Figure 3 ) and multiple behavioral generation ( N = 6; Figures 4 and 5 ) was determined by referring to the previous study , by Briggman and Kristan , where single-sided VSD imaging was performed to obtain functional maps of the leech ganglion ( Briggman and Kristan , 2006 ) . In experiments where only local bending was the target behavior , we dissected out short chains of ganglia from segments 8 through 12 . In experiments involving swimming or crawling , we isolated whole nerve cords , including the head brain , all 21 segmental ganglia , and the tail brain . In all cases , the blood sinus surrounding the nervous system was dissected away around segmental ganglion M10 . We removed the sheath from the ventral and dorsal surface of this ganglion before applying voltage-sensitive dyes . To induce swimming , a dorsal posterior ( DP ) nerve root in one of ganglia M11 through M13 was stimulated through a suction electrode . Brief electrical pulses ( 3 ms ) were delivered at 50 Hz in a 3-s-long train , with an amplitude of 7–8 V . To elicit crawling , several nerves from the tail brain were stimulated using the same stimulus parameters as for DP nerve stimulation . The preparation was put on a disk-shaped plate ( diameter: 13 mm , thickness: 0 . 65 mm ) with an open window ( 1 . 6 mm x 2 . 6 mm ) on its center , made of PDMS ( Sylgard 184 , Dow Corning , Midland , MI ) . The target ganglion for imaging was placed over this window so that the PDMS substrate did not touch the bottom side of the ganglion and did not disturb focusing of the bottom objective . For imaging , we put this PDMS plate together with a preparation on the center of a glass-bottom dish ( 38 mm in diameter ) whose periphery had PDMS substrate . Whole nerve cords or short chains of leech ganglia can move slightly because muscle cells are embedded in the nerve cord . We therefore stabilized the ganglion to be imaged by tightly pinning down blood sinus tissue to the PDMS substrate and by sandwiching adjacent connectives between small pieces of medical dressing ( Tegaderm , 3M , Maplewood , MN ) , which was also pinned down , to minimize any motion artifacts . When imaging short chains of ganglia in local bend experiments , we disconnected inter-segmental interactions by pinching and crushing axon bundles inside of anterior/posterior connectives from M10 using forceps to reduce the ganglion’s movement and variability of responsiveness of neurons in the ganglion . Throughout the dissection and during imaging , preparations were maintained in chambers filled with cold leech saline consisting of the following ( in mM ) : 115 NaCl , 4 KCl , 1 . 8 CaCl2 , 2 MgCl2 , 10 glucose , and 10 HEPES , at pH 7 . 4 . Only before crawling was induced , we temporarily replaced the cold saline ( 2–5°C ) with room temperature ( 20–23°C ) saline to obtain the most natural crawling rhythm . We bath loaded 800 nM VF2 . 1 ( OMe ) . H ( Woodford et al . , 2015 ) ( provided by Evan Miller ) in leech saline containing 1% pluronic acid ( PowerloadTM Concentrate 100x , Thermo Fisher Scientific , Waltham , MA ) . To help with dye penetration into the cell membranes , we circulated the solution using a pair of peristaltic pumps ( approximately 1 . 1 mL/min flow rate ) with outflows directed at the dorsal and ventral surfaces of the ganglion , for 20 min total . The criteria for inclusion of preparation for data were ( 1 ) no obvious damage on neurons or connectives during desheathing or VSD staining was observed , and ( 2 ) a sufficient number of fictive behaviors was induced in a preparation for analysis . We did not systematically examine when dye rundown occurs under certain level of exposure light brightness , but we could image fictive behavioral pattern with no major loss of imaging quality at least until the total light exposure reached at 250 s . For the left/right local bend experiment , single experiments usually took approximately 40 min of real time from the first trial , of which approximately 100 s was imaging time . For the multiple behaviors experiment , the time varied depending how smoothly all behaviors were successfully induced , but they typically took 30 min from the first trial , of which approximately 250 s was imaging time . According to Miller et al . ( 2012 ) , VoltageFluor dyes are less toxic than FRET-based dyes ( Miller et al . , 2012 ) . Still , the combination of repetitive high-amplitude current injection ( needed to reliably elicit multiple action potentials ) and bright light exposure sometimes destroyed dye-stained P cells by causing injury bursting . That is why we protected P cells from phototoxicity by leaving sheath around it . Compared with P cells , dye-stained N cells were found to be relatively resistant to current injection and light exposure . We recorded intracellularly from up to three neurons simultaneously using 20–50 MΩ glass microelectrodes filled with 3 M potassium acetate and 60 mM potassium chloride , using Neuroprobe amplifiers ( Model 1600; A-M systems , Sequim , WA ) . Intracellular recordings provided additional information regarding the behavioral state of the preparation as well as confirmation of the corresponding optical signals . We recorded extracellularly using suction electrodes and a four-channel differential amplifier ( Model 1700; A-M Systems ) . All electrical signals were digitized at 10 kHz using a 16-bit analog-to-digital board ( NI USB-6221; National Instruments , Austin , TX ) and VScope software ( Wagenaar , 2017 ) . We outlined the images of individual cell bodies manually as regions of interest using VScope ( Wagenaar , 2017 ) . Pixel values within each cellular outline were then averaged in each frame , yielding a raw fluorescence signal . Signals were processed to remove artifacts from micromotion ( next section ) , and to correct for slow reduction of overall fluorescence intensity due to dye bleaching . The latter was achieved by subtracting locally fitted third-order polynomials using the SALPA algorithm ( Wagenaar and Potter , 2002 ) with a time constant of 1 to 15 s . In addition , brightness averaged across the areas of the ganglion outside of ROIs was subtracted for each frame to reduce global noise due to fluorescent crosstalk among top and bottom images ( Lippert et al . , 2007 ) . Finally , signals were normalized to their average value and expressed as a percent change in fluorescence ( ΔF/F ) . As mentioned above , motion artifacts were reduced by removing fans from CCD cameras and by pinning down ganglia tightly on the PDMS substrate . However , even very small motions can cause highly detrimental artifacts in VSD recordings . To correct for small motions , we designated the middle frame of any recording as a reference frame , and generated a pair of artificial frames by shifting the reference frame one pixel to the left or to the right . Let IRandIL be vectors consisting of the intensity values of the pixels in the right- and left-shifted reference frames , and let I′ be the intensity vector of an arbitrary frame in the recording . As long as the motion is small ( less than or approximately equal to one pixel ) , Δx=2 ( I′−IL ) ⋅ ( IR−IL ) /‖IR−IL‖2−1 , where · is the vector product and ‖I‖ is the vector norm , is a good estimate for the motion in the x-direction between the frame under study and the reference frame . ( The reason is that an image shifted by Δx pixels can be approximated asI′=[ ( 1−Δx ) IL+ ( 1+Δx ) IR]/2as long as \|Δx\|⪅1 . The first equation is derived from the second by minimizing with respect to Δx . ) The same method can of course be used for motion in the y-direction . More interestingly , the method can be used for other affine distortions as well . For instance , if we calculate artificial frames by rotating the reference frame by ±0 . 1° , the above procedure would yield estimates of image rotation ( in units of 0 . 1° ) . Using this method , we estimated and corrected for small motions that may occur within the preparation or even due to vibrations in the microscope , thus preventing motion artifacts in the extracted VSD traces ( Figure 1—figure supplement 2 ) . In our experiments on the encoding of stimulus identity by individual neurons , we performed 10 trials stimulating the left PV cell and 10 stimulating the right PV cell , in order ( LR ) ( RL ) ( LR ) ( RL ) …Eight preparations with which we successfully performed 10 left Pv and 10 right Pv trials were collected in approximately 6 weeks . To calculate how well each cell ‘predicted’ the stimulus identity ( i . e . , ‘left’ or ‘right’ ) , we calculated the average ΔF/F during the first 0 . 5 s of each stimulus relative to the preceding reference phase , separately for each trial . Taking each trial in turn , we then took that trial and its ‘partner’ trial out , and calculated the average ΔF/F for the 9 ‘left’ stimuli out of the remaining 18 trials and also for the ‘right’ stimuli . The ‘partner’ trial was the next trial for odd-numbered trials , and the preceding trial for even-numbered trials . If the ΔF/F in the trial under consideration was closer to the average ΔF/F of the ‘left’ trials in the training set than to the average of the ‘right’ trials , the neuron was considered correct in its ‘prediction’ of stimulus identity if the trial under consideration was in fact a ‘left’ trial , and conversely for ‘right’ trials . The percentage of trials in which a cell correctly predicted stimulus identity in this sense was used as a measure of prediction success . Any cell that correctly predicted stimulus identity in at least 75% of trials ( 50% being change performance ) was considered to encode stimulus identity . We used multitaper spectral analysis ( Taylor et al . , 2003 ) to estimate the coherence between optical signals from individual cells with a common reference . That reference was the stimulus train for local bending , the optical signal of a DI-1 motor neuron for swimming , or the intracellular electrode signal of an AE motor neuron for crawling . For each recording , we calculated the 95% confidence interval for the magnitude of estimated coherence under the null hypothesis that a signal was not coherent with the reference ( Cacciatore et al . , 1999 ) . A cell was considered to be involved in the behavior expressed during a given trial if its measured coherence exceeded this confidence interval . The overall layout of neurons within leech ganglia is highly conserved between ganglia within an animal as well as between animals , but the precise geometry does vary . In order to identify cells seen in the VSD image sequence ( Figure 3—figure supplement 1a ) with neurons in the canonical map , we developed a graphical user interface in GNU Octave ( version 4 . 00 ) that allows us to proceed as follows . First , we mark all the visible cells as regions of interest on the image ( Figure 3—figure supplement 1b ) . Then , we overlay the canonical map over this ( Figure 3—figure supplement 1c ) . To the trained eye , the identification of many of the larger cells is immediately obvious , so we register these identities ( using a drag-and-drop mechanism in the GUI; ( Figure 3—figure supplement 1d ) ) . Cell identification in this step can be confirmed based on physiological properties observed in its optical signals . This partial mapping of ROIs to identified neurons allows the program to do a coarse alignment between the canonical map and the actual image using affine transformations local to each of the four packets of cells ( Figure 3—figure supplement 1e ) . ( The ganglion is divided by giant glial cells into six packets ( Kristan et al . , 2005 ) , the boundaries of which are indicated on the canonical map . ) This preliminary alignment enables us to identify several other neurons with high confidence , after which the computer can perform a local alignment step . Finally , the computer assigns putative identities to the remaining ROIs , leading to a nearly complete mapping between ROIs ( orange dots in Figure 3—figure supplement 1f ) and identified neurons ( cross marks ) . This mapping technique can be applied to other preparations if a two-dimensional canonical map of any target aspect of the nervous system is available . The following steps ( 1-3 ) of this procedure deserve further explanation . For each neuron in each animal , we determined whether its coherence exceeded the 95% confidence interval of the null hypothesis that a given neuron was not involved in a given behavior . Six preparations with which we successfully induced the full set of behaviors ( local bending , swimming and crawling ) were collected in approximately 2 months . If a neuron exceeded that threshold for a given behavior in four out of six animals , it was considered to be involved in that behavior ( Briggman and Kristan , 2006 ) ( Figure 4i ) . Since swimming and crawling are both symmetric behaviors , we included both members of a homologous pair if ( and only if ) at least one member exceeded the 97 . 5% C . I . We clustered cells based on the matrix of the correlation coefficients of their activity patterns , separately for each behavior ( by constructing a dendrogram based on the correlation distance followed by tree cutting ) . We then assigned a dorsoventrality index ( DVI ) to each cell , which was equal to the fraction of dorsally located cells in that cell’s cluster . This is what is shown in the histograms of Figure 5b . Cells in clusters with fewer than three members were ignored for this calculation; the results did not change qualitatively if this threshold was changed to two or five . Based on the DVI , we calculated the integration coefficient ( CI ) of Figure 5c as:CI=⟨1−2\|DVI−1/2\|⟩ , where \|⋅\| denotes absolute value and ⟨ · ⟩ denotes the average across all cells ( except those not in clusters of size three or more ) . All data processing and statistical analysis were performed in GNU Octave , version 4 . 0 . 0 . Comparison of the number of stimulus-discriminating neurons between the stimulus phase and the control phase for Figure 3g was conducted using a paired sample t-test . Integration coefficients were compared among three fictive behaviors for Figure 5c using ANOVA followed by Tukey’s test . Comparison of swimming parameters between single- and double-sided imaging for Figure 2—figure supplement 1c–f was carried out using ANOVA . Except where otherwise noted , the significance level was set at 0 . 05 .	In every animal , networks of nerve cells work together to interpret signals from the environment and to coordinate responses . Being able to record the activity of all the neurons in a brain at once would greatly advance our understanding of how the brain works . Yet it is not possible to do this for a human brain , which contains several billion neurons . The medicinal leech , on the other hand , has a much simpler nervous system . It has 21 brain-like units called segmental ganglia , which control how the parts of its body move , and each one contains about 400 neurons arranged on a single layer . The activity of large populations of neurons can be monitored using a technique called fluorescent imaging . Most fluorescent dyes , however , are not sensitive enough to report low levels of activity or fast enough to track individual nerve impulses . Also , current microscopy techniques only allow one surface to be imaged at any one time . These limitations constrain the kinds of questions that neuroscientists can ask about how networks of nerve cells function . Tomina and Wagenaar have now developed a double-sided fluorescent microscope system that allows a ganglion in a medicinal leech to be viewed from both sides at once . Using a new generation of dyes , which rapidly change their brightness as individual neurons become active or are inhibited , subtle changes in the activity of hundreds of individual neurons were monitored at the same time . In a test of the system , Tomina and Wagenaar recorded activity for different leech behaviors , like bending , swimming and crawling . For the first time , the relationships between neurons on both sides of the ganglion could be seen . This new technique for examining the activity in neuronal circuitry will allow complex networks of neurons to be studied in more detail . The data that these images generate could then be analyzed mathematically to better understand how the brain processes information from its senses and generates behavior .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "tools", "and", "resources", "neuroscience" ]	2017	A double-sided microscope to realize whole-ganglion imaging of membrane potential in the medicinal leech
Unresolved ER stress followed by cell death is recognized as the main cause of a multitude of pathologies including neonatal diabetes . A systematic analysis of the mechanisms of β-cell loss and dysfunction in Akita mice , in which a mutation in the proinsulin gene causes a severe form of permanent neonatal diabetes , showed no increase in β-cell apoptosis throughout life . Surprisingly , we found that the main mechanism leading to β-cell dysfunction is marked impairment of β-cell growth during the early postnatal life due to transient inhibition of mTORC1 , which governs postnatal β-cell growth and differentiation . Importantly , restoration of mTORC1 activity in neonate β-cells was sufficient to rescue postnatal β-cell growth , and to improve diabetes . We propose a scenario for the development of permanent neonatal diabetes , possibly also common forms of diabetes , where early-life events inducing ER stress affect β-cell mass expansion due to mTOR inhibition . β-Cell failure is the fundamental pathophysiological factor of both type 1 ( T1D ) and type 2 diabetes ( T2D ) ( Cerasi and Luft , 1967; Accili et al . , 2010; Rhodes , 2005; Mathis et al . , 2001 ) . There also exist less frequent , monogenic forms of diabetes resulting from loss-of-function mutations in β-cell function genes . An example is proinsulin mutations which lead to proinsulin misfolding , inducing β-cell ER stress and consequently permanent neonatal diabetes , also called mutant-insulin diabetes of the young ( MIDY ) ; its animal counterpart is the Akita mouse ( Liu et al . , 2010; Weiss , 2013 ) . β-Cells have a highly developed endoplasmic reticulum ( ER ) to cope with the demand to secrete high amounts of insulin . In diabetes , the proinsulin burden on the ER is increased and proinsulin folding is impaired due to altered β-cell redox state , hence leading to accumulation of misfolded proinsulin and consequently to ER stress . Therefore , proinsulin misfolding/ER stress also plays an important role in the pathophysiology of T1D and T2D ( Eizirik et al . , 2008; Scheuner and Kaufman , 2008 ) . Clarifying how ER stress leads to β-cell failure in Akita diabetes can have important implications for the common forms of diabetes . β-Cell mass is reduced in diabetes ( Rahier et al . , 2008; Butler et al . , 2003 ) , albeit with very large variation between subjects , even in T1D ( Campbell-Thompson et al . , 2016 ) . Several mechanisms are implicated , including impaired programming of the endocrine pancreas in utero ( Sandovici et al . , 2013; Alejandro et al . , 2014 ) , increased β-cell apoptosis ( Butler et al . , 2003; Jurgens et al . , 2011; Donath et al . , 1999 ) , reduced β-cell proliferation ( Butler et al . , 2007 ) , and dedifferentiation of mature β-cells ( Talchai et al . , 2012 ) . The quantitative contribution of the different mechanisms to β-cell loss in diabetes is controversial . More important , it is uncertain whether β-cell loss precedes the onset of diabetes or develops during later stages of the disease secondary to hyperglycemia , and thus can rather be viewed as a complication of diabetes . β-Cell mass expands rapidly in the newborn and then adjusts to changes in metabolic demand , probably also in humans ( Bonner-Weir et al . , 2016; Cigliola et al . , 2016 ) . In mice , islet and β-cell numbers are increased more than 3-fold between 10 days of age and adulthood; this is associated with high β-cell replication , which is drastically decreased during adulthood ( Herbach et al . , 2011; Teta et al . , 2005; Saisho et al . , 2013 ) . β-Cell mass expansion is mainly mediated via proliferation of mature β-cells ( Dor et al . , 2004 ) . It has been recently suggested that insulin demand drives β-cell proliferation via the unfolded protein response ( UPR ) , which senses insulin production . UPR activation during ER stress correlated with and triggered β-cell proliferation in response to glucose , probably through ATF6 ( 24 ) . Others showed that reducing the proinsulin load by deleting the insulin gene decreased UPR along with increased β-cell proliferation ( Szabat et al . , 2016 ) , suggesting that ER stress is implicated in the regulation of β-cell proliferation . Herein , we exploited the Akita mouse model of diabetes to study how ER stress affects β-cell mass expansion and differentiation during early life . We found that exposure to ER stress during the neonatal period dramatically reduces β-cell growth and functional maturation . This was associated with transient inhibition of the key signaling complex mTORC1 which governs postnatal β-cell growth and differentiation . Impairment of early β-cell growth and maturation leads to permanent β-cell dysfunction with subsequent development of diabetes; restoration of mTORC1 activity in Akita neonates was sufficient to prevent β-cell loss and ameliorate diabetes . Metabolic state and islet morphometry were analyzed in 2- to 3-month-old Akita mice . Adult Akita mice develop severe insulin-deficient diabetes with fed blood glucose ~ 400 mg/dl along with a 90% decrease of pancreatic insulin content ( Figure 1—figure supplement 1a–c ) . In adult Akita mice , β-cell mass was decreased by 70% compared to age-matched controls ( Figure 1a ) . We studied whether decreased β-cell mass is mediated via impaired β-cell proliferation , increased apoptosis or dedifferentiation . The rate of β-cell proliferation measured by Ki67 staining was < 1% and similar in control and Akita mice ( Figure 1b ) . In agreement with a previous study ( Izumi et al . , 2003 ) , there was a slight increase in the number of TUNEL+ β-cells in Akita mice ( Figure 1b ) . Most islets contained no or only a single TUNEL+ β-cell . We counted 2592 β-cells in wild type and 1754 cells in Akita mice and found that the frequency of TUNEL+ cells was 0 . 1% in Akita mice , whereas no TUNEL+ cells were observed in control mice; the difference between groups was not statistically significant ( p=0 . 1 ) . Thus , the frequency of apoptotic events based on TUNEL was fairly low in Akita β-cells . Apoptotic cells are rapidly cleared by macrophages; therefore , the true rate of apoptosis is very difficult to assess in all models of diabetes . We cannot exclude that cumulative low-grade apoptosis throughout life contributes to β-cell loss in adult animals; however , this finding was somewhat surprising , considering the fact that irreparable proinsulin misfolding generates severe ER stress associated with β-cell loss and insulin deficiency mimicking T1D . Next , we studied by lineage tracing whether β-cell loss results from β-cell degranulation or trans-differentiation . We generated RIP-Cre:Rosa26-Yfp reporter mice on the background of wild-type and Akita mice , in order to monitor the fate of β-cells in adult animals . We stained pancreatic sections of β-cell reporter mice for insulin , glucagon and somatostatin and quantified the percentage of genetically labeled β-cells ( YFP+ ) expressing insulin or non-β-cell hormones ( Figure 1c ) . In Akita mice , the number of genetically labeled β-cells that stained negative for insulin ( INS ) increased by 2 . 6-fold compared with wild-type mice ( Figure 1d ) . Part of the YFP+/INS- cells expressed glucagon or somatostatin ( 0 . 3% of YFP+/INS- cells; 9 out of 3233 cells ) in Akita compared to 0 . 04% in controls; 3 out of 8091 cells ) . The percentage of β-cells expressing transcription factors required for β-cell maturation and function , including PDX-1 and NKX6 . 1 , was decreased in Akita mice ( Figure 1—figure supplement 2 ) . These findings suggest that some degree of β-cell dedifferentiation/reprogramming does take place in diabetic Akita mice; nevertheless , 98 . 7% of genetically labeled Akita β-cells remained insulin positive ( Figure 1d ) ; therefore , these alterations could not explain the 70% decrease in β-cell mass . Collectively , decreased β-cell mass in diabetic Akita mice is not due to alterations in β-cell proliferation , survival or differentiation in adulthood . We therefore assessed β-cell dynamics during the early postnatal period . Developmental insults during gestation , such as malnutrition , low-protein diet and increased exposure to glucocorticoids , are known to restrict the number of β-cells formed in the fetal pancreas , which is maintained in adulthood ( Alejandro et al . , 2014; Dumortier et al . , 2011; Garofano et al . , 1998 ) . We envisioned that proinsulin misfolding in the embryo after the initiation of insulin biosynthesis at day E11 might lead to ER stress with subsequent impairment of β-cell growth in utero . We analyzed β-cell mass , proliferation and apoptosis in Akita and control newborns at P1-2 . At this stage , Akita mice have normal body and pancreatic weight ( Figure 2a–b ) and are strictly normoglycemic ( Figure 2c ) . β-Cell mass in Akita newborns was similar to that in control mice ( Figure 2d ) . Furthermore , β-cell proliferation was approximately 8-fold higher than in adult animals and was similar in Akita and control mice ( Figure 2e ) ; TUNEL+ β-cells were found neither in control nor in Akita mice ( n = 2592 control and 1754 Akita β-cells were counted ) . In Akita mice , the percentage of NKX6 . 1 expressing β-cells was similar to that in control mice , whereas there was a small decrease in the percentage of PDX-1 expressing β-cells ( Figure 2f–g ) . Altogether , these findings indicate that in uteroβ-cell development in Akita mice is only minimally impaired , and that β-cell loss must occur after birth . We then hypothesized that ER stress might impair β-cell growth during the postnatal period prior to development of diabetes . To test this hypothesis , we assessed the metabolic state and β-cell mass in Akita compared to control mice at P19-21 , prior to weaning . At this stage , body weight and fed and fasting blood glucose in Akita mice were still normal ( Figures 2a and 3a–b ) ; however , the mice exhibited marked β-cell dysfunction , evident by glucose intolerance associated with blunt insulin response to glucose and decreased pancreatic insulin content ( Figure 3b–d ) . Islets isolated from pre-diabetic Akita mice also showed marked attenuation of glucose-stimulated insulin secretion along with reduced insulin content ( Figure 3e ) . β-Cell mass was decreased by 60% compared to controls , which is similar to the relative decrease in β-cell mass in adult mice ( Figure 3f ) . This was accompanied by a parallel decrease in β-cell proliferation based on Ki67 , PCNA and phospho-Histone-H3 immunostaining ( Figure 4a and Figure 4—figure supplement 1a ) . In control mice , β-cell proliferation remained high in the first 3 weeks of life; this was accompanied by two-fold increase in β-cell mass ( Figure 4d–e ) . The decline in β-cell proliferation in Akita mice completely prevented the expected early increase of β-cell mass ( Figure 4e ) . The proliferation rate in the exocrine tissue was similar in wild type mice and in Akita mice ( Figure 4—figure supplement 1b ) , indicating that the effect of the Akita mutation on proliferation is cell autonomous . Consistently , the weight of the pancreas , which mainly contains exocrine tissue , was similar in Akita and control mice ( Figure 2b ) . To assess β-cell size , we used insulin staining to mark β-cells and E-cadherin to highlight cell boundaries ( Figure 4b–c ) . In control mice , β-cell size remained unchanged during the first 3 weeks of life and increased 3-fold in adult animals . In Akita mice , β-cell size decreased during the early postnatal period , but increased after weaning while developing diabetes . Akita β-cells were smaller than control at P1-2 , P19-21 and in 2- to 3-month old animals: a 33% reduction in β-cell size was observed in adult Akita mice ( Figure 4b–c ) . Notably , β-cell mass increased after weaning both in wild type and in Akita mice ( Figure 4e ) ; however , the overall increase in β-cell mass in Akita mice was attenuated compared to controls due to the lower rate of β-cell proliferation and smaller increase in β-cell size . We did not detect any TUNEL+ β-cells at P19-21 , neither in control nor in Akita mice ( n = 2676 control and 1447 Akita β-cells were counted ) , indicating that apoptotic events were quite rare even in Akita mice . In summary , Akita β-cell mass is decreased due to impaired postnatal β-cell growth early in life , prior to the onset of full-blown diabetes . To understand the mechanisms underlying Akita β-cell growth arrest and dysfunction prior to development of diabetes , we isolated islets from pre-weaning mice at P19-21 , and analyzed gene expression by RNA-seq . It has been previously reported that in heterozygous Akita mice β-cell loss is accompanied by decreased α-cell number and that islet composition remained unchanged ( Kayo and Koizumi , 1998 ) . Consistently , we found that β-cell number per islet area and β/α cell ratio were similar in neonate Akita and wild-type mice ( Figure 5—figure supplement 1 ) , indicating that transcriptomic analysis mainly reflects the changes in the genetic signature of the β-cells ( ~70% of all islet cells ) and is not influenced by alterations in islet composition . The list of differentially expressed genes between Akita and control islets is shown in Table 1 . We performed geneset enrichment and pathway analyses using Genomica and Ingenuity software . ER-stress-related genes were upregulated in Akita islets , along with modest enrichment of genes involved in apoptosis ( Figure 5a ) . Intriguingly , total steady state mRNA levels of Xbp1 , the main ER stress-sensing transcription factor , were decreased with only a modest increase in Xbp1 splicing ( Figure 5b ) . The most prominent upregulated UPR gene was Homocysteine-responsive endoplasmic reticulum-resident ubiquitin-like domain member one protein ( Herpud1 ) ( log 2FC 1 . 8; p=4 . 6×10−24 ) . HERPUD1 functions as a hub for membrane association of ER associated degradation ( ERAD ) machinery components and for the interactions between misfolded proteins and ERAD . The expression of chaperones , including Dnajc3 ( Hsp40 ) , Manf and Hspa5 ( Bip ) was upregulated in Akita islets ( Figure 5b ) ; the protein level of the latter was also markedly increased ( Figure 6a ) . There was a mixed response of genes that regulate apoptosis in ER stress: Atf6 , Atf3 , Ddit3 ( Chop ) , Txnip and Bbc3 ( Puma ) were upregulated , whereas Atf4 and pro-apoptotic Bax were downregulated ( the changes in Atf3 , Atf4 , Atf6 and Bax were not statistically significant ) ( Figure 5b ) . P85α is a regulatory unit of PI3 kinase; it has been shown that P85α deficiency protects β-cells from ER-stress-induced apoptosis ( Winnay et al . , 2014 ) . In Akita islets , the expression of Pik3r1 gene encoding for P85α was decreased ( Figure 5b ) , probably promoting β-cell survival . It has been previously shown that in Akita mice ER stress-induced apoptosis is mediated via CHOP; however , CHOP expression was increased only after development of diabetes , but not during the neonatal period ( Oyadomari et al . , 2002 ) , further suggesting that young β-cells adapt to chronic ER stress without robust stimulation of the terminal , pro-apoptotic UPR . Strikingly , we found that genes regulating β-cell differentiation and function were downregulated in pre-weaning , neonate Akita islets ( Figure 5c–d and Table 1 ) . This included the transcription factors Nkx6 . 1 , Nkx2 . 2 and Mafa , proinsulin ( Ins1 and Ins2 ) , pancreatic convertase 1/3 ( Pcsk1 ) and Glut2 ( Slc2a2 ) , as well as genes involved in calcium signaling , insulin granule formation and secretion ( Table 1 ) . Consistently , target genes of NKX6 . 1 and PDX-1 transcription factors , master regulators of β-cell differentiation and function , were also downregulated , suggesting impairment of β-cell differentiation ( Figure 5c–d ) . RNA-seq showed that in Akita neonates , the mRNA level of Pdx1 was not significantly downregulated ( Figure 5b ) , whereas PDX-1 protein level was markedly reduced ( Figure 6a ) . Immunostaining showed that the number of β-cells expressing NKX6 . 1 and PDX-1 was decreased by ~50% ( Figure 6b ) , indicating that the lower expression of β-cell transcription factors is not due to decreased β-cell number per se . Finally , we treated adult and neonate islets and the β-cell line INS-1E with low-dose thapsigargin; this decreased PDX-1 protein level ( Figure 6c–d ) . Chemical chaperones ( TUDCA and 4-PBA ) had variable effects on BiP expression; however , both compounds failed to prevent the effect of ER stress on PDX-1 expression ( Figure 6c ) . Mitochondrial activity has been implicated in cell proliferation including that of β-cells ( Walter et al . , 2015; Klochendler et al . , 2016 ) and is instrumental for β-cell functional maturation and for the development of mitogenic and secretory responses to glucose ( Stolovich-Rain et al . , 2015 ) . Several genes encoding subunits of the electron transport chain including Ndufs2 , Sdhc , Cox6a2 , Cox6c , as well as the key anaplerotic enzyme pyruvate carboxylase ( Pcx ) were downregulated in neonate Akita islets . On the contrary , the expression of pyruvate dehydrogenase kinases ( Pdk1 , 2 and 4 ) , which inhibit oxidative phosphorylation by phosphorylating pyruvate dehydrogenase , was upregulated ( Table 1 ) . In summary , ER stress leads to decreased expression of key β-cell transcription factors and mitochondrial genes along with impaired postnatal β-cell differentiation and functional maturation . Consistent with abrogated β-cell growth , the expression of proliferation and cell cycle genes was reduced in pre-weaning Akita neonates ( Table 1 ) . It has been reported that insulin-like growth factor 1 and 2 ( IGF1 and IGF2 ) and epidermal growth factor ( EGF ) receptors are necessary for normal β-cell growth and differentiation ( Kulkarni et al . , 2002; Miettinen et al . , 2008 ) . The expression of EGF , IGF1 and IGF2 receptors was indeed decreased in Akita islets , whereas the expression of the insulin receptor remained unchanged ( Table 1 ) . Growth factors mediate their effects via IRS proteins with subsequent activation of PI3 kinase and its downstream target AKT . The expression of Pik3r1 encoding for the regulatory unit of PI3 kinase ( P85α ) was decreased ( Figure 5b ) , along with marked inhibition of AKT activity ( Figure 7a ) . mTORC1 is a protein complex that integrates signals from nutrients , growth factors , hormones and stress to regulate cell growth and proliferation , which is indispensable for embryonic and postnatal β-cell growth and maturation ( Ni et al . , 2017 ) . Western blotting showed that also mTORC1 activity was markedly inhibited in neonatal Akita islets , evident by decreased protein levels and Ser240/244 phosphorylation of ribosomal S6 ( Figure 8a ) . Eukaryotic translation initiation factor 4E binding protein ( 4E-BP1 ) dephosphorylation was reflected in the shift from the highly phosphorylated γ-band to the nonphosphorylated β-band as previously described ( Ni et al . , 2017 ) ( Figure 8a ) . Immunostaining showed that the number of phospho-S6+ β-cells was high in newborn β-cells and decreased over time ( Figure 8b ) . On the contrary , mTORC1 activity ( S6 phosphorylation ) in the exocrine pancreas was low during the neonatal period and was markedly enhanced in adult mice ( Figure 8b ) . The number of phospho-S6+ β-cells was lower in Akita mice than in controls already at P1-2 and at P19-21 , further indicating that mTORC1 activity was decreased in neonate Akita β-cells ( Figure 8b ) . In adult islets the number of S6+ β-cells was small and mTORC1 activity was increased in Akita , despite sustained inhibition of AKT signaling ( Figure 8a and 7a ) . We and others have previously shown that in diabetes hyperglycemia stimulates mTORC1 activity ( Fraenkel et al . , 2008; Yuan et al . , 2017 ) . Treatment of adult Akita mice with the glucosuric drug dapagliflozin for 72 hr decreased blood glucose and abrogated S6 phosphorylation ( Figure 8c ) , indicating that mTORC1 activation in diabetic Akita β-cells is mediated via hyperglycemia . Consistent with the findings in neonate Akita islets , treatment of INS-1E cells with low-dose thapsigargin for 48 hr did not affect IRS2 protein level and inhibited AKT and S6 phosphorylation ( Figure 7b–c ) , suggesting that ER stress inhibits AKT-mTORC1 signaling . We next studied whether treatment with chemical chaperones can prevent the downregulation of mTORC1 and increase β-cell proliferation in Akita neonates . Intriguingly , both TUDCA and PBA further decreased mTORC1 activity in Akita β-cells ( Figure 8—figure supplement 1a ) . In vivo , treatment of Akita neonates with TUDCA for 48 hr decreased β-cell proliferation ( Figure 8—figure supplement 1b ) . In summary , in Akita islets mTORC1 is inhibited during the neonatal period in parallel to the β-cell growth arrest . Treatment with chemical chaperones failed to correct the early β-cell growth arrest . The TSC1/TSC2 complex is a key negative upstream regulator of mTORC1 . Constitutive activation of mTORC1 by Tsc2 knockout in β-cells modulates β-cell mass in a biphasic manner ( Bartolomé et al . , 2014; Shigeyama et al . , 2008 ) . In young mice , constitutive mTORC1 activation increases β-cell number and size , whereas in old mice the animals develop diabetes due to increased β-cell apoptosis . Because ER stress inhibited mTORC1 and β-cell growth in neonates , we studied whether stimulation of mTORC1 could rescue diabetes in Akita mice . We generated heterozygous and homozygous βTsc1 knockout Akita mice ( RIP-Cre:Tsc1flox/+:Akita and RIP-Cre:Tsc1flox/flox:Akita mice ) by crossing Akita mice with Rosa-26-floxed Tsc1 mice and with RIP-Cre mice . βTsc1+/+ , βTsc1+/- and βTsc1-/- mice were used as controls . It has been previously reported that the RIP-Cre alone without recombination at lox sites is associated with glucose intolerance and even frank diabetes ( Lee et al . , 2006 ) . We found that in wild-type mice , expression of RIP-Cre induced only modest glucose intolerance even in adult mice ( Figure 9—figure supplement 1a ) . Moreover , it did not affect fed blood glucose either in wild-type or in Akita mice ( Figure 9—figure supplement 1b ) , and the insulin sensitivity of Akita mice was unaltered ( Figure 9—figure supplement 1c ) . We therefore believe that this is a valid model to test the effects of mTORC1 activation on diabetes and β-cell function in Akita mice . We first studied the effect of Tsc1 knockout on mTORC1 activity in neonates at P19-21 . In Akita neonates TSC1 deficiency increased mTORC1 activity compared to Akita controls , evident by S6 phosphorylation ( Figure 9a ) . Activation of mTORC1 did not affect the expression of BiP ( Figure 9b ) , suggesting that this did not have a major effect on β-cell ER stress . TSC1 deficiency in Akita mice increased β-cell size ( Figure 9c ) and proliferation ( Figure 9d ) . At P30-35 , mTORC1 activation did not affect β-cell proliferation either in heterozygous or homozygous Tsc1 knockout mice ( Figure 9d ) , indicating that stimulation of mTORC1 induced β-cell proliferation only in young mice prior to weaning . We then tested the effects of β-cell TSC1 deficiency on the metabolic state of Akita mice after weaning . IPGTT performed at the age of 3–4 weeks showed that glucose tolerance was improved or normalized in βTsc1+/- and βTsc1-/-Akita mice ( Figure 10a–b ) . TSC1 deficiency doubled pancreatic insulin content in control mice and increased it fivefold in Akita mice ( Figure 10c–d ) . Islet insulin content of Akita mice crossed with the Tsc1 null was twofold higher compared to Akita islets ( Figure 10e ) . Glucose-stimulated insulin secretion remained markedly reduced in vivo and ex vivo ( Figure 10f–g ) , indicating that stimulation of mTORC1 improved the metabolic state by increasing β-cell mass and islet insulin content without affecting the fundamental defects in the insulin response to stimulus . Intriguingly , activation of mTORC1 in pre-diabetic Akita islets did not affect PDX-1 and NKX6 . 1 expression ( Figure 10—figure supplement 1a–d ) . Collectively , these findings indicate that activation of mTORC1 improved glycemia by increasing β-cell mass and islet insulin content despite persistent ER-stress-induced β-cell dysfunction . We followed a small number of Akita mice with restored mTORC1 for 3 months . Part of the mice became mildly hyperglycemic , whereas others developed overt diabetes with severe hyperglycemia ( Figure 10—figure supplement 2 ) . Thus , life-long ER stress might eventually lead to diabetes despite the initial increase in β-cell mass with heterogeneity in the timing of appearance and severity of hyperglycemia . We found that the ER stress of neonate β-cells interrupted their proliferation and cell size growth , resulting in low β-cell mass accompanied by severe insulin deficiency . The decline in β-cell proliferation along with attenuation of β-cell hypertrophy caused a ~ 70% decrease in β-cell mass together with marked depletion of islet insulin content and blunt insulin response to glucose . These deficiencies culminated in diabetes when nutrient load was increased after weaning . β-Cell growth arrest in Akita neonates was associated with transient inhibition of mTORC1 . Interestingly , in adult Akita mice mTORC1 activity was increased , most probably due to hyperglycemia , and β-cell growth was resumed , albeit without ‘catch-up’ growth , hence β-cell mass remained reduced . Importantly , partial restoration of mTORC1 activity in neonate Akita β-cells was sufficient to rescue β-cell expansion with marked improvement of glucose tolerance despite ongoing ER stress and β-cell dysfunction , indicating that mTORC1 inhibition plays a key role in the pathophysiology of neonatal diabetes ( Figure 10h ) . Our findings highlight the importance of postnatal β-cell growth and differentiation for normal glucose homeostasis in adult life . During the neonatal period , β-cells expand rapidly by proliferation , followed by hypertrophy after the transition from suckling to regular chow . These dynamics of β-cell number and size culminate in ~ 8-fold increase of β-cell mass in young adult animals , which seems to be a sine qua non condition for coping with the increased insulin demand of adult life . In our model ( Akita mice ) ER stress is induced by the expression of a mutant , unfoldable insulin which creates protein aggregates in the ER . Insulin is expressed at day E11 . 5 and therefore some degree of ER stress is expected to occur already in fetal Akita β-cells . Nevertheless , neonate Akita mice had normal β-cell mass . Also affected human subjects with Akita diabetes are born with normal body weight and are normoglycemic at birth ( see accompanying paper by Balboa et al ) , indicating that β-cell dysfunction develops after birth . Several lines of evidence indicate that fetal and neonate β-cells respond to stress by slowing replication . A striking example is intrauterine growth retardation ( IUGR ) , where placental insufficiency generates hypoxia and nutrient deprivation , resulting in decreased β-cell proliferation and mass in utero ( Thompson et al . , 2010 ) . Infants with IUGR exhibit impaired insulin secretion and show a high incidence of T2D in adulthood ( Barker , 2006 ) . Similarly , malnutrition and low-protein diet during pregnancy restrict the number of β-cells in the fetal pancreas ( Alejandro et al . , 2014; Dumortier et al . , 2011; Garofano et al . , 1998 ) . The Wolcott-Rallison syndrome results from mutations in PERK ( EIF2AK3 ) , leading to permanent neonatal diabetes due to β-cell ER stress . Similar to Akita mice , PERK-deficient mice exhibited severe defects in neonatal β-cell proliferation , resulting in low β-cell mass and β-cell dysfunction ( Zhang et al . , 2006 ) . Most importantly , in the accompanying paper Balboa et al show that the proliferation rate of β-like cells derived from induced-pluripotent stem ( iPS ) cells from human subjects carrying missense INS mutations , which disrupt the proinsulin inter-chain disulphide bonds formation similar to the Akita mutation , was reduced compared to control cells in which the mutation was corrected by CRISPR . These findings strongly suggest that the proliferation inhibitory response to ER stress is a general phenomenon , and indeed relevant to disease pathophysiology in man . We performed an unbiased transcriptomic analysis in Akita neonates and studied insulin/IGF-1 signaling to clarify how ER stress induces β-cell growth arrest . Strikingly , we found that the genetic program that governs β-cell growth , including growth factor receptors ( IGF-1R , IGF-2R and EGFR ) and other replication genes , was downregulated . Moreover , AKT-mTORC1 signaling was vigorously suppressed . Previous reports showed that ER stress leads to suppression of insulin receptor signaling in peripheral tissues through hyperactivation of c-Jun N-terminal kinase ( JNK ) and subsequent serine phosphorylation of insulin receptor substrate-1 ( IRS-1 ) ( Ozcan et al . , 2004 ) . It has been recently shown that growth factor receptor bound protein 10 ( GRB10 ) , a key negative regulator of insulin , IGF1 and mTORC1 signaling is activated by ER stress via an ATF4-mediated increase in Grb10 transcription ( Luo et al . , 2018 ) . Interestingly , GRB10 has been implicated in the regulation of β-cell proliferation and function ( Zhang et al . , 2012; Prokopenko et al . , 2014 ) . We found that in neonate Akita islets , the expression of the regulatory unit of PI3 kinase , Pik3r1 ( p85α ) , which is essential for PI3 kinase activation by growth factors is decreased . P85α directly interacts with sXbp1 and mediates its transport to the nucleus ( Winnay et al . , 2014; Park et al . , 2010 ) , hence P85α may have a dual role in the regulation of insulin/IGF1-1 and UPR signaling in response to ER stress . Collectively , multiple mechanisms might be involved in the inhibition of insulin/IGF-1 signaling by ER stress . Of note , mTORC1 activity was also decreased in β-like cells derived from iPS cells from human subjects carrying the INS C96R ( Akita ) mutation ( accompanying paper ) . Accumulating data suggest that mTORC1 is a master regulator of β-cell growth during early development . mTORC1 inhibition by β-cell-specific deletion of Raptor disrupts mitochondrial function , and postnatal β-cell growth and functional maturation ( Ni et al . , 2017 ) , thus mimicking our findings in neonate Akita islets . S6K1 deficiency in mice results in IUGR and impairment of β-cell growth in utero ( Um et al . , 2015 ) . Feeding pregnant mice with a low protein diet decreased β-cell proliferation , mass and function in the offspring in an mTORC1-dependent manner ( Alejandro et al . , 2014 ) . Consistent with this ubiquitous role of mTORC1 in regulating cell size and proliferation , decreased mTORC1 activity in Akita neonate islets was associated with β-cell growth arrest . Others and we have previously shown that mTORC1 promotes ER stress , and its inhibition may prevent apoptosis under ER stress conditions ( Yuan et al . , 2017; Bachar et al . , 2009; Bachar-Wikstrom et al . , 2013; Guha et al . , 2017 ) . Therefore , mTORC1 down-regulation in neonate Akita islets can be viewed as an adaptive response aimed to alleviate ER stress and promote β-cell survival by halting anabolic , energy consuming processes . However , during early stages of development this adaptive mechanism is counter-productive , resulting in marked impairment of β-cell expansion , and consequently leads to future development of diabetes . Therefore , mTORC1 may be viewed as a double-edged sword in the context of β-cell ER stress: on one hand mTORC1 activation may promote ER stress , on the other hand its inhibition early in life impairs β-cell growth and differentiation . Interestingly , it has been recently suggested that mature tissues universally respond to cellular injury by first shutting down mTORC1 , followed by its reactivation which is required for cell cycle entry and tissue repair; this process was termed paligenosis ( Willet et al . , 2018 ) . In Akita mice , reactivation of mTORC1 occurred after weaning and the development of hyperglycemia when the β-cells already lost their ability to proliferate , resulting in permanent β-cell deficiency . Differentiation of Akita β-cells was impaired , evident by decreased expression of genes regulating β-cell identity and function . These changes preceded the development of overt diabetes and are most likely secondary to ER stress per se; this is in contrast with the common view that β-cell dedifferentiation in diabetes is secondary to chronic hyperglycemia ( Wang et al . , 2014 ) . β-Cell transcription factors , including PDX-1 and NKX6 . 1 and their downstream target genes , were decreased in normoglycemic Akita neonates . PDX-1 and NKX6 . 1 instruct β-cell differentiation during development , but are also essential for maintaining β-cell identity and function in adult animals . The latter has also been implicated in the regulation of β-cell proliferation ( Tessem et al . , 2014 ) . Restoration of mTORC1 activity in Akita β-cells increased β-cell expansion and ameliorated diabetes without increasing PDX-1 and NKX6 . 1 expression and glucose-stimulated insulin secretion , further indicating that β-cell growth arrest induced by mTORC1 inhibition plays a key role in the pathophysiology of permanent postnatal diabetes . Our findings have implications not only for the pathophysiology of rare monogenic forms of diabetes , but also for T2D . There is extreme heterogeneity in β-cell mass in healthy individuals as well as subjects with T1D and T2D , which is poorly understood ( Campbell-Thompson et al . , 2016; Cigliola et al . , 2016 ) . Adult β-cell mass is likely a key factor in the risk of developing T2D in the context of obesity and insulin resistance . Since β-cell proliferation is low in the adult , especially in humans , it is generally accepted that impaired β-cell proliferation plays a minor role in the pathophysiology of diabetes . However , genome-wide association studies do point to the importance of β-cell proliferation as a determinant of T2D ( Thomsen et al . , 2016 ) . β-Cell expansion during the fetal and early neonatal period is extensive , and therefore impairment of β-cell proliferation during these early developmental stages will strongly impact the ultimate β-cell mass and function . Our data , although emanating from a neonatal diabetes model can be relevant also to other forms of diabetes , for example T2D in which interplay of genetics ( variants in cell cycle genes ) and environmental factors like viral infections , nutritional stressors or noxious chemicals during the early postnatal period , may induce silent but detrimental effects on β-cell mass via the ER stress-mTOR pathway , predisposing to diabetes in adulthood . Mouse strains used included RIP-Cre ( Gannon et al . , 2000 ) , Rosa26-LSL-Yfp ( Srinivas et al . , 2001 ) , Akita ( Ins2WT/C96Y ) ( The Jackson Laboratory ) , Tsc1fl/fl ( a kind gift from Dr . B . Tirosh , The Hebrew University , Jerusalem ) . The genetic background of the Tsc1fl/fl mice is 129S4/SvJae strain . Ins2C96Y Akita and the RIP-Cre mice were generated on the background of C57BL/6J mice . The Rosa26-LSL-Yfp mice are a mixture of the 129 × 1/SvJ and of C57BL/6J as previously described ( Srinivas et al . , 2001 ) . Akita males were selectively chosen for all analyses , since they develop a more severe form of diabetes compared to females . Mice were housed under similar conditions with 12 hr light/dark cycles with free access to food and water at The Hebrew University animal care unit . For assessment of glucose tolerance , mice fasted for 16 hr or 4 hr were given 1 . 0 or 1 . 5 g/kg glucose IP followed by consecutive blood glucose measurements . Tail blood glucose was monitored using an Accuchek glucometer ( Roche Diagnostics GmbH , Mannheim , Germany ) . For measurement of serum insulin blood samples were collected either from the tail or from the facial vein using heparin coated capillaries or tubes at the start and 15 min after glucose injection . Plasma samples were analyzed using ultrasensitive insulin kits ( Mercodia , Uppsala , Sweden and Crystal Chem Inc . , IL ) . Pancreatic insulin content was analyzed in whole pancreas extracts . Pancreases were isolated , homogenized and insulin was extracted overnight in acid ethanol at 4˚C . Insulin content was determined by an ELISA kit ( Mercodia ) . Animal use was approved by the Institutional Animal Care and Use Committee of the Hebrew University . The rat insulinoma cell line INS-1E was kindly provided by Prof . M . Walker ( The Weizmann Institute of Science , Rehovot , Israel ) and grown as previously described ( Luo et al . , 2018 ) . Mycoplasma contamination was examined periodically and the tests showed no evidence for contamination . Functionality of the cell line was validated by checking periodically their glucose stimulated insulin secretion . Islets were isolated by ductal perfusion of collagenase P ( Roche ) . Hand-picked islets were plated for overnight recovery in RPMI-1640 medium containing 11 . 1 mmol/l glucose ( Biological Industries ) supplemented with 10% FBS , L-glutamine and penicillin-streptomycin in a 37°C , 5% CO2 incubator before experimental procedures . For static glucose-stimulated insulin secretion tests , batches of 25 islets in triplicates or quadruplicates were pre-incubated for 60 min in RPMI-1640 containing 3 . 3 mmol/l glucose , then consecutively incubated at 3 . 3 mmol/l and 16 . 7 mmol/l glucose for 1 hr at 37°C in 200 µl modified Krebs-Ringer bicarbonate buffer containing 20 mmol/l HEPES and 0 . 25% BSA ( KRBH-BSA ) . Medium was collected , centrifuged , and frozen at −20°C and islets were lysed using 0 . 1% BSA-GB/NP-40 . Insulin in medium and islet lysates was determined by ELISA . Pancreases were fixed with zinc-formalin ( neonates ) or 4% buffered formaldehyde ( weaning and adults ) for 3 hr . Paraffin sections ( 5 µm thick ) were rehydrated and antigen retrieval was performed using a Biocare pressure cooker and citrate buffer ( pH = 6 ) . The following antibodies were used: guinea pig anti-insulin 1:200 ( DakoCytomation , Glostrup , Denmark ) , rabbit anti-Ki67 1:200 ( Thermo Scientific , Kalamazoo , MI ) , goat anti PDX-1 1:200 ( kindly provided by Dr . C . V . Wright , Vanderbilt University , TN ) , mouse anti-NKX6 . 1 1:200 ( Developmental Studies Hybridoma Bank ) , PS6 ( Cell Signaling , MA ) , mouse anti E-cadherin 1:100 ( BD Bioscences , NJ ) , mouse anti PCNA 1:500 ( DakoCytomation , Glostrup , Denmark ) rabbit anti H3P 1:100 ( Cell Signaling , MA ) . TUNEL staining was performed with the Roche Cell Death Detection Kit ( Roche Diagnostics ) , cell nuclei were visualized with DAPI staining . Secondary antibodies are all from Jackson Immuno Research Laboratories . Digital images of pancreatic islets were obtained with a Zeiss LSM-710 and Nikon A1R confocal microscope using a x40 oil objective . For analysis of β-cell proliferation and apoptosis , β-cells were counted using Adobe Photoshop CS6 software . To determine β-cell mass , consecutive paraffin sections 75 µm ( in young and adult mice ) or 50 µm ( in newborns ) apart spanning the entire pancreas were stained for insulin and hematoxylin . Digital images were obtained at an original magnification of × 4 with a Nikon C1 confocal microscope , stitched using NIS-Elements software ( Nikon , Melville , NY ) , and the percent area covered by insulin was determined . β-Cell mass was calculated as the product of pancreas weight and percentage insulin area . Protein levels were assessed using antibodies against: total and phospho S6 ribosomal protein ( Ser240/244 ) , insulin receptor substrate 2 ( IRS2 ) , total and phospho-AKT/protein kinase B ( Ser473 and Thr 308 ) , phospho-4EBP-1 ( Thr37/46 ) , BiP , PDX-1 , tubulin , and Hsp90 . Peroxidase-conjugated AffiniPure goat anti-rabbit , anti-chicken and anti-mouse IgG from Jackson ImmunoResearch Laboratories ( West Grove , PA ) were used as secondary antibodies . RNA was extracted using TRI Reagent ( Biolab , Jerusalem , Israel ) and an RNeasy Micro Kit ( Qiagen ) ; samples of 260 ng total RNA were reverse transcribed using a high capacity cDNA Reverse Transcription Kit ( qScript , Quantabio , Beverly , MA ) . Quantitative real-time RT-PCR for total and spliced Xbp1 was performed on a Prism 7000 Sequence Detection System using the Power SYBR Green PCR Master Mix ( Applied Biosystems , Foster City , CA ) . All samples were corrected for glyceraldehyde-3-phosphate dehydrogenase . The following oligonucleotides were used for the PCR of total and spliced Xbp1: forward T-Xbp1 , 5'- AAGAACACGCTTGGGAAT-3' and reverse t-Xbp1: 5'- ACTCCCCTTGGCCTCCAC-3; forward s-Xbp1: 5′-GAGTCCGCAGCAGGTG-3′ and reverse s-Xbp1: 5′- GTGTCAGAGTCCATGGGA-3′ . RNA sequencing libraries were constructed from 120 ng of total RNA using the TruSeq RNA V2 sample prep kit ( Illumina ) . Single read sequencing was performed on Illumina hiSeq2500 to 50 bp . Reads were aligned to the mouse genome GRCm38 using STAR ( v2 . 5 . 2b ) . Quantification of read counts per gene was performed using htseq-count ( version 0 . 7 . 2 ) and differentially expressed genes were identified using DESeq2 package ( version 1 . 12 . 4 ) for normalization and evaluation of differential expression . The significance threshold for comparisons was taken as p value < 0 . 05 . Gene set enrichment analyses were done using Genomica ( http://www . genomespace . org ) and GSEA ( http://software . broadinstitute . org/gsea/index . jsp ) and pathway analyses were carried out using the software Ingenuity Pathway Analysis ( IPA; Ingenuity Systems , http://www . ingenuity . com ) . Statistical analysis was performed using GraphPad Prism 6 . 01 software ( GraphPad Software , La Jolla , CA ) . Differences between multiple groups were analyzed by one-way ANOVA . Two-tailed paired Student’s t test was used to compare differences between two groups . One-sample Student’s t test was performed to validate statistical differences in experiments expressing data as relative to control . Data in graphs and tables are presented as means ±SEM ( standard error of the mean ) . p<0 . 05 was considered significant .	Insulin is a hormone that is crucial for maintaining normal blood sugar levels and is produced by so called β-cells in the pancreas . If the body stops making insulin , or cells stop responding to it , blood sugar levels rise , leading to diabetes . A form of diabetes known as type 1 diabetes , where the body stops making insulin , usually starts in childhood and can sometimes appear during the first six months of life . Infants affected by this early onset of diabetes have mutations in one copy of the gene that encodes insulin . They can still produce half of the amount of insulin , which should be sufficient to control blood sugar to a certain extent . Instead , insulin production stops almost completely after a few months . Scientists believe that this is because the mutant insulin has a toxic effect on β-cells . Mutations in the insulin gene affect the structure of insulin . As a result , insulin builds up in the β-cells , which may eventually cause the cells to die . But the mutant insulin might also cause a problem with a molecule called mTORC1 , which helps β-cells to grow . To investigate this further , Riahi et al . used a mouse model of this form of diabetes to study how stress affects β-cells from birth to adulthood . Mutant β-cells slowed down their rate of cell growth and division early after birth , but did not die more frequently . The results also revealed that β-cells had lower levels of mTORC1 , which probably is the main cause of the reduced cell division and growth . When mTORC1 levels were boosted experimentally , the β-cells started to grow and produce more insulin . Understanding β-cell biology and the link between stress and growth , especially early in life , is a key step in understanding diabetes . In a separate study , Balboa et al . found that human β-cells with insulin mutations also had low mTORC1 and struggled to grow . If boosting mTORC1 could rescue β-cell growth in humans , it could lead to new ways to prevent diabetes .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "cell", "biology", "genetics", "and", "genomics" ]	2018	Inhibition of mTORC1 by ER stress impairs neonatal β-cell expansion and predisposes to diabetes in the Akita mouse
The unconventional secretory pathway exports proteins that bypass the endoplasmic reticulum . In Saccharomyces cerevisiae , conditions that trigger Acb1 secretion via this pathway generate a Grh1 containing compartment composed of vesicles and tubules surrounded by a cup-shaped membrane and collectively called CUPS . Here we report a quantitative assay for Acb1 secretion that reveals requirements for ESCRT-I , -II , and -III but , surprisingly , without the involvement of the Vps4 AAA-ATPase . The major ESCRT-III subunit Snf7 localizes transiently to CUPS and this was accelerated in vps4Δ cells , correlating with increased Acb1 secretion . Microscopic analysis suggests that , instead of forming intraluminal vesicles with the help of Vps4 , ESCRT-III/Snf7 promotes direct engulfment of preexisting Grh1 containing vesicles and tubules into a saccule to generate a mature Acb1 containing compartment . This novel multivesicular / multilamellar compartment , we suggest represents the stable secretory form of CUPS that is competent for the release of Acb1 to cells exterior . A large number of signal sequence lacking proteins are secreted by eukaryotic cells and the examples include: Acyl CoA binding protein 1 ( Acb1 ) , also known as diazepam binding inhibitor ( DBI ) that acts as an allosteric effectors of GABA ionotrophic receptor ( Costa and Guidotti , 1991; Gandolfo et al . , 2001 ) ; FGF1and FGF2 that are required for angiogenesis ( Jackson et al . , 1992; Schäfer et al . , 2004 ) ; the β-galactoside–specific lectins galectin 1 and 3 , blood coagulation factor XIIIa , macrophage migration inhibitory factor ( MIF ) , interleukin ( IL ) -1ß , and the engrailed homeoprotein ( Dupont et al . , 2011; Flieger et al . , 2003; Grundmann et al . , 1988; Joliot et al . , 1997; Rubartelli et al . , 1990; Lutomski et al . , 1997; Menon and Hughes , 1999; Manjithaya et al . , 2010; Duran et al . , 2010; Kinseth et al . , 2007; Nickel and Seedorf , 2008; Rabouille et al . , 2012; Subramani and Malhotra , 2013; Zang et al . , 2015 ) . How are these proteins that cannot enter the endoplasmic reticulum - Golgi complex pathway of protein secretion exported from cells ? Is there a common pathway of their export ? Does their release from the cytoplasm to the extracellular space involve a membrane bounded vesicular intermediate or are they translocated directly from the cytoplasm to the extracellular space via a translocator ? Progress in our understanding of these issues of fundamental importance has been slow because these proteins are released in exceptionally small quantities in a cell type specific and a signal dependent manner ( Malhotra , 2013 ) . This class of proteins is typically detected in the extracellular space by a functional bioassay or ELISA , but these assays do not distinguish between processes involved in the secretion of full-length proteins from those that process the respective cargoes for their presentation in a functional active form . These technical limitations have hampered our understanding of the mechanism of unconventional protein secretion . In 2007 , we made a surprising finding that an ER exit site and Golgi membrane associated peripheral protein GrhA was required for nutrient starvation induced secretion of signal sequence lacking Acyl CoA binding protein ( AcbA ) of 87 amino acids by Dictyostelium discoideum ( Kinseth et al . , 2007 ) . In the extracellular space , AcbA is proteolytically processed into a 34 amino acid peptide , SDF-2 ( spore differentiation factor 2 ) . SDF-2 is required for rapid encapsulation of the prespore cells ( Anjard and Loomis , 2005; Anjard et al . , 1998 ) . Subsequent analysis revealed that secretion of AcbA orthologs of the yeasts Saccharaomyces cerevisiae and Pichia pastoris also required GrhA ortholog called Grh1 ( Duran et al . , 2010; Manjithaya et al . , 2010 ) . In addition , a subset of genes that ordinarily function in the biogenesis of multi-vesicular body ( MVB ) , targeting of membranes to endosomes , fusion of membranes with the plasma membrane , and autophagosome formation were also required for Acb1 secretion ( Duran et al . , 2010; Manjithaya et al . , 2010 ) . However , the secretion of Acb1 was measured by an assay that detected the activity of SDF-2 or an SDF-2-like peptide . This procedure does not distinguish proteins required directly for Acb1 secretion from those with a role in its modification or processing to generate a functional SDF-2 . In our subsequent analyses , we discovered that Grh1 , upon incubation of yeast in starvation medium , translocated from its normal ER exit site/early Golgi residence to one or two larger membrane bound compartments . Based on the shape of the membranes containing Grh1 , we have called these compartments CUPS ( Compartment for Unconventional Protein Secretion ) ( Bruns et al . , 2011 ) . In addition to Grh1 , CUPS contain the early Golgi components Bug1 , Uso1 and Sed5 , but form independent of COPII and COPI dependent vesicular transport ( Cruz-Garcia et al . , 2014 ) . The biogenesis of CUPS requires the PI 4-kinase Pik1 and the Arf-GEF Sec7 . Interestingly , in a vps34Δ mutant CUPS form but breakdown indicating the requirement of PI3P production by Vps34 in the stability of the CUPS ( Bruns et al . , 2011; Cruz-Garcia et al . , 2014 ) . We have now developed a procedure to measure full length secreted Acb1 by extracting the yeast cell wall without causing cell lysis . We have used this assay to characterize the role of the ESCRT proteins in CUPS biogenesis and Acb1 secretion . Our findings reveal that ESCRT-I , -II and –III are involved in Acb1 secretion . In contrast neither ESCRT-0 nor Vps4 are required for this process . These results indicate a Vps4 independent role of ESCRT-III in membrane remodeling . We present the ultra structural analysis of CUPS and the findings that Snf7 , the ESCRT-III component , attaches to CUPS during maturation and is required for their stability . The stable CUPS are found to contain Acb1 . The description and the significance of our findings follow . We were unable to detect full-length Acb1 or SDF-2 directly in the medium of starving S . cerevisiae by immunoprecipitation , western blotting and mass spectrometry ( data not shown ) . We reasoned that full-length Acb1 was likely secreted into the periplasmic space that is between plasma membrane and the cell wall and this pool was cleaved to generate SDF-2 . Once processed , SDF-2 could diffuse into the medium because of its small size ( 34 amino acids ) and/or charge . The cell wall of yeast is composed of glucans , chitin and an outer layer of highly negatively-charged mannoproteins ( Lipke and Ovalle , 1998 ) . Incubating cells in alkaline buffer loosens the cell wall and releases a population of non-covalently bound cell wall proteins ( Figure 1A ) ( Klis et al . , 2007; Mrsă et al . , 1997 ) . In fact , this procedure has been used to report the secretion of signal sequence lacking gluconeogenic , glycolytic enzymes , and the exogenously expressed human Galectin-1 ( Cleves et al . , 1996; Giardina et al . , 2014 ) . But how much of these proteins are released as a result of cell lysis by this procedure ? 10 . 7554/eLife . 16299 . 003Figure 1 . A quantitative assay for Acb1 secretion . ( A ) The cell wall is a highly-charged , porous meshwork of glucans , chitin , and mannoproteins . Incubation in high pH buffers loosens the cell wall , thus allowing some non-covalently bound proteins to be released . ( B ) Standard cell wall extraction procedures employed thus far cause cell lysis . Wild type cells were grown to mid-logarithmic phase , washed twice , and cultured in 2% potassium acetate for 2 . 5 hr ( starvation ) . Cell wall proteins were extracted from equal numbers of growing and starved cells in 100 mM Tris-HCl pH 9 . 4 , 10 mM DTT with mixing at 350 rpm for 30 min at 37°C , followed by precipitation with TCA . Lysates ( L ) and cell wall-extracted proteins ( W ) were analyzed by western blot . ( C ) Mild cell wall extraction conditions do not cause lysis and reveal starvation-specific release of Acb1 . Wild type cells were grown to mid-logarithmic phase , washed twice , and incubated in 2% potassium acetate for 2 . 5 hr ( starvation ) . Cell wall proteins were extracted from equal numbers of growing and starved cells in 100 mM Tris-HCl pH 9 . 4 , 2% sorbitol for 10 min on ice followed by precipitation with TCA . Lysates ( L ) and cell wall-extracted proteins ( W ) were analyzed by western blot . ( D ) Time course of Acb1 secretion during starvation . Wild type cells were grown to mid-logarithmic phase , washed twice , and cultured in 2% potassium acetate for the indicated times . Cell wall proteins were extracted in the mild conditions described in ( C ) and analyzed by western blot . ( E ) The ratio of wall/lysate Acb1 was determined and the average amount of secreted Acb1 at 3 hr was calculated to be 0 . 94% ( SEM = 0 . 16% ) and at 4 hr 0 . 93% ( SEM = 0 . 16% ) of total cellular Acb1 ( n = 3 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 16299 . 003 To distinguish secreted Acb1 from that which leaks into the extracellular space due to cell lysis , we compared the presence of Acb1 in the extracellular space to cofilin ( Cof1 ) , which is not secreted . Acb1 and Cof1 are both small proteins of 10 . 1 kDa and 15 . 9 kDa , respectively , they have similar predicted isoelectric points , and are abundant cytosolic proteins estimated at 142817 and 201065 molcules/cell , respectively ( Kulak et al . , 2014 ) . Cell leakage , rupture of the plasma membrane or lysis during the experimental procedures should have similar effects on Acb1 and Cof1 . Yeast were grown to mid-logarithmic phase and either left untreated or washed twice and starved of nitrogen and glucose by incubation in 2% potassium acetate ( hereafter referred to as starvation ) . After 2 . 5 hr equal number of growing and starved cells were harvested . The cell wall was extracted by the standard procedure for removal of non-covalently bound proteins by incubation with a pH 9 . 4 Tris-HCl buffer in combination with reducing agents and mixing at 37°C ( Cleves et al . , 1996; Giardina et al . , 2014 ) . The proteins extracted under these conditions were analyzed by western blotting with anti-Acb1 and Cof1 antibodies , respectively . The results revealed high levels of Cof1 and Acb1 in the cell wall extracts of both growing and starved cells ( Figure 1B ) , suggesting partial cell lysis . We therefore modified the procedure and incubated cells in pH 9 . 4 Tris-HCl buffer without reducing agents and in the presence of 2% sorbitol to protect against the osmotic stress resulting from loosening the cell wall . The mixture was kept on ice to further preserve the integrity of the plasma membrane . Under these conditions Acb1 , but not Cof1 , was detected in the cell wall extract of starving cells ( Figure 1C ) . Next we monitored secretion of Acb1 during starvation over time comparing cell wall extracts of growing cells and cells starved for up to 4 hr . The spore viability assay previously indicated a burst in SDF-2 activity at 2 . 5 hr post-induction of starvation ( Duran et al . , 2010 ) . Consistently , secretion of full-length Acb1 into the cell wall/periplasm followed similar dynamics , peaking at 3 hr of starvation ( Figure 1D ) . The levels of a soluble cell wall protein , Bgl2 , were also analyzed as a marker of cell wall extraction efficiency . The amount of Bgl2 extracted from starved cells remained constant throughout starvation , indicating that secretion of Acb1 increased during starvation ( Figure 1D ) . The levels of cell wall extracted and lysate Acb1 were quantitated and the amount of Acb1 secreted after 3 hr of starvation was calculated to be on average 1% of the total intracellular Acb1 ( Figure 1E ) . This procedure to detect Acb1 is not without limitations however . Culturing cells at 37°C , prior to extraction , resulted in Cof1 release ( data not shown ) , which precludes testing of temperature-sensitive yeast mutants for their involvement in Acb1 secretion . High concentrations of DMSO , which is often used as a solvent to dissolve chemicals such as Latrunculin A or Brefeldin A , also led to Cof1 release ( data not shown ) . Therefore , DMSO soluble chemical inhibitors , in general , cannot be tested for their effects on Acb1 secretion . Finally , some yeast mutant strains are hyper osmosensitive and cannot therefore be tested by our procedure as they reveal a persistent Cof1 signal upon cell wall extraction . Regardless of these caveats , this assay for the first time provides a quantitative and a reliable measure of Acb1 secretion . We first tested the involvement of Grh1 that has been shown previously to be involved in Acb1 secretion based on the spore viability assay in D . dictyostelium ( Kinseth et al . , 2007; Duran et al . , 2010 ) . Wild type and grh1Δ cells were grown to mid-logarithmic phase , starved in potassium acetate and extracted by the procedure described above . We chose to monitor secretion after 2 . 5 hr starvation to further avoid the potential of lysis associated with longer periods of starvation . The ratios of cell wall to lysate Acb1 from potassium acetate cultured cells were quantitated and the effect of loss of Grh1 was normalized to percent of wild type . No Acb1 was detected in the cell wall of growing grh1Δ cells ( data not shown ) , while grh1Δ cells incubated in potassium acetate showed ~80% reduction in Acb1 secretion compared to wild-type cells ( Figure 2A–B ) . This confirms the involvement of Grh1 in Acb1 secretion . 10 . 7554/eLife . 16299 . 004Figure 2 . Acb1 secretion requires Grh1 and a subset of ESCRT proteins . Wild type and deletion strains were grown to mid-logarithmic phase , washed twice , and incubated in 2% potassium acetate for 2 . 5 hr . Cell wall proteins were extracted from equal numbers cells in 100 mM Tris-HCl pH 9 . 4 , 2% sorbitol for 10 min on ice followed by precipitation with TCA . Lysates and cell wall-extracted proteins were analyzed by western blot . The ratio of wall/lysate Acb1 during starvation was determined and compared to that of wild type in each experiment . Statistical analyses were performed for each gene deletion and are represented as% of wild type ( paired student’s t-test ) . ( A ) Grh1 is required for Acb1 secretion . Representative cell wall extractions monitored by western blot of wild type and grh1Δ cells . ( B ) grh1Δ cells secreted on average 80 . 3% less Acb1 than wild-type cells ( p = 0 . 0045 , SEM = 3 . 8% , n= 4 ) . ( C ) Acb1 secretion requires ESCRT complexes I , II and III but not ESCRT-0 . Representative cell wall extractions monitored by western blot of wild type and one deletion strain from each ESCRT complex . ( D ) Quantification of all ESCRT deletion strains tested ( n = 4 or more ) . ESCRT-0; vps27Δ ( +17% , SEM = 3 . 2% , p > 0 . 2 ) , hse1Δ ( +8% , SEM = 0 . 9% , p > 0 . 2 ) . ESCRT-I; vps23Δ ( -81 . 8% , SEM = 7 . 9% , p = 0 . 0046 ) , vps37Δ ( -81 . 5% , SEM = 8 . 4% , p = 0 . 001 ) , vps28Δ ( -69 . 4% , SEM = 11 . 8% , p = 0 . 026 ) . ESCRT-II; vps25Δ ( -78 . 8% , SEM = 14 . 1% , p = 0 . 021 ) , vps36Δ ( -81 . 7% , SEM = 2 . 9% , p = 0 . 0055 ) , vps22Δ ( -71 . 6% , SEM = 7 . 7% , p = 0 . 032 ) . ESCRT-III; snf7Δ ( -68 . 5% , SEM = 4 . 4% , p = 0 . 055 ) , vps20Δ ( -71 . 5% , SEM = 6 . 5% , p = 0 . 022 ) , vps24Δ ( -77 . 4% , SEM = 7 . 5% , p = 0 . 049 ) , vps2Δ ( -66 . 7% , SEM = 11 . 3% , p = 0 . 0006 ) . ( E ) Vps4 and accessory proteins are not required for Acb1 secretion . Representative cell wall extractions monitored by western blot of wild type and indicated deletion strains . ( F ) Quantification of indicated deletion strains ( n = 4 or more ) . No differences were determined to be statistically significant . DOI: http://dx . doi . org/10 . 7554/eLife . 16299 . 004 Next , we tested the role of the endosomal sorting complexes required for transport ( ESCRTs ) that are involved in the formation of intraluminal vesicles in endosomes to generate a multivesicular body ( MVB ) , some of which we previously reported to play a role in Acb1 secretion ( Duran et al . , 2010 ) . Specifically , we tested Vps23 , Vps37 and Vps28 of ESCRT-I , which showed 82% , 82% and 69% reduction in Acb1 secretion , respectively . All members of ESCRT-II , consisting of Vps36 , Vps25 and Vps22 , were tested resulting in 82% , 79% and 72% decrease in Acb1 secretion , respectively . Similarly , all four core components of ESCRT-III , Snf7 , Vps20 , Vps24 and Vps2 , were examined and all showed similar reductions in secretion of 69% , 72% , 78% and 67% respectively ( Figure 2C–D ) . Vps27 and Hse1 of ESCRT-0 were not required for Acb1 secretion ( Figure 2C–D ) . To complete the assessment of the role of the MVB pathway in Acb1 secretion we tested the role of Vps4 , the AAA-ATPase required for the disassembly of ESCRT-III ( Babst et al . , 1998 ) . Yet , loss of Vps4 did not affect Acb1 secretion ( Figure 2E–F ) . Vps4 activity in the MVB pathway is regulated by interactions with various accessory proteins . Did2 and Ist1 proteins modulate ESCRT-III disassembly by recruiting Vps4 , while Vta1 and Vps60 form a complex that positively regulates the activity of Vps4 on endosomes ( Xiao et al . , 2008; Azmi et al . , 2006; Rue et al . , 2008; Shestakova et al . , 2010; Nickerson et al . , 2006 ) . Deletion strains for each of these accessory proteins were tested for Acb1 secretion by the cell wall extraction assay and none exhibited a reduction in Acb1 secretion ( Figure 2E–F ) . These results suggest that Acb1 secretion requires ESCRT-III function but not Vps4 activity . This is an unexpected finding since all other known ESCRT dependent processes require ESCRT-III and Vps4 . Our previous findings revealed the requirement of a subset of ESCRT proteins in CUPS biogenesis ( Bruns et al . , 2011 ) . In our previous analysis Vps27 , Hse1 , Vps23 , Mvb12 , Vps23 , and Vps4 were found to have no role in CUPS biogenesis , whereas Vps36 , Vps25 , Vps2 and Vps20 deletions revealed varying , intermediate effects at 4 hr after starvation , which we know now to be much longer than required for Acb1 secretion or CUPS biogenesis ( Bruns et al . , 2011; Cruz-Garcia et al . , 2014 ) . In the current analysis we examined the requirements of ESCRT proteins in CUPS formation at 30 min and 2 . 5 hr of starvation to discern effects on biogenesis and stability and included ESCRT proteins that had not been tested so far . Cells were quantified for presence of CUPS ( 1–3 large punctae ) , intermediate CUPS formation ( large punctae and small punctae ) or no CUPS ( only small punctae ) . None of the ESCRT proteins tested affected the initial CUPS formation and organization after 30min of starvation ( Figure 3A ) . After 2 . 5 hr of starvation , we observed a requirement of ESCRT-II ( Vps36 , Vps22 and Vps25 ) and ESCRT-III complexes ( Vps20 , Vps2 , Vps24 and Snf7 ) , but not ESCRT-0 ( Vps27 and Hse1 ) , ESCRT-I ( Vps23 and Vps28 ) or Vps4 proteins , as also reported previously after 4 hr starvation ( Figure 3A ) and Bruns et al . ( 2011 ) . The loss of Vps20 and Snf7 of ESCRT-III had the most dramatic effect on CUPS formation after 2 . 5 hr of starvation ( Figure 3A–B ) . 10 . 7554/eLife . 16299 . 005Figure 3 . The involvement of ESCRTs in CUPS formation , stability and disassembly . ( A ) Wild type and the indicated deletion strains expressing Grh1-2xGFP were grown to mid-logarithmic phase , washed twice , and incubated in 2% potassium acetate for 30 min and 2 . 5 hr to assess CUPS biogenesis and stability , respectively . Cells were grouped in 3 classes: CUPS ( 1–3 large punctae per cell ) , intermediate ( large and small punctae ) , no CUPS ( multiple small punctae ) . Between 50–200 cells were counted for each strain/condition from 3 independent experiments . ( B ) Effect of loss of Snf7 of ESCRT-III at 30 min and 2 . 5 hr starvation . ( C ) Wild type and vps4Δ cells were starved for 2 . 5 hr as in ( A ) to allow CUPS formation . Cells were washed once and cultured in rich media for 45 min . Scale bars = 2 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 16299 . 005 We have shown previously that upon shifting cells from starvation to growth medium CUPS relocated to the ER by a Sly1 and COPI dependent pathway ( Cruz-Garcia et al . , 2014 ) . We assessed whether ESCRT proteins affected CUPS absorption into the ER upon re-growth . The same cells as above were starved for 2 . 5 hr and then incubated in growth medium for 45 min . The wild type cells revealed the expected relocation of Grh1-2xGFP to ER exit sites/early Golgi membranes . This relocation was unaffected by loss of any ESCRT protein , including Vps4 ( data not shown and Figure 3C ) . Therefore , CUPS maturation during starvation depends on ESCRT-II and III complex members , but CUPS reabsorption into the ER upon re-growth is independent of ESCRT proteins . Loss of Snf7 exhibited a strong defect in CUPS morphology after 2 . 5 hr of starvation , and because it is the most abundant ESCRT-III protein , contributing 50% of the predicted 450 kDa ESCRT-III complex , we decided to localize Snf7 in starving cells ( Teis et al . , 2008 ) . The C-terminal fusion of bulky tags such as GFP or RFP to all four ESCRT-III subunits is known to act as dominant-negative with respect to MVB sorting by interfering with auto inhibition of the C-terminal tail ( Teis et al . , 2008 ) . We examined Grh1-2xGFP localization when Snf7-RFP was integrated in the genome and found it also acted as dominant-negative with respect to CUPS stability , exhibiting the same phenotype as loss of Snf7 ( Figure 4A ) . Snf7-RFP localized to multiple punctate elements , representing dysfunctional MVBs . Interestingly , upon starvation , the Grh1 and Snf7 containing structures were often juxtaposed or slightly overlapping ( Figure 4A ) . An established procedure to overcome this inhibitory effects is to co-express wild type and Snf7-RFP in the same cells ( Guizetti et al . , 2011 ) . To ensure that co-expression of Snf7-RFP did not affect MVB pathway we assessed the trafficking of MVB cargo Cps1 in cells expressing empty vector or Snf7-RFP under the control of its own promoter . Our findings reveal that expression of Snf7-RFP did not block trafficking of GFP-Cps1 to the vacuole ( Figure 4—figure supplement 1 ) . We therefore co-expressed exogenous Snf7-GFP with untagged endogenous Snf7 in cells expressing Grh1-2xmCherry . Under these conditions CUPS formation after 2 . 5 hr of starvation was largely restored , with more than 70% of cells exhibiting morphological normal CUPS ( Figure 4B ) . In growth conditions , Snf7-GFP localized to small punctate elements and the vacuole membrane ( Figure 4B ) . The vacuole membrane location is likely a result of inefficient disassembly of tagged Snf7 . Upon starvation , Snf7 localized predominantly to 1–3 large structures per cell , and the vacuole membrane labeling was largely reduced ( Figure 4B ) . Cells with very high Snf7-GFP expression displayed altered CUPS morphology ( data not shown ) , we therefore focused on cells that exhibited normal CUPS and found that ~15% of such cells co-localized Snf7-GFP to Grh1-2xmCherry containing CUPS ( Figure 4B ) . To further ascertain the location and dynamics of Snf7 to Grh1 containing CUPS , we performed time-lapse confocal imaging during starvation . Our results showed that Snf7 and Grh1 exist on separate compartments for the majority of the time during starvation . However , these compartments periodically contacted and/or coalesced ( Figure 4C–D ) . Quantitative analysis of cells with Snf7-GFP expression and Grh1-2xmCherry on CUPS indicated that co-localization occurred transiently only after 1 hr of starvation ( see also Figure 6A ) . Between 1 and 2 . 5 hr of starvation , analysis of 62 cells revealed Snf7 and Grh1 compartments co-localized or overlapped rapidly 17 times in 17 different cells for 10–20 s . Interestingly , between 2 . 5 and 3 hr of starvation , the time when Acb1 secretion peaks , analysis of 96 cells revealed 37 co-localization events in 25 cells , meaning co-localization was observed more than once per cell in some cases . Of these 37 co-localization events , 22 were rapid , lasting only 10–30 s ( examples shown in Figure 4C–D , arrowheads ) , while 15 were stable , sometimes lasting for up to 10-minutes ( examples shown in Figure 4C–D , arrows ) . In summary , the frequency and duration of co-localization increased throughout starvation , correlating with both the timing of Acb1 secretion and the requirement of Snf7 in CUPS stability but not with the initial steps of CUPS biogenesis ( Figures 2 and 3 ) . Due to photo bleaching , it was not possible to visualize the same cell throughout the time-course of starvation . Despite this , from the combined data we propose that all CUPS at one point contact or fuse with the Snf7 containing compartment and this is necessary for the final step of maturing CUPS into a secretion competent organelle . In the absence of Snf7 , this final maturation might be blocked and Grh1 containing CUPS disintegrate . One possibility is that Snf7 is required for sealing of the membranes that compose CUPS to generate a compartment sealed from the cytoplasm . 10 . 7554/eLife . 16299 . 006Figure 4 . Snf7 localizes transiently to CUPS . ( A ) Genomically integrated Snf7-RFP and Grh1-2xGFP were visualized by fluorescence microscopy during growth in mid-logarithmic phase and after incubation in 2% potassium acetate for 2 . 5 hr . ( B ) Snf7-GFP was expressed exogenously in cells expressing genomically integrated Grh1-2xmCherry ( Grh1-2xCh ) and visualized by fluorescence microscopy during growth in mid-logarithmic phase and after incubation in 2% potassium acetate for 2 . 5 hr . ( C–D ) The same cells from ( B ) were visualized over time throughout starvation by confocal spinning disk microscopy . Cells with Grh1-2xmCherry on CUPS and Snf7-GFP expression were quantitated for frequency and duration of co-localization and/or overlap of the two structures . Rapid ( less than 30 s ) and stable ( more than 30 s ) co-localization examples are indicated by arrowheads and arrows , respectively . All scale bars = 2 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 16299 . 00610 . 7554/eLife . 16299 . 007Figure 4—figure supplement 1 . GFP-Cps1 transport is unaffected in pSnf7-RFP expressing cells . Wild type cells co-expressing GFP-Cps1 ( pRS415 ) with either empty pRS416 vector or pSnf7-RFP were grown mid-logarithmic phase and visualized by fluorescence microscopy . Scale bar = 2 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 16299 . 007 The ordered assembly of ESCRT complexes for sorting of transmembrane cargoes into MVBs is well defined ( reviewed in Hurley and Emr , 2006; Piper and Katzmann , 2007 ) . ESCRT-0 binds and recognizes ubiquitinated cargoes that in turn recruit ESCRT-I , followed by ESCRT-II ( Katzmann et al . , 2003; Kostelansky et al . , 2006 ) . Vps20 of ESCRT-III binds directly to Vps25 of ESCRT-II and initiates ESCRT-III filament formation by recruiting Snf7 and nucleating its homo oligomerization ( Teis et al . , 2008; 2010 ) . Binding of Vps2 and Vps24 caps ESCRT-III filament formation and followed by recruitment of Vps4 to constrict the necks of budding ILVs and to finally disassemble the ESCRT-III filaments ( Babst et al . , 1998; Saksena et al . , 2009; Adell et al . , 2014 ) . Is the attachment of Snf7 to CUPS similar to the order of events as in MVB biogenesis ? Snf7-RFP was expressed exogenously from its own promoter in wild type and ESCRT deleted Grh1-2xGFP expressing strains . The heterogeneity in Snf7-RFP expression levels was more pronounced in the deletion strains compared to wild type cells . A small percentage of mutant cells with very high Snf7-RFP levels had aberrant accumulations and were excluded from further investigation . Snf7 recruitment to membranes in growth conditions was as expected for the various ESCRT deletions ( Figure 5 ) . We found that loss of ESCRT-I ( Vps23 , Vps28 ) , -II ( Vps36 , Vps25 ) or Vps20 ( ESCRT-III ) proteins abolished the vacuolar membrane localization of Snf7 and resulted in a more diffuse cytosolic pattern with an occasional faint spot ( Figure 5 ) . Loss of ESCRT-III components Vps2 or Vps24 resulted in Snf7 localization that was very similar to wild type cells , with vacuole membrane accumulation and an occassional discrete spot . Cells lacking Vps4 displayed a distinct Snf7 localization compared to other ESCRT deletions . In this instance , Snf7 accumulated only in punctate structures and not the vacuole membrane or cytosol ( Figure 5 ) . 10 . 7554/eLife . 16299 . 008Figure 5 . Recruitment of Snf7 to membranes for Acb1 secretion and MVB pathway share a similar mechanism . Snf7-RFP was expressed exogenously from its own promoter in wild type and the indicated ESCRT deletion strains expressing Grh1-2xGFP . Cells were visualized by fluorescence microscopy during growth in mid-logarithmic phase ( growth ) and after incubation in 2% potassium acetate ( starvation ) for 2–2 . 5 hr . Scale bar = 2 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 16299 . 00810 . 7554/eLife . 16299 . 009Figure 6 . Snf7 recruitment to CUPS and Acb1 secretion are accelerated in vps4Δ cells . ( A ) Snf7-RFP was expressed exogenously from its own promoter in wild type and vps4Δ cells expressing Grh1-2xGFP . Cells were starved and immediately visualized by time-lapse confocal microscopy ( Scale bar = 2 μm ) . ( B–C ) Wild type and vps4Δ were grown to mid-logarithmic phase , washed twice , and incubated in 2% potassium acetate for 1 hr and 2 . 5 hr . Cell wall proteins were extracted as before and analyzed by western blotting . Acb1 levels were quantified and the ratio of wall/lysate Acb1 was determined . n=3 ( D–E ) Snf7 polymerization is required for Acb1 secretion . snf7Δ cells expressing wild type Snf7 , Snf7-L121D or empty vector were grown to mid-logarithmic phase , washed twice , and incubated in 2% potassium acetate for 2 . 5 hr . Cell wall proteins were extracted as before and analyzed by western blotting . The ratio of wall/lysate Acb1 was determined ( n=3 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 16299 . 009 We then monitored Snf7 location in ESCRT deleted cells starved for 2–2 . 5 hr , which is the time with frequent location of Snf7 to Grh1 containing CUPS ( Figure 5 ) . In wild type cells , Snf7-RFP localization changes to 1–3 prominent foci with reduced vacuole membrane staining ( see also Figure 4B–D ) . As mentioned above , we observed mild perturbation in CUPS formation in wild type cells expressing Snf7-RFP . Similarly , Snf7-RFP expression exacerbated the defects in CUPS morphology in most ESCRT mutants ( Figure 5 ) . Even loss of ESCRT-I components , which normally did not appear to affect CUPS morphology , resulted in more cells with incomplete CUPS formation upon Snf7-RFP expression . Similar to the situation in growth conditions , ESCRT-I mutants displayed more cytosolic distribution of Snf7-RFP and there was no vacuole membrane localization compared to wild type cells . This suggests that , as in growth , Snf7 recruitment to membranes is impaired in these cells upon starvation . However , Snf7 was recruited to 1–2 foci in some cells , although less efficiently , and this could , at times , be co-localized to Grh1 containing CUPS ( Figure 5 ) . This is expected , as loss of Snf7 recruitment should lead to breakdown of CUPS . The CUPS morphology was severely affected in ESCRT-II mutants and vps20Δ cells . Snf7 was mostly cytosolic in these cells , similar to ESCRT-I mutants , however much larger accumulations in 1–2 dots were often observed , but these aberrant structures did not appear to co-localize with Grh1 . Such accumulations were also occasionally observed in growth conditions in these deletion strains and we suggest the corresponding structures are dysfunctional that result from sequestration of Snf7 in the absence of the nucleator Vps20 . Snf7-RFP in cells lacking Vps24 or Vps2 did not change localization upon starvation and was retained mostly at the vacuole membrane and multiple small foci ( Figure 5 ) . Notably , Snf7 was not detected in larger foci as in wild type cells ( Figure 5 ) . In vps4Δ cells Snf7-RFP localization did not change dramatically when compared to growth and again displayed a unique localization when compared to other ESCRT deletions . Cells with very high Snf7-RFP levels also altered CUPS morphology , similar to ESCRT-I mutants , however , unlike in vps2/24Δ cells , Snf7 localized mostly to larger foci that could be observed to co-localize with Grh1 at times ( Figure 5 ) . Altogether the data imply that recruitment of Snf7 to a structure for unconventional secretion follows a similar hierarchy , where ESCRT-I and II enhance Snf7 recruitment and Vps20 acts as the nucleator . Loss of Vps2 or Vps24 resulted in Snf7 immobilization upon starvation which could mean Snf7 was sequestered in growth before starvation or Vps2/Vps24 are directly required for Snf7 relocalization in starvation . Snf7 did localize to punctate elements in vps4Δ cells , albeit not exactly as in wild type cells , but as CUPS morphology was not affected and Snf7 co-localization to CUPS could be observed this begins to explain the independence of Vps4 in this process . We wanted to examine in more detail the recruitment of Snf7 to CUPS during starvation in the absence of Vps4 . We therefore performed time-lapse confocal imaging as in Figure 4 and found Snf7 was recruited to CUPS membranes much earlier in starvation than in wild type cells ( Figure 6A ) . We examined wild type and vps4Δ cells exogenously expressing Snf7-RFP during the first hour of starvation . In wild type cells there was no detectable localization of Snf7-RFP to CUPS , consistent with our findings in Figure 4 . Specifically , analysis of 65 wild type cells between 10 and 50 min of starvation revealed no co-localization of Grh1 and Snf7 ( Figure 6A – WT 25 min ) . In fact , Snf7 foci had not completely formed as later in starvation , while in vps4Δ cells Snf7 localized almost exclusively to such foci ( Figure 6A ) . Analysis of 72 vps4Δ cells between 10 and 50 min of starvation revealed Snf7 localized to CUPS in 20 cells ( in 2 cases twice per cell ) , usually rapidly , for 10–30 s ( example shown in Figure 6A – vps4Δ 15 min – arrowheads ) . We observed 2 stable co-localization events but these were associated to larger Snf7 accumulations ( example shown Figure 6A – vps4Δ 20 min – arrow ) . Later in starvation there was no appreciable difference in the frequency or duration of Snf7 localization to CUPS in wild type versus vps4Δ cells ( data not shown ) . Therefore Snf7 recruitment to CUPS is accelerated in vps4Δ cells . This prompted us to also examine Acb1 secretion early in starvation . Indeed , we observed a modest acceleration of Acb1 secretion in vps4Δ cells . After 1 hr starvation wild type cells had secreted very little Acb1 , whereas vps4Δ cells secreted more than two fold more Acb1 at this time point ( Figure 6B–C ) . We had consistently observed a trend of slight hyper-secretion of Acb1 in vps4Δ cells at 2 . 5 hr starvation ( Figure 2E–F and Figure 6B–C ) , which could be explained as a result of an increased kinetics of Snf7 recruitment to CUPS and thus an increased rate of Acb1 export . To further explore the function of Snf7 for Acb1 secretion we tested a point mutant ( L121D ) that is unable to homooligomerize and is therefore defective in ESCRT-III filament formation ( Saksena et al . , 2009 ) . We expressed wild type or mutated Snf7 in snf7Δ cells and examined Acb1 secretion after 2 . 5 hr of starvation . Cells expressing empty vector or Snf7-L121D were unable to secrete Acb1 , in contrast to cells expressing wild type Snf7 ( Figure 6D–E ) . Therefore the ability of Snf7 to polymerize into filaments is required for Acb1 secretion , although Vps4 mediated disassembly is not . In fact , inhibiting the latter leads to slightly accelerated secretion . Immunoelectron microscopy of cells incubated in potassium acetate revealed the presence of Grh1 to a cup-shaped membranous compartment and a collection of small vesicles ( Bruns et al . , 2011 ) . In order to better characterize- in 3D - the Grh1 containing membranes , we made use of the correlative light electron microscopy procedure ( CLEM ) . Cells expressing Grh1-2xmCherry were subjected to high pressure freezing and thin sectioning . The 300 nm sections were first visualized by fluorescence microscopy to locate areas of interest to be analyzed by electron tomography microscopy , followed by three-dimensional reconstruction of the Grh1 containing membranes . In growing cells , as expected , Grh1 was found predominantly in the vicinity of typical Golgi-like cisternae and vesicles that were indistinguishable in wild type and snf7Δ cells ( Figure 7A – WT growth , Figure 7—figure supplement 1 and data not shown ) . 10 . 7554/eLife . 16299 . 010Figure 7 . Ultrastructure of CUPS and the involvement of Snf7 in their stability . CUPS are revealed as a spheroidal collection of highly curved membranes of average 200 nm diameter ( WT starvation 1 , 3D Model ) . ( A ) Grh1-2xmCherry expressing cells were grown to mid-logarithmic phase , washed twice , and incubated in 2% potassium acetate for 2 . 5–3 hr . Growing or starved cells were subjected to cryofixation and correlative light and electron microscopy ( CLEM ) ( see Materials and methods ) . High magnification tomograms were acquired and 3D models were reconstructed of the membranes positive for mCherry signal . In the case of wild type cells , 13 tomograms were acquired in growth and 16 during starvation . For snf7Δ cells , 14 tomograms were acquired in starvation ( the organization of Grh1-2xmCherry positive membranes during growth was indistinguishable from wild type ) . In 50% of CUPS structures identified in wild type cells a saccule surrounding CUPS was observed ( WT starvation ) . ( B ) Classification of Grh1-positive membranes and measurements of CUPS structures at 45 min and 2 . 5 hr starvation . At 45 min starvation 12 tomograms each from wild type and snf7D cells were analyzed . 'Vesicles' refers to vesicles and small cisternae . The diameter of the CUPS structures was measured along the longest axis . In the case of 'CUPS + saccule' the CUPS structures were small , with an average diameter of 200 nm . DOI: http://dx . doi . org/10 . 7554/eLife . 16299 . 01010 . 7554/eLife . 16299 . 011Figure 7—figure supplement 1 . Remaining 3D models of Grh1-positive membranes from Figure 7A . DOI: http://dx . doi . org/10 . 7554/eLife . 16299 . 01110 . 7554/eLife . 16299 . 012Figure 7—figure supplement 2 . Examples of CLEM analysis from wild type and snf7Δ cells at 45 min of starvation . Grh1-positive membranes from 12 tomograms each from wild type and snf7Δ cells of were analyzed . DOI: http://dx . doi . org/10 . 7554/eLife . 16299 . 012 We identify CUPS as a convoluted network of tubules and vesicles with an overall spheroidal shape , a cavity at the center , and an average diameter of ~200 nm ( 14 of 16 tomograms at 2 . 5 hr starvation ) ( Figure 7A – WT starvation 1 and Video 1 ) . We followed the development of CUPS , based on the location of Grh1-2xmCherry at 45 min and 2 . 5 hr of starvation . At 45 min of starvation the Grh1 containing structures were more heterogeneous in size , but in general became larger throughout starvation ( Figure 7A–B and Figure 7—figure supplement 2 ) . As the structure became larger , the 'sphere' appeared smoother , or less fenestrated . Of particular note was the presence of a large fenestrated cisternae or saccule that seemed to engulf the Grh1-positive membranes ( Figure 7A – WT starvation 2 and Video 2 ) . This was present in just 15% of CUPS at 45 min starvation and was increased to 50% of the CUPS structures at 2 . 5 hr ( Figure 7B ) . The 'engulfed' CUPS were on average ~200 nm in size . 10 . 7554/eLife . 16299 . 013Video 1 . Morphology of CUPS Full tomogram and 3D reconstruction of Grh1-2xmCherry positive membranes corresponding to Figure 7A – WT – starvation 1 – 'CUPS alone' . DOI: http://dx . doi . org/10 . 7554/eLife . 16299 . 01310 . 7554/eLife . 16299 . 014Video 2 . Morphology of stable CUPS Full tomogram and 3D reconstruction of Grh1-2xmCherry positive membranes corresponding to Figure 7A – WT – starvation 2 – 'CUPS + saccule' . DOI: http://dx . doi . org/10 . 7554/eLife . 16299 . 014 In cells lacking Snf7 at 45 min of starvation there were no significant differences found in the organization of Grh1-positive membranes , as we observed by fluorescence microscopy ( Figure 7B and Figure 7—figure supplement 2 , see also Figure 3A–B ) . However , after 2 . 5 hr of starvation more than 80% of the Grh1-positive membranes were identified as small vesicles ( Figure 7A – snf7Δ starvation ) . Moreover , the large saccule , which was observed associated with Grh1 containing membranes in wild type cells , was never identified in snf7Δ cells at 2 . 5 hr starvation ( Figure 7A–B ) . Our data shows that approximately 1% ( roughly 1400 molcules of Acb1 per cell ) are secreted upon nutrient starvation ( Figure 1 ) . Thin sections of fixed yeast cultured in potassium acetate for 2 . 5 hr were visualized by immunoelectron microscopy with anti-GFP and anti-Acb1 specific antibodies that can bind and detect the cognate protein only when the latter is concentrated . The results reveal three major features . 1 , Grh1-2xGFP ( 15 nm gold ) is contained in tubules and vesicles , 2 , Grh1 is proximal to a cup-shaped membrane , and 3 , Grh1 is encased in a saccule , which contains Acb1 ( 10 nm gold ) ( Figure 8 ) . 10 . 7554/eLife . 16299 . 015Figure 8 . Immunoelectron microscopy identifies Acb1 in stable CUPS . Grh1-2xGFP expressing cells were starved for 2 . 5 hr and processed for immunogold labeling and electron microscopy ( see Materials and methods ) . GFP = 15 nm gold , Acb1 = 10 nm gold . Stage 1 – Grh1 labels tubulo-vesicular structure . Stage 2 – A sheet/saccule approaches and initiates engulfment of the Grh1 labelled structure ( white arrows ) . Stage 3 – The Grh1 structure is completely engulfed and contains Acb1 – mature or stable CUPS . DOI: http://dx . doi . org/10 . 7554/eLife . 16299 . 015 Based on this data alone , it would be inaccurate to propose these as sequential steps in the pathway of Acb1 secretion , but these images , combined with the CLEM analysis are suggestive of a possible pathway for the formation of CUPS . The initial Grh1 positive tubulo-vesicular clusters resemble the typical mammalian ERGIC compartment ( Figure 8 – Stage 1 , Figure 9 – immature CUPS ) . At this stage Acb1 is rarely found in these structures . In the next stage , a tubular structure ( probably a sheet like structure in 3D ) approaches and initiates the engulfment of the Grh1 positive immature CUPS ( Figure 8 – Stage 2 , white arrows ) . We believe this to be the saccule identified by our CLEM analysis ( Figure 7A – WT starvation 2 and Video 2 ) . This Grh1 encasing compartment also contains Acb1 ( Figure 8 – Stage 3 , Figure 9 – stable CUPS ) . 10 . 7554/eLife . 16299 . 016Figure 9 . A schematic presentation of steps in Acb1 secretion . Grh1 containing vesicles and tubules assemble into foci that are clearly visible as one to two spots per cell . This collection of membranes are formed and consumed constitutively during growth ( immature CUPS ) . However , upon shifting cells to starvation medium , the immature CUPS become encased in a saccule ( yellow membrane ) by an Snf7 mediated reaction . We call this membrane-bounded compartment 'stable CUPS' . This stable form of CUPS is in fact found to contain Acb1 . Stable CUPS are quite different from a standard MVB . The stable CUPS are double membrane bounded and contain internal membranes of different sizes and shapes . An MVB on the other hand , is composed of a single bilayer containing uniformed sized vesicles . The key difference also is that Vps4 is required for the formation of an MVB , but not for the Snf7 dependent engulfment of Grh1 containing immature CUPS . The stable CUPS then release Acb1 to the exterior of the cells . DOI: http://dx . doi . org/10 . 7554/eLife . 16299 . 016 There are numerous reports of exosomes as vehicles for the release of cytoplasmic proteins in the extracellular space . These extracellular vesicles are derived from the fusion of multivesicular bodies ( MVBs ) to the cell surface ( reviewed in Bobrie et al . , 2011 ) . While a number of issues on the role of exosomes in secretion remain unclear , our new findings strongly indicate that canonical MVBs are not involved in Acb1 secretion . Acb1 secretion requires ESCRT-I , -II and -III proteins but not ESCRT-0 or Vps4 proteins ( Figure 3 ) . The lack of ESCRT-0 involvement is not surprising as its main function is to recognize transmembrane MVB cargoes . Moreover , other ESCRT dependent membrane fission events , such as HIV budding , cytokinesis and nuclear envelope re-sealing after mitosis also require only a subset of ESCRT proteins and are independent of ESCRT-0 ( Morita et al . , 2007; Garrus et al . , 2001; Vietri et al . , 2015 ) . However , the function of Vps4 is essential for all aforementioned ESCRT-dependent functions . The pathway of Acb1 secretion requires ESCRT-III , but is independent of Vps4 . Grh1 relocates from ER exit site/intermediate compartment like structures in growing yeast cells to a new compartment called CUPS upon starvation ( Bruns et al . , 2011; Cruz-Garcia et al . , 2014 ) . The lack of requirement for ESCRT-0 and Vps4 in CUPS formation correlates with our secretion data indicating these proteins , and thereby the MVB pathway , are not required for Acb1 secretion . Our findings also show that loss of Vps4 promotes an earlier location of Snf7 to CUPS and this correlates with faster Acb1 secretion ( Figure 6 ) . Deletion of Snf7 did not affect the initial assembly of CUPS from small Grh1 containing vesicles , but these immature CUPS were unstable , did not proceed to become fully mature CUPS and finally broke into small clusters of vesicles ( Figures 3 and 7 ) . This indicates a role of Snf7/ESCRT-III in maintaining the structural integrity of CUPS . We observed that Grh1 and Snf7 reside on different compartments that contact and at times coalesce ( Figures 4 and 6 ) . Based on our findings , we suggest that Grh1 containing membranes contact and are encased in Snf7 containing compartments and this produces stable CUPS . Based on the CLEM data we suggest that Grh1 containing immature CUPS likely form constitutively , even during growth , as evident by the presence of CUPS like structures in 2 of 16 tomograms during growth , but are rapidly dismantled if their function is not required for the purpose of unconventional secretion ( Figure 7 ) . Upon starvation their consumption to the ER is blocked and they mature - by ultimately being encased in a membrane by Snf7 dependent reaction to form stable CUPS ( Figures 7–9 ) . We have been unable to detect Acb1 in the immature Grh1 containing vesicles and tubules . However , once they are encased in a saccule by a process that correlates with co-localization of Grh1 and Snf7 , the ensuing compartment is found to contain Acb1 ( Figure 8 ) . We do not know the origin of the saccular membrane , but our findings strongly suggest that acquisition of Acb1 into the secretory pathway is post-production of Grh1 containing vesicular membranes . One possibility is that Acb1 is attached to the saccule that engulfs Grh1 containing membranes and this casing of saccule is sealed by an Snf7 dependent reaction to produce a compartment whose contents are separated from the cytoplasm . This doubled membrane bounded multivesicular / multilamellar compartment is clearly morphologically distinct from an MVB , that is composed of a single membranes containing uniformed sized vesicles . Moreover its formation is different from the mechanism by which cells produce an MVB . The role of Vps4 in the biogenesis of these compartments is the most significant difference . While Vps4 is required for MVB biogenesis , it is not involved in mature CUPS formation . We call this compartment stable CUPS to distinguish them from the collection of Grh1 containing vesicles and tubules that are observed at 30 min post starvation ( Figure 9 ) . The stable CUPS containing sealed Acb1 are then the source for the release of Acb1 into the exterior of the cell . Cells exposed to alkaline pH activate the Rim signaling pathway , which also depends on a subset of ESCRT proteins . Briefly , the arrestin-like protein Rim8 is recruited to plasma membrane via the plasma membrane sensor Rim21 ( Obara et al . , 2012; Obara and Kihara , 2014; Herrador et al . , 2010 ) . Rim8 directly recruits Vps23 of ESCRT-I , bypassing ESCRT-0 ( Herrador et al . , 2010 ) . This ultimately leads to ESCRT-III dependent recruitment of a protease ( Rim13 ) that cleaves a transcription factor ( Rim 101 ) , which then regulates the genes that provide protection from the elevated pH ( Lamb et al . , 2001; Weiss et al . , 2009 ) . This process is independent of ESCRT-0 but always active in the absence of Vps2 , Vps24 or Vps4 , which cause constitutive recruitment and cleavage of Rim101 ( Hayashi et al . , 2005 ) . Acb1 secretion and CUPS formation is also independent of ESCRT-0 , but do require Vps2/24 proteins . Although Vps4 is not required for Acb1 secretion or CUPS formation , the loss of Vps4 does not induce Acb1 secretion or CUPS formation in growing conditions . We did , however , observe modest acceleration of Acb1 secretion and Snf7 recruitment to CUPS upon starvation in vps4Δ cells . It is therefore unclear whether the conditions that trigger Acb1 secretion are similar to cellular response to pH alterations . Moreover , we have tested loss of Rim8 , Rim21 and Rim101 directly for Acb1 secretion and unfortunately these gene deletions caused extensive cell lysis thereby precluding their evaluation in Acb1 secretion by our cell wall extraction assay ( data not shown ) . Clearly , the mechanism by which cells sense and signal in response to nutrient starvation to generate CUPS and promote Acb1 secretion is an interesting challenge and might well share some genetic requirements of the other well known signaling pathways such as the high pH triggered Rim signaling pathway . It has recently been shown that siRNA dependent knockdown of GRASP55 , GRAPS65 , Hrs ( Vps27 , ESCRT-0 ) and Tsg101 ( Vps23 , ESCRT-I ) affected unconventional IL-1ß secretion ( Zhang et al . , 2015 ) . This confirms some of the data previously reported for Acb1 secretion ( Duran et al . , 2010; Manjithaya et al . , 2010 ) . Based on their data , Schekman and colleagues have suggested that IL-1ß is translocated into a vesicle , which grows into a phagophore and is converted into an autophagosome that contains IL1ß between the inner and the outer membrane . The autophagosome then either fuses directly with the cell surface or first with an endosome/MVB to generate an amphisome that later fuses with the plasma membrane to release soluble IL-1ß ( Zhang et al . , 2015 ) . Although , many mechanistic questions remained unsolved: the authors have not reported the requirement of other ESCRTS , the genes required for complete formation of an autophagosome , the proposed translocon for transferring IL-1ß across the membrane , the involvement of endosomes , amphisomes , or the fusion of membranes with the cell surface . Regardless , how does this pathway relate to the secretion of Acb1 in yeast ? Our findings indicate that Grh1 containing membranes , after their production by COPI and COPII independent reaction ( Cruz-Garcia et al . , 2014 ) are first collected into a focus and then appear more tubulo vesicular ( Figure 9 – immature CUPS ) . These membranes then grow and appear to be more connected . The phagophores implicated by us previously ( Duran et al . , 2010; Bruns et al . , 2011 ) , and described recently by Schekman and colleagues , could represent these immature CUPS structures . Our data reveal that Grh1 containing membranes ( immature CUPS ) are encased in a saccule by Snf7 dependent process . This step is required for full maturation of CUPS . We call this compartment stable CUPS that contain Acb1 ( Figure 9 ) . We speculate that this compartment might be the functional ortholog of the amphisomes proposed by Schekman and colleagues for IL-1ß secretion . Built into this common theme are also other features with respect to specific cargoes that might utilize different chaperones and means for their capture into the starting stage ( the vesicles ) , which could explain the utility of ESCRT-0 for IL1ß , but not Acb1 secretion . It is important to note that the compartments of this pathway do not need to be identical in organization and shape: the yeast Golgi membranes do not appear anything like the Golgi stacks of higher eukaryotes . So the overall - evolving - pathway of unconventional protein secretion is likely more similar than presently appreciated . Importantly , this pathway does not use the conventional Vps4 and MVB mediated release of exosomes into the extracellular space . Yeast cells were grown in synthetic complete ( SC ) media ( 0 . 67% yeast nitrogen base without amino acids , 2% glucose supplemented with amino acid drop-out mix ( SIGMA-Aldrich , St . Louis , MO , USA ) . All strains are derived from the BY4741 background ( MATa his3∆1 leu2∆0 met15∆0 ura3∆0 ) . Deletion strains were from the EUROSCARF collection with individual genes replaced by KanMx4 . Strains expressing C-terminally 2xyeGFP- and/or 2xyomCherry-tagged Grh1 were constructed by a PCR-based targeted homologous recombination and have been described previously ( Cruz-Garcia et al . , 2014 ) . The Snf7-RFP strain was a kind gift from Dr . Erin O’Shea ( Harvard University ) . All subsequent strains were generated by mating and sporulation , followed by selection of clones with appropriate markers , and confirmation of haploidy . The pRS415-GFP-Cps1 under the control of its own was provided by David Teis . The pRS416-Snf7-GFP expressed from its own promoter was kindly provided by Dr . Scott Emr ( Cornell University , Ithaca , USA ) . The Snf7-RFP expression construct was generated by PCR amplification of Snf7 , with its own promoter , mRFP tag and ADH1 terminator sequences from genomic DNA derived from the Snf7-RFP strain and ultimately cloned into pRS416 . Similarly , to generate Snf7-L121D construct , Snf7 with its own promoter and terminator was first amplified from wild type genomic DNA and cloned into pRS416 . The latter was mutated using the Gibson assembly method to change codon 121 from CTT to GAT ( Gibson et al . , 2009 ) . SnapGene software ( from GSL Biotech , Chicago , IL; available at www . snapgene . com ) was used for molecular cloning design . Yeast cells were grown to mid-logarithmic phase by at a density of 0 . 003–0 . 006 OD600 in SC medium at 25°C . The following day , when cells had reached OD600 of 0 . 4–0 . 7 equal numbers of cells ( 15 OD600 units ) were harvested , washed twice in sterile water , resuspended in 1 . 5 mL of 2% potassium acetate and incubated for 2 hr . Concomitant to this , growing cells were diluted in SC medium , continued growing in logarithmic phase and 15 OD600 units were harvested as before . The cell wall extraction buffer ( 100 mM Tris-HCl , pH 9 . 4 , 2% sorbitol ) was always prepared fresh before use and kept on ice . To ensure no loss of cells and to avoid cell contamination in the extracted buffer , 2 mL tubes were siliconized with Sigmacote ( SIGMA-Aldrich ) prior to collection . Cells were harvested by centrifugation at 3000xg for 3 min at 4°C , medium or potassium acetate was removed and 1 . 5 mL of cold extraction buffer was added . Cells were resuspended gently by inversion and incubated on ice for 10 min , after which they were centrifuged as before , 3000xg for 3 min at 4°C , and 1 . 3 mL of extraction buffer was removed to ensure no cell contamination . The remaining buffer was removed and the cells were resuspended in 0 . 75 mL of cold TE buffer ( Tris-HCl , pH 7 . 5 , EDTA ) with protease inhibitors ( aprotinin , pepstatin , leupeptin [SIGMA-Aldrich] ) and 10 μL was boiled directly in 90 μL of 2x sample buffer ( lysate ) . To the extracted protein fraction , 30 μg of BSA ( bovine serum albumin [SIGMA-Aldrich] ) carrier protein was added and 0 . 2 mL of 100% Trichloroacetic acid ( SIGMA-Aldrich ) . Proteins were precipitated on ice for 1 hr , centrifuged 16 , 000xg for 30 min and boiled in 45 μL 2x sample buffer . For detection , proteins ( 10 μL each of lysate or wall fractions ) were separated in a 16 . 5% Tris-tricine peptide gel ( Bio-Rad ) allowing separation of Cof1 ( 15 . kDa ) from Acb1 ( 10 . kDa ) , before transfer to nitrocellulose . Rabbit anti-Cof1 antibody was a generous gift from Dr . John Cooper ( Washington University in St . Louis , USA ) . Rabbit anti-Bgl2 was a gift from Dr . Randy Schekman ( UC Berkeley , CA , USA ) . Rabbit anti-Acb1 antibody was generated by inoculating rabbits with recombinant , untagged Acb1 , purified from bacteria . Specificity of the serum was confirmed by testing lysates prepared from acb1Δ cells . After incubation in the appropriate medium cells were harvested by centrifugation at 3000 g for 3 min , resuspended in a small volume of the corresponding medium , spotted on a microscope slide , and imaged live with a DMI6000 B microscope ( Leica microsystems , Wetzlar , Germany ) equipped with a DFC 360FX camera ( Leica microsystems ) using an HCX Plan Apochromat 100x 1 . 4 NA objective . Images were acquired using LAS AF software ( Leica microsystems ) and processing was performed with ImageJ 1 . 47n software . After incubation in starvation medium for 20 min , ~0 . 05 OD600 nm of cells were plated in starvation medium on Concanavalin A–coated ( SIGMA-Aldrich ) Lab-Tek chambers ( Thermo Fisher Scientific , Waltham , MA , USA ) and were allowed to settle for 20 min at 25°C . Due to issues of bleaching , fields of cells were continuously imaged up to 10 min throughout starvation . Whole cell Z stacks with a step size of 0 . 3 μm were continuously acquired ( 10 s frames ) using a spinning-disk confocal microscope ( Revolution XD; Andor Technology , Belfast , United Kingdom ) with a Plan Apochromat 100× 1 . 45 NA objective lens equipped with a dual-mode electron-modifying charge-coupled device camera ( iXon 897 E; Andor Technology ) and controlled by the iQ Live Cell Imaging software ( Andor Technology ) . Processing was performed with ImageJ 1 . 47n software . For CLEM analysis , yeast cells were filtered into a paste and cryoimmobilised with HPM 010 high pressure freezing machine ( Bal-Tec , Los Angeles , CA , USA ) . Freeze-substitution with 0 . 1% Uranyl Acetate in acetone and embedding in Lowicryl resin was performed in an AFS2 machine ( Leica microsystems ) as described in ( Kukulski et al . , 2011 ) . 300 nm thick sections were cut from the polymerized resin block and picked up on carbon coated mesh grids . 50 nm TetraSpeck fluorescent microspheres ( Lifetechnologies , Carlsbad , CA , USA ) were adsorbed to the grid for the subsequent fiducial-based correlation between light and electron microscopy images . The fluorescence microscopy ( FM ) imaging of the sections was carried out as previously described ( Kukulski et al . , 2011; Avinoam et al . , 2015 ) using a widefield fluorescence microscope ( Olympus IX81 ) equipped with an Olympus PlanApo 100X 1 . 45 NA oil immersion objective . Images were collected with mcherry-specific settings as well as in the green channel ( Tetraspecks could be distinguished from the mcherry-specific signal by their fluorescence in both channels ) . Grids were then incubated with 10 nm protein A-coupled gold beads and stained with Uranyl Acetate and Reynolds lead citrate . Tilt series were then acquired with a FEI Tecnai F30 electron microscope . Lower magnification series ( 3900X ) for tetraspecs-based correlation as well as high magnification series ( 20000X ) were acquired . Tomograms were then reconstructed using the IMOD software package ( Kremer et al . , 1996 ) . Areas of interest were identified with a fiducial-based correlation performed as described previously using in-house written MATLAB scripts ( Kukulski et al . , 2011; 2012 ) . Briefly , the position of Tetraspeck microspheres was manually assigned in both the FM and low magnification EM images . The coordinate of fiducials pairs in the two imaging modalities were used to calculate a linear transformation , which allowed to map the coordinates of the fluorescent spot of interest on the electron tomogram . The gold beads were then used to calculate the transformation between low and high magnification tomograms , therefore allowing the overlay of the FM image with the high resolution tomogram . Once the Grh1-2xmCherry fluorescent signal was mapped on the high magnification tomogram , the membranes in the area of interest were manually segmented with IMOD and a 3D model was reconstructed . Samples were processed essentially as described earlier ( Peters et al . , 2006 ) . In brief , yeast cells expressing Grh1-2xGFP cultured in starvation conditions ( see above ) were fixed with 4% paraformaldehyde in PBS for 2 hr . The cells were then washed with PBS/0 . 02 M glycine , and resuspended in 12% gelatin in PBS , and then embedded in the same solution . The embedded cells were cut in 1 mm blocks and infiltrated with 2 . 3 M sucrose at 4°C , mounted on aluminum pins , and frozen in liquid nitrogen . The samples were then sectioned and the ultrathin cryosections were picked up in a mixture of 50% sucrose and 50% methylcellulose and incubated with antibodies to Acb1 and GFP ( to monitor Grh1 ) followed by protein A gold ( 15 nm and 10 nm respectively ) in this sequential order . The labelled sections were then imaged in FEI Tecnai-12 electron microscope .	Cells produce thousands of different proteins with a variety of different roles in the body . Some proteins , for example the hormone insulin , perform roles outside of the cell and are released from cells in a process that has several stages . In the first step , newly-made insulin and many other “secretory” proteins enter a compartment called the endoplasmic reticulum . Once inside , these proteins can then be loaded into other compartments and transported to the edge of the cell . There is another class of secretory proteins that are released from the cell without first entering the endoplasmic reticulum , in a process termed “unconventional protein secretion” . A protein called Acb1 is released from yeast cells in this manner . Previous research identified a compartment that might be involved in this process . However , it is not clear how this compartment ( named CUPS ) forms , and what role it plays in unconventional protein secretion . Curwin et al . investigated how CUPS form in yeast cells , and whether the compartment contains Acb1 proteins . The experiments reveal that after CUPS form they need to mature into a form that is involved in the release of Acb1 proteins from the cell . This maturation process involves some , but not all , of the same genes as those involved in producing another type of compartment in cells called a multivesicular body . Acb1 is only found in the mature CUPS and multivesicular bodies are not involved in the release of this protein from the cell . Curwin et al . ’s findings shed some light on how Acb1 and other secretory proteins can be released from cells without involving the endoplasmic reticulum . Future challenges are to reveal how CUPS capture cargo and find out how Acb1 leaves the CUPS to exit the cell .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "cell", "biology" ]	2016	ESCRT-III drives the final stages of CUPS maturation for unconventional protein secretion
The rat parahippocampal region ( PHR ) and retrosplenial cortex ( RSC ) are cortical areas important for spatial cognition . In PHR , head-direction cells are present before eye-opening , earliest detected in postnatal day ( P ) 11 animals . Border cells have been recorded around eye-opening ( P16 ) , while grid cells do not obtain adult-like features until the fourth postnatal week . In view of these developmental time-lines , we aimed to explore when afferents originating in RSC arrive in PHR . To this end , we injected rats aged P0-P28 with anterograde tracers into RSC . First , we characterized the organization of RSC-PHR projections in postnatal rats and compared these results with data obtained in the adult . Second , we described the morphological development of axonal plexus in PHR . We conclude that the first arriving RSC-axons in PHR , present from P1 onwards , already show a topographical organization similar to that seen in adults , although the labeled plexus does not obtain adult-like densities until P12 . The parahippocampal region ( PHR ) is important for learning and memory . It consists of two functionally different networks , one of which , involved in spatial functions , comprises the presubiculum ( PrS ) , parasubiculum ( PaS ) , medial entorhinal cortex ( MEC ) and postrhinal cortex ( POR ) . The other important for object representation and processing of contextual information , comprising lateral entorhinal cortex ( LEC ) and perirhinal cortex ( PER; Eichenbaum et al . , 2007; Canto et al . , 2008; Ranganath and Ritchey , 2012; Knierim et al . , 2014; Witter et al . , 2014 ) . PrS , PaS , and MEC harbor cells representing an animal’s position ( grid cells ) , direction ( head direction cells ) , borders in the environment ( border cells ) and speed of the animal ( speed cells; Fyhn et al . , 2004; Solstad et al . , 2008; Boccara et al . , 2010; Kropff et al . , 2015 ) . Even though a complete mechanistic understanding on how these spatial codes emerge is still lacking , it is believed that both intrinsic connectivity and extrinsic afferents are necessary to produce the receptive fields observed ( Brandon et al . , 2011; Koenig et al . , 2011; Bonnevie et al . , 2013; Couey et al . , 2013; Newman et al . , 2014 ) . One approach applied to identify the critical elements underlying the functioning of these cell types , has been to study the developmental aspects of PHR networks . The different spatially modulated neuron types in PHR emerge at different periods during development . Border cells and head-direction cells can both be observed at the end of the second- and beginning of the third postnatal week ( Bjerknes et al . , 2014; Bjerknes et al . , 2015; Tan et al . , 2015 ) . In contrast , adult-like grid cells in layer II of MEC first appear during the fourth postnatal week ( Langston et al . , 2010; Wills et al . , 2010 ) . The early presence of head-direction cells is apparently paralleled by a similarly early developed shared connectivity ( Bjerknes et al . , 2015 ) , while the late development of grid cells is paralleled by a corresponding late development of the relevant intrinsic connectivity in MEC ( Langston et al . , 2010; Couey et al . , 2013 ) . Regarding main inputs , sparse connections between the hippocampal formation ( HF ) and PHR are present from birth , reaching adult-like morphological features during the second postnatal week ( Deng et al . , 2007; O'Reilly et al . , 2013; O'Reilly et al . , 2015 ) . A similar developmental timeline has also been reported for functional connections from PrS/PaS to MEC ( Canto et al . , 2011 ) . However , the timeline of development of cortical afferents to HF and PHR is still unknown . One of the most prominent cortical inputs to PHR originates in the retrosplenial cortex ( RSC ) and spatially modulated cells have been found also in the latter cortical domain ( Cho and Sharp , 2001; Sugar et al . , 2011; Alexander and Nitz , 2015 ) . Lesions of RSC result in impairments in navigational tasks ( Vann et al . , 2009 ) . In addition , RSC in rodents is necessary for fear conditioning , both when context or complex multimodal stimuli are used as conditional stimuli ( Keene and Bucci , 2008a; 2008b; Corcoran et al . , 2011; Cowansage et al . , 2014; Robinson et al . , 2014 ) and in rabbits RSC neurons are responsive to auditory cues when used as a CS in a memory task ( Gabriel et al . , 1991 ) , suggesting that RSC has a general role in memory processes . The effect of lesioning RSC on navigation and memory performance is surprisingly similar to that seen after lesions that inflict the HF-PHR . It has thus been postulated that interactions between RSC and HF-PHR are crucial for spatial processing . In the adult , RSC projections target preferentially POR , PrS , PaS , and MEC ( Jones and Witter , 2007; Sugar et al . , 2011; Kononenko and Witter , 2012; Czajkowski et al . , 2013 ) . In this paper , we aimed to ascertain the relevance of the RSC-PHR projection for the development of the functionally different neuron types in PHR . We hypothesized that if inputs from RSC are important for the development of head-direction- and/or border cells , these inputs should be present before eye-opening . Alternatively , if inputs from RSC are only important for the formation of stable grid cells , these inputs might develop after eye-opening , likely reaching adult-like morphology during the third and fourth week . We injected RSC of rats at different postnatal ages , ranging from postnatal day 0 to 28 ( P0-P28 ) , with anterograde tracers . Using retrograde tracing , we identified RSC neurons originating the developing PHR projections . The anterograde experimental material was used to analyze the organization of RSC-PHR projections in postnatal rats and to compare these results with data previously obtained in the adult . We further analyzed the development of axon morphology and densities of axonal plexus in PHR . Several nomenclatures have been used to describe subdivisions of RSC . These nomenclatures mainly follow the same cytoarchitectonic- and histochemical criteria and are therefore directly comparable . For a summary and direct comparison of the different nomenclatures , we refer to a recent review on the RSC-HF-PHR connectivity ( Sugar et al . , 2011 ) . In the current manuscript , we chose to use the nomenclature of Vogt ( 2004 ) . We defined the border between area ( A ) 29 and A30 as the area where layer II changes from being very condensed in A29 , to less condensed in A30 . The rostral border of RSC , towards the anterior cingulate cortex ( ACC ) was defined as the area where layer II/III widens and where layer IV shifts from being clearly demarcated in RSC towards being more diffusely organized in ACC . The ventral border of RSC with PHR , more specifically with PrS , was defined by the appearance of a cell free lamina dissecans in PrS , not present in RSC . In adult rats , A29 can be further subdivided into three different cytoarchitectonic subdivisions; A29a , b and c , which are involved in different cognitive functions ( van Groen et al . , 2004 ) . However , in the immature cortex , these cytoarchitectonic areas are not apparent . We therefore chose to define a continuous measure of the dorsoventral positions within RSC ( see methods for details ) . This measure is indirectly related to the classical cytoarchitectonic subdivision since the classical borders of A29a , b and c follow approximately our continuous definition of the dorsoventral axis of RSC . To investigate the development of RSC projection patterns in PHR we aimed to inject anterograde tracers in different locations within RSC of differently aged pups . Of the 82 animals used in this study , 20 animals either did not survive surgery or no injection sites were observed in RSC . In the remaining 62 animals we obtained 113 injections in RSC . Eight of these injections only involved layers I and/or layers II-III and did not result in any labeled fibers in HF-PHR . These experiments were therefore excluded from further analyses . The remaining 105 injections all covered at least parts of layer V of RSC and involved different parts of RSC . We obtained one ( n=21 ) , two ( n=31 ) , three ( n=6 ) or four ( n=1 ) ipsilateral injections in RSC of each brain ( Figure 1 ) . In our analyses , we regarded each of these injections as independent experiments . Most of the experiments ( n=93 ) were performed in animals aged P15 or younger since other comparable corticocortical projections are developed before eye-opening and the functional cell types are present in HF-PHR at this age ( Langston et al . , 2010; O'Reilly et al . , 2013; Bjerknes et al . , 2014; O'Reilly et al . , 2014 ) . However , we also obtained injections in older pups ( n=12 ) with a maximum age of P28 . 10 . 7554/eLife . 13925 . 003Figure 1 . Location of the center of injections . ( A ) The location of the center of each injection was normalized to a standard 3D atlas of the rat brain ( Waxholm space; Papp et al . , 2014; See video 1 and 2 ) . Lateral ( A1 ) , dorsal ( A2 ) and para-caudal view ( A3 ) of the 3D atlas brain with the center of each injection ( colored spheres ) . The injections are color-coded according to age ( see color code bottom left ) . Injections were performed in either the left or right hemisphere but to ease visualization , all injections were plotted in the right hemisphere . Dentate gyrus ( DG ) , CA3-1 , subiculum ( SUB ) , fasciola cinereum ( FC ) , pre- and parasubiculum ( PrS and PaS ) , medial- and lateral entorhinal cortex ( MEC and LEC ) , A35-36 and postrhinal cortex ( POR ) are color coded , while the rest of the brain is colored green . ( B ) Midsagital view of the center of the injections projected to the pial surface . The laminar position is disregarded to allow the injections to be plotted in 2D . Light blue , light red and grey line depict respectively the dorsal border of A30 , the border between A29 and A30 and the ventral border of A29 . Injections are color coded according to age . Triangle explained in C . ( C ) One example , as shown in B , of the algorithm used for calculating normalized 2D coordinates of the injections . The pial surface area of A29 and A30 was divided into triangles ( grey area ) . The shortest vector ( black line ) between the injection ( red sphere ) and the cortical surface was calculated . Thereafter , we calculated the coordinate of the intersection of the vector and the plane within the triangle ( yellow dot ) which represented the 'transposed' location of the injection . The normalized dorsoventral coordinate of each injection was defined by calculating the shortest vector from the transposed injection to dorsal ( ddorsal ) and ventral border of A29 or A30 ( dventral ) . The normalized rostrocaudal coordinate was obtained by first calculating a line along the rostrocaudal extend of A29 and A30 , positioned equally distant from the dorsal and ventral borders ( magenta line ) . Next , we calculated the shortest vector between the injection and this line ( red line ) and found the intersection between the two . The rostrocaudal coordinate was obtained by calculating the cumulative distance from the cross section to the rostral ( drostral ) and caudal end of RSC ( dcaudal ) . ( D ) Normalized flatmap of the injections . The 3D RSC is converted to a 2D normalized flatmap to obtain relative rostrocaudal and dorsoventral positions of the injections . The figure is oriented with rostral RSC ( left ) , caudal RSC ( right ) , dorsal RSC ( top ) ventral RSC ( bottom ) to each of the sides of the rectangle . Grey line depicts the border between A29 and A30 in RSC . In all figures , each injection is color coded according to the bottom left color scheme; light grey colored injections represent injections in pups aged P1 , green colored injections represent injections in pups aged P8 while cyan colored injections represent injections in pups aged close to P15 . Grey injections represent injections in pups older than P15 . DOI: http://dx . doi . org/10 . 7554/eLife . 13925 . 00310 . 7554/eLife . 13925 . 004Video 1 . Representation of all injection sites in RSC represented in a 3D rendering of the rat brain ( Waxholm space; Papp et al . , 2014; 2015 ) . The injection sites are color coded for age ( for code see Figure 1 ) DOI: http://dx . doi . org/10 . 7554/eLife . 13925 . 00410 . 7554/eLife . 13925 . 005Video 2 . Representation of all injection sites in RSC represented in a 3D rendering of the rat brain ( Waxholm space; Papp et al . , 2014; 2015 ) . The injection sites are color coded for age ( for code see Figure 1 ) DOI: http://dx . doi . org/10 . 7554/eLife . 13925 . 005 To compare the location of injections in brains of different ages , we age-normalized the position of the injections by the use of a 3D-atlas brain ( see methods and Figure 1A , Video 1 and 2 ) . This was achieved by identifying atlas-sections containing landmarks and cytoarchitectonic borders present in the histological section containing the center of each injection . In the atlas-section , we recorded the coordinate of the center of each injection . Since the caudal RSC cortex is curved both along the dorsoventral and rostrocaudal axis we divided the pial surface of RSC in the atlas brain into triangles and used these triangles to calculate normalized rostrocaudal and dorsoventral coordinates of the each injection ( Figure 1B and C ) . For visualization , we plotted the normalized injection coordinates into a schematic representation of RSC ( Figure 1D ) . We analyzed the anterograde labeling resulting from all 105 injections . In some ( n=44 ) , we observed a few retrogradely labeled neurons within the dorsal half of subiculum ( SUB ) . This potentially may lead to false positive labeling , which will be addressed in the detailed descriptions below . In accordance with previous studies in adults , we observed anterogradely labeled fibers in the striatum , anterior nuclei of the thalamus , anterior cingulate cortex , parietal cortex and visual cortices and in the brainstem ( van Groen and Wyss , 1990; 1992; 2003; Jones et al . , 2005 ) . However , a detailed assessment of these projections is outside the scope of this paper and here we will detail the projections to PHR . We observed anterogradely labeled fibers in layers I , III and V-VI of PrS ( in all of the experiments included in the analyses , n=105 ) , layers V-VI of PaS ( n=90 ) , layers V-VI of MEC ( n=90 ) , layers V-VI of medial LEC ( n=14 ) and in all layers of posterior POR ( n=75 ) . In some cases , we also observed single fibers in layers I and/or III of MEC ( n=34 ) . Additionally , we observed few fibers ( if present typically one or two fibers in an experiment ) in the dorsal half of SUB ( n=41 ) and/or in the dorsal CA fields ( n=16 ) . In contrast to previously published data ( Burwell and Amaral , 1998b ) , we did not observe any labeled fibers in PER in any of our experiments . This includes the oldest aged pups ( P27-28 ) and the adult cases , suggesting that this lack of perirhinal projections is not a developmental feature . It is thus obvious that all experiments shared RSC-PHR projection patterns , but there were also marked differences from case to case depending on the location of injection within RSC . This will be systematically described in the next sections . Injections in the rostral half of A30 ( n=23 ) all resulted in comparable patterns of labeling in PHR . Labeled fibers were present in layers I , III and V-VI of the dorsal one-third of PrS . In layers I and III of PrS , labeled fibers were predominantly located in distal PrS and the densest plexus was generally observed in the dorsal extreme of PrS . Moderate numbers of labeled fibers were present in layers V-VI of dorsal PaS and layers V-VI of dorsomedial MEC . Among the injections in the rostral half of A30 , the most caudal ones usually resulted in more widespread labeling in MEC . In none of the cases did we observe labeled fibers in the ventral two-thirds of PHR . We observed single labeled fibers in layer III of MEC in some of the experiments ( n=3 ) , in POR ( n=16 ) and in HF ( n=5 ) . In a representative animal ( 18433; P13 ) , DA-A488 was injected in layers I-V in rostral A30 ( Figure 2A–D , magenta ) . From the injection site , fibers continued caudally in layer VI of RSC and in the cingular bundle towards PHR . At the dorsal pole of PHR , fibers entered into the superficial layers and branched extensively in layer I and superficial layer III of intermediate proximodistal parts of the dorsal pole of PrS . At this dorsal level , labeled fibers were also present in layers V-VI of PrS . Single fibers continued into deep layers of dorsal PaS and deep layers of posterior POR and deep layers of dorsomedial MEC . No fibers were observed in more ventral levels of PHR , such that the ventral 75% of PHR did not show any labeled fibers . 10 . 7554/eLife . 13925 . 006Figure 2 . Representative examples of injections in RSC . ( A ) Normalized flatmap ( see Figure 1D ) of the locations of the injections in RSC , shown in B and E . Injections are located in the rostral A30 ( magenta ) , intermediate rostral A30 ( green ) , intermediate caudal A30 ( cyan ) and the intermediate rostral quarter of A29 ( yellow ) . ( B ) Horizontally cut and Nissl stained section at the level of the injections overlaid with a neighboring fluorescent section containing the center of an injection in rostral A30 ( magenta ) and intermediate-rostral A30 ( green ) within the same animal . Grey line depicts delineation of A30 . ( C ) The projections after the two injections shown in B were traced and represented in a dorsoventral series of drawings of horizontal sections through the PHR . After injections in rostral A30 ( magenta ) labeled fibers were mostly observed in the dorsal PrS layers I and III ( C1 , D1 ) . After injections in intermediate-caudal A30 ( green ) the densest plexus was located more ventrally in PHR and in addition to labeled fibers in layers I , III and V-VI of PrS , labeled fibers also extended into layers V-VI of PaS , POR and MEC ( C2-3 and D3 ) . Numbers above sections indicate the dorsoventral position of the section relative to the total dorsoventral extent of PHR . The yellow boxes in C1 and C3 indicate the position of high power digital images obtained from the actual sections ( D1 and D3 ) . Grey lines depict borders between the HF-PHR subdivisions , the border between cortex and white matter and lamina dissecans . ( D ) High power images of plexus depicted in the sections shown in C and E . Roman numbers indicate cortical layers . Grey lines depict borders between layers . ( D1 ) Labeled fibers in superficial layer III of PrS after injections in rostral A30 ( magenta ) . Additionally a few fibers are seen originating in the intermediate-caudal quarter of RSC ( green ) . ( D2 ) Labeled fibers in proximal PrS deep layer III and layers V-VI after injection in intermediate-rostral A29 ( yellow ) and labeled fibers in distal PrS deep layer III and layers V-VI after injection in intermediate-caudal A30 ( cyan ) . ( D3 ) Labeled fibers in layers I and III after injection in intermediate-caudal A30 ( green ) . No fibers originating in the rostral A30 were observed . ( D4 ) After injection in intermediate-caudal A30 labeled fibers were observed in medial MEC ( cyan ) , while after injection in intermediate-rostral A29 labeled fibers were observed in lateral MEC ( yellow ) . ( E ) Top: the projections after two injections ( bottom ) were traced and represented in a dorsoventral series of drawings of horizontal sections through the PHR . After injections in intermediate-caudal A30 ( cyan ) labeled fibers were observed in distal PrS dorsally ( E1-3 ) . At more ventral levels fibers also extended into deep layers of PaS and medial MEC ( E3-4 ) . After injections in intermediate-rostral A29 ( yellow ) the densest plexus was located in proximal parts of PrS dorsally ( E1-2 ) . At more ventral levels the plexus in PrS layers I and III disappeared while in the deep layers the plexus shifted to lateral parts of EC at successively more ventral levels ( E3-4 ) . Numbers below sections indicate the dorsoventral position of the section relative to the total dorsoventral extent of PHR . The magenta boxes in E2 and E3 indicate the position of high power digital images obtained from the actual sections ( D2 and D4 ) . Grey lines depict borders between the HF-PHR subdivisions , the border between cortex and white matter and lamina dissecans . Bottom: Horizontally cut and Nissl stained sections at the level of the injection overlaid with neighboring fluorescent sections containing the center of an injection in intermediate-caudal A30 ( cyan ) and intermediate-rostral A29 ( yellow ) within the same animal . Gray line depicts delineation of A29 and A30 . Scale bars equal 100 μm ( high power images ) and 1000 µm ( low power images ) . DOI: http://dx . doi . org/10 . 7554/eLife . 13925 . 006 Injections in rostral A30 thus resulted in labeling in the dorsal third of PHR , particularly in posterior POR , distal PrS , the complete transverse extent of PaS , and medial parts of MEC . We analyzed 32 injections in caudal A30 . Overall , injections in caudal A30 resulted in comparable projection patterns . Labeled axons were observed in layers I , III and V-VI in distal parts of PrS , layers V-VI of PaS and in layers V-VI medially in MEC . Compared to the labelling observed after rostral injections , the labeling resulting from caudal injections extended more ventrally in PrS , PaS and MEC ( compare magenta and green topography in Figure 2C ) . The total extend of all plexus covered approximately the dorsal half of PrS and PaS and the ventromedial part of MEC . The area receiving the densest projections was also shifted more ventrally . The maximum density of the plexus in PrS was usually located in the dorsoventral middle of its distal part . Only in one case , labeled fibers were present in the ventral one-third of PHR . At all dorsoventral levels , labeling covered the distal part of PrS , in layers I , III and V-VI , the proximodistal extent of PaS , and medial parts of MEC . After injections in caudal A30 , we observed single labeled fibers in layer III of MEC in some experiments ( n=12 ) , in POR ( n=29 ) , and in HF ( n=15 ) . In a representative animal ( 18453; P13 ) , BDA was injected in layers I-V of intermediate-caudal A30 ( Figure 2A and D-E , cyan ) . From the injection site labeled fibers continued caudally and ventrally in the cortex . Arriving in PHR , fibers continued ventrally in layer I and the lamina dissecans of PrS as well as in deep white matter . At the dorsal pole of PrS , labeled fibers occasionally entered into layer III . A dense plexus was labeled in layers V-VI of the dorsal one-third of distal PrS , in layers V-VI of PaS and deep layers of medial POR . At more ventral levels , labeled fibers also extended into layers V-VI of MEC . In ventral parts of MEC , only a few labeled fibers were observed . No labeled fibers were observed in the most ventral one-third of PHR . Injections in caudal A30 thus resulted in a labeling pattern in PHR comparable to that seen in case of rostral A30 injections , but extending to more ventral parts of PHR . Injections in the rostral half of A29 ( n=15 ) resulted in a comparable pattern of labeling in PHR . Labeled fibers were present in the dorsal half of PrS layers I , III and V-VI , in layers V-VI of the dorsal half of PaS and in layers V-VI of MEC . One injection also resulted in labeled fibers in dorsal LEC . In PrS , the fibers tended to be located more proximally compared to the projections originating from A30 at the same rostrocaudal level ( compare cyan and yellow fibers in Figure 2E1–2 ) . Among the injections in rostral half of A29 , the most caudal injections usually resulted in more extensive labeling of axons in MEC . In those cases , the plexus in MEC was located at intermediate mediolateral levels , more lateral compared to the MEC plexus seen after A30 injections ( compare cyan and yellow fibers in Figure 2E3–4 ) . In some cases , we also observed single labeled fibers in layer III of MEC ( n=6 ) , in POR ( n=8 ) and in HF ( n=9 ) . In a representative animal ( 18453; P13 ) , DA-A488 was injected in layers I-VI of intermediate-rostral A29 ( Figure 2A and D–E , yellow ) . From the injection site , labeled fibers ran caudally and ventrally in the cingular bundle and in layer VI of RSC . At the dorsal pole of PrS , labeled axons continued ventrally in layer I , lamina dissecans , and in the deep white matter . At this level , labeled fibers entered PrS and branched in layers I , III and V-VI of proximal PrS . Single fibers extended into deep layers of PaS and POR . At approximately the dorsoventral middle of PHR , the density of labeled fibers in proximal PrS layers I and III gradually decreased , while in the deep layers , labeled fibers gradually shifted position at successively more ventral levels . More specifically , moving from dorsal to ventral levels of PHR , labeled fibers occupied proximal PrS at dorsal levels , and distal PrS , PaS , medial MEC , lateral MEC and finally LEC at successively more ventral levels . It is thus apparent that injections in rostral A29 resulted in labeling mainly in the dorsal third of PHR , including posterior POR and the transverse extent of PaS , similar to what was observed following injections in rostral A30 . The distribution in PrS and MEC however differed from that resulting from injections in rostral A30 in that labeling was present in proximal PrS and more lateral parts of MEC . Caudal A29-injections ( n=35 ) resulted in comparable projection patterns in PHR . Labeled fibers were mostly located in the dorsal half of PrS , and the dorsal two-thirds of layers V-VI of PaS and MEC . However , in some experiments labeled fibers were also observed in ventral MEC and dorsal LEC . Compared to injections in caudal A30 , injections in caudal A29 resulted in labeling also in more proximal parts of dorsal PrS . Additionally , after injections in caudal A29 , very few fibers were observed in deep layers of PaS , while a dense patch of fibers was usually observed in MEC . The densest projection to PHR usually targeted the deep layers of dorsal PrS and intermediate dorsoventral MEC . We further observed in some cases single labeled fibers in layer III of MEC ( n=13 ) , in POR ( n=22 ) and in HF ( n=13 ) . In a representative animal ( 18576; P15 ) BDA was injected in layers I-VI of caudal A29 ( Figure 3 ) . Labeled fibers left the injection site and traveled through layer VI caudally and ventrally towards the RSC-PrS border . Single fibers penetrated the lamina dissecans and branched in layers I and III of dorsal PrS . Single fibers continued into deep layers of PaS and into all layers of POR close to the PaS border . A few fibers were also present in layer III of PaS . The density of labeled fibers increased at more ventral levels and reached its maximum in the middle of the dorsal half of PHR where proximal PrS layers I-III , layers V-VI of PrS were covered by labeled fibers . A few fibers were also labeled in layers V-VI of dorsomedial MEC . Single fibers invaded SUB , layer III of PaS and layer III of MEC . We observed a change in the density and position of labeled fibers along the dorsoventral extent of dorsal PHR . The density of labeling in PHR decreased at successively more ventral levels , while the position of labeled fibers in deep layers shifted from PrS and PaS at dorsal levels towards MEC at intermediate dorsoventral levels . In the ventral third of MEC labeling was only located in lateral parts of MEC . A few labeled fibers were observed in LEC . 10 . 7554/eLife . 13925 . 007Figure 3 . Representative example of an injection in caudal A29 . ( A ) Normalized flatmap with location of an injection in caudal A29 ( magenta ) . ( B ) Horizontally cut and Nissl stained section at the level of the injection overlaid with an adjacent section containing the center of the fluorescent tracer injection in caudal A29 ( magenta ) . Grey lines depict delineation of A29 and A30 . ( C ) The projections after the injection were traced and represented in a dorsoventral series of drawings of horizontal sections through the PHR . Labeled fibers are located in more proximal parts of layers I , III and V-VI of PrS compared to after projections in A30 ( compare C1-3 and Figure 2 ) . At successively more ventral levels , fibers also extended into increasingly more lateral parts of layers V-VI of MEC compared with injections in A30 ( compare C2-4 and Figure 2 ) . Grey lines depict borders between the HF-PHR subdivisions , the border between cortex and white matter and lamina dissecans . Scale bars equal 1000 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 13925 . 007 All injections in caudal A29 thus resulted in labeling in PHR , comparable to what was seen in case of injections in rostral A29 . However , the labeling from caudal A29 extended more ventrally in PHR . The different labeling patterns in PHR , observed after injections in different parts of RSC indicate that RSC is heterogeneous with respect to the terminal distribution of PHR projections . To systematically analyze this , we measured , in each experiment , the location of the labeled plexus in PHR . Thereafter , we produced normalized flatmaps of the location of the labeled plexus in each experiment ( see methods for details ) . These flatmaps consisted of multiple bins spanning the dorsoventral and transverse axis of PHR ( Figure 4A ) . Each bin was assigned a value between 0 ( black ) and 1 ( yellow ) reflecting the density of labeled fibers in PHR ( Figure 4B ) . This approach allowed us to compare flatmaps across animals of different ages and pool flatmaps of groups of interest . For statistical analysis , we calculated the center of mass of the labeled fields and tested the relationships between this measure and the coordinates of the injection and the age of the animal . 10 . 7554/eLife . 13925 . 008Figure 4 . Standardized representation of location of labeled axons in PHR . ( A ) Extents of PrS , deep layers of PaS and deep layers of EC were measured in the horizontal plane ( 1; Nissl stained sections from case 18427 , 3; Nissl stained sections from case 18589 ) . In every section , the extents of PrS , PaS , MEC and LEC along the transverse axis were measured . Based on the measurements of all sections in all brains ( 2 and 4 ) we binned the PHR along the dorsoventral and transverse axis and made an average representation of PHR ( 5; left: layers I and III of PrS ( light blue ) , right: layers V-VI of PrS ( light blue ) , PaS ( dark blue ) , MEC ( light green ) and LEC ( dark green ) . Dorsal , ventral , proximal ( prox ) , distal ( dist ) , medial ( med ) , lateral ( lat ) indicates dorsoventral and transverse axis of the flatmaps . s# , a refers to individual measurements of layers I and III of PrS in C1 . ( B ) Locations of labeled fibers were obtained by measuring the distance between the plexus and the borders of the field in which the plexus was located ( 1; case 18427 , note that we inserted magenta labeled structures to illustrate labeled fibers and their respective positions on the flatmaps ) . The measurements were performed in every section containing a plexus . The bins between the boundaries of each plexus were given a value ranging from 1 to 3 reflecting weak to dense labeling respectively ( color coded in 2 , 1 = brown; 2 = orange; 3 = bright yellow ) . Bins outside the plexus boundaries were given the value 0 ( black , no plexus ) . In experiments in which we observed single labeled axons or a sparse plexus , we measured the distance from each labeled axon to one of the borders of the field in which the plexus was located ( 3; case 18589; repeating patterns in top sections are artefacts due to stitching of digitized images ) . We inserted magenta labeled fibers and their respective positions on the flatmaps ) . We gave each bin in the flatmap a value corresponding to the number of labeled single axons ( 4; bins not containing any labeled fibers: black; bins with the highest number of labeled axons: bright yellow ) . For all flatmaps , the centers of mass for layers I and III in PrS , layers V-VI of PrS and PaS combined and layers V-VI of MEC and LEC combined were calculated ( red dots ) . To compare groups of injections , flatmaps of individual injections were normalized to the highest valued bin , transformed to the average flatmap and added together ( 5; see methods for further details ) . ( C ) Binning of layers I and III of PrS along the dorsoventral and along the transverse axis . The example is based on animals 18427 ( left ) and 18589 ( right; both shown Figure 4A and B ) . All subdivisions are binned according to the same algorithm . First the transverse measurements ( s# , a , visualized by black lines in C1 ) and the dorsoventral measurement ( dva ) in all sections in all animals were normalized to the longest measurement of the respective subdivision in the particular animal ( red ( transverse ) and blue ( dorsoventral ) numbers in C1 ) . Next , we binned the dorsoventral axis of each PHR subdivision ( represented by turquois lines in C1 ) in the same amounts of bins as the maximum number of sections containing PHR in a single series ( smax; 23 in the example ) and calculated the transverse extends of each row of bins in each animal ( ti , a , numbers not shown ) . Thereafter , we calculated the means of the normalized transverse measurements across all animals for each bin ( t¯i , red numbers in C2 ) , and the mean across all animals of the normalized dorsoventral extend of PHR ( dv¯; blue number in C2 ) . The ratio of the mean transverse extends and the mean dorsoventral extend was calculated such that each dorsoventral level was expressed as a value relative to the dorsoventral extent of PHR ( Ti; red numbers in C3 ) . Finally , the number of bins along the transverse axis for each dorsoventral level ( Tibin ) was calculated ( C4 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 13925 . 008 We first performed a cluster analysis to investigate whether the patterns of labeling in PHR were clustered dependent on the location of the injection site or the age of the animal . Using the distribution patterns as represented in the flatmaps ( see methods for details ) , we identified five clusters . However , the injection sites in RSC associated with each of the clusters were not clustered in RSC such that several of the injections associated with one cluster of distribution patterns partly overlapped with injection sites associated with other labeling clusters in PHR . None of the clusters of labeling in PHR was associated with distinct age groups of animals . We therefore concluded that RSC does not contain regions having distinct projection patterns to PHR , but more likely has a continuous topographical organization of projections to PHR . Therefore , we next assessed whether the labeling patterns in PHC changed systematically in relationship to the rostrocaudal and/or the dorsoventral position of the injections in RSC . For the analysis of the projection patterns , we initially subdivided the injections into two groups depending on whether the injections were located in A29 ( n=50 ) or A30 ( n=55 ) . We also subdivided each of the two areas into four equally sized rostrocaudal regions; the rostral quarter of RSC ( n=16 ) , intermediate-rostral quarter of RSC ( n=22 ) , intermediate-caudal quarter of RSC ( n=48 ) and caudal quarter of RSC ( n=19 ) . To evaluate the projection patterns we analyzed the data in two different ways . First , we pooled the projection patterns of all experiments in each of the eight injection groups described above . We subsequently analyzed the 'mean' projection pattern of each group , as represented on flatmaps of PHR ( Figure 5—figure supplement 1A ) . However , the pooled flatmaps are sensitive to the number of bins covered by the labeled axons in each experiment . Experiments in young animals with single labeled fibers in PHR or experiments with small injections resulting in few labeled axons in PHR will contribute fewer 'labeled bins' to the pooled flatmaps , and the pooled flatmaps might therefore be biased towards the experiments with many bins containing labeled axons . Therefore , we also calculated , for each experiment , the center of mass of the projections to respectively layers I and III of PrS , layers V-VI of PrS and PaS combined , and layers V-VI of MEC and LEC combined . The centers of mass served as a quantifiable measure of the location of the labeled axons in PHR , which is independent of the number of fibers labeled in each experiment . After injections in the rostral part of A29 and A30 , labeled fibers were only seen in the dorsal one-third of PHR . Labeling was mainly present in superficial and deep layers of PrS with approximately equal densities , while only a moderate number of labeled fibers was observed in deep layers of PaS and dorsomedial MEC ( Figure 5—figure supplement 1A ) . Injections in the three more caudal subdivisions of A29 and A30 resulted in labeling in PHR at successively more ventral levels . Labeling was seen in layers I , III and V-VI of the dorsal half of PrS , in layers V-VI of the dorsal half of PaS and of the medial part of MEC ( Figure 5—figure supplement 1A ) . When comparing injections in A29 and A30 , we did not see a systematic relationship between the placement of the injection and the dorsoventral distribution of the labeling ( Figure 5—figure supplement 1A ) . The results thus suggest that the rostrocaudal placement of injection in RSC is related to the dorsoventral location of the labeled plexus in PrS , PaS , and MEC . Multiple regressions confirmed a relationship between the rostrocaudal placement of the injection and the dorsoventral location of the center of mass of the labeled axon terminals . More rostrally placed injections resulted in centers of mass of the labeled plexus that were located more dorsally in layers I and III of PrS ( Figure 5A and Figure 5—figure supplement 1B , β=−0 . 201 , 95% confidence interval ( CI ) [−0 . 290 , −0 . 111] , t99=−4 . 452 , p<0 . 001 ) . The dorsoventral placement of injections was not significantly related to the dorsoventral location of the centers of mass of the projections ( β=−0 . 052 , 95% CI [−0 . 144 , 0 . 040] , t99=−1 . 120 , p=0 . 265 ) . There was no significant interaction effect of rostrocaudal-by-dorsoventral placement of the injection on the resulting patterns of anterograde labeling in PHR . 10 . 7554/eLife . 13925 . 009Figure 5 . Topographical organization of projections . The position of the centers of mass of labelling in layers I and III of PrS , layers V-VI of PrS and PaS and layers V-VI of MEC and LEC is plotted . Each dot is color coded with respect to the rostrocaudal ( A ) or dorsoventral position ( B ) of the injections in RSC ( rostral half; blue , caudal half; red , A29; yellow , A30; green ) . In layers I and III of PrS more caudally placed injections result in projections located more ventral compared to rostrally placed injections ( A; p<0 . 001 ) , while ventrally placed injections result in projections located more proximal compared to dorsally placed injections ( B; p<0 . 001 ) . In layers V-VI of PrS and PaS caudally placed injections result in projections located more ventral compared with rostrally placed injections ( A; p<0 . 001 ) , while ventral and rostrally placed injections result in projections located significantly more proximal compared with dorsally and caudally placed injections ( B; ventral: p<0 . 001 , caudal: p=0 . 033 ) . In MEC and LEC layers V-VI , caudal and ventrally placed injections result in projections located significantly more ventral compared to rostral and dorsally placed injections ( A; caudal: p<0 . 001 , ventral: p=0 . 032 ) , while ventrally placed injections result in projections located significantly more lateral compared to dorsally placed injections ( B; p=0 . 020 ) . Multiple regression was used for all statistical tests ( Figure 5—source data 1–12 ) . For flatmaps of the projection patterns see Figure 5—figure supplement 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 13925 . 00910 . 7554/eLife . 13925 . 010Figure 5—source data 1 . Datapoints used in multiple regressions . DOI: http://dx . doi . org/10 . 7554/eLife . 13925 . 01010 . 7554/eLife . 13925 . 011Figure 5—source data 2 . Datapoints used in multiple regressions . DOI: http://dx . doi . org/10 . 7554/eLife . 13925 . 01110 . 7554/eLife . 13925 . 012Figure 5—source data 3 . Datapoints used in multiple regressions . DOI: http://dx . doi . org/10 . 7554/eLife . 13925 . 01210 . 7554/eLife . 13925 . 013Figure 5—source data 4 . Datapoints used in multiple regressions . DOI: http://dx . doi . org/10 . 7554/eLife . 13925 . 01310 . 7554/eLife . 13925 . 014Figure 5—source data 5 . Datapoints used in multiple regressions . DOI: http://dx . doi . org/10 . 7554/eLife . 13925 . 01410 . 7554/eLife . 13925 . 015Figure 5—source data 6 . Datapoints used in multiple regressions . DOI: http://dx . doi . org/10 . 7554/eLife . 13925 . 01510 . 7554/eLife . 13925 . 016Figure 5—source data 7 . Datapoints used in multiple regressions . DOI: http://dx . doi . org/10 . 7554/eLife . 13925 . 01610 . 7554/eLife . 13925 . 017Figure 5—source data 8 . Datapoints used in multiple regressions . DOI: http://dx . doi . org/10 . 7554/eLife . 13925 . 01710 . 7554/eLife . 13925 . 018Figure 5—source data 9 . Datapoints used in multiple regressions . DOI: http://dx . doi . org/10 . 7554/eLife . 13925 . 01810 . 7554/eLife . 13925 . 019Figure 5—source data 10 . Datapoints used in multiple regressions . DOI: http://dx . doi . org/10 . 7554/eLife . 13925 . 01910 . 7554/eLife . 13925 . 020Figure 5—source data 11 . Datapoints used in multiple regressions . DOI: http://dx . doi . org/10 . 7554/eLife . 13925 . 02010 . 7554/eLife . 13925 . 021Figure 5—source data 12 . Datapoints used in multiple regressions . DOI: http://dx . doi . org/10 . 7554/eLife . 13925 . 02110 . 7554/eLife . 13925 . 022Figure 5—figure supplement 1 . Topographical organization of RSC-PHR projections in pups . ( A ) Injections were organized into 2 x 4 groups ( A29/A30 x rostrocaudal ) based on the location of the injection and the average projection pattern for each group was calculated . Injections in rostral RSC ( left column ) resulted in labeled fibers within dorsal PrS layers I , III and V-VI , while more caudal injections ( the three right columns ) resulted in labeling in more ventral parts of PrS , PaS and MEC . Injections in A30 ( top row ) resulted in labeled fibers in distal PrS , PaS and medial MEC . In contrast , injections in A29 ( bottom row ) resulted in labeled fibers in more proximal parts of PrS and more lateral parts of MEC . Dorsal , ventral , proximal ( prox ) , distal ( dist ) , medial ( med ) , lateral ( lat ) indicates dorsoventral and transverse axis of the flatmaps . ( B-G ) Centers of mass of the labeling resulting from each injection were calculated for PrS layers I and III ( B and C ) , PrS and PaS layers V-VI ( D and E ) and MEC and LEC layers V-VI ( F and G ) . In the plots , each injection is plotted according to its location in RSC and colorcoded according to the dorsal ( blue ) , ventral ( red ) , distal/medial ( green ) and proximal/lateral ( yellow ) position of the center of mass of the projection in PHR ( see color maps in bottom right of each figure ) . DOI: http://dx . doi . org/10 . 7554/eLife . 13925 . 022 Regarding terminal labeling in layers V-VI of PrS and PaS , more rostrally placed injections in RSC resulted in terminal centers of mass located more dorsally in layers V-VI of PrS and PaS ( Figure 5A and Figure 5—figure supplement 1D , β=−0 . 133 , 95% CI [−0 . 203 , −0 . 063] , t99=−3 . 784 , p<0 . 001 ) . The dorsoventral placement of the injection showed no significant relationship with the dorsoventral location of the centers of mass of the labeling ( β=0 . 020 , 95% CI [−0 . 052 , 0 . 091] , t99=0 . 547 , p=0 . 585 ) . There was no significant interaction effect of rostrocaudal-by-dorsoventral placement of the injection . In layers V-VI of MEC and LEC , more rostrally placed injections resulted in terminal centers of mass located more dorsally in layers V-VI of MEC and LEC ( Figure 5A , Figure 5—figure supplement 1F , β=−0 . 186 , 95% CI [−0 . 269 , −0 . 104] , t85=−4 . 498 , p<0 . 001 ) . The dorsoventral placement of the injection showed a weaker , but significant relationship with the dorsoventral location of the centers of mass of the labeling since more ventrally placed injections in RSC had centers of mass located more ventral in PHR ( Figure 5B , Figure 5—figure supplement 1F , β=0 . 089 , 95% CI [0 . 008 , 0 . 171] , t85=2 . 181 , p=0 . 032 ) . There were no significant interaction effects of rostrocaudal-by-dorsoventral placement of the injection . Based on these data we conclude that the rostrocaudal position of injections in RSC determined the dorsoventral location of the labeled plexus in PrS , PaS and MEC . We subsequently analyzed whether the location of the injection site influenced the transverse position of the labeled axons in the identified PHR subdivisions . A visual analysis of the plotted centers of mass and the pooled flatmaps showed that in layers I and III of PrS the labeling was generally located distally in case of injections in A30 ( Figure 5B , Figure 5—figure supplement 1A ) . After injections in A29 , proximal PrS was also covered by labeled fibers . This suggested that the dorsoventral position of the injection in RSC is related to the transverse position of the labeled fibers in PHR . This suggestion was substantiated through multiple regression analysis , showing that following ventral injections , the centers of mass of the labeled plexus in layers I and III of PrS were significantly more proximal compared to those following injections in dorsal RSC ( Figure 5B and Figure 5—figure supplement 1C; β=0 . 392 , 95% CI [0 . 267 , 0 . 518] , t99=6 . 200 , p<0 . 001 ) . The dorsoventral placement of the injections showed no significant relationship with the dorsoventral location of the centers of mass of the labeling ( β=0 . 117 , 95% CI [-0 . 005 , 0 . 239] , t99=1 . 902 , p=0 . 060 ) . There was no significant interaction effect of rostrocaudal-by-dorsoventral placement of the injection . In layers V-VI of PrS and PaS , the centers of mass were located significantly more proximal after injections in ventral RSC compared to injections in dorsal ( Figure 5B and Figure 5—figure supplement 1E; β=0 . 292 , 95% CI [0 . 192 , 0 . 392] , t99=5 . 780 , p<0 . 001 ) . The rostrocaudal placement of the injection did show a weaker , but significant relationship with the proximodistal location of the labeled fibers ( Figure 5A and Figure 5—figure supplement 1E , ( β=0 . 107 , 95% CI [0 . 009 , 0 . 205] , t99=2 . 164 , p=0 . 033 ) ; injections in caudal RSC had centers of mass located more distal in PrS layers V-VI , compared to what was seen following injections in rostral RSC . Additionally , we observed a significant rostrocaudal-by-dorsoventral placement of injection interaction effect , since injections located more dorsal and more caudal had centers of mass in significantly more distal parts of layers V and VI of PrS and PaS ( β=1 . 039 , 95% CI [0 . 529 , 1 . 548] , t99=4 . 049 , p<0 . 001 ) . The dense and extensive labeling in layers V-VI of MEC was clearly seen in case of injections in A29 , while a more restricted area , medially in MEC , was covered after injections in A30 ( Figure 5—figure supplement 1A ) . Multiple regression analysis confirmed that the centers of mass of the terminating axons were located more lateral after injections in ventral RSC compared to after injections in dorsal RSC ( Figure 5B and Figure 5—figure supplement 1G; β=−0 . 121 , 95% CI [−0 . 222 , −0 . 020] , t85=−2 . 378 , p=0 . 020 ) . However , the rostrocaudal placement of the injection had no significant relationship with the mediolateral location of the labeled fibers ( β=−0 . 048 , 95% CI [−0 . 150 , 0 . 054] , t=−0 . 937 , p=0 . 351 ) . There was no significant interaction effect of rostrocaudal-by-dorsoventral placement of the injection . The overall analysis thus supported the conclusion that the dorsoventral position of the injection in RSC determines the transverse position of the labeled fibers in PrS , PaS , and MEC . By plotting each of the transverse coordinates of the centers of mass against the dorsoventral coordinate of the injection we did not observe any discrete 'jumps' ( data not shown ) . This suggests that the topographical organization of projections from RSC to PHR is not organized into discrete projection patterns from each of A30 or A29 or its subdivisions a , b or c , but rather is organized as a continuous dorsoventral gradient , similar to what has been reported for the adult situation . Next , we aimed to investigate how the topographical organization of the projections developed during the postnatal period . We binned the experiments in three age groups , aged P1-P6 ( n=33 ) , P7-P13 ( n=52 ) , and injections in animals older than P14 ( n=20 ) . The labelling patterns from all experiments in each of these age groups were in general similar ( Figure 6—figure supplement 1 , All areas ) . In all age groups , layers I , III and V-VI of dorsal PrS , layers V-VI of dorsal PaS , and layers V-VI of medial MEC were labeled following injections in RSC . Additionally , for all age groups , the dorsal PrS was the most commonly labeled part of PHR . The distribution of labeling was comparable between all age groups . We subsequently refined this analysis to study differences between rostral and caudal RSC with respect to labeling patterns in PHR . We plotted the centers of mass of the labeled projections observed after injections in the rostral half ( n=10 , 1st week; n=32 , 2nd week; n=5 , 3rd week and older ) and caudal half of RSC ( n=23 1st week , n=30 2nd week , n=15 3rd week and older; Figure 6A ) . The overall patterns across the age groups were similar and resembled the data including all injections in each subgroup . After injections in the rostral half of RSC , the labeling patterns in all age groups were limited to layers I , III and V-VI of dorsal PrS , layers V-VI of dorsal PaS and layers V-VI of MEC ( Figure 6—figure supplement 1 R ) . However , in MEC , animals aged younger than a week had a tendency to only display labeled fibers in the most dorsomedial part while older animals also displayed labeled fibers in more lateral and ventral parts of MEC . 10 . 7554/eLife . 13925 . 023Figure 6 . Development of topographies . The position of the centers of mass of the labelling in layers I and III of PrS , layers V-VI of PrS and PaS and layers V-VI of MEC and LEC is plotted ( see also Figure 6—figure supplement 2 ) . Each dot is color coded with respect to the rostrocaudal ( A ) or dorsoventral position ( B ) of the injections in RSC ( rostral half; blue , caudal half; red , A29; yellow , A30; green ) . Left column; animals aged P1-6 , middle column; animals aged P7-13 , right column; animals aged P14-28 . Age is not related to the dorsoventral position of the plexus ( PrS LI-III: p=0 . 876; PRS and PaS LV-VI; p=0 . 187; EC LV-VI: p=0 . 198 ) or the transverse position of the plexus ( PrS LI-III: p=0 . 641; PrS and PaS LV-VI; p=0 . 325; EC LV-VI: p=0 . 402 ) . Multiple regression was used for all statistical tests ( Figure 5—source data 1–12 ) . For flatmaps of the projection patterns see Figure 6—figure supplement 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 13925 . 02310 . 7554/eLife . 13925 . 024Figure 6—figure supplement 1 . Flatmaps of projection patterns of different age groups . All injections were subdivided into three groups based on the age of the injected animal . 1st column shows flatmaps of averaged labeling patterns in animals injected during the 1st postnatal week , 2nd column shows flatmaps of animals injected during the 2nd postnatal week , 3rd column shows flatmaps of animals injected during the 3rd postnatal week or later and the 4th column contains animals across all ages . The 1st row shows flatmaps of labeling patterns independent of the location of the injection . Numbers under the flatmap depict the number of experiments in each group . The 2nd row shows flatmaps of the labeling patterns observed after injections in A30 , while the 3rd row shows flatmaps of the labeling patterns observed after injections in A29 . In row 3 and 4 the numbers under the flatmap depict the number of injections in rostral quarter of RSC + the number of injections in the intermediate-rostral quarter of RSC + the number of injections in the intermediate-caudal quarter of RSC + the number of injections in the caudal quarter of RSC . The 4th row shows flatmaps of the labeling pattern observed after injections in the rostral half ( R ) of RSC while the 5th row shows injections in the caudal half ( C ) of RSC . In rows 4 and 5 the numbers under the flatmap depict the number of injections in A29 + the number of injections in A30 in the respective group . Dorsal , ventral , proximal ( prox ) , distal ( dist ) , medial ( med ) , lateral ( lat ) indicates dorsoventral and transverse axis of the flatmaps . DOI: http://dx . doi . org/10 . 7554/eLife . 13925 . 02410 . 7554/eLife . 13925 . 025Figure 6—figure supplement 2 . Development of topographies . ( A–C ) Centers of mass of the labeled plexus after each injection were calculated for PrS layers I and III ( A ) , PrS and PaS layers V-VI ( B ) and MEC and LEC layers V-VI ( C ) . In the plots each injection is plotted according to the age of the animal and the rostrocaudal position in RSC and color coded according to the dorsal ( blue ) , ventral ( red ) position of the center of mass of the projection in PHR ( see color maps in bottom left of each figure ) . ( D–F ) Centers of mass of the labeled plexus after each injection were calculated for PrS layers I and III ( D ) , PrS and PaS layers V-VI ( E ) and MEC and LEC layers V-VI ( F ) . In the plots each injection is plotted according to the age of the animal and the dorsoventral position in RSC and color coded according to the proximal/lateral ( yellow ) and distal/medial ( green ) position of the center of mass of the projection in PHR ( see color maps in top right of each figure ) . DOI: http://dx . doi . org/10 . 7554/eLife . 13925 . 025 Next , we split each age group in two different subgroups with respect to the location of the injection in A29 or A30 , combining data on rostral and caudal RSC . We plotted the centers of mass of the labeling after injections in A30 ( n=9 1st week , n=31 2nd week , n=15 3rd week and older ) and A29 ( n=24 1st week , n=21 2nd week , n=5 3rd week and older , Figure 6B ) . Following injections in A30 in all age groups , the labeled plexus were located in layers I , III and V-VI ( Figure 6—figure supplement 1 A30 ) . In animals aged younger than a week , a tendency for less dense labeling was seen in ventral PHR compared to the older age groups . After injections in A29 , all age groups showed comparable labeling patterns . In PrS layers I , III , and V-VI , the terminating axons in all age groups were located more proximally compared to the projection patterns seen after injections in A30 ( Figure 6—figure supplement 1 A29 ) . Similar to injections in A30 , in animals aged younger than a week , A29 injections resulted in less labeling in ventral levels of PHR . This effect was most obvious in MEC . After injections in caudal A29 and A30 , the labeling patterns were in general similar across all age groups ( Figure 6—figure supplement 1C ) . Injections in caudal RSC resulted in preferred labeling in more ventral parts of PHR compared to injections in rostral RSC . A descriptive assessment of the projection patterns thus suggested that the different age groups had comparable patterns of labeling . The topography along the transverse axis of PrS and PaS seen after A29 and A30 injections was observed already during the first postnatal week . In addition , in all age groups , caudal injections resulted in more extensive labeling in ventral PHR compared to rostral injections . However , injections in younger animals tended to have less labeled fibers in ventral PHR . To investigate this phenomenon more carefully , we plotted the coordinates of the centers of mass of the axonal labeling as a function of age and the location of the injection ( Figure 6 and Figure 6—figure supplement 2 ) . This analysis indicated that already from the earliest postnatal ages , the position of the centers of mass are organized as described above . Already at the earliest ages , we observed that different rostrocaudal levels of RSC project to different dorsoventral levels of PrS and PaS . Even though we observed a tendency for the youngest animals to not display labeling more ventrally in PHR , age had no significant effect on the position of the centers of mass along the dorsoventral axis in neither layers I and III of PrS ( Figure 6—figure supplement 2A , β=0 . 000 , 95% CI [−0 . 005 , 0 . 004] , t99=−0 . 157 , p=0 . 876 ) , in layers V-VI of PrS and PaS ( Figure 6—figure supplement 2B , β=−0 . 002 , 95% CI [−0 . 006 , 0 . 001] , t99=−1 . 328 , p=0 . 187 ) , nor in layers V-VI of MEC and LEC in older animals ( Figure 6—figure supplement 2C , β=−0 . 003 , 95% CI [−0 . 007 , 0 . 001] , t85=−1 . 299 , p=0 . 198 ) . To check if animals aged younger than a week had less labeled fibers in ventral PHR compared to older animals , we converted the continuous age variable to a discrete variable were animals aged younger and older than a week were considered as two different groups . However the age groups were not significantly related to the location of the center of mass ( lowest p-value=0 . 221 ) . In neither of the regression analyses , we observed any significant location-by-age interaction effects . Already in the youngest cases , the centers of mass in layers I and III of PrS and layers V-VI of PrS and PaS after injections in A29 were located more proximal compared to those resulting from injections in A30 ( Figure 6B and Figure 6—figure supplement 2D and E ) . Age did not predict the location of the centers of mass along the transverse axis for neither layers I and III of PrS ( Figure 6—figure supplement 2D , β=0 . 001 , 95% CI [-0 . 005 , 0 . 008] , t99=0 . 468 , p=0 . 641 ) , layers V-VI of PrS and PaS ( Figure 6—figure supplement 2E , β=0 . 002 , 95% CI [-0 . 002 , 0 . 007] , t99=0 . 990 , p=0 . 325 ) , nor layers V-VI of MEC and LEC ( Figure 6—figure supplement 2F , β=0 . 002 , 95% CI [-0 . 003 , 0 . 008] , t85=0 . 843 , p=0 . 402 ) . However , for layers V-VI of PrS and PaS there was a significant interaction of rostrocaudal placement of injection-by-age ( β=-0 . 033 , 95% CI [-0 . 056 , -0 . 11] , t99=-2 . 995 , p=0 . 004 ) , indicating that at older ages , respectively more rostral injections resulted in labeling in more distal parts of layers V-VI PrS and PaS . Based on these results we conclude that the heterogeneous projection patterns observed in later postnatal stages is also present in the first postnatal stages . We next asked whether the projection patterns in PHR , observed in young animals , are also present in adult animals ( approximately 3 months ) . To this end , we reanalyzed a dataset described in earlier publications from our lab ( Jones et al . , 2005; Jones and Witter , 2007 ) . Analyses of the cases individually , suggested that the organization of the projection pattern as we described in young animals was also present in adult animals . After injections in A30 , a dense fiber plexus was observed distally in layers I and III of dorsal PrS ( Figure 7A ) . Additional labeled fibers were observed in layers V-VI of distal PrS , PaS and medially in MEC , comparable to the projection pattern we observed after injections in A30 of pups ( Figure 5—figure supplement 1A and Figure 6—figure supplement 1 ) . Compared to injections in A30 , injections in A29 resulted in labeled fibers more proximally in PrS and more laterally in MEC ( Figure 7B ) . This topographical organization seems to be graded , since the injection located at the border between A29 and A30 displayed labeled fibers in a mediolateral position between the dorsally- and ventrally injected cases ( compare Figure 7C with A and B respectively ) . Compared to rostrally placed injections ( Figure 7A and B ) , we observed that caudally placed injections displayed more labeled axons in ventral PrS and MEC , and the area receiving the densest projection in both PrS and MEC was shifted more ventrally ( Figure 7C ) . Taken together , these findings suggest that the RSC to PHR projections in the postnatal brain are similarly organized to the adult ones . 10 . 7554/eLife . 13925 . 026Figure 7 . Labelling patterns in the adult . ( A ) An injection in intermediate rostrocaudal A30 resulted in dense labelling in layers I and III of distal PrS and a few fibers in deep PrS , PaS and POR ( A1 ) . More ventrally , fibers also invaded the MEC with the most fibers located medially in MEC ( A3 ) . ( B ) An injection in intermediate rostrocaudal A29 resulted in a dense fiber plexus in layers I and III of proximal PrS and POR ( B2 ) . At this dorsoventral level a few fibers also invaded PaS . More ventrally , a fiber plexus was present in lateral MEC . ( C ) An injection at the border of A29 and A30 in caudal RSC resulted in fibers located in layer I of PrS and deep PaS and POR dorsally ( C1 ) . More ventrally , a dense projection to layers I and III of PrS ( C2 and 3 ) and a moderate projection to intermediate mediolateral MEC was present ( C4 ) . In A-C numbers above each section depict normalized dorsoventral position of the section and the line at the bottom of the figure represents the relative dorsoventral position of each section . Scale bar equals 1000 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 13925 . 026 We aimed to investigate whether there was a chronological development of the number of RSC-axons which could be traced within PHR . In the first postnatal week , very few fibers were labeled in PHR . At P1-P3 we only observed single unbranched fibers in PHR ( Figure 8A–D ) . Along the fibers , axonal growth cone-like swellings were observed , independent of whether the fibers were located in the cortex or in the white matter . At P3 , we observed the first branching fibers in PHR , although these fibers typically branched only once or twice in each section . After P3 , the plexus gradually increased in density and complexity . However , we did not observe an adult-like plexus until P12 ( Figure 8F ) . At this age , we observed dense plexus , with fibers displaying numerous branching points and each fiber containing fine extensions and protrusions , similar to what was seen in older animals and adults ( Figure 8G ) . Based on these observations we concluded that the number of RSC fibers in PHR increase gradually from single fibers at P1 to adult-like densities around P12 . 10 . 7554/eLife . 13925 . 027Figure 8 . Examples of the densest fiber plexus observed in PHR at different ages . At birth , single unbranching fibers were present in superficial layers of PrS ( A ) , deep layers of PrS ( B ) and deep layers of MEC ( C ) . At P3 , some branching fibers were observed ( D , PrS ) . These fibers typically branched only once or twice . At P11 ( E , PrS ) the complexity of the fiber plexus increased as fibers have multiple branching points and thin fibers are seen within the plexus . At P12 ( F , PrS ) the first plexus which was comparable to plexus in adolescent animals ( G , PrS ) and adults ( data not shown ) was observed . In the experiment shown in G , the terminal distribution of three differentially labeled projections is illustrated . A plexus resulting from an injection in intermediate-rostral A29 ( yellow ) terminated throughout layer III of PrS , while the plexus observed after an injection in intermediate-rostral A30 terminated in the center of layer III of PrS ( cyan ) . Fibers originating from a third injections in intermediate-caudal RSC ( magenta ) are observed , however the densest fiber plexus is located in a section at more ventral levels of PrS . The large , overlapping magenta and yellow structures are endothelium which take up alexa-preconjugated dextran amines . White line depicts the deep and superficial borders of layer III . Scale bars equal 100 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 13925 . 027 Since most of the RSC-PHR projections seemed to be fully developed very early during postnatal development , we aimed to investigate whether the PHR projecting neurons in RSC were located in layer V , similar to adults ( Burwell and Amaral , 1998a ) . After fast blue injections in PHR of pups we identified labeled neurons in RSC in superficial layer V at P5 ( Figure 9A ) and P11 ( Figure 9B ) , which suggested that the RSC-PHR projections are adult-like with respect to the layer in which the neurons are located already during the first postnatal week of development . These observations are in line with our anterograde material in which we observed that all injections that did not cover parts of layer V ( n=8 ) did not result in any labeled axons in PHR . 10 . 7554/eLife . 13925 . 028Figure 9 . RSC projections to PHR arise from neurons in layer V . ( A ) A FB injection in deep layers of mainly PrS and PaS ( top right; P5 , coronal section ) resulted in retrograde labeling of neurons in CA1 , SUB and RSC ( top row ) . Coronal sections shown are from intermediate-rostral RSC ( 1 ) , intermediate-caudal RSC ( 2 ) and caudal RSC ( 3 ) . Cyan squares depict size and position of high-power images in the bottom row . Bottom row: High power images of retrogradely labeled neurons in superficial layer V of RSC . Dashed lines depict the pia and the border between cortex , white matter and the corpus callosum . Solid lines depict borders between layer I , layers II-III and layer V and border between A29 and A30 . Scale bars equal 100 μm ( high power images ) and 1000 µm ( low power images ) . ( B ) FB was injected in an intermediate dorsoventral level in MEC and PaS ( top right; P11 , horizontal sections ) and resulted in retrogradely labeled neurons in superficial layer V of caudal RSC ( top row ) . Horizontal sections are organized from dorsal ( 1 ) to ventral ( 3 ) . Cyan squares depict location of high-power images in the bottom row . White lines depicts borders of RSC and border between A29 and A30 . Bottom row: High power images of retrogradely labeled neurons in superficial layer V of RSC . Dashed lines depict the pia and the border between cortex and white matter . Solid lines depict borders between layer I , layers II-III and layer V . Scale bars equal 100 μm ( high power images ) and 1000 µm ( low power images ) . DOI: http://dx . doi . org/10 . 7554/eLife . 13925 . 028 To study the development of RSC projections to HF-PHR , we injected anterograde tracers in RSC of rats aged P0-28 . We conclude that the postnatal RSC projects densely to all layers of PrS and posterior POR and deep layers of PaS and MEC and weakly to deep layers of LEC and to SUB . Our retrograde experiments showed that the origin of these projections were neurons located in superficial parts of layer V of RSC . These findings are in accordance with previous work in the adult ( Wyss and Van Groen , 1992; Shibata , 1994; Burwell and Amaral , 1998a; Jones and Witter , 2007 ) . Additionally , we report that the RSC projections to PrS , PaS and EC are topographically organized similarly already in the youngest postnatal rats and in adult rats ( Figure 10 ) . The notion that rostral RSC only projects to dorsal PHR , while caudal RSC projects to additional more ventral parts of PHR is in accordance with previous work in the adult ( Wyss and Van Groen , 1992; Shibata , 1994; Jones and Witter , 2007 ) . Second , dorsal RSC ( A30 ) projects preferentially to distal PrS , PaS and medial MEC , while more ventral parts of RSC ( A29 ) project significantly more to proximal PrS and more lateral parts of MEC . To our knowledge , such topographies have not been reported in earlier studies . 10 . 7554/eLife . 13925 . 029Figure 10 . Summary of topographical organization of projections from RSC to PHR in the developing and adult brain . Schematic representation of the organization of the projections from RSC ( top ) to PHR ( bottom ) . Projections from RSC to PHR terminate mainly in PrS , PaS and MEC . Projections originating from rostral RSC ( purple and black ) terminate in dorsal PHR , while projections originating from caudal RSC ( yellow and grey ) terminate in more ventral parts of PHR . Projections originating from ventral RSC ( black and yellow ) terminate in proximal PrS and in lateral parts of MEC , while projections originating from dorsal RSC ( purple and grey ) terminate in distal PrS and in medial parts of MEC . DOI: http://dx . doi . org/10 . 7554/eLife . 13925 . 029 The observation that A30 is preferentially connected to distal PrS , PaS and medial MEC , while A29 is preferentially connected to proximal PrS and more lateral parts of MEC , but not PaS is in line with other connectional and functional differences . The PrS to MEC projection is topographically organized such that distal PrS projects to medial MEC , while proximal PrS projects to more lateral parts of MEC ( Shipley , 1975; Honda and Ishizuka , 2004 ) . Furthermore , these partner domains in PrS and MEC are selectively innervated by different areas of SUB along its transverse axis such that distal SUB projects to distal PrS and medial MEC , while more proximal parts of SUB , with the exclusion of the very proximal part , project to proximal PrS and more lateral MEC ( Witter , 2006; O'Reilly et al . , 2013 ) . This indicates the existence of a connectional route linking A30 , medial MEC , distal SUB and distal PrS to each other . A parallel route links A29 with more lateral MEC , more proximal SUB and proximal PrS . Although no clear transverse gradients in spatial modulation have been reported in MEC , electrophysiological properties of neurons show a transverse gradient , indicative that more medially positioned grid cells might be more precisely spatially tuned ( Canto and Witter , 2012 ) . Furthermore , neurons in distal SUB are more spatially modulated compared to those in proximal SUB ( Sharp and Green , 1994 ) . All observations thus point to a differentiation of A29 and A30 where A30 is connected preferentially to more spatially modulated neurons in medial MEC and distal SUB compared to A29 which is more connected to neurons in lateral MEC and more proximal SUB . Additional cortical and subcortical connections are in line with this proposed differentiation between both areas of RSC . The projections to anterior cingulate cortex are differentially organized for A29 and A30 and while A30 is connected to visual area A17 and A18b , A29 is only connected to A18b . With respect to thalamic connections , A29 is connected to the anterodorsal and anteroventral nuclei while A30 preferentially connects with the anteromedial nucleus ( van Groen and Wyss , 1990 , 1992; Shibata , 1998; van Groen and Wyss , 2003; Jones et al . , 2005 ) . Taken together , these connectional differences are in line with reported functional differences between the two areas ( van Groen et al . , 2004; Vann and Aggleton , 2005; Pothuizen et al . , 2009; Pothuizen et al . , 2010 ) . Such a functional differentiation between A29 and A30 could thus possibly result in functional differences along the transverse axes of PrS , PaS and MEC . During the postnatal period we observed an overall increase in the density of labeled axonal branches in all PHR subregions . Even though we did not perform a formal quantification of the number of labeled fibers in PHR , we only observed single unbranching fibers in animals aged P1-2 . During the first postnatal week , the number of axons generally increased , and the fibers displayed several branching points towards the end of the first postnatal week . The first terminal plexus with adult-like densities were observed in P12 animals . Even though several other factors , such as tracer type and the number of layer V RSC neurons involved in the injection also had an impact on the number of fibers observed in each experiment , we are confident that the most important predictor of labeled fiber density was the age of the animal . Interestingly , the time window of increased density of RSC afferents in PHR is paralleled by several anatomical and physiological changes in PHR . EC afferents originating in PaS and PrS become functional from P8 and mature gradually until they are fully adult-like around P14 ( Canto et al . , 2011 ) . Similar developmental timescales have also been reported for the functional development of intralaminar projections within EC ( O'Reilly et al . , 2010 ) . Our temporal analysis revealed that RSC axons destined for PHR migrate directly into their area of termination and thereafter keeps their position constant while the number of axonal branches and the total axonal spread increases gradually until they reach adult-like plexus features . This observation is supported by our center of mass analyses of the early perinatal RSC to PHR projections , showing that projections originating in different parts of RSC show a striking terminal topography already during the first postnatal week . Head-direction cells and border cells are all present in PHR when electrodes are lowered into the brain during the second postnatal week ( Bjerknes et al . , 2014; Bjerknes et al . , 2015 ) , while grid cells mature during the third and fourth postnatal week ( Langston et al . , 2010; Wills et al . , 2010 ) . Even though no published experiments have investigated whether border cells and head-direction cells are present before the second postnatal week , our data indicate that the topographical organization of RSC terminals in PHR is present before spatially modulated neurons are present in PHR and that adult-like axonal densities can be observed approximately at the same time-point as the first border cells and head-direction cells are observed in PHR . This conclusion is comparable to what has been observed for intrinsic HF-PHR connectivity which is also topographically organized already at early postnatal periods , demonstrating increased plexus densities for intrinsic HF-PHR projections during ongoing development ( Fricke and Cowan , 1977; Borrell et al . , 1999; O'Reilly et al . , 2013; O'Reilly et al . , 2014 ) . It is also comparable to what has been reported in several other developing brain systems . For instance , for thalamocortical projections in the visual system of both monkeys , cats and ferrets , the first arriving axons in the visual cortex already show a topographical organization into ocular dominance columns ( Horton and Hocking , 1996; Crowley and Katz , 2000; Crair et al . , 2001 ) . Similarly , the rat ventral posterior thalamic nucleus issue projections to the somatosensory cortex in which the first arriving axons target distinct areas later forming a defined barrel ( Catalano et al . , 1996 ) . Sensory inputs to the olfactory bulb are present long before neurons display a receptive field . Moreover , the development of these inputs is independent of activity in sensory receptors , which suggests that the development of topographies in the olfactory bulb is experience independent ( Lin et al . , 2000 ) . The same conclusion apparently holds for the projections from RSC to PHR , These findings are different from what has been reported for the retinogeniculate projection in several species . In this projection , there is an overshoot of axonal terminals , which are initially diffuse and later pruned into an adult-like topography ( Rakic , 1976; Linden et al . , 1981; Shatz , 1983; Godement et al . , 1984 ) . These differences might imply that distinct molecular- or activity based principles governs the axonal termination patterns in different neural projections . The results presented in this study thus lead us to conclude that the topographical organization of PHR connectivity is present when the first RSC axons arrive in PHR . The densities of the terminal plexus appear to develop gradually without any clear signs of pruning , though it remains to be determined whether connectional reorganization occurs at the synaptic level during development . The first plexus with adult-like densities can be observed around P12 which is around the time when the first spatially modulated neurons in PHR have been observed ( P11 ) , but before eye-opening and before the animals starts to navigate ( around P15 ) and thus before adult like grid cells are observed in PHR ( after P25; Langston et al . , 2010; Wills et al . , 2010; Bjerknes et al . , 2014; Bjerknes et al . , 2015 ) . These findings suggest that RSC afferents might be important for the development of head-direction and border cells . Eighty two female and male Long Evans rats aged between P0 and P28 were used in this study . Additionally , we reanalyzed anterograde injections performed in approximately 3 months old adult Wistar rats ( Jones et al . , 2005; Jones and Witter , 2007 ) and retrograde injections performed in Long Evans pups ( O'Reilly et al . , 2014 ) . We refer to these original publications for a detailed description of the experimental protocols . The pups were bred in-house and housed in enriched cages together with their parents and littermates . Cages were checked every morning and evening for pups and the day pups were observed was considered P0 . We use the day of perfusion designating the age of the animal . To avoid unnecessary stress for the animals , litters with more than ten pups were culled to ten pups at P0 or P1 . At P21 , the pups were separated from their parents and moved to cages together with littermates of the same sex . The animals lived in a controlled environment ( 22 ± 1°C; humidity 60%; lights on from 8:00 P . M . to 8:00 A . M . ) . Food and water were available ad libitum . The experimental protocols followed the European Communities Council Directive and the Norwegian Experiments on Animals Act and local directives of the responsible veterinarian at the Norwegian University of Science and Technology . All surgeries were conducted under isoflurane gas anesthesia . Animals were placed in an induction chamber and fully anesthetized before they were moved to a stereotaxic frame . The head was fixed using a neonatal mask and mouthpiece ( model 973-B; Kopf , Tujunga , CA ) and zygoma ear cups ( model 921; Kopf ) . Animals older than P18 were mounted in a small-sized adult mask and the head was fixed with blunted ear bars . Before incision , the skin was disinfected with 2% iodine in 65% ethanol , and a local analgesic bupivacain ( 0 . 2 ml per 100 g bodyweight of a 0 . 5 mg/ml solution; Marcain , Astra Zeneca , London , UK ) was injected subcutaneously at the place of incision . The skin was opened with a small-sized and sharp tipped scissor . After incision , the mouthpiece and ear cups were adjusted so that bregma and lambda were aligned horizontally . Before injecting , bone over the place of injection and over the posterior extreme of the sagittal sinus was removed . The exact place of injection was measured using the junction of the transverse- and sagittal sinus as a reference for the anteroposterior coordinate , the lateral edge of the midsagital sinus as a reference for the mediolateral coordinate and the level of the dura as a reference for the dorsoventral coordinate . Before injection , the dura was punctured , and glass micropipettes with an outer diameter of 20–25 μm were lowered into the brain ( 30–0044 , Harvard Apparatus , Holliston , MA; pulled with a PP-830 puller , Narishige , Japan ) . The anterograde tracers biotinylated dextran amine ( BDA; 5% in phosphate buffer ( PB; 0 . 125M in H20; pH 7 . 4 ) , 10000 MW , D1956 , Invitrogen , Eugene , OR ) or preconjugated dextran amines ( all 5% in PB , 10 000 MW; Alexa-488 DA , D22910; Alexa-546 DA , D22911; Alexa-647 DA , D22914; Invitrogen ) were iontophoretically injected through the micropipettes into RSC ( 4–6 μA , alternating currents , 6 s on/6 s off , for 5–15 min , 51595; Stoelting , Wood Dale , IL ) . Throughout the surgery , appropriate amounts of sterile saline ( room temperature ) were administered subcutaneously to avoid dehydration . Animals were also administered carprofen during surgery as a post-surgery analgesic ( 1 ml per 100 g bodyweight of a 0 . 5 mg/ml solution; Rimadyl , Pfizer , New York , NY ) . After surgery , the incision was sutured and the pups were allowed to recover under a heating lamp . When fully awake , the animals were returned to maternal care until the time of sacrifice . We euthanized the animals 18–30 hr after surgery under a terminal anesthesia with isoflurane . The thorax was opened and cold Ringer’s solution ( 8 . 5 g NaCl , 0 . 25 g KCl and 0 . 2 g NaHCO3 per liter of H2O , pH 6 . 9 ) was transcardially perfused through the body . When the liver turned pale the perfusion solution was changed to a 4% solution of freshly depolymerized paraformaldehyde in PB ( pH 7 . 4 ) . In case of P0-P2 animals , 0 . 1% glutaraldehyde was added to the fixative . The brain was removed from the skull and postfixed overnight at 4°C in the same fixative . Twenty-four hours after perfusion , the brains were transferred to PB containing 2% dimethyl sulfoxide ( DMSO; VWR , Radnor , PA ) and 20% glycerol ( VWR ) . Brains were cut with a freezing microtome ( HM-430 Thermo Scientific , Waltham , MA ) in 40 or 50 µm thick horizontal sections . Depending on the age of the animal sections were collected in four to six equally spaced series . One of the series was mounted directly on superfrost slides ( 10149870 , Thermo Scientific ) . The remaining series of sections were collected in vials containing 2% DMSO and 20% glycerol in PB and stored at -20°C until further usage . The mounted series was Nissl-stained by first dehydrating the sections in increasing ethanol solutions ( 50% , 70% , 80% , 90% , 100% , 100% , 100% ) followed by two minutes in xylene ( VWR ) to clear the sections . Thereafter , the sections were rehydrated in decreasing ethanol solutions ( opposite order as the dehydration protocol ) and placed in cresyl violet solution for two to six minutes . Subsequently , the sections were rinsed quickly in water and placed in 50% ethanol containing acetic acid to differentiate the staining . The sections were dehydrated in ethanol , cleared in xylene and finally coverslipped with Entellan ( 107961 , Merck , Darmstadt , Germany ) . In case of brains with BDA injections , one series of sections was rinsed three times for 10 min in PB and then three times for 10 min in tris ( hydroxymethyl ) aminomethane ( Tris ) -buffered saline ( 50 mM Tris ( Merck ) and 150 mM NaCl in H2O ) containing 2% Triton X-100 ( TBS-Tx; Merck , pH 8 . 0 ) . In experiments were multiple tracers were injected , the sections were incubated with Alexa-conjugated streptavidin ( Alexa-405 S32351 , Alexa-488 , S11223; Alexa-546 , S11225; Alexa-633 , S21375 , Invitrogen ) in a 1:200 solution with TBS-Tx overnight at 4°C . In experiments were BDA was the only tracer injected in the brain , sections were incubated for 90 min in TBS-Tx with avidin-biotin-peroxidase ( Vectastain Standard PK-4000 ABC kit; Vector , Burlingame , CA ) according to the manufacturer’s instructions . Subsequently , sections were rinsed three times for 10 min in TBS-Tx and two times for 5 min in Tris-HCl ( 50 mM Tris in H2O , pH adjusted to 7 . 6 by adding HCl ) and incubated for approximately 15 min in a diaminobenzidine tetrahydrochloride ( DAB ) –peroxidase solution containing 5 mg DAB ( D5905 , Sigma-Aldrich , St . Louis , MO ) and 3 . 3 µl H2O2 ( H1009 , Sigma-Aldrich ) in 10 ml Tris-HCl . Irrespective of whether the incubations were carried out with Alexa-conjugated streptavidin or DAB , the sections were rinsed two times for 5 min in Tris-HCl , and subsequently mounted on glass slides from a 0 . 2% gelatin solution in Tris-HCl . After overnight drying , they were cleared in toluene and coverslipped with Entellan ( Merck ) . Sections were inspected with fluorescence illumination at the appropriate excitation wavelength or conventional brightfield illumination ( Zeiss Axio Imager M1/2 ) . Digital images of successful injections and anterogradely labeled plexus in HF-PHR were obtained using a slide scanner equipped for either brightfield or fluorescent imaging ( Zeiss Mirax Midi; objective 20X; NA 0 . 8 ) . For illustrative purposes , images of labeled cells and Nissl stained tissue were exported using Panoramic Viewer software ( 3DHistech , Budapest , Hungary ) and processed in Adobe Photoshop and Illustrator ( CS6 , Adobe Systems , San Jose , CA ) . After the sections were digitized , we aimed to obtain , for each experiment , realistic estimates of the location of the anterogradely labeled axons in PHR and of the locations of the injections in RSC . To achieve this , we first produced an average flatmap of PHR based on measurements of each PHR subdivision in all animals ( Figure 4A ) . Second , we measured the location of labeled fibers within PHR and represented the position of the labeled fibers on the average flatmap ( Figure 4B ) . Third , we plotted the injections in a reference atlas brain ( Figure 1 ) . In order to create an average flatmap , we produced individual flatmaps of all animals . To this end , we delineated all subdivisions within PHR using cytoarchitectonic differences between the different subdivisions ( Boccara et al . , 2015 ) . Borders were established in fluorescent- or DAB-stained sections overlaid with the neighboring Nissl-stained section . In all sections , we measured ( using Panoramic Viewer software , 3DHistech , Budapest , Hungary ) the extent , along the transverse axis , of subdivisions containing labeled RSC axons , i . e . superficial layers of PrS , deep layers of PrS , deep layers of PaS , deep layers of MEC and deep layers of LEC ( Figure 4A1–4 ) . The transverse measurements where obtained from all sections of all brains , stored in excel files and the data were further processed using MatLab software ( R2015b , MathWorks , Natick , MA ) . We next aimed to make one average flatmap , constituting of square bins , representing the mean 'shape' of PHR across all brains . Since brains of animals of different ages have different sizes , we first converted the absolute measurements into normalized values . For this , we divided each measurement by the maximum measured extent of the respective subdivision for the particular animal ( Figure 4C1 ) . Next , we binned the dorsoventral axis of each PHR subdivision in 29 equally sized bins , since the maximum number of sections containing PHR in a single series was 29 . Subsequently , we calculated the mean of the normalized transverse measurements , across all animals , for each of the 29 dorsoventral levels ( Figure 4C2 ) . This procedure was repeated for each subdivision . To obtain bins each representing a square area of the brain , we first calculated the ratio between the total dorsoventral extent of PHR and the maximum measured extent along the transverse axis of each animal ( Figure 4C1 ) . Thereafter , we calculated the mean of these ratios across all animals ( Figure 4C2 ) . Next , the 29 normalized transverse measurements were divided by the mean of the dorsoventral extents ( Figure 4C3 ) . This procedure was repeated for each subdivision . This procedure thus resulted in , for each subdivision , 29 dorsoventral levels with different transverse extents , expressed as a value relative to the dorsoventral length of PHR . The transverse extents were subsequently turned into square bins ( calculated as the average extent along the transverse axis x 29 and rounded to the nearest integer , Figure 4C4 ) so that each subdivision could be represented as a flatmap containing 29 rows of bins with a variable amount of square bins in each row ( Figure 4A5 ) . We represented the location of labeled axons in PHR for each experiment within the average flatmap in two ways . In experiments where a dense labeled plexus was present , the distances between the boundaries of the plexus and the borders of the respective PHR subdivisions were measured along the transverse axis ( Figure 4B1 ) . We did this for each section containing a labeled plexus . The density of labeling in each plexus in each section was subsequently given a value from 1–3 depending on a subjective evaluation . The densest plexus in each experiment was given a value of '3' , while weaker plexus were valued '1' or '2' depending on the density relative to the densest plexus observed in the experiment . Alternatively , in experiments in which we observed only single labeled axons or a sparsely labeled plexus , the transverse measurements were obtained as follows . In PrS , the distance from each labeled axon to the proximal border of PrS was measured , while in deep layers of PaS the distance from each labeled axon to the distal border of PaS was measured . In deep layers of MEC and LEC , the distance from each labeled axon to the medial border of the respective subdivision was measured ( Figure 4B3 ) . In cases of missing or damaged sections , we estimated the putative projection pattern in the section by using the average projection pattern of the sections directly above and below . To identify the bin ( s ) in the average flatmap representing the location of labeled fibers , their absolute position , as established above , was normalized within its respective area . In cases where we observed a dense plexus , each bin was given the value corresponding to what was described above ( Figure 4B2 ) . In cases where we measured the location of labeled fibers , each bin was given a value equal to the numbers of labeled axons present in the location represented by the respective bin ( Figure 4B4 ) . All bins not containing any labeled axons or located outside the labeled plexus were given the value 0 . To be able to directly compare or pool flatmaps representing either single fibers or dense plexus , we normalized the values of all bins in each experiment to the maximum valued bin in the respective experiment , resulting in bins in each experiment with values ranging between 0 , representing no plexus or fibers , or 1 , representing the bin with the densest labeling . To compare different sources of variability for the projection patterns , we organized the flatmaps into different groups of interest . For each group , we summed the values of bins representing the same location in all flatmaps ( Figure 4B5 ) . Finally , we normalized the 'summed' flatmap to the bin with the highest value , similar to the procedure described above . In flatmaps representing projection patterns of several subgroups , normalized flatmaps were summed , so that each subgroup had similar impact on the summed flatmap . For illustrative purposes , the flatmaps of individual experiments or groups of experiments were plotted using MatLab . In all experiments , we calculated the coordinates for the 'center of mass' of the projections along the transverse and dorsoventral axis for respectively layers I and III of PrS , layers V-VI of PrS and PaS combined and for the combined layers V-VI of MEC and LEC . The center of mass-values of the axonal plexus were used for statistical analyses . In scatter plots of center of mass-values , we defined the extremes of the color scale as ± 2 standard deviations from the mean value . All values between these extremes were plotted using a linear color scale while more extreme values were thresholded to the extreme colors . Since brains of different ages have different sizes , we aimed to normalize the position of the injections . In adult rats , RSC can be subdivided into four different cytoarchitectonic subdivisions A30 and A29a , b and c . In the adult brain , these four subdivisions are positioned along the dorsoventral extend of RSC . In the immature cortex , only the cytoarchitectonic border between A29 and A30 is clearly identifiable . We therefore chose to define the dorsoventral location of the injections in A29 and A30 as a continuous coordinate . This measure is thus indirectly related to the classical discrete cytoarchitectonic subdivisions . We used the recently released reference 3D-atlas brain ( Papp et al . , 2014; 2015 ) to map all injections in a standardized space , irrespective of age . First , we identified coordinates of the dorsal , ventral , rostral and caudal border of respectively A29 and A30 in the atlas brain . The lines between the respective coordinates were smoothed using local regression . Next , we calculated the cutting angles of our experimental brains relative to the atlas brain and made sections of the standard atlas brain with the same cutting angles . We identified atlas-sections containing landmarks and cytoarchitectonic borders present in the section containing the center of each injection . The atlas coordinate of the center of each injection was recorded and served as an age-normalized 3D point-measure of the injection location within RSC . For illustrative purposes , the coordinates of all injections were plotted within the 3D volume ( Figure 1A , ITK-SNAP , NIH ) . Since the caudal RSC cortex is curved both along the dorsoventral and rostrocaudal axis we flattened RSC and transposed each injection onto a 2D plane . This was done by dividing the surface area of A29 and A30 in the atlas brain into multiple triangles ( Figure 1B ) . The coordinates of the dorsal and ventral borders of A29 and A30 determined the coordinates of each triangle . For each injection , we calculated the shortest vector between the injection and the cortical surface within any of the triangles ( Figure 1C ) . Thereafter , we calculated the intersection of the vector and the plane within the triangle . This coordinate represented the “transposed” location onto the cortical surface of each injection ( Figure 1B and C ) . The normalized 2D coordinate of each injection was defined as follows ( Figure 1C ) ; the dorsoventral coordinate was defined as , dvdv+dd , where dv and dd represent the distance from the transposed injection to the ventral and dorsal border respectively . The rostrocaudal coordinate was obtained by first calculating a line along the rostrocaudal extend of A29 and A30 , positioned equally distant from their respective dorsal and ventral borders . Next , we calculated the shortest vector between the transposed injection and the line and found the intersection between the two . The rostrocaudal coordinate was defined as drdr+dc , where dr and dc is the cumulative distance from the cross section to the rostral and caudal end of RSC . This resulted in the normalized map of the positions of all injection sites analyzed in this study ( Figure 1D ) . For cluster analyses , we smoothed each flatmap by applying a Gaussian filter of 5 x 5 bins with a standard deviation of 1 . 5 . Thereafter we calculated pairwise correlations between each possible pair of flatmaps . The correlation matrix was subsequently clustered by using the k-means clustering algorithm in MatLab . The number of clusters was subjectively decided by evaluating the ratio between within- and between cluster variance . The number of clusters was fixed when adding more clusters would not result in a substantial decrease in the within- to between cluster variance ratio . Our dataset contained experiments in animals of all ages between P1 and P19 and injections in animals aged P27 and P28 . Since we did not have samples between P19 and P27 we only included experiments in animals aged P19 or younger in the statistical analyses . Experiments that did not result in labeled axons in any of the areas of PHC where not included in the analyses . The dorsoventral and transverse coordinates of centers of mass were thereafter analyzed independently . For each of the coordinates , we fitted a linear multiple regression model to the data using age and the normalized location of the injection as predictors . A model including the main predictors and two-way interactions was fitted to the data using SPSS ( version 20 , IBM , Armonk , NY ) . We subsequently removed from the model the two-way interaction with the lowest standardized β-value and were insignificantly different from zero ( t-test ) , and a new model was fitted . This procedure was repeated until the model only consisted of the main effects and significant two-way interaction terms . In the regression models , the center of mass coordinates ranged from 0 ( ventral ) to 1 ( dorsal ) and from 0 ( proximal and medial ) to 1 ( distal and lateral ) , the coordinates of the injection cite ranged from 0 ( rostral ) and 1 ( caudal ) and from 0 ( ventral ) and 1 ( dorsal ) and the age of the animal was measured in postnatal age in days . Regression coefficients are reported unstandardized with 95% confidence intervals . To assess whether the data met the assumption of linear relationships , we visually assessed scatter plots in which each of the independent variables were plotted versus each of the center of mass coordinates . Additionally , we evaluated the standardized predictor values plotted against the standardized regression residuals to assess whether the assumptions of linearity , homoscedasticity , independent residuals and normally distributed residuals were met . We additionally tested whether the residuals were normally distributed by visually inspecting frequency histograms and normal probability plots . Next , we tested multicollinearity of the independent variables by correlating each of the independent variables . No multicollinearity was assumed if the Pearson’s correlations were between -0 . 70 and 0 . 70 . In addition , we assessed the variance inflation factor ( VIF ) of each model . The highest calculated VIF was 1 . 197 , which is well below recommended cutoffs . Statistical significance of regression coefficients was determined using two sided t-test with p<0 . 05 as criterion .	Our ability to navigate critically depends on part of the brain called the parahippocampal region . Within this region , there are several different types of brain cells ( or neurons ) whose activity “codes” different aspects of navigation , such as position , direction and speed . To understand how parahippocampal neurons are able to form these activity patterns , we need to understand how they develop connections with neurons from other brain regions that are important for navigation , such as the retrosplenial cortex . If inputs from retrosplenial neurons are important for generating the activity patterns observed in the parahippocampal region , the connections between the two groups of neurons should be fully mature before the activity patterns emerge . In rats , this should occur around 11–16 days after birth . Sugar and Witter have now assessed how the retrosplenial inputs are organized in the parahippocampal region of rats . This revealed that , when the rats are born , there are very few retrosplenial inputs present in the parahippocampal region . However , the few inputs that are present are organized similarly to how they eventually will be organized in adults . After birth , the number of inputs gradually increases until the rats are approximately 12 days old , at which point the pattern of connections is indistinguishable from what we observe in adults . Thus it appears that retrosplenial inputs are fully mature before activity patterns emerge in the parahippocampal region . In the future , Sugar and Witter would like to investigate how inputs to the parahippocampal region are able to organize themselves during early development . The importance of retrosplenial inputs could also be investigated by manipulating them during development and adulthood .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "neuroscience" ]	2016	Postnatal development of retrosplenial projections to the parahippocampal region of the rat
Cavin-3 is a tumor suppressor protein of unknown function . Using both in vivo and in vitro approaches , we show that cavin-3 dictates the balance between ERK and Akt signaling . Loss of cavin-3 increases Akt signaling at the expense of ERK , while gain of cavin-3 increases ERK signaling at the expense Akt . Cavin-3 facilitates signal transduction to ERK by anchoring caveolae to the membrane skeleton of the plasma membrane via myosin-1c . Caveolae are lipid raft specializations that contain an ERK activation module and loss of the cavin-3 linkage reduces the abundance of caveolae , thereby separating this ERK activation module from signaling receptors . Loss of cavin-3 promotes Akt signaling through suppression of EGR1 and PTEN . The in vitro consequences of the loss of cavin-3 include induction of Warburg metabolism ( aerobic glycolysis ) , accelerated cell proliferation , and resistance to apoptosis . The in vivo consequences of cavin-3 knockout are increased lactate production and cachexia . Cavin-3 ( PRKCDBP , hSRBC ) is a tumor suppressor protein of unclear function . In humans , cavin-3 is encoded in the 11p15 . 5 tumor suppressor locus and loss of cavin-3 expression is common in many epithelial and glial derived cancers ( Zochbauer-Muller et al . , 2005; Lee et al . , 2008; Martinez et al . , 2009; Tong et al . , 2010; Caren et al . , 2011; Lee et al . , 2011 ) . Cavin-3 expression is also absent in many cancer cell lines and ectopic expression of cavin-3 in these cells is sufficient to suppress their tumorigenesis in athymic mice ( Xu et al . , 2001; Lee et al . , 2011 ) . How cavin-3 expression suppresses tumorigenesis is not clear; however , forced over-expression of cavin-3 can induce G1 arrest and promote apoptosis ( Lee et al . , 2011 ) , suggesting that cavin-3 may suppress mitogenic signaling . Cavin-3 is one of four cavin family members , all of which are localized to caveolae ( Bastiani et al . , 2009 ) . Caveolae are invaginated , lipid-raft microdomains of the plasma membrane that may play roles in mitogenic signaling because a population of EGF , PDGF , and insulin receptors have been visualized by immuno-EM either in or adjacent to caveolae ( Liu et al . , 1996; Foti et al . , 2007; Nagy et al . , 2010 ) . Caveolae are not , however , required for cell proliferation because caveolae are not present in all cell types , caveolae do not appear until late in embryogenesis and animals with mutations that prevent caveolae formation are viable and of normal size ( Engelman et al . , 1998; Drab et al . , 2001; Razani et al . , 2001 , 2002; Fang et al . , 2006; Liu et al . , 2008 ) . Caveolae may instead limit cell proliferation because some mutations that prevent caveolae formation are associated with hyperplasia and fibrosis in lung ( Drab et al . , 2001; Razani et al . , 2001 ) . Cavins likely serve both structural and functional roles in caveolae through their interactions with the caveolin family of integral membrane proteins . Cavin-1 provides a structural function by binding to the principal caveolin , caveolin-1 , stabilizing the invaginated morphology of caveolae and providing an interaction surface for other cavins ( Hill et al . , 2008; Liu and Pilch , 2008; Bastiani et al . , 2009 ) . Cavin-2 and Cavin-4 show restricted expression and may serve cell-type specific functions ( Bastiani et al . , 2009; Hansen et al . , 2013 ) . By contrast , cavin-3 is broadly expressed and has been proposed to function in caveolae internalization ( McMahon et al . , 2009 ) . How cavin-3 might participate in cellular signaling is not clear . Here we show that cavin-3 dictates the balance between ERK and Akt signaling with consequences for cell metabolism , apoptosis and cell proliferation . We also characterize the molecular mechanisms by which cavin-3 influences cellular signaling . We employed two model systems to investigate whether loss of cavin-3 influences cell signaling . The first model system examined acute effects by depleting cavin-3 with siRNA in human SV589 fibroblasts over a time course of 15 days ( Figures 1–3 ) . The second model system examined chronic effects by knocking out the cavin-3 locus in the mouse and characterizing the consequences of loss of cavin-3 in embryonic fibroblasts ( Figure 4 ) . Because loss of cavin-3 is associated with cancer in multiple tissues and cancer is associated with elevated mitogenic signaling , we hypothesized that loss of cavin-3 might augment signaling in response to growth factors . 10 . 7554/eLife . 00905 . 003Figure 1 . Knockdown of cavin-3 suppresses mitogen-dependent ERK activation . ( A ) Loss of cavin-3 suppresses IE response . Human SV589 fibroblasts were mock treated ( 0 day ) or treated with cavin-3 siRNA for 2 , 7 or 14 days . Knockdown was maintained by splitting and re-transfecting cells with cavin-3 siRNA on days 5 , 9 , and 12 . Cells were then serum starved for 20 hr and RNA from cells was either harvested immediately or following treatment with 100 ng/ml EGF for 1 or 3 hr . Bars indicate the number of transcripts whose mean expression increased or decreased twofold following 1 hr , but not 3 hr , treatment . Solid and open bars indicate the number of transcripts common to transcripts induced or suppressed in the absence of knockdown . Hashed bars indicate knockdown-specific transcripts . Microarrays were performed in triplicate for 47 , 323 transcripts . Complete microarray data is provided ( Dryad: Michaely et al . , 2013 ) . ( B ) Leading edge analysis shows that loss of cavin-3 impairs ERK activation by EGF . Transcripts induced by 1 hr but not 3 hr of EGF stimulation were ordered based upon fold-induction using microarray data collected from cells without knockdown ( No KD ) . Fold-inductions for the top 15 transcripts are shown together with fold-inductions for the same transcripts in the 3-day , 8-day and 15-day knockdowns . All data are means ± SD , n = 3 . ( * ) indicates genes for which published data has identified transcriptional regulation by ERK ( Agarwal et al . , 1995; Gille et al . , 1995; Cohen , 1996; Ochsner et al . , 2003; Lin et al . , 2004; Hosokawa et al . , 2005; Stockhausen et al . , 2005; Bradley et al . , 2008 ) . ( C ) Cavin-3 knockdown impairs ERK translocation to the nucleus . Fibroblasts were mock treated ( No KD ) or treated with cavin-3 siRNA for 2 days , starved of serum for 20 hr , stimulated with EGF for 15 min , fixed and immunostained for total ERK . Nuclei were stained using DAPI . ( D ) Knockdown impairs cFos induction . Fibroblasts were mock treated ( No KD ) or treated with cavin-3 siRNA for 2 days , starved 20 hr for serum and stimulated with EGF for the indicated time . Cells were then lysed and immunoblotted for the indicated protein . ( E ) Loss of cavin-3 suppresses ERK activation by diverse mitogens . Fibroblasts were mock treated ( No KD ) or treated with cavin-3 siRNA , cultured for 2 days in serum , serum starved overnight and stimulated with the indicated mitogen . Cell lysates were immunoblotted for indicated proteins . DOI: http://dx . doi . org/10 . 7554/eLife . 00905 . 00310 . 7554/eLife . 00905 . 004Figure 2 . Knockdown of cavin-3 activates Akt . ( A ) Loss of cavin-3 first slows then accelerates cell proliferation . Fibroblasts were mock treated or treated with siRNA against cavin-3 and counted daily . Cells were re-treated and re-plated on day 5 . On day 9 , cells that had been treated with cavin-3 siRNA were either re-treated or allowed to recover from cavin-3 depletion ( Release ) . Doubling times ( td ) are indicated . ( B ) Protein profile of cavin-3 knockdown cells over time . SV589 fibroblasts were mock treated ( No KD ) or treated with cavin-3 siRNA . Cells were split and re-treated with siRNA on days 5 , 9 , and 12 . Cell lysates were prepared when indicated and immunoblotted for the indicated proteins . ( C ) Knockdown augments pAkt at the expense of pERK . pERK , ERK , pAkt , Akt and cavin-3 immunoblot staining was quantified by densitometry . Data were normalized to No KD ( 0 day ) and are means ± SEM , n = 3 . p<0 . 05 as compared to No KD . ( D ) Knockdown confers resistance to PD98059 and sensitivity to LY294002 . PD98059 interrupts the signaling pathway from growth factor receptors ( GFRs ) to ERK by inhibiting MEK . LY294002 interrupts the signaling pathway from GFRs to Akt by inhibiting the p110 subunit of PI3K . Cells were depleted of cavin-3 for the indicated number of days and treated with the indicated concentrations of PD98059 or LY294002 for 24 hr . Data are shown as a percentage of untreated and are means ± SEM , n = 6 . ( E ) Knockdown suppresses TNFα-dependent apoptosis . The indicated cells were treated overnight with 10 μg/ml cyclohexamide alone ( − ) or in combination with 10 ng/ml TNFα ( + ) and assayed for apoptotic cells by TUNEL staining ( top ) and PARP1 cleavage ( arrow , bottom ) . TUNEL data are means ± SEM , from three independent experiments . p<0 . 05 as compared to no TNFα . DOI: http://dx . doi . org/10 . 7554/eLife . 00905 . 00410 . 7554/eLife . 00905 . 005Figure 3 . Knockdown of cavin-3 increases transcription of biosynthetic genes and induces aerobic glycolysis . ( A ) Knockdown of cavin-3 progressively alters gene expression . RNA from SV589 fibroblasts grown in normal medium was harvested from mock-transfected cells or cells treated with cavin-3 siRNA for 3 , 8 , or 15 days . Transcripts that were increased or decreased twofold in triplicate microarrays are presented as Venn diagrams . Numbers indicate the number of transcripts common or unique to each knockdown . Complete microarray data is available at Dryad ( Michaely et al . , 2013 ) . ( B ) Prolonged knockdown augments many protein and nucleic acid biosynthetic components . The percent of gene transcripts with either >20% increase or >20% decrease over no knockdown in heat maps for protein and nucleic acid synthesis are plotted . Heat maps are provided in Figure 3—figure supplement 1 . ( C ) Knockdown increases fermentative glycolysis . Glucose consumption and lactate production were measured over 8 hr by colorimetric assay . Data are means ± SEM , n = 6 . p<0 . 05 as compared to No KD . H1299 cells serve as a positive control . DOI: http://dx . doi . org/10 . 7554/eLife . 00905 . 00510 . 7554/eLife . 00905 . 006Figure 3—figure supplement 1 . Heat maps of gene transcripts involved in protein and nucleic acid biosynthesis . Heat maps were generated using ratios of mean transcript levels in 3-day , 8-day and 15-day knockdowns relative to no knockdown and masking these ratios onto KEGG pathway gene sets for protein and nucleic acid biosynthetic pathways . Colors in the heat map correspond to transcript ratios as follows: <0 . 4 , dark blue; 0 . 4–0 . 56 , blue; 0 . 56–0 . 8 , cyan; 0 . 8–1 . 25 , green; 1 . 25–1 . 75 , yellow; 1 . 75–2 . 5 , orange; and >2 . 5 , red . Genes with multiple transcripts in the microarray were summed prior to ratio comparison . DOI: http://dx . doi . org/10 . 7554/eLife . 00905 . 00610 . 7554/eLife . 00905 . 007Figure 4 . MEFs from Cavin-3 KO animals recapitulate phenotypes observed following long-term knockdown in human fibroblasts . ( A ) Diagram of the targeting strategy used to generate germline knockout of the Prkcdbp ( cavin-3 ) gene . SA and LA indicate short arm and long arm regions of homology used for homologous recombination . Recombination replaced exon 1 , most of exon 2 and the intron between the two coding exons with the neomycin resistance cassette . ( B ) Protein profiles of MEFs show that Cavin-3 KO MEFs have changes in protein distribution with respect to WT MEFs that are similar to the changes observed in human fibroblasts following 15-day knockdown . ( C ) Quantification of pERK and pAkt changes show that Cavin-3 KO MEFs have fourfold more pAkt and 3 . 7-fold less pERK than WT MEFs . Data are means ± SEM , n = 3 . p<0 . 05 as compared to WT MEFs . ( D ) Cavin-3 KO MEFs are more resistant to PD98059 and more sensitive to LY294002 than WT MEFs . Data are means ± SEM , n = 6 . ( E ) Cavin-3 KO MEFs proliferate faster that WT MEFs . ( F ) Cavin-3 KO MEFs are more glycolytic than WT MEFs . Data are means ± SEM , n = 6 . p<0 . 05 as compared to WT MEFs . ( G ) Cavin-3 KO MEFs are more resistant to TNFα than WT MEFs . Arrow indicates cleaved PARP1 . TUNEL data are means ± SEM from three independent experiments . p<0 . 05 as compared to no TNFα . All assays were performed as in Figures 1–3 . DOI: http://dx . doi . org/10 . 7554/eLife . 00905 . 007 Contrary to expectation , microarrays showed that knockdown of cavin-3 suppressed the ability of EGF to induce most immediate early ( IE ) response genes ( Figure 1A ) . Suppressed transcripts fell into two categories: those that were fully suppressed by 3 days of knockdown and those that fell gradually during the 15-day time course . Many IE transcripts in the first group encoded proteins whose expression is driven by ERK ( Figure 1A , B ) . This observation suggested that loss of cavin-3 suppressed ERK signaling and 3 days of knockdown proved sufficient to inhibit EGF-dependent phosphorylation of ERK ( pERK ) , translocation of ERK to the nucleus and induction of the ERK-responsive transcription factor , cFos ( Figure 1C , D ) . The impact of cavin-3 depletion on ERK signaling was not specific to EGF because 3-day knockdown also impaired ERK activation by serum and diverse stimuli ( Figure 1E ) . Cavin-3 expression is thus necessary for efficient ERK activation . pERK is a potent driver of cell proliferation and knockdown of cavin-3 initially slowed cell growth; however , proliferation returned to a normal rate after 3 days and exceeded the normal rate after 10 days of knockdown ( Figure 2A ) . pERK levels did not recover during the time course ( Figure 2B ) , indicating that cavin-3 depleted cells compensated through other means . To identify the mechanism , we queried the IE microarray data and noted that the transcript whose expression showed the greatest suppression following cavin-3 depletion was early growth response protein 1 ( EGR1 ) ( Figure 1B ) . EGR1 is an ERK-induced transcription factor that drives expression of phosphatase and tensin homolog protein ( PTEN ) ( Yu et al . , 2011 ) . PTEN opposes the action of phosphatidylinositol 3-kinase ( PI3K ) by dephosphorylating phosphatidylinositol 3 , 4 , 5-trisphosphate ( PIP3 ) back to phosphatidylinositol 4 , 5-bisphosphate ( PIP2 ) . PIP3 generation by PI3K is necessary for the recruitment and activation of Akt ( Engelman et al . , 2006 ) . Knockdown of cavin-3 caused progressive loss of both EGR1 and PTEN . Loss of EGR1 and PTEN coincided with a fivefold increase in steady state levels of activated Akt ( pAkt ) ( Figure 2B , C ) . The switch from ERK to Akt signaling resulted in cell growth that was resistant to the MAPK/ERK Kinase ( MEK ) inhibitor , PD98059 , and more sensitive to the PI3K inhibitor , LY294002 ( Figure 2D ) . These findings show that loss of cavin-3 shifts cellular signaling to an Akt-dominated state . Activation of Akt can suppress apoptosis through inhibition of cytochrome c release from mitochondria and induction of inhibitor of apoptosis proteins ( IAPs ) ( Kennedy et al . , 1999; Papapetropoulos et al . , 2000 ) . To determine whether loss of cavin-3 suppresses apoptosis , we compared cell sensitivity to tumor necrosis factor-α ( TNFα ) , an apoptosis-inducing cytokine . Treatment of SV589 fibroblasts with TNFα potently induced apoptosis as evidenced by the robust cleavage of the caspase-3 target , Poly [ADP-Ribose] Polymerase 1 ( PARP1 ) , and the prevalence of TUNEL positive cells ( Figure 2E ) . Knockdown of cavin-3 progressively reduced cell sensitivity to TNFα in both assays . Resistance to TNFα correlated with induction of the IAP , survivin ( Figure 2B ) , one of several anti-apoptosis proteins induced by Akt ( Duronio , 2008; Guha and Altieri , 2009 ) . These findings show that loss of cavin-3 inhibits apoptosis . Akt can also promote cell proliferation through activation of the mammalian target of rapamycin complex 1 ( mTORC1 ) ( Manning and Cantley , 2007 ) . mTORC1 activates many biosynthetic pathways through phosphorylation and activation of S6K ( Duvel et al . , 2010 ) . Loss of cavin-3 potently increased levels of phosphorylated S6K ( pS6K ) ( Figure 2B ) , indicating that cavin-3 loss activates mTORC1 . Microarrays comparing transcript profiles cells depleted of cavin-3 for 0 , 3 , 8 , and 15 days showed that the rapid proliferation of 15-day knockdown cells was associated with substantial changes in gene transcription ( Figures 1A and 3A ) . Many of transcripts that were up-regulated in the 15-day knockdowns are part of protein and nucleic acid biosynthetic pathways ( Figure 3B ) . Cells fuel these biosynthetic pathways with metabolites derived from glycolysis . To increase metabolite levels , some rapidly dividing cells switch to aerobic glycolysis ( Warburg metabolism ) , a metabolic state that is characterized by increased use of fermentative glycolysis even under normoxic conditions ( DeBerardinis et al . , 2008 ) . Entry into aerobic glycolysis normally requires hypoxia induced factor 1 ( HIF1 ) , a transcription factor whose activity is controlled by the protein level of its α-subunit ( HIF1α ) ( Lunt and Vander Heiden , 2011 ) . Akt and mTORC1 increase HIF1α protein levels ( Schleicher et al . , 2009; Duvel et al . , 2010 ) and knockdown of cavin-3 progressively increased both HIF1α levels and the use of fermentative glycolysis ( Figures 2B and 3C ) . Glucose consumption and lactate production reached rates that were equivalent to those observed in H1299 cells , a non-small cell lung carcinoma cell line that lacks cavin-3 expression ( Xu et al . , 2001 ) . These findings show that prolonged loss of cavin-3 activates the Akt/mTORC1/HIF1 pathway and induces aerobic glycolysis . Activation of these pathways likely facilitates rapid cell proliferation . To test whether chronic loss of cavin-3 could recapitulate phenotypes observed in acute knockdown experiments , a germline knockout of the cavin-3 gene ( Prkcdbp ) was generated in mice ( Figure 4A ) . As observed with 15-day knockdowns in SV589 cells , embryonic fibroblasts ( MEFs ) from cavin-3 knockout ( Cavin-3 KO , Prkcdbp−/− ) animals displayed reduced levels of pERK , EGR1 and PTEN , elevated levels of survivin , pAkt , pS6K and HIF1α , increased rates of fermentative glycolysis , faster proliferation , heightened sensitivity to LY294002 , resistance to PD98059 and insensitivity to TNFα as compared to wild-type MEFs ( Figure 4B–G ) . Thus , genetic ablation of cavin-3 expression recapitulates phenotypes induced following long-term knockdown of cavin-3 by siRNA . To characterize the molecular mechanisms by which loss of cavin-3 alters cellular signaling , we first explored how cavin-3 facilitates ERK signaling . The loss of pERK but not pAkt that is associated with loss of cavin-3 ( Figures 2B and 4B ) suggested that cavin-3 acts downstream of mitogen receptors . Consistent with this conclusion , knockdown of cavin-3 did not inhibit EGF-dependent autophosphorylation of EGF receptors ( Figure 5A ) . Furthermore , inhibition of protein phosphatase 2A ( PP2A ) , which normally prevents the basal activity of mitogen receptors from activating downstream signaling cascades ( Wang et al . , 2003; Van Kanegan et al . , 2005 ) , was unable to activate ERK despite potent activation of Akt in cells depleted for cavin-3 ( Figure 5B ) . These findings indicate that loss of cavin-3 disrupts signal transduction coupling between mitogen receptors and ERK . 10 . 7554/eLife . 00905 . 008Figure 5 . Cavin-3 anchors caveolae to F-actin via myosin-1c , thereby positioning MEK and ERK for activation by mitogen receptors . ( A ) 3-day knockdown of cavin-3 cripples signal transduction to ERK , but has little effect on receptor autophosphorylation . Fibroblasts ( SV589 ) were mock treated or treated with cavin-3 siRNA , cultured in normal medium for 2 days , serum starved overnight , stimulated or not with 100 ng/ml EGF for 15 min , lysed and immunoblotted for the indicated protein . ( B ) Knockdown of cavin-3 prevents calyculin-A dependent activation of ERK . SV589 cells were treated or not with cavin-3 siRNA for 2 days , serum starved overnight , and treated or not with calyculin-A for 30 min followed by the addition of 100 nM insulin for the indicated times . Cell lysates were immunoblotted for the indicated protein . ( C ) 3-day knockdown of cavin-3 or myosin-1c redistributes caveolin-1 to the cell interior by immunofluorescence and reduces the abundance of morphological caveolae by thin section EM . Cells were treated or not with the indicated siRNA for 3 days and processed either for caveolin-1 immunofluorescence or thin section EM . Arrowheads indicate morphological caveolae . Quantification of caveolae abundance from 10 random fields was calculated as caveolae number per mm of plasma membrane length . Knockdown of cavin-3 or myosin-1 reduced caveolae abundance to similar extents ( 87% for cavin-3 siRNA and 90% for myosin-1c siRNA ) . ( D ) EGF-dependent activation of ERK requires cavin-3 , myosin-1c and F-actin . Untreated , 3-day cavin-3 knockdown , 3-day myosin-1c knockdown or 30 min latrunculin-A treated cells were induced or not with EGF , lysed and immunoblotted for pERK and ERK . ( E ) Myosin-1c co-localizes with caveolin-1 . Myosin-1c and caveolin-1 were localized by immunofluorescence . ( F ) Myosin-1c associates with cavin-3 . SV589 fibroblasts were lysed and immunoprecipitated ( IP ) with the indicated antibody . Immunoprecipitants were immunoblotted ( IB ) for cavin-3 and myosin-1 . ( G ) Co-fractionation of MEK and ERK with EGFR requires cavin-3 , myosin-1c and F-actin . Cell membranes from untreated , 3-day cavin-3 knockdown , 3-day myosin-1c knockdown and 30 min latrunculin-A treated cells were isolated , shattered by sonication , separated on 5–30% Iodixanol gradients , fractionated and immunoblotted for EGFR , MEK , ERK , and Caveolin-1 ( Cav-1 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 00905 . 008 Cavin-3 is localized in caveolae ( Bastiani et al . , 2009; McMahon et al . , 2009 ) , suggesting that the influence of cavin-3 on signaling involves caveolae . Caveolae have been implicated in both ERK and Akt signaling , though whether caveolae promote or suppress either pathway is not clear ( Liu et al . , 1996; Mineo et al . , 1996; Engelman et al . , 1998; Furuchi and Anderson , 1998; Teixeira et al . , 1999; Park et al . , 2000; Cohen et al . , 2003 ) . To determine whether cavin-3-dependent facilitation of ERK signaling involves caveolae , we used light and electron microscopy to test whether loss of cavin-3 influences caveolae abundance . We found that knockdown of cavin-3 both reduced caveolae abundance and re-localized the principal caveolae coat protein , caveolin-1 , to the Golgi region ( Figure 5C ) . The reduction in caveolae abundance correlated with loss of ERK responsiveness to EGF ( Figure 5D ) . These findings indicate that cavin-3 is necessary for surface caveolae and suggest that caveolae facilitate signal transduction to ERK . Caveolae associate with actin filaments at the plasma membrane ( Rohlich and Allison , 1976; Singer , 1979 ) and disruption of F-actin impaired mitogen-dependent stimulation of ERK to the same extent as cavin-3 depletion ( Figure 5D; Aplin and Juliano , 1999 ) , suggesting that cavin-3 may increase caveolae abundance by anchoring caveolae to F-actin at the cell surface . Cavin-3 binds to the caveolar coat ( Bastiani et al . , 2009; McMahon et al . , 2009 ) ; however , cavin-3 lacks a canonical actin-binding site . To determine whether cavin-3 associates with an actin-binding protein , we tested candidate actin-binding proteins that had previously been identified in a proteomic screen as lipid raft/caveolae proteins ( Foster et al . , 2003 ) for their ability to satisfy the following five criteria: ( i ) co-localization with caveolin-1 , ( ii ) reciprocal co-immunoprecipitation with cavin-3 , ( iii ) knockdown that mislocalizes caveolin-1 , ( iv ) knockdown that reduces surface caveolae and ( v ) knockdown that impairs ERK activation . Myosin-1c satisfied all five criteria ( Figure 5C–F ) , indicating that myosin-1c participates in a cavin-3 linkage that anchors caveolae to peripheral actin . To determine how caveolae might participate in ERK signaling , we reasoned that the role of caveolae should be downstream of ras because ras participates in both ERK and Akt activation . MEK and ERK have been shown to co-fractionate with mitogen receptors and caveolin-1 in cellular membranes of low protein-to-lipid density ( Liu et al . , 1997 ) . Consistent with these published observations , we found that EGFR , caveolin-1 , MEK and ERK co-fractionated in low-density membrane fractions isolated from untreated SV589 fibroblasts; however , knockdown of cavin-3 , knockdown of myosin-1c or brief treatment with the actin-depolymerization compound , latrunculin-A , caused MEK and ERK to accumulate in the high-density fractions , which contain the bulk of cellular membrane protein ( Figure 5G ) . The distribution of EGFR and caveolin-1 in the gradients became broader and heavier , but the majority of both EGFR and caveolin-1 remained in lighter fractions as compared to MEK and ERK . The EGFR remained on the plasma membrane because EGF treatment induced autophosphorylation of EGFRs in the absence of cavin-3 ( Figure 5A ) . These observations indicate that a cytoskeletal linkage involving cavin-3 , myosin-1c and F-actin positions MEK and ERK for activation by mitogen receptors , most likely by anchoring caveolae at the cell surface . Cavin-3 associates with caveolae as part of a complex with caveolin-1 and other cavins ( Bastiani et al . , 2009; McMahon et al . , 2009 ) , suggesting that caveolin-1 and other cavins may participate in the cavin-3 linkage . Mammalian genomes encode four cavins of which two ( cavin-1 and cavin-3 ) are readily detectable in fibroblasts grown under normal culture conditions . We tested whether cavin-1 and caveolin-1 are components of the cavin-3 linkage using immunoprecipitation with antibodies to myosin-1c and found that in addition to cavin-3 both cavin-1 and caveolin-1 co-precipitated with myosin-1c ( Figure 6A ) . To test whether cavin-1 and caveolin-1 are necessary for the signaling function of the cavin-3 linkage , pERK and pAkt levels were compared in fibroblasts treated with siRNA against cavin-1 , cavin-3 , caveolin-1 or myosin-1c . Like the cavin-3 and myosin-1c knockdowns , knockdowns of caveolin-1 and cavin-1 increased pAkt at the expense of pERK ( Figure 6B , C ) . However , unlike the myosin-1c knockdown , knockdown of cavin-1 reduced the protein level of cavin-3 , while knockdown of caveolin-1 reduced protein levels of both cavin-1 and cavin-3 . To test whether the effects of cavin-1 and caveolin-1 knockdowns on pERK and pAkt depend upon changes in cavin-3 protein levels , we over-expressed cavin-3 in SV589 fibroblasts ( Figure 6D ) and repeated the four knockdowns . Cavin-3 over-expression partially protected cavin-3 levels from knockdown of cavin-1 , cavin-3 or caveolin-1 . This protection muted the effects of these knockdowns on pERK and pAkt levels ( Figure 6E , F ) . Importantly , cavin-3 levels correlated better with pERK/ERK and pAkt/Akt ratios than did levels of either cavin-1 or caveolin-1 ( Figure 6C , F , G ) . The strength of this correlation suggests that cavin-3 is the limiting component of the cavin/caveolin complex for caveolar influence on ERK and Akt signaling . Both caveolin-1 and cavin-1 are required for normal abundance of caveolae ( Drab et al . , 2001; Razani et al . , 2001; Hill et al . , 2008; Liu et al . , 2008 ) . Caveolin-1 is an integral membrane protein that polymerizes to form filaments that coat the cytosolic surface of caveolae ( Fernandez et al . , 2002 ) . Cavin-1 has been proposed to serve as an adaptor that links other cavins to the caveolin coat ( Bastiani et al . , 2009 ) . We propose that cavin-3 promotes efficient signal transduction to ERK by bridging between the cavin-1/caveolin-1 complex and myosin-1c ( Figure 6H ) . 10 . 7554/eLife . 00905 . 009Figure 6 . Linkage function requires myosin-1c , cavin-1 , cavin-3 , and caveolin-1 . ( A ) The cavin-3 linkage involves caveolin-1 , cavin-1 , and myosin-1c . SV589 fibroblasts were lysed , immunoprecipitated ( IP ) with the indicated antibody and immunoblotted ( IB ) for the indicated protein . ( B ) Knockdown of cavin-1 , cavin-3 , caveolin-1 or myosin-1c suppresses pERK and augments pAkt levels . SV589 fibroblasts were treated with siRNA against cavin-1 , cavin-3 , myosin-1c or caveolin-1 for 3 days and immunoblotted for the indicated proteins . ( C ) Quantification of the effects of knockdowns on pERK/ERK and pAkt/Akt levels . Data are means ± SEM , n = 3 . p<0 . 05 as compared to No KD . ( D ) Stable over-expression of cavin-3 in SV589 fibroblasts ( S/Cavin-3 cells ) increases cavin-3 levels 2 . 5-fold over parental SV589 cells . ( E ) Over-expression of cavin-3 mutes effects of cavin-1 , cavin-3 , and caveolin-1 siRNAs on pERK and pAkt levels . S/Cavin-3 cells were treated with siRNA against cavin-1 , cavin-3 , caveolin-1 or myosin-1c for 3 days and blotted for the indicated proteins . The exposure of the cavin-3 blot was selected based upon similarity of the No KD controls in panels B and E . ( F ) Quantification of the effects of knockdowns on pERK/ERK and pAkt/Akt levels . Data are means ± SEM , n = 3 . p<0 . 05 as compared to No KD . ( G ) The mean values for pERK/ERK and pAkt/Akt for cavin-1 , cavin-3 , caveolin-1 , and myosin-1c knockdowns from panels C and F were plotted against cavin-3 protein levels . Linear regression was performed on the six pERK/ERK data points and six pAkt/Akt data points for the cavin-1 , cavin-3 , and caveolin-1 knockdowns . R2 values for the pERK/ERK and pAkt/Akt lines are 0 . 937 and 0 . 930 , respectively . ( H ) Model of the cavin-3 linkage between caveolae and F-actin in the context of the signaling pathways leading to ERK and Akt activation . DOI: http://dx . doi . org/10 . 7554/eLife . 00905 . 009 Loss of cavin-3 is common in cancer cells ( Xu et al . , 2001; Zochbauer-Muller et al . , 2005; Lee et al . , 2008; Martinez et al . , 2009; Tong et al . , 2010; Caren et al . , 2011; Lee et al . , 2011 ) and many cancer cells show elevated Akt/mTORC1 signaling , aerobic glycolysis and resistance to apoptosis ( Manning and Cantley , 2007; DeBerardinis et al . , 2008; Duronio , 2008 ) . To test whether loss of the cavin-3 linkage is necessary for the altered cell signaling , metabolism and apoptosis phenotypes of cancer cells we tested whether reconstitution of the cavin-3 linkage was sufficient to normalize ERK/Akt signaling , cell metabolism and apoptosis . To facilitate this reconstitution , we looked for a cancer cell line that lacks cavin-3 , but expresses normal levels of all other linkage components . While many cancer cell lines lack cavin-3 ( Xu et al . , 2001; Bastiani et al . , 2009; Lee et al . , 2011 ) , some lines such as PC-3 lacked additional linkage components ( Figure 7A; Bastiani et al . , 2009 ) . H1299 cells are a line of non-small cell lung carcinoma cells that lacked detectable cavin-3 , but exhibited levels of cavin-1 , caveolin-1 , myosin-1c and actin that were similar to the endogenous levels of SV589 fibroblasts ( Figure 7A ) . Comparison of the pERK and pAkt responses to mitogens showed that H1299 cells had pERK and pAkt responses that were similar to SV589 fibroblasts following cavin-3 knockdown ( Figure 7—figure supplement 1 ) . To reconstitute cavin-3 in H1299 cells , retroviral vectors were used to stably express either GFP alone ( H/GFP ) or GFP and cavin-3 ( H/Cavin-3 ) . Use of a weak promoter-expression system generated levels of cavin-3 expression in H/Cavin-3 cells that were similar to endogenous levels expressed in SV589 fibroblasts ( Figure 7B ) . Re-expression of cavin-3 restored the cavin-3 linkage as evidenced by the ability of myosin-1c antibody to co-precipitate cavin-1 , cavin-3 , and myosin-1c from lysates of H/Cavin-3 cells , but not H/GFP cells ( Figure 7C ) . Restoration of the cavin-3 linkage resulted in a 7 . 6-fold increase in surface caveolae and redistribution of caveolin-1 to the plasma membrane ( Figure 7D ) . As compared to parental H1299 or H/GFP cells , H/Cavin-3 cells displayed higher levels of pERK and EGR1 ( Figure 7B , E ) ; lower levels of pAkt , HIF1α and pS6K ( Figure 7B , E ) ; and increased sensitivity to PD98059 and resistance to LY294002 ( Figure 7F ) . These signaling changes correlated with reductions in the rate of cell growth , glucose uptake and lactate production ( Figure 7G , H ) . Cavin-3 re-expression also suppressed survivin levels and sensitized cells to TNFα-dependent apoptosis ( Figure 7B , I ) . These findings show that restoration of the cavin-3 linkage in cancer cells can normalize ERK/Akt signaling , cell metabolism and apoptosis sensitivity . 10 . 7554/eLife . 00905 . 010Figure 7 . Stable expression of cavin-3 normalizes multiple phenotypes in cancer cells . ( A ) Comparison of cavin-1 , cavin-3 , myosin-1c , and caveolin-1 in SV589 fibroblasts ( Fibroblast ) , PC-3 cells and H1299 cells . Representative immunoblots of the indicated proteins are shown . Comparison of pERK and pAkt responses to mitogen stimulation is provided in Supplement 1 . ( B ) Expression of cavin-3 in H1299 cells reverts cellular signaling . Cell lysates from human SV589 fibroblasts ( Fibroblast ) , H1299 cells , H1299 stably expressing cavin-3 ( H/Cavin-3 cells ) and H1299 stably expressing GFP ( H/GFP cells ) were immunoblotted for the indicated proteins . ( C ) Association of myosin-1c with cavins and caveolin-1 requires cavin-3 . The indicated antibodies ( IP ) were used for immunoprecipitation from the indicated cells and immunoblotted for the indicated proteins ( IB ) . ( D ) Cavin-3 expression increases caveolin-1 staining at the plasma membrane and increases the abundance of surface caveolae by 7 . 6-fold . ( E ) Quantification of changes in pERK/ERK and pAkt/Akt shows that expression of cavin-3 normalizes pERK and pAkt levels . Densitometry was performed on three replicates of the immunoblots from panel B . Data are means ± SEM . p<0 . 05 as compared to SV589 fibroblasts . ( F ) Cavin-3 expression increases sensitivity to PD98059 and decreases sensitivity to LY294002 . Data are means ± SEM , n = 6 . ( G ) Cavin-3 expression decreases proliferation rate . ( H ) Cavin-3 expression suppresses glycolysis . Data are means ± SEM , n = 6 . p<0 . 05 as compared to SV589 fibroblasts . ( I ) Cavin-3 expression sensitizes cells to TNFα . Arrow indicates cleaved PARP1 . TUNEL data are means ± SEM from three independent experiments . p<0 . 05 as compared to no TNFα . All assays were performed as in Figures 1–3 . DOI: http://dx . doi . org/10 . 7554/eLife . 00905 . 01010 . 7554/eLife . 00905 . 011Figure 7—figure supplement 1 . pERK and pAkt responses to EGF or Insulin . SV589 fibroblasts were mock transfected or transfected with cavin-3 siRNA . Fibroblasts , PC-3 and H1299 cells were then cultured 2 days in serum , serum starved overnight and stimulated with either 100 nM insulin or 100 ng/ml EGF for the indicated times . Cells were lysed , run on SDS-PAGE gels and immunoblotted for pAkt , Akt , pERK , and ERK . DOI: http://dx . doi . org/10 . 7554/eLife . 00905 . 011 Loss of EGR1 and PTEN paralleled the increase in pAkt ( Figures 2B and 4B ) , suggesting that loss of cavin-3 drives Akt activation through loss of the pERK-EGR1-PTEN axis . To test this hypothesis , we examined whether ectopic expression of EGR1 could suppress pAkt in the absence of cavin-3 . Retroviruses were used to stably express ( i ) GFP alone , ( ii ) GFP and EGR1 or ( iii ) GFP and cavin-3 in Prkcdbp−/− ( Cavin-3 KO ) MEFs and H1299 cells . Expression of EGR1 in either Cavin-3 KO MEFs or H1299 cells suppressed pAkt levels to the same extent as expression of cavin-3 ( Figure 8A , B ) . These observations indicate that EGR1 acts downstream of cavin-3 to suppress Akt activation . Interestingly , while expression of either EGR1 or cavin-3 restored PTEN expression to a normal level in Cavin-3 KO MEFs , expression of neither EGR1 nor cavin-3 substantially improved PTEN expression in H1299 cells despite potent suppression of pAkt . The PTEN promoter in H1299 cells is hypermethylated ( Soria et al . , 2002 ) and this methylation may limit the ability of EGR1 to drive PTEN expression . The ability of cavin-3 and EGR1 to nonetheless suppress Akt activation indicates that EGR1 suppresses Akt activation through mechanisms that are independent of PTEN protein level . 10 . 7554/eLife . 00905 . 012Figure 8 . Loss of cavin-3 promotes Akt activation through loss of EGR1 . ( A ) EGR1 expression is sufficient to suppress the Akt/mTORC1/HIF1α pathway . Immunoblotting of the indicated proteins was used to compare protein profiles of WT MEFs to Cavin-3 KO MEFs stably expressing GFP alone ( KO/GFP ) , GFP and EGR1 ( KO/EGR1 ) or GFP and cavin-3 ( KO/Cavin-3 ) and human SV589 fibroblasts to H1299 cells stably expressing GFP alone ( H/GFP ) , GFP and EGR1 ( H/EGR1 ) or GFP and cavin-3 ( H/Cavin-3 ) . ( B ) Quantification of pERK/ERK and pAkt/Akt levels show that expression of EGR1 normalizes pAkt levels , but not pERK levels , in cavin-3 deficient cells . Data are means ± SEM , n = 3 . p<0 . 05 as compared to either WT MEFs ( WT ) or SV589 fibroblasts ( Fibroblast ) . ( C ) Expression of EGR1 is sufficient to suppress aerobic glycolysis . Glucose uptake and lactate production data are means ± SEM , n = 6 . p<0 . 05 relative to WT MEF ( WT ) or SV589 fibroblast ( Fibroblast ) controls . ( D ) Expression of EGR1 is not sufficient to normalize TNFα-induced apoptosis . Arrow indicates cleaved PARP1 . TUNEL data are means ± SEM , n = 3 experiments . p<0 . 05 relative to cells not treated with TNFα . ( E ) Expression of EGR1 is not sufficient to normalize caveolin-1 distribution . Indicated cells were processed for caveolin-1 immunofluorescence . All assays were performed as in Figures 1–3 . DOI: http://dx . doi . org/10 . 7554/eLife . 00905 . 012 Expression of EGR1 was sufficient to suppress aerobic glycolysis in both Cavin-3 KO MEFs and H1299 cells ( Figure 8 ) . EGR1 expression suppressed both pS6K and HIF1α levels in both cell lines ( Figure 8A ) and loss of HIF1α correlated with reductions in glucose consumption and lactate production ( Figure 8C ) . Akt and mTORC1 induce HIF1α and the ability of EGR1 to suppress Akt activation indicates that loss of cavin-3 induces aerobic glycolysis via loss of EGR1-dependent suppression of the Akt/mTORC1/HIF1 pathway . In contrast to the effects of EGR1 on cell metabolism , only cavin-3 re-expression was able to rescue sensitivity to TNFα ( Figure 8D ) , indicating that cavin-3 supports an EGR1-independent process that is necessary for TNFα-sensitivity . Expression of EGR1 also did not restore pERK levels ( Figure 8A , B ) or drive caveolin-1 to the plasma membrane ( Figure 8E ) . Active ERK facilitates apoptosis through both the intrinsic and extrinsic pathways ( Cagnol and Chambard , 2010 ) and the ability of cavin-3 to support normal apoptosis sensitivity may require both the EGR1-dependent reduction in pAkt and a caveolae-dependent increase in pERK . Together , these findings show that cavin-3 activates at least two processes: ( i ) an EGR1-dependent process that suppresses the Akt/mTORC1/HIF1 pathway; and ( ii ) an EGR1-independent process that is necessary for normal apoptosis . The signaling changes that were observed in cell culture following loss of cavin-3 were also observed in vivo . Lung tissue from Prkcdbp−/− ( Cavin-3 KO ) animals showed decreased pERK , EGR1 , and PTEN levels and increased pAkt and HIF1α levels as compared to lung tissue from wild-type animals ( Figure 9A , B ) . The elevated HIF1α of Cavin-3 KO lung tissue was associated with increased fermentative glycolysis ex vivo ( Figure 9C ) . Thus , in vivo loss of cavin-3 promoted Akt signaling at the expense of ERK and increased glycolytic metabolism . However , these signaling changes were not associated with developmental defects , as would be expected if apoptosis were compromised , or hyperplasia , as would be expected if cell proliferation were augmented . Cavin-3 KO mice did have shorter lifespan than control animals ( Figure 9D ) and the principal cause of death was cachexia as exemplified by a 40% reduction in body weight and severe lipodystrophy ( Figure 9E , F ) . Lipodystrophy is frequently associated with hepatic steatosis ( Huang-Doran et al . , 2010 ) and areas of steatosis were observed in livers of Cavin-3 KO animals ( Figure 9G ) . Despite the strong association of lung cancer with loss of cavin-3 expression ( Zochbauer-Muller et al . , 2005 ) , we did not observe lung cancers and saw no differences in lung structure or alveolar density ( Figure 9G , H ) . Masson’s trichrome stain of lung sections also did not show differences in collagen fiber content , as would be expected for fibrosis ( data not shown ) . A survey of additional tissues by H&E staining did not show differences between normal and Cavin-3 KO mice at either 4 months or 2 years of age ( Figure 9G and data not shown ) . Thus , genetic ablation of cavin-3 expression increases Akt signaling at the expense of ERK , increases the use of fermentative glycolysis in tissues and causes cachexia , but is not sufficient to cause substantial de novo tumorigenesis . 10 . 7554/eLife . 00905 . 013Figure 9 . Cavin-3 KO animals have shortened lifespan resulting from late onset cachexia . ( A ) Protein profiles of lung tissue from 6-week old animals . ( B ) Quantification of pERK/ERK and pAkt/Akt in lung tissue from 6-week old animals show that loss of cavin-3 increases pAkt levels and decreases pERK levels by twofold . Data are means ± SEM , n = 6 ( three males and three females ) . ( C ) Lung tissue from Cavin-3 KO animals is more glycolytic than normal . Lung tissue from 6-week old animals was excised and assayed for glucose consumption and lactate production over 4 hr in ex vivo culture . Data are means ± SEM , n = 6 ( three males and three females ) . ( D ) Kaplan-Meier plot showing that Cavin-3 KO mice have decreased lifespan ( n = 12 ) . ( E ) Cavin-3 KO animals have a 40% reduction in body mass . *p<0 . 05 as compared to WT animals . ( F ) Cavin-3 KO animals have lipodystrophy . Shown are dissected abdominal fat pads . ( G ) H&E staining of abdominal fat pad , lung , liver , and small intestine . ( H ) Lung tissue from Cavin-3 KO animals does not show hyperplasia . Slides of WT and Cavin-3 KO lung sections were coded and imaged ( four fields from six sections for WT and four fields from four sections for Cavin-3 KO ) by a blinded observer . Mean linear intercepts of coded images were measured by a blinded observer . Data shown are means ± SD . p value is 0 . 6006 . DOI: http://dx . doi . org/10 . 7554/eLife . 00905 . 013 The major finding of this study is that the tumor suppressor protein , cavin-3 , controls the balance between ERK and Akt signaling with consequences for cell proliferation , metabolism , and apoptosis ( Figures 1–4 , 7 ) . Cavin-3 promotes ERK signaling by anchoring the ERK activation module of caveolae to F-actin at the plasma membrane ( Figures 5 and 6 ) and suppresses Akt signaling by promoting EGR1 and PTEN expression ( Figures 7 and 8 ) . The in vitro consequences of loss of cavin-3 include induction of Warburg metabolism , faster cell proliferation and resistance to apoptosis ( Figures 1–4 , 7 ) . The in vivo consequences of loss of cavin-3 include elevated use of glycolysis and cachexia ( Figure 9 ) . The cavin-3 dependent coupling of caveolae to F-actin is analogous to linkages that generate other specialized domains of the plasma membrane . Peripheral F-actin is crosslinked by spectrins into a viscoelastic network ( the membrane skeleton ) that associates with the plasma membrane through diverse linkages . The spatial organization of specific linkages allows the membrane skeleton to generate specializations within the plasma membrane and loss of different organizing linkages results in loss of specific specialized domains ( Bennett and Healy , 2008 ) . Our findings show that cavin-3 is part of a linkage necessary for the caveolae specialization ( Figures 5–7 ) . Intriguingly , insulin receptors also associate with F-actin and are enriched at the border between caveolae and the rest of the plasma membrane ( Foti et al . , 2007 ) . The combination of a receptor-actin linkage with the cavin-3 linkage suggests that the skeleton assembles mitogen receptors with the caveolar ERK activation module to facilitate signal transduction coupling ( Figure 6H ) . Other adaptors may link additional downstream signaling modules to the skeleton and different assemblies of receptors and signaling modules may dictate temporal , spatial , and cell-specific responses to common environmental cues . Skeleton-dependent integration of signaling pathways may promote cell differentiation because the sophistication of the skeleton correlates with the degree of cellular differentiation ( Bennett and Healy , 2008 ) and defects in the skeleton are associated with dedifferentiation , increased cell proliferation and resistance to anoikis ( Mishra et al . , 2005; Kumar et al . , 2011 ) . Whereas most linkages to the membrane skeleton are static , the cavin-3 linkage involves the motor molecule , myosin-1c , and the use of this motor may contribute to caveolae function in at least two ways . First , myosin-1c participates in the transport of lipid raft components ( Bond et al . , 2013 ) and the use of myosin-1c may couple delivery of caveolar components with caveolae anchorage . Second , caveolae are endocytic structures and myosin-1c may control movement of caveolae vesicles within the cortical actin network . The majority of caveolae-derived vesicles remain at the cell periphery where they re-fuse with the plasma membrane in a cyclic process referred to as potocytosis ( Anderson et al . , 1992; Boucrot et al . , 2011 ) . Myosin-1c may drive caveolae-derived vesicles back to the cell surface in a calcium-regulated manner because calcium channels are enriched in caveolae ( Isshiki et al . , 1998; Pani and Singh , 2009 ) , the Donnan effects generated by endocytosis promote calcium channel opening ( Saito et al . , 2007 ) and calcium promotes both myosin-1c motor activity ( Barylko et al . , 1992; Zhu et al . , 1998 ) and recycling of caveolae-derived vesicles back to the cell surface ( Lin et al . , 2012 ) . Potocytosis of caveolae may regulate the strength of signal transduction from surface receptors to ERK or allow ERK signal transduction to be moved to different sites along the plasma membrane . Importantly , the effects of cavin-3 on ERK signaling are matched by inverse changes in Akt signaling . EGR1 can participate in a reciprocal relationship between ERK and Akt ( Yu et al . , 2011 ) and our findings show that cavin-3 suppresses pAkt via EGR1 ( Figure 8 ) . EGR1 induces PTEN expression ( Baron et al . , 2006 ) ; however , our findings show that induction of PTEN expression is not the only mechanism by which EGR1 suppresses Akt activation . How EGR1 suppresses Akt activation is currently under investigation . Alterations in cavin-3 protein levels have proportionate effects on pERK and pAkt levels ( Figure 6G ) and the tightness of this correlation may be the result of positive and negative feedback loops involving EGR1 and cavin-3 . EGR1 can promote ERK activation ( Shen et al . , 2011 ) and ERK activation drives EGR1 expression ( Cohen , 1996 ) . This positive feedback loop requires cavin-3 because both pERK and EGR1 expression are suppressed in the absence of cavin-3 ( Figures 1–4 , 7 ) . Cavin-3 and EGR1 also act within a negative feedback loop because EGR1 down-regulates cavin-1 and caveolin-1 expression ( Joshi et al . , 2012 ) and loss of either cavin-1 or caveolin-1 suppresses cavin-3 expression ( Figure 6 ) . Regulation within these feedback loops may set the balance point between pERK and pAkt . Interestingly , modulation of cellular phenotypes in response to changes in this balance point involves substantial hysteresis . Treatment with cavin-3 siRNA reduces cavin-3 protein levels by more than 80% within 3-days; however , the phenotypic effects of cavin-3 depletion require 10–15 days to manifest fully ( Figures 1–3 ) . Many of these phenotypic effects involve extensive changes in gene expression ( Figures 1A and 3A ) . While the extent of these cellular changes may require some time to complete , our data suggest that more active processes are responsible for the hysteresis . EGR1 protein levels fall slowly during the 15-day time course , despite rapid loss of pERK ( Figure 2B ) and the short half-life of EGR1 protein ( Waters et al . , 1990 ) . EGR1 expression can be supported by means other than ERK ( Hallahan et al . , 1991; Guillemot et al . , 2001 ) and these mechanisms may slow the loss of EGR1 . The gradual loss of EGR1 coincides with gradual increases in survivin , pS6K and HIF1α ( Figure 2B ) . The slowly evolving changes in these and other factors likely dictate the time-dependence for the manifestations of resistance to apoptosis , aerobic glycolysis and acceleration in cell proliferation . The length of time required to reach the final phenotypic state also provides a general note of caution with respect to knockdown experiments . siRNA knockdowns are typically assayed 3 days post-transfection; however , our data show that this time window may capture an intermediate state that differs substantially from chronic loss-of-function . Characterization of intermediate states may improve understanding of how proteins such as cavin-3 impact the panoply of cellular processes . The ability of cavin-3 to influence cell signaling , proliferation , metabolism , and apoptosis provides explanations for how cavin-3 functions as a tumor suppressor; however , Cavin-3 KO animals do not show substantial increases in spontaneous cancers . These observations imply that loss of cavin-3 is not sufficient for tumorigenesis . Consistent with this conclusion , loss of cavin-3 expression is more prevalent in late-stage/high-grade cancers than in early-stage/low-grade cancers ( Lee et al . , 2008; Caren et al . , 2011; Wikman et al . , 2012 ) . A potential clue as to the role of cavin-3 in cancer comes from the observation that 35–55% of glial , lung , gastric , ovarian , breast , and colorectal cancers show hypermethylation in their cavin-3 promoters ( Xu et al . , 2001; Zochbauer-Muller et al . , 2005; Lee et al . , 2008; Martinez et al . , 2009; Tong et al . , 2010; Lee et al . , 2011 ) . Methylation of the cavin-3 promoter also occurs normally in trophoblasts when they invade the endometrium during pregnancy ( Grigoriu et al . , 2011 ) . Trophoblast invasion shares many features with cancer cell metastasis ( Ferretti et al . , 2007 ) and the degree of methylation in the cavin-3 promoter correlates with both trophoblast invasion potential and cancer metastasis ( Wikman et al . , 2012; van Dijk et al . , 2012 ) . The accelerated cell proliferation , induction of Warburg metabolism and resistance to apoptosis that results from loss of cavin-3 may facilitate the ability of invading cells to survive and proliferate in new environments and may thus provide strong selection for loss of cavin-3 function in cancer cells . Strong selection pressure for loss of cavin-3 in cancer cells is suggested by the observation that while 41% of primary non-small cell lung carcinomas show methylation of their cavin-3 promoters , 81% of these carcinomas ( N = 93 ) lack detectable cavin-3 expression by immunohistochemistry ( Zochbauer-Muller et al . , 2005 ) . Lung cancers are not commonly detected until late in disease progression and loss of cavin-3 may facilitate stage progression to metastatic disease . The most apparent defect in Cavin-3 KO animals is cachexia as evidenced by a 40% reduction in weight and severe lipodystrophy ( Figure 9 ) . Lipodystrophies have also been noted in humans and animals lacking either cavin-1 or caveolin-1 ( Cao et al . , 2008; Kim et al . , 2008; Liu et al . , 2008; Hayashi et al . , 2009; Asterholm et al . , 2012 ) and the association of cavin-3 with cavin-1 and caveolin-1 ( Bastiani et al . , 2009 ) combined with the dependence of cavin-3 protein on cavin-1 and caveolin-1 ( Figure 6 ) suggests that these lipodystrophies are caused by a common mechanism . Loss of cavin-3 linkage components may cause lipodystrophy through selective death in adipocytes ( Martin et al . , 2012 ) ; however , lipid mobilizing factors are elevated in the circulation of Cavin-3 KO animals ( data not shown ) , suggesting that lipolysis is responsible for the loss of triglyceride stores . Cavin-3 KO animals exhibit increased use of fermentative glycolysis ( Figure 9 ) and this increase may promote lipolysis for the purpose of clearing lactate . Glycolysis generates lactate and hepatocytes convert lactate back to glucose through the Cori cycle , which fuels the necessary gluconeogenesis via oxidative phosphorylation of fatty acids . Loss of caveolae increases flux through the Cori cycle as evidenced by the increased rates of lactate production , hepatic gluconeogenesis and adipocyte lipolysis in Cav1−/− ( Caveolin-1 KO ) mice ( Asterholm et al . , 2012 ) . Whole body lactate production increases with age ( Wallace , 2005 ) and loss of cavin-3 may exasperate lactate production to a point after which fatty acid demand for gluconeogenesis triggers lipolysis of triglycerides stored in adipose tissue . As lactate production continues to increase , host responses to lactate may drive cannibalization of protein from muscle , which in the absence of cancer is the source of most lactate production . Cancers that commonly elicit cachexia include lung , breast , and colorectal cancers ( Fox et al . , 2009 ) and these cancers frequently lack cavin-3 expression ( Xu et al . , 2001; Zochbauer-Muller et al . , 2005; Lee et al . , 2011 ) . Thus , the absence of cavin-3 in tumors may predispose patients to the development of cancer-associated cachexia , a condition that is the immediate cause of death for more than 20% of all cancer patients ( Tisdale , 2002 ) . Recent work has shown that caveolae of different tissues and cell types have different cavin compositions ( Bastiani et al . , 2009; Hansen et al . , 2013 ) . The fibroblasts and epithelial cells examined here express cavin-1 and cavin-3 , but lack cavin-2 and cavin-4 . By contrast , adipocytes abundantly express cavin-2 , myocytes have abundant cavin-4 and endothelial cells have different compositions depending upon tissue localization ( Stan et al . , 1999; Ogata et al . , 2008; Bastiani et al . , 2009; Hansen et al . , 2013 ) . Use of different cavins may provide alternative linkages that can support surface caveolae or provide novel functions for caveolae in different cell types . Many functions have been ascribed to caveolae including mitogen signaling , mechanosensing , nitric oxide signaling , endocytosis , and transcytosis ( Boscher and Nabi , 2012; Kiss , 2012; Mineo and Shaul , 2012; Nassoy and Lamaze , 2012; Parton and del Pozo , 2013 ) . Different cavin compositions may support different functions in different cellular settings . We show here that cavin-3 plays a critical role in the signal transduction function of caveolae and that cells , which normally express cavin-3 , rely upon cavin-3 for normal ERK and Akt signaling with consequences for cell metabolism , apoptosis , and cell proliferation . The following antibodies were used in this study . Anti-EGFR; Anti-phospho-EGFR ( pY1068 ) ; Anti-phospho-Erk1/2 ( pT202/pY204 ) ; Anti-PTEN; Anti-phospho-Akt1 ( pS473 ) ; Anti-Akt1; Anti-MEK1/2; Anti-survivin; Anti-PARP1; and Anti-c-Fos , ( all from Cell Signaling , Danvers , MA ) ; Anti-caveolin-1 ( BD Transduction Laboratories , San Jose , CA ) ; Anti- ( human cavin-3 ) ( Bethyl Labs , Montgomery , TX ) ; Anti- ( mouse cavin-3 ) ( Proteintech Group , Inc . , Chicago , IL ) ; Anti-ERK1/2 ( Millipore , Billerica , MA ) ; Anti-HIF1α ( Bethyl Labs ) ; Anti-Cavin-1 ( AbCam , Cambridge , MA ) ; Anti-Myo1c ( Santa Cruz Biotechnology , Santa Cruz , CA ) ; Anti-Tubulin; and Anti-Actin ( Sigma-Aldrich , St . Louis , MO ) . Epidermal growth factor ( MP Biomedicals ) ; platelet-derived growth factor BB ( Millipore ) ; 12-O-Tetradecanoylphorbol-13-Acetate ( Cell Signaling ) ; Lysophosphatidic acid ( Sigma ) ; Recombinant human insulin ( Sigma ) ; tumor necrosis factor alpha ( Cell Signaling ) Latrunculin-A ( Sigma ) ; Glucose Assay Kit ( Sigma ) ; L-Lactate Assay Kit ( Megazyme International , Wicklow , Ireland ) . Clean-Blot IP Detection Kit ( Pierce Biochemicals , Rockford , IL ) were purchased . siRNA against cavin-3 was from Dharmacon RNAi technologies ( Thermo Scientific , Pittsburgh , PA ) and consists of an equal molar mixture of the following three oligos: 5′-UGGCCAAGGCGGAGCGCGU , 5′-GCGGGAAGCUCCACGUUCU and 5′-GCACCGGAUUGCAGAAGGU . Each of the three oligos was individually active against cavin-3 . siRNA against cavin-1 ( sc-76293 ) , myosin-1c ( sc-44604 ) and caveolin-1 ( sc-29241 ) were obtained from Santa Cruz Biotechnologies . Mouse fetuses were harvested from 14-day pregnant Cavin-3 KO and wild-type mice . Fetuses were removed from dissected uteruses , usually between 3–5 fetuses per uterus , and heads and liver of fetuses were removed and blood clots removed by washing with 5 ml sterile phosphate buffered saline . Approximately three fresh embryos were placed in a single 10 cm sterile culture dish and minced with a sterile single edge razorblade into 0 . 5–1 mm slices . Minced tissue was digested with 0 . 05% Trypsin-EDTA by adding 5 ml of Trypsin-EDTA solution to each culture dish and incubating at 37°C under 5% CO2 for 20 min with periodic agitation . After trypsin treatment , tissue suspensions were homogenized by passage through a 10 ml pipet . Tissue suspensions were plated in DMEM ( low glucose ) + 10% FBS and incubated for 5 hr at 37°C/5% CO2 . Medium was then replaced with fresh medium and cells were passaged 1:3 every 4 days for six passages , when cells entered crisis . During crisis , cells were re-fed every 3 days and split 1:2 once per week ( or when plates reached confluence ) for 3 months . The results of the Cavin-3 KO MEF line shown in Figure 4 are typical of derived Cavin-3 KO MEF lines . MEFs and the human fibroblast cell line , SV589 ( Yamamoto et al . , 1984 ) , were grown in Delbecco’s Modified Eagle Medium ( DMEM ) supplemented with 10% fetal bovine serum ( FBS ) at 37°C with 5% CO2 . The lung cancer cell line , H1299 , was obtained from the American Tissue Culture Collection ( Manassas , VA ) . H1299 cells were grown in RPMI-1640 growth medium supplemented with 10% FBS at 37°C with 5% CO2 . Cell number was counted using a Bright Line Hemocytometer ( Hausser Scientific , Horsham , PA ) . For transfection of siRNAs into cells , an RNA-liposome suspension was prepared for each sample by mixing 2 ml of serum-reduced OPTI-MEM I media ( Gibco , Grand Island , NY ) with 480 nmol of siRNA and 36 μl of Lipofectamine RNAiMAX Reagent ( Invitrogen , Grand island , NY ) followed by incubation at room temperature for 20 min . The suspension was added to a 10 cm dish prior to layering 10 ml of cell suspension atop the RNA suspension . Cell suspensions were prepared by trypsination and suspension in DMEM supplemented with 10% FBS . Final siRNA concentration was 40 nM . 250 , 000 cells per 10 cm dish were added . After 5 hr incubation with RNA suspension at 37°C , culture medium was replaced with fresh growth medium . For prolonged knockdown , the described siRNA treatment was repeated on days 5 , 9 , and 12 such that assays used cells 3 days post siRNA treatment at all time points , unless otherwise indicated in the figure legend . For all mitogen stimulation experiments , test cells were grown to near confluence , then serum starved for 20 hr prior to addition of serum deficient medium containing mitogens indicated in figure legends . Mitogens were used at the following working concentrations: EGF at 100 ng/ml; PDGF at 20 ng/ml; Insulin at 100 nM; TPA at 200 nM; and LPA at 20 μM . Cells that received Latrunculin-A were pretreated with this actin sequestering agent for 20 min at 25 nM final concentration prior to EGF stimulation . After mitogen treatment , cells were washed twice with PBS and gently scraped into 5 ml PBS supplemented with phosphatase and protease inhibitor cocktails ( RPI Corporation , Mount Prospect , IL ) . Cells were recovered by centrifugation at 1 , 000 × g for 3 min and then resuspended in 1 ml PBS supplemented with protease and phosphatase inhibitor cocktails at 4°C . Protein concentrations were determined by Bradford Assay , cell suspensions were diluted to 1 mg/ml concentration in SDS Sample buffer and denatured by heating to 100°C for 10 min . Equal protein loads of SDS denatured whole cell lysates were resolved by SDS-PAGE , and transferred onto PVDF membranes ( Millipore ) . PDVF membranes were blocked for 30 min with 5% non-fat dry milk , washed with PBS , and incubated with appropriate concentrations of primary antibody overnight at 4°C . Blots were then washed , incubated with anti-rabbit or anti-mouse secondary IgGs conjugated with HRP ( Biorad , Hercules , CA ) for 1 hr at room temperature , washed and visualized on film using the Pierce ECL Chemiluminescence Substrate Kit ( Thermo Scientific , Pittsburgh , PA ) . Quantification was performed by densotometry . All experiments were performed at least three times with a representative experiment shown in figures . Cells were subcultured onto 6 cm plates at an initial plating density of 40% confluency in 3 ml DMEM supplemented with 10% FBS and allowed to proliferate for 24 hr in incubators at 37°C and 5% CO2 . Growth medium was replaced with 1 . 5 ml of fresh DMEM supplemented with 10% FBS and incubated at 37°C in 5% CO2 for a further 8 hr . 1 ml of fresh medium and 1 ml of conditioned media from each plate were centrifuged at 10 , 000 × g for 5 min to remove trace insoluble materials and assayed for Glucose and Lactate content , using Glucose ( Sigma ) and Lactate ( Megazyme ) Assay Kits . Cellular membranes from 4 × 15 cm confluent dishes of untreated , siRNA treated or Latrunculin treated cells were separated by density using published protocols ( Smart et al . , 1995 ) . Immunoprecipitations were performed as follows . Cell lysates were prepared from two 15 cm confluent dishes of cells by washing twice with 10 ml ice-cold PBS followed by scraping of cells into 5 ml of PBS with protease and phosphatase inhibitor cocktails . Scrapped cells were pooled , pelleted at 700 × g , and resuspended in 5 ml of TETN/OG ( 25 mM Tris-HCl , pH 7 . 5; 5 mM EDTA , 150 mM NaCl , 1% triton X-100 , 60 mM octylglucoside ) supplemented with 10 mM CaCl2 . Cell suspensions were incubated on ice for 1 hr , with votex mixing every 15 min to lyse cells . Nuclei were removed by centrifugation at 2000 × g for 5 min . The post-nuclear supernatant was divided into aliquots . One aliquot was TCA precipitated and resuspended in 0 . 2 ml SDS sample buffer and is designated ‘Input’ . The remaining aliquots were incubated with no antibody or 10 μg non-specific rabbit IgG , rabbit anti-cavin-3 , rabbit anti-tubulin or rabbit anti-myosin-1c . Suspensions were incubated at 37°C for 1 hr with gentle mixing . 25 μl of a 50% slurry of Protein A/G beads ( Santa Cruz Biotechnologies ) was then added and incubated with gentle mixing for 4 hr at 4°C . Beads were pelleted at 2000 × g and washed twice with TETN500 ( TET + 500 mM NaCl ) , twice with TETN250 ( TET + 250 mM NaCl ) , and twice with Tris/EDTA ( 10 mM Tris , pH 7 . 5 , 5 mM EDTA ) . Final pellets were dried at 55°C for 1 hr , resuspended in 0 . 2 ml SDS sample buffer and boiled . Eluted material was separated from beads by centrifugation at 10 , 000 × g for 5 min . 1/20 of each sample was resolved on 4–15% Polyacrylamide SDS gels , transferred to PVDF membranes and incubated with indicated primary antibodies overnight at 4°C . Detection of primary antibodies used the Clean-Blot IP Detection Kit ( Pierce Biochemicals ) , which suppresses detection of the denatured IgG heavy and light chains of precipitating antibodies . Cells were grown on circular 12 mm optical borosilicate glass coverslips in 12 well cell culture plates . siRNA treatments were conducted on glass coverslips with appropriate number of cells to reach a confluence of 60% after 3 d culture at 37°C and 5% CO2 . After culturing , coverslips were washed with PBS two times , fixed with 3% paraformaldehype on ice for 15 min and permeabilized with 0 . 1% Triton X100 on ice for 10 min . Coverslips were then blocked with 1% normal goat serum ( NGS ) diluted in PBS for 30 min followed by incubation with primary antibody diluted into PBS supplemented with 1% NGS for 1 hr at room temperature . Coverslips were washed three times with PBS supplemented with 0 . 1% NGS , incubated with Alexaflur conjugated secondary antibodies ( diluted in PBS + 1% NGS ) for 1 hr at room temperature , washed three times with PBS + 0 . 1% NGS , washed twice with PBS , stained with 150 nM DAPI ( 4′ , 6-diamidino-2-phenylindole ) in PBS for 5 min , washed three times with PBS and mounted using Fluoromount-G ( Southern Biotech , Birmingham , AL ) . Cell images were taken on a Zeiss AxioImager M1 fluorescent microscope . Whole cells were gently scraped off culture dishes , fixed with 2% glutaraldehyde ( EM Sciences , Fort Washington , PA ) in PBS at room temperature for 1 hr , post-fixed with 1% Uranyl acetate in PBS for 1 hr , embedded in K4M epoxy resin , sectioned , and viewed with a Tecnai G2 Spirit 120 kV transmission electron microscope . The wild-type cavin-3 or EGR1 was stably expressed in SV589 fibroblasts , H1299 cells and Cavin-3 KO MEFs using a retroviral system as previously described ( Zhao and Michaely , 2008 ) . Briefly , cDNAs for human cavin-3 and human EGR1 were subcloned into the pMX-IRES-GFP bicistronic retroviral vector ( Liu et al . , 1997 ) . Cavin-3 , EGR1 or vector control retroviral vectors were co-transfected with the pAmpho packaging vector ( Clontech , Mountain View , CA ) into 293T cells to produce infectious , replication-defective retroviruses . Recombinant retroviruses were used to infect H1299 cells , which are a metastatic human non-small cell lung carcinoma cell line that does not express detectable cavin-3 ( Xu et al . , 2001 ) , SV589 fibroblasts and Cavin-3 KO MEFs . The IRES element allows both the gene of interest and GFP to be translated from the same mRNA and thus cells that express GFP also express the gene of interest following successful genomic integration of the virus . GFP positive cells were purified using two rounds of fluorescence activated cell sorting ( FACS ) with a BD FACSAria cell sorter ( Becton Dickinson ) . Two rounds of sorting generated populations that were at least 96% GFP positive . Genomic DNA clones of mouse strain 129Sv/J were obtained by PCR into the pCR II vector ( Invitrogen ) . From clones #F7 ( 3 . 2 kb DNA fragment ) and #R7 ( 4 . 1 kb DNA fragment ) , an aligned 7 . 0 kb DNA fragment covering the full Prkcdbp ( Cavin-3 ) gene including both exon1 and exon 2 was assembled and confirmed by sequencing with the mouse genome database ( MGI ) . A pJB1 cassette vector expressing the neomycin resistant gene ( Neor ) flanked by two LoxP sites ( a gift from Joachim Herz , UTSW ) was used to construct the Cavin-3 targeting vector using three steps . ( i ) The 1 . 2 kb Avr II-Xho I fragment ( short arm , SA ) from genomic clone #F7 was subcloned into the Xba I-Xho I sites of the pBS2-SK vector ( Stratagene , La Jolla , CA ) , then subcloned using Not I into pJB1 to generate pJB1/SA . ( ii ) The 3 . 7 kb Xho I fragment ( long arm , LA ) from the full length Cavin-3 clone was subcloned into the Xho I site of pJB1 to generate pJB1/LA . ( iii ) The 7 . 0 kb Bam HI-Pac I fragment cut from clone pJB1/SA4 , was ligated with the 8 . 0 kb Bam HI-Pac I fragment cut from clone pJB1/LA9 to complete the Cavin-3 targeting vector construction . The targeting vector construct was electroporated into J1 ES cells derived from 129Sv/J mice by the Transgenic Core Facility under the direction of Robert Hammer on our campus . 600 ES cell clones resistant to both G418 and gancyclovir were expanded and analyzed by PCR resulting in the identification of two Prkcdbp+/− clones . These cells were injected into C57BL/6 blastocysts , which produced 14 chimeras as assessed by coat color . Germline transmission was determined by coat color and by PCR using the following primer sets: Exon II-mRNA-Fwd1 ( F1 ) , 5′- CAGATCAGCCAGAGGATGAAG-3′ , Exon II-mRNA-Rev1 ( R1 ) , 5′- GGTAGGTTGAGGAGGTTCTGG-3′ , and neo-S3 ( neoF1 ) , 5′- CAGAGGCCACTTGTGTAGCGCC-3′ . The 272 bp ( F1/R1 , wt ) and 532 bp ( neoF1/R1 , KO ) amplification products were verified by sequencing . The null allele was then backcrossed through eight generations onto the C57BL/6 background . Cells were freshly harvested and total RNAs were immediately extracted using an RNeasy Mini kit ( Qiagen , Valencia , CA ) following the manufacturer’s instructions . RNA quality was checked using Bioanalyzer Chip ( Agilent , Santa Clara , CA ) and gene expression data were obtained using HumanHT-12 v4 Expression BeadChip ( Illumina , San Diego , CA ) through the UTSW Microarray Core Facility on campus . Mean linear intercepts were calculated using H&E sections of normal and Cavin-3 KO lung at 600 × magnification by a blinded observer using ImageJ software . Four measurements per section were made using sections obtained from three animals per group . p values were obtained by two-tailed Student’s t-test using GraphPad Prism 5 . Sedated animals were sacrificed by exsanguination . Lungs were then removed , sliced , weighed , and cultured in D-MEM with antibiotics for 4 hr . Glucose and lactate production were assessed as in the in vitro cell culture experiments . Duplicate plates of cells were treated with 10 μg/ml cycloheximide alone or in combination with 10 ng/ml TNFα for 15 hr in a humidified , 37°C CO2 incubator . Cells were then either processed as whole cell lysates for PARP1 analysis or processed as whole cells for TUNEL . For TUNEL assays cells were washed with PBS , scraped into PBS , washed , fixed with 4% paraformaldehyde ( 10 min on ice ) , washed with PBS and permeabilized with ethanol ( 70% ethanol 15 hr −20°C ) . TUNEL was performed using a TUNEL kit ( #A23210; Invitrogen ) with Pacific Blue conjugated anti-BrdU ( #B35129; Invitrogen ) . FACS analysis of TUNEL samples used a BD LSR II Flow cytometer at 405 nm . 10000 cells were counted for each sample and fluorescent profiles quantified using FlowJo software . All apoptosis data are derived from three independent experiments .	The plasma membrane separates cells from their environment , and surface receptors in this membrane allow cells to respond to changes in their environment by converting external cues into intracellular signals . This process , which is known as signal transduction , plays a central role in the biology of cells , and abnormal signaling is a common cause of human disease . In cancer for example , signals tend to be too strong or they are sent at the wrong time . Signal transduction frequently occurs at specialized regions of the plasma membrane . Caveolae are small indentations of the plasma membrane that comprise one type of signaling specialization . A protein that is concentrated in caveolae , cavin-3 , suppresses tumor formation and is commonly absent from cancer cells . These observations suggest that cavin-3 participates in signal transduction and pathways that are associated with cancer , but the details of this involvement are not well understood . Hernandez et al . now show that cavin-3 controls the balance between two key intracellular signals , ERK and Akt . High levels of cavin-3 promote activation of the ERK signaling pathway but suppress activation of the Akt signaling pathway . Loss of cavin-3 has the opposite effect , activating Akt at the expense of ERK . The consequences of loss of cavin-3 include accelerated cell proliferation , the induction of Warburg metabolism ( a metabolic state that supports rapid cell division ) , and the suppression of the apoptosis pathway . ( Suppression of this pathway , which leads to cell death , allows cancer cells to proliferate in the body . ) While deletion of the cavin-3 gene in mice is not sufficient to cause spontaneous cancer , animals that are deficient in cavin-3 die prematurely of cachexia , a tissue wasting sequela experienced by nearly half of all cancer patients . Hernandez et al . also show that cavin-3 influences cellular signaling by linking caveolae to the membrane skeleton—a network of proteins that underlies the plasma membrane . This linkage is necessary to ensure that cells have the correct abundance of caveolae , and it also facilitates signal transduction to the ERK signaling pathway . ERK activation in this context drives expression of two proteins , EGR1 and PTEN , which suppress Akt signaling . Hernandez et al . propose that the membrane skeleton functions as a scaffold that adaptors , such as cavin-3 , use to assemble signaling modules with surface receptors for the purpose of controlling the signal transduction output from these receptors .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "cell", "biology" ]	2013	Cavin-3 dictates the balance between ERK and Akt signaling
Humans usually assess things not according to their absolute value , but relative to reference points – a main tenant of Prospect Theory . For example , people rate a new salary relative to previous salaries and salaries of their peers , rather than absolute income . We demonstrate a similar effect in an insect: ants expecting to find low-quality food showed higher acceptance of medium-quality food than ants expecting medium quality , and vice versa for high expectations . Further experiments demonstrate that these contrast effects arise from cognitive rather than mere sensory or pre-cognitive perceptual causes . Social information gained inside the nest can also serve as a reference point: the quality of food received from other ants affected the perceived value of food found later . Value judgement is a key element in decision making , and thus relative value perception strongly influences which option is chosen and ultimately how all animals make decisions . We all compare options when making both large and small decisions , ranging from career choices to the choices of an evening’s entertainment . Understanding how options are compared has thus been central to the study of behavioural economics . Theories explaining the mechanisms by which options are compared and decisions are made have a long tradition ( Vlaev et al . , 2011 ) , with Expected Utility Theory ( EUT ) being the most widely used theory in economic models ( Mankiw , 2011; von Neumann and Morgenstern , 1944 ) . EUT suggests that decisions are made by evaluating and comparing the expected utility from each option . A rational decision maker then chooses the option resulting in the best end state: the option providing the greatest utility ( von Neumann and Morgenstern , 1944 ) . However , over the past decades economic research on human decision making has started to shift away from a view of ( absolute ) utility maximisation towards more nuanced notions of relative utility , such as reference-dependent evaluations . Kahneman and Tversky ( 1979 ) made a major contribution to this shift by introducing Prospect Theory , suggesting that decision making is not based on absolute outcomes , but rather on relative perceptions of gains and losses . In contrast to EUT , the utility attributed to options being evaluated is determined relative to a reference point , such as the status quo or former experience ( Vlaev et al . , 2011; Kahneman and Tversky , 1979; Parducci , 1984; Tversky and Kahneman , 1992; Ungemach et al . , 2011 ) . Various examples of relative value perception have been described . For example , satisfaction gained from income is perceived not absolutely , but relative to the income of others in the social reference group – such as one’s colleagues ( Boyce et al . , 2010 ) . Overall , Prospect Theory has enriched our understanding of human decision making by conceptualising it as more nuanced and less rational than previously assumed ( Tversky and Kahneman , 1974; Tversky and Kahneman , 1981 ) . The concept of malleable value perception is not just relevant to humans . Value judgments in animals are also influenced by factors apparently independent of the absolute value of options . For example , capuchin monkeys refuse otherwise acceptable pay ( cucumber ) in exchanges with a human experimenter if they had witnessed a conspecific obtain a more attractive reward ( grape ) for equal effort ( Brosnan and De Waal , 2003 ) . Rats , starlings , and ants , like humans , place greater value on things they work harder for ( Aw et al . , 2011; Czaczkes et al . , 2018a; Lydall et al . , 2010 ) , and starlings , fish and locusts demonstrate state-dependent learning , wherein they show a preference for options experienced when they were in a poor condition ( Aw et al . , 2009; Pompilio et al . , 2006; Schuck-Paim et al . , 2004 ) . Roces and Núñez ( Roces , 1993; Roces and Núñez , 1993 ) aimed to show that in leaf cutting ants perceived value can be influenced by other ants . Ants recruited to higher quality food sources ran faster , deposited more pheromone , but cut smaller leaf fragments , even if the food source the recruits find is replaced by a standardised food source ( Roces , 1993; Roces and Núñez , 1993 ) . However , in these experiments , the absolute value and nature of the reference remains unclear , and indeed pheromone presence may have caused the observed behaviours without influencing the ants’ expectations or value perception at all . Healey and Pratt ( 2008 ) showed that colonies of the house-hunting ant species Temnothorax curvispinosus move into a nest of mediocre quality faster when they were previously housed in a high-quality nest compared to colonies which were previously housed in a poor-quality nest ( Healey and Pratt , 2008 ) . In contrast , Stroeymeyt et al . ( 2011 ) showed that colonies of Temnothorax albipennis developed an aversion towards mediocre-quality nests available in their environment when they were housed in a high-quality nest , whereas colonies housed in a low-quality nest did not , and thus show an experience-dependent flexibility in nest choice ( Stroeymeyt et al . , 2011 ) . However , critically missing from the existing works is a systematic description of value judgment relative to a reference point . ‘Value distortion by comparison’ effects have been studied for decades using the successive contrasts paradigm , in which animals are trained to a quality or quantity of reward which is then suddenly increased ( positive incentive contrast ) or decreased ( negative incentive contrast ) ( Bentosela et al . , 2009; Bitterman , 1976; Couvillon and Bitterman , 1984; Crespi , 1942; Flaherty , 1982; Flaherty , 1999; Mustaca et al . , 2000; Weinstein , 1970a ) . Many mammals , including apes , monkeys , rats , and dogs ( Brosnan and De Waal , 2003; Bentosela et al . , 2009; Crespi , 1942; Flaherty , 1999; Mustaca et al . , 2000; Pellegrini and Mustaca , 2000; Weinstein , 1970b ) have been shown to respond to successive negative contrast by disrupting their behaviour compared to control animals which had not experienced a change in reward . The animals display behaviour akin to disappointment – slower running speeds to a reward ( Bower , 1961 ) , depressed licking behaviour ( Flaherty et al . , 1985; Vogel et al . , 1968 ) , or reward rejection ( Tinklepaugh , 1928 ) . Contrast effects were also successfully described in invertebrates ( Bitterman , 1976; Couvillon and Bitterman , 1984; Richter and Waddington , 1993 ) . Bitterman ( 1976 ) found negative incentive contrast effects in honey bees which were trained to a high-quality feeder and then received a downshift to a lower quality feeder . In contrast , bees which experienced an upshift in feeder quality did not show any feeding interruptions ( Bitterman , 1976; Couvillon and Bitterman , 1984 ) . While negative successive contrast effects – akin to disappointment – have been well described in animals , positive successive contrast effects – akin to elation – have often proved elusive ( Bower , 1961; Black , 1968; Capaldi and Lynch , 1967; Dunham , 1968; Papini et al . , 2001 ) . There are several factors which may prevent positive contrast effects from being detected . Firstly , ceiling effects may occur when the performance of animals receiving a large reward is at or near a physical limit . The absence of positive contrast effects may then not be due to the absence of perceived positive contrast , but rather due to an artefact of experimental design ( Bower , 1961; Campbell et al . , 1970 ) . Secondly , neophobia counteracts positive contrast effects: animals may be reluctant to eat a novel food – even if the food is of higher quality than normal ( Flaherty , 1999; Oberhauser and Czaczkes , 2018 ) . Finally , generalisation decrement may prevent stronger responses to positive contrast . Generalisation decrement occurs when animals are trained under one set of stimuli and then tested under another . The strength of the tested response may decrease with increasing differences between the training and testing stimuli ( Kimble , 1961 ) , which may then result in weaker positive contrast effects following a reward shift . Thus , the reward change itself may lead to a decrease in responding just as would any other change in context , such as a change in the brightness of the runway or scent of the food ( Oberhauser and Czaczkes , 2018; Capaldi , 1978; Premack and Hillix , 1962 ) . Even though positive contrast effects proved to be hard to demonstrate in laboratory experiments , there are good theoretical reasons for expecting both positive and negative contrast effects to evolve ( McNamara et al . , 2013 ) . Incentive contrasts have also been demonstrated for rewards other than food . Females become more ( or less ) likely to accept a mate of given quality if they have prior experience of better ( or worse ) mates . Such mate quality contrast effects are reported in both vertebrates ( Collins , 1995 ) and invertebrates ( Dukas , 2005; Reid and Stamps , 1997; Wagner et al . , 2001 ) . In this study , we investigate positive and negative contrast effects using the successive contrasts paradigm , and , in addition to demonstrating positive and negative contrast effects , define the first relative value curve in an invertebrate; the ant Lasius niger . We conduct a critical control experiment to rule out physiological or psychophysical effects which may lead to the same pattern ( see experiment 2 ) and thus provide strong evidence for a purely cognitive relative value effect in a non-human animal . Furthermore , we demonstrate that information flowing into the nest can influence value perception in outgoing foragers . This suggests that food sources are not only valued based on individual experiences , but also based on social information gained inside the nest . The perceived value of a food source influences social information dissemination , by affecting the strength of pheromone trails which then lead further ants to the food source . Thus , the way in which value is judged is likely to strongly affect the foraging mechanics of a whole colony . Eight stock colonies of the black garden ant Lasius niger were collected on the University of Regensburg campus . The colonies were kept in 30 × 30 × 10 cm foraging boxes with a layer of plaster covering the bottom . Each box contained a circular plaster nest box ( 14 cm diameter , 2 cm height ) . The colonies were queenless with around 1000–2000 workers and small amounts of brood . Queenless colonies still forage and lay pheromone trails and are frequently used in foraging experiments ( Devigne and Detrain , 2002; Dussutour et al . , 2004 ) . The colonies were fed with ad libitum 0 . 5M sucrose solution and received Drosophila fruit flies once a week . Water was available ad libitum . One sub-colony of 500 individuals was formed from each stock colony , and these eight fixed-size sub-colonies were used for our experiments . Sub-colonies were maintained identically to the stock colonies , but did not receive any Drosophila fruit flies to prevent brood production , and were starved 4 days prior to the experiments in order to achieve a uniform and high motivation for foraging ( Mailleux et al . , 2006; Josens and Roces , 2000 ) . During starvation , water was available ad libitum . Any ants which died or were removed from the sub-colonies were replaced with ants from the original stock colonies . The general setup used for all of our three experiments was identical and consisted of a 20 × 1 cm long paper-covered runway which was connected to the sub-colony’s nest box via a 40 cm long drawbridge ( Figure 1A ) . A 5 mm diameter drop of sucrose solution ( Sigma-Aldrich ) was placed on an acetate feeder at the end of the runway ( 60 cm from the nest ) . The molarity of the sucrose droplet depended on the experiment , treatment and on the ants’ number of visit to the food source . To begin an experiment , 2–4 ants were allowed onto the runway , and the first ant to reach the feeder was marked with a dot of acrylic paint on its gaster . This procedure may select for the more active foragers , but does not introduce any selection bias between treatments . The marked ant was allowed to drink to repletion at the food source , while all other ants were returned to the nest . Food acceptance scores as a measure of perceived value were noted for each ant and visit as follows: Full acceptance ( 1 ) was scored when the ant remained in contact with the drop from the moment of contact and did not interrupt drinking within 3 s of initial contact ( see Video 1 ) . Partial acceptance ( 0 . 5 ) was scored if feeding was interrupted within 3 s after the first contact with the food source , but the ant still filled its crop within 10 min ( as can be seen by the distention of the abdominal tergites ) . Ants which interrupt feeding within the first seconds after contacting the food usually show successive feeding interruptions and generally show a rather ‘impatient’ behaviour compared to ants which show a food acceptance score of 1 ( see Video 2 ) . Lastly , rejection ( 0 ) was scored if the ant refused to feed at the sucrose solution and either returned to the nest immediately or failed to fill its crop within 10 min . When the ant had filled its crop or decided not to feed at the sucrose droplet , it was allowed to return to the nest . Inside the nest , the ant unloaded its crop to its nestmates and was then allowed back onto the runway for another visit . The drawbridge was now used to selectively allow only the marked ant onto the runway . In addition to measuring food acceptance , we also measured pheromone deposition . Individual pheromone deposition behaviour correlates with the ( perceived ) quality of a food source ( Beckers et al . , 1993; Hangartner , 1970; Czaczkes et al . , 2015 ) . Individual ants can adapt the strength of a pheromone trail by either depositing pheromone or not , or varying the intensity of a pheromone trail through number of pheromone depositions ( Beckers et al . , 1993; Hangartner , 1970 ) . Pheromone deposition behaviour in L . niger is highly stereotypic . To deposit pheromone , an ant briefly interrupts running to bend its gaster and press the tip of the gaster onto the substrate ( Beckers et al . , 1992 ) . This allows the strength of a pheromone trail to be quantified by counting the number of pheromone depositions over the 20 cm runway leading to the feeder . Pheromone depositions were measured each time the ant moved from the food source back to the nest ( inward trip ) , and each time the ant moved from the nest towards the food source ( outward trip ) . Because L . niger foragers almost never lay pheromone when they are not aware of a food source ( Beckers et al . , 1992 ) , we did not measure pheromone depositions for the very first outward trip ( visit 1 ) . The presence of trail pheromone on a path depresses further pheromone deposition ( Czaczkes et al . , 2013 ) . Thus , each time an ant had passed the 20 cm runway , the paper overlay covering the runway was replaced by a fresh one every time the ant left the runway to feed at the feeder or returned to the nest . All experimental runs were recorded with a Panasonic DMC-FZ1000 camera to allow for later video analysis . Each tested ant was observed until all experimental runs were finished and then discarded from the colony before switching to the next ant . If an ant did not return before finishing all experimental runs , we waited for 15 min , then discarded it from the colony and moved to the next ant . Statistical analyses were carried out in R v . 3 . 4 . 1 ( R Development Core Team , 2016 ) using Generalized Linear Mixed Models ( GLMMs ) in the LME4 package ( Bates et al . , 2014 ) to analyse pheromone depositions data and Cumulative Link Mixed Models ( CLMMs ) in the ordinal package ( Christensen , 2015 ) to analyse food acceptance scores . CLMMs were used to analyse the acceptance data since we used an ordered factor with three levels ( 1 = full acceptance , 0 . 5 = partial acceptance , 0 = rejection ) . As multiple ants were tested per colony , colony identity was added as a random effect to each model . GLMMs were tested for fit , dispersion and zero inflation using the DHARMa package ( Hartig , 2017 ) . The model predictors and interactions were defined a priori , following Forstmeier and Schielzeth ( 2011 ) . All p-values presented were corrected for multiple testing using the Benjamini–Hochberg method ( Benjamini and Hochberg , 1995 ) . A total of 1070 ants were tested , with 829 in experiment 1 , 73 in experiment 2 and 168 in experiment 3 ( Supplementary file 1 ) . Sample sizes were set ahead of time by deciding how much time we will invest in data collection ( 1 day per treatment per colony ) . Depending on the experiment , we either used treatment ( experiment 1 and 3 = Reference Molarity; experiment 2 = expected molarity triggered by a scented runway and the odours presented on the runway ) or an interaction between treatment and visit number , and the odours presented on the runway ( training visits of experiment 2 ) or trophallaxis time ( experiment 3 ) as fixed factors . The interaction between expected molarity and visit number in the training runs of experiment 2 was added , because experience with a food source is likely to affect the behaviour at a food source . The odours presented on the runway were added as fixed factors to test for odour preferences regardless of sucrose molarity . The interaction between trophallaxis time and reference molarity in experiment 3 was added because trophallaxis time may affect food acceptance through crop load and information gained through trophallaxis ( for the effects of trophallaxis time on food acceptance see Figure 5—figure supplement 1 , and Table S4 in Figure 5—source data 1 ) . Because individual ants were tested multiple times in experiments 1 and 2 , we included AntID nested in colony as a random factor for statistical analyses of the training visits . We used the following general model formula ( this formula varied depending on experiment as described above ) :FoodAcceptance∼treatment+ ( randomfactor:colony ) As the pheromone deposition data is count data , they were analysed using a GLMM with a Poisson distribution . Depending on the experiment , we either used treatment ( experiment 1 = Reference Molarity; experiment 2 = expected molarity triggered by a scented runway and the odours presented on the runway ) or an interaction between treatment and visit number ( training visits of experiment 2 ) as fixed factors . The interaction between expected molarity and visit number in the training runs of experiment 2 was added , because experience with a food source is likely to affect the behaviour at a food source . The odours presented on the runway were added as fixed factors to test for odour preferences regardless of sucrose molarity . Because individual ants were tested multiple times in experiment 2 , we included AntID nested in colony as a random factor for statistical analyses of the training visits . For statistical analysis of experiment 1 , we also added a variable indicating if ants deposited more or less pheromone compared to the average to correct for individual strength of pheromone depositions and overdispersion . The variable was calculated as follows:Differencetoaverage= ( ( NumberPheromoneDepositions1stvisit−meannumberPheromoneDepositions1stvisit ) + ( NumberPheromoneDepositions2ndvisit−meannumberPheromoneDepositions2ndvisit ) ) /2 We used the following model formulae in the model: Experiment 1:NumberPheromoneDepositions∼treatment+Differencetoaverage+ ( Differencetoaverage ) 2+ ( randomeffects:colony/AntID ) Experiment 2:NumberPheromoneDepositions∼scentassociatedtomolarity+ ( randomeffects:colony ) The number of drinking interruptions was quantified via video analysis in experiment 2 ( see below ) . This was analysed statistically in a manner identical to number of pheromone depositions . Trophallaxis time in seconds in experiment three were used in full seconds and treated as count data . We performed a GLMM with Poisson distribution and Reference Molarity as a fixed effect , while colony identity was added as a random factor:TrophallaxisTimeseconds∼ReferenceMolarity+ ( randomeffects:colony ) Ants made two initial training visits to a feeder at the end of a runway in order to set their reference point ( Figure 1A ) . The quality of the sucrose solution was varied between ants , with each ant receiving the same quality twice successively . Twelve different molarities were used: 0 . 1 , 0 . 2 , 0 . 3 , 0 . 4 , 0 . 5 , 0 . 6 , 0 . 7 , 0 . 8 , 0 . 9 , 1 , 1 . 5 or 2M ( also referred to as pre-shift solution or reference point ) . Lasius niger workers learn the quality of a feeder within two visits ( Wendt and Czaczkes , 2017 ) . On the third visit ( test visit ) , the food source was replaced by a 0 . 5M sucrose solution droplet for all ants ( also referred to as post-shift solution ) . Thus , ants trained to qualities < 0 . 5M experienced a positive successive contrast , ants trained to > 0 . 5M experienced a negative successive contrast , and the ants trained to 0 . 5M constituted the control ( no contrast ) . 97% of ants successfully finished the training procedure and participated in the test visit ( third visit ) . Ants seemed to value sucrose solution droplets relative to their reference point ( Figure 2—figure supplement 1 ) . In the training visits , acceptance scores increased significantly with increasing molarity of the reference quality ( CLMM: estimate = 1 . 97 , z = 9 . 65 , p<0 . 001 , Figure 2 ) . However , in the test ( contrast ) visit , acceptance scores decreased significantly with increasing molarity of the reference quality ( CLMM: estimate = −2 . 59 , z = −13 . 57 , p<0 . 001 , Figure 2 ) . Ants which were trained to the lowest molarity ( 0 . 1M: p<0 . 001 ) showed significantly higher acceptance of 0 . 5M sucrose than control ants , while ants trained to high molarities ( 1 . 5M: p<0 . 001 , 2M: p<0 . 001 ) showed lower acceptance of 0 . 5M than the control group ( see Table S1 in Figure 2—source data 1 for all pairwise comparisons ) . A similar pattern was found for pheromone deposition behaviour on the way back to the nest ( Figure 3 ) . In the training visits , number of pheromone depositions increased significantly with increasing molarity of the reference solution ( GLMM: estimate = 0 . 86 , z = 13 . 87 , p<0 . 001 ) . By contrast , on the test visit pheromone depositions decreased significantly with increasing molarity of the reference solution ( GLMM: estimate = −0 . 82 , z = −9 . 75 , p<0 . 001 , Figure 3 ) . Ants which deposited more pheromone during the training visits generally deposited more pheromone on the test visit compared to ants which deposited less pheromone during the training visits ( GLMM: estimate = 0 . 16 , z = 15 . 99 , p<0 . 001 ) . Ants which were trained to a low molarity ( 0 . 2M: p<0 . 01 ) deposited significantly more pheromone in the test visit than control ants , while ants trained to high molarities ( 1M: p<0 . 001 , 1 . 5M: p<0 . 001 , 2M: p<0 . 001 ) deposited less pheromone than the control group ( see Table S2 in Figure 3—source data 1 for pairwise comparisons ) . These results are consistent with relative value perception stemming from the psychological effects of successive contrasts . We could further define a relative value perception curve similar to that described in Prospect Theory , as well as showing positive contrast effects for both food acceptance and number of pheromone depositions . However , there is another possible explanation for these results: non-random selection of individuals with different acceptance thresholds . Different individuals from the same colony may have different acceptance thresholds . Animals with lower acceptance thresholds may readily exploit low-quality food sources while animals with higher thresholds may not . When training to lower molarity sucrose , ants with high thresholds may not have completed training , leaving only a non-random subset of ants with low acceptance thresholds at the test phase ( Robinson et al . , 2009 ) . Thresholds may also be influenced by experience , by which animals use the best experienced option as a threshold for accepting a new option or not ( Stroeymeyt et al . , 2011; Robinson et al . , 2011 ) . However , we can exclude this possibility , as the proportion of ants not completing training was uniformly low and did not vary with treatment ( see Supplementary file 1 ) . This may occur in ants which were trained to higher molarity food due to the blocking of more sweetness receptors compared to low molarity sucrose . The more sweetness receptors are blocked by a sweet reference solution , the fewer receptors will fire when confronted with a post-shift reward , thus making solutions taste less sweet for ants trained to high-molarity solutions , and sweeter for ants which were trained to low molarities ( Bitterman , 1976 ) . Ants may not only have stored sucrose solutions in their crop while foraging , but may also have ingested small amounts of it , leading to an increase of haemolymph-sugar levels . Higher blood-sugar levels negatively affect sweetness perception in humans ( Mayer-Gross and Walker , 1946; Melanson et al . , 1999 ) , and a similar effect could cause a post-shift solution to taste less sweet to animals trained on high sucrose concentrations . The contrast effects shown in experiment one could also derive from simple psychophysical mechanisms ( Fechner , 1860; Zwislocki , 2009 ) , and thus arise from sensory perceptual mechanisms rather than higher level cognitive processing of value . Sensory judgements are usually made relative to reference points and through constant comparisons with former stimuli ( Vlaev et al . , 2011; Helson , 1964 ) . Thus , identical stimuli may be perceived differently depending on the context they are presented within . The position of the reference point in the range of stimuli may thus bias how the stimulus , and thus the value , of a post-shift reward is perceived ( Zwislocki , 2009 ) . For example , the sweetness of a sucrose solution may be perceived as much stronger when the reference point to which it is compared is low . Psychophysical sensory contrasts are physiological or low-level cognitive phenomena , found in all animal taxa studied , and even potentially in bacteria ( Akre and Johnsen , 2014; Kojadinovic et al . , 2013; Mesibov et al . , 1973 ) . Animals may rationally expect the pre-shift reward to be available in the future again and therefore rationally show lower acceptance towards the post-shift reward , because they are waiting for the pre-shift reward to reoccur . All these alternative factors would lead to the same behavioural patterns found in experiment one without relative value perception necessarily being present . Experiment two was designed to rule out these alternative explanations . To rule out the alternative non-psychological explanations for the contrast effects we described above , we needed to change the expectations of the ants while exposing all ants to identical training regimes . This would provide a reference point for testing relative value perception while keeping sensory saturation , haemolymph-sugar levels , psychophysical effects , future expectations , and ant subsets the same until the testing phase . Ants were allowed to make eight training visits . The quality of the sucrose solution offered at the end of the runway alternated each visit , always beginning with the low-quality solution . The solutions were scented using either rosemary or lemon essential oils ( 0 . 05 µl essential oil per ml sucrose solution , rosemary: Rosmarinus officinalis; Lemon: Citrus limon , Markl GbR , Grünwald ) . In half the trials the 1 . 5M solution was scented with lemon and the 0 . 25M with rosemary , and vice versa for the other trials . In addition , to support learning and to allow solution quality anticipation , we also scented the paper overlays covering the runway leading to the feeder . Paper overlays were scented by storing them for at least 1 day in an airtight box containing a droplet of essential oil on filter paper in a petridish . Finally , in addition to odours cuing sucrose molarity , visual cues were also provided . These consisted of printed and laminated pieces of paper ( 22 × 16 . 5 cm , displayed in Figure 1B ) displayed at the end of the runway , directly behind the sucrose droplet . On the 9th ( test ) visit , the odour of the runway and the visual cue signified either 1 . 5M or 0 . 25M , while the sucrose solution provided was unscented and of intermediate ( 0 . 5M ) quality . Runway scents in the test visit were varied systematically between ants , but each ant was confronted with only one of the two runway scents coupled with unscented 0 . 5M sucrose . While the ant fed at the sucrose droplet , the scented runway overlay was replaced with an unscented overlay in order to eliminate possible effects of scent association on pheromone deposition behaviour . Previous work has shown that L . niger foragers can form robust expectations of upcoming reward quality based on runway odour after four visits to each odour/quality combination ( Czaczkes et al . , 2018b ) . Nonetheless , to ensure that learning had taken place , on the 10th visit , we carried out a memory probe . The linear runway was replaced with a Y-maze ( Figure 1B ) , with two 10 cm long arms and a 10 cm long stem . The Y-maze stem was covered with an unscented paper overlay while one arm was covered with the 1 . 5M-associated odour overlay , and the other with the 0 . 25M-associated odour overlay . The matching visual cues were placed directly behind the relevant Y-maze arms . Trained ants were allowed to walk onto the Y-maze and their arm choice was noted . We used two decision lines to define arm choice – an initial decision line ( Figure 1B , 2 . 5 cm after the bifurcation ) and a final decision line ( 7 . 5 cm after the bifurcation ) . After testing on the Y-maze , the ants were permanently removed from the colony . 97 . 2% of ants successfully finished the training procedure and participated in the last test visit . Additionally to the other measures , on the 9th ( test ) visit of this experiment we counted the number of food interruptions made by an ant from the moment of first hitting the food source until it had finished feeding at the sucrose droplet . The number of food interruptions are likely to reflect and support the behaviour encoded in food acceptance scores and was thus investigated to give stronger support for the results of this experiment . During training , ants behaved as expected , showing higher acceptance and pheromone deposition for 1 . 5M compared to 0 . 25M on all but the very first visit to 0 . 25M ( Food acceptance: CLMM: estimate = −7 . 34 , z = −8 . 9 , p<0 . 001; pheromone depositions outward journey: GLMM: estimate = 0 . 23 , z = 2 . 89 , p<0 . 01; pheromone depositions inward journey: GLMM: estimate = −2 . 49 , z = −19 . 46 , p<0 . 001 , Figure 4A , C & E ) . Furthermore , food acceptance and pheromone depositions both on the outward and inward journeys decreased with increasing experience with the 0 . 25M feeder and increased with increasing experience with the 1 . 5M feeder ( Food acceptance: CLMM: estimate = −2 . 84 , z = −3 . 63 , p<0 . 001; pheromone depositions outward journey: GLMM: estimate = −0 . 94 , z = −10 . 00 , p<0 . 001; pheromone depositions inward journey: GLMM: estimate = −0 . 53 , z = −4 . 41 , p<0 . 001 ) . On the outward journey of the 9th ( test ) visit , ants walking towards the feeder while exposed to 1 . 5M sucrose-associated cues deposited more pheromone ( median = 15 , Figure 4D ) compared to ants exposed to 0 . 25M-associated cues ( median = 2 , GLMM: estimate = −1 . 31 , z = −12 . 94 , p<0 . 001 ) . Moreover , in the learning probe , 87% of ants chose the 1 . 5M associated arm . This demonstrates that ants formed a robust expectation of food molarity based on the cues learned during training . Ants exposed to 1 . 5M-associated cues during the 9th visit showed significantly lower food acceptance towards the unscented 0 . 5M feeder than ants exposed to 0 . 25M-associated cues ( CLMM: estimate = 1 . 04 , z = 2 . 049 , p<0 . 05 , Figure 4B , Supplementary file 1 ) . Although ants exposed to high molarity associated cues – presented through scented runways on the way to the food – showed a significantly higher number of pheromone depositions on their return journey than ants confronted with low molarity scent ( GLMM: estimate = −1 . 65 , z = −3 . 03 , p<0 . 01 , Figure 4E & F ) , the number of pheromone depositions decreased drastically for both treatments compared to training visits ( median 1 . 5M = 0 , median 0 . 25M = 0 , Figure 4E and F , Supplementary file 1 ) . Even after controlling for alternative explanations , ants still show contrast effects depending on the quality of the post shift solution . This is in spite of all ants undergoing identical training experiences . The only difference between the groups was the odour of the runway on the 9th ( test ) visit . It is thus unlikely that sensory saturation , increased haemolymph-sugar levels , simple psychophysical effects or ants expecting pre-shift solutions to return can fully explain the behaviour of the ants in our experiments . All videos were re-analysed by a naive scientific assistant and this blind analysis of the ants behaviour confirmed the stated results ( CLMM: estimate = 1 . 42 , z = 2 . 35 , p=0 . 019 ) , and also found that ants interrupted drinking significantly more often when expecting high rather than low food qualities ( GLMM , estimate = 0 . 36 , z = 2 . 74 , p=0 . 006 , see Figure 4—figure supplement 1—source data 1 and Figure 4—figure supplement 1 ) . Non-random selection of individuals with different acceptance thresholds can also be excluded for the results of this experiment as the proportion of ants not completing training was again uniformly low ( see Supplementary file 1 ) and all ants had to taste both low and high molarities in order to complete training . An ant was allowed to feed at an unscented sucrose solution droplet of either 0 . 16 , 0 . 5 or 1 . 5M ( also referred to as pre-shift solution or reference point ) and return to the nest to unload its crop via trophallaxis . When trophallaxis began , we noted the time spent in trophallaxis with the first trophallactic partner . When trophallaxis stopped , the receiving trophallactic partner ( receiver ) was gently moved from the nest and placed onto the start of a runway offering unscented 0 . 5M sucrose solution at the end ( also referred to as post-shift solution ) . As the receiver fed , we noted its food acceptance . Acceptance scores of receivers towards 0 . 5M decreased with increasing molarity of the sucrose solution received through food exchanges inside the nest ( CLMM: estimate = −0 . 57 , z = −3 . 07 , p<0 . 01 ) . The interaction of reference molarity and trophallaxis time significantly predicted acceptance ( CLMM: estimate = −0 . 48 , z = −2 . 33 , p=0 . 02 , Figure 5 ) and longer trophallaxis times led to lower food acceptance in ants as well ( CLMM: estimate = −0 . 70 , z = −3 . 62 , p<0 . 001 ) . Ants which received 0 . 16M inside the nest showed significantly higher acceptance of 0 . 5M sucrose than ants which received 1 . 5M ( p<0 . 01 , see Table S3 in Figure 5—source data 1 for pairwise comparisons ) . The time spent in trophallaxis with the receiver increased significantly with increasing molarity ( GLMM: estimate = 0 . 13 , z = 4 . 79 , p<0 . 001 , see Figure 5—source data 1 ) . Ants valued a standard quality food source relative to the molarity which they received from a returning forager inside the nest . This suggests that information about the quality of a food source received through trophallactic interactions inside the nest can be used by naive foragers when evaluating new food sources outside the nest . Thus , the nest serves as an information hub in which information about available food sources can be gathered , processed , and disseminated . The introduction of Prospect Theory ( Kahneman and Tversky , 1979 ) contributed to a major shift in economic research by suggesting that humans do not perceive value in absolute terms , but relative to reference points . Here , we demonstrate parallel findings in an insect . To the best of our knowledge , we provide the first detailed description of relative value perception in an invertebrate based on individual experience , but also induced by social information . Furthermore , we demonstrate the elusive positive contrast effects in ants which were trained to low molarities ( Figure 2 and 3 ) . Similar results in house-hunting ants were explained by a simple threshold rule ( Stroeymeyt et al . , 2011; Robinson et al . , 2009; Robinson et al . , 2011 ) which suggests that individuals have different acceptance thresholds and ants with lower thresholds accept lower quality options . The higher the quality of the option , the more often it exceeds the acceptance threshold of individual ants , and thus the option is accepted more readily . This could have potentially affected our results in experiment 1 , as we would expect fewer individuals to accept very low reference points . Ants which did not accept the low-quality sucrose would thus not be tested . Therefore , at low reference points , we would only select individuals with very low acceptance thresholds , while no threshold selection would occur at high reference points . When confronting ants with medium-quality food after training , the differently selected acceptance thresholds may lead to the same pattern as we observed . However , 97% of all ants finished both the training and the test phases and no higher proportion of cancelled training can be seen at lower reference molarities ( see Supplementary file 1 ) . It is thus unlikely that a simple threshold rule leads to the results shown in experiment 1 ( Figure 2 and 3 ) . While a second major prediction of Prospect Theory , that ‘losses loom larger than gains’ ( Tversky and Kahneman , 1992 ) , is not supported by the data of our main experiment , it is also not ruled out . We believe ants do overemphasise losses , but , due to limitations in the experimental design and physiological limitations of the animals , we cannot make strong claims about this ( Collins , 1995; Dukas , 2005; Reid and Stamps , 1997 ) . The lack of strong evidence for losses being overemphasised may stem from the psychophysics of our study system: a basic tenant of psychophysics is that the Just Noticeable Difference ( JNDs ) between two stimuli is a function of the relative difference between the stimuli ( Fechner , 1860; Zwislocki , 2009; Stevens , 1957 ) . Thus , ants shifted from 0 . 1M to medium ( 0 . 5M ) quality experience a fivefold increase in molarity , while those down-shifted from 0 . 9M to 0 . 5M experience less than a twofold decrease , although the absolute change was of the same magnitude . This would predict larger shift-changes , in terms of absolute molarity change , for gains than for losses . Indeed , the fact that this is also not seen may imply that losses are indeed – relatively speaking – looming larger than gains for the ants . Finally , it must be kept in mind that acceptance scores are unlikely to be linear , and that pheromone deposition behaviour shows large variation ( Beckers et al . , 1992 ) , making it difficult to use either of these factors to quantitatively test for over- and undervaluation of gains and losses . The results of experiment 2 allow us to exclude all but a cognitive relative value effect ( Figure 4 ) . This cognitive effect is subjectively very familiar to humans , and its presence in an invertebrate is at first glance surprising . However , insects have been shown to display many cognitive traits in parallel with humans ( Brosnan and De Waal , 2003; Aw et al . , 2011; Premack and Hillix , 1962; Dukas , 2005; Reid and Stamps , 1997 ) , and contrast effects are likely selected for ( McNamara et al . , 2013 ) . The smaller effect size in experiment two is presumably driven either by the exclusion of the additional driving factors ( see experiment two description ) , or the additional complications involved in an extensive training regime , or both . Specifically , the expectations leading to contrast effects in experiment two were driven by differential learning of odour-quality associations , rather than a simple one-component memory of food quality as may have been the case in experiment 1 . This may have weakened the observed effect . Another possible explanation for smaller effect sizes may be that in experiment 2 ants had access to two reference points ( 0 . 25M and 1 . 5M ) to use for value judgement of the medium-quality food in the control experiment , while in experiment 1 they only had one reference . Thus , while the odour cue may have overemphasised the role of the associated quality as reference , the competing reference quality may have been acting as a second reference . Additional reference points are likely to affect the scale post-shift rewards are compared to ( Zwislocki , 2009 ) . This possibility is supported by the acceptance data collected during training in experiment 2 . On the first training visit , all ants encountered low quality food and showed a high food acceptance towards the feeder ( Figure 4A ) . However , as soon as ants had experienced a high-quality sucrose solution , the previously acceptable low-quality food became unattractive , and food acceptance scores decreased from a mean of 0 . 99 to 0 . 39 . This strongly suggests that the ants were valuing the training solutions in relation to each other , and may therefore have used both reference points to judge the value of an unscented medium-quality food source . It is possible that the ants may have calculated an average from both reference points , and used the average as comparison to judge the value of the post-shift reward ( Flaherty , 1999 ) , as shown in rats ( Peters and McHose , 1974 ) . However , the fact that medium quality elicited different food acceptance scores depending on the runway scent makes it unlikely that this would be the only factor affecting acceptance scores . Lastly , masking effects may also explain the smaller contrast effects of experiment 2 compared to experiment 1: learning theory suggests that neutral cues associated to positive stimuli will elicit positive responses even when no reward is given and vice versa ( Rescorla and Wagner , 1972 ) . Therefore , since ants were confronted with the scent associated to high-quality food , food acceptance may have been affected by the scent itself , leading to an elevated food acceptance compared to ants tested in experiment 1 which did not receive a positive cue , but only medium-quality food . The reduced pheromone deposition seen in the final return in experiment 2 may be due to the change in environment ( scented runways to unscented runways ) causing a disruption in recruitment behaviour , perhaps due to generalisation decrement ( Kimble , 1961; Capaldi , 1978 ) or neophobia ( Barnett , 1958; Johnson , 2000; Mitchell , 1976; Pliner and Loewen , 1997 ) . Furthermore , since only the scented paper overlays were replaced by unscented ones , but not the runways themselves , it is possible that small portions of the odours were still present , driving the ants to deposit pheromone according to the remaining odours , with higher deposition rates for the high-quality associated odour . In a separate experiment , such pheromone deposition directly related to quality-associated odours on runways was clearly demonstrated ( Wendt and Czaczkes , 2019 ) . This would explain why pheromone depositions were higher for ants returning to the nest from a high molarity scent than in ants returning from a low molarity scent . Information about sucrose concentrations gained through trophallactic interactions inside the nest can affect the way newly discovered food sources are valued outside the nest ( Figure 5 ) , as well as providing other information ( Provecho and Josens , 2009; Josens et al . , 2016; LeBoeuf et al . , 2016 ) . By taking into account information gained inside the nest , recruited workers are able to evaluate newly discovered food sources in relation to other food sources available in the environment . Ants could thus forego food sources which are of lower quality than the average available food sources ( Wendt and Czaczkes , 2017 ) . Even though higher trophallaxis times led to lower acceptance scores and trophallaxis times were higher at high reference molarities , this does not necessarily imply that ants ingested more sucrose at higher references and were thus less hungry or motivated . Higher sucrose solutions are more viscose and thus ants take longer to ingest the same amount of sucrose compared to low molarities ( Josens et al . , 1998 ) . If , however , more sucrose solution was transferred between the returning forager and the recruit at longer trophallaxis times , it is likely that information input increases and food acceptance decreases . The longer the trophallaxis time , the more the recruit can fill its crop through trophallaxis and therefore the food acceptance may decrease , because the recruit is less starved than an ant which showed a short trophallaxis time . However , even if more food was transferred , the food acceptance scores are a measure of the first assessment of ants at a food source , not the ingested volume . Thus , while some ants may have had less space in their crop left , this may not necessarily affect the food acceptance score , while it is very likely to affect ingested volume after trophallaxis . Additionally , if longer trophallaxis times lead to more ingested sucrose solution , it is also more likely that a higher amount of information about the past food quality is transferred . Thus , more transferred food during trophallaxis may have led to better informed ants reaching the post-shift solution and thus stronger contrast effects . Since the data shows clear effects of both trophallaxis time and reference solution on the food acceptance of 0 . 5M sucrose , longer trophallaxis times cannot be the only factor driving the contrast effects found in this experiment ( see Figure 5—source data 1 and Figure 5—figure supplement 1 ) . Even at high trophallaxis times , ants with a 0 . 16M reference showed no low food acceptance scores , unlike ants with high reference solution after long trophallaxis times . Ultimately , we see the nest serving as an information hub , in which information about currently available food sources can be collected , synthesised , and fed back to outgoing foragers . Relative value perception can therefore be expected to have strong effects not only on the individual behaviour of animals , but also on the collective behaviour of insect colonies . For example , colonies of house-hunting ants developed an aversion towards mediocre nests when housed in high-quality nests , but not when they were housed in low-quality nests . Such mediocre nests are then avoided when colonies have to find a new nest site while newly discovered mediocre nests are readily accepted ( Stroeymeyt et al . , 2011; Robinson et al . , 2011 ) . However , while in house-hunting the reference resource is directly experienced by scouts only , we demonstrate that information brought back to the nest can set a reference point for ants which have not directly experienced the resource in situ . A broad range of behaviours relevant to behavioural economics have been described in invertebrates ( Czaczkes et al . , 2018a; Pompilio et al . , 2006; Wendt and Czaczkes , 2017; Czaczkes et al . , 2018b; Cheng et al . , 2002 ) . We propose that invertebrates make attractive models for a broader understanding of behavioural economics in humans . The benefits of an interdisciplinary approach will likely flow both ways . Using animal models allows researchers to avoid pitfalls associated with studies on humans , such as cultural and educational differences ( Carter and Irons , 1991; Guiso et al . , 2006 ) , second-guessing of experimenters , and non-relevant reward sizes ( Levitt and List , 2007 ) as well as relaxing ethical concerns . The game-like designs of many economic experiments are highly artificial and the incentive magnitudes that can be provided are limited ( Kahneman and Tversky , 1979; Levitt and List , 2007 ) . While there has been much progress in field studies on humans to clearly measure causal relationships ( Harrison and List , 2004 ) , the usefulness of these new techniques is constrained by the range of questions and settings to which they can be applied . Hence , while behavioural studies on invertebrates also have their limitations ( for example , in that inducing expectations is more of a challenge ) , they can be easily designed to be ecologically meaningful , and offer rewards which are in line with the real-life budgets under which the animals operate . Finally , due to human complexity , building economic models which accurately predict human behaviour is challenging . Insect economic behaviours are demonstrably similar to that of humans , but likely simpler . We therefore propose that economic models to predict invertebrate decision making may be a complementary step on the way to predicting human behaviour . There is a well-developed tradition of integrating economics and biology ( Aw et al . , 2011; Czaczkes et al . , 2018a; Lydall et al . , 2010; Aw et al . , 2009; Wendt and Czaczkes , 2017; Cheng et al . , 2002; Evans and Westergaard , 2006 ) . Here we provide a systematic description of value judgment relative to a reference point in ants , define a relative value curve as described in Prospect Theory , and provide some of the first strong evidence for a purely cognitive element to relative value judgement . Reference points can not only be set by individual experiences but also through social information such as pheromone trails or through trophallactic contacts inside the nest . We feel a critical mass of evidence is now available to consider comparative behavioural economics as a relevant discipline for both biologists and economists .	We make many decisions every day , often by comparing options and choosing the one with the greatest profit . But how much we value something often does not depend solely on our needs . Instead , this value may depend on our expectations or other arbitrary reference points . For example , how satisfied you are with your income might depend on how much your colleagues or friends earn . Animals , including insects , also make decisions when feeding , choosing a partner , or finding a nesting site . Sometimes animals behave in ways that look like disappointment . For example , monkeys may reject a cucumber as a reward if they have seen another monkey get a grape for completing the same task . But it is hard to tell if this behavior reflects a value judgment . To investigate whether insects evaluate their options against their expectations , Wendt et al . offered black garden ants sugar water over multiple trials . Some ants grew to expect low quality sugar water ( containing little sugar ) ; some expected medium quality; and others expected high quality sugar water ( containing a high concentration of sugar ) . Ants that expected to find low quality sugar water were more likely to accept medium quality options than ants that expected the medium quality sugar water . Similarly , ants that expected high quality sugar water were less likely to accept lower quality sugar water . Further experiments confirmed that the ants were not using physical cues such as satiation to guide their behavior . Furthermore , Wendt et al . found that ants that returned to the nest after foraging passed on information that altered the expectations of the next group of foragers about nearby food . This suggests that the value that ants place on food sources depends both on individual experiences and on information gained from others . Studies of decision making in humans can be difficult to perform and interpret , because volunteers may try to second-guess what the experimenters want to find , and culture and education may also influence choices . Studying ants instead could help to avoid these pitfalls , as the results presented by Wendt et al . suggest they make decisions in similar ways to humans . Future work building on these findings could also help researchers to predict how insects behave , particularly in rapidly changing environments .	[ "Abstract", "Introduction", "Materials", "and", "methods", "Experiment", "1", "–", "Defining", "a", "relative", "value", "perception", "curve", "Experiment", "2", "–", "ruling", "out", "alternative", "explanations", "using", "scent", "training", "Experiment", "3", "–", "expectation", "setting", "via", "trophallaxis:", "the", "nest", "as", "an", "information", "hub", "Discussion" ]	[ "ecology", "evolutionary", "biology" ]	2019	Positive and negative incentive contrasts lead to relative value perception in ants
Understanding the extent to which enzyme evolution is reversible can shed light on the fundamental relationship between protein sequence , structure , and function . Here , we perform an experimental test of evolutionary reversibility using directed evolution from a phosphotriesterase to an arylesterase , and back , and examine the underlying molecular basis . We find that wild-type phosphotriesterase function could be restored ( >104-fold activity increase ) , but via an alternative set of mutations . The enzyme active site converged towards its original state , indicating evolutionary constraints imposed by catalytic requirements . We reveal that extensive epistasis prevents reversions and necessitates fixation of new mutations , leading to a functionally identical sequence . Many amino acid exchanges between the new and original enzyme are not tolerated , implying sequence incompatibility . Therefore , the evolution was phenotypically reversible but genotypically irreversible . Our study illustrates that the enzyme's adaptive landscape is highly rugged , and different functional sequences may constitute separate fitness peaks . The controversy surrounding evolutionary reversibility pertains to one of the fundamental questions in evolutionary biology: the extent to which selection pressure determines evolutionary outcomes ( Teotonio and Rose , 2001; Gould , 2007; Collin and Miglietta , 2008; Lobkovsky and Koonin , 2012 ) . Also , through understanding reversibility on the levels of both phenotype and genotype , one could catch a glimpse at the structure of the respective fitness ( or adaptive ) landscape . The extent of ruggedness of adaptive landscapes—that is , the prevalence of epistasis , and thus historical contingency—have recently received considerable attention ( Whitlock et al . , 1995; Poelwijk et al . , 2007; de Visser et al . , 2011; Breen et al . , 2012; Harms and Thornton , 2013; McCandlish et al . , 2013; Kaltenbach and Tokuriki , 2014 ) . While the unlikelihood of reversing a historical pathway taken by evolution has been demonstrated ( Bridgham et al . , 2009 ) , a large number of sequences can encode functionally identical proteins ( ‘genotypic redundancy’ ) and phenotypic reversion can still occur via alternative pathways ( Clarke , 1985; Lenski , 1988; Crill et al . , 2000; Teotonio and Rose , 2000; Kitano et al . , 2008 ) . Yet , the evolutionary dynamics underlying phenotypic reversion have not been addressed . Does phenotypic reversion lead back to the ancestral peak on the adaptive landscape or to a new peak ( Carneiro and Hartl , 2010; Lobkovsky and Koonin , 2012 ) ? In other words , to what extent are the sequences of the ancestral and reverse-evolved proteins accessible via a neutral network—that is , are amino acid exchanges between the two proteins tolerated or result in loss of function ? The inability to exchange amino acids between homologous proteins due to epistasis represents ‘genotypic incompatibility’ and can result in a non-functional enzyme , a phenomenon which can be compared to the ‘Dobzhansky-Muller effect’ of hybrid incompatibility ( Orr , 1995; Kondrashov et al . , 2002 ) . Another important aspect to be explored is the underlying molecular mechanism of phenotypic reversibility . Restoration of function can either be brought about by the same structure and mechanism as in the ancestor , or by a distinct , alternative state . Structural convergence would indicate that functional requirements exist , which deterministically lead to one particular structural solution . On the other hand , structural divergence would imply the accessibility of various solutions that can bring about efficient catalysis . Thus , understanding the molecular basis for ( ir ) reversibility and ( in ) compatibility would provide valuable insights into protein sequence-function-structure relationships . What are the molecular requirements for a specific function ? What structural changes are required to switch from one function to another ? Identifying such changes , which are often based on subtle effects ( e . g . , on mutations occurring in remote locations , or mutations which only show a favorable effect in combination ) , remains a great challenge in protein science . What is the molecular basis underlying mutational epistasis , which leads to alternative evolutionary outcomes ? Directed evolution is a powerful tool to address these questions and explore adaptive landscapes because it allows the study of evolution in a highly controlled setup ( Peisajovich and Tawfik , 2007; Romero and Arnold , 2009; Kawecki et al . , 2012 ) . High selection pressure can prevent fixation of neutral , functionally irrelevant mutations , resulting in an adaptive trajectory without mutational noise . All evolutionary intermediates ( the ‘molecular fossil record’ ) are obtained , so the evolutionary dynamics and their molecular basis can be characterized in detail . Performing evolution in both the forward and reverse direction and comparing the changes in each direction provides a unique handle for identifying such effects . Understanding these phenomena would improve our ability to design and engineer novel proteins in the laboratory . Here , we experimentally test the reversibility of enzyme evolution and investigate its molecular basis . We previously evolved the enzyme PTE , a phosphotriesterase , into an arylesterase ( Roodveldt and Tawfik , 2005; Tokuriki et al . , 2012; Wyganowski et al . , 2013 ) . In this work , we applied a selection pressure to restore the original phosphotriesterase activity . We characterized the entire trajectory including both the forward and reverse process in terms of phenotypic reversibility ( function or enzymatic activity ) , genotypic irreversibility ( sequence ) , as well as in terms of the underlying molecular basis ( structure and mechanism ) . We find that PTE has a rugged adaptive landscape on which the accessibility of functional mutations is severely limited , and describe the mechanisms that lead to genotypic irreversibility and incompatibility . We previously reported the laboratory evolution of PTE ( wtPTE ) into a highly efficient arylesterase for 2-naphthyl hexanoate ( 2NH ) ( Roodveldt and Tawfik , 2005; Tokuriki et al . , 2012; Wyganowski et al . , 2013 ) . In the course of the trajectory , the original phosphotriesterase activity decreased drastically ( 104-fold ) although no selection pressure was applied against it . In this work , we first completed the functional transition by further decreasing the remaining phosphotriesterase activity ( ∼10-fold ) by four additional rounds of directed evolution for maintaining arylesterase but reducing phosphotriesterase activity ( Supplementary file 1 ) . Briefly , libraries were generated by error-prone PCR and transformed into Escherichia coli ( BL21 ( DE3 ) ) . As a pre-screen for arylesterase activity , protein expression was induced in the bacterial colonies on agar plates , and a mixture of the substrate 2NH and a product stain ( Fast Red ) was added as previously described ( Figure 1A ) ( Roodveldt and Tawfik , 2005; Tokuriki et al . , 2012; Wyganowski et al . , 2013 ) . Upon hydrolysis of 2NH , Fast Red forms a red complex with the naphtholate leaving group , meaning colonies that develop a red color contain active arylesterase variants . In each round , 2000–10 , 000 colonies were screened in this fashion , theoretically covering most single point mutations in the 330 amino acid PTE gene . Positive colonies were then re-grown and re-assayed in 96-well plates and initial rates of both 2NH and paraoxon hydrolysis were determined in clarified lysate . In our experience , activity increases >1 . 3-fold compared to the respective parent yielded reliably improved variants . The variant with the largest improvement in initial rate was then used as the template for the next round of error-prone PCR or several variants were subjected to DNA shuffling . To buffer the destabilizing effects of functional mutations and minimize reductions in soluble protein expression levels , we used GroEL/ES overexpression as previously described ( Supplementary file 1 ) ( Tokuriki and Tawfik , 2009; Wyganowski et al . , 2013 ) . In total , with 22 rounds of ‘forward evolution’ , the accumulation of 26 mutations from wtPTE resulted in a highly efficient and specialized arylesterase ( AE ) with a ∼105-fold increase in arylesterase rates ( kcat/KM for 2NH >106 M−1s−1 ) and an overall ∼105-fold decrease in phosphotriesterase activity ( kcat/KM for paraoxon ≈102 M−1s−1 , Figure 1B , C ) . Because selection was specific for arylester hydrolysis until round 18 , the change in phosphotriesterase activity was stochastic: many mutations decreased phosphotriesterase activity ( 11 mutations ) , some were neutral ( nine mutations ) , and others increased phosphotriesterase activity ( six mutations ) . Starting from AE , we then performed the reverse evolution to restore phosphotriesterase activity using an experimental setup equivalent to the forward process ( Figure 1A ) with the following modifications: the pre-screen was carried out using a fluorogenic phosphotriester as a surrogate for paraoxon ( Supplementary file 1 ) and then validated in 96-well format as described above . The selection criterion was now an increased initial rate of paraoxon hydrolysis . In our evolutionary model system , variant fitness is defined as the level of enzymatic activity in cell lysate . All variants were also purified and the kinetic parameters determined , which correlated well with lysate activity ( Figure 1—figure supplement 1 , Supplementary file 2 ) . 10 . 7554/eLife . 06492 . 003Figure 1 . Activity and sequence changes of PTE over the evolution . ( A ) Overview of the experimental evolution . Libraries were generated and transformed into Escherichia coli . Proteins were expressed and screened for paraoxon and/or 2NH hydrolysis in bacterial lysates . Several thousand variants were screened per round , theoretically covering most single point mutations in the ∼1000 bp PTE gene . Details are given in Supplementary file 1 . ( B ) Activity changes during the forward ( screening for arylesterase hydrolysis ) and reverse evolution ( screening for re-increase in phosphotriesterase hydrolysis ) . Steady-state kinetic parameters for all variants are provided in Supplementary file 2A . ( C ) Type , position , and order of occurrence of the 33 mutations obtained in the evolution . Mutations are shown relative to wtPTE ( GenBank accession number KJ680379 ) with lower case italics denoting the amino acid found in wtPTE . Note that wtPTE was obtained in previous screens for improved expression levels in E . coli and contains six mutations relative to the naturally occurring PTE ( I106L , F132L , K185R , D208G , R319S ) ( Roodveldt and Tawfik , 2005; Tokuriki et al . , 2012 ) . The following mutations occurred in individual variants , but were not fixated after DNA shuffling: R7a: a204G , R7c: a102V , R19: a78T , v143A , t311A , revR1: c59Y , s238R , revR5: i176V , revR8a: d264E , revR8b: i296V . All additional variants characterized and sequenced in each round are shown in Supplementary file 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 06492 . 00310 . 7554/eLife . 06492 . 004Figure 1—figure supplement 1 . Correlation between activities measured in cell lysate and using purified enzyme for all variants selected over the evolution ( Supplementary file 2 ) . ( A ) Phosphotriesterase activity . ( B ) Arylesterase activity . All measurements were carried out at 200 μM substrate . Activities in cell lysate are given relative to wtPTE . DOI: http://dx . doi . org/10 . 7554/eLife . 06492 . 004 The restoration of phosphotriesterase activity in the reverse evolution followed a pattern similar to that observed for arylesterase activity in the forward evolution: increasing smoothly and gradually through the stepwise accumulation of mutations ( Figure 1B , C ) . Moreover , it followed a ‘diminishing returns’ pattern characteristic for the development of a function under selection—that is , the activity gain per mutation gradually decreased in later stages of the functional transition , where fitness reached a plateau ( Figure 1B ) ( Stebbins , 1944; MacLean et al . , 2010; Chou et al . , 2011; Khan et al . , 2011; Tokuriki et al . , 2012 ) . Furthermore , trade-offs between the two activities were weak in the early rounds of evolution , resulting in a generalist , bifunctional intermediate ( Aharoni et al . , 2005; Khersonsky and Tawfik , 2010 ) . In the forward evolution , trade-offs then became stronger , leading to specialization of the arylesterase . The reverse evolution , however , retained characteristics of a generalist: the large increase in phosphotriesterase activity ( >104-fold ) was accompanied by only a small ( five-fold ) reduction in arylesterase activity . A possible reason for this is that the reverse evolution is still at an early phase of the functional transition after 12 rounds ( vs . 22 in the forward evolution ) . Because we were unable to isolate any variant with further improved phosphotriesterase activity , it might be necessary to impose a negative selection pressure to specialize the enzyme . The molecular basis of substrate binding and trade-offs is described further below ( see also Figure 2 and Figure 2—figure supplement 1 ) . Overall , we obtained a new efficient , enzyme ( neoPTE ) on par with wtPTE ( kcat/KM>106 M−1s−1 for paraoxon in both cases ) . The recovery of identical phosphotriesterase rates in neoPTE compared to wtPTE establishes that evolution of the phenotype was fully reversible . 10 . 7554/eLife . 06492 . 005Figure 2 . Reshaping of the PTE active site over the evolution . ( A ) WtPTE ( PDB ID: 4PCP ) features an active site which is well adapted for paraoxon hydrolysis , but suboptimal for 2NH . ( B ) In the forward evolution , selection for arylesterase activity leads to several changes in the binding pocket from wtPTE to AE ( PDB ID: 4PCN ) . ( C ) The reverse evolution leads to restoration of the ancestral state in neoPTE ( PDB ID: 4PBF ) . The four regions of change are highlighted in different colors . Top row: the 2NH analogue ( yellow ) was modeled into the three structures by superposition with PTE-R18 in complex with the analogue ( PDB ID: 4E3T ) ( Tokuriki et al . , 2012 ) . Bottom row: the paraoxon analogue diethyl 4-methoxyphenyl phosphate ( yellow ) was modeled into the structures by superposition with Agrobacterium radiobacter PTE in complex with the analogue ( PDB ID: 2R1N ) ( Hong and Raushel , 1996 ) . Amino acids found in wtPTE are shown in lower case italics . DOI: http://dx . doi . org/10 . 7554/eLife . 06492 . 00510 . 7554/eLife . 06492 . 006Figure 2—figure supplement 1 . Details of the active site changes . By overlaying the structures of wtPTE , AE , and neoPTE with the structure of A . radiobacter PTE in complex with the paraoxon analogue diethyl 4-methoxyphenyl phosphate ( see Figure 2 ) , regions important for paraoxon binding could be identified and the effect of mutations derived . The loss of phosphotriesterase in AE and its restoration in neoPTE is achieved mainly by changes in shape complementarity between enzyme and substrate , changes in hydrophobicity , and π-π stacking ( A–C ) . It is likely that the movement of the β-metal also influences catalysis , although the exact effects of the metal displacement on catalysis are as yet unclear ( D ) . ( A ) Interaction between paraoxon and residues 306 and 308 . Substitution of the bulky Phe306 by Ile improves 2NH binding in AE , but results in a loss of interaction with paraoxon . In neoPTE , rather than reversion of f306I , s308 is mutated to the more hydrophobic Cys , improving interaction with the para-nitrophenyl group ( Figure 2 pink region , Figure 7A ) . ( B ) Steric hindrance between Phe271 and paraoxon . Substitution of Leu271 to the larger Phe improves 2NH binding in AE , but causes steric hindrance with paraoxon . In neoPTE , in addition to reversion to the ancestral Leu , repositioning of the loop through a combination of remote mutations results in a ‘downward’ movement of Leu271 , further enlarging the pocket ( see also Figure 2 orange region , Figure 7A ) . ( C ) Shift in position of Leu106/Trp131/Leu132 ( Figure 2 purple region , Figure 7B ) . While wt- and neoPTE feature edge-to-face π-π stacking between Trp131 and the para-nitrophenyl ring , in AE the shift in position brings Trp131 closer to the partially positive edges of the ring , resulting in electrostatic repulsion . Moreover , the shift in Leu106 brings it closer to the ethoxy group of the substrate in AE . ( D ) Shift in position of the β-metal . The inter-metal distance is reduced from 3 . 8 Å in wtPTE to 3 . 3 Å in AE through a movement of His201 and the β-metal . In neoPTE , the original spacing seen in wtPTE is restored ( Figure 2 light blue region , Figure 7C ) , perhaps because the decreased distance in AE destabilized the transition state for paraoxon hydrolysis . DOI: http://dx . doi . org/10 . 7554/eLife . 06492 . 00610 . 7554/eLife . 06492 . 007Figure 2—figure supplement 2 . Overlay of electron density maps for the active sites of ( A ) wtPTE ( salmon ) and AE ( cyan ) and ( B ) wtPTE ( salmon ) and neoPTE ( magenta ) . Electron density for wtPTE is shown as grey isosurface ( 2 s ) , while electron density of AE and neoPTE is shown as isomesh ( 2 s ) , for contrast between the two maps . Comparison between ( A ) and ( B ) illustrates the shift of the β-metal ion closer to the α-metal ion in AE , and the shift back to the wtPTE position in neoPTE . Likewise comparison between the positions of Leu106 , Trp131 and Leu132 in ( C ) wtPTE ( salmon ) and AE ( cyan ) and ( D ) wtPTE ( salmon ) and neoPTE ( magenta ) illustrates that these sidechains adopt alternative positions in AE , but return to their original conformations in neoPTE . DOI: http://dx . doi . org/10 . 7554/eLife . 06492 . 00710 . 7554/eLife . 06492 . 008Figure 2—figure supplement 3 . Development of B-factors over the evolution . In wtPTE , loop 7 shows the maximum B-factor . The forward evolution for 2NH activity resulted in stabilization of loop 7 , whereas flexibility of loops 4 and 5 increased . In neoPTE , the original dynamics of the structure were restored as shown by the increased flexibility of loop 7 as well as the reduced B-factor of loops 4 and 5 . DOI: http://dx . doi . org/10 . 7554/eLife . 06492 . 00810 . 7554/eLife . 06492 . 009Figure 2—figure supplement 4 . Linear free energy relationships of wtPTE , AE , and neoPTE . ( A ) Arylester hydrolysis . The kcat/KM of all three variants is independent of the leaving group pKa . ( B ) Phosphotriester hydrolysis . In neoPTE , the break in leaving group dependence around pH 7 , which is characteristic for wtPTE ( Hong and Raushel , 1996; Tokuriki et al . , 2012 ) , is restored . Information about each substrate is provided in Supplementary file 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 06492 . 009 To examine the genetic changes causing phenotypic reversion , the sequence of all evolutionary intermediates was determined . Only five of the 26 mutations that accumulated in the forward evolution were reverted to the original sequence ( ‘reversions’ , A49v , I172t , Q180h , L271f , M314t , Figure 1C; amino acids shown in lower case italics denote the wtPTE state , while amino acids not present in the wild type are shown in capital letters ) . Nine additional ‘new mutations’ accumulated , two of which occurred in positions that were mutated in the forward evolution ( V130M—originally leu , I306M—originally phe ) , and seven were in positions that were not previously mutated ( p135S , y156H , g174D , a203E , m293K , s258N , s308C ) . Overall , neoPTE is separated further from wtPTE ( 28 out of 333 amino acids ) than AE from wtPTE ( 26 amino acids ) . Additional rounds of evolution failed to yield more reversions or activity increases ( Supplementary file 1 ) . In the forward evolution , the loss of phosphotriesterase activity was largely a side product of the property under selection , the increase in arylesterase activity . Therefore , not all mutations decreased phosphotriesterase activity ( Figure 1B ) , and it is not surprising that phenotypic reversion did not require full genotypic reversion . However , a number of mutations that did contribute to decreasing phosphotriesterase in the forward process were also not reverted . Moreover , the new mutations are located in the same mutational clusters seen in the forward evolution ( Figure 1C ) , indicating they may be alternative solutions to the same functional requirement and replace reversions , as detailed further below . Taken together , although the phenotype was reversible , PTE evolution was genotypically irreversible , but an alternative trajectory was readily taken . To unravel the molecular basis of the observed genotypic irreversibility , we solved crystal structures of wtPTE , AE , and neoPTE ( Supplementary file 3 ) . We compared the structures and modeled both a paraoxon and a 2NH analogue into each active site ( by superposition with structures containing these analogues [Hong and Raushel , 1996; Tokuriki et al . , 2012] ) . The phosphotriester paraoxon is characterized by tetrahedral ground-state geometry and P–O cleavage proceeds via a trigonal bipyramidal transition state . The arylester 2NH is planar and C–O bond hydrolysis proceeds via a tetrahedral transition state . The structural comparison indicates that AE adapted to the planar substrate 2NH in the forward evolution , but that this came at a cost of phosphotriesterase activity , as the bulky paraoxon is no longer efficiently recognized ( Figure 2 ) . We identify several regions of the active site that may be responsible for the functional transition ( Figure 2 and Figure 2—figure supplement 1 ) . First , a binding pocket for the naphthyl leaving group of 2NH was excavated through the combined action of h254R and d233E ( Figure 2A , B , green region ) ( Hong and Raushel , 1996; Tokuriki et al . , 2012 ) . Leaving group coordination was further improved through a subtle ∼1 . 0 Å shift of Trp131 ( Figure 2A , B , purple region ) . Moreover , the pocket was elongated through the f306I mutation ( Figure 2A , B , pink region ) and narrowed by l271F ( Figure 2A , B , orange region ) , resulting in better accommodation of the long hexanoate chain of 2NH . These changes may lead to the reduction of phosphotriesterase activity through loss of interactions ( either shape complementarity , hydrophobicity , or π-π stacking ) in several regions and steric hindrance in others , as described in further detail below ( Figure 2—figure supplement 1 ) . Additionally , the distance between the two active site zinc ions decreased from 3 . 8 Å to 3 . 3 Å ( Figure 2A , B , light blue region and Figure 2—figure supplement 1 ) . The observed structural changes are subtle , at the sub-angstrom scale , and their contributions to catalysis unquantified . However , the dispersion precision indicator ( DPI; Cruickshank , 1999 ) for each of the structures is less than one-tenth of an angstrom , meaning that the observed distance changes ( including the 0 . 5 Å shift in the metal position ) are significant ( Figure 2—figure supplement 2 ) . In neoPTE , the part of the active site necessary for phosphotriesterase activity has converged back towards its original state . The regions of suboptimal binding were re-optimized for paraoxon and the metal distance was restored to the 3 . 8 Å ( Figure 2C ) . Moreover , the pattern of loop flexibility that is characteristic of wtPTE was also restored in neoPTE ( Figure 2—figure supplement 3 ) . Furthermore , we measured linear free energy relationships for wtPTE , AE , and neoPTE for both arylester and phosphotriester hydrolysis ( Figure 2—figure supplement 4 ) , that is , the dependence of the catalytic parameters kcat/KM on the pKa of the leaving group . For phosphotriester hydrolysis by wtPTE , a break in pKa dependence around 7 is consistent with the rate-limiting step changing on either side of this break ( Hong and Raushel , 1996; Tokuriki et al . , 2012 ) . By contrast , AE shows a continuous , linear dependence over the whole pKa range , indicating that the rate-limiting step does not change . In neoPTE , the pattern characteristic for wtPTE was restored . Together with the observed structural convergence , the simplest assumption must be that the very similar active site environment enables similar residue contributions to catalysis in wt- and neoPTE on phosphotriesterase activity . However , active site convergence is not complete , as the naphthyl binding pocket remains intact ( Figure 2C , green region , Arg254 and Glu233 ) , which likely explains why neoPTE is still bifunctional . It should be pointed out that , at this stage , we do not know the extent to which the modification of each structural element contributes to the overall >104-fold activity change . Also , we cannot exclude the existence of alternative substrate binding modes from our model , as well as the role protein dynamics play in the functional switch . However , in the combined forward and reverse evolution , which involved a change in catalytic activity of >104 M−1s−1 in each direction , only four mutations were located in the active site . Instead , most functional mutations occur in more remote positions . Therefore , it is likely that fine-tuning of the active site by these remote mutations contributes significantly to the activity changes . Taken together , the restoration of all structural elements key for phosphotriesterase activity as well as the catalytic mechanism occurred despite the alternative genotypic trajectory , suggesting that biophysical requirements exist for this particular active site shape , and that phosphotriesterase activity may otherwise be inefficient . To further investigate whether mutational accessibility is dictated by the necessity for structural convergence to the wild-type active site , a parallel evolutionary experiment was performed . In this experiment , we attempted to restore phosphotriesterase activity by a trajectory containing only new mutations . To this end , we sequenced the improved variants after each round and removed all those containing reversions . This trajectory only resulted in a 70-fold improvement in five rounds ( Figure 3A ) , after which the activity plateaued and no further improved variants could be found . This failure to reach wild-type activity levels without reversions , as well as the fact that three out of the five new mutations obtained ( p135S , a203E , s308C , Figure 3B ) were identical to the successful trajectory containing reversions , emphasizes that the number of adaptive trajectories that lead to a wild-type level fitness peak from AE are highly limited . However , trajectories involving neutral mutations , or trajectories which do not pass through the best variant in each round but through less improved intermediates , may exist . It is likely that a wild-type-like paraoxon binding pocket is compulsory to achieve efficient phosphotriesterase activity , and only a small set of mutations ( e . g . , reversions or the combination of reversions and new mutations that we identified ) can provide such a solution . 10 . 7554/eLife . 06492 . 010Figure 3 . An alternative experimental evolution , where fixation of back-to-wild-type reversions was prohibited , failed to restore the original level of PTE activity . ( A ) Activity changes in the alternative trajectory . After five rounds , PTE activity plateaued at a 65-fold improvement ( trajectory 2 ) , 340-fold lower than the main trajectory ( trajectory 1 ) . ( B ) Mutations accumulated in the alternative trajectory . All clones containing reversions , which occurred frequently , were removed after sequencing and thereby prohibited from fixating . Three of the five new mutations also fixated in the main trajectory . The mutations c59Y and s238R occurred in variants revR1 and revR2b , but were not fixated after DNA shuffling . Amino acids found in wtPTE are shown in lower case italics . All additional variants characterized and sequenced in each round are shown in Supplementary file 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 06492 . 010 Next , we set out to answer the question how the two enzymes exhibiting identical phosphotriesterase activity , wt- and neoPTE , are connected on the adaptive landscape . If they populate the same fitness plateau , amino acid exchanges between them should be neutral . On the other hand , a loss of function upon interconversion between amino acids would indicate genotypic incompatibility ( Kondrashov et al . , 2002; Lunzer et al . , 2010; Wellner et al . , 2013 ) , meaning that the two enzymes occupy distinct positions on the landscape that are poorly connected through a neutral network . To this end , we characterized the effect of all 28 single point exchanges separating the two enzymes in each background ( 56 mutants in total , Figure 4 and Supplementary file 2 ) . Mutations are considered non-neutral if they cause a >1 . 3-fold change in phosphotriesterase activity in lysate compared to the parent background because , in our screening system , this cut-off enabled us to reliably identify improved variants . Moreover , we have performed a statistical analysis of the mutational effects , which confirms that a >1 . 3-fold change is significant ( p-values <0 . 05 ) in almost all cases ( statistics are provided in Figure 4—source data 1 and Supplementary file 2B ) . According to this analysis , only eight of 28 exchanges were compatible; they were neutral in both backgrounds ( Figure 4A ) . The remaining 20 positions showed incompatibility , 15 of which were partially incompatible , as the exchange was neutral in one background but deleterious in the other ( Figure 4B , C ) . Five exchanges were completely incompatible; they severely decreased activity in both backgrounds ( Figure 4D ) . Taken together , despite >90% sequence identity between wt- and neoPTE , the reverse trajectory led to a functional sequence that is poorly connected with the original one . It remains unknown whether the two sequences comprise completely separate peaks on the adaptive landscape or are connected through a neutral network , that is , if the neutral exchanges would permit the subsequent occurrence of initially deleterious exchanges . However , because >70% of the mutated positions cause incompatibility , only one out of the 54 exchanges confers higher fitness , and this exchange ( neoPTE + f306M ) would require two simultaneous base changes , it is unlikely that an evolutionary transition between the two could easily occur by adaptive or strong purifying selection . 10 . 7554/eLife . 06492 . 011Figure 4 . Genotypic incompatibility between wtPTE and neoPTE . ( A–D ) The effect of the 28 amino acid exchanges separating the two enzymes was tested in the background of wtPTE and neoPTE , respectively . Activities are given relative to the parent mutational background , wtPTE or neoPTE . Amino acids found in wtPTE are shown in lower case italics . Color code as in Figure 1 . ( A ) Compatible exchanges , neutral in both backgrounds . ( B , C ) Partially incompatible exchanges , neutral in one background but deleterious for another . ( D ) Mutually incompatible exchanges . Mutations causing a >1 . 3-fold change compared to the respective parent mutant ( dotted line ) are considered non-neutral . p-values compared to each parent ( Supplementary file 2B ) and p-values for the effect of each mutation Figure 4—source data 1 . in the two backgrounds were calculated . Note that the effect of i313F , which causes a significant decrease in wtPTE , is statistically not significant between wtPTE and neoPTE . DOI: http://dx . doi . org/10 . 7554/eLife . 06492 . 01110 . 7554/eLife . 06492 . 012Figure 4—source data 1 . Comparison of the effect of mutations in wt- and neoPTE . Fold-changes between the two backgrounds as well as p-values calculated according to the t-test are given . [a] Only mutations with an average >1 . 3-fold difference between backgrounds AND a p-value <0 . 05 are considered significant . Non-significant values are underlined . These belong to the ‘PANEL A’ series , which consists of mutations that are neutral in both cases . The only exception is i313F from PANEL C , which is significantly different from wtPTE , but not significantly different between wt- and neoPTE . DOI: http://dx . doi . org/10 . 7554/eLife . 06492 . 012 To understand how convergence to the original function and architecture was achieved despite genotypic irreversibility and incompatibility , we performed a comprehensive mutational analysis . All 33 mutated positions were examined in the background of the three enzymes ( wtPTE , AE , and neoPTE ) , and in the background in which they originally occurred in the evolution ( i . e . , in the different rounds ) to identify mutations that are epistatic—that is , change their effect depending on the genetic background ( Figure 5 and Supplementary file 2 ) . To determine whether or not the measured changes were significant , the same stringent cut-off as described above for the comparison between wt- and neoPTE was applied ( statistics are provided in Supplementary file 2B–G ) . Furthermore , we analyzed the crystal structures to determine which mutations caused the divergence and convergence of the active site configuration . 10 . 7554/eLife . 06492 . 013Figure 5 . Changes in phosphotriesterase activity upon mutations in five different backgrounds: wtPTE in the forward evolution , AE in the reverse evolution , and neoPTE . Thirty-three positions were mutated in the entire evolution , two of which ( 130 and 306 ) were mutated to two different amino acids . Amino acids found in wtPTE are shown in lower case italics . Numbers indicate the fold change in activity caused by a mutation in a certain background ( Supplementary file 2B–F ) . Mutations causing a >1 . 3-fold change compared to the respective parent mutant are considered non-neutral . p-values compared to each parent were calculated ( Supplementary file 2B , D , F ) . The mutations T341i in AE , l140M and t199I in the forward evolution , and V49a and s258N in the reverse evolution are not significant ( p-values >0 . 05 ) . Therefore , out of 144 mutations , only five show a >1 . 3-fold effect , but are statistically not significant . Boxes that are crossed out indicate that a mutation did not occur in this background . For direct comparison , the activity changes resulting from a mutation are adjusted to the same direction—from the amino acid found in AE to the respective other amino acid ( label at the top ) . To illustrate , the effect of R254h was measured as follows: AE and neoPTE contain Arg254 , and thus the effect of R254h is directly calculated based on the comparison between AE and AE-R254h ( Fold changeR254h = ActicityAE-R254h/ActivityAE ) and between neoPTE and neoPTE-R254h ( Fold changeR254h = ActicityneoPTE-R254h/ActivityneoPTE ) . However , because wtPTE and the forward evolution background already contain His254 , the effect of introducing this amino acid has to be calculated ‘in reverse’ by first assuming to remove this mutation and then adding it back in , that is , based on the comparison between wtPTE-h254R and wtPTE ( Fold changeR254h = ActicitywtPTE/ActivitywtPTE-h254R ) . All mutational effects that were calculated in this ‘reverse’ way are underlined . Note that wtPTE-h254R is identical to the round 1 variant and therefore the effect in the forward evolution is the same as in the wtPTE background . Because R254h did not occur in the reverse evolution , no effect could be calculated in this background and the respective box is crossed out . DOI: http://dx . doi . org/10 . 7554/eLife . 06492 . 013 The analysis revealed extensive epistasis during the forward and reverse evolution . In the forward evolution , the effect of mutations is significantly altered after their fixation due to epistasis caused by mutations subsequently accumulated in the trajectory . For example , some mutations initially increased ( t172I in round 6 and l271F in round 14 ) or were neutral to ( l130V in round 14 ) phosphotriesterase activity when they occurred in the trajectory , and were thus unfavorable to revert as their reversion would not change ( V130l ) or decrease ( I172t , F271l ) activity ( Figure 6A ) . However , reversion of these mutations became possible ( i . e . , would lead to an increase in activity ) in the background of AE ( Figure 6A ) . On the contrary , h254R decreased phosphotriesterase activity when it occurred in round 1 and therefore its reversion ( R254h ) would initially be favorable . However , the effect of this reversion switched to unfavorable ( R254h ) when it was tested in AE ( Figure 6B ) . Moreover , mutations in the forward evolution had a permissive effect on the accumulation of new mutations and , in this way , opened up a path towards the alternative trajectory taken in the reverse process; all new mutations had a neutral or negative effect on phosphotriesterase activity in the genetic background of wtPTE but most of them become positive in AE ( Figure 6C ) ; for example , AE-s308C ( 6 . 4-fold ) , AE-V130M ( 5 . 4-fold ) and AE-p135S ( 2 . 9-fold ) . Because these mutations can compete with the most favorable reversions ( >1 . 3–8-fold effect , Supplementary file 2 ) , they were selected in the early rounds of the reverse evolution , laying the foundation for the alternative trajectory . 10 . 7554/eLife . 06492 . 014Figure 6 . Epistasis between mutations in the forward evolution restricts some reversions while permitting others as well as new mutations . ( A ) Several reversions change their effect from unfavorable upon their initial occurrence in the forward evolution to favorable in AE . ( B ) Other reversions change their effect from favorable to unfavorable . Note that , in the forward evolution , mutations occurred in the opposite direction as shown ( l130V , t172I , l271F , and h254R ) , but are given in the same direction as AE for direct comparison . Phosphotriesterase activity was too low to be determined in AE + R254h , but at least 10-fold reduced . ( C ) The effect of new mutations changes from wtPTE ( small panel ) to AE ( large panel ) . Relative activities were calculated by comparing a variant containing a certain mutation with one lacking only this mutation . Mutations causing a >1 . 3-fold change compared to the respective parent mutant ( dotted line ) are considered non-neutral . p-values compared to each parent ( Supplementary file 2B ) and p-values for the effect of each mutation in the two respective backgrounds shown in each panel were calculated ( Figure 6—source data 1 , 2 ) . Note that the mutation m293K , which causes a significant increase in AE , does not have a significantly different effect in the two backgrounds . Amino acids found in wtPTE are shown in lower case italics . Color code as in Figure 1 . All other mutational effects in the different backgrounds are given in Figure 5 and Supplementary file 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 06492 . 01410 . 7554/eLife . 06492 . 015Figure 6—source data 1 . Comparison of the effect of mutations in the forward evolution and in AE ( panels A , B ) . Fold-changes between the two backgrounds as well as p-values calculated according to the t-test are given . [a] Only mutations with an average >1 . 3-fold difference between backgrounds and a p-value <0 . 05 are considered significant . [b] Phosphotriesterase activity was too low to be determined in AE + R254h , but at least 10-fold reduced . DOI: http://dx . doi . org/10 . 7554/eLife . 06492 . 01510 . 7554/eLife . 06492 . 016Figure 6—source data 2 . Comparison of the effect of mutations in wtPTE and AE ( panel C ) . Fold-changes between the two backgrounds as well as p-values calculated according to the t-test are given . [a] Only mutations with an average >1 . 3-fold difference between backgrounds and a p-value <0 . 05 are considered significant . Non-significant values are underlined . Note that the effect of m293K , which causes a significant increase in AE , is statistically not significant between wtPTE and AE . DOI: http://dx . doi . org/10 . 7554/eLife . 06492 . 016 In the reverse evolution , the active site architecture necessary for phosphotriesterase activity was restored largely through new mutations , which restricted the reversion of mutations accumulated in the forward evolution . Overall , nine of the 10 reversions that were initially favorable in the background of AE lost their favorable effect in neoPTE because of epistasis during the reverse evolution ( Figure 7A ) . We were able to trace the molecular basis of this effect in several cases as described in the following . First , f306I enlarged the active site in the forward evolution , resulting in a loss of shape complementarity to paraoxon . In the reverse evolution , the nearby s308C offsets this effect by increasing the hydrophobicity of the pocket ( Figure 2 , Figure 7C , Figure 2—figure supplement 1A ) . The redundancy of the mutations f306I and s308C was also evidenced by combinatorial mutational analysis; incorporation of s308C restricts subsequent reversion of I306f due to sign epistasis ( Figure 7B ) . While this reversion would have been favorable in isolation , phosphotriesterase activity of the double mutant AE-I306f-s308C is reduced compared to AE-s308C . Second , the active site was narrowed in the forward evolution by l271F and several other mutations in loop7/8 including l272M and i313F ( Figure 7C and Figure 2—figure supplement 1B ) , causing steric hindrance for paraoxon . The pocket was re-opened initially by the reversion F271l . Subsequently , the new mutation s258N destabilized and altered the conformation of loop 7 and further enlarged the pocket ( Figure 2 , Figure 2—figure supplement 1B ) . We also observed that incorporation of s308C and F271l restricted the reversion of both l272M and f313I ( M272l and I313f , Figure 7B ) . Third , the active site was reshaped by a subtle ∼1 . 0 Å shift of Leu106 and Trp131 , which was likely triggered by a cluster of remote mutations occurring in the same loops ( s102T , l130V , m138I , s137T , and v140M , Figure 7D , and Figure 2—figure supplement 1C ) . In the reverse evolution , these residues are shifted back to their original positions through two new remote mutations , p135S and V130M ( Figure 7D ) . Again , the two mutations are redundant and mutually exclusive; p135S restricts the reversion of m138I ( I138m , Figure 7B ) . Fourth , the distance between the two active site zinc ions decreased from 3 . 8 to 3 . 3 Å in the forward evolution through displacement of the metal-chelating His201 and the β-metal ( Figure 7E ) , which was likely triggered by the combined action of several remote mutations in loops 4 and 5 ( t172I , q180H , t199I , and a204G , Figure 2 ) . In the reverse evolution , the positions of His201 and the β-metal , as well as the original inter-metal distance of 3 . 8 Å , were restored through the reversion I172t and formation of a new hydrogen bonding network with two additional new mutations , a203E and g174D ( Figure 7E ) . These examples demonstrate that rewiring the intramolecular interaction network of the protein can result in the same physical solution in key elements in the active site . Rewiring occurs because new mutations act as ‘epistatic ratchets’ ( Bridgham et al . , 2009 ) for potential reversions , restricting their fixation and thus leading to the incompatible new enzyme neoPTE . 10 . 7554/eLife . 06492 . 017Figure 7 . Convergence to the original active site configuration in the reverse evolution through rewiring of the molecular interaction network leads to genetic incompatibility . ( A , B ) Epistasis during the reverse evolution causes irreversibility and incompatibility . ( A ) The activity change of mutations that were favorable in the initial stage of reverse evolution , but not reverted . neoPTE background: small panel; AE background: large panel . Color code as in Figures 1 , 2 . Mutations causing a >1 . 3-fold change compared to the respective parent mutant ( dotted line ) are considered non-neutral . p-values compared to each parent ( Supplementary file 2B ) and p-values for the effect of each mutation in the two respective backgrounds shown in each panel were calculated ( Figure 7—source data 1 ) . ( B ) Combinations of mutations that constrained the evolutionary trajectory due to sign epistasis . Phosphotriesterase activity is shown on a linear scale . p-values are given in Supplementary file 2G . Note that the two mutants AE + F271l + s308C and AE + M272l + s308C have non-significant p-values compared to the ‘double mutant’ in this series , AE + F271l + M272l + s308C . However , determination of kcat/KM values confirms sign epistasis in this series ( see also Supplementary file 2G ) . ( C–E ) Amino acid changes in the forward ( left panel ) and reverse evolution ( right panel ) . ( C ) Reorganization of loops 7 and 8 . A new mutation , s258N , caused the reorganization ( see also Figure 2—figure supplement 1A , B ) . ( D ) Different combinations of remote mutations in loop 3 resulted in identical positioning of Leu106 , Trp131 , and Leu132 in wtPTE and neoPTE ( see also Figure 2—figure supplement 1C ) . ( E ) Rewiring the interaction network in neoPTE by remote mutations in loops 4 and 5 led to β-metal displacement ( see also Figure 2—figure supplement 1D ) . Amino acids found in wtPTE are shown in lower case italics . DOI: http://dx . doi . org/10 . 7554/eLife . 06492 . 01710 . 7554/eLife . 06492 . 018Figure 7—source data 1 . Comparison of the effect of mutations in wtPTE and AE ( panel A ) . Fold-changes between the two backgrounds as well as p-values calculated according to the t-test are given . [a] Only mutations with an average >1 . 3-fold difference between backgrounds and a p-value <0 . 05 are considered significant . Non-significant values are underlined . DOI: http://dx . doi . org/10 . 7554/eLife . 06492 . 018 Our work demonstrated that a >104-fold loss in phosphotriesterase activity , which accompanied the functional transition to a distinct chemical reaction—arylester hydrolysis—via accumulation of 26 mutations , is readily restored when the selection pressure is reverted . Phenotypic reversal has been observed in previous cases ( Clarke , 1985; Lenski , 1988; Crill et al . , 2000; Teotonio and Rose , 2000; Kitano et al . , 2008 ) , supporting the notion that the phenotype is largely subject to deterministic forces . The likelihood of evolutionary reversibility depends on the complexity of the system and the distance in sequence and function from the ancestor , and it is possible that starting from a more distantly evolved arylesterase would have failed to restore phosphotriesterase function . Moreover , we modulated protein stability throughout the entire trajectory using overexpression of GroEL/ES to avoid evolutionary dead ends caused by stability bottlenecks ( Socha and Tokuriki , 2013; Wyganowski et al . , 2013 ) . In the absence of chaperones , adaptation may have occurred through a different pathway . Another limitation of our work is that we only examined two evolutionary trajectories ( the main trajectory and the trajectory without reversions ) . One could imagine conducting multiple parallel evolutionary experiments to shed light on the repeatability of the trajectory taken , but unfortunately our screening system is not amenable to such a throughput . Nevertheless , our experiment shows that the genotype is subject to strong constraints: an alternative mutational pathway was taken which prevented retracing of the original pathway . Genotypic irreversibility was caused by several factors . First , because selection in the forward evolution was only for increased arylesterase activity ( except in rounds 19–22 ) , the effect of the mutations on phosphotriesterase activity was stochastic: most decreased phosphotriesterase activity , some did not affect it , and some increased phosphotriesterase activity . Therefore , even if one were to revert the mutations in the reverse order of their occurrence ( from rounds 22 back to 1 ) , the lack of continuous activity increases would prevent a gradual adaptive trajectory . Second , by the end of the forward evolution , several new mutations able to increase paraoxonase activity emerged due to epistasis . The fixation of these mutations then acts as an epistatic ratchet ( Bridgham et al . , 2009 ) that prevents reversions . Therefore , as soon as the first new mutation accumulates , the trajectory deviates further from the original path . Our work suggests that only certain sets of mutations are able to cause phenotypic reversal . Although genotypic redundancy was observed , the presence of at least some reversions was essential for complete restoration of catalytic activity , and several new mutations were shared between the two trajectories examined . Similarly , other experimental evolution studies that examined parallel evolutionary trajectories starting from the same sequence often resulted in accumulation of the same mutations ( Bull et al . , 1997; Salverda et al . , 2011; Dickinson et al . , 2013; Khanal et al . , 2014 ) . These observations indicate that a number of functional mutations accessible from a particular starting point are highly limited , and that the genotype is also subjected to deterministic forces to some extent . In our case , the limited accessibility to functional mutation can be explained by the requirement to adapt the wild-type active site configuration in order to obtain efficient phosphotriesterase hydrolysis . Recent work by Harms et al . showed that the accessibility of functional and permissive mutations on hormone receptors is also strongly constrained by biophysical requirements imposed on the binding pocket as well as by protein dynamics ( Harms and Thornton , 2014 ) . Understanding such biophysical requirements and , in the case of enzymes , imperatives of chemical reactivity , is essential to develop our knowledge of evolutionary dynamics and constraints , although the exact nature of such requirements may be unique to each protein . In the case of PTE , the combination of multiple subtle changes is required to fulfill these biophysical requirements and completely switch the enzyme's ability to recognize two different substrates ( paraoxon vs . 2NH: tetrahedral vs . planar , P–O bond vs C–O bond cleavage , trigonal bipyramidal vs . tetrahedral transition state geometry ) . All but four of the 33 mutations occur in locations remote from the active site and act by fine-tuning rather than directly changing the active site configuration . Some changes occur at the sub-Å level ( e . g . , the shift in Trp131 , Leu132 , and the β-metal ) , and possibly act by influencing the dynamics of the active site loops . It may be that only remote mutations can achieve such subtle optimization . A mutation directly in the active site would result in a larger , more disruptive change ( e . g . , even a single additional carbon center would fill an additional 4 Å radius ) and therefore be unable to provide the necessary fine-tuning . Other directed evolution studies also observed the accumulation of remote mutations ( Morley and Kazlauskas , 2005 ) , suggesting that fine-tuning of the active site may be a common strategy to implement a new function . Our work reveals that the adaptive landscape of PTE is highly rugged: even single amino acid changes can regulate activities upwards or downwards and also predetermine the potential effect of subsequent mutations . As discussed above , because multiple mutations can directly or indirectly affect the same key component for catalysis , their effects are likely to be epistatic . Therefore , the alternative trajectory is caused by epistasis between mutations: frequently , those mutations that accumulate first have a permissive or restrictive effect on subsequent mutations . Overall , >70% of mutations have highly variable effects on phosphotriesterase activity , depending on the genetic background ( 26 out of 33 positions , Figure 5 ) , and ∼40% showed sign epistasis ( 7 out of 33 ) . The role of epistasis in natural evolution has recently received much attention , but its extent and prevalence are still under debate ( Whitlock et al . , 1995; Poelwijk et al . , 2007; de Visser et al . , 2011; Breen et al . , 2012; Harms and Thornton , 2013; McCandlish et al . , 2013; Kaltenbach and Tokuriki , 2014 ) . Our findings suggest a high frequency of strong epistatic interactions during functional adaptation and therefore support the view that epistasis is paramount in shaping evolution . However , while restrictive mutations block many of the possible evolutionary trajectories , as has been previously emphasized , permissive mutations simultaneously open up new pathways , avoiding ‘evolutionary dead-ends’ and contributing to the diversity of enzyme homologs found in nature . Moreover , our study demonstrates how genotypic irreversibility leads to the emergence of a functional sequence incompatible with the original one ( Kondrashov et al . , 2002; Maheshwari and Barbash , 2011 ) , and implies the importance of evolutionary contingency on the genotypic level . In nature , the environment never ceases to change and a temporary relaxation in selection pressure ( i . e . , the level and type of nutrients or toxins ) followed by re-adaptation ( through both reversions and new compensatory mutations ) may be common ( Akashi et al . , 2012 ) . Higher levels of organization ( such as metabolic or regulatory networks ) might be subject to similar contingency; restoration of a certain function may be achieved by alternative mutations in other parts of the protein structure , in other domains , or in a different protein altogether . If the mutations are mutually exclusive , sequence incompatibilities may arise rapidly . Therefore , in addition to proposed mechanisms such as genetic drift ( Akashi et al . , 2012 ) , genotypic irreversibility may contribute to the prevalence of incompatibility between orthologous enzymes ( Lunzer et al . , 2010; Kvitek and Sherlock , 2011; Corbett-Detig et al . , 2013; Wellner et al . , 2013; Schumer et al . , 2014; Shafee et al . , 2015 ) . Finally , our observations have important implications for the engineering of highly efficient enzymes—for example , how fine-tuning of multiple active site regions can confer significant activity changes , and how context-dependent such changes are . As our understanding of protein sequence-structure-function relationships grows , further rational and computational approaches need to be developed to address the role of remote mutations and epistasis to enhance our ability to create tailor-made proteins . Error-prone PCR libraries were generated using nucleotide analogues ( 8-oxo-2′-deoxyguanosine-5′-triphosphate [8-oxo-dGTP] and 2′-deoxy-P-nucleoside-5′-triphosphate [dPTP] ) or Mutazyme ( GeneMorph II Random Mutagenesis kit , Agilent , Santa Clara , CA , United States ) . A typical protocol using nucleotide analogues can be found in Tokuriki and Tawfik ( 2009 ) . A typical protocol using Mutazyme starts with a 50 μl PCR reaction containing 50 ng of pET-Strep-PTE template and 0 . 8 μM of outer primers ( forward TTCCCCATCGGTGATGTC , reverse GTCACGCTGCGCGTAAC ) . Cycling conditions were: initial denaturation at 95°C for 2 min followed by 10 cycles of denaturation ( 30 s , 95°C ) , annealing ( 30 s , 63°C ) and extension ( 1 min , 72°C ) , and a final extension step at 72°C for 10 min . Plasmid was removed by treatment with Dpn I ( NEB , Ipswich , MA , United States ) . The PCR product was purified using the QIAquick PCR purification kit ( Qiagen , Netherlands ) , and amplified further with BIOTAQ DNA polymerase ( Bioline , United Kingdom ) using inner primers ( forward ACGATGCGTCCGGCGTA , reverse GCTAGTTATTGCTCAGCG ) and starting from 20 ng of template in a 100 μl reaction volume . Cycling conditions were: initial denaturation at 95°C for 2 min followed by 20 cycles of denaturation ( 30 s , 95°C ) , annealing ( 1 min , 58°C ) and extension ( 30 s , 72°C ) , and a final extension step at 72°C for 2 min . This gave an average of two amino acid substitutions per gene . PTE genes of selected variants were amplified by PCR from pET-Strep-PTE plasmids using the outer primers and BIOTAQ DNA polymerase . Cycling conditions were: initial denaturation at 95°C for 2 min followed by 25 cycles of denaturation ( 30 s , 95°C ) , annealing ( 1 min , 63°C ) and extension ( 1 min , 72°C ) , and a final extension step at 72°C for 2 min . PCR products were purified using the QIAquick PCR purification kit and mixed at equal amounts . Before the preparative digest , conditions were optimized by digesting 1 μg of template DNA with a range of DNase I concentrations ( Fermentas , Waltham , MA , United States ) . DNase digest buffer ( 10× ) consists of 0 . 5 M Tris-HCl pH 7 . 5 supplemented with 0 . 5 mg/ml BSA . In addition , reactions contained 10 mM MnCl2 . Reactions were incubated for 10 min at 37°C , stopped by addition of 1/5 vol of stop buffer ( 30 mM EDTA pH 8 . 0 , 30% glycerol and ≈0 . 6× of a DNA loading buffer ) and analyzed by agarose gel electrophoresis in TBE buffer ( 2% agarose gel , 45 mM Tris , 45 mM boric acid , and 1 mM EDTA pH 8 . 0; for all other agarose gel electrophoresis procedures , we used 1% agarose gels and TAE buffer , which is 40 mM Tris , 20 mM acetic acid , and 1 mM EDTA pH 8 . 0 ) . Reactions were scaled up to 10–15 μg of DNA and the digest repeated at the appropriate DNase dilution to give fragments in the range of 50–150 bp . Fragments were purified by gel extraction and 60–80 ng used in a 20 μl assembly PCR . This PCR was performed with Herculase I ( Stratagene , La Jolla , CA , United States ) . Cycling conditions were: initial denaturation at 96°C for 90 s followed by 35 cycles of denaturation ( 30 s , 94°C ) , annealing ( incremental 3°C steps from 65°C down to 41°C , 90 s each ) and extension ( 2 min , 72°C ) , and a final extension step at 72°C for 10 min . Full-length assembly products were amplified using the inner primers and BIOTaq DNA polymerase under the following cycling conditions: initial denaturation at 95°C for 2 min followed by 25 cycles of denaturation ( 30 s , 95°C ) , annealing ( 1 min , 58°C ) and extension ( 1 min , 72°C ) , and a final extension step at 72°C for 2 min . The amount of assembly product used as template for this reaction was varied and product formation verified by 1% agarose gel electrophoresis . Fractions containing product were pooled and purified using the QIAquick PCR purification kit . Mutants were constructed by site-directed mutagenesis as described in the QuikChange Site-Directed Mutagenesis manual ( Agilent ) . PCR products and pET-Strep-ACP vector were digested with Fermentas FastDigest Nco I and Hind III ( or Kpn I , see Supplementary file 1 ) for 1 hr at 37°C . The vector was treated with CIP ( calf-intestinal alkaline phosphatase , NEB , Ipswich , MA , United States ) for an additional hour and subsequently insert and vector were purified from 1% agarose gel using the QIAquick gel extraction kit followed by the Qiagen PCR purification kit . Ligations were performed at a vector:insert mass ratio of 1:1 using T4 DNA ligase ( NEB , Ipswich , MA , United States ) supplemented with 0 . 5 mM ATP ( NEB , Ipswich , MA , United States ) for 2 hr at 22°C or 16°C overnight . Prior to transformation , reactions were purified by ethanol/glycogen ( Fermentas , Waltham , MA , United States ) precipitation . Transformation into electrocompetent E . cloni 10G ( Lucigen , Middleton , WI , United States ) yielded at least 105 colonies . Plasmids were extracted and re-transformed into E . coli BL21 ( DE3 ) containing pGro7 plasmid for overexpression of the GroEL/ES chaperone system . Transformation reactions were plated on an average of 10 agar plates ( 140 mm diameter ) containing 100 μg/ml ampicillin ( or 50 μg/ml kanamycin , see Supplementary file 1 ) and 34 μg/ml chloramphenicol such that each plate contained 200–1000 colonies , leading to a final library size of 2000–10 , 000 variants . Colonies were transferred onto nitrocellulose membrane ( BioTrace NT Pure Nitrocellulose Transfer Membrane 0 . 2 μm , PALL Life Sciences , Port Washington , NY , United States ) , which was then placed onto a second plate additionally containing 1 mM isopropyl β-D-1-thiogalactopyranoside ( IPGT ) 200 μM ZnCl2 ( to ensure availability of Zn2+ ions necessary for enzymatic activity ) , and either 20% ( wt/vol ) arabinose for chaperone overexpression or 20% ( wt/vol ) glucose for repression of chaperone expression . After expression overnight at room temperature for plates containing arabinose or for 1 hr at 37°C for plates containing glucose , the membrane was placed into an empty petri dish . For low activity levels of the parent gene where a maximum signal is desirable , cells were lysed prior to the activity assay by alternating three times between storage at −20°C and 37°C . For higher activities , the lysis step was omitted , making it easier to differentiate between different colonies . For the activity assay , 20–25 ml of 0 . 5% Agarose in 50 mM Tris-HCl buffer , pH 7 . 5 containing 200 μM 2NH ( Sigma , St . Louis , MO , United States ) and Fast Red ( Sigma , St . Louis , MO , United States ) was poured onto the membrane . Red color developed within 30 min . To screen for phosphotriesterase activity , the buffer contained varying concentrations of fluorogenic phosphotriester instead of 2NH/Fast Red as indicated in Supplementary file 1 . Turnover of O-fluoresceinyl-O , O-diethyl-thiophosphate ( fluoresceinyl-DETP , excitation 495 nm , emission 520 nm ) was detected in a Typhoon 9400 scanner ( GE Healthcare , Wauwatosa , WI , United States ) after an appropriate incubation time ( 0–3 hr ) . In the case of 7-O-diethylphosphoryl-3-cyano-4-methyl-7-hydroxycoumarin ( Me-DEPCyC ) , activity was detected in an agarose gel imager ( excitation 365 nm ) using a SYBR Safe filter . Colonies exhibiting high enzymatic activity identified in the pre-screen were picked and re-grown in four to six 96-deep well plates overnight at 30°C , leading to a library of 400–600 pre-selected variants . Wells contained 200 μl lysogeny broth ( LB ) supplemented with 100 μg/ml ampicillin and 34 μg/ml chloramphenicol . Subsequently , deep well plates containing 500 μl LB per well supplemented with ampicillin , chloramphenicol , and 20% ( wt/vol ) arabinose or glucose ( depending on whether chaperone overexpression was to be induced or repressed ) were inoculated with 25 μl of pre-culture and grown for 2–3 hr at 37°C until the OD600 reached ∼0 . 6 . Expression of PTE variants was induced by adding IPTG to a final concentration of 1 mM and cultures were incubated for an additional 2 hr at 30°C or for 1 hr at 37°C in rounds aimed at reducing chaperone dependence . Cells were spun down at 4°C at maximum speed ( 3320×g ) for 5–10 min and the supernatant was removed . Pellets were frozen for a minimum of 30 min at −80°C and subsequently lysed by addition of 200 μl 50 mM Tris-HCl pH 7 . 5 supplemented with 0 . 1% ( wt/vol ) Triton-X100 , 200 μM ZnCl2 , 100 μg/ml lysozyme , and ∼1 μl of benzonase ( 25 U/μl , Novagen , Madison , WI , United States ) per 100 ml . After 30 min of lysis at room temperature , cell debris was spun down at 4°C at 3320×g for 20 min . Depending on the activity level of the library , clarified lysate was diluted prior to the activity assay to obtain a good signal in the initial linear phase of the reaction . Reactions were performed in transparent 96-well plates containing 200 μl per well ( 20 μl lysate + 180 μl of 200 μM substrate in 50 mM Tris-HCl , pH 7 . 5 supplemented with 0 . 02% Triton-X100 in the case of paraoxon and 0 . 1% in the case of 2NH/FR ) . Paraoxon hydrolysis was monitored at 405 nm; 2NH hydrolysis was monitored at 500 nm via complex formation with Fast Red . Improvements >1 . 3-fold relative to the previous round were considered significant . The best clones were picked and re-grown in triplicate . The observed initial rates were normalized to cell density ( determined by the OD600 ) and the average values determined . Approximately 10 improved variants were sequenced after each round . A description of each directed evolution round including selection criteria , the mutations found in each sequenced variant , and mention of the variants chosen as templates for the next library generation can be found in Supplementary file 1 . pET-Strep-PTE plasmids were transformed into E . coli BL21 ( DE3 ) and grown at 37°C in TB medium containing 100 µg/ml ampicillin and 200 μM ZnCl2 . Expression was induced with 0 . 4 mM IPTG when cell density reached an OD600 of 0 . 6 units and cells grown overnight at 20°C . Cells were harvested by centrifugation at 3320×g and 4°C for 10 min , resuspended and lysed for 1 hr at room temperature using a 1:1 mixture of B-PER Protein Extraction Reagent ( Thermo Scientific , Waltham , MA , United States ) and 50 mM Tris-HCl buffer , pH 7 . 5 containing 200 μM ZnCl2 , 100 μg/ml lysozyme and ∼1 μl of benzonase per 100 ml . Cell debris was removed by centrifugation at 30 , 000×g and 4°C for 45 min and the clarified lysate passed through a 45 μm filter before loading onto a Strep-Tactin Superflow High capacity column ( 1 ml column volume ) . After several washes with 50 mM Tris-HCl buffer , pH 7 . 5 containing 200 μM ZnCl2 , Strep-PTE variants were eluted in the same buffer containing 2 . 5 mM desthiobiotin according to the manufacturer's instructions ( IBA BioTAGnology , Germany ) . Protein was dialyzed overnight against 50 mM Tris-HCl buffer , pH 7 . 5 containing 100 mM NaCl and concentrated if necessary . This protocol was adapted for purification in 96-well format by using AcroPrep 96 Filter Plates ( Pall Life Sciences , Port Washington , NY , United States ) according to the manufacturer's instructions . Lysates were clarified using Lysate Clearance plates ( 3 μm GxF , 0 . 2 μm Supor ) and transferred to filter plates ( 0 . 45 μm GHP ) containing 50 μl Strep-tactin resin per well . Wells were washed 3× with 50 mM Tris-HCl pH 8 . 5 containing 100 mM NaCl and 200 μM ZnCl2 and 3× with pH 7 . 5 buffer . After elution , samples were concentrated and elution buffer removed using ultrafiltration plates ( Omega 10K membrane ) . Paraoxon , 2NH , and Fast Red were purchased from Sigma ( St . Louis , MO , United States ) . Substrates for linear free energy relationships were gifts from Dan Tawfik's laboratory and their synthesis is described in Khersonsky and Tawfik ( 2005 ) . Absorbance wavelengths and extinction coefficients are given in Supplementary file 2 . For determination of initial rates in lysate , cells were grown and assayed in at least duplicate as described under the section ‘Screens in 96-well plates’ . The experiment was repeated and the average change relative to the respective parent variant and the standard deviation were determined ( Supplementary file 2 ) . A Student's t-test was performed to obtain p-values . Where applicable , p-values were also calculated to determine whether the effect of a certain mutation in two different backgrounds ( rather than compared to the parent mutant lacking this mutation ) is significant . For determination of initial rates using purified enzyme , variants were expressed and purified in 96-well format in at least duplicate and assayed as described above in ‘Screens in 96-well plates’ . For determination of Michaelis–Menten parameters , reactions were performed in triplicate at a range of substrate concentrations ( 0–2000 μM ) . Reactions were initiated by addition of 180 μl of substrate solution ( in 50 mM Tris-HCl , pH 7 . 5 supplemented with 0 . 02–0 . 1% Triton X-100 ) to 20 μl of enzyme in 50 mM Tris-HCl , pH 7 . 5 supplemented with 200 μM ZnCl2 and 0 . 02% Triton X-100 . Data were fit to Michaelis–Menten kinetics in Kaleidagraph . AE and neoPTE genes were cloned into pET32-trx plasmid without Strep-tag using FastDigest NcoI and HindIII as described above , transformed into E . coli BL21 ( DE3 ) , and grown for 72 hr at 30°C in TB medium containing 100 μg/ml ampicillin and 500 μM ZnCl2 . Cells were harvested by centrifugation at 3320×g and 4°C for 10 min , resuspended in 20 mM Tris-HCl pH 8 containing 100 μM ZnCl2 and lysed by sonication ( OMNI Sonic Ruptor 400 , Thermo Scientific , Waltham , MA , United States , 3× 30 s on/60 s off , amplitude 40% ) . Cell debris was removed by centrifugation at 30 , 000×g and 4°C for 45 min and lysate filtered through 45 μm filters ( Millipore ) . The lysate was loaded onto two HiPrep Q FF columns ( GE Healthcare , Wauwatosa , WI , United States ) in series . PTE elutes in the flow through as well as the early wash fractions . Active fractions were pooled and passed through a 45 μm filter . The sample was concentrated over a Millipore spin column ( MWCO 30 , 000 ) and purified by gel filtration ( HiLoad 16/60 Superdex 200 prep grade , GE Healthcare , Wauwatosa , WI , United States ) . Protein was concentrated to 12 mg/ml and stored at 4°C . Crystals of wtPTE , AE , and neoPTE were obtained by vapor diffusion from a solution containing protein ( 10 mg/ml ) plus 20–30% wt/vol 2-methane-4-pentane diol ( MPD ) , buffered to pH 6 . 5 by 0 . 1 M sodium cacodylate , as described previously ( Tokuriki et al . , 2012 ) . Serial microseeding was performed to increase crystal size ( Bergfors , 2003 ) . Crystals grew to approximately 200 micrometers and were soaked in 40% MPD , 0 . 1 M sodium cacodylate as cryoprotectant for 5–10 min and then flash-cooled to 100 K in the gaseous nitrogen cryostream of a cooling device ( Oxford Cryosystems , United Kingdom ) . Data were collected from frozen crystals on beamline MX1 of the Australian Synchrotron ( AS ) . The data were indexed and integrated by XDS ( Kabsch , 2010 ) and Aimless ( Evans and Murshudov , 2013 ) , with data cut-off being made at the highest resolution that retained a mean half dataset correlation coefficient ( CC1/2 ) of at least 0 . 5 in the outer shell ( Supplementary file 3 ) ( Karplus and Diederichs , 2012 ) . Although all three crystals were crystallized in the same conditions , neoPTE crystallized in a different space group with different unit cell dimensions ( p65 , with a h , −h−k , l merohedral twin operator , vs C2221 ) . A starting model for refinement ( R18; PDB ID: 4E3T ) ( Tokuriki et al . , 2012 ) was used to provide initial phases . Structures were refined with phenix . refine ( Afonine et al . , 2012 ) and validated with molprobity ( Chen et al . , 2010 ) , as implemented in the phenix software suite ( Supplementary file 3 ) ( Adams et al . , 2010 ) .	Enzymes in bacteria and other organisms are built following instructions contained within each cell's DNA . Changes in the DNA , that is to say , mutations , can alter the shape and activity of the enzymes that are produced , which can ultimately affect the ability of the organism to survive and reproduce . Mutations that are beneficial to the organism are more likely to be passed on to future generations , which can lead to populations changing over time . The DNA sequences that an organism carries are referred to as its ‘genotype’ and the resulting physical characteristics of the organism are known as its ‘phenotype’ . Studies of evolution tend to focus on how particular species or molecules become more different over time . However , one area that remains controversial is whether it is possible for evolution to be reversed so that an organism or molecule returns to a previous form . An enzyme called PTE is said to have phosphotriesterase activity because it catalyzes this particular type of chemical reaction . Recently , a group of researchers used a method called ‘directed evolution’ to demonstrate that it is possible for PTE to evolve in a way that means it loses its phosphotriesterase activity and becomes able to catalyze a different type of chemical reaction . Here , Kaltenbach et al . —including some of the researchers from the previous work—investigated whether it was possible to use the same method to reverse this evolution and restore the enzyme's original activity . The experiments show that reverse evolution is possible as phosphotriesterase activity was restored to the PTE enzyme from the previous study . However , although the phenotype of the final enzyme matched that of the original PTE enzyme , the genotypes did not match as the DNA sequences of the genes that encode these enzymes differ . The DNA does not revert to its original sequence because the effect of individual mutations on the phenotype depends on what other mutations are present . For example , as the enzyme evolved its new activity , additional mutations accumulated that did not alter enzyme activity . During the reverse evolution experiment , some of these mutations could have started to exert influence on the phenotype so that different mutations were required to restore the phosphotriesterase activity . In the future , Kaltenbach et al . 's findings may aid efforts to engineer artificial enzymes for use in medicine or industry .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "evolutionary", "biology", "biochemistry", "and", "chemical", "biology" ]	2015	Reverse evolution leads to genotypic incompatibility despite functional and active site convergence
In studies of voluntary movement , a most elemental quantity is the reaction time ( RT ) between the onset of a visual stimulus and a saccade toward it . However , this RT demonstrates extremely high variability which , in spite of extensive research , remains unexplained . It is well established that , when a visual target appears , oculomotor activity gradually builds up until a critical level is reached , at which point a saccade is triggered . Here , based on computational work and single-neuron recordings from monkey frontal eye field ( FEF ) , we show that this rise-to-threshold process starts from a dynamic initial state that already contains other incipient , internally driven motor plans , which compete with the target-driven activity to varying degrees . The ensuing conflict resolution process , which manifests in subtle covariations between baseline activity , build-up rate , and threshold , consists of fundamentally deterministic interactions , and explains the observed RT distributions while invoking only a small amount of intrinsic randomness . The reaction time ( RT ) represents the total time taken to perform all of the mental operations that may contribute to a particular action , such as stimulus detection , attention , working memory , or motor preparation . Although the importance of the RT as a fundamental metric for inferring the mechanisms that mediate cognition cannot be overstated ( Welford , 1980; Meyer et al . , 1988 ) , such reliance is a double-edged sword . Under appropriate experimental conditions , differential measurements of RT may be used as a readout for changes in the ( mean ) time consumed by any one of the aforementioned operations , but a particular RT value is hard to interpret because it may be that not all of the operations involved are known , and those that are relevant may overlap in time to varying degrees . Furthermore , each operation may have its own , independent source of variability , making it very difficult to attribute the measured variance in RT to a particular cause ( e . g . , Krajbich et al . , 2015 ) . In the case of eye movements , this ambiguity is likely to be more severe than previously appreciated . There is a firm mechanistic account that describes how saccades are triggered , but according to the present results , that account lacks a crucial ingredient — ongoing motor conflict — and assumes , incorrectly , that in response to the same stimulus , the fundamental reason why some saccades are triggered very quickly whereas others take much longer simply boils down to noise in the underlying neuronal activity . The neural dynamics that give rise to eye movements are well established . In essence , the preparation to make a saccade of a particular direction and amplitude is equal to a gradual rise in the activity of oculomotor neurons that are selective for the corresponding movement vector . If this rising activity , referred to as a motor plan , ramps up rapidly , the saccade is initiated quickly; if the motor plan grows more slowly , the saccade starts later . Quantitatively , this corresponds to a negative correlation between saccadic RT and build-up rate ( Hanes and Schall , 1996; Fecteau and Munoz , 2007; Ding and Gold , 2012; Heitz and Schall , 2012; Costello et al . , 2013 ) . Notably , neurons that encode motor plans seem to reach a consistent firing level just before the onset of a saccade , particularly in the superior colliculus ( SC ) and the frontal eye field ( FEF ) ( Hanes and Schall , 1996; Brown et al . , 2008; Stanford et al . , 2010; Ding and Gold , 2012 ) . That is , there appears to be a fixed activity threshold that serves as a trigger for eye movements ( Lo and Wang , 2006 ) . Thus , it is widely thought that , for simple saccades to lone , unambiguous stimuli , i . e . , reactive saccades , the variance of the RT distribution is predominantly determined by the variance of the FEF/SC build-up rates across trials ( Carpenter and Williams , 1995; Hanes and Schall , 1996; Fecteau and Munoz , 2007; Sumner , 2011 ) . This rise-to-threshold process is of enormous conceptual importance , as it is the key building block of virtually all models of decision-making in which multiple-choice alternatives , typically guided by perceptual information , are evaluated over time ( Gold and Shadlen , 2001; Erlhagen and Schöner , 2002; Smith and Ratcliff , 2004; Brown and Heathcote , 2008; Stanford et al . , 2010; Krajbich and Rangel , 2011; Thura et al . , 2012; Brunton et al . , 2013; Miller and Katz , 2013 ) . Nevertheless , the variance of this process in its simplest possible instantiation — reactive saccades — remains a mystery ( Sumner , 2011 ) , because it seems too large to reflect noise or intrinsic randomness ( in the build-up rates of oculomotor neurons ) alone . One possibility is that the randomness is purposeful , that unpredictability in saccade timing somehow entails a behavioral advantage ( Carpenter , 1999 ) . Alternatively , the RT of reactive saccades may fluctuate , at least in part , because of underlying neural mechanisms that have simply not been identified yet . Here we describe such mechanisms . We recorded activity from single FEF neurons in an elegant paradigm ( Lauwereyns et al . , 2002; Hikosaka et al . , 2006 ) that produces a large spread in saccadic RT simply by varying the subject’s spatial expectation of reward . A model based on competitive dynamics quantitatively reproduced the temporal profiles of the evoked neural responses , as well as their dependencies on RT , reward expectation , and trial outcome ( correct/incorrect ) — this , while simultaneously matching the monkeys’ full RT distributions across experimental conditions . The results indicate that RTs vary because the stimulus-driven activity does not start from a consistent , neutral state , but rather from a dynamic , biased state in which incipient , internally driven motor plans are already developing . In other words , when the target appears , the monkey is already thinking of looking somewhere . This conflict between motor alternatives ( 1 ) requires varying amounts of time to be resolved , depending on how advanced and how congruent the bias-driven plans are relative to the target-driven response , ( 2 ) is initially defined by the baseline levels of activity ( measured during fixation ) across spatial locations , and ( 3 ) dictates the build-up rate and threshold of the subsequent rise-to-threshold process . Thus , the noise in the build-up rate is much more modest than that predicted by extant frameworks , and the high variability of saccades to single targets is , to a large degree , deterministic , a direct consequence of motor selection mechanisms that allow voluntary saccades to be driven by both sensory events and internal biases . Two rhesus monkeys were trained on the one-direction rewarded ( 1DR ) task ( Figure 1a ) , in which a saccade to a lone , unambiguous target must be made but a large liquid reward ( primary reinforcer ) is available only when the target appears in one specific location ( Lauwereyns et al . , 2002; Hikosaka et al . , 2006 ) . The rewarded location remains constant over a block of trials and then changes . Of note , ours is a RT version of the task whereby the go signal ( offset of fixation point ) , which means ‘move now ! ’ is simultaneous with target onset . Also , it involves four locations and variable block length . This task generates errors and a large spread in RT ( Figure 2 ) under minimalistic sensory stimulation conditions . We exploit this to investigate how variance in saccadic performance relates to variance in FEF activity . When the target and rewarded locations coincided ( congruent trials; Figure 2a , red traces ) , the monkeys consistently moved their eyes very quickly ( monkey G , 158±33 ms , mean RT ± 1 SD; monkey K , 146±21 ms ) , and essentially never missed ( monkey G , 99 . 8% correct , n=7234 congruent trials; monkey K , 99 . 6% , n=5837 ) . By contrast , when the rewarded and target locations were either diametrically opposed or adjacent ( incongruent trials; Figure 2b–d , cyan traces ) , both the mean RT and the spread increased dramatically ( monkey G , 269±84 ms; monkey K , 236±77 ms ) , as did the percentage of incorrect saccades away from the target ( monkey G , 18 . 3% incorrect , n=16905 incongruent trials; monkey K , 8 . 1% , n=12708 ) . The symmetric condition in which all directions were equally rewarded ( ADR; Figure 1b ) produced RT distributions that were intermediate between those of congruent and incongruent trials ( Figure 2a–d , gray; monkey G , 192±40 ms; monkey K , 174±36 ms ) . These results recapitulate the puzzle mentioned in the Introduction: if saccades can be very fast , why , under identical stimulation conditions , are they sometimes very slow ? Also note that , compared to those of correct saccades , the RTs of incorrect saccades were neither consistently fast , as might be expected on the basis of strong anticipation , nor consistently slow , as might be expected from a protracted deliberation process ( Figure 2e ) . Instead , the RTs during errors fell squarely in the middle of the distributions of correct RTs ( for correct trials , 90% of RTs were inside the ranges 158–432 ms and 146–404 ms for monkeys G and K , respectively; for errors , the ranges were 180–336 and 152–388 ms ) . Mechanistically , it is not obvious how this could be accomplished . One of the aims of the model presented below is to reproduce the RT data shown in Figure 2 and to resolve this conundrum . In short , monkeys are highly sensitive to the spatially asymmetric value associated with otherwise identical target stimuli ( for additional evidence of this , see Figure 2—figure supplement 1 ) . This manifests primarily as large differences in RT between correct congruent , correct incongruent , and incorrect incongruent trials ( for other manifestations , see Figure 2—figure supplement 2 ) . The pattern of results suggests that , when a target appears at an unrewarded location , a conflict arises because the monkey wants to look at the rewarded location instead . In what follows , we ask: what features of the evoked FEF activity reflect such conflict , and do they account for the measured variability in the direction and timing of the elicited saccades ? We recorded single-unit activity from 132 FEF neurons in the two monkeys ( 67 in monkey G; 65 in monkey K ) during performance of the 1DR task . The analysis in this section focuses on a population of 62 neurons that satisfied two conditions: they had standard visual and saccade-related responses , and both correct and incorrect trials were collected for them ( for details of cell and trial selection procedures , see 'Materials and methods , Neuronal classification and selection' ) . In this section , we show that , during the rise-to-threshold process associated with reactive saccades , the build-up rate , the apparent threshold , and the baseline firing rates measured during fixation ( before the target is presented ) exhibit coordinated variations that have not been appreciated before and which , beyond the specifics of our task , are likely to be key for determining RTs in general . We refer to the baseline levels in plural because the target location is not the only relevant one; neurons with RFs at other locations may be activated by internally driven biases , creating motor conflict , that is , activity that competes with and impacts the target-driven response — and the RT . Before describing these results , a note about nomenclature . Whereas just two distinct experimental conditions , congruent and incongruent , are relevant for behavioral analysis , neurophysiologically there are eight conditions to consider , depending on whether the target , expected reward , and saccadic movement were inside or outside a recorded cell’s response field ( RF ) . Thus , for example , we use the abbreviation IOI to denote the target/reward/saccade combination in/out/in; that is , trials for which the target was in , the reward was expected out , and the saccade was into the RF ( see icons in Figure 3a–c ) . Such IOI trials are incongruent , because target and reward locations do not match , and correct , because the target and saccade locations match . Only six of the eight possible combinations are considered because congruent trials were virtually devoid of errors , so IIO and OOI conditions are absent . With this notation at hand , we now turn to the activity of FEF across the six remaining conditions . Although each saccade in the 1DR task involves a single target , the motor preparation process in FEF can ( and should ) be understood as a motor selection process involving not one but at least two populations of neurons , those that contribute to the actual saccadic choice and those that favor the opposite choice ( Figure 3a–c , see icons ) . During congruent trials , when the target appears at the rewarded location , the evoked activity rises most rapidly and reaches the highest firing rate ( Figure 3a , III trials , red trace ) . The saccade is essentially always correct and no evidence of conflict is discernible because the neurons favoring the opposite choice , away from the target , show little ( if any ) response before the eye movement ( Figure 3a , OOO trials , green trace ) . In this case , naively , it would appear as if only the one response that rises to threshold is important . Notably , though , a difference in activity between the two complementary populations is already evident before the go signal/target onset . This is the internal bias signal created by reward expectation , which in this case is spatially congruent with both the target and the saccade . This baseline signal is consistent with previous neurophysiological studies using the 1DR task ( Takikawa et al . , 2002; Sato and Hikosaka , 2002; Ding and Hikosaka , 2006 ) , and may be interpreted as a neural correlate of spatial attention ( Maunsell , 2004; Peck et al . , 2009; Preciado et al . , 2017 ) . By contrast , during incongruent trials , when the target is presented opposite to the rewarded location , a conflict arises early in the trial in the form of a higher baseline favoring the rewarded location ( Figure 3b , note green trace above red before go signal ) . During correct trials this conflict is appropriately resolved as the target-driven activity increases and overtakes the competition ( Figure 3b , IOI trials , red trace ) , but the rise proceeds more slowly , that is , it has a lower build-up rate and ultimately reaches a lower peak firing level ( Rp; see 'Materials and methods , Peak response' ) than that observed when the bias and the target are congruent ( Figure 3d , IOI vs . III ) . Finally , the conflict is even more extreme during incongruent trials that end in erroneous choices toward the rewarded location ( Figure 3c ) . In that case , the initial bias in baseline activity is maximized ( Figure 3e , IOO vs . OII ) and the evoked target-driven activity ( Figure 3c , magenta trace ) is considerably weaker than that observed during correct saccades ( Figure 3d , IOO vs . IOI ) . The neural response associated with the ( wrong ) saccadic choice rises very slowly ( Figure 3c , cyan trace ) and reaches a modest threshold level prior to saccade onset , but this activity is nonetheless slightly above that associated with the opposite ( correct ) motor alternative ( Figure 3c , right panel; Figure 3d , OII vs . IOO ) . This ambivalence between motor plans is reminiscent of that observed during choice tasks ( Thompson et al . , 2005; Ding and Gold , 2012; Costello et al . , 2013 ) , as if the monkeys had struggled to make a choice between two competing targets , even though only one was displayed . These results were based on recordings from 62 neurons with diverse visuomotor properties , but were qualitatively similar when the averaging across cells was restricted to units that were either predominantly visual or predominantly motor ( Figure 3—figure supplement 1 ) . Those two populations differed in the time at which they fired maximally and in the magnitude of their baseline activity , but qualitatively , their responses changed similarly across conditions and outcomes . The mean peak and baseline responses , Rp and Rb , demonstrated remarkably similar variation patterns ( Figure 3 , compare [d] and [e] ) . Overall , these comparisons based on population responses suggest that the baseline firing rates ( at both the target and opposite spatial locations ) , build-up rate , and threshold of the rise-to-threshold process vary in a coordinated way across experimental conditions . But then a crucial question arises: do such covariations also occur from trial to trial within each condition ? We sought evidence of this by analyzing the responses of individual neurons . First we found that , during incongruent trials , the baseline activity , Rb ( firing rate in a 250 ms window preceding target onset ) , is strongly predictive of outcome . When the rewarded location coincided with the RF and the target was presented outside , most neurons ( 34 of 53 , p= 0 . 03 , binomial test ) had a higher baseline rate before incorrect as opposed to correct trials ( Figure 3f , cyan dots , OII vs . OIO; p=0 . 002 , permutation test ) , as if an excessive Rb triggered a ( wrong ) saccade into the RF . Conversely , when the rewarded location was opposite to the RF and the target was subsequently presented inside , most neurons ( 33 of 42 , p= 0 . 001 , binomial test ) had a lower Rb preceding incorrect trials ( Figure 3f , magenta dots , IOO vs . IOI; p=0 . 009 ) , as if the lack of baseline activity precluded a ( correct ) saccade into the RF . Second , for each recorded cell ( V , VM , and M ) , we calculated Spearman correlation coefficients between pairs of neural response measures across trials ( 'Materials and methods , Statistical analyses' ) . Although some of those correlations were too noisy to resolve ( e . g . , between baseline activity and build-up rate ) , that between baseline activity and peak response , ρ ( Rb , Rp ) , was highly robust; it was strongly positive not only in IOI trials ( Figure 4a ) but also in other conditions ( III , OOO , and OIO; p<0 . 001 in all cases ) , and there was no evidence of a dependence on the visuomotor properties of the recorded cells ( Figure 4a , scatterplot; p>0 . 27 for linear regression ) . This confirms that when the baseline activity of a neuron is high , the response evoked later , after the target appears , typically reaches a higher firing level than when the baseline is low . At the single-cell level , we also found strong associations between responsivity and RT . In IOI trials , the correlation between build-up rate and RT , ρ ( RBU , RT ) , was strongly negative ( Figure 4b ) . This is in agreement with the common finding reported by previous studies , that the build-up rate is the main link between oculomotor activity and saccade latency . Furthermore , the correlations ρ ( Rb , RT ) and ρ ( Rp , RT ) were also predominantly negative , consistent with the observation that , across conditions with movements into the RF , shorter RTs are simultaneously associated with higher baseline , higher build-up rate , and higher threshold ( Figure 3a , b ) . These dependencies , and important deviations observed across trial types , are analyzed in detail below . In summary , during the 1DR task , the activity in FEF demonstrated characteristic covariations in the three main features of the rising saccade-related activity: the ‘baseline’ firing rates at both the target and opposite spatial locations , the build-up rate of the target-driven response , and the maximum activity reached before movement onset . This suggests a causal relationship between the baseline and the subsequent response , because the baseline signal arises earlier ( before target presentation ) and because it is predictive of outcome ( Figure 3f ) . Although these covariations were more evident across task conditions than across trials , our contention is that they are always present because they reflect fundamental dynamics of the oculomotor circuitry . Next , to test this , we present a mechanistic model that generalizes these novel interrelationships to all trials , and is thereby able to relate neuronal activity to RT with remarkable detail . A saccadic competition model was developed to investigate the mechanistic link between the FEF activity and the monkeys’ behavior in the 1DR task ( 'Materials and methods , Saccadic competition model' ) . Such a bridge requires that multiple constraints are satisfied . First of all , the model must reproduce the neurophysiological results presented in the previous section . Thus , it considers two neural populations whose responses may rise to a threshold . One population generates saccades toward location T , where the target stimulus is presented , and the other toward D , the diametrically opposite location ( Figure 5a , icon ) . In any given trial , if the target-driven response , RT ( Figure 5a–c , red traces ) , reaches threshold first , a correct saccade to the target is produced , whereas if the bias-driven activity , RD ( Figure 5a–c , blue traces ) , reaches threshold first , the result is an incorrect saccade away from the target . In this way , the RT and RD variables correspond to the population responses recorded with the target inside ( Figure 3a–c , reddish traces ) and outside ( greenish traces ) of the RF , respectively . Another important feature of the recorded data is the evident asymmetry between the two motor plans: the target-driven activity is typically strong and never fully suppressed , whereas the internally driven activity favoring the opposite location is typically suppressed and only rarely of moderate strength . In the model , this asymmetry is captured by two suppression mechanisms that constrain when and how RD can rise . One of them ( 'Materials and methods , Saccadic competition model , Rule 1' ) simply prescribes that , once RT is rising , it can fully suppress RD . That is , the moment RT advances past RD , RD stops rising altogether . The other mechanism is about the timing of RD . The target-driven response , RT , begins to rise shortly after target onset ( after an afferent delay of 35 ms ) , whereas its counterpart , RD , begins to rise later , partly because of a somewhat longer afferent delay ( 50 ms ) but mostly because of a transient , stimulus-driven suppression . This ( partial ) suppression is based on abundant evidence ( reviewed by Salinas and Stanford , ( 2018 ) ; see also Reingold and Stampe ( 2002 ) ; Dorris et al . ( 2007 ) ; Stanford et al . ( 2010 ) ; Bompas and Sumner ( 2011 ) ; Buonocore and McIntosh ( 2012 ) ; Buonocore et al . , ( 2017 ) ) indicating that ongoing saccade plans , RD in this case , are briefly inhibited by stimuli that appear abruptly , just like the target in our experiment . In the model , the inhibition lasts 115 ms ( Figure 5a–c , gray shades ) , after which RD may rise in full force — if it was not overtaken by RT in the interim . These two suppression mechanisms give the target-driven activity an advantage over its competing , internally driven counterpart , and they are necessary to reconcile the occurrence of errors with the late-onset , weak activity away from the RF seen during correct saccades ( Figure 3a , b , greenish traces ) . For instance , without the transient , stimulus-driven inhibition , the model would produce incorrect saccades at a rate that is vastly higher than that observed experimentally . Finally , the model must also capture the variations in baseline activity , build-up rate , and threshold observed in the FEF data ( Figures 3d , e and 4a ) , and this is where the crucial conceptual leap takes place . What we found empirically was that , for the target-driven response , a higher baseline was accompanied by both a higher build-up rate and a higher threshold , with the baseline activity of the alternative motor plan having opposite effects . The model generalizes these dependencies by making two assumptions . First , that similar relationships hold across all trials , rather than just across the three experimental conditions examined , and second , that because the baseline signal is present before target onset , any variations in build-up rate and threshold can be formulated mainly as the result of variations in baseline activity . Thus , the model can be thought of as designed to test whether the differences in the rise-to-threshold process observed across experimental conditions ( Figure 3 ) are the average manifestations of similar but more general dynamical relationships between the three relevant variables , where the variance is primarily derived from the baselines . So , in practice , the general idea is that the baselines fluctuate stochastically and determine the ensuing rise-to-threshold excursion in each trial . The resulting dynamics between competing motor plans can be intuitively appreciated with three example trials ( Figure 5a–c ) . The simplest situation is when , during fixation , the baseline at the target location , BT , is larger than that at the opposite location , BD ( Figure 5a ) . This is typically the case when the target and rewarded locations coincide , but what matters in the model is simply the actual baseline values ( more on this below ) . The condition BT>BD has two specific consequences: ( 1 ) it yields a high build-up rate for the target-driven activity , RT ( red trace ) , and ( 2 ) it sets the saccade threshold , Θ , to a high value ( Equations 6 , 7 ) . Thus , because of the high build-up rate , RT rises sharply and quickly triggers a saccade , in spite of the high Θ . The D plan ( blue trace ) is always suppressed in this case , so no overt conflict is visible . This is how correct saccades with very short RTs are produced . The more interesting scenario occurs when the bias-driven plan starts with the higher baseline , as typically happens when the reward is expected at the D location ( but again , the dynamics are dictated just by the baseline values ) . In that case , the saccade can be either correct or incorrect , depending on how big the lead is . When BD is much larger than BT ( Figure 5c ) , the consequences are essentially the opposite of those in the previous example: ( 1 ) RT has a low build-up rate , so the target-driven response ( red trace ) rises slowly , and ( 2 ) the saccade threshold , Θ , is low . In this way , RD is able to advance steadily after the suppression interval and win the race from wire to wire , reaching a relatively low firing level before saccade onset . This is how incorrect saccades are produced . By contrast , if the baseline BD is only moderately higher than BT ( Figure 5b ) , then the combined effect of the baselines is intermediate relative to that in the two previous examples: ( 1 ) the initial build-up rate of RT is moderate , neither as high as that in the first example nor as low as that in the second , and ( 2 ) the value of Θ is also intermediate . The target-driven plan ( red trace ) rises at a rate that allows it to overtake the competing plan ( blue trace ) and to win the race by coming from behind . Importantly , in this case , RT slows down as it goes past RD ( note the slight change in the slope of the red trace during the shaded interval ) . Although RT wins the race , overtaking the competing plan exacts a toll , and the lower its initial build-up rate , the higher that toll ( Equation 11 ) . This is the one mechanism that was introduced into the model specifically to satisfy key behavioral constraints . In this case , slowing down the winner target-driven plan is necessary to produce correct saccades with very long RTs — longer than those of incorrect saccades . These examples illustrate how the baseline levels BT and BD quantitatively regulate both the build-up rate of the target-driven activity and the saccade threshold . Nevertheless , it is important to stress that , at the same time , the baselines convey information about the location of the expected reward in a manner that is consistent with the experimental data . In the simulations , the baseline values across trials are characterized by their mean and variance , which are fixed ( Equation 5 ) . However , the two mean values are assigned to the T and D locations according to a simple prescription: the rewarded location gets the higher mean ( 'Materials and methods , Saccadic competition model' ) . Thus , in simulations of the congruent case , BT is typically — although not always — larger than BD ( as in Figure 5a ) , and the majority of trials are fast and correct; whereas in simulations of the incongruent case , the roles are reversed , BT is ( on average ) lower than BD ( as in Figure 5b , c ) , which results in a combination of correct and incorrect slower responses . Other than that , the simulations proceed in exactly the same way in the two bias conditions , as they should . With all of these elements in place , the model parameters were adjusted to fit all of the experimental data discussed so far ( 'Materials and methods , Correspondence between data and model parameters' ) . In this way , when trials were sorted by bias and outcome , the simulated neural responses reproduced the covariations in baseline , build-up rate , and threshold across conditions ( Figure 5d–f; for quantification , see Figure 5—figure supplement 1 ) . This demonstrates that , as intended , the hypothesized coupling across trials is indeed consistent with the observed coupling across experimental conditions . In addition , the average RT ( t ) and RD ( t ) traces matched the trajectories of the recorded population responses in great detail ( compare to Figure 3a–c ) . The proposed interaction mechanisms between the two motor plans resulted in average traces with the appropriate magnitude , time course , and degree of ambivalence . But most critically , at the same time , the model fully accounted for the behavioral data: ( 1 ) it generated correct and incorrect saccades in proportions similar to those found experimentally ( ∼0% and ∼10% errors in congruent and incongruent conditions ) , and ( 2 ) it generated simulated distributions of RTs ( Figure 5g–i ) that closely mimic their behavioral counterparts ( as assessed by mean , median , SD , and skewness ) . Note , in particular , that the RTs in incorrect trials ( panel i ) are neither too fast , because the stimulus-driven suppression mechanism prevents fast errors , nor too slow , because the slowest responses ( which occur when RT slows down ) are correct . The model explains the behavioral data in terms of the neural data , accurately replicating both . The results show that , as an intrinsic part of the motor competition process , the baseline activity , build-up rate , and threshold vary in a coordinated fashion to generate the wide range of RTs observed in the task . In the rest of the paper , we show that this fact explains many other , seemingly odd features of the neural data . A key assumption of the model is that fluctuations in baseline activity result in fluctuations in build-up rate and threshold . To further characterize the interdependencies between these three variables and better understand how they impact the RT , we compared the responses evoked in congruent versus incongruent trials before and after equalizing their RTs . The FEF responses recorded during III and IOI trials were quite distinct ( Figure 3a , b ) , even though both involved correct saccades in the same direction . The differences could be due to the different expected reward locations in the two conditions and/or to the different RTs generated ( Figure 6a ) . To eliminate the differences due to RT , we devised a simple sub-sampling procedure ( 'Materials and methods , RT matching' ) that resulted in IOI and III data subsets with identical numbers of trials and nearly identical RT distributions ( Figure 6b ) . Then , we made comparisons across conditions with and without matching the RTs . What should be expected on the basis of the model ? In the standard case , without RT matching ( NM condition ) , the target-driven response , RT , has a higher baseline and reaches a higher threshold in congruent trials than in incongruent trials ( Figure 6c ) . However , because they have very different build-up rates ( note the steeper rise of the magenta curve ) , when aligned on saccade onset , the corresponding response trajectories intersect each other twice . By contrast , when their respective RT distributions are the same ( YM condition ) , the shapes of the trajectories are much more similar and no longer intersect; now , across the two conditions , the differences in baseline and threshold are smaller , and the build-up rates are nearly identical ( Figure 6d ) . These results make perfect sense within our modeling framework . First , shorter RTs are associated with higher baseline , higher build-up rate , and higher threshold , but the RT equalization procedure only retains the fastest IOI trials , so naturally the resulting IOI trajectory simultaneously increases its baseline , build-up rate , and threshold . Second , the build-up rate changes the most because it relates to RT most directly ( Figure 4b ) . And third , the residual difference between congruent and incongruent conditions , which is exclusively related to the internal bias signal , is consistent with the strong coupling between the threshold and the baseline across trials ( Figure 4a; Equation 6 ) . Now consider the same analysis but for 84 V , VM , and M FEF neurons that had sufficient RT-matched trials . Qualitatively , the results are just as expected from the model: when the RTs are not matched , the responses in III and IOI trials differ patently in baseline , threshold , and build-up rate ( Figure 6e ) , and the corresponding curves intersect each other twice when the spikes are aligned on saccade onset ( right panel ) . In contrast , when the RTs are matched , the differences in build-up rate practically disappear , and those in baseline and threshold become smaller ( Figure 6f ) . These results were qualitatively similar across FEF cell categories ( Figure 6—figure supplement 1 ) . These findings are consistent with the dynamics of the model , which dictate that shorter RTs are generally associated with a higher baseline ( at the target location ) , higher build-up rate , and higher threshold , where the correlation between build-up rate and RT is strongest . To test the model more stringently , we exploited the wide range of RTs produced in the 1DR task to generate predictions for how the evoked neural activity should be expected to vary as a function of RT . The rationale for these predictions is straightforward: instead of calculating the mean activity averaged across all trials in , say , the IOI condition ( Figure 5e ) , we consider similar traces based on subsets of trials within narrow RT bins ( 'Materials and methods , Continuous firing activity' ) . Assuming that the activity in FEF directly contributes to triggering each saccade , as happens in the model , the resulting response profiles should vary systematically across those RT bins , and any patterns should be consistent with the correlations among baseline , build-up rate , and threshold instantiated by the model , as well as with its other mechanisms ( e . g . , RD suppression ) . Indeed , when the simulated IOI trials are sorted and averaged according to RT and the resulting curves are color-coded , a characteristic pattern emerges ( Figure 7a ) : steeply rising trajectories precede short-latency saccades ( black ) , and more shallow , protracted trajectories precede long-latency saccades ( red ) . This is largely because , in the model , the RT depends critically on the build-up rate of the target-driven activity . Notably , the apparent threshold reached by these curves at saccade onset is also modulated by RT ( Figure 7a , right panel ) , with fast choices ( black ) leading to higher activity levels than slow ones ( red ) . This , in turn , is consistent with the correlated fluctuations in build-up rate and threshold built into the model . Qualitatively , this prediction for the IOI condition — that is , the pattern resulting from the simultaneous dependencies of build-up rate and threshold on RT — is highly robust to parameter variations ( of at least ±30% , Equations 5–11 ) . To test this prediction , the recorded trials from 84 FEF neurons were sorted by RT in the same way as the simulated trials , and the corresponding traces were averaged across cells ( 'Materials and methods , Continuous firing activity' ) . The population curves that resulted ( Figure 7e ) showed the same smooth transitions across RT bins as the simulated curves . Both the build-up rate and the threshold increased with shorter RTs as expected from the model . More generally , when comparing across narrow RT bins , the agreement between the simulations and the overall population activity in FEF was always tight and evident — even though the model predictions varied widely across trial types . This was true in three important respects . ( 1 ) For the activity evoked in III trials . When the target and rewarded locations were congruent , the neural responses into the RF ( Figure 7g ) were much less sensitive to RT than those in the corresponding incongruent trials ( Figure 7e ) , with the variations in threshold essentially disappearing ( Figure 7g , right panel ) . According to the model ( Figure 7c ) , this much weaker dependence resulted from intrinsic randomness , or noise , in the build-up rate , that is , variability that is independent of the baseline ( the term η in Equations 8 and 9 ) . In the model , such randomness is proportionally stronger in III than in IOI trials , and it blurs the effects created by the baseline-dependent fluctuations . This is explained in more detail in the last section of the 'Results' . Here , we simply emphasize that , although the sensitivity to RT manifested quite differently in III and IOI trials , there was close agreement between the simulated and neural data in both . ( 2 ) For the baseline activity . The model postulates that the fluctuations in baseline translate into fluctuations in build-up rate and threshold . As a consequence , when sorted by RT , the simulated baseline levels spread accordingly . For the target-driven response , RT , a higher baseline always corresponds to shorter RTs ( Figure 7a , c , left panels , note black lines above red before the go signal ) , so the correlation ρ ( Rb , RT ) is negative . By contrast , for the baseline of the opposite motor plan , RD , the correlation is either positive ( Figure 7b , note red lines above black ) or zero ( Figure 7d , note overlapping red and black curves ) , because of the competitive nature of the interactions between the T and D motor plans . Again , even though the effects were expected to be small for all conditions , the actual baseline activity measured in FEF was highly consistent with the model predictions ( Figure 7e–h , left panels ) . This is easier to visualize when the data are magnified appropriately ( Figure 8a ) . In quantitative terms ( Figure 8b ) , on average , there was a significant negative correlation between baseline activity and RT in both IOI ( p=10−5 , signed-rank test ) and III ( p=0 . 004 ) trials , as predicted . There was no net correlation in OOO trials ( p=0 . 8 ) , again as predicted . In OIO trials , although the trend was not strong ( p=0 . 16 ) , it was toward a positive correlation , as expected . ( 3 ) For the activity elicited during saccades away from the RF . The model predicts that , when the target is outside of the RF , the low-intensity evoked response should display the same dependencies on RT as the baseline activity preceding it ( Figure 7b , d ) . Once again , the neural data were very similar to the model simulations ( Figure 7f , h; Figure 7—figure supplement 1 ) , and the agreement was confirmed statistically ( Figure 7—figure supplement 1 ) . In summary , the FEF activity averaged across V , VM , and M cells demonstrated varying degrees of sensitivity to RT , depending on the specific experimental condition considered , but in all cases , the neuronal data conformed closely to the simulation results . Such agreement supports several key features of the model , including the competitive interactions between the target- and bias-driven responses , the limited yet visible impact of intrinsic randomness ( in build-up rate ) on the evoked responses , and a central hypothesis of the model — that the fluctuations in baseline at the two relevant locations are directly linked and possibly causal to the subsequent movement-related dynamics and , ultimately , to the RTs generated . We examined two aspects of the model that could potentially limit its significance . The first is the degree to which it depends on the precise mix of V , VM , and M cells included in the analyses . This is an issue because it is unclear whether all cell types contribute equally to the FEF output , that is , to the signal that is thought to cross a threshold to trigger a saccade . For example , the purely visual ( V ) neurons could have a lesser weight than those with movement-related activity . To investigate this , we repeated all analyses excluding all of the V neurons ( n=26 cells with visuomotor index < 0 . 46 ) from the population averages , and re-fitted the model in accordance to this more restricted data set . Quantitatively , the main difference was that variations in baseline were diminished , but qualitatively , the results were similar to those based on the larger population ( compare Figure 7 and Figure 7—figure supplement 2 ) . Furthermore , the model was still able to simultaneously replicate the neural activity and the RT distributions accurately ( Figure 7—figure supplement 2 ) . Thus , neither the neuronal averages nor the model results are overly sensitive to the magnitude of the visual component of the population response . The second potential concern is whether the model generalizes to other tasks or experimental conditions . Consistency with prior studies indicates that the reward manipulation in the 1DR task simply exaggerates and exposes mechanisms that are always operating ( see 'Discussion' ) . However , to test this more directly , we explored the conditions needed for the model to reproduce the data in the ADR task , in which all directions are equally rewarded ( Figure 1b ) . In the ADR condition , the average baseline level was constant ( Figure 7—figure supplement 3 , compare [c] vs . [d] ) , as expected given that there was no spatial asymmetry in that case . Accordingly , in the model , the two mean baseline levels were set equal to the measured experimental value ( BT=BD=0 . 2 in Equation 5 ) . Apart from that , matching the model to the ADR data required only two additional parameter adjustments: lowering the variability of the baselines , and slightly decreasing the maximum build-up rate of the stimulus-driven response ( Figure 7—figure supplement 3 ) . Everything else was as in the 1DR simulations . With those changes , the model generated less extreme variations in baseline , build-up rate , and threshold , and was able to replicate both the neural activity and the RT distribution measured in ADR trials ( Figure 7—figure supplement 3 ) . These ADR results provide an important consistency check , showing that the same competitive dynamics postulated for the congruent and incongruent 1DR conditions are fully compatible with the simpler , unbiased case . In characterizing the functional roles of specific brain circuits , one of the main challenges is dealing with the inevitable heterogeneity of cell types and their specializations ( Zeng and Sanes , 2017 ) . Not surprisingly , single FEF neurons showed a variety of relationships to RT in the 1DR task . Remarkably , however , the model accounted for much of this diversity on the basis of a simple intuition: the build-up rate of the target-driven activity is determined by two factors , one that is coupled to the baseline and another that is not , and those factors are weighted in different proportions across single neurons . In this section . we first describe the range of RT preferences measured in single FEF cells and then show that those diverse preferences naturally fall out of the elements already built into the model . We examined the responses of individual FEF neurons during correct saccades into the RF , and found that their dependencies on RT could deviate quite substantially from those of the average population . This is illustrated with two example cells for which the maximum level of activity across trials was modulated strongly — and in opposite directions ( Figure 9a–f ) . To view all the responses recorded from a given neuron simultaneously , these were arranged as activity maps in which color corresponds to firing intensity and trials are ordered according to RT ( Figure 9a , d ) . In this way , it is clear that both cells were most active shortly before the saccade ( white marks on the right ) and that their firing rates were very different for the fastest versus the slowest responses ( top vs . bottom trials ) . One cell preferred fast trials , that is , it fired at higher rates for short RTs , whereas the other preferred slow trials , that is , it fired at higher rates for long RTs . The contrast is also apparent when the same data are plotted as collections of firing rate traces sorted and color-coded by RT , as done in previous figures ( Figure 9b , e; compare to Figure 7e ) . For each recorded neuron , sensitivity to RT was quantified by ρ ( Rp , RT ) , the Spearman correlation between the peak response and RT across trials ( 'Materials and methods , Statistical analyses' ) . Negative values correspond to a preference for short RTs , as for the first example cell ( Figure 9c ) , whereas positive values correspond to a preference for long RTs , as for the second example cell ( Figure 9f ) . Across our FEF sample , the resulting distribution of correlation coefficients ( mean ρ=−0 . 12 , p=0 . 0005 , signed rank test ) was notable in two ways . First , it contained many more significant correlations , both positive and negative , than expected just by chance ( 43 of 132 cells ∼33% were significant with p< 0 . 05 , as opposed to 6 . 75 expected by chance; p=10−22 , binomial test ) . Thus , a substantial fraction of the FEF neurons had robust temporal preferences , with both modulation types represented ( Figure 9g , colored points ) . Second , the distribution was approximately the same for all the standard FEF cell types . The proportion of positive and negative correlations , as well as the fraction of significant neurons , was statistically the same for the V , VM , M , and other categories ( Figure 9g; p>0 . 2 , binomial tests ) . So , as far as we could tell , the sensitivity to RT spanned a similar range for all the elements of the FEF circuitry . These results explain the moderate RT sensitivity seen in the average population activity during IOI trials ( Figure 7e ) as the sum of two opposing contributions from subpopulations with temporal preferences that partially offset each other . Within the fast-preferring group ( ρ<0 ) , the pattern of response trajectories of many cells was qualitatively similar to that of the average population but showed more extreme variations across RT bins ( Figure 10f; compare to Figure 7e ) . By contrast , only a few of the neurons within the slow-preferring group ( ρ>0 ) demonstrated a strong correlation with RT ( Figures 9d–f and 10g ) ; for most of them , the dependence on RT , particularly during the ∼100 ms before movement onset , was more modest ( Figure 10h ) . Thus , when the responses of the fast- and slow-preferring neurons are combined , their temporal dependencies partially cancel out , and the overall population activity ends up resembling an attenuated version of the fast-preferring responses . Similar results were obtained in the congruent condition . That is , both fast- and slow-preferring cells were also found during III trials ( Figure 10i , j ) , except that in this case , the complementary modulations canceled out more fully upon averaging ( Figure 7g ) . The temporal heterogeneity just discussed was readily replicated by the model . For incongruent trials , strong modulation could be simulated favoring either short ( Figure 10a ) or long RTs ( Figure 10b ) , but more modest temporal sensitivity like that exhibited by the majority of slow-preferring neurons could be reproduced too ( Figure 10c ) . In all of these cases , the simulated response trajectories matched the experimental results extremely well ( compare to Figure 10f–h ) . The most visible ( but still minor ) discrepancy between the simulated and neuronal trajectories was due to the discontinuity of the threshold-crossing event in the former , as opposed to the sharp but smooth turn around the peak of activity of the latter . Analogous results were obtained in congruent trials , for which both fast- ( Figure 10d ) and slow-preferring ( Figure 10e ) model responses similar to the experimental ones were also generated ( compare to Figure 10i , j ) . In both congruency conditions , the simulated slow-preferring responses are particularly notable because , although they are still target-driven , they would seem to require mechanisms that are completely different from those described earlier . How did the model capture such wide-ranging heterogeneity ? The short answer is that , without changing any of the parameters , the target-driven activity in the model , RT , can be naturally expressed as the sum of two components: ( 1 ) RT ( t ) =RTU ( t ) +RTC ( t ) where the equality holds at every point in time . For one component ( indicated by the C superscript ) , the variations in build-up rate are coupled to the fluctuations in baseline ( BT ) , whereas for the other component ( U superscript ) , the variations in build-up rate are random , uncoupled from the baseline . These two components correspond to the two neuronal types with opposite RT preferences . To see the correspondence , consider once more the RT-sensitive responses of the FEF cells , now focusing on how the response trajectories fan out when they are aligned on the go signal: for the fast-preferring examples , the slopes of the curves increase progressively as the RTs become shorter , and the spread is visible from the moment the activity begins to rise ( Figure 10a , d , f , i , left panels ) ; by contrast , for the slow-preferring examples , all the curves begin to rise with approximately the same slope , and the modulation by RT begins to manifest only later , ∼100 ms after the go signal ( Figure 10b , e , g , j , left panels ) . According to the model , this feature , the variability of the initial build-up rate , is the fundamental mechanistic distinction between the fast- and slow-preferring FEF neurons . A more elaborate intuition can be gleaned from the analytical expression that determines the initial build-up rate of the target-driven activity , GT , in each trial . This build-up rate can be written as ( 2 ) GT=f1+ϕ+f2 BT ( see Equations 8 , 9 ) . Here , the terms f1 , f2 and ϕ are not necessarily constant , but what matters is that they do not depend on the baseline at the target location , BT . The term ϕ , which for now is assumed to be relatively small , represents noise in GT , that is , the random fluctuations in build-up rate mentioned earlier . Intuitively , then , Equation 2 says that the initial build-up of RT is the result of two influences , a term that depends on the baseline BT and a relatively constant drive that is independent of it . The former , f2 BT , leads to much higher variability in build-up rate across trials — and stronger covariance with RT — than the latter , f1+ϕ . Now , the coupled and uncoupled components in Equation 1 differ exclusively in their initial build-up rates , which are given by the two terms just discussed:GTU=αf1+f1+ϕ ( 3 ) GTC=−αf1+f2 BT . A key property of these build-up rates is that their sum , GTU+GTC , is always equal to GT , as given by Equation 2 . This is true for any value of the newly introduced parameter α , which serves as an offset by means of which the weights of the two components may be adjusted . Splitting the target-driven activity in this way allowed us to simulate neuronal responses RTU and RTC that had opposite RT preferences but were otherwise identical , as they had the same initial conditions , afferent delays , evolution equations , and so on . Furthermore , by varying α , we could vary the strength of the resulting modulation — without altering either the original target-related activity , RT ( t ) , or the outcomes and RTs of the competitions in any way . In other words , the split via Equations 1–3 produces paired sets of target-driven responses with different RT sensitivities , parameterized by α , but all the pairs thus generated are compatible with the same summed activity ( Figure 7 ) and the same distributions of outcomes and RTs ( Figure 5g–i ) . On the basis of this simple decomposition of RT into pairs of components , the model can generate a wide range of RT-sensitive responses , which are strikingly similar to those found in the FEF population . In closing this section , we emphasize the distinct role that intrinsic randomness plays in the model , and why it is necessary . During incongruent trials ( or , more precisely , when BT≤BD ) . the variance in RT is so strongly coupled to the fluctuations in baseline activity that the noise in the build-up rate has a negligible impact ( in Equation 2 , ϕ≪f2 ) . In that case , even large variations ( of ∼100% ) in noise have relatively little consequence , and RTC and RTU correspond almost perfectly to the fast- and slow-preferring neurons , respectively . By contrast , during congruent trials ( or , more precisely , when BT>BD ) , a relatively large amount of independent noise ( ϕ>f2 ) is necessary to reproduce the lack of sensitivity to RT in the average population activity ( Figure 7g ) . In that case , the preferences for short and long RTs of the simulated responses ( Figure 10d , e ) are noticeably sensitive to modest variations ( of ∼20% ) in noise , and the temporal heterogeneity of the RTU and RTC components is more complicated; the details are beyond the scope of this report . Nevertheless , the conclusion is clear: the large variance in RT observed during incongruent trials vastly exceeds that associated with intrinsic noise in the rising activity , and is fundamentally determined by the covert conflict between competing saccade plans and the ensuing dynamics; whereas the much smaller variance in RT observed during congruent trials is best explained by noise that is independent of the competitive interactions , which make just a modest contribution in that case . We examined single-neuron activity in FEF , a cortical area whose role in saccade generation and attentional deployment is firmly established ( Bruce and Goldberg , 1985; Tehovnik et al . , 2000; Squire et al . , 2012 ) , and where the activity of movement-related neurons is perhaps most emblematic of the idealized rise to threshold ( Hanes and Schall , 1996; Fecteau and Munoz , 2007; Stanford et al . , 2010; Ding and Gold , 2012; Costello et al . , 2013 ) . We found that the three main components of this process — the baseline activity preceding target presentation , the build-up rate of the evoked response , and the saccade threshold — fluctuate in a coordinated fashion . This is already a significant departure from the simpler , standard framework in which the only source of variability ( within a given experimental condition ) is the build-up rate . But the problem is more complicated , because by themselves , the interrelationships between these three variables are insufficient to explain the variations in RT accurately; for that , it is critical to consider not only the target-driven response but also the weaker , internally driven activity favoring saccades to alternative locations . During reactive saccades , it may seem as if only one motor plan is possible , but this is rather illusory . When the target appears , oculomotor activity begins to grow in response to it , but this ramping process ( represented by RT in the model ) does not start from the same neutral state every time; instead , it occurs while other incipient , internally driven motor plans ( represented by RD ) are also developing , and the time necessary to resolve the ensuing conflict depends on how advanced and how congruent those budding , bias-driven plans are relative to the target-driven response . Thus , the high variance in RT results not from noisy representations or sloppy computations , but rather from the normal operation of a well-oiled motor-selection machine ( Najemnik and Geisler , 2005; Oostwoud Wijdenes et al . , 2016; Tian et al . , 2016 ) . Making the target unique , highly visible , and task-relevant minimizes potential variance related to the sensory detection step and enhances the priority of the target-related plan , but still leaves those alternative internal plans relatively unconstrained . What the 1DR task does is to align the internal biases with a specific direction — that of the expected reward — and it is under those conditions that the motor selection process becomes more apparent . Contrary to current ideas , we found that noise in the build-up rate is not the main source of RT variance . Such noise was discernible but only during congruent trials , when the target-driven activity typically starts with the higher baseline and there is minimal motor ambiguity to begin with ( Figure 3a ) . More typically , the proportion of variance in RT due to such intrinsic randomness is smaller ( ADR task ) , and may become negligible ( 1DR task , incongruent trials ) as motor conflict increases . The model revealed completely novel mechanistic details of the motor selection process but , on average , its manifestations during the rise to threshold were nevertheless quite subtle , particularly during correct saccades . Consider , for example , how the longest RTs in correct trials are produced ( Figure 5b , h ) . That was quite a puzzle . In that case , the target-driven activity , which starts with a lower baseline than the opposing plan , must somehow rise at about the lowest possible build-up rate and yet still win the competition by a large margin . The solution is for the target-driven response to rise quickly initially , suppress the opposing plan early on , and then slow down immediately afterward — all of which happens when the competing plans are far from threshold . This is a major departure from standard choice models , in which the outcome is not determined until activity is close to or at threshold . It means that critical dynamical interactions may occur at quite low levels of activity , where they are less readily apparent and much more difficult to characterize without prior knowledge of their signature features . The model has numerous moving parts , but consider the scope of the data that it reconciles ( 'Materials and methods , Correspondence between data and model parameters' ) . Behaviorally , the model generated errors at the appropriate rates and reproduced three distinct RT distributions in their entirety ( Figure 5g–i ) . Neurophysiologically , it replicated the average response trajectories in all conditions ( Figure 5d–f ) , the isolated effect of the spatial bias ( Figure 6c–f ) , the dependence on RT of the average activity ( Figure 7 ) , and most remarkably , the responses of individual FEF neurons , which showed a wide range of RT preferences ( Figure 10 ) . With minimal adjustment , the model also replicated the empirical results in the ADR task ( Figure 7—figure supplement 3 ) . Thus , a large number of disparate observations are subsumed into one coherent framework for resolving the kind of saccadic conflict that must typify naturally occurring oculomotor behaviors . Our results indicate that the saccade threshold is not constant . Instead , it fluctuates quite dramatically , and in tandem with other elements of the circuitry . Although the proposed dynamics remain to be tested directly , substantial agreement can already be found with extant data . For instance , in the model , a stronger internally driven activity promotes a lower threshold , and indeed , movement-related activity preceding memory-guided saccades , anti-saccades , or saccades triggered by blinks is considerably weaker than that for stimulus-driven saccades ( Edelman and Goldberg , 2001; Jantz et al . , 2013; Jagadisan and Gandhi , 2017 ) . In addition , the presaccadic activity measured during visual search is less vigorous for incorrect than for correct responses to the same location ( Thompson et al . , 2005 ) , presumably because the former involve a stronger internal ( and erroneous ) drive that promotes a lower threshold , just as in our data . And , again in the context of visual search , one study ( Heitz and Schall , 2012 ) showed multiple differences in FEF activity across task conditions similar to those found here; that is , when the mean RT was shorter , the baseline , build-up rate , and threshold were all higher . These findings place significant constraints on the trigger mechanism that converts a saccade plan into a committed , uncancelable command . For instance , a popular idea is that adjustments in threshold may serve to trade speed against accuracy during choices ( Lo and Wang , 2006; Bogacz et al . , 2010; Heitz and Schall , 2012; but see Salinas et al . , 2014; Thura et al . , 2014; Thura and Cisek , 2016 , 2017 ) . This is simply because , everything else being equal , activity that ramps toward a higher threshold should take longer to reach it , thus providing more time for deliberation . However , the results in the 1DR task are entirely antithetical to these notions: first , congruent trials produce shorter RTs and higher accuracy than incongruent ones , and second , the observed variations in threshold are linked to variations in baseline and build-up rate in ways that , according to the standard characterization of the speed-accuracy tradeoff , are entirely inconsistent . A higher threshold , which lengthens the deliberation period , is typically accompanied by a higher baseline and a higher build-up rate , both of which shorten that period . These considerations are significant because , although the threshold is a universal feature of decision-making models , the strongest evidence of its existence is precisely the behavior of oculomotor neurons in FEF and SC ( Hanes and Schall , 1996; Lo and Wang , 2006; Brown et al . , 2008; Stanford et al . , 2010; Ding and Gold , 2012 ) . In other related circuits , either no threshold is apparent ( Ding and Gold , 2010; Stuphorn et al . , 2010 ) or its implementation is much less evident ( Afshar et al . , 2011; Hayden et al . , 2011 ) . Sumner ( 2011 ) has pinpointed why explaining saccadic latency distributions has been so challenging: it is well established that Gaussian variability in the build-up rate of a rise-to-threshold process accurately reproduces their characteristic skewed shapes ( Carpenter and Williams , 1995; Hanes and Schall , 1996; Fecteau and Munoz , 2007 ) , but most factors that are known to affect saccadic RT in simple tasks are normally modeled as baseline shifts ( Sumner , 2011 ) . Our results suggest that the dichotomy is false . The fluctuations in baseline , build-up rate , and RT are inextricably linked ( Figures 4 , 6 , 7 and 8 ) ; they involve alternate , covert motor plans and depend on multiple neural mechanisms acting in concert . According to the model , the major source of randomness across trials is the variability of the baselines ( Equation 5 ) . The noise in the target-driven build-up rate ( ϕ in Equation 2 ) makes a distinct contribution , but in the incongruent condition , in particular , nearly all of the variance observed experimentally — in RT , in saccadic choice , in threshold level , and in the build-up rates and peak responses of the neurons — is determined by the computational amplification of the initial baseline fluctuations . This is a simplification , of course . First , it is possible that the link is not strictly causal , that is , that an unidentified factor drives the fluctuations in baseline and in the other dynamic variables . And second , other sources of variability are likely to contribute too; for instance , variations in response onset ( Pouget et al . , 2011; Peel et al . , 2017 ) , which would correspond to fluctuations in the afferent delays of the model . These contributions , however , are likely to be very small in comparison ( Pouget et al . , 2011 ) . It is certainly possible to add noise to the afferent delays or to other components of the model without substantially altering the results , but what is notable is that this is not necessary . Arguably , the baselines reflect multiple cognitive elements , including expectation , anticipation , and the allocation of attention and other resources ( Bruce and Goldberg , 1985; Kastner et al . , 1999; Coe et al . , 2002; Maunsell , 2004; Rao et al . , 2012; Zhang et al . , 2014; Thura and Cisek , 2016 ) . In the model , these elements set the initial spatial priorities such that the neurons with the higher baseline have a higher probability of triggering the next saccade . This probability is a complicated function of the baseline values and the upcoming stimulus , but qualitatively , the effect is very much in agreement with the findings of other single-neuron studies ( Everling et al . , 1998; Everling and Munoz , 2000 ) and with the results of subthreshold microstimulation experiments ( Glimcher and Sparks , 1993; Dorris et al . , 2007 ) . It is also in line with theoretical studies showing that the background activity in recurrent circuits can have profound dynamical and amplification effects on evoked responses ( Salinas and Abbott , 1996; Chance et al . , 2002; Salinas , 2003; York and van Rossum , 2009 ) . The model proposes a tight relationship between initial state and subsequent oculomotor dynamics , and interestingly , mounting evidence demonstrates a similar phenomenon in motor cortex , whereby the initial neural state is predictive of an ensuing arm movement and of the trajectories of the underlying neural signals over time ( Churchland et al . , 2006 , 2010 , 2012; Afshar et al . , 2011; Ames et al . , 2014; Sheahan et al . , 2016; Stavisky et al . , 2017; Wang et al . , 2018 ) . In that case , the dynamics develop within a very high-dimensional space , such that the preparatory activity is only weakly related to specific kinematic parameters ( Churchland et al . , 2006 , 2010 ) . Saccades are simpler because they are lower dimensional and largely stereotyped , and because the activity of any given cell generally corresponds to a fixed movement vector . However , we propose that their dynamical behavior is qualitatively similar in that the initial state of the system — that is , the configuration of baseline levels across RFs during fixation — fundamentally determines its subsequent temporal evolution , including its interaction with new incoming sensory information ( Sheahan et al . , 2016 ) and the eventual outcome ( Churchland et al . , 2006; Afshar et al . , 2011 ) . Overall , the current results suggest that , when oculomotor circuits receive new visual information , ongoing saccade plans and internal settings ( e . g . , threshold level and attentional locus ) radically shape the impact of that information , even when it is behaviorally relevant , expected , and unambiguous . Deeper understanding of the underlying network dynamics will be critical for further elucidation of how saccades are triggered and , more generally , of how perceptually guided choices are made . Experimental subjects were two adult male rhesus monkeys ( Macaca mulatta ) . All experimental procedures were conducted in accordance with the NIH Guide for the Care and Use of Laboratory Animals , USDA regulations , and the policies set forth by the Institutional Animal Care and Use Committee ( IACUC ) of Wake Forest School of Medicine . An MRI-compatible post served to stabilize the head during behavioral training and recording sessions . Analog eye position signals were collected using a scleral search coil ( Riverbend Instruments , Birmingham , AL ) and infrared tracking ( EyeLink 1000 , SR Research , Ottawa , Ontario , Canada ) . Stimulus presentation , reward delivery , and data acquisition were controlled by a purpose-designed software/hardware package ( Ryklin Software , New York , NY ) . Target stimuli were displayed by a 48×42 array of tri-color light-emitting diodes . Saccade onset was identified as the time at which eye velocity exceeded 50∘/s; having detected the start of a saccade , its end was identified as the time at which eye velocity fell below 40∘/s . Eye movements were scored as correct if the saccade endpoint fell within 5∘ of the target stimulus . Neural activity was recorded using single tungsten microelectrodes ( 2–4 MΩ , FHC , Bowdoin , ME ) driven by a hydraulic microdrive ( FHC ) . Individual neurons were isolated on the basis of the amplitude and/or waveform characteristics of the recorded and filtered signals ( FHC; Plexon , Inc , Dallas , TX ) . Putative FEF neurons were selected from areas in which saccade-like movements could be evoked by low-current microstimulation ( 70 ms stimulus trains at 350 Hz , with amplitude equal to 50 μA ) ( Bruce and Goldberg , 1985; Costello et al . , 2013 ) . Neurons were recorded unilaterally in both monkeys . The majority of RFs ( 76% ) were located at 10∘ of eccentricity . In the 1DR task ( Figure 1a ) , all trials began with the appearance of a centrally located stimulus . Monkeys had to maintain their gaze on this fixation spot for 1000 ms . The disappearance of the fixation spot ( go signal ) was simultaneous with the appearance of a second , target stimulus in one of four possible locations ( Figure 1a ) . Subjects were required to make a saccade to the peripheral target within 500 ms of the go signal in order to receive a liquid reward . Target locations varied pseudorandomly from trial to trial . In each block of trials , only one of the four target locations was associated with a large reward; the other three were unrewarded ( Monkey G ) , or yielded a much smaller reward ( Monkey K ) . For brevity , we refer to these simply as the ‘rewarded’ and ‘unrewarded’ locations . The rewarded location changed pseudorandomly from one block to another . Block length was highly variable ( range: 10–140 trials ) ; the average was 70 trials per block . In the all-directions-rewarded task ( ADR ) , the events were the same as in the 1DR , but the four target locations were rewarded equally ( Figure 1b ) . Blocks of ADR trials were run sporadically , interleaved with those of 1DR trials . In the delayed-saccade task , each trial began with fixation of a central spot , followed by the appearance of a single stimulus at a peripheral location during continued fixation . After a variable delay ( 500 , 750 , or 1000 ms ) , the fixation spot was extinguished ( go signal ) and the subject received a liquid reward if a saccade was made to the peripheral target . In each experimental session , the delayed-saccade task was run first to locate the RF of the recorded neuron , and subjects performed the 1DR task after the initial spatial characterization . The four target locations in the 1DR task were chosen on the basis of the RF of each recorded neuron . One location always corresponded to the RF center . The others had equal eccentricity and were 90∘ , 180∘ , and 270∘ away from the RF relative to fixation . The RT was always measured from the go signal until the onset of the saccade . For all analyses not specifically examining sequential effects and block transitions , we discarded the first 8 trials of each 1DR block , during which the monkeys may have been discovering the new rewarded location ( Figure 2—figure supplement 1a ) . This guaranteed that all behavioral and neural metrics reflected a stable expectation , and that erroneous saccades were not due to spatial uncertainty . More stringent exclusion criteria produced qualitatively similar results . Continuous ( or instantaneous ) firing rate traces , also known as spike density functions , were computed by convolving evoked spike trains with a Gaussian function ( σ=15 ms ) with unit area . Continuous mean traces for each neuron were generated by averaging across trials . To produce equivalent population responses ( e . g . , Figure 3a–c ) , the continuous traces of individual cells were first normalized by each neuron’s overall maximum firing rate and were then averaged across neurons . To visualize how RT modulated the activity of each cell , families of firing rate traces ordered and color-coded by RT were generated ( Figure 9b , e ) . For this , the trials in the relevant experimental condition ( e . g . , IOI , III ) were sorted by RT and distributed over 20 evenly spaced , overlapping bins defined by RT quantiles , where each bin contained 20% of the trials . Thus , the first bin was centered on the 10th percentile and included the fastest 20% of the recorded trials , the next bin included the 20% of the trials around the 14th percentile , and so on , with the last bin being centered on the 90th percentile and including the slowest 20% of the recorded trials . Then , a continuous firing rate trace was produced for each of the 20 RT bins/quantiles . To generate equivalent families of curves for populations of neurons ( Figure 7e–h ) , the traces for each participating cell were first normalized by that cell’s overall maximum firing rate , and then , for each quantile , a population trace was compiled by averaging across neurons . This method — based on quantiles , as opposed to standard binning using fixed RT values —reveals more clearly the actual modulation range of the neural responses and their smooth dependence on RT , because the numbers of trials and neurons remain nearly constant across bins . Also , if the dependence on RT is monotonic , as the data indicate , this procedure can only underestimate the magnitude of the modulation . For the simulated data , families of response curves ordered by RT were generated in the same way . For each neuron , an activity map ( Figure 9a , d ) for a given condition ( e . g . , IOI , III ) was assembled by aligning all spike trains to the go signal , converting each one to a continuous firing rate , sorting the trials by RT , putting the sorted firing rate traces into a single matrix , and displaying the matrix as a heat map with color indicating intensity . For display purposes , activity maps were also smoothed with a Gaussian function in the vertical direction , that is , across trials ( σ=2 trials ) , but this was exclusively for ease of viewing; trials were kept independent in all analyses . All data analyses were performed using customized scripts in Matlab ( The MathWorks , Natick , MA ) . For comparisons across any two conditions , significance was typically evaluated using permutation tests for paired or unpaired samples ( Siegel and Castellan , 1988 ) , as appropriate . Because 100 , 000 permutations were used , the smallest significance value in this case is reported as p<10−5 . The relationship between neuronal activity and RT for each neuron was evaluated separately for each experimental condition ( e . g . , III , IOI ) using the Spearman rank correlation coefficient . The Matlab function corr was used for this . This coefficient serves to identify any monotonic relationship between two variables . As measures of activity , for each cell we considered: the baseline response , Rb ( firing rate in a 250 ms window preceding the go signal ) ; the mean response , Rm ( firing rate computed over the full RT interval ) ; the presaccadic response , Rsac ( firing rate in a 50–70 ms window preceding saccade onset ) ; the build-up rate , RBU ( described below ) ; and the peak response , Rp ( described below ) . The correlation between Rp and RT is denoted as ρ ( Rp , RT ) . Correlations in other activity measures are denoted similarly ) . Neurons with ρ ( Rp , RT ) <0 and ρ ( Rp , RT ) >0 were designated as fast- and slow-preferring , respectively . The standard presaccadic firing rate , Rsac , which is calculated in a fixed time window anchored to saccade onset , would seem to be the most direct indicator of saccade threshold . However , for a purely visual neuron responding to target onset , the use of Rsac could produce a negative correlation with RT even if that cell was always activated with the same temporal profile . To avoid this spurious correlation due to temporal misalignment between the visually driven spikes and saccade onset across trials , we computed the peak response , Rp , which is insensitive to the alignment of the spike trains . Although , on average , results based on Rp and Rsac were highly consistent ( e . g . , compare Figure 3d and Figure 5—figure supplement 1a ) , the former is a more veridical indicator of response modulation on a single-cell basis . For each neuron , the value of Rp in each trial was equal to the cell’s firing rate computed in an interval centered on the time point Tp , which we call the time of peak response . This is simply the time along a trial ( with the go signal at t=0 ) at which the cell was most likely to fire at the highest rate ( Figure 9a , d , black marks ) . Tp is described as a linear function of RT , ( 4 ) Tp=β0+β1×RTwhere the coefficients β0 and β1 characterize the timing of each neuron . For example , for a typical visual cell with β0=80 and β1=0 , the maximum rate in a trial is observed 80 ms after the go signal , regardless of RT , whereas for a typical movement-related cell with β0=−30 and β1=1 , the highest discharge occurs 30 ms before the saccade . The coefficients β0 and β1 were obtained in two steps: ( 1 ) finding , from the activity map of the cell , the maximum instantaneous firing rate in each trial and the time , relative to the go signal , at which that rate was achieved , Tmax , and ( 2 ) fitting Tmax as a linear function of RT . All trials in which a saccade was made into the cell’s RF were included , regardless of bias condition . The coefficients resulting from the fit , that is , the intercept and slope , were β0 and β1 . Finally , to determine Rp in a given trial , first , the corresponding Tp was found by plugging the RT from that trial into Equation 4; then a firing rate was calculated by counting the spikes in a time window centered on Tp and dividing this by the window length . The result was Rp . The window length was 100 ms for most cells ( ∼80% ) but was set to 200 ms for a minority that had more prolonged responses ( e . g . , postsaccadic cells ) . The build-up rate , RBU , was computed for each cell and each trial by calculating the excursion in firing rate from the initial response to the peak , Rp−Ron , and dividing it by the time interval Tp−Ton between the onset of the rise and the time of peak response . For the peak response , Rp and Tp are as defined above , whereas for the onset of activity , Ron and Ton , are as follows . The onset time corresponds to the latency of the cell’s response , that is , to the time point between target onset and Tp at which the activity of the neuron starts ramping up . This latency , Ton , was computed separately for each trial using the method developed by Rowland et al . ( 2007 ) . The firing rate Ron was calculated by counting the spikes in a time window centered on Ton and dividing this by the window length ( 50 ms ) . Several other methods were tested for computing the build-up rate in individual trials . The results reported ( Figure 4b ) were robust across methods . The 132 FEF neurons were classified by comparing their responses ( mean firing rate in windows of 100–250 ms ) during fixation , during the RT interval , and after the saccade . Multiple-comparison tests were performed by ANOVA . Accordingly , cells that were maximally active before the go signal were classified as fixation neurons ( n=12 ) ; cells that responded significantly above baseline , but only after the saccade , were deemed postsaccadic ( n=18 ) ; and neurons that began to respond shortly after the go signal and that were still significantly active after the saccade were deemed wide-profile ( n=10 ) . The latter group could conceivably have been included in the visuomotor category described below , but given their peculiar lack of sensitivity to the saccade , they were analyzed separately . The rest of the neurons ( n=92 ) had standard visuomotor properties and responded significantly above baseline between the go signal and saccade onset . A visuomotor index , which was equated to the coefficient β1 in Equation 4 , was used to characterize them ( Figure 9g ) . This coefficient naturally serves as a visuomotor index because it describes the degree to which a neuron is activated by a stimulus in its RF ( in which case β1≈0 ) as opposed to an eye movement toward it ( in which case β1≈1 ) — which is the classic criterion used to classify FEF cells ( Bruce and Goldberg , 1985 ) . So , based on their β1 values , the remaining 92 neurons were classified as either visual ( V; n=26 ) , visuomotor ( VM; n=43 ) , or motor ( M; n=23 ) . As expected from previous studies ( Bruce and Goldberg , 1985; Costello et al . , 2013 ) , during delayed-saccade trials , the neurons thus classified as V responded briskly to the presentation of the target stimulus in the RF and gradually decreased their activity thereafter; M neurons showed no activity linked to stimulus presentation but responded intensely just before movement onset; and the cells in the VM group showed both visual and presaccadic activation . Two of the 92 neurons with standard visuomotor properties lacked well-defined RFs and were excluded from the population averages . Of the remaining 90 units , only the 62 that had error trials ( IOO , OII ) were considered for comparisons between correct and incorrect responses ( Figure 3 ) . Further comparisons between the model and neural populations presented in the main text were based on 84 neurons , with the 6 units with the lowest visuomotor indices ( β1<0 . 1 ) also being excluded from the main pool of 90 , as these were unlikely to carry any trace of the motor signal described by the model . However , these 6 cells were included in all single-cell analyses and when explicitly dividing the population according to visual versus motor activity ( Figure 3—figure supplement 1 ) . To tease apart the effects of RT and spatial bias on FEF activity , we devised a procedure for equalizing the RT distributions of the congruent and incongruent conditions . For each recorded cell , the observed distributions in III and IOI trials ( Figure 6a ) were sub-sampled as follows . An III trial was selected randomly and the IOI trial with the most similar RT was identified; then , if the RT difference was smaller than 15 ms , the two trials were accepted into the respective sub-samples and removed from the original pools , or else the III trial was discarded . After probing all of the III trials like this , the resulting sub-sampled pools ( Figure 6b ) contained equal numbers of trials with virtually identical RT sets . To account for variations due to random resampling , all results based on RT matching were repeated 50 times and averaged . The model consists of two populations of FEF neurons that trigger saccades toward locations T ( where the target stimulus appears ) and D ( diametrically opposite to T ) , their activities represented by variables RT and RD . After the go signal is given , both motor plans begin to increase , and the first one to reach a threshold Θ wins the competition , thus determining the direction of the evoked saccade and the RT . If RT wins , the saccade is correct , toward T , whereas if RD wins , the saccade is incorrect , toward D . Each simulated race corresponds to one trial of the 1DR task . The actual trajectories followed by RT and RD in each trial are dictated by the dynamics and interactions described below . Matlab scripts for running the model are provided as supplementary source code files . In each simulated trial , three key quantities need to be specified before the race between RT and RD can take place: the baseline firing levels , which serve as the initial values for RT and RD , the initial build-up rates of the two motor plans , and the threshold , Θ . Simulated neural responses are scaled so that the firing activity at threshold is around 1 . The baselines for the target and distracter locations , BT and BD , are specified first , drawn according to the following expressions , BT=⟨ BT ⟩ [1+σ ϵT]0 ( 5 ) BD=⟨ BD ⟩ [1+σ ϵD]0where the variability across trials is determined by σ=0 . 28 , and ϵT and ϵD are random Gaussian samples ( different for each trial ) with negative correlation equal to −0 . 5 , zero mean , and unit variance . Here and in the expressions below , the square brackets with a subscript indicate that there is a minimum floor value beyond which the argument cannot drop , that is , [ x ]a=max{x , a} . This ensures , for instance , that quantities commensurate with firing activity are not negative . The mean baseline levels in Equation 5 , ⟨ BT ⟩ and ⟨ BD ⟩ , are set to be consistent with the spatial bias: in incongruent trials ⟨ BT ⟩=0 . 16 and ⟨ BD ⟩=0 . 34 , so the target side has the lower baseline on average , whereas in congruent trials ⟨ BT ⟩=0 . 34 and ⟨ BD ⟩=0 . 16 , so the target side takes the higher value . Stated differently , the higher mean baseline is always assigned to the rewarded location . Once the baselines are drawn , the other key quantities , the threshold and the initial build-up rates , can be set for the trial . The threshold is given by ( 6 ) Θ=[1 . 185+1 . 2 ( BT−BD ) ]0 . 73where Θ cannot drop below 0 . 73 , as indicated by the square brackets . This expression means that the threshold for triggering a saccade increases with the baseline level on the target side and decreases with the baseline level on the opposite side , but cannot be less than a certain minimum . The initial build-up rates ( or gains ) of the motor plans also depend on the baselines . For the internally driven plan , RD , the build-up rate is ( 7 ) GD=0 . 001 [1 . 4+1 . 7 ( BD−BT ) ]0so the activity favoring D rises most steeply when the baseline at that location is high and the baseline at the target location is low . For the target-driven motor plan , RT , the rise in activity has two distinct regimes . In the first one , when BT≥BD , the initial build-up rate is given by ( 8 ) GT=0 . 001 ( 6 . 16+0 . 55 η+2 . 5 BT ) where η is a random Gaussian sample ( zero mean , unit variance ) that varies stochastically across trials and represents intrinsic , baseline-independent fluctuations in build-up rate . In the second regime , when BT<BD , ( 9 ) GT=0 . 001 3 . 0+0 . 3 η+23 . 25 BT1+1 . 3 BDwhere the numerator has the same form as in Equation 8 but now BD appears in the denominator . The rationale for using two distinct expressions for GT is simply that the rise of the target-driven activity is very different when RT is already above the competition before the race begins compared to when it starts below the competition ( Figure 3a , b ) . In the former case ( BT≥BD , regime 1 ) the rise is always steep , the dependence on BT is weak , and there is no further opposition from the D plan ( note the absence of a BD term in Equation 8 ) . By contrast , when the target side starts at a disadvantage ( BT<BD , regime 2 ) , the initial build-up rate depends strongly on the actual baseline level , BT , and there is sizable competition from the D plan , instantiated as divisive suppression ( note the dependence on BD in the denominator of Equation 9 ) . It is important to realize that , because the baselines fluctuate across trials ( Equation 5 ) , in general , Equation 8 applies most often , but not uniquely , to congruent trials . Similarly , Equation 9 applies most often , but not uniquely , to incongruent trials . The build-up rate GT depends only on the baseline values themselves , regardless of the label assigned to the spatial configuration of each trial . In other words , the local competition process has no knowledge of what determines the baselines , it simply takes them as input and evolves accordingly . The main variables , RT and RD , which represent the activities of the competing populations , are updated in each time step Δt ( set to 1 ms ) as followsRT ( t+Δt ) =RT ( t ) +Δt VT ( 10 ) RD ( t+Δt ) =RD ( t ) +Δt VDwhere VT and VD are the instantaneous build-up rates . Initially ( that is , during fixation ) , the activities are equal to their respective baseline values , and they remain constant until the target/go signal is presented ( at t=0 ) . Thus , the initial conditions are RT=BT , RD=BD , VT=0 and VD=0 . The motor plans begin to advance thereafter , but not right away , because there is an afferent delay between the go signal and the actual onset of ramping activity . The target-driven plan , RT , begins to rise after a short delay AT=35 ms , and does so with the build-up rate prescribed by Equation 8 or 9 , whichever applies ( which can be coded as: if t≥AT , then VT=GT ) . The bias-driven plan , RD , begins to rise after a slightly longer delay AD=50 ms , but at the beginning , it is partially inhibited by the cue presentation . During this partial inhibition , which occurs between Ion=40 and Ioff=155 ms , RD rises slowly , at 38% of its nominal build-up rate , GD ( if t≥AD and t∈[Ion , Ioff] , then VD=0 . 38 GD ) . This dynamic is based on evidence indicating that the abrupt appearance of a visual stimulus , the target in this case , briefly interrupts or suppresses ongoing saccade plans ( reviewed by Salinas and Stanford , 2018 ) , a phenomenon known as ‘saccadic inhibition’ or the ‘remote distracter effect’ . After the inhibition period has elapsed , the bias-driven activity may rise in full force ( if t>Ioff , then VD=GD ) . The two motor plans then continue to advance until one of them reaches threshold . However , their build-up rates may change mid-flight as one plan goes past the other . These changes are dictated by two rules that describe the two possible ways in which the competition may end . Rule 1 ( T wins ) : if the target-driven firing rate , RT , exceeds the competing one at any point after its afferent delay has elapsed , then two things happen . First , RD is fully suppressed , so it stops increasing altogether ( if t>AT and RT>RD , then VD=0 ) . And second , the build-up rate of the T plan is adjusted so that ( 11 ) VTwin=−0 . 0088+2 . 6 GT ( if t>AT and RT>RD , then VT=VTwin ) . In this case , RT wins the race and the evoked saccade is correct , toward the target . The coefficients in Equation 11 are such that VTwin is typically smaller than GT . This means that the target-driven motor plan typically slows down after it overtakes the competition , and the lower its initial build-up rate , the more it slows down . This change in build-up rate represents the difficulty , or cost , of resolving the conflict for the T plan . Rule 2 ( D wins ) : if the internally driven firing rate , RD , exceeds the competing one at any point after its afferent delay has elapsed and outside of the transient inhibition interval , then RT can no longer advance past RD . In this case , RD simply continues to rise , winning the race without any further change in its build-up rate . The evoked saccade is incorrect , away from the target . The target-driven motor plan also keeps rising , but may suffer a minimal amount of suppression ( the amount needed to ensure that RT stays below RD for the remainder of the trial ) . Finally , to account for the characteristic fall in activity seen postsaccadically in FEF , after the winner reaches threshold , both motor plans decay exponentially toward a firing level of 0 . 2 with a time constant of 120 ms . Note that the model generates different outcomes and RTs based on just three quantities ( random numbers ) that vary across trials: ϵT and ϵD , which determine the baseline firing levels ( Equations 5 ) , and η , which adds independent noise to the build-up rate of RT ( Equations 8 , 9 ) . There are no other sources of variability . All parameter values in the model were adjusted to fit the experimental data . Each parameter typically has multiple effects , often on both the behavioral and neurophysiological responses simulated . For example , the afferent delays determine the short pause between the go signal and the onset of the rise to threshold ( seen experimentally in Figure 3a–c , left panels ) , but they also pin the left tails of the RT distributions ( Figure 5g–i ) because they determine the shortest possible RTs . Similarly , the variance of the baselines ( σ ) generally determines the correlation between activity and RT ( Figures 3 and 7 ) , but it also influences the widths of the RT distributions during incongruent trials ( Figure 5h , i ) . With this in mind , note that the parameters in Equations 5 , 6 were primarily set to match the baseline and threshold values measured from the FEF population across conditions ( Figure 5—figure supplement 1 ) . The parameters in Equations 7 , 8 , and 9 mainly determine the shapes of the RT distributions for incorrect incongruent , correct congruent , and correct incongruent trials , respectively ( Figure 5g–i ) . The parameters that describe the stimulus-driven suppression of RD determine the frequency of incorrect saccades , the shape of the corresponding RT distribution , and the shape of the simulated response trajectories in those incorrect trials . Finally , the parameters associated with Rule 1 scale the RT distribution for correct incongruent trials , and also determine the steepness of the simulated target-driven rise in activity . In all , 22 model parameters were adjusted to satisfy 23 basic experimental constraints: six baseline and six threshold values across conditions ( Figure 5—figure supplement 1 ) , two error rates , and three RT distributions ( Figure 5g–i ) , each of which requires , at a minimum , three parameters to be characterized . But note that the model accounted for many more features ( i . e . , degrees of freedom ) in the data , pertaining to the specific shapes of response trajectories , the effect of RT equalization , the correlation between firing activity and RT before and after target onset , and the response heterogeneity across individual FEF neurons .	As we examine the space around us our eyes move in short steps , looking toward a new location about four times a second . Neurons in a region of the brain called the frontal eye field help initiate these eye movements , which are known as saccades . Each neuron contributes to a saccade with a specific direction and size . Before a saccade , the relevant neurons in the frontal eye field steadily increase their activity . When this activity reaches a critical threshold , the visual system issues a command to move the eyes in the appropriate direction . So a saccade that moves the eyes to the right requires a specific group of neurons to be strongly activated – but , at the same time , the neurons responsible for movement to the left need to be less active . Imagine that you have to move your eyes as quickly as possible to look at a spot of light that appears on a screen . Some of the time your eyes will start to move about 100 milliseconds after the light appears . But on other attempts , your eyes will not start moving until 300 milliseconds after the light came on . What causes this variability ? To find out , Hauser et al . recorded from neurons in monkeys trained to perform such a task . When the spot of light appeared many different neurons were active , suggesting there is conflict between the plan that would move the eyes toward the target and plans to look at other locations . That is , when the target appears , the monkey is already thinking of looking somewhere . The time required to resolve this conflict depends on how far apart the target and the competing locations are from one another , and on how much the competing neurons have increased their activity before the target appears . Similar mechanisms are likely to operate when we sit at the dinner table and look for the salt shaker , for example , and so the results presented by Hauser et al . will help us to understand how we direct our attention to different points in space . Understanding how these processes work in more detail will help us to discern what happens when they go wrong , as occurs in attention deficit disorders like ADHD .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "neuroscience" ]	2018	Motor selection dynamics in FEF explain the reaction time variance of saccades to single targets
The control of cellular growth is central to multicellular patterning . In plants , the encapsulating cell wall literally binds neighbouring cells to each other and limits cellular sliding/migration . In contrast to its developmental importance , growth regulation is poorly understood in plants . Here , we reveal that the phytohormone auxin impacts on the shape of the biggest plant organelle , the vacuole . TIR1/AFBs-dependent auxin signalling posttranslationally controls the protein abundance of vacuolar SNARE components . Genetic and pharmacological interference with the auxin effect on vacuolar SNAREs interrelates with auxin-resistant vacuolar morphogenesis and cell size regulation . Vacuolar SNARE VTI11 is strictly required for auxin-reliant vacuolar morphogenesis and loss of function renders cells largely insensitive to auxin-dependent growth inhibition . Our data suggests that the adaptation of SNARE-dependent vacuolar morphogenesis allows auxin to limit cellular expansion , contributing to root organ growth rates . Symplastic growth , characterised by cells that do not alter their relative position to each other , is typical in plant tissue expansion ( Priestley , 1930; Erickson , 1986 ) . Such development implies supra-cellular ( above the level of single cells ) regulation , which has an enormous impact on cellular growth control for plant patterning . Despite their importance , molecular mechanisms that restrict cellular and tissue growth are poorly understood in plants . The phytohormone auxin is a crucial growth regulator and central in differential growth processes . TRANSPORT INHIBITOR RESISTANT1 ( TIR1 ) and its homologs AUXIN F-BOX PROTEINS ( AFBs ) have been unequivocally demonstrated to be auxin receptors ( Kepinski and Leyser , 2005; Dharmasiri et al . , 2005a ) . Genomic auxin responses are initiated by auxin binding to TIR1/AFBs , promoting its interaction with proteins of the AUXIN/INDOLE ACETIC ACID ( Aux/IAA ) family . Auxin-dependent formation of such a co-receptor pair triggers the ubiquitination and subsequent degradation of Aux/IAA proteins . In the absence of auxin , Aux/IAAs form inhibitory heterodimers with AUXIN RESPONSE FACTOR ( ARF ) family transcription factors . Thus , auxin-dependent Aux/IAA degradation leads to the release of ARF transcription factors and subsequent transcriptional responses ( for reviews , see Quint and Gray , 2006; Sauer et al . , 2013 ) . Intriguingly , auxin-signalling events promote and inhibit cellular growth in a cell-type- and auxin concentration-dependent manner . Physiological auxin levels induce growth in light grown aerial tissues , while most root tissues show growth repression in response to the very same auxin concentrations ( Sauer et al . , 2013 ) . The ‘acid growth’ theory proposes that auxin causes extracellular acidifications and subsequent cell wall remodelling , ultimately driving turgor-dependent cellular expansion ( Sauer and Kleine-Vehn , 2011 ) . This theory is based on tissues showing auxin-dependent growth induction . In contrast , relatively little is known of how auxin inhibits cellular growth in other tissues . The higher plant vacuole is , due to its size , the most prominent plant organelle , and shares its lytic function with its counterparts in yeast and animal lysosomes ( Marty , 1999 ) . It may likewise be hypothesised that multifunctional plant vacuoles also contribute to cellular size regulation , as the volume of vacuoles correlates with individual cell size in plant cell cultures ( Owens and Poole , 1979 ) . The root epidermis is a suitable model to study such processes ( Löfke et al . , 2013 ) , because it is regularly spaced into shorter tricho- and longer atrichoblast cell files in the late meristematic zone ( Berger et al . , 1998 ) , which intriguingly , show smaller and larger vacuolar structures , respectively ( Berger et al . , 1998; Löfke et al . , 2013 ) . However , the functional importance of this correlation remains to be addressed . Here we use this cell biological model system to reveal and subsequently characterise auxin-dependent vacuolar morphogenesis and its requirement for limiting cellular growth . Epidermal cells show smaller and larger vacuolar structures in shorter tricho- and longer atrichoblast cells , respectively ( Berger et al . , 1998; Löfke et al . , 2013 ) ( Figure 1A ) . We hypothesised that if the vacuolar morphology contributes to cellular size , the growth regulator auxin may impact on its regulation . 10 . 7554/eLife . 05868 . 003Figure 1 . Auxin triggered changes in vacuolar morphology correlate with its effect on cell size . ( A–C ) Seedlings treated with the solvent DMSO ( A ) , auxin ( B ) ( NAA 250 nM; 20 hr ) or auxin biosynthesis inhibitor kynurenin ( C ) ( Kyn ) ( 2 µM; 20 hr ) . Tonoplast localised VAMP711-YFP ( orange ) as vacuolar marker and propidium iodide stain ( green ) for decorating the cell wall were used for confocal imaging of tricho-/atrichoblast ( T/A ) cells ( A–C ) . ( D ) Vacuolar morphology ( vac . morph . [µm2] ) index of tricho/atrichoblast cells after pharmacological manipulation of auxin levels . ( E–G ) Vacuolar morphology of estradiol ( 10 µM; 20 hr ) induced YUCCA6 overexpression ( YUC6ox ) ( F ) and the respective empty vector control ( pER8 ) ( E ) . Tonoplast membrane stain MDY-64 ( orange ) was used for confocal imaging . ( G ) Vacuolar morphology ( vac . morph . [µm2] ) index after genetic manipulation of auxin levels . ( H and I ) Quantification of cell length change in tricho-/atrichoblast ( T/A ) cells following pharmacological ( H ) or genetic manipulation of auxin levels ( I ) . For statistical analysis , treated cells were compared to untreated tricho-/atrichoblast . n = 40 cells in 10 individual seedlings for cell length measurement and n = 40 cells in eight individual seedlings for vacuolar morphology index quantification . Error bars represent s . e . m . Student's t-test p-values: p < 0 . 05 , p < 0 . 01 , *p < 0 . 001 . Scale bar: 15 µm ( A–C , F , G ) . DOI: http://dx . doi . org/10 . 7554/eLife . 05868 . 00310 . 7554/eLife . 05868 . 004Figure 1—figure supplement 1 . Quantification of epidermal cell length and vacuolar morphology . ( A ) From the onset of pronounced elongation ( atrichoblast cell which is 2 . 5 times longer than wide ) , eight atrichoblast cells ( depicted in yellow numbers ) were counted towards the root tip . At this position , the length of four tricho-/atrichoblast cells ( depicted in white numbers ) were measured and averaged . Propidium-iodide-stained cell walls ( orange ) . ( B ) The same positional information was used for quantifying the vacuolar morphology . Longest and widest distance was measured in the largest depicted vacuolar structure in four tricho-/atrichoblast cells per root . Vacuolar morphology was depicted by multiplying the distances ( C ) or by dividing the cell length by width ( D ) . In this manuscript we mainly used ( C ) as vacuolar morphology index ( vac . morph . index ) . ( E ) Three dimensional representation ( orthogonal sectioning ) of VAMP711-YFP expressing root epidermis; cross-hair depicts the region of optical sections used for all figures . T refers to trichoblast and A to atrichoblast cell file . Scale bar in ( A ) 50 µm; in ( B ) 10 µm; in ( E ) 20 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 05868 . 00410 . 7554/eLife . 05868 . 005Figure 1—figure supplement 2 . Auxin does not affect vacuolar morphology of epidermal root cells in the differentiation zone . DMSO ( A and C ) or NAA ( B and D ) ( 250 nM , 20 hr ) -treated pUBQ10::VAMP711-YFP ( orange ) expressing seedlings imaged at the onset of root hair bulging ( differentiation zone ) . Propidium-iodide ( PI ) -stained cell walls ( green ) . ( C and D ) Overlay of YFP and PI . Scale bar: 50 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 05868 . 005 To specifically investigate the role of auxin in limiting cellular size in roots , we initially assessed how auxin impacts on late meristematic epidermal cells . To allow cellular development under high auxin conditions , we exogenously applied synthetic auxin , 1-naphtylacetic acid ( NAA ) , in nanomolar ranges for 20 hr and screened for subcellular effects that showed differential regulation in neighbouring cells . Remarkably , exogenous application of auxin ( NAA [250 nM] ) or endogenous elevation of YUCCA-dependent auxin biosynthesis led to a dramatic change in vacuolar appearance in root epidermal cells , which consequently displayed smaller luminal vacuolar structures ( Figure 1A , B , E , F ) . We established a vacuolar morphology index in epidermal cells based on the biggest luminal structure to further evaluate the apparent auxin effect on vacuolar shape ( Figure 1—figure supplement 1 ) . This analysis revealed that high auxin conditions affect vacuolar structures , particularly in atrichoblasts ( Figure 1D , G ) . Adversely , pharmacological depletion of auxin caused visibly larger vacuolar structures in both cell types , but was more pronounced in trichoblasts ( relative to the untreated control ) ( Figure 1A , C , D ) . This data shows that auxin differentially affects vacuolar shape in neighbouring epidermal cells . Notably , the differential effect of auxin on vacuoles correlated exactly with a differential effect on cellular size . High auxin levels reduced the cell length of atrichoblast cells ( Figure 1A , B , E , F , H , I ) , whereas cell lengths of the smaller trichoblasts were not significantly affected ( Figure 1A , B , E , F , H , I ) . Conversely , pharmacological reduction in auxin biosynthesis mainly increased the cell length in trichoblasts ( Figure 1A , C , H ) . Our data shows that the auxin effect on vacuolar morphology correlates with its negative effect on late meristematic epidermal cell size . Auxin treatments manifestly did not reverse vacuolar morphology of fully elongated epidermal root cells in the differentiation zone ( Figure 1—figure supplement 2 ) . This finding suggests that auxin mainly shapes vacuoles in growth competent cells . Next we investigated the auxin effect on vacuoles in the course of time . As the auxin effect on vacuoles was most pronounced in atrichoblasts , we focused our analysis ( from here onwards ) mainly on this cell-type . Notably , auxin imposed in time steadily increasing effects on vacuolar appearance ( Figure 2—figure supplement 1 ) . Auxin induced detectable changes in vacuolar morphology already after 15–30 min ( Figure 2A ) . On the other hand the auxin effect on late meristematic cell size was slightly later starting to be significantly affected around 45 min ( Figure 2B ) . 10 . 7554/eLife . 05868 . 006Figure 2 . Auxin effect on vacuoles precedes cell size regulation . ( A and B ) Time course imaging of 250 nM NAA treated seedlings were performed every 15 min . Image acquisition took 10 min per time point . Graphs depict vacuolar morphology index ( A ) and cell length of atrichoblasts ( B ) . Untreated seedlings were imaged before and after recording the auxin treated samples and resulting average was defined as T0 . Error bars represent s . e . m . For statistical analysis DMSO and NAA treatments were compared . n = 50 cells in 10 individual seedlings for each time point . Student's t-test p-values: p < 0 . 05 p < 0 . 01 , p < 0 . 001 . DOI: http://dx . doi . org/10 . 7554/eLife . 05868 . 00610 . 7554/eLife . 05868 . 007Figure 2—figure supplement 1 . Auxin effect on vacuolar morphology increases in time . Auxin ( 250 nM NAA ) induced changes in vacuolar morphology over time . Error bars represent s . e . m . For statistical analysis DMSO and NAA treatments were compared . n = 50 cells in 10 individual seedlings for each time point . Student´s t-test p-values: p < 0 . 001 . DOI: http://dx . doi . org/10 . 7554/eLife . 05868 . 007 Based on our time course experiments we conclude that the auxin effect on vacuolar morphology precedes the auxin impact on late meristematic cell size . We subsequently further characterised the unprecedented role of auxin in vacuolar morphogenesis . The auxin effect on vacuoles was in the time frame of fast transcriptional responses and , subsequently , we tested whether the TRANSPORT INHIBITOR RESISTANT1 ( TIR1 ) /AUXIN F-BOX PROTEINS ( AFB ) auxin receptors ( Leyser , 2006; Mockaitis and Estelle , 2008 ) are required for the auxin effect on vacuoles . It has been suggested that auxin analogue 5-F-IAA preferentially triggers genomic auxin responses via TIR1/AFBs ( Robert et al . , 2010 ) and it indeed induced small luminal vacuoles ( Figure 3A , B , E ) . Correspondingly , auxinole , a designated inhibitor of TIR1/AFBs auxin receptors ( Hayashi et al . , 2012 ) , blocked the auxin effect on vacuolar morphology ( Figure 3A , C , D , E ) , and , the genetic reduction of TIR1/AFBs functions in tir1 afb1 afb3 triple mutants prompted partial resistance to the auxin-induced changes in vacuolar appearance ( Figure 3F–J ) . 10 . 7554/eLife . 05868 . 008Figure 3 . Auxin affects vacuolar morphology in a TIR1/AFBs-dependent manner . ( A–D ) Seedlings treated with DMSO ( A ) , auxin analogue 5-F-IAA ( B ) ( 250 nM; 20 hr ) , TIR1/AFBs antagonist auxinole ( C ) ( 20 µM; 20 hr ) and concomitant with NAA and auxinole ( D ) . Tonoplast localised VAMP711-YFP ( orange ) as vacuolar marker and propidium iodide ( green ) for decorating the cell wall was used for confocal imaging ( A–D ) . ( E ) Vacuolar morphology ( vac . morph . [µm2] ) index of treatments used in A–D . For statistical analysis DMSO and treatments were compared . ( F–I ) DMSO ( F ) or NAA ( G ) ( 250 nM; 20 hr ) treated control seedlings compared to tir/afb1/afb3 triple mutants treated with DMSO ( H ) or NAA ( I ) ( 250 nM; 20 hr ) . Tonoplast localised VAMP711-RFP ( orange ) as vacuolar marker was used for confocal imaging in F–I . ( J ) Vacuolar morphology ( vac . morph . [µm2] ) index of treatments shown in F–I . For statistical analysis either DMSO or NAA treatments were compared between control and indicated mutant . n= 40 cells in eight individual seedlings . Error bars represent s . e . m . Student's t-test P-values: p < 0 . 001 . Scale bar: 15 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 05868 . 008 This set of data indicates that TIR1/AFBs-dependent auxin signalling is required for the auxin effect on vacuolar morphogenesis . In the following we got interested in SNAP ( Soluble NSF Attachment Protein ) Receptor ( SNARE ) complexes at the vacuole . Proximity of adjacent membrane allows the interaction of v ( vesicle ) - and t ( target ) -SNAREs to form a complex , allowing the fusion of vesicles to specific target membranes . SNAREs are essential for eukaryotic vesicle trafficking and according to structural features SNAREs are divided in R ( arginine ) - and Q ( glutamine ) -SNAREs ( Martens and McMahon , 2008 ) . In yeast , the SNARE complex is furthermore central in homotypic vacuolar membrane remodelling and proteomic approaches have identified conserved SNARE complexes at the plant tonoplast ( Carter et al . , 2004 ) . Ergo , we tested whether auxin affects vacuolar SNAREs in Arabidopsis . Remarkably , increased auxin biosynthesis or exogenous application of auxin increased the fluorescence intensity of tonoplast localised SNAREs , such as VAMP711-YFP , SYP21-YFP and SYP22-GFP ( Figure 4A–L , N , O ) . Auxin severely impacts on vacuolar appearance and , hence , it could be that membrane crowding induces higher fluorescence . To address this question we performed co-localisation of VAMP711-RFP/YFP and membrane dyes , such as FM4-64 and MDY-64 . Notably , VAMP711-RFP/YFP , but not the membrane dyes showed auxin-induced signal intensities , suggesting that the auxin effect on vacuolar SNAREs does not rely on membrane crowding ( Figure 4—figure supplement 1 ) . 10 . 7554/eLife . 05868 . 009Figure 4 . Auxin posttranslationally stabilises tonoplast localised SNAREs . ( A–L ) Tonoplast localised SNAREs and marker protein NET4A under high auxin conditions . pUBQ10::VAMP711-YFP ( A and B ) , 35S::SYP21-YFP ( C and D ) , SYP22::SYP22-GFP ( in syp22 ) ( E and F ) and NET4A::NET4A-GFP ( G and H ) expressing seedlings treated with DMSO ( A , C , E , G ) or NAA ( B , D , F , H ) ( 500 nM; 20 hr ) . ( I and J ) Estradiol does not affect VAMP711-YFP abundance . pUBQ10::VAMP711-YFP expressing seedlings treated with DMSO ( I ) or estradiol ( J ) ( 10 µM estradiol; 20 hr ) . YUCCA6 expression under control of an estradiol inducible promoter in pUBQ10::VAMP711-YFP expressing seedlings after DMSO ( K ) or estradiol ( L ) ( 10 µM; 20 hr ) treatment . Propidium iodide ( green ) for decorating the cell wall was used for confocal imaging of atrichoblast cells . ( M ) Western-blot ( anti-GFP ) representing VAMP711-YFP , SYP21-YFP , SYP22-GFP and NET4A-GFP protein abundance after NAA ( 500 nM; 20 hr ) and control treatment as well as corresponding alpha-tubulin abundance for normalization . ( N ) Mean grey value of vacuolar localised SNAREs and marker protein NET4A after auxin treatments ( 500 nM; 20 hr ) compared to DMSO treatments . ( O ) Mean grey value of VAMP711-YFP after YUCCA6 induction ( 10 µM; 20 hr ) . ( P ) Western-blot quantification ( mean grey values ) . n = 3 biological replicates each consisting of a pool of 40–50 roots . Error bars represent s . e . m . For statistical analysis DMSO and NAA treatments were compared . For confocal analysis ( N-O ) : n = 32 cells in eight individual seedlings . Student's t-test p-values: p < 0 . 001 . Scale bar: 15 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 05868 . 00910 . 7554/eLife . 05868 . 010Figure 4—figure supplement 1 . Increase in SNARE intensity is independent of membrane crowding . ( A–J ) Simultaneous imaging of VAMP711-RFP/YFP and tonoplast staining dyes under untreated and high auxin conditions . pUBQ10::VAMP711-RFP ( A and B ) and pUBQ10::VAMP711-YFP ( F and G ) expressing seedlings were counterstained either with MDY-64 ( C and D ) or FM4-64 ( H and I ) and treated with DMSO ( A and C ) or NAA ( B and D ) ( 500 nM; 20 hr ) . ( E ) Quantification of mean grey value of VAMP711-RFP and MDY-64 after NAA treatments ( 500 nM ) compared to DMSO control . ( J ) Quantification of mean grey value of VAMP711-YFP and FM4-64 after NAA treatments ( 500 nM ) compared to DMSO control . Error bars represent s . e . m . For statistical analysis DMSO and NAA treatments were compared . n = 40 cells in 10 individual seedlings . Student's t-test p-values: p < 0 . 001 . Scale bar: 15 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 05868 . 01010 . 7554/eLife . 05868 . 011Figure 4—figure supplement 2 . Auxin affects cellular abundance of vacuolar SNAREs . ( A–H ) Maximum projection of tonoplast localised SNAREs and marker protein NET4A under high auxin conditions . pUBQ10::VAMP711-YFP expressing seedlings treated with DMSO ( A ) or NAA ( B ) ( 500 nM; 20 hr ) . 35S::SYP21-YFP expressing seedlings treated with DMSO ( C ) or NAA ( D ) ( 500 nM; 20 hr ) . SYP22::SYP22-GFP ( in syp22 ) expressing seedlings treated with DMSO ( E ) or NAA ( F ) ( 500 nM; 20 hr ) . NET4A::NET4A-GFP expressing seedlings treated with DMSO ( G ) or NAA ( H ) ( 500 nM; 20 hr ) . 17–20 steps of 1 µm size were used for image acquisition . ( I ) Mean grey value of vacuolar SNAREs and tonoplast marker NET4A in DMSO control and auxin treatments ( 500 nM; 20 hr ) . Error bars represent s . e . m . For statistical analysis DMSO and NAA treatments were compared . Student's t-test p-values: p < 0 . 01 , p < 0 . 001 . Scale bar: 15 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 05868 . 01110 . 7554/eLife . 05868 . 012Figure 4—figure supplement 3 . Induction of a single SNARE component is not sufficient to affect vacuolar morphology . ( A–D ) Estradiol induction of VAMP711 . pMDC7::citrus-flag ( control ) after DMSO ( A ) or estradiol ( B ) ( 2 µM ) treatment . VAMP711 expression under control of an estradiol inducible promoter after DMSO ( C ) or estradiol ( D ) ( 2 µM ) incubation . Vacuoles were decorated with FM4-64 ( orange ) . ( E ) Semi-quantitative RT-PCR of VAMP711 in pMDC7::citrus-flag and pMDC7::VAMP711 expressing plants with ( + ) and without ( − ) estradiol induction ( 2 µM; 20 hr ) . UBQ-5 expression was used for normalisation . Depicted values represent quantification of mean grey values of corresponding bands . ( F ) Vacuolar morphology ( vac . morph . [µm2] ) index of estradiol induced pMDC7::citrus-flag and pMDC7::VAMP711 expressing seedlings compared to the uninduced control . Scale bar , 15 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 05868 . 012 Exogenous application of auxin does not detectably impact on cytosolic pH ( Gjetting et al . , 2012 ) . However , to fully preclude that pH sensitivity may affect fluorescence of VAMP711-YFP ( YFP faces the cytosol ) , we also utilized more pH resistant RFP fusions . The auxin effect on VAMP711 was detectable in both pH sensitive YFP and pH resistant RFP fusions ( Figure 4—figure supplement 1A , B; Figure 5A , B ) , suggesting that the auxin effect on SNAREs does not indirectly rely on cytosolic pH . 10 . 7554/eLife . 05868 . 013Figure 5 . TIR1/AFBs-dependent auxin perception is required for posttranslational VAMP711 stabilisation . ( A–D ) Pharmacological inhibition of TIR1/AFBs dependent signalling . pUBQ10::VAMP711-YFP expressing seedlings treated with DMSO ( A ) , NAA ( B ) ( 500 nM; 20 hr ) , auxinole ( C ) ( 20 µM; 20 hr ) or auxinole/NAA co-treatment ( D ) . ( E ) Mean grey value of VAMP711-YFP abundance after NAA or NAA/auxinole co-treatments . ( F–I ) Genetic inhibition of TIR1/AFBs signalling . VAMP711-RFP expressing control seedlings ( for H and I ) treated with DMSO ( F ) and NAA ( G ) ( 500 nM; 20 hr ) . VAMP711-RFP abundance in tir1-1/afb1-3/afb3-4 mutant background after DMSO ( H ) or NAA ( I ) ( 500 nM; 20 hr ) treatment . ( J ) Mean grey value of treatments in F–I . VAMP711-YFP/RFP ( orange ) as a vacuolar marker and propidium iodide ( green ) for decorating the cell wall were used for confocal imaging of atrichoblast cells . n = 32 cells in eight individual seedlings . Error bars represent s . e . m . For statistical analysis either DMSO or NAA treatments were compared between control and indicated mutant/treated seedlings . Student's t-test p-values: p < 0 . 001 . Scale bar: 15 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 05868 . 013 To assess whether auxin treatments affect the overall cellular abundance of vacuolar SNAREs , we performed defined z-stack imaging . Subsequent maximum projections and intensity measurements confirmed that auxin increases cellular SNARE abundance at the tonoplast ( Figure 4—figure supplement 2A–I ) . To further emphasize on this finding , we also performed western blots on excised root tips , similarly confirming our conclusion that auxin increases vacuolar SNARE abundance ( Figure 4M , P ) . Notably , tonoplast marker NET4A-GFP ( Deeks et al . , 2012 ) did not show auxin-induced stabilization ( Figure 4G , H; Figure 4—figure supplement 2G , H ) , suggesting certain specificity for the auxin effect on vacuolar SNAREs . This set of data indicates that auxin affects vacuolar SNARE function . Induction of a single SNARE component , such as VAMP711 , did not affect vacuolar morphogenesis ( Figure 4—figure supplement 3 ) , possibly indicating the joint requirement of several complex components . Furthermore , auxin modulated SNARE proteins also when expressed under constitutive promoters ( Figure 4A–D , I–N ) . This data implies that auxin affects the vacuolar SNAREs posttranslationally . Next we tested whether TIR1/AFBs-dependent auxin perception mechanisms are required for the auxin effect on vacuolar SNAREs . Concomitant treatments with auxin and the TIR1/AFBs antagonist auxinole comprehensively interfered with auxin-induced stabilisation of VAMP711-YFP ( Figure 5A–E ) . Concurrently , the auxin effect on vacuolar SNAREs was also significantly reduced in the tir1 afb1 afb3 triple mutant ( Figure 5F–J ) . Hence , pharmacologic and genetic interference with TIR1/AFBs did not only inhibit the auxin effect on vacuoles , but also abolished the posttranslational effect of auxin on VAMP711 . We conclude that the TIR1/AFBs-dependent auxin signalling triggers higher SNARE abundance at the tonoplast . It has been suggested that several vacuolar SNARE components act redundantly ( Yano et al . , 2003; Uemura et al . , 2010 ) and also in our conditions most analysed SNARE single mutants displayed vacuolar morphology reminiscent to wild type ( Figure 6—figure supplement 1 ) . In contrast , vti11 mutant alleles display roundish vacuoles in untreated conditions ( Yano et al . , 2003; Zheng et al . , 2014 ) ( Figure 6A , C ) . Despite these apparent defects , vacuoles remained differentially controlled in vti11 mutant tricho- and atrichoblast cells ( Figure 6—figure supplement 2 ) , indicating that the cell type-dependent regulation of vacuolar morphology is at least partially operational in vti11 mutants . 10 . 7554/eLife . 05868 . 014Figure 6 . SNARE-dependent vacuolar morphogenesis is required for auxin regulated cell size determination . ( A–F ) Vacuolar morphology and cell size determination in the vti11 mutant . Control treatment of Col-0 with DMSO ( A ) or NAA ( B ) ( 250 nM; 20 hr ) . vti11 mutants were treated with DMSO ( C ) or NAA ( D ) ( 250 nM; 20 hr ) . MDY-64 ( orange ) and propidium iodide ( green ) were used for confocal imaging of atrichoblast cells ( A–D ) . ( E ) Vacuolar morphology ( vac . morph . [µm2] ) index of prior treatments in Col-0 and vti11 mutant . ( F ) Cell length change of Col-0 and vti11 atrichoblast cells after NAA ( 250 nM; 20 hr ) treatment compared to DMSO control . n = 40 cells in eight individual seedlings for vacuolar morphology index quantification and n = 40 cells in 10 individual seedlings for cell size measurements . Error bars represent s . e . m . For statistical analysis either DMSO or NAA treatments were compared between control and indicated mutant/treated seedlings . Student's t-test p-values: p < 0 . 01 , p < 0 . 001 . Scale bar: 15 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 05868 . 01410 . 7554/eLife . 05868 . 015Figure 6—figure supplement 1 . Auxin affects vacuolar morphology in several vacuolar snare single mutants . ( A–H ) Overview of the vacuolar morphology in single snare mutants after auxin treatment . Control treatment of Col-0 with DMSO ( A ) or NAA ( B ) ( 250 nM; 20 hr ) . syp21 treated with DMSO ( C ) or NAA ( D ) ( 250 nM; 20 hr ) . syp22 treated with DMSO ( E ) or NAA ( F ) ( 250 nM; 20 hr ) and vamp711 treated with DMSO ( G ) or NAA ( H ) ( 250 nM; 20 hr ) . Vacuoles were decorated with MDY-64 ( orange ) and propidium iodide ( PI ) was used for counterstaining the cell walls ( green ) . ( I ) Vacuolar morphology ( vac . morph . [µm2] ) index of DMSO ( control ) and NAA ( 250 nM; 20 hr ) treated Col-0 and mutant atrichoblast cells . ( J ) Semi-quantitative RT-PCR of VAMP711 in Col-0 and vamp711 mutant line showing knock down of VAMP711 transcripts . UBQ-5 expression was used for normalisation . Depicted values represent quantification of mean grey values of corresponding bands . n = 40 atrichoblast cells in eight individual roots for vacuolar morphology index determination . Error bars represent s . e . m . Student's t-test p-value: p < 0 . 001 . Scale bar: 15 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 05868 . 01510 . 7554/eLife . 05868 . 016Figure 6—figure supplement 2 . The vacuolar morphology in the vti11 mutant remained differentially controlled in tricho-/atrichoblast root epidermal cells . ( A and B ) Root epidermal tricho- and atrichoblast cells of Col-0 ( A ) or vti11 mutant ( B ) were decorated with MDY-64 . ( C ) Vacuolar morphology ( vac . morph . [µm2] ) index of Col-0 and vti11 tricho- and atrichoblast cells . ( D ) Mean cell length of Col-0 and vti11 tricho- and atrichoblast cells . For statistical analysis vti11 tricho- and atrichoblast cells were compared to wild-type tricho- and atrichoblast cells . T refers to trichoblast and A to atrichoblast cell files . n = 40 quantified cells in eight seedlings for vacuolar morphology index measurements and n = 40 quantified cells in 10 seedlings for cell length quantification . Error bars represent s . e . m . Student's t-test p-value: p < 0 . 001 . Scale bar: 15 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 05868 . 01610 . 7554/eLife . 05868 . 017Figure 6—figure supplement 3 . The pVTI11:VTI11-GFP complements auxin phenotypes in vti11 mutant . Control treatment of Col-0 ( A and B ) and vti11 ( C and D ) with DMSO ( A and C ) or NAA ( B and D ) ( 250 nM; 20 hr ) . ( E–J ) VTI11::GFP-VTI11 expressing vti11 seedlings under control and high auxin conditions . ( E ) DMSO control treatment of VTI11-GFP expressing vti11 cells . ( H ) NAA treatment ( 250 nM; 20 hr ) of VTI11-GFP expressing vti11 cells . Samples were counterstained with FM4-64 ( F and I ) and respective overlay ( G and J ) is shown . ( K ) Vacuolar morphology index of DMSO control and NAA ( 250 nM; 20 hr ) treated Col-0 and mutant atrichoblast cells . ( L and M ) Complementation of vti11 . ( L ) Root growth of Col-0 , vti11 and VTI11::GFP-VTI11 in vti11 after control and NAA ( 125 nM ) treatment . ( M ) Quantification of root growth inhibition in Col-0 , vti11 and VTI11::GFP-VTI11 in vti11 after NAA ( 125 nM ) treatment . n = 40 atrichoblast cells in eight individual roots for vacuolar morphology index determination and n = 20–25 roots for root growth inhibition . For statistical analysis Col-0 were compared with mutant seedlings . Error bars represent s . e . m . Student's t-test p-value: p < 0 . 001 . Scale bar , in A–J: 15 µm; in L: 1 cm . DOI: http://dx . doi . org/10 . 7554/eLife . 05868 . 017 We , hence , have chosen vti11 mutants for further investigation and tested if VTI11 function is required for the auxin effect on vacuoles . Auxin treatments were less effective to modulate vacuolar morphology in vti11 mutants ( Figure 6A–E ) . Notably , pVTI11:VTI11-GFP expression in vti11 mutant cells induced reversion to auxin sensitive vacuolar morphology ( Figure 6—figure supplement 3 ) . This data indicates that auxin does not only affect SNARE abundance , but requires functional Q-SNARE VTI11 to modulate vacuolar shapes . Vacuoles partially escaped auxin regulation in vti11 mutants , allowing us to assess the requirement of VTI11 function for auxin-dependent limitation of meristematic cell size . Interestingly , vti11 mutants were not only partially resistant to the auxin effect on vacuoles , but in addition , less sensitive to the negative impact of auxin on late meristematic cell size ( Figure 6A–D , F ) . This data suggests that VTI11 function is required for the auxin effect on vacuolar shape and late meristematic cell size . Several phosphatidylinositol ( PI ) -dependent processes have been previously shown to play a role in vacuolar biogenesis in yeast ( Mayer et al . , 2000 ) and also impact on vacuolar morphology in plants ( Nováková et al . , 2014; Zheng et al . , 2014 ) . PI3/4 kinase inhibitor Wortmannin ( WM ) affects vacuolar morphology and has been recently proffered as affecting processes upstream of vacuolar SNAREs in plants ( Feraru et al . , 2010; Zheng et al . , 2014 ) . WM treatments led to larger luminal vacuoles and abolished the auxin effect on vacuoles ( Figure 7A–E ) . It may be noted that the negative effect of auxin on limiting late meristematic cell size was also abolished after low doses of WM [2 µM] ( Figure 7F ) . This data suggests that WM sensitive processes may contribute to auxin-dependent vacuolar morphogenesis and cell size regulation . 10 . 7554/eLife . 05868 . 018Figure 7 . PI4-kinase function is required for auxin dependent vacuolar morphology , cell size determination and control of posttranslational VAMP711 abundance . ( A–G ) Effect of wortmannin ( WM ) on auxin regulated vacuolar morphology , cell growth inhibition and VAMP711 abundance . Control treatment of VAMP711-YFP atrichoblasts with DMSO ( A ) or NAA ( B ) ( 250 nM; 20 hr ) . VAMP711-YFP expressing seedlings after WM ( C ) ( 10 µM; 20 hr ) or NAA/WM co-treatment ( D ) . Quantification of vacuolar morphology ( vac . morph . [µm2] ) index ( E ) and cell length change ( F ) . ( G ) Relative mean grey value of VAMP711-YFP abundance after NAA ( 500 nM; 20 hr ) and/or WM ( 10 µM; 20 hr ) treatment . Corresponding images are shown in Figure 7—figure supplement 1 . ( H–M ) Effect on auxin regulated vacuolar morphology , cell growth inhibition and VAMP711 abundance in pi4kß1/2 plants . Control treatment of VAMP711-YFP atrichoblasts with DMSO ( H ) or NAA ( I ) ( 100 nM; 20 hr ) for comparability . VAMP711-YFP expression in pi4kß1/2 mutant background after DMSO ( J ) or NAA ( K ) ( 100 nM; 20 hr ) treatment . Quantification of vacuolar morphology ( vac . morph . [µm2] ) index ( L ) and cell length change ( M ) . ( N ) Absolute mean grey value of VAMP711-YFP abundance after NAA ( 500 nM; 20 hr ) treatment in the pi4kß1/2 mutant background . Corresponding images are shown in Figure 7—figure supplement 1 . VAMP711-YFP ( orange ) as a vacuolar marker and propidium iodide ( green ) for decorating the cell wall were used for confocal imaging of atrichoblast cells . n = 32 cells in eight individual seedlings for cell length measurements and n = 40 cells in eight individual seedlings for vacuolar morphology index quantification . Error bars represent s . e . m . For statistical analysis either DMSO or NAA treatments were compared between control and indicated mutant/treated seedlings . Student's t-test p-values: p < 0 . 01 , p < 0 . 001 . Scale bar: 15 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 05868 . 01810 . 7554/eLife . 05868 . 019Figure 7—figure supplement 1 . PI4 kinases posttranslationally control VAMP711 abundance . ( A–D ) Wortmannin inhibits the auxin effect on VAMP711 stabilisation . VAMP711-YFP signal after DMSO ( A ) or NAA ( B ) ( 500 nM; 20 hr ) ( control for C and D ) treatment . VAMP711-YFP seedlings treated with wortmannin ( WM ) ( C ) ( 10 µM; 20 hr ) or WM and NAA ( D ) ( 500 nM; 20 hr ) . ( E–H ) Reduced VAMP711 abundance in pi4kß1/2 mutants . Control treatment of VAMP711-YFP with DMSO ( E ) or NAA ( F ) ( 500 nM; 20 hr ) . pi4kß1/2 seedlings expressing pUBQ10::VAMP711-YFP treated with DMSO ( G ) or NAA ( H ) ( 500 nM; 20 hr ) . VAMP711-YFP ( orange ) as a vacuolar marker and propidium iodide ( green ) for decorating the cell wall were used for confocal imaging of atrichoblast cells . n = 32 cells in eight individual seedlings . Error bars represent s . e . m . For statistical analysis either DMSO or NAA treatments were compared between control and indicated mutant/treated seedlings . Student's t-test p-values: p < 0 . 001 . Scale bar: 15 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 05868 . 01910 . 7554/eLife . 05868 . 020Figure 7—figure supplement 2 . Tricho-/atrichoblast cell length in wortmannin treated samples . ( A and B ) Root epidermal tricho-/atrichoblast cells of VAMP711-YFP expressing seedlings after DMSO ( A ) or WM ( B ) ( 10 µM; 20 hr ) treatment . ( C ) Mean epidermal cell length in the root meristem of DMSO or WM ( 2 µM; 20 hr ) treated seedlings . n = 40 quantified cells in eight seedlings . Error bars represent s . e . m . Student's t-test p-value: *p < 0 . 001 . Scale bar: 15 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 05868 . 020 To substantiate this pharmacological data , we subsequently screened the relevant literature for WM sensitive molecular components , which may affect root epidermal processes . PI4Kß1 and PI4Kß2 are expressed in root epidermal cells and redundantly control mature root hair morphology ( Preuss et al . , 2006 ) . We therefore tested the auxin effect on vacuolar morphology in pi4kß1/2 double mutants . Compared to wild-type seedlings , pi4kß1/2 double mutants showed partially auxin resistant vacuolar appearance ( Figure 7H–L ) . Moreover , the genetic interference with PI4Kß-function was reminiscent of WM treatments and also abolished the auxin effect on reducing late meristematic cell size ( Figure 7M ) . Based on our pharmacological and genetic data , we conclude that PI-dependent processes affect auxin-dependent vacuolar morphology . We subsequently addressed whether the obstruction of PI homeostasis may impair auxin-dependent posttranslational regulation of SNARE proteins . Concomitant treatments with PI3/4-kinase inhibitor WM and auxin completely diminished the auxin effect on vacuolar SNAREs even at relatively high auxin levels ( NAA [500 nM] ) ( Figure 7G; Figure 7—figure supplement 1A–D ) . This data suggests that WM sensitive processes interfere with both the auxin effect on vacuolar morphology and the auxin-dependent regulation of vacuolar SNARE abundance . Analogously , the pi4kß1/2 double mutants showed reduced VAMP711-YFP abundance in untreated and auxin treated conditions ( Figure 7N; Figure 7—figure supplement 1E–H ) . We accordingly conclude that auxin requires PI4K activity to modulate vacuolar SNAREs . In contrast to WM , SNARE stability was still increased in response to high auxin ( NAA [500 nM] ) in pi4kß1/2 double mutants ( Figure 7N; Figure 7—figure supplement 1E–H ) . Notably , pi4kß1/2 double mutants showed auxin sensitive vacuolar morphogenesis at these higher auxin levels ( Figure 7—figure supplement 1G , H ) . Based on this data , we assume that PI4 kinase activity and possibly other , WM sensitive , PI-dependent processes affect auxin-dependent posttranslational regulation of vacuolar SNARE proteins . We conclude that the PI-dependent interference with the auxin effect on vacuolar SNAREs completely coincides with a reduced impact of auxin on vacuolar morphology and meristematic cell size . Here we show that auxin limits cellular vacuolisation , correlating with its negative impacts on meristematic cell size . Prolonged auxin treatments shift the cell length ratio of tricho- and atrichoblasts . Assuming that there is no cellular sliding , the auxin effect on meristematic cell size might not only depend on cellular expansion , but also on altered division rates . To exclude cell division and to assess whether the auxin effect on vacuolar morphology could limit cellular growth , we subsequently concentrated solely on cellular expansion of epidermal cells in the elongation zone and recorded their maximum expansion under high auxin conditions . As expected , auxin repressed cellular elongation in wild-type epidermal cells ( Figure 8A , B ) . In contrast , pharmacological inhibition of PI3 and PI4 kinases led to reduced sensitivity to auxin-dependent inhibition of epidermal growth ( Figure 8C , D , I ) . pi4kß1/2 double mutants were similarly resistant to the auxin-dependent inhibition of cellular expansion ( Figure 8E , F , J ) . 10 . 7554/eLife . 05868 . 021Figure 8 . Auxin dependent vacuolar morphogenesis links auxin dependent growth inhibition . ( A–H ) Fully elongated root cells in the differentiation zone of Col-0 after DMSO ( A ) , NAA ( B ) ( 250 nM ) , WM ( C ) ( 2 µM ) , WM+NAA ( D ) and pi4kß1/2 after DMSO ( E ) or NAA ( F ) ( 250 nM ) , as well as vti11 after DMSO ( G ) or NAA ( H ) ( 250 nM ) treatments . ( I ) Cell length change of Col-0 trichoblast cells after NAA ( 250 nM; 20 hr ) and/or WM ( 2 µM; 20 hr ) treatment . ( J ) Cell length change of Col-0 and pi4kß1/2 trichoblast cells after NAA ( 250 nM; 20 hr ) treatment . ( K ) Cell length change of Col-0 and vti11 trichoblast cells after NAA ( 250 nM; 20 hr ) treatment . ( L ) NAA mediated root growth inhibition of Col-0 , pi4kß1/2 and vti11 after NAA ( 125 nM ) and/or WM ( 2 µM ) treatment . ( M ) Quantification of root growth inhibition in Col-0 after NAA ( 125 nM ) and/or WM ( 2 µM ) treatment ( germinated; 12 DAG ) . ( N ) Quantification of root growth inhibition Col-0 and pi4kß1/2 after NAA ( 125 nM ) treatment ( germinated; 8 DAG ) . ( O ) Quantification of root growth inhibition in Col-0 and vti11 after NAA ( 125 nM ) treatment ( germinated; 8 DAG ) . Propidium iodide ( red ) for decorating the cell wall was used for confocal imaging of epidermal cells . For cell length change n = 30 seedlings out of three independent experiments with approximately 90–120 quantified cells in total; and n = 20–25 roots for root growth inhibition . Error bars represent s . e . m . For statistical analysis either NAA or WM treatments were compared between control and/or indicated mutant/treated seedlings . Student's t-test p-values: p < 0 . 05 , *p < 0 . 001 . Scale bar: 50 µm ( A–H ) ; 1 cm ( L ) . DOI: http://dx . doi . org/10 . 7554/eLife . 05868 . 02110 . 7554/eLife . 05868 . 022Figure 8—figure supplement 1 . Root growth of pi4kß1/2 mutants reacts to auxin in a dose dependent manner . ( A–D ) Root growth and length quantification of Col-0 and pi4kß1/2 seedlings under high auxin conditions . ( A ) Root growth of Col-0 seedling according to different NAA concentrations ( 300 nM and 600 nM NAA ) . ( B ) Root growth of pi4kß1/2 seedling according to different NAA concentrations ( 300 nM and 600 nM NAA ) . ( C ) Mean root growth inhibition of Col-0 and pi4kß1/2 in responce to different auxin concentrations ( 300 nM and 600 nM NAA ) . ( D ) Relative root growth inhibition of Col-0 and pi4kß1/2 in response to different auxin concentrations ( 300 nM and 600 nM NAA ) . n = 20–25 roots per treatment . Error bars represent s . e . m . Student's t-test p-value: *p < 0 . 001 . Scale bar: 1 cm . DOI: http://dx . doi . org/10 . 7554/eLife . 05868 . 022 Subsequently , we tested root organ growth in response to auxin . Exogenous application of auxin strongly reduced the root length of wild-type seedlings , but was less effective following pharmacological ( WM treated wild-type ) and genetic interference ( pi4kß1/2 double ) with PI kinases ( Figure 8L–N ) . Notably and comparable with our cellular analysis , increased auxin concentrations led to root growth inhibition also in pi4kß1/2 double mutants ( Figure 8—figure supplement 1A–D ) . In conclusion , this data suggests that PI-dependent processes contribute to auxin-dependent inhibition of cellular and root organ growth . Thereafter we likewise addressed auxin dependent inhibition of cellular expansion in VTI11 deficient roots . In agreement with our data on meristematic cell size control , cellular elongation was also less sensitive to auxin in vit11 mutants ( Figure 8G , H , K ) . The negative effect of auxin on root organ growth was also reduced in vti11 mutants ( Figure 8L , O ) , but was complemented by VTI11-GFP expression ( Figure 6—figure supplement 3 ) . Therefore , we conclude that auxin requires VTI11 function to inhibit cellular expansion and moreover root organ growth . We have demonstrated that auxin interferes with SNARE abundance at the tonoplast , correlating with its effect on vacuolar morphology . What is striking is that auxin-dependent modulation of vacuolar shape precisely coincides with auxin dependent growth inhibition . Pharmacological and genetic interference with the auxin-dependent regulation of SNARE abundance corresponds with auxin resistant vacuolar morphology . Moreover , abrogation of the cellular auxin effect on vacuolar morphology interrelates with lower sensitivity to auxin-dependent inhibition of cellular growth . We propose , based on these independent lines of evidence , that auxin signalling utilises SNARE-dependent vacuolar morphogenesis to restrict cellular expansion . Such regulation could have widespread developmental consequences , such as determining root organ growth rates . It has been noted that the morphology of plant vacuoles correlate with cell size ( Owens and Poole , 1979; Berger et al . , 1998; Löfke et al . , 2013 ) and , hence , it was tempting to postulate that vacuoles may even drive cellular growth ( Marty , 1999 ) . Surprisingly , little is actually known about mechanisms controlling vacuolar morphology and whether the vacuoles are indeed involved in growth regulation . Plant vacuoles , inter alia , are claimed to be important cellular osmoregulators , and accordingly , have been hypothesised as contributing to turgor-dependent cellular growth induction ( Marty , 1999 ) . This may explain the correlation between vacuolarisation with cellular size . Though appealing , also this hypothesis awaits experimental validation . Hence , the potential role of auxin in turgor regulation should also be carefully assessed . Just such a role has been insinuated in recent papers related to lateral root emergence . Auxin responses have been reported to reduce cellular pressure in cell files facing lateral root primordia ( Peret et al . , 2012 ) and vacuolar morphology alterations have been recently described in endodermal tissues during lateral root emergence ( Peret et al . , 2012; Vermeer et al . , 2014 ) . However , these proposed auxin responses are not related to cellular growth regulation and it remains to be seen whether the auxin impact on vacuolar morphology described here is related to interference with turgor pressure . Here we show that auxin controls the overall vacuolar morphology . Several pieces of independent evidence suggest that the modulation of the vacuolar SNAREs allows the phytohormone auxin to limit cellular growth . Future studies will further assess the precise molecular role of vacuolar SNAREs in shaping the plant vacuole . Auxin impacts on vacuolar SNAREs in a posttranslational manner and it is tempting to speculate that auxin could impose a conformational change on the SNARE complex , possibly affecting its stability and activity . Accordingly , the auxin effect on vacuolar SNAREs could impact on homotypic vacuolar fusion events , leading to smaller luminal structures . Alternatively , also a structural role of SNAREs has been envisioned ( Di Sansebastiano , 2013 ) , which could be yet another mechanism to impact on vacuolar organisation . Notably , a recent study suggested that most vacuolar structures are interconnected within an untreated cell ( Viotti et al . , 2013 ) . Hence , it needs to be seen how precisely auxin impacts on vacuolar morphology and whether these structures are still interconnected in high auxin conditions . We assume that auxin utilises the plant vacuole as a cellular effector to restrict cellular expansion . Hence , we conclude that previous correlative observations on cell size and vacuolar morphology ( Owens and Poole , 1979; Berger et al . , 1998; Löfke et al . , 2013 ) may not reflect that vacuoles drive growth , but rather rely on a vacuole-dependent cellular mechanism that limits growth . Our data extends the current view of auxin biology , suggesting that auxin coordinates extracellular and intracellular components , such as cell wall acidification ( Sauer and Kleine-Vehn , 2011; Spartz et al . , 2014 ) and vacuolar morphogenesis , for driving and restricting cellular growth . In this light , the luminal increase of plant vacuoles alone may not be sufficient to induce larger cell sizes ( Figure 6—figure supplement 2; Figure 7—figure supplement 2 ) due to cell wall limitations . In contrast , limiting cellular vacuolarisation appears sufficient to restrict cellular growth . Such a dual growth mechanism would allow plants to dynamically de- and accelerate cellular expansion , integrating multiple distinct , possibly conflictive internal and external triggers . We used Arabidopsis thaliana of ecotype Columbia 0 ( Col-0 ) . vamp711 ( N673991 ) , obtained from the Nottingham Arabidopsis stock centre and the decrease of VAMP711 transcript was shown by RT-PCR . Other plant lines were described previously , pUBQ10::VAMP711-YFP/RFP ( Wave 9Y/R ) ( Geldner et al . , 2009 ) , pER8 and pER8::YUC6 ( Mashiguchi et al . , 2011 ) , tir triple: tir1-1/afb1-3/afb3-4 ( Dharmasiri et al . , 2005b ) , pi4kß1/2 ( Preuss et al . , 2006 ) , 35S::SYP21-YFP ( Robert et al . , 2008 ) , SYP22::SYP22-GFP in syp22 background ( Uemura et al . , 2010 ) , syp21 ( pep12 ) ( Uemura et al . , 2010 ) , syp22 ( vam3-1 ) ( Uemura et al . , 2010 ) , vti11 ( zigzag ) ( Yano et al . , 2003 ) , VTI11::GFP-VTI11 in vti11 ( Niihama et al . , 2005 ) , NET4A::NET4A-GFP ( Deeks et al . , 2012 ) . The pi4kß1/2 and pER8::YUC6 were crossed into Wave 9Y , respectively . tir triple plants expressing Wave 9R were obtained by floral dipping in Agrobacterium tumefaciens liquid cultured cells harbouring the Wave 9R vector and were subsequently selected on Basta ( Bayer , Germany ) supplemented MS plates . Seeds were stratified at 4°C for 2 days in the dark and grown on vertically orientated ½ Murashige and Skoog ( MS ) medium plates under a long-day regime ( 16 hr light/8 hr dark ) at 20–22°C . Gateway cloning was used to construct pMDC7_B ( pUBQ ) ::VAMP711 ( Destination vector from Barbez et al . , 2012 ) . The VAMP711 fragment was amplified with the high-fidelity polymerase ‘I proof’ ( BioRad , CA , USA ) from the Wave line vector harbouring VAMP711 ( Wave 9Y ) . The primers were designed using life technologies OligoPerfect ( www . lifetechnologies . com/oligoperfect ) and are given below . The fragment obtained was introduced into the pDONR221 ( Invitrogen , CA , USA ) . Coding VAMP711 sequence from the entry vector was subsequently introduced in the gateway compatible destination vector pMDC7_B ( pUBQ ) using the Invitrogen LR clonase ( + ) and the resulting construct was transformed into Col-0 plants by floral dipping in Agrobacterium tumefaciens liquid culture . The T1 generation obtained was selected on hygromycin B supplemented MS plates . All chemicals were dissolved in DMSO and were applied in solid ½ MS-medium . Only dyes were applied in liquid ½ MS-medium before imaging . 1-naphthaleneacetic acid ( NAA ) was obtained from Duchefa ( Netherlands ) , 5-F-IAA , estradiol , FM4-64 , L-kynurenine ( Kyn ) and propidium iodide ( PI ) from Sigma ( MO , USA ) , wortmannin ( WM ) from Cayman Chemical ( MI , USA ) , MDY-64 from life technologies ( CA , USA ) and auxinole was kindly provided by Ken-ichiro Hayashi ( Hayashi et al . , 2012 ) . For analysing the vacuolar morphology and cell length change , all the experiments were carried out on 7 days old seedlings . For analysing the vacuolar morphology index , optical confocal sections ( above the cell nucleus ) of the root epidermis were acquired and processed in imageJ . Images were taken in the late meristematic zone ( position shown in Figure 1—figure supplement 1 ) . The largest luminal structure in five tricho- and/or atrichoblast cells were quantified by measuring the longest to widest distance and processed by multiplying the values . Means and standard error were calculated and statistical significance was evaluated by the Student's t-test using graphpad ( http://www . graphpad . com/quickcalcs/ ) . Figures display a representative experiment ( out of three independent repetitions ) utilising eight individual roots . The cell length change in the late meristematic zone ( position shown in Figure 1—figure supplement 1 ) was quantified in the median epidermal confocal section decorated with propidium iodide to visualise the cell wall . Cell length measurements in the late meristematic zone were performed in imageJ by quantifying the length of four tricho- and/or atrichoblast cells which were averaged . Means and standard error were calculated and statistical significance was evaluated by the Student's t-test using graphpad ( http://www . graphpad . com/quickcalcs/ ) . Figures show a representative experiment ( out of three independent repetitions ) utilising 10 individual roots . To estimate the position for the cell length measurements in the elongation zone , seedlings were stained in propidium iodide for 5 min and subsequently imaged choosing the position were no propidium iodide ( 0 . 02 mg/ml ) entered the vasculature , showing fully developed endodermal diffusion barriers . Length measurements of epidermal root hair cells were performed in imageJ by quantifying the length of individual cells which were averaged . Means and standard error were calculated and statistical significance was evaluated by the Student's t-test using graphpad ( http://www . graphpad . com/quickcalcs/ ) . Figures show a representative experiment ( out of three independent repetitions ) utilising 10 individual roots . In each root approximately three cells were quantified . For analysis of the root length , seedlings grown in vertically orientated plates were scanned on a flat-bed scanner and measurements were performed in imageJ . Per condition , 20–25 seedlings were analysed ( for each experiment ) 8 days after germination . Means and standard error were calculated and statistical significance was evaluated by the Student's t-test using graphpad ( http://www . graphpad . com/quickcalcs/ ) . Figures show a representative experiment ( out of three independent repetitions ) . Root samples ( ∼50 mg each ) of 7 days old pUBQ10::VAMP711-YFP expressing seedlings were shock-frozen and homogenised in liquid nitrogen , then 150 µl extraction buffers ( 67 mM TRIS pH 6 . 8 , 133 mM DTT , 2 . 7% SDS , 13% glycerol , 0 . 01% bromophenol blue ) were added and immediately incubated at 95°C for 5 min . After centrifugation , 15 µl of the extracts were separated by SDS-PAGE ( 10% gel ) and blotted onto a 0 . 2 µm polyvinylidene difluoride membrane ( BioRad ) . After blocking with a solution of 5% skim milk powder in TBS-T ( 150 mM NaCl , 10 mM TRIS/HCl pH = 8 . 0 , 0 . 1% Tween 20 ) the membrane was probed with an anti-GFP antibody ( JL-8; Clontech; Takara Bio Europe , Japan ) diluted 1:2000 in skim milk TBS-T solution or anti-alpha-tubulin ( B-5-1-1; Sigma ) diluted 1:50000 in skim milk TBS-T solution . Horseradish peroxidase-conjugated goat anti-mouse antibody ( Dianova , Germany ) was employed as secondary antibody ( 1:20000 ) . For detection , the SuperSignal West Pico chemiluminescent detection reagent ( Thermo scientific , MA , USA ) was used . Three independent root samples were quantified with imageJ and normalized to alpha-tubulin , means and standard error were calculated and statistical significance was evaluated by the Student's t-test using graphpad ( http://www . graphpad . com/quickcalcs/ ) . Figure shows a representative experiment ( out of three independent repetitions ) . For live cell imaging , wherever possible , roots were mounted in a propidium iodide ( PI ) solution ( 0 . 02 mg/ml ) for counterstaining the cell walls . MDY-64 and FM4-64 staining was performed as described ( Scheuring et al . , 2015 ) . For 3D imaging , epidermal cells were recorded with a step size of 1 µm with approximately 17–20 single images . For image acquisition a Leica DM6000 CS , TCS AOBS confocal laser scanning microscope was used , equipped with a HCX PL APO CS 20 . 0 0 . 70 IMM UV or a HCX PL APO CS 63 . 0 × 1 . 20 WATER objective . Fluorescence signals of GFP ( excitation 488 nm and emission 500 nm–546 nm ) , YFP ( excitation 514 nm and emission 525 nm–578 nm ) , RFP ( excitation 561 nm and emission 574 nm–618 nm ) , propidium iodide ( excitation 561 nm and emission 577 nm–746 nm ) and MDY-64 ( excitation 458 nm and emission 465 nm–550 nm ) were processed with the Leica software LAS AF 3 . 1 or with ImageJ ( http://rsb . info . nih . gov/ij/ ) and data was statistically evaluated by Student's t-test using graphpad ( http://www . graphpad . com/quickcalcs/ ) . Representative images are shown . VAMP711_ATTB1_fwd , GGGGACAAGTTTGTACAAAAAAGCAGGCTCGATGGCGATTCTGTACGCC , rev , GGGGACCACTTTGTACAAGAAAGCTGGGTCTTAAATGCAAGATGGTAGAGTAGGTC; UBQ5 fwd , GACGCTTCATCTCGTCC , rev , GTAAACGTAGGTGAGTCCA . Destination vector: pMDC7_B ( pUBQ ) ( Barbez et al . , 2012 ) .	In plants and animals , the way that cells grow is carefully controlled to enable tissues and organs to form and be maintained . This is especially important in plants because the cells are attached to each other by their cell walls . This means that , unlike some animal cells , plant cells are not able to move around as a plant's organs develop . Plant cells contain a large storage compartment called the vacuole , which occupies 30–80% of a cell's volume . The volume of the vacuole increases as the cell increases in size , and some researchers have suggested that the vacuole might help to control cell growth . A plant hormone called auxin can alter the growth of plant cells . However , this hormone's effect depends on the position of the cell in the plant; for example , it inhibits the growth of root cells , but promotes the growth of cells in the shoots and leaves . Nevertheless , it is not clear precisely how auxin controls plant cell growth . Here , Löfke et al . studied the effect of auxin on the appearance of vacuoles in a type of plant cell—called the root epidermal cell—on the surface of the roots of a plant called Arabidopsis thaliana . The experiments show that auxin alters the appearance of the vacuoles in these cells so they become smaller in size . At the same time , auxin also inhibits the growth of these cells . Löfke et al . found that auxin increases the amount of certain proteins in the membrane that surrounds the vacuole . These proteins belong to the SNARE family and one SNARE protein in particular , called VTI11 , is required for auxin to be able to both alter the appearance of the vacuoles and restrict the growth of root epidermal cells . Enzymes called PI4 kinases were also shown to be involved in the control of the SNARE proteins in response to the auxin hormone . Löfke et al . 's findings suggest that auxin restricts the growth of root epidermal cells by controlling the amount of SNARE proteins in the vacuole membrane . The next challenge will be to understand precisely how the shape of the vacuole is controlled and how it contributes to cell growth .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "plant", "biology", "cell", "biology" ]	2015	Auxin regulates SNARE-dependent vacuolar morphology restricting cell size
The replisome is a multiprotein machine that carries out DNA replication . In Escherichia coli , a single pair of replisomes is responsible for duplicating the entire 4 . 6 Mbp circular chromosome . In vitro studies of reconstituted E . coli replisomes have attributed this remarkable processivity to the high stability of the replisome once assembled on DNA . By examining replisomes in live E . coli with fluorescence microscopy , we found that the Pol III* subassembly frequently disengages from the replisome during DNA synthesis and exchanges with free copies from solution . In contrast , the DnaB helicase associates stably with the replication fork , providing the molecular basis for how the E . coli replisome can maintain high processivity and yet possess the flexibility to bypass obstructions in template DNA . Our data challenges the widely-accepted semi-discontinuous model of chromosomal replication , instead supporting a fully discontinuous mechanism in which synthesis of both leading and lagging strands is frequently interrupted . DNA replication is carried out by a multifunctional machine , the replisome ( Beattie and Reyes-Lamothe , 2015 ) . The E . coli replisome has been characterized in vitro and in vivo and is composed of more than 12 different proteins ( Kurth and O'Donnell , 2013; Reyes-Lamothe et al . , 2010 ) . DNA synthesis is performed by the Pol III polymerase ( αεθ ) . Three copies of Pol III are incorporated into the replisome through an interaction with the τ subunit of the pentameric clamp loader complex ( τ3δδ’ ) . Together , these constitute the Pol III* subassembly ( ( αεθ ) 3-τ3δδ’ ) . The clamp loader is also responsible for loading the β clamp dimer onto DNA , which is required for processive synthesis by Pol III . Addition of β clamp to Pol III* forms the Pol III holoenzyme . At the core of the E . coli replisome is the replicative helicase , DnaB , which encircles the lagging strand template and unwinds parental DNA . The Pol III holoenzyme associates with DnaB through the τ subunit of the clamp loader ( Figure 1A ) . In addition , the DnaB helicase recruits the primase , DnaG , which synthesizes RNA primers . Due to the antiparallel nature of DNA , synthesis of one the strands – the leading strand – occurs co-directionally with progression of the replication fork , while the second strand – the lagging strand – is synthesized by repeated cycles of primer synthesis and DNA extension . 10 . 7554/eLife . 21763 . 003Figure 1 . Most replisome subunits exchange frequently with the diffusing pool . ( A ) Model illustrating the architecture of a replisome at the E . coli replication fork . ( B ) Representative fluorescence images of FRAP experiments for the Pol III α subunit and the DnaB helicase . Cell boundaries shown as white lines , red circle shows the location of the bleached focus . ( C ) Representative examples of the FRAP curves for Pol III α subunit ( N = 48 ) and DnaB ( N = 96 ) . Red line shows a reaction-diffusion model fit to the data , dashed grey lines show SE for the model . Dashed blue line represents the estimated maximum possible fluorescence recovery after correcting for photobleaching . ( D ) Analysis summary of the replisome by FRAP . Bars represent average bound-times . Red squares represent level of recovery normalised to the intensity before bleaching . Dashed blue line represents maximum possible fluorescence recovery . It was not possible to estimate the bound-time for DnaB . Error bars represent SE . DOI: http://dx . doi . org/10 . 7554/eLife . 21763 . 00310 . 7554/eLife . 21763 . 004Figure 1—figure supplement 1 . Artificial elongation of cells by cephalexin treatment does not interfere with DNA replication or protein expression . ( A ) Representative images of non-treated ( M9-Gly ) or cells treated with 40 µg/ml of cephalexin ( M9 Gly + Cephalexin ) are shown . The position of the ori1 locus ( mapped at 3908 kb in the chromosome ) , labelled by TetR-mCerulean bound to a tetO array , is shown in red , while the position of the replisome component SSB labelled with YPet is shown in green . Top panels do not show the phase contrast . Scale bar represents 2 µm . A table summarizes analysis of this data , showing the number of ori1 and SSB foci per cell , and the ratio between them . ( B ) Distribution of the total intensity of DAPI signal against cell length in cells untreated ( blue dots ) and treated with cephalexin for one hour ( red dots ) . Ethanol fixation was used to ensure homogenous permeability to the dye . Fitting to a linear model is shown in the respective colours . The inset shows the distribution of mean intensities per pixel for both conditions . ( C ) Distribution of the total intensity of ε-YPet signal against cell length in cells untreated ( blue dots ) and treated with cephalexin for one hour ( red dots ) . Fitting to a linear model is shown in the respective colours . The inset shows the distribution of mean intensities per pixel for both conditions . DOI: http://dx . doi . org/10 . 7554/eLife . 21763 . 00410 . 7554/eLife . 21763 . 005Figure 1—figure supplement 2 . Minimal contribution of YPet photoblinking during FRAP . ( A ) Representative fluorescence images of a cell carrying a tetO operator array and expressing TetR-YPet . Cells were fixed with formaldehyde to avoid protein exchange . Cell boundaries are represented with a white line . The point of localized bleaching is shown with a red circle . ( B ) Average distribution of fluorescence recovery after photobleaching of 49 cells . Note that intensity increase after bleaching is minimal ( <5% of the total intensity ) , consistent with stochastic fluctuations and experimental measurement error . Error bars represent SE . DOI: http://dx . doi . org/10 . 7554/eLife . 21763 . 00510 . 7554/eLife . 21763 . 006Figure 1—figure supplement 3 . Growth rate and replication time of E . coli in our experimental conditions . ( A ) Growth curve of AB1157 in M9-Glycerol at 22°C is shown . Samples were taken every hour for 7 hr . ( B ) Distribution of the number of spots per cell of a strain carrying ε-YPet grown in M9-Glycerol at 22°C ( N = 1403 cells ) . We estimated the replication time by taking into account the generation time and the number of cells with spots . We made the assumption that initiation of DNA replication occurs at cell birth to account for the uneven distribution of cell ages in the population . ( C ) Representative images obtained from a time-lapse experiment of a strain carrying SSB-YPet grown on a 1% agarose pad in M9-Glycerol at 22°C . Pictures were taken at 10 min intervals . Replication time was determined from the time point of first appearance of SSB-YPet spot to its subsequent disappearance ( N = 56 cells ) . The average of the two methods ( 150 min ) is reported in the main text . DOI: http://dx . doi . org/10 . 7554/eLife . 21763 . 006 Replication of the circular chromosome of E . coli proceeds bidirectionally from a single , defined locus: oriC . Multiple mechanisms tightly restrict DnaB loading , and therefore replisome assembly , to the oriC locus during initiation , with a single initiation event per cell cycle ( Costa et al . , 2013 ) . Two sister replisomes are assembled at oriC during initiation , and each is responsible for replicating half of the ~4 . 6 Mbp chromosome . At 37°C , it takes 40–60 min of continuous DNA synthesis to complete chromosomal replication , at rates of 600–1000 bp s−1 . Given the extent of DNA synthesis required , it has been assumed that the replisome is a stable protein complex capable of replicating large fragments of the chromosome without disassembling . This is supported by in vitro data showing that a single purified replisome , once assembled on DNA , is capable of synthesizing DNA with an average length of 70 kbp without requiring replacement of the Pol III* subassembly or DnaB ( Tanner et al . , 2011; Yao et al . , 2009 ) . Even greater stability has been inferred from in vivo experiments that suggest infrequent replication fork collapse during chromosome replication in E . coli ( Maisnier-Patin et al . , 2001 ) . Chromosomal DNA presents multiple potential obstacles to replisome progression . DNA lesions can result in the stalling of the replisome due to Pol III’s inability to use damaged DNA as a template ( Moore et al . , 1981 ) . In addition , the replisome frequently encounters DNA-bound proteins , potentially resulting in disassembly or pausing ( Mettrick and Grainge , 2016; Gupta et al . , 2013 ) . Multiple mechanisms have been proposed that allow replisome integrity to be maintained during bypass of such obstacles ( Heller and Marians , 2006a; Yeeles and Marians , 2011; Pomerantz and O'Donnell , 2008 ) and that remove bound proteins from DNA ( Gupta et al . , 2013 ) . In cases where these strategies are insufficient , the cell also has mechanisms to mediate the reassembly of the replisome at specific DNA structures that arise following replisome collapse ( Heller and Marians , 2006b ) . The frequency at which replisomes encounter these obstacles and the efficiency of the bypass mechanisms are still unclear . It is also uncertain in which way the architecture and stability of replisome play a role during these events . The replisome is likely to be affected by multiple factors present inside cells which have not been accounted for in reconstituted systems . However , a direct measurement of the stability of the replisome has not been undertaken in vivo . Here we measure the binding kinetics of replisome subunits during DNA replication using two independent fluorescence-based methods in living cells . Our results show that the entire Pol III* subassembly is replaced within the replisome at a frequency equivalent to a few cycles of Okazaki fragment synthesis . This leads us to conclude that DNA replication is a discontinuous process on both strands . We also find that the DnaB helicase remains bound to DNA for tens of minutes , preventing disassembly of the replisome likely by serving as a dock during Pol III* subassembly turnover . We propose that this dynamic stability provides the replisome with flexibility to bypass frequent obstacles on DNA while maintaining the necessary processivity for chromosomal replication . To assess the stability of the E . coli replisome when replicating chromosomal DNA in vivo , we first measured the binding kinetics of replisome subunits using fluorescence recovery after photobleaching ( FRAP ) in strains possessing fluorescent YPet derivatives of key replisome components ( Reyes-Lamothe et al . , 2010 , 2008 ) . Using actively replicating cells in growth conditions that permit a single replication event per cell cycle , we bleached individual foci of fluorescent replisome subunits using a focused laser pulse and measured their recovery over time ( Figure 1B ) . The dimensions of E . coli – in our conditions typically ~0 . 7 µm diameter and few microns in length – and the low number of replisome subunit molecules per cell – a few hundred for most subunits ( Reyes-Lamothe et al . , 2010 ) – increased the difficulty of selectively bleaching replisome foci without affecting the remaining fluorescent pool . To minimize photobleaching of the diffusing pool of fluorescent proteins we increased cell volume by treatment with cephalexin; this did not affect DNA replication ( Figure 1—figure supplement 1 ) . To our surprise , we found that the initial focus fluorescence recovered in a few seconds for Pol III and clamp loader components ( Figure 1B ) . Fluorescence recovery is not explained by the photophysical properties of YPet , like photoblinking , and instead it represents protein exchange ( Figure 1—figure supplement 2 ) . We used a reaction-diffusion model in a reaction-limited regime to fit the average fluorescence recovery curve of individual subunits , and calculated a time constant for binding ( bound-time ) which represents the average time that a molecule is bound to the replisome before exchanging . The bound-time was 4 ± 2 and 6 ± 2 s ( mean ± SE ) for the α and ε subunits of Pol III , respectively . Similarly , the τ , δ and χ subunits of the clamp loader had bound times of 6 ± 3 , 3 ± 2 and 6 ± 4 s , respectively ( Figure 1C–D ) . Molecules of the β clamp exchanged at a slower rate , remaining associated for an average of 36 ± 21 s , consistent with its binding to newly-synthesized DNA behind the replisome ( Moolman et al . , 2014; Su'etsugu and Errington , 2011 ) . The timescale of Pol III holoenzyme exchange is in striking contrast to the ~150 min required for two replisomes to complete chromosomal replication under our microscopy conditions ( Figure 1—figure supplement 3 ) . To confirm these results , we used single-particle tracking Photoactivated Localization Microscopy ( sptPALM ) ( Manley et al . , 2008 ) to determine the bound-times of replisome subunits ( Uphoff et al . , 2013 ) . We constructed E . coli strains with functional fusions of replisome subunits and the photoconvertible fluorescent protein , mMaple ( McEvoy et al . , 2012 ) ( Figure 2—figure supplement 1 ) . We used a single low intensity pulse of 405 nm-laser activation per experiment to switch , on average , a single molecule per cell into a red fluorescence state . Long ( 500 ms ) camera exposure times – to motion-blur fast diffusing molecules – were used , spaced by 1 s or 5 s intervals , to track non-diffusing replisome-associated molecules as foci ( Figure 2A ) . This illumination protocol did not perturb cell growth ( Figure 2—figure supplement 2 ) . Track duration distributions for labelled replisome subunits were calculated from the number of frames individual molecules appeared as foci ( Figure 2B–C ) . Bound times were calculated by correcting for the disappearance of foci due to photobleaching , which was characterized using a strain carrying the transcriptional repressor LacI fused to mMaple and a chromosomal array of lacO binding sites . We also assessed the effect photoblinking using this same strain ( Figure 2—figure supplement 3A–C ) . We expect that in the timescale of our experiments , the lifetime of LacI-mMaple foci will be dictated by photobleaching , with dissociation from DNA being negligible ( Hammar et al . , 2014 ) . 10 . 7554/eLife . 21763 . 007Figure 2 . Exchange of Pol III* subassembly and DnaB occur on different timescales . ( A ) Diagram illustrating the sptPALM experimental design used to measure bound-times . ( B ) Representative example of the focus life span for the Pol III ε subunit . ( C ) Representative examples of the distribution of fluorescent foci life-spans ( blue bars ) for Pol III ε subunit and DnaB , showing fitting of a single-exponential decay model ( red line ) , the estimated bleaching rate in the same conditions ( blue line ) and the corrected estimated bound-time ( purple line ) . Note that to improve accuracy in single-molecule detection tracks shorter than four localizations were removed in the case of ε but corrected during curve fitting , hence the lower bar near 0 s time point . ε data was collected using 500 ms exposure time and 1 s intervals ( N = 143 ) , DnaB data was collected using 2 s exposure time and 10 s intervals ( N = 86 ) . The plot for DnaB shows binned data for presentation purposes . ( D ) Summary of estimated average bound-times . Errors in the table represent SE . DOI: http://dx . doi . org/10 . 7554/eLife . 21763 . 00710 . 7554/eLife . 21763 . 008Figure 2—figure supplement 1 . Characterisation of mMaple fusions . ( A ) Plot showing growth curves of the AB1157 and derivative strains carrying mMaple fusions to replisome components . Experiment was done in M9-Glycerol at 37°C . Average and standard deviation of three experiments are shown . ( B ) Table that summarizes the results from the growth curve experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 21763 . 00810 . 7554/eLife . 21763 . 009Figure 2—figure supplement 2 . Minimal exposure to 405 nm activation light allows continuation of cell growth . ( A ) Images obtained from an sptPALM experiment of a strain carrying mMaple-DnaB using 2 s exposure times of the 561 nm laser and 2 min intervals ( time in minutes ) . Note the growth of cells despite exposure to a single event of 405 nm wavelength activation and multiple exposures to 561 nm wavelength light . Scale bar = 2 µm . ( B ) Plot showing lengths of cells over time for eight different cells . The average of doubling-time is similar to the generation time of AB1157 at 22°C . DOI: http://dx . doi . org/10 . 7554/eLife . 21763 . 00910 . 7554/eLife . 21763 . 010Figure 2—figure supplement 3 . Estimation of photoblinking , test for two binding kinetic regimes and characterisation of the effect of longer 2 s capture rates in our estimation of bound-times . ( A ) Frequency of detected short gaps , likely representing photoblinking , during the tracking of a population of LacI-mMaple molecules using 500 ms capture rates . We applied a cut-off threshold at 2 . 6 s for the maximum duration of photoblinking based on previous characterisation of mMaple ( Durisic et al . , 2014 ) . More than 75% of the molecules did not show photoblinking ( N = 148 molecules ) . ( B ) Distribution of gap times between subsequent localizations at the same location of the field of view . Note that most events lasted for only one frame . N = 60 events . ( C ) We fitted the distribution of gap times to a single exponential function using a truncated form of MLE . This was done to account for the fraction of events shorter than 500 ms , which would be missed in our experiments . Using a 1-frame memory parameter we estimate that our analysis will prematurely terminate less than 7 . 5% , 3% and 0 . 0001% of the tracks due to blinking when using a 1 s , 2 s , and 5 s intervals , respectively . ( D ) Semi-log plots of the data presented in Figure 2C for ε and DnaB . The plots show a relatively linear relation between number of cases and time , which is indicative of a single regime of binding for both subunits . Further support of a single binding behaviour is presented in Supplementary file 1C . ( E ) Plots showing the PDF curves of bound , bleaching , and tracking times for representative results from a single experiment of ε imaged with 500 ms ( left ) ( N = 143 ) and 2 s exposure ( right ) ( N = 415 ) . The bound-time was 7 . 44 s ( SE ±1 . 07 s ) and 12 . 34 s ( SE ±1 . 36 s ) for 500 ms and 2 s , respectively . The plot for the 500 ms example is presented to facilitate comparison and is identical to that in Figure 2C . DOI: http://dx . doi . org/10 . 7554/eLife . 21763 . 01010 . 7554/eLife . 21763 . 011Figure 2—figure supplement 4 . Slow diffusion of DnaB helicase complicates correct assignment of immobile molecules at sub-second capture rates . ( A–F ) Analysis from PALM experiments of ε-mMaple and mMaple-DnaB using a 21 ms capture rate . ( A ) Calculated MSDs for the two proteins . In both cases , curves plateau at around 600 nm in agreement with the dimension of the short axis of the cell . ( B ) Comparison of step lengths in the tracks of diffusing molecules analysed . ( C–D ) Examples of detected tracks for ε and DnaB . Lines of different colours are used to facilitate the observation of individual tracks . A red line was only used to show the position of tracks representing immobile tracks where the apparent diffusion coefficient is close to 0 . The outline of the cell is shown in grey . ( E–F ) Distribution of the apparent diffusion coefficients calculated . 2344 tracks obtained from 77 cells and 2467 tracks obtained from 90 cells were used for ε and DnaB , respectively . ( G ) Distribution of mean PSFs for the x-axis for tracks of ε-mMaple or mMaple-DnaB obtained from experiments done with 500 ms capture rates . In the case of ε , a clear peak close to 100 nm shows the PSF dimensions of immobile molecules , while a second peak close to 200 nm represents diffusing molecules . In contrast , the dimensions of the PSFs from immobile and diffusive molecules is less clear for DnaB . The dashed line shows the threshold used to assign immobile and diffusing molecules in our analysis . DOI: http://dx . doi . org/10 . 7554/eLife . 21763 . 011 The single-molecule results are consistent with our FRAP data . Pol III subunit and clamp loader components indeed exchanged rapidly , with ε , τ and δ remaining replisome-associated for only 10 ± 0 . 7 , 10 ± 0 . 7 and 12 ± 0 . 9 s ( mean ± SE ) , respectively ( Figure 2B and D ) . We found no strong evidence for multiple binding behaviors of individual subunits ( Figure 2—figure supplement 3D and Supplementary file 1C ) , suggesting that both leading and lagging strand polymerases behave similarly . As with FRAP , we observed similar bound-times for all subunits of both the DNA polymerase III and clamp loader complexes despite a difference in stoichiometry – δ , τ and ε are present in 1 , 3 and 3 copies per replisome , respectively ( Figure 1A ) . As such , exchange of individual subunits independently from one another , although still possible , does not easily explain our results . We therefore propose that the unit of exchange of Pol III and clamp loader subunits is the Pol III* subassembly ( ( αεθ ) 3-τ3δδ’ ) . This idea is supported by in vitro data that shows that exchange of Pol III at the replication fork requires it to be part of the Pol III* subassembly ( Yuan et al . , 2016 ) . Cells have an excess of free Pol III that does not interact with the clamp loader ( Maki and Kornberg , 1985 ) . Consistent with the notion that only Pol III subunits found within a Pol III* subassembly are competent for exchange , we observed re-binding of single molecules of ε , often to a different replisome , at a much higher frequency than would be predicted if all of the ~270 molecules of ε in the cell were in direct competition ( Reyes-Lamothe et al . , 2010 ) ( Figure 3A and Figure 3—figure supplement 1 ) . 10 . 7554/eLife . 21763 . 012Figure 3 . DnaB may serve as a platform for frequent Pol III* subassembly re-binding . ( A ) Representative example of a cell where a single activate copy of ε-mMaple shows multiple cycles of binding and unbinding ( time in seconds ) . Obtained from an experiment using 2 s intervals between consecutive images . Frame average shows the cellular location of two replisomes . ( B ) Example of a cell where a single activated mMaple-DnaB molecule remains localized as a focus for several minutes ( time in minutes ) . Obtained from an experiment using 10 s intervals between consecutive images . Boundaries of the cells at the beginning of the experiment are shown as white outlines . DOI: http://dx . doi . org/10 . 7554/eLife . 21763 . 01210 . 7554/eLife . 21763 . 013Figure 3—figure supplement 1 . Re-binding of copies of ε at the same position are unexpectedly frequent . ( A ) Plot showing the number of rebinding events during a single experiment . ( B ) Distribution of gap times , calculated from data obtained using 2 s exposure time . Rebinding analysis was done similarly to the track linkage analysis , except no frame threshold was applied . Time between re-binding events was obtained by determining the time interval between the end of a single-molecule track and the beginning of a new track at the same position in the cell . Note that in reality the frequency of binding will likely be higher , since molecules have a similar probability of binding to a different replisome between consecutive events . We manually analysed cells with two replisome spots and estimated that 52% of consecutive spot reappearance events occur at a different position in cells with two replisomes ( N = 30 cells; 199 re-binding events ) . DOI: http://dx . doi . org/10 . 7554/eLife . 21763 . 013 The β clamp again showed a longer bound-time of 47 ± 3 s in sptPALM . Our estimate is broadly consistent with a previous estimate from E . coli , although ~4 times shorter ( Moolman et al . , 2014 ) . Assuming similar numbers of DNA-bound β as in that earlier report , we estimate from our results that a new β dimer is loaded at each replication fork every ~2 s . Considering an average rate of fork progression of ~260 bp s−1 – calculated from the duration of replication in the conditions used ( Figure 1—figure supplement 3 ) – we find an average Okazaki fragment length of 520 bp , in close agreement with an in vitro measurement of 650 bp at room temperature ( Yao et al . , 2009 ) . In contrast , we estimate from FRAP and sptPALM an average replication fork progression of 1–3 kbp prior to Pol III exchange , which would allow completion of multiple Okazaki cycles . We therefore think it unlikely that subunit exchange within the replisome is exclusively linked to the dynamics on the lagging strand . The DnaB helicase displayed very different dynamics to other replisome subunits when assessed by FRAP . Crucially , we never observed full fluorescence recovery of DnaB over 5 min of measurement ( Figure 1B–D ) , indicating that it is stably associated with the replisome on this timescale . Our analysis showed an initial recovery of fluorescence with a 7 ± 4 s time constant , which we attribute to the signal from diffusing molecules moving into the bleached area . However , we did not observe a significant increase in the intensity after this initial time point , preventing us from accurately estimating a bound-time for DnaB ( Figure 1C ) . We conclude that in contrast to Pol III holoenzyme subunits , replisome-associated DnaB does not exchange frequently , and is instead a stable component of the replisome . Analysis of DnaB by sptPALM confirmed that it is the most stable subunit in the replisome . To eliminate incorrect assignment of DnaB fluorescence by our software we used even longer ( 2 s ) exposure times , thus further blurring slow-diffusing molecules ( Figure 2—figure supplement 4 ) . Control experiments with ε under the same conditions had no significant effect on calculated bound-times ( Figure 2—figure supplement 3E ) . Crucially , we estimate that single molecules of DnaB remained bound to the replisome for 913 ± 508 s , significantly longer than any other component ( Figure 2B–D ) . The width on the distribution for this estimate is inherently large due to a close similarity between DnaB foci lifetimes and the bleaching time of mMaple ( Figure 2C ) . However , our long bound-time estimate is supported by frequent examples of DnaB fluorescent foci that last for tens of minutes ( Figure 3B ) . Currently we cannot assess the extent at which the turnover detected represents PriABC mediated re-loading of helicase ( Heller and Marians , 2006b ) . Altogether , our data supports a role for DnaB as the primary determinant of replisome integrity . To determine if subunit exchange occurs as a consequence of active DNA synthesis , we measured bound-times by FRAP and sptPALM in cells treated with the DNA polymerase inhibitor hydroxyurea ( HU ) ( Sinha and Snustad , 1972 ) . Components of the Pol III* subassembly and the β-clamp showed bound-times two- to six fold higher than in untreated cells ( Figure 4A–B ) . HU had no apparent effect on DnaB FRAP estimates ( Figure 4A ) . We conclude that the exchange of replisome components is at least partially dependent on active DNA synthesis . Remaining turnover may reflect residual DNA synthesis after HU treatment . In addition , it may indicate that the replisome is intrinsically dynamic as a multiprotein complex , with DNA synthesis further increasing subunit exchange . 10 . 7554/eLife . 21763 . 014Figure 4 . Replisome dynamics are partly dependent on active DNA synthesis . ( A ) Summary of the average bound-times in cells treated with HU , estimated by FRAP . The results for untreated ( blue bars ) and treated cells ( green bars ) is shown . Data for the untreated condition is presented to facilitate comparison and is identical to that in Figure 1D . Red squares represent the normalised level of fluorescence recovery . Dashed blue line shows estimated maximum possible recovery . It was not possible to estimate the bound-time for DnaB . ( B ) Summary of bound-times estimated by sptPALM ( weighted average ) . Results for the untreated ( blue bars ) and treated cells ( green bars ) is shown . Data obtained using cells treated with the RNA Polymerase inhibitor Rifampicin is also shown for cells carrying ε-mMaple ( purple bars ) . Data for the untreated condition is presented to facilitate comparison and is identical to that in Figure 2D . Error bars represent SE . DOI: http://dx . doi . org/10 . 7554/eLife . 21763 . 01410 . 7554/eLife . 21763 . 015Table 1 . Strains used for this study . DOI: http://dx . doi . org/10 . 7554/eLife . 21763 . 015StrainRelevant genotypeSourceAB1157thr-1 , araC14 , leuB6 ( Am ) , DE ( gpt-proA ) 62 , lacY1 , tsx-33 , qsr'-0 , glnV44 ( AS ) , galK2 ( Oc ) , LAM- , Rac-0 , hisG4 ( Oc ) , rfbC1 , mgl-51 , rpoS396 ( Am ) , rpsL31 ( strR ) , kdgK51 , xylA5 , mtl-1 , argE3 ( Oc ) , thi-1Dewitt and Adelberg , 1962RRL27holC-ypet kanReyes-Lamothe et al . ( 2008 ) RRL30holE-ypet kanReyes-Lamothe et al . ( 2008 ) RRL32ssb-ypet kanReyes-Lamothe et al . ( 2008 ) RRL33holA-ypet kanReyes-Lamothe et al . ( 2008 ) RRL34holD-ypet kanReyes-Lamothe et al . ( 2008 ) RRL35dnaE-ypet kanReyes-Lamothe et al . ( 2008 ) RRL36dnaQ-ypet kanReyes-Lamothe et al . ( 2008 ) RRL51dnaX-ypet kanReyes-Lamothe et al . ( 2008 ) RRL196frt ypet-dnaNReyes-Lamothe et al . ( 2010 ) RRL368frt-ypet-dnaBReyes-Lamothe et al . ( 2010 ) RRL537dnaQ-mMaple kanThis studyRRL538holA-mMaple kanThis studyRRL541tetR-ypet kan , [tetO240-gm]852This studyRRL553dnaX-mMaple kanThis studyRRL557frt mMaple-dnaBThis studyRRL558frt mMaple-dnaNThis studyTB44dnaB-mMaple kanThis studyTB54lacI-mMaple kan , [lacO240-hyg]2735::ΔpheAThis studyPlasmidFeaturesSourcepKD46Expression of lambda red genesDatsenko and Wanner , 2000pCP20Expression of Flp recombinaseDatsenko and Wanner , 2000pROD61mYPet Kan R6K gamma ori . For C-ter insertionsThis studypROD83YPet Kan R6K gamma ori . For N-ter insertionsThis studypROD93mMaple Kan R6K gamma ori . For C-ter insertionsThis studypROD160mMaple Kan R6K gamma ori . For N-ter insertionsThis study Obstructions in template DNA , particularly protein-DNA complexes , have been shown to cause frequent pausing of the E . coli replisome in vivo ( Gupta et al . , 2013 ) . In vitro studies have demonstrated that the E . coli replisome is capable of bypassing such obstacles by interrupting leading strand synthesis and resuming extension downstream of the obstruction from a leading strand primer deposited by DnaG or an mRNA synthesized by RNA polymerase ( Heller and Marians , 2006a; Yeeles and Marians , 2011; Pomerantz and O'Donnell , 2008 ) . Our data is entirely consistent with this mechanism , whereby bypass could be achieved through detachment of the stalled Pol III* subassembly from DnaB and its replacement downstream of the obstacle with another Pol III* from solution . Because DnaB translocates on the lagging strand , small lesions and large protein blockages on the leading strand can both be bypassed by Pol III* exchange . Note however that we did not observe any apparent effect on the dynamics of Pol III after inhibiting transcription , suggesting that this process is not the main cause of exchange ( Figure 4B ) . On the lagging strand , small template lesions capable of passing through the central pore of DnaB may also be bypassed through Pol III* exchange . In contrast , obstacles on the lagging strand that destabilize DnaB , such as proteins stably bound to DNA or strand discontinuities , would likely result in the disassembly of the replisome . We propose that obstacle bypass along template DNA may be the primary selection pressure that has driven the evolution of a dynamic replisome . In addition to the model above , we acknowledge that other processes may also exert selective pressure for the generation of the observed replisome binding kinetics . First , unbinding of Pol III* subassembly may result from build-up of helical torsion in the template DNA generated by the coupled synthesis of both DNA strands ( Kurth et al . , 2013 ) . This is consistent with longer binding times when synthesis was inhibited by HU . Unbinding of a single polymerase from DNA or release of the whole Pol III* subassembly would have the same effect on stress relief . Second , the dynamics observed may be a byproduct of the highly regulated interaction between Pol III and β-clamp . Even though Pol III tightly binds the β-clamp to ensure highly processive synthesis , these two proteins rapidly unbind from each other upon completion of the duplex DNA . The strength of this protein-protein interaction is modulated by the OB domain in the α subunit of Pol III , which binds to ssDNA ahead of the catalytic domain , and the C-terminus of the τ subunit of the clamp loader ( Georgescu et al . , 2009 , Leu et al . , 2003 ) . Premature activation of such a switch in both leading and lagging strand polymerases would weaken the grip of the Pol III* subassembly on the replication fork and potentially result in its displacement . This idea is consistent with the presence of ssDNA gaps in the lagging strand ( Li and Marians , 2000 ) , which may be explained by a premature loss of Pol III processivity . Presumably , exchange of Pol III* subassembly within the replisome occurs rapidly enough to minimize potentially deleterious ssDNA gaps between fragments of nascent DNA on the leading strand . However , the rate of DNA unwinding by DnaB decreases by more than 10-fold when DnaB is detached from the τ subunit of the clamp loader ( Kim et al . , 1996a ) , providing a potential safety mechanism to limit DNA unwinding and exposure of ssDNA during Pol III* subassembly exchange . It remains to be determined if a newly associated copy of Pol III uses the existing 3’ end at the leading strand or requires the activity of primase to resume synthesis . Our data apparently contradict in vitro studies which have demonstrated that a single reconstituted E . coli replisome can operate without subunit exchange in synthesizing an average of ~80 kbp ( Tanner et al . , 2011; Yao et al . , 2009 ) . Measurements in those reports were performed by removing all diffusing Pol III* subassembly and DnaB subunits from the reaction . In contrast , in the cell there is a permanent excess of diffusing replisome subunits . We believe this explains the differences observed with our in vivo data . Competition for binding sites between DNA-bound and diffusing molecules has been shown to change the DNA-binding kinetics of proteins such as Fis , HU and NHP6A ( Graham et al . , 2011 ) , EcoRI ( Sidorova et al . , 2013 ) , RPA ( Gibb et al . , 2014 ) and the transcription factor CueR ( Chen et al . , 2015 ) . Furthermore , mathematical modelling has shown that it is theoretically possible for a replisome to be stable under conditions in which no extra subunits are present , as in vitro , and yet undergo frequent subunit exchange in the presence of extra subunits , as in vivo , due to subunit competition ( Åberg et al . , 2016 ) . We think that active synthesis may enhance exchange with the diffusing pool , consistent with our results using HU . Frequent exchange of DNA polymerases in the presence of extra subunits has been observed in the replisomes of bacteriophages T4 and T7 in vitro ( Yang et al . , 2004; Johnson et al . , 2007 ) . In T7 , this occurs through a mechanism in which extra DNA polymerases associate with the bacteriophage DNA helicase and exchange with the active DNA polymerase through competition for DNA binding ( Geertsema et al . , 2014; Loparo et al . , 2011 ) . It will be interesting in the future to determine the mechanisms that exist in E . coli to ensure efficient capture and exchange of the low-abundance Pol III* subassembly . One predicted consequence of the Pol III* subassembly exchanging as a single unit is that synthesis of both leading and lagging strands will be frequently interrupted , resulting in discontinuities on both strands . This contrasts the widely-accepted semi-discontinuous model of DNA replication . However , while this model is strongly supported by in vitro experiments ( Wu et al . , 1992 ) , the mechanism that operates in vivo has long been unclear ( Yeeles , 2014; Wang , 2005 ) . Okazaki and colleagues’ original characterization of replication intermediates demonstrated that all DNA is initially synthesized as short fragments , supporting fully discontinuous DNA replication ( Okazaki et al . , 1968 ) . More recent in vivo experiments performed in the absence of DNA ligase support the idea that discontinuities are produced on both leading and lagging strands during DNA replication in E . coli ( Amado and Kuzminov , 2013 , 2006 ) . Our data provide a mechanistic explanation for these observations , and supports a discontinuous model of DNA synthesis in E . coli . All strains used are derivatives of AB1157 . Cells were routinely grown in LB or in M9 minimal media . M9 was supplemented with glycerol ( final concentration 0 . 2% ) ; 100 µg/ml of amino acids threonine , leucine , proline , histidine and arginine; and thiamine ( 0 . 5 µg/ml ) . When required , antibiotics were added at the following concentrations: ampicillin ( 100 µg/ml ) , kanamycin ( 30 µg/ml ) , chloramphenicol ( 25 µg/ml ) , cephalexin ( 40 µg/ml ) , rifampicin ( Rif ) ( 300 µg/ml ) and hydroxyurea ( HU ) ( 60–100 mM ) . For microscopy , cells were spotted on a 1% agarose pad in M9-Glycerol . DAPI was used at a working concentration of 300 nM as recommended by manufacturer . Ethanol fixation was done using 70% ethanol in water , followed by two washes with PBS . For TetR-YPet strain , fixation was done using 4% formaldehyde and incubating 15 min at room temperature , 15 min on ice , followed by two washes with PBS . Chromosomal replacement of replisome genes by fluorescent derivatives was done by lambda red ( Reyes-Lamothe et al . , 2008; Datsenko and Wanner , 2000 ) . In short , we used plasmids carrying a copy of ypet ( Reyes-Lamothe et al . , 2008; Nguyen and Daugherty , 2005 ) or mMaple ( McEvoy et al . , 2012 ) followed or preceded by a kanamycin resistance cassette flanked by frt sites as PCR templates . Flexible peptides with sequences SAGSAAGSGEF ( YPet C-ter fusions ) , SAGSAAGSGAV ( mMaple C-ter fusions ) or SAGSAAGSGSA ( YPet and mMaple N-ter fusions ) were used as a linker between the fluorescent protein ( FP ) and the protein targeted . Primers carrying 40-50nt tails with identical sequence to the chromosomal locus for insertion were used to amplify the linker-FP-kanR ( or kanR-FP-linker in the case of N-terminal fusions ) from template plasmids . The resulting PCR product was transformed by electroporation into a strain carrying the lambda red-expressing plasmid pKD46 . Colonies were selected by kanamycin resistance and ampicillin sensitivity , screened by PCR using primers annealing to regions flanking the insertion , and sequenced . In the case of N-terminal fusions , in order to minimize the effect of the insertion on the expression levels of the gene we removed the kanamycin cassette by expressing the Flp recombinase from plasmid pCP20 ( Datsenko and Wanner , 2000 ) . Gene fusions did not have any apparent detrimental effect on cell growth ( Figure 2—figure supplement 1 ) . LacI-mMaple was generated through lambda red using the strain MG1655 . The gene fusion was then transduced , using P1 phage , into an AB1157 derivative carrying a 256-lacO array replacing the pheA gene ( chromosomal position 2735 kb ) ( Wang et al . , 2006 ) . Similarly , a TetR-YPet fusion expressed from a lac promoter ( Reyes-Lamothe et al . , 2014 ) was transduced into a strain carrying a 256-tetO operator array at R3 ( chromosomal position 852 kb ) ( Wang et al . , 2006 ) . Cells were grown in M9-Glycerol at 30°C , treated with cephalexin for 2 hr , harvested at early log-phase ( OD6000 . 1–0 . 2 ) , concentrated and spotted onto a pad of 1% agarose in M9-Glycerol , contained in a gene frame ( Thermo Scientific ) . Treatment with hydroxyurea was done on the agarose pad by mixing HU with media and agarose . Cells were incubated on the slides for 10 min before imaging . Most FRAP experiments , except for the TetR-YPet control , were performed using a spinning disk imaging system ( PerkinElmer ) with a 100x NA1 . 35 oil objective and an ImagEM EMCCD camera ( Hamamatsu Photonics ) . Images were acquired using Volocity imaging software . An image was acquired in the brightfield channel at the beginning of the experiment to serve as a reference . FRAP was performed by pulse-bleaching using a 488 nm laser for 10–15 ms and 30–50% laser intensity ( radius of the spot was diffraction limited at ~300 nm ) . Two pre-bleach images were captured , the bleach spot was centered on one replisome focus and recovery of the bleached region was recorded at different intervals after bleaching . Image capture was done at a 300 ms frame rate ( 4–6% 515 nm laser ) for most replisome components except for DnaB helicase , for which 500 ms capture rate was used ( 2% 515 nm laser ) . For α , ε , τ , δ and χ , we used intervals between pictures of 2 s , 5 s and 10 s . For DnaB and β , we used intervals between pictures of 5 s , 10 s and 20 s . Experiments were done at room temperature . FRAP to control for photoblinking was done using an epifluorescence system , Leica DMi8 , with a 100x oil objective ( Leica 100x/NA 1 . 47 HL PL APO ) and an iXon Ultra 897 EMCCD camera ( Andor ) . FRAP was performed using an iLas2 unit ( Roper Scientific ) using an ILE laser combiner ( Andor ) and a 150 mW 488 nm laser . Both bleaching and excitation of YPet were done using the 488 nm laser . Acquisition was done using 100 ms exposure at 5 s intervals . Initial position of spots was manually selected using the coordinates for localized bleaching in the image recorded by the acquisition software . Tracking was then done automatically using a previously developed custom program in MATLAB ( Mathworks ) , ADEMS code ( Miller et al . , 2015 ) ( freely available at https://sourceforge . net/projects/york-biophysics/ ) . Most experiments analyzed had a pixel size of 100 nm , for which we used a search window with a radius of 5 pixels and an initial guess for the PSF of 3 pixels when fitting candidate spots . For a minority of the experiments , with pixel size of 140 nm , analysis was done using 4-pixel search window and a 2-pixel radius for initial fitting . Intensity traces were filtered to retain only those where clear bleaching was observed . We removed any trace where the intensity at any of the pre-bleach time points was below the value of the ROI immediately after bleaching ( 0 s time point ) . In addition , the intensity at the 0 s time point had to be below 40% of the mean pre-bleach intensity . FRAP data were then normalized by the average intensity of the pre-bleached data points . To estimate the maximum possible fluorescence recovery ( Max recovery ) , we used the corresponding brightfield image to draw a polygon in ImageJ ( Schneider et al . , 2012 ) around cells containing a bleached spot . We used these ROIs to obtain the intensities across the experiment in the fluorescence channel . Max recovery was calculated by dividing the intensity of the cell at 0 s time point by the average intensity before bleaching . An average Max recovery value was obtained from all the bleached cells in the experiment . To correct for photobleaching during the experiment , a different set of spots was manually selected in cells not exposed to localized bleaching , so they could serve as a baseline control . An average bleaching curve was produced using the intensity traces from these fluorescent foci . All data used to generate the bleaching curve were obtained in the same day using the same strain , excitation settings and interval between pictures as for the FRAP experiment . The average curve was fitted to an exponential decay function . FRAP intensity traces were corrected by dividing each time point by the corresponding normalized value in the fitted bleaching curve . Data from the same set of experiments were averaged . Data from experiments performed the same day , but having different intervals between pictures , were collated into a single recovery curve . Data were then fitted by an exponential solution of the reaction-diffusion equation in a reaction-limited regime ( equation 1 ) using MATLAB: ( 1 ) y=c−ae−bt where c is the asymptote for recovery , a the amplitude of recovery , and b the rate of unbinding ( i . e . koff ) . Bound-times are the reciprocal of koff . Upper boundary for c during fitting was set to the Max recovery ( see above ) , plus ten percent of this value to account for measurement error . In addition to R squared , which is not recommended for non-linear models , goodness of fit was assessed using the Kolmogorov-Smirnov test by measuring the normality in the distribution of residuals ( Andrae et al . , 2010 ) . Standard errors and 95% confidence intervals ( CI ) on the parameter estimates were calculated using the variable values previously obtained , as initial estimates , and bootstrap sampling was performed over 10 , 000 samples ( Supplementary file 1A ) . The values reported in the figures are weighted averages of all the experiments done for the same subunit . We expect that co-localization of sister replisome will have no effect on the rates calculated since the intensity of every spot is normalized against itself in FRAP , and the average rate of recovery is the same at every replisome . Similarly , in sptPALM binding time of individual molecules should not be influenced by a nearby replisome , resulting only a minimal increase in the probability of re-binding to the same place . Cells were harvested from early log-phase cultures in M9-Glycerol ( OD6000 . 1–0 . 2 ) , concentrated and spotted onto a pad of 1% agarose in M9-Glycerol , contained in a gene frame . Coverslips cleaned with versa-clean , acetone and methanol were used to minimize fluorescent background . Treatment with hydroxyurea was done on the agarose pad , by mixing HU with media and agarose . Imaging was performed at room temperature on an inverted Olympus IX83 microscope using a 60x oil objective lens ( Olympus Plan Apo 60X NA 1 . 42 oil ) or 100x oil objective lens ( Olympus Plan Apo 100X NA 1 . 40 oil ) . Images were captured using a Hamamatsu Orca-Flash 4 . 0 sCMOS camera . Excitation was done from an iChrome Multi-Laser Engine from Toptica Photonics . Laser triggering was done through a real-time controller U-RTCE ( Olympus ) . Experiments were done using HiLo illumination setup ( Tokunaga et al . , 2008 ) from a single-line cell^TIRF illuminator ( Olympus ) . Olympus CellSens 2 . 1 imaging software was used to control the microscope and lasers . For experiments with replisome subunits fused to mMaple , a single 405 nm wavelength activation event , typically lasting less than 20 ms , was followed by multiple 561 nm wavelength excitation events with camera captures of 500 ms spaced by 1 s or 5 s intervals , or camera captures of 2s with continuous excitation ( 2 s rates ) or 10 s intervals . Low levels of exposure to violet-blue light were used to minimize photoxicity and allow cells to continue growing during the experiments ( Figure 1—figure supplement 3 ) . To image LacI , we used continuous illumination of 561 nm wavelength after a single 405 nm wavelength activation event at capture rates and intervals of 500 ms or 2 s . We also used 2s capture with 10 s intervals to characterize LacI bleaching in long experiments . Rifampicin experiments were done in a similar manner except Rifampicin was added to the M9-Gly agarose pad , and imaging was done after a 20 min incubation on the agarose pad . We noticed that fewer spots were detected , consistent with inhibition of replication initiation through Rifampicin . Images were first segmented in order to remove out-of-cell noise coming from contaminants on the coverslip . Binary masks were created using ImageJ , either from the differential interference contrast ( DIC ) channel or the green fluorescent channel of mMaple . For DIC , alignment was done by first obtaining a maximum-intensity projection of the PALM timelapse , and subsequently aligning it to the reference DIC , using ImageJ . Each slice of the PALM timelapse was then multiplied by the binary mask , to retain intensities within cells only . An average value of the out-of-cell background was added to regions outside of the ROIs to minimize incorrect assignment by the detection program due to sharp intensity increases . PALM tracking was performed using previously developed software ( Uphoff et al . , 2013 ) , based on the DAOSTORM ( Holden et al . , 2011 ) localization algorithm . An intensity threshold was used to find candidate molecules . The positions of the candidate molecules were then used as initial guesses for a 2D-elliptical Gaussian fit . The fitted parameters were: x-position , y-position , x-standard deviation , y-standard deviation , intensity , brightness , elliptical rotation angle , and background . Tracking was done based on a widely used algorithm ( Crocker and Grier , 1996 ) . Localizations were linked if they appeared within a 300 nm radius between consecutive frames , using a memory parameter of one frame to account for blinking or missed localizations ( i . e . the molecule can go missing for one frame and still be linked ) . Further refinement of the recorded tracks was done to analyze only those that represented immobile single-molecules . To remove slow-diffusing molecules , we plotted a histogram of the PSFs in x and y for all localizations , and performed a two-component Gaussian mixture fit using Maximum Likelihood Estimation . The component with the smaller mean PSF likely represents bound molecules , whereas the other component represents unbound molecules . The two-component Gaussian mixture model has the following form: ( 2 ) p ( 1σ12π ) e− ( x−μ1 ) 22 ( σ1 ) 2+ ( 1− p ) ( 1σ22π ) e− ( x−μ2 ) 22 ( σ2 ) 2 Where p is the mixture probability , σ1 and µ1 are the standard deviation and mean of normal distribution 1 , respectively . Likewise , σ2 and µ2 are the standard deviation and mean of normal distribution 2 , respectively . From the fit , we identified the mean and standard deviation of the component representing bound molecules . We then took 2 standard deviations above the mean to obtain an initial estimation of the threshold . We assessed the accuracy of tracking by manually comparing the tracking results for a subset of fluorescent spots with their lifetime in the original images . Using this method we determined that a threshold of x ≤ 170 and y ≤ 215 , placed on the mean PSF over the track , helped to eliminate most of the unbound molecules from subsequent analysis . In addition , we varied the threshold on the number of localizations for track acceptance across different time-intervals of capture . Our reasoning was that the probability that a track represents a genuinely bound molecule becomes higher as the time interval used increases ( Mazza et al . , 2012 ) . Therefore , the thresholds for removing tracks were <4 , <3 , <2 , and <2 localizations for interval times of 1 s , 2 s , 5 s , and 10 s , respectively . The thresholds were selected by comparing the raw image by eye to the tracks found by the tracking software . Technically no tracks were removed for 5 s and 10 s since tracks with 1 localization cannot be used to calculate track durations . To quantify only single-molecule tracks , we plotted a histogram of the mean intensity of a track and fit using a Gaussian Mixture Model ( GMM ) , utilizing the Expectation-Maximization ( EM ) algorithm . The intensity values were clustered based on membership probabilities ( i . e . the probability of belonging to a particular Gaussian component ) . We used a 2 component GMM fit for most cases and isolated the cluster having the lowest mean , as the intensity values from this cluster likely represent single molecules . We used a 3 component GMM fit in some cases where a significant portion of the molecules seemed out of focus , resulting in a sharp spike of low intensity values in the histogram . In such cases , we isolated the cluster with the second lowest mean . This was especially important when studying proteins with long bound-times , where track fragmentation has a greater relative effect in underestimating the real track duration . We also performed a Bayesian Information Criterion ( BIC ) test to confirm that the 3 component GMM fit better than the 2 component model . We used only track durations with single molecule intensity values for subsequent analysis . To avoid track fragmentation in the analysis of proteins with long bound-times , as in LacI and DnaB , caused by fluctuations in intensity or the molecule moving transiently out-of-focus , we determined the typical length of time that the localization software misses spot detection during long tracks ( gap time ) . We did this by manually comparing the outcome of the analysis to the lifetime of a subset of spots in the original images . We found that on average , the gap was ~4 frames . Therefore , we linked tracks based on the criteria that their mean positions were ≤300 nm apart and gap time between them was <= 4 frames . For these data sets , we performed the GMM fit for isolation of single-molecule tracks after track linkage . The track durations of multiple samples taken on the same day and time-interval were amalgamated into one data set . In order to get the average track duration , we fitted the track durations using MLE . The reason for our choice of MLE over the more commonly used Least Squares-Estimation ( LSE ) method is that it is invariant to the bin size ( i . e . the parameter estimate is the same regardless of how we bin the data ) and it allows us to infer what the population parameter is . Essentially , we use information from our sample data ( track duration times and track acceptance threshold ) as input into MLE , in order to find the population probability density function ( PDF ) , that makes our data the most likely . The fitted lines represent this PDF ( Myung , 2003; Woody et al . , 2016 ) . Histograms were binned based on the square-root rule , where the number of bins is equal to the square root of the sample size . We binned our data for presentation purposes only , in order to reduce noise associated with a finite sample size and reveal our sample distribution more clearly . The PDF of the track durations is related to bleaching and unbinding as follows: ( 3 ) ktracke−ktrackt= ( koff+kbleach ) e− ( koff+kbleach ) t Where ktrack is the rate of track durations ending , koff is the rate of unbinding , and kbleach is the rate of bleaching . The model PDF we used for fitting was a left-truncated exponential distribution . This was used to compensate for the fact that we removed short duration tracks from analysis . The general form of this PDF is: ( 4 ) ( 1τ ) e− ( x−L ) τ where τ is the mean time , and L is the truncation point/origin of exponential distribution ( Balakrishnan and Basu , 1995 ) . Note that the equation has the same form as expected for a translated exponential distribution and so we used this form for all data sets . To correct for photobleaching , we used LacI tagged with mMaple . LacI is expected to have a binding time significantly longer than the bleaching time of mMaple in our experimental conditions ( Hammar et al . , 2014 ) . Therefore , since the koff term is much smaller than the kbleach term , the average track duration is equivalent to the average bleach time for mMaple . Note that previous estimates of LacI bound-time at the lacO operator were determined using a single copy of the operator , while we used an array composed of 256 copies of lacO . This should result in even longer apparent binding of the repressor protein and increase the likelihood that focus disappearance is solely due to bleaching . We obtained a constant exposure bleaching curve , which we used for the 1 s , and 5 s intervals ( 500 ms exposure data ) . We scaled the average bleach time from the constant exposure bleaching data , in order to use it for different time intervals . The constant exposure bleaching time is related to the average bleaching time as follows: ( 5 ) Tbleach= ( tintervaltexp ) Tconstant Where tinterval is the interval time , texp is the exposure time , and Tconstant is the constant exposure bleaching time . We then calculated the average bound time using the following equation: ( 6 ) Tbound=TtrackTbleachTbleach−Ttrack In order to calculate the SE and 95% CI on the parameter estimates , we used the right-hand side of Equation 3 . We used the bleaching times and bound times calculated previously , as initial estimates , and then performed bootstrap sampling over 10 , 000 samples , in order to calculate the standard errors and confidence intervals on the bound time estimates . We used the ‘bias and accelerated percentile method’ ( BCA ) algorithm when calculating CI , in order to compensate for any bias or skewness in the bootstrap distribution . Previous characterization of photoblinking of mMaple found 49% probability of blinking and an average of 3 . 4 blinking events per molecule ( Durisic et al . , 2014 ) . This same study set a cutoff time of 2 . 6 s to account for over 99% of the blinking events . We expected to detect fewer blinking events since the shorter ones will be recorded only as intensity fluctuations , and not discontinuities in the track , due to the use of longer capture rates , 500 ms instead of 100 ms . In addition , lower exposure intensity would likely contribute to a decrease rate of blinking ( Garcia-Parajo et al . , 2000 ) . To estimate the effect of blinking in our analysis , we used the data of LacI using 500 ms capture-times . We analysed the data as previously except that we did not apply the one-frame memory parameter during tracking . We then determined the number of frames between two consecutive tracks at the same position of the field of view . We used a 2 . 6 s cutoff in our data since longer gap times likely represent new binding events instead of blinking . For DnaB , since 500 ms capture-times were not efficient at preventing diffusing molecules from being detected , even after the PSF threshold was applied ( Figure 2—figure supplement 4G ) , we used exposure times of 2 s and spaced capture by intervals of 2 s and 10 s . We used Pol III ε subunit as a control to ensure that increasing the exposure time does not significantly alter the bound time estimates ( Figure 2—figure supplement 3 ) . Since the track durations of DnaB are similar to those of LacI , we determined a weighted average of the track duration times obtained and for each data set of DnaB performed a constrained fit ( i . e . fitting with bounds placed on the estimates ) . We calculated a bound time from the weighted average in order to generate an initial estimate of the bound time , which was then used for the constrained fit . We allowed for 20% variation in the bleaching time in order to determine physically reasonable estimates . The lower and upper bounds for the DnaB bound time were 1 s and 90 min , respectively . For the fitting procedure , we calculated the negative log-likelihood function of the two parameter ( bleach time , bound time ) left-truncated exponential distribution , as well as the gradients . We then used the MATLAB minimization function , fmincon , in order to find the parameters that minimize the negative log-likelihood function . This was done to improve the convergence to the correct solution , especially if the initial estimates were far from the actual solution , and to simplify the estimation procedure . We subsequently performed bootstrap sampling as discussed before to calculate standard errors and confidence intervals ( Supplementary file 1B ) . The final estimates for bound times were calculated by doing a weighted average of data taken on multiple days and with different time intervals . We performed a chi-square goodness of fit test under the null hypothesis that our data is sampled from a single-exponential distribution , and the alternative hypothesis that it does not . It is possible however that even if the fit is good , that a different model fits the data better ( e . g . two exponential model ) . We wanted to determine the best model for the data and we performed two tests in this regard: ( 1 ) Log-Likelihood Ratio ( LLR ) test and ( 2 ) Bayesian Information Criterion ( BIC ) test . Log-Likelihood Ratio Test- The LLR test tries to test if an unconstrained model statistically significantly fits the data better than the constrained model , by comparing the likelihood values obtained from the unconstrained versus the constrained . In our case , the unconstrained model is the two-exponential model while the one-exponential model is the constrained model , as shown below: ( 7 ) p ( 1τ1 ) e− ( x−L ) τ1+ ( 1−p ) ( 1τ2 ) e− ( x−L ) τ2 where τ1= ( Tbleach+Tboundα ) /TbleachTboundα , τ2= ( Tbleach+Tboundβ ) /TbleachTboundβ , p is the mixture probability , and L is the truncation point . Note that if we constrain p=1 , we recover the single-exponential model . Bayesian Information Criterion ( BIC ) test- The BIC test determines which model fits the data better , but penalizes for greater complexity ( i . e . more parameters ) , to prevent over-fitting the data . The lowest number obtained through the test indicates the model that fits the data the best with the least complexity . When calculating the MLE estimates and log-likelihood values for the two-exponential model , the lower and upper bounds for the two timescales were 0 . 1 and 5400 s , respectively . The bounds for the bleaching constant were placed such that it allowed for 20% variation in the estimate . The criterion we used to judge if the two-exponential model was the better model , was if the BIC test gave the lowest value for the two-exponential model and the LLR test gave a p<0 . 01 . Also , the estimates obtained from the two-exponential should be sensible , and especially , they should not give us simply the values of the bounds , as that indicates that no estimates were found . We found a few cases where the dataset passed the criterion . The timescales estimated were not consistent however , and upon further examination we realized that it was due to a few noticeable outliers in the dataset , possibly from noise due to dirt still on the coverslip . When we removed the outliers , it resulted in these datasets not passing the criterion , but without significantly changing the bound times previously obtained in the single-exponential fits . We used LacI data collected with 500 ms exposure as fast as possible ( ~500 ms interval time ) , to characterize mMaple under our acquisition settings . The mean positions of single-molecule tracks initiating at the first frame were used as ROIs around which a 7 × 7 pixel window was drawn to extract intensity-time traces . Fluorescence from a bound molecule was identified as being 2 standard deviations above the mean cellular background , and the signal had to be above this threshold for >3 localizations ( similar to track acceptance threshold previously described ) . Gap durations were calculated as the number of frames between bound fluorescence signals . Fits to the gap durations were done through MLE using a truncated exponential model . The resulting fit was used to calculate the probability of a gap duration lasting greater than a specified value , through integration . To estimate the effective loading rate we followed the following equation described elsewhere ( Moolman et al . , 2014 ) : ( 8 ) 1Teff_load=β2boundTunload where Teff_load represents the loading rate , β2bound represents the number of copies of β clamp at the fork and Tunload represents the bound-time of a β clamp . In our estimations we assume that there are 23 β dimers per fork as previously estimated ( Moolman et al . , 2014 ) .	New cells are created when an existing cell divides to produce two new ones . During this process the original cell must copy its DNA so each new cell inherits a full set of genetic material . DNA is made up of two strands that twist together to form a double helix . These strands need to be separated so they can be used as templates to make new DNA strands . An enzyme called DNA helicase is responsible for separating the two DNA strands and another enzyme makes the new DNA . These enzymes are part of a group of proteins collectively called the replisome that controls the whole DNA copying process . The replisome must be extremely reliable to avoid introducing mistakes into the cell’s genes . Previous research using replisomes extracted from cells indicated that replisomes are effective at copying DNA because the proteins they contain are strongly bound together and remain attached to the DNA for a long time . However , the behavior of replisomes in living cells has not been closely examined . Beattie , Kapadia et al . used microscopy to observe how the replisome copies DNA in a bacterium called Escherichia coli . The experiments revealed that most of the proteins within the replisome are constantly being replaced during DNA copying . The exception to this is DNA helicase , which stays in place at the front of the replisome , providing a landing platform for all the other parts of the machine to come and go . Future work will investigate why the parts of the replisome are replaced so frequently . This may allow us to alter the stability of the bacterial replisome , which may lead to new medical treatments and biotechnologies .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "chromosomes", "and", "gene", "expression", "biochemistry", "and", "chemical", "biology" ]	2017	Frequent exchange of the DNA polymerase during bacterial chromosome replication

Subsets and Splits

No community queries yet

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